ɑ1,3-mannosyltransferase promotes the malignant progression of bladder cancer through activating TNF signaling pathway

Abstract

Background

Bladder cancer (BCa) is the most prevalent malignancy of the urinary system. Aberrant glycosylation, driven by specific glycosyltransferases (GTs), plays a pivotal role in various carcinogenic processes. However, the role of GTs-related glycobiomarkers and their underlying mechanisms in BCa remain poorly understood.

Methods

A diagnostic model based on GTs was constructed and validated using multiple bioinformatics tools. The diagnostic and prognostic value, biological functions and potential targeted drugs were assessed using R packages, K–M plotter and molecular docking. The functional impact and mechanism of ALG3 in BCa were investigated through functional assays, RNA sequencing, immunoprecipitation, and lectin pull down assays.

Results

A diagnostic model comprising six GTs (ALG3, POMT2, UGCG, XXYLT1, COLGALT1, and A4GALT) was established, demonstrating high diagnostic accuracy for BCa (AUC: 0.966; sensitivity: 88.5%; specificity: 92.6%), which was further validated. Among these, ALG3 and POMT2 belong to the mannosyltransferase family, with ALG3 identified as a more reliable diagnostic glycobiomarker than POMT2 for BCa detection. GSVA analysis revealed that ALG3 was significantly enriched in Umbrella cells, suggesting its potential role in influencing BCa cell fate. Overexpression of ALG3 promoted cell proliferation and metastasis by modulating CD44 N-glycosylation and activating the TNF signaling pathway, confirming its role as a tumor promoter and oncogene in BCa progression. Moreover, ALG3 was identified as a novel target of miR-142-5p. Four potential small molecule inhibitors of ALG3 were identified, with selumetinib emerging as a promising candidate.

Conclusions

ALG3 contributed to BCa progression via CD44 N-glycosylation and TNF pathway, positioning it as a promising serum glycobiomarker, a feasible therapeutic target, and a valuable reference for personalized and precision medicine in BCa.

Introduction

Bladder cancer, one of the most prevalent malignancies of the urinary system, exhibits high prevalence and increasing morbidity [1, 2]. It is among the most costly diseases to manage over a patient’s lifetime, posing a significant burden on healthcare system. Recurrence and a high propensity for metastasis are defining characteristics of bladder cancer [3, 4]. Based on the extent of invasion, bladder cancer is classified into two categories: non-muscle invasive bladder cancer (NMIBC), which has a relatively favorable prognosis, and muscle invasive bladder cancer (MIBC), which is highly aggressive. Studies reported that 60–80% of NMIBC patients experience recurrence, while 20–40% progress to MIBC, where the 5-year survival rate drops to just 60% [5]. Currently, cystoscopy and urine cytology are the primary techniques for bladder cancer diagnosis. However, cystoscopy is invasive and expensive, whereas urine cytology, although minimally invasive and highly specific, suffers from low sensitivity [6]. Consequently, there is an urgent need for targeted detection methods that offer both high sensitivity and specificity to enhance the clinical diagnosis of bladder cancer. In addition, the toxicity and side effects of chemotherapy drugs, along with associated complications, limit patients’ tolerance and willingness to undergo aggressive treatment. The absence of reliable predictive biomarkers to guide treatment selection and outcome evaluation contributes to high rates of recurrence, progression and metastasis, leading to poor long-term prognosis. This highlights the critical need to identify novel diagnostic biomarkers to improve clinical detection, as well as therapeutic targets to support personalized treatment strategies for bladder cancer.

Glycosylation, one of the most common post-translational modifications, is regulated by the coordinated action of specific glycosyltransferases and glycosidases. Based on the linkages between oligosaccharides and proteins, glycosylation is categorized into two main forms: N-linked glycosylation and O-linked glycosylation. Abnormal glycosylation has been implicated in a wide range of diseases [7,8,9,10], particularly cancer, which remains one of the most life-threatening conditions globally. As a key hallmark of cancer, glycosylation plays a critical role in essential carcinogenic processes, including epithelial–mesenchymal transition [11], angiogenesis [12], malignant transformation [13] and immune escape [14]. Targeting glycosylation offers a promising strategy for advancing both the diagnosis and treatment of cancer.

As key glycoenzymes responsible for protein glycosylation, multiple glycosyltransferase families have been identified to date, including fucosyltransferases, mannosyltransferases, and sialyltransferases, etc. Aberrant glycosylation, resulting from the dysregulated expression of these glycosyltransferases, has been observed in various cancers. For instance, He et al. reported that fucosyltransferase 2 inhibited epithelial–mesenchymal transition and metastasis in colorectal cancer through inducing α−1,2 fucosylation of LRP-1, suggesting its potential as a therapeutic target [15]. Li et al. demonstrated that protein O-fucosyltransferase 1 promoted immune evasion by enhancing the stability of PD-L1, thereby guiding liver cancer toward immunotherapy [16]. Scott et al. identified GALNT7 as a novel biomarker for prostate cancer diagnosis, with GALNT7-mediated O-glycosylation playing a critical role in prostate cancer progression [17]. Wu et al. suggested that inhibition of ALG3 enhanced the responsiveness of breast cancer cells to 5-fluorouracil treatment, highlighting its potential as a prognostic indicator, a biomarker for immune infiltration, and a promising therapeutic strategy for cancer [18]. These findings underscore the importance of comprehensive and systematic research on the expression profiles and functions of glycosyltransferases, which will help elucidate the pathogenesis of cancer and identify effective biomarkers for cancer diagnosis, prognosis and treatment evaluation.

In the current study, 212 GTs were compiled, and bladder cancer data sets (TCGA–BLCA, GSE13507, GSE188715 and GSE135337) were downloaded. A GTs-associated diagnostic model for bladder cancer was developed and validated. Subsequently, the expression profiles, diagnostic value, and prognostic significance of two key members—ALG3 and POMT2, were assessed. Ultimately, ALG3 was identified as a promising glycobiomarker, driving the malignant behaviors of bladder cancer via modulation of CD44 N-glycosylation and activation of the TNF signaling pathway. Moreover, miR-142-5p/ALG3 regulatory axis was identified as a critical molecular switch governing this pathogenic cascade.

Materials and methods

Source of data sets

The bladder cancer data sets utilized in this study were obtained from the Cancer Genome Atlas (TCGA) database (https://portal.gdc.cancer.gov/) and the Gene Expression Omnibus (GEO) database (https://www.ncbi.nlm.nih.gov/geo/). The TCGA–BLCA data set, comprising 411 tumor tissues and 19 normal bladder tissues, was applied for machine learning analysis. The GSE188715 data set, which includes 57 tumor tissues and 13 normal bladder tissues, was employed to validate the established diagnostic model based on six essential GTs. In addition, the single-cell RNA sequencing (scRNA-seq) data set, GSE135337, containing 7 primary tumor samples and 1 paracancerous tissue sample, was utilized for cell group annotation. The GSE211692 data set, containing 399 bladder cancer serum samples and 5643 healthy controls, was utilized for miR-142-5p expression validation.

GTs extraction

A total of 212 GTs were extracted from the GGDB GlycoGene database (https://acgg.asia/ggdb2/). Differentially expressed GTs were identified using the R packages “limma” and “wilcox” on the TCGA–BLCA data set, with a threshold of |Log2 FC|≥ 1 and padj < 0.05.

Establishment of the diagnostic model based on differentially expressed GTs

To develop an effective diagnostic model for bladder cancer detection, the most diagnostically significant GTs were selected using the R package “caret”. This package facilitated the implementation of Gradient Boosting Machine (GBM), least absolute shrinkage and selection operator (LASSO), Random Forest (RF), and eXtreme Gradient Boosting (XGB) analysis. The importance of each gene was evaluated, and the top 15 genes from each method were identified. Common genes shared across all methods were then used to construct a diagnostic model through binary logistic regression analysis. Receiver operating characteristic (ROC) curves were generated for individual genes and the combined gene set using the R package “timeROC” to assess the model’s diagnostic performance. The model was further validated using the GSE188715 data set.

Associations of six critical GTs with clinicopathological parameters of bladder cancer

The relationships between the expression of the six GTs and the clinical features of bladder cancer, including cancer stage and lymph nodal metastasis status, were analyzed using the UALCAN database (https://ualcan.path.uab.edu/analysis.html).

Expression and prognostic value of ALG3 and POMT2 in bladder cancer

The expression levels of ALG3 and POMT2 across various cancers were assessed using the TIMER database (https://cistrome.shinyapps.io/timer/). The normalized gene expression matrix of BLCA patients was utilized to extract the mRNA levels of ALG3 and POMT2, and the results were visualized using the R package “ggplot2”. To evaluate their prognostic significance in bladder cancer, overall survival curves based on ALG3 and POMT2 expression were generated using the R packages “survival” and “survminer.

Gene ontology and gene set enrichment analysis

Patients with bladder cancer were stratified into high-ALG3 or POMT2 and low-ALG3 or POMT2 groups based on the median expression values derived from the TCGA–BLCA data set. Differentially expressed genes between these two groups were identified using the R package “limma”, with a threshold of |Log2 FC|≥ 1 and padj < 0.05. Functional annotation was performed through Gene Ontology (GO) analysis using the R package “clusterprofiler”. In addition, Gene Set Enrichment Analysis (GSEA) was conducted to explore differences in signaling pathways and biological processes. This analysis utilized the fold change of each gene calculated with the R package “limma”, with Hallmark Gene Sets and C2 Kegg Gene Sets as references, and was performed using GSEA software.

Correlation analysis between ALG3 or POMT2 and immune infiltration, immune checkpoints, and immunotherapy

The relationships between ALG3 or POMT2 expression and the infiltration levels of 22 immune cell types were analyzed using the R package “IOBR”. In addition, 46 immune checkpoint-related genes and 59 immunomodulators were compiled, and their correlations with ALG3 or POMT2 expression were evaluated using the R package “corrplot”. To predict the response of bladder cancer patients to immunotherapy based on ALG3 or POMT2 expression, data from the TCGA–BLCA data set were normalized using the R package “scale”. The immunotherapy response of high-ALG3 or POMT2 groups and low-ALG3 or POMT2 groups to anti-PD1 and anti-CTLA4 treatments was then assessed.

Single-cell RNA sequencing data analysis

The normalized RNA expression matrix of the GSE135337 data set was converted into scRNA-seq data using the R package “seurat”. Quality control was performed to filter out low-quality or biased cells, resulting in a final data set of 19,459 cells for subsequent analysis. Dimensionality reduction was achieved through principal component analysis using the R package “RunPCA”, followed by t-distributed stochastic neighbor embedding with the R package “tSNER”, and “Uniform Manifold Approximation and Projection (UMAP)” to identify cellular subpopulations. These subpopulations were annotated using the R package “SingleR”, with reference data from CellMarker (http://xteam.xbio.top/CellMarker/) based on the expression of marker genes.

Analysis of communication between cell subgroups

Following the identification of cellular subpopulations, intercellular communication among diverse cell types and receptor–ligand pairs was explored using the R package “CellChat”. The probability of cell communication was computed to quantify the strength of these interactions.

Clinical serum samples

Serum samples from 77 healthy controls and 96 bladder cancer patients were collected from the First Affiliated Hospital of Dalian Medical University between December 2022 and February 2024. All patients were diagnosed through pathological examination and had not undergone any preoperative treatment. This study was approved by the Ethics Committee of the First Affiliated Hospital of Dalian Medical University (No. PJ-KS-KY-2022–439), and written informed consent was obtained from all participants.

Enzyme-linked immunosorbent assay

Serum levels of ALG3 and POMT2 in bladder cancer patients were quantified using commercial double-antibody sandwich enzyme-linked immunosorbent assay (ELISA) kits (Westang, Shanghai, China). Serum samples (100 μl) from both patients and healthy controls, along with standard solutions, were added to 96-well microplates pre-coated with ALG3 or POMT2 antibodies and incubated at 37 ℃ for 40 min. After washing 4–6 times, 50 μl of biotin-labeled primary antibody was added to each well and incubated at 37 ℃ for 20 min. Subsequently, 100 μl of streptavidin–HRP-labeled second antibody was added to each well and incubated for 10 min at 37 ℃. Afterward, 100 μl of tetramethylbenzidine substrate solution was added to initiate the colorimetric reaction, and the signals were developed in the dark at 37 ℃ for 15 min. Absorbance at 450 nm (OD450nm) was measured and recorded after the reaction was terminated by adding stop solution. Serum concentrations of ALG3 and POMT2 were calculated using a standard curve, and statistical analysis was performed using an independent Student’s t test.

Cell culture and transfection

The human bladder cancer cell lines T24 (JNO-H0484) and UMUC3 (JNO-H0486) were purchased from Jennio Biotechnology (Guangzhou, China). T24 cells were maintained in RPMI-1640 (Gibco, California, USA) medium, and UMUC3 cells were cultured in DMEM-H (Gibco) medium, both supplemented with 10% fetal bovine serum (FBS, Gibco) in a 5% CO₂ atmosphere at 37 ℃. The culture medium was refreshed every 2–3 days depending on cell growth conditions.

For transfection, ALG3 cDNA was used at a final concentration of 4 μg, ALG3 siRNA was applied at a final concentration of 50 pmol, miR-142-5p mimics and inhibitor was used at a final concentration of 50 pmol, using Lipofectamine 3000 reagent (Invitrogen, Carlsbad, CA, USA) according to the manufacturer’s instructions. The ALG3 siRNA sequences were as follows: 5′-GCGUCAUCAAUGGUACCUATT-3′ (sense), 5′-UAGGUACCAUUGAUGACGCTT-3′ (anti-sense). miR-142-5p mimics: 5′-CAUAAAGUAGAAAGCACUACU-3′ (sense); 5′-UAGUGCUUUCUACUUUAUGUU-3′ (anti-sense); miR-142-5p inhibitor: 5′-AGUAGUGCUUUCUACUUUAUG-3′ (sense).

Quantitative real-time PCR

Total RNA was extracted using RNAiso Plus (TAKARA, Japan) according to the manufacturer’s instructions. After reverse transcription (PrimeScript™ RT reagent Kit, RR037Q, TAKARA, Japan), quantitative real-time PCR was performed to evaluate changes in ALG3 expression using TB Green^® Premix Ex Taq™ II (RR820 A, TAKARA, Japan). Data were normalized to GAPDH and analyzed using the 2^−△△CT method. All experiments were performed in triplicate. The primer sequences used were as follows: ALG3: Forward (F): 5′-CACCTTCTGGGTCATTCACAGG-3′; Reverse (R): 5′-GTGTCACCCTGCAGTTGGGTATAGT-3′; GAPDH: Forward (F): 5′-GCACCGTCAAGGCTGAGAAC-3′; Reverse (R): 5′-TGGTGAAGACGCCAGTGGA-3′. hsa-miR-142-5p (F): 5′-AGCTCGCGCATAAAGTAGAAAG-3′; hsa-miR-142-5p (R): 5′-TATGGTTGTTCTCGTCTCTGTGTC-3′. U6 (F): 5′-CAGCACATATACTAAAATTGGAACG-3′; U6 (R): 5′-ACGAATTTGCGTGTCATCC-3′.

Western blot

Total protein was extracted using RIPA lysis buffer (Beyotime Biotechnology, Beijing, China) supplemented with 1 mM PMSF (Beyotime Biotechnology, Beijing, China). Protein concentration was determined, and proteins were separated by 12% SDS–PAGE before being transferred onto a nitrocellulose membrane (Pall Corporation, NY, USA). The membranes were blocked with 5% skim milk at room temperature for 2 h, and then incubated with primary antibodies at 4℃ overnight: anti-ALG3 (1:1000, DF14333, Affinity Biosciences, Jiangsu, China), anti-TNF-α (1:1000, A22227, ABclonal Technology, Wuhan, China), anti-TNFR1 (1:1000, A1540, ABclonal Technology), anti-TRADD (1:1000, A18626, ABclonal Technology), anti-MMP3 (1:500, A11418, ABclonal Technology), anti-MMP9 (1:500, A23535, ABclonal Technology), and anti-GAPDH (1:5000, A19056, ABclonal Technology). After three washes with 1 × TBST for 10 min each, the membranes were incubated with HRP-conjugated goat anti-rabbit IgG (1:5000, AS014, ABclonal Technology) at room temperature for 45 min. Following additional washes, protein bands were visualized using an enhanced chemiluminescence system, and band intensity was quantified using Image J software. Statistical analysis was performed based on the quantified data.

Cell proliferation assay

After transfection with ALG3 cDNA (4 μg), cells were trypsinized and seeded into 96-well plates at a density of 3 × 10³ cells per well. Cell proliferation was assessed using the Cell Counting Kit-8 (CCK8, Beyotime Biotechnology) at 24, 48, 72, and 96 h, following the manufacturer’s protocol. At each timepoint, 10 μl of CCK8 reagent was mixed with 90 μl of fetal bovine serum-free medium and added to each well, followed by incubation at 37℃ for 2 h. The OD450nm was measured, and growth curves were generated.

Colony formation assay

Following ALG3 cDNA transfection, cells were seeded into 3 cm plates at a density of 1 × 10³ per plate and cultured for 10–14 days in a 5% CO₂ atmosphere at 37 ℃. Once colonies formed and contained more than 50 cells, they were washed with PBS, fixed with 100% methanol for 20 min, and stained with 0.5% crystal violet. Colony numbers were quantified after imaging under an inverted microscope (Olympus, Japan).

Cell migration and invasion assays

Cell migration and invasion capabilities were evaluated using transwell assays with chambers featuring an 8 μm pore size (Corning Costar). For the migration assay, chambers were left uncoated, while for the invasion assay, 50 μl of diluted Matrigel (1:8, BD Biosciences) was applied to the upper chambers. Briefly, 5 × 10⁴ cells in 200 μl of serum-free RPMI-1640 medium (for T24 cells) or DMEM-H medium (for UMUC3 cells) were seeded into the upper chambers, while 600 μl of complete medium was added to the lower chambers. After incubation at 37℃ for the specified durations, cells were fixed with 100% methanol for 20 min, stained with 0.5% crystal violet, and imaged under an inverted microscope. The number of cells that migrated or invaded to the lower side of the chamber was quantified from ten randomly fields using Image J software. Statistical analysis was performed accordingly.

Animal experiments

Animal experiments were conducted in accordance with protocols approved by the Animal Ethics Committee of Dalian Medical University (No. AEE23158). Male BALB/c nude mice (aged 4–6 weeks) were purchased and housed in the SPF center, with all procedures approved by the Ethics Committee of Dalian Medical University. A total of 3 × 10⁶ cells (oeNC and oeALG3 T24 cells stably expressing shNC and ALG3, respectively) in 100 μl of suspension were mixed with 100 μl of Matrigel and subcutaneously injected into the flanks of the nude mice. Tumor growth was monitored, and tumor length and width were measured every other day. Tumor volume was calculated using the formula: volume = (length × width²)/2. Images of the nude mice and excised tumor tissues were captured and harvested for subsequent experiments.

RNA and microRNA sequencing

RNA sequencing (RNA-seq) was conducted on ALG3-overexpressing UMUC3 cells, while microRNA sequencing (microRNA-seq) was performed on ALG3-downregulation UMUC3 cells. Total RNA was extracted and sent to Novogene Corporation (Tianjin, China) for library preparation and sequencing. Differentially expressed genes (DEGs) for RNA-seq were identified using the following criteria: |Log2 FC|≥ 1 and padj < 0.05. Differentially expressed miRNAs were identified using |Log2 FC|≥ 0.59 and padj < 0.05.

Drug screening and molecular docking

Potential small molecule inhibitors targeting ALG3 were screened using the GSCA database (http://bioinfo.life.hust.edu.cn/GSCA/#/). The structures of small molecules interacting with ALG3 were retrieved from the PubChem database (https://pubchem.ncbi.nlm.nih.gov/), and the ALG3 protein structure (ID: Q92685) was downloaded from the Uniprot database (https://www.uniprot.org/) for molecular docking. Protein preparation was performed using the “Protein Preparation Wizard” module in Schrödinger software, followed by binding site prediction using the “SiteMap” tool. Molecular docking was conducted using the “standard precision” mode in Schrödinger software, with the “GlideScore” function used for scoring. Ten conformations were generated for each small molecule, and binding energy minimization was performed post-docking. The docking results were visualized using PyMol software.

Immunoprecipitation

Cells were lysed on ice for 20 min and centrifuged to collect the supernatant. Protein concentration was measured using a BCA assay, and 500 μg of protein was incubated with 50 μl of CD44-coupled magnetic beads to enrich CD44 proteins from the total protein lysate for 2 h at room temperature. The immunoprecipitated proteins were analyzed by Western blot, with GNA lectin used to assess changes in CD44 N-glycosylation.

Lectin pull down

Total proteins were extracted using RIPA lysis buffer supplemented with protease inhibitors Cocktails and PMSF. After measuring protein concentration, equal amounts of protein samples (500 μg) from each group were incubated with 50 μl of GNA-coated agarose beads (Vector Labs, California, USA) at 4 ℃ overnight. Following elution and boiling at 100℃, the pulled-down proteins were separated by 12% SDS–PAGE. The proteins were then transferred to a nitrocellulose membrane and blocked with 5% non-fat milk. The membrane was incubated with the primary antibody against CD44 (1:2000, ABclonal Technology, Wuhan, China) at 4 ℃ overnight, followed by incubation with an HRP-conjugated secondary antibody for 45 min. Protein bands were visualized using an enhanced chemiluminiscent system.

Dual-luciferase reporter assay

To validate the direct interaction between miR-142-5p and ALG3, we conducted dual-luciferase reporter assays using HEK293 T cells. Cells were co-transfected with either wild-type ALG3 3′-UTR luciferase reporter plasmid (WT, GenePharma, Shanghai, China) containing the predicted miR-142-5p binding site or mutant type (MUT) ALG3 3'-UTR luciferase reporter plasmid (MUT, GenePharma, Shanghai, China) constructed with site-directed mutations in the seed region, along with miR-142-5p mimics (50 pmol) or negative control (NC). The pRL-TK Renilla luciferase plasmid was included for normalization. Luciferase activities were quantified 48 h post-transfection using the dual-luciferase reporter assay system (Solarbio, Beijing, China), confirming sequence-specific binding of miR-142-5p to the ALG3 3′-UTR.

Cell viability assay

Cells were seeded into 96-well plates at a density of 5 × 10³ cells per well and treated with varying concentrations of selumetinib (0, 1, 5, 10, 20, 40, 80, 160 μM) for 48 h. Subsequently, 10 μl of CCK-8 reagent was mixed with 90 μl of FBS-free medium and added to each well, followed by incubation at 37 ℃ for 2 h. Absorbance at 450 nm was measured, and the data were recorded. The IC₅₀ values were calculated, and growth curves were generated using GraphPad Prism 10.0 software.

Statistical analysis

All data were presented as mean ± standard deviation (SD) from at least three independent experiments. Statistical analysis was performed using SPSS 22.0 and GraphPad Prism 10.0, with Student’s t test or one-way ANOVA applied based on the data type. ROC curve analysis was conducted to assess the diagnostic value of ALG3 and POMT2 for bladder cancer. Survival curves were generated using the Kaplan–Meier method, and differences between the curves were evaluated using the log-rank test. A P value < 0.05 was considered statistically significant.

Results

Establishment of a bladder cancer diagnostic model based on differentially expressed GTs

Timely and accurate diagnosis is crucial for achieving favorable clinical outcomes in bladder cancer. Given the pivotal role of glycosylation in tumor progression, a total of 212 GTs were retrieved, and their differential expression was analyzed using the"wilcox"method (101 GTs) (Fig. 1A) and the"limma"method (54 GTs) (Fig. 1B). A total of 51 overlapping GTs were identified (Fig. 1C) and further analyzed using four machine learning algorithms (GBM, LASSO, RF and XGB) to identify genes with diagnostic potential (Fig. 1D). Ultimately, six overlapping GTs (ALG3, POMT2, UGCG, XXYLT1, COLGALT1, and A4GALT) were selected (Fig. 1E), and a diagnostic model was developed. The clinical utility of this model was assessed using ROC curves. When evaluated individually, ALG3 demonstrated a significantly higher AUC of 0.939 compared to the other five GTs (POMT2: 0.764, UGCG: 0.799, XXYLT1: 0.787, COLGALT1: 0.780, A4GALT: 0.758), with high sensitivity (84.1%) and specificity (88.9%) (Table 1). These results suggested that ALG3 hold substantial potential as a standalone diagnostic glycobiomarker for bladder cancer. When the six GTs were combined, the AUC increased to 0.966, with sensitivity rising to 88.5% and specificity to 92.6%, outperforming the individual GTs (Fig. 1F). The clinical applicability of this model was further validated using the GSE188715 data set (AUC: 0.960; sensitivity: 92.8%; specificity: 87.3%) (Fig. 1G) (Table 1). The diagnostic score was calculated using the following formula: diagnostic score = −1.652 + (0.136 × ALG3 expression) + (−0.018 × UGCG expression) + (0.132 × POMT2 expression) + (0.018 × XXYLT1 expression) + (0.008 × COLGALT1 expression) + (−0.019 × A4GALT expression). In summary, we successfully established a bladder cancer diagnostic model based on six differentially expressed GTs, which might help reduce the clinical burden of detection and improved diagnostic accuracy.

Fig. 1

Construction of a diagnostic model for bladder cancer based on six critical differentially expressed GTs. A, B Differentially expressed GTs in bladder cancer were screened using the R packages “wilcox” (A) and “limma” (B) with the TCGA–BLCA data set obtained from the TCGA database. C Venn analysis was performed to identify common GTs screened by both the “wilcox” and “limma” methods in bladder cancer. D Four machine learning algorithms (GBM, LASSO, RF, and XGB) was applied to screen essential differentially expressed GTs with diagnostic value for establishing the bladder cancer diagnostic model. E Venn analysis was used to visualize the common genes selected by the four machine learning algorithms (GBM, LASSO, RF, and XGB). F Diagnostic model for bladder cancer was established using six common critical GTs based on the TCGA–BLCA data set as the training set, with diagnostic accuracy evaluated through ROC curves. G Diagnostic accuracy of the constructed model was validated using the GSE188715 data set as the testing set, as assessed by ROC curves

Table 1 ROC curve analysis of the clinical application value of the established diagnostic model based on the differentially expressed GTs for bladder cancer

Elevated expression of intersecting GTs-ALG3 and POMT2 in bladder cancer

From the bladder cancer diagnostic model, six key GTs were identified. Among these, four GTs exhibited higher expression in patients with advanced-stage bladder cancer: ALG3 (Supplementary Fig. 1 A), POMT2 (Supplementary Fig. 1B), XXYLT1 (Supplementary Fig. 1D), and COLGALT1 (Supplementary Fig. 1E). In contrast, two GTs showed lower expression: UGCG (Supplementary Fig. 1 C) and A4GALT (Supplementary Fig. 1 F). However, no significant differences in the expression of these six GTs were observed between normal tissues and tissues from patients without lymph node metastasis (Supplementary Fig. 1G–L). In addition, four GTs–ALG3 (Supplementary Fig. 1G), UGCG (Supplementary Fig. 1I), COLGALT1 (Supplementary Fig. 1 K) and A4GALT (Supplementary Fig. 1L) also showed no significant changes between patients without lymph node metastasis and those with lymph node metastasis. For POMT2, a significant difference was observed between patients without lymph node metastasis and those with 4 to 9 axillary lymph node metastasis (Supplementary Fig. 1H). Similarly, for XXYLT1, a significant difference was noted between patients without lymph node metastasis and those with 10 or more axillary lymph node metastasis (Supplementary Fig. 1 J). Despite these findings, none of the six GTs proved to be ideal indicators for evaluating lymph node metastasis status. Notably, both ALG3 and POMT2 belong to the mannosyltransferase family. We, therefore, analyzed the expression profiles of mannosyltransferase family members in bladder cancer. Nine members (ALG1, ALG3, ALG5, ALG6, ALG8, ALG10, ALG14, POMT1 and POMT2) were found to be upregulated, while only one member (ALG9) was downregulated (Fig. 2A). Subsequently, we focused on ALG3 and POMT2. The TIMER database was utilized to examine their expression profiles, revealing that both ALG3 (Supplementary Fig. 2 A) and POMT2 (Supplementary Fig. 2B) were significantly upregulated in a majority of cancers, including bladder cancer. High expression of ALG3 (Fig. 2B, C) and POMT2 (Fig. 2E, F) was further validated in the GSE13507 and GSE188715 data sets, as well as in paired tissue samples from the TCGA–BLCA data set (Fig. 2D, G).

Fig. 2

Expression profile, prognostic value and enrichment analysis of ALG3 and POMT2 in bladder cancer. A Expression profile of mannosyltransferase family members (ALG1-14, POMT1-2) in bladder cancer. B, C Expression of ALG3 in non-paired bladder cancer tissues using the GSE13507 data set (B) and GSE188715 data set (C) from the GEO database. D Expression of ALG3 in paired bladder cancer tissues using the TCGA–BLCA data set from the TCGA database. E, F Expression of POMT2 in non-paired bladder cancer tissues using the GSE13507 data set (E) and GSE188715 data set (F) from the GEO database. G Expression of POMT2 in paired bladder cancer tissues using the TCGA–BLCA data set from the TCGA database. H, I Kaplan–Meier survival curves showing overall survival based on ALG3 (H) and POMT2 (I) expression in bladder cancer. J uniCox regression analysis assessing the prognostic value of age, gender, stage, and ALG3 expression for bladder cancer survival, presented with p values and hazard ratios. K uniCox regression analysis assessing the prognostic value of age, gender, grade, and POMT2 expression for bladder cancer survival, presented with p values and hazard ratios. L multiCox regression analysis evaluating the prognostic value of age, gender, stage, and ALG3 expression as independent markers for bladder cancer survival, presented with p values and hazard ratios. M multiCox regression analysis evaluating the prognostic value of age, gender, grade, and POMT2 expression as independent markers for bladder cancer survival, presented with p values and hazard ratios. N, O Circle diagrams of GO enrichment analysis showing functional enrichment results of differentially expressed genes based on ALG3 (N) and POMT2 (O) expression, including biological processes, molecular functions, and cellular components. P Five GSEA enrichment results highlighting ALG3-related pathways associated with cancer development and progression, including ROS, glycolysis, DNA repair, G2/M checkpoint and inflammatory response. Q Five GSEA enrichment results of POMT2-related pathways associated with cancer development and progression, including EMT, G2/M checkpoint, hypoxia, inflammatory response and KRAS activation. *P < 0.05, **P < 0.01, ***P < 0.001

Beyond confirming the elevated expression of ALG3 and POMT2, we further investigated their potential as independent prognostic biomarkers for predicting survival outcomes in bladder cancer patients. Kaplan–Meier survival curves, constructed using the TCGA–BLCA data set, demonstrated that patients with low expression of ALG3 (Fig. 2H) or POMT2 (Fig. 2I) exhibited significantly better overall survival rates compared to those with high expression. uniCox regression analysis identified ALG3 (HR = 1.422; P = 0.016), age (HR = 1.039; P < 0.001), and tumor stage (HR = 1.844; P < 0.001) as significant prognostic factors for bladder cancer (Fig. 2J). These findings were further confirmed by multiCox regression analysis (Fig. 2L), establishing ALG3 as a potential independent predictor of overall survival. Similarly, POMT2 (HR = 1.063; P = 0.007) and age (HR = 1.034; P < 0.001) were also identified as significant prognostic factors (Fig. 2K), with their independent prognostic value confirmed through multiCox regression analysis (Fig. 2M). These results highlighted that both ALG3 and POMT2 served as potential independent prognostic glycobiomarkers with promising clinical applications for predicting outcomes in bladder cancer patients.

After observing the expression changes and prognostic significance of ALG3 and POMT2, we stratified the TCGA–BLCA data set into high- and low-expression groups for both markers. Differentially expressed genes (DEGs) were then identified, and GO enrichment analysis was performed. The results showed that ALG3-related DEGs were primarily enriched in glycosaminoglycan and heparin binding, as well as various enzyme activities involved in metabolic processes, such as endopeptidase, serine-type endopeptidase and serine hydrolase activity (Fig. 2N). In contrast, POMT2-related DEGs were linked to cytoskeletal organization, skin development, antigen binding, receptor–ligand activity and serine-type peptidase activity (Fig. 2O). GSEA results further revealed that ALG3-related DEGs were significantly enriched in pathways related to glycolysis, DNA repair, the G2/M checkpoint, inflammatory response, and reactive oxygen species production (Fig. 2P). Similarly, POMT2-related DEGs were linked to epithelial–mesenchymal transition, the G2/M checkpoint, hypoxia, inflammatory response and KRAS activation (Fig. 2Q). These findings suggested that ALG3 and POMT2 regulated multiple tumor-promoting processes, providing valuable insights into the mechanisms underlying bladder cancer progression and highlighting potential targets for future therapeutic interventions.

Correlation analysis of ALG3 and POMT2 with the tumor microenvironment in bladder cancer

The tumor microenvironment (TME) plays a crucial role in tumor initiation and metastasis, comprising immune cells, tumor-associated fibroblasts, and extracellular matrix, etc. We investigated the relationships between the expression of ALG3 and POMT2 and the infiltration levels of 22 immune cell types. Our analysis revealed that ALG3 expression was significantly positively correlated with M0, M1 and M2 macrophages, while showing negative correlations with activated dendritic cells, resting memory CD4 T cells, naive CD4 T cells, plasma cells, regulatory T cells, and naive B cells (Fig. 3A). In contrast, POMT2 expression exhibited significant positive correlations with resting NK cells and M2 macrophages, but negative correlations with plasma cells and naive B cells (Fig. 3B). These findings suggested that ALG3 and POMT2 could influence bladder cancer progression by modulating the infiltration of specific immune cell populations within the TME. Given that immune cells in the TME produce and secrete cytokines and chemokines that regulate tumor development, we further evaluated the associations between ALG3 and POMT2 expression and 59 chemokines. ALG3 showed the strongest correlations with CXCL5, CXCL1 and CCL24, while POMT2 was most significantly correlated with CXCL16, CCL7, CCL5 and CCL28 (Fig. 3C). These chemokines might underlie the observed differences in immune cell infiltration within the TME.

Fig. 3

Correlations between two common GTs and immune infiltration, immunomodulators, immune checkpoints, and immunotherapy. A, B Relationships between ALG3 (A), POMT2 (B) expression and the abundance of 22 immune cell types, analyzed using Pearson correlation analysis. C Correlation between ALG3 or POMT2 expression and 59 immunomodulators. D, E Associations between ALG3 (D) and POMT2 (E) expression and 46 immune checkpoints, evaluating using Pearson correlation analysis. F, G Responsiveness of patients with low- or high-ALG3 (F) or POMT2 (G) expression to immunotherapy, including anti-CTLA-4 and anti-PD1 treatment. *P < 0.05, **P < 0.01, ***P < 0.001

Immune checkpoints play a critical role in distinguishing self from non-self, and tumor cells often exploit these immune checkpoints to evade immune detection. We analyzed the correlation between ALG3 and POMT2 expression and 46 immune checkpoint molecules. ALG3 exhibited significant positive correlations with 11 immune checkpoint molecules (CD276, CD44, TNFRSF18, LAG3, HAVCR2, CD86, PDCD1LG2, TNFRSF8, CD274, LAIR1 and ICOSLG), and negative correlations with 7 molecules (CD28, TNFSF15, CD40LG, BTNL2, CD200, BTLA and ADORA2 A) (Fig. 3D). In contrast, POMT2 showed significant positive correlations with 16 immune checkpoint molecules (CD44, CD276, NRP1, TNFRSF25, TNFRSF18, CD274, PDCD1LG2, CD70, LAG3, CD160, TNFSF9, CD80, CD200R1, ADORA2 A, TNFSF14 and TNFSF18), and negative correlations with 5 molecules (BTLA, TNFSF15, CD40LG, VTCN1 and CD244) (Fig. 3E). These results illustrated that ALG3 and POMT2 might contribute to immune evasion by interacting with immune checkpoint molecules. Given the growing prominence of immune checkpoint inhibitors (ICIs) as promising cancer therapeutics, we assessed the response of bladder cancer patients to ICIs based on ALG3 (Fig. 3F) and POMT2 (Fig. 3G) expression. Our analysis revealed that only patients with high POMT2 expression exhibited a reduction in the immune-related signature (IPS) score following anti-CTLA-4 treatment (Fig. 3G). These findings suggested that ALG3 and POMT2 might not be suitable targets for immunotherapy in bladder cancer.

Intercellular communication between diverse cell subtypes investigated with scRNA-seq data set

After quality control and multiple filtering steps, 11 distinct cell clusters were identified and annotated using the GSE135337 data set (Fig. 4A, B). These clusters included epithelial cells, basal cells, fibroblasts, luminal cells, umbrella cells, tumor-associated macrophages, P53-like cells, endothelial cells, T cells, myofibroblasts and B cells. The intercellular communication networks between these clusters were visualized (Fig. 4C), revealing mutual regulatory interactions between tumor cells and the TME. We hypothesized that the complex communication and coordinated contributions among these diverse cell clusters collectively influence tumor progression. To further explore these interactions, we analyzed the incoming and outgoing signals between the 11 cell clusters. Cytokines or ligands secreted by one cluster acted as signal senders, while extracellular ligands binding to receptors on other clusters acted as signal receivers (Fig. 4D, E). This analysis underscored the essential roles these cell clusters play in tumor development and highlighted the intricate signaling networks within the TME.

Fig. 4

Clustering and annotation of single-cell RNA sequencing data in GSE135337. A t-SNE and UMAP plots displaying merged cells from bladder cancer tissues and normal bladder tissues, with 11 cell types identified using specific cell markers. B Bubble plot illustrating marker genes expressed in the 11 identified cell types. C Number of interactions and interaction weights/strength among 11 cell subtypes within the tumor microenvironment. D Bubble plot highlighting the essential incoming and outgoing signaling patterns of the 11 cell subpopulations. E Sankey diagram depicting the multiple communication pathways between different cell types

We next examined the expression patterns of the six critical GTs across different cell types (Fig. 5) (Supplementary Fig. 3), with a particularly focus on ALG3 and POMT2. ALG3 was predominantly expressed in Umbrella cells and Basal cells (Fig. 5A), while POMT2 was mainly expressed in Fibroblasts and Basal cells (Fig. 5B). To further explore the biological processes associated with ALG3 and POMT2, we isolated Umbrella cells, Basal cells, and Fibroblasts into tumor and normal groups, and performed GSEA analysis. For ALG3, processes such as inflammatory response, epithelial–mesenchymal transition, glycolysis and apoptosis were activated in ALG3 + Umbrella cells (Fig. 5C), while the G2M checkpoint and E2F targets were suppressed in ALG3 + Basal cells (Fig. 5D). In contrast, for POMT2, processes including inflammatory response, androgen response, angiogenesis, epithelial–mesenchymal transition and apoptosis were activated, while DNA repair was suppressed in POMT2 + fibroblasts (Fig. 5E). In addition, the IL2–STAT5 signaling pathway, DNA repair and epithelial-mesenchymal transition were inhibited in POMT2 + Basal cells (Fig. 5F). These findings suggested that ALG3 + and POMT2 + cell subgroups might play significant roles in the initiation and metastasis of bladder cancer by modulating specific biological processes.

Fig. 5

Predominant expression of ALG3 and POMT2 and GSEA analysis to screen their cancer-associated biological processes. A, B Expression of ALG3 (A) and POMT2 (B) in specific identified cell types, analyzed using the GSE135337 data set and visualized via UMAP plots. C, D GSEA enrichment analysis of cancer-associated biological processes related to ALG3 in umbrella cells (C) and basal cells (D). E, F GSEA enrichment analysis of cancer-associated biological processes related to POMT2 in fibroblasts (E) and basal cells (F)

ALG3 as a potential diagnostic glycobiomarker for bladder cancer

After validating the expression levels and identifying potential target drugs, we analyzed serum samples to evaluate the diagnostic value of ALG3 and POMT2 for bladder cancer. The results showed that serum levels of ALG3 (Fig. 6A) and POMT2 (Fig. 6B) were significantly elevated in bladder cancer patients, aligning with the tissue expression patterns observed in the database. Furthermore, ALG3 levels were notably higher in the invasive subtype compared to the non-invasive subtype (Fig. 6C), whereas POMT2 levels showed no significant difference (Fig. 6D). These findings suggested that ALG3 might play a role in promoting muscle invasion in bladder cancer, highlighting its potential as a diagnostic glycobiomarker.

Fig. 6

Diagnostic values of ALG3 and POMT2 for bladder cancer. A, B ELISA assays measuring the serum concentrations of ALG3 (A) and POMT2 (B) in bladder cancer patients and healthy controls. C, D ELISA assays measuring the serum levels of ALG3 (C) and POMT2 (D) in non-invasive and invasive subtypes of bladder cancer patients. E–G ROC curve analysis evaluating the diagnostic value of ALG3 and POMT2 for bladder cancer based on expression data from ELISA assays (E), the TCGA–BLCA data set (F) and the GSE13507 data set (G). *P < 0.05, ***P < 0.001

The diagnostic performance of ALG3 and POMT2 was further evaluated. When assessed individually, ALG3 demonstrated a significantly higher AUC (0.896) compared to POMT2 (0.781), with ALG3 also exhibiting superior sensitivity (95.8% vs. 76%). Although both markers showed comparable specificity (ALG3 = 77.3% vs. POMT2 = 78.7%), the combined analysis of sensitivity, specificity, and AUC highlighted ALG3 as a more valuable and reliable diagnostic glycobiomarker for bladder cancer detection (Fig. 6E, Table 2). This conclusion was further supported by database analysis. In both the TCGA–BLCA (Fig. 6F, Table 2) and GSE13507 data sets (Fig. 6G, Table 2), the AUC for ALG3 was significantly higher than that for POMT2 (TCGA–BLCA: 0.944 vs. 0.775; GSE13507: 0.779 vs. 0.594). In addition, ALG3 outperformed POMT2 in sensitivity (TCGA–BLCA: 89.5% vs. 53.3%; GSE13507: 71.5% vs. 28.5%) and specificity (TCGA–BLCA: 89.5% vs. 89.5%; GSE13507: 98.3% vs. 81.0%), further confirming its superior diagnostic value. When considering combined detection (ALG3 + POMT2), the AUC (0.920) was higher than of either marker alone (ALG3: 0.896; POMT2: 0.781). Although the sensitivity of the combined marker was slightly lower than ALG3 alone (88.5% vs. 95.8%), its specificity improved, increasing from 77.3% (ALG3) or 78.7% (POMT2) to 82.7% (ALG3 + POMT2) (Fig. 6E, Table 2). Similar trends were observed in the ROC curves generated from the TCGA–BLCA (Fig. 6F, Table 2) and GSE13507 data sets (Fig. 6G, Table 2). In summary, ALG3 emerged as a more ideal glycobiomarker for individual diagnostic purposes, while the combined use of ALG3 and POMT2 offered a balanced approach with improved specificity, making it a suitable option for enhanced diagnostic accuracy.

Table 2 ROC curve analysis of the diagnostic value of ALG3 and POMT2 for bladder cancer

ALG3 promoted malignant behaviors of bladder cancer cells through activating the TNF signaling pathway

Following the observation of abnormal expression, diagnostic value, and prognostic significance of ALG3, we further explored its functional impact on bladder cancer cell behaviors. After transfecting cells with ALG3 cDNA and siRNA, quantitative real-time PCR (Fig. 7A, B) and Western blot (Fig. 7C, D) were performed to confirm successful upregulation and downregulation of ALG3 expression. Moreover, ALG3 overexpression significantly enhanced cell proliferation (Fig. 7E) and colony formation (Fig. 7G) compared to the mock group, while ALG3 knockdown reduced these effects compared to the scramble group (Fig. 7F, H). Similarly, transwell assays revealed that ALG3 overexpression increased cell migration and invasion (Fig. 7I), whereas ALG3 knockdown suppressed these capabilities (Fig. 7J). To further validate the tumor-promoting role of ALG3 in vivo, a tumorigenesis model of bladder cancer was established using stable ALG3-overexpressing T24 cells generated via lentiviral transduction. After subcutaneous injection, tumor growth was monitored, measured, and recorded every 2 days. Overexpression of ALG3 led to significant tumor growth promotion (Fig. 8A, B), as evidenced by larger tumor size (Fig. 8C), increased tumor weight (Fig. 8D), and a higher growth rate (Fig. 8E) compared to the control group. Collectively, these findings highlighted the tumor-promoting effects of ALG3 in bladder cancer, emphasizing its role in enhancing proliferation, migration, invasion and tumor growth.

Fig. 7

ALG3 promoted cell proliferation, migration, and invasion of bladder cancer cells. A, B Quantitative real-time PCR analysis of ALG3 expression in T24 and UMUC3 cells after transfection with ALG3 cDNA (A) or ALG3 siRNA (B). C, D Western blot analysis of ALG3 expression in T24 and UMUC3 cells after transfection with ALG3 cDNA (C) or ALG3 siRNA (D) for 48 h. E, F CCK8 assays assessing changes in cell proliferation ability after treatment with ALG3 cDNA (E) or ALG3 siRNA (F). G, H Colony formation assays evaluating cell proliferation ability after transfection with ALG3 cDNA (G) or ALG3 siRNA (H). I, J Transwell assays measuring cell migration and invasion abilities after treatment with ALG3 cDNA (I) or ALG3 siRNA (J). **P < 0.01, ***P < 0.001

Fig. 8

ALG3 enhanced the tumorigenesis of bladder cancer in vivo. A, B Overview of tumor formation in nude mice following subcutaneous injection of T24 cells transfected with oeNC (control) and oeALG3 (ALG3 overexpression). C Images of tumors extracted from the nude mice. D Comparison of tumor weights between the oeNC and oeALG3 groups. E Comparison of tumor volumes between the oeNC and oeALG3 groups at 7, 9, 11, 13, 15, 17, 19, 21, 23, 25, 27 and 29 days post-injection. *P < 0.05, **P < 0.01, ***P < 0.001

To uncover the signaling mechanisms mediated by ALG3, RNA samples were collected for RNA-seq analysis following ALG3 cDNA transfection in UMUC3 cells. A total of 234 upregulated genes and 217 downregulated genes were identified using the criteria |Log2 FC|≥ 1 and padj < 0.05 (Fig. 9A). These 451 DEGs were subsequently subjected to GO, Reactome and KEGG pathway analysis. GO enrichment analysis indicated that ALG3-related DEGs were associated with the regulation of vasculature development, cytokine production, angiogenesis and response to wounding (Fig. 9B). Reactome analysis revealed that ALG3-related DEGs were involved in extracellular matrix organization, cell–cell junction organization and cell–cell communication (Fig. 9C), processes crucial for tumor development and progression. KEGG pathway analysis demonstrated that multiple signaling pathways were activated following ALG3 expression changes, including the TNF signaling pathway, MAPK signaling pathway, IL-17 signaling pathway and NF-kappa B signaling pathway (Fig. 9D). Combining these findings with the signaling pathways identified in the scRNA-seq GSEA results (Fig. 5C), we focused on the TNF signaling pathway. As illustrated, increased expression of key factors such as TNF-α, TNFR1, TRADD, MMP3 and MMP9 was observed (Fig. 9E). These results confirmed that ALG3 activated the TNF signaling pathway to promote malignant behaviors in bladder cancer.

Fig. 9

ALG3 activated the TNF signaling pathway to promote bladder cancer cell behaviors. A Vocanol plot of differentially expressed genes after ALG3 overexpression in UMUC3 cells, obtained from RNA-seq analysis. B, C GO enrichment analysis (B) and Reactome analysis (C) of ALG3-related DEGs to identify altered biological processes. D KEGG enrichment analysis of signaling pathways altered by ALG3-related DEGs. E Western blot analysis of key factors (TNF-α, TNFR1, TRADD, MMP3 and MMP9) in the TNF signaling pathway after transfection with ALG3 cDNA. **P < 0.01, ***P < 0.001

CD44 was a potential target protein for ALG3

CD44, a highly glycosylated protein primarily localized on the cell membrane, was investigated as a potential target of ALG3. We first conducted molecular docking to assess their interaction (Fig. 10A). Subsequently, potential N-glycosites on CD44 were predicted using the UniProt database, identifying nine sites: N25, N57, N100, N110, N120, N350, N548, N599 and N636 (Fig. 10B). To assess changes in CD44 N-glycosylation, we employed magnetic beads coated with CD44 antibody to enrich CD44 proteins from total cellular proteins, and GNA lectin antibody was utilized to assess the changes in its N-glycosylation modification level. Notably, ALG3 upregulation significantly increased N-glycosylation level of CD44 (Fig. 10C). This finding was further validated using GNA lectin pull down assay, where GNA-coated agarose beads were used to enrich proteins containing N-glycosylated modification from total cellular proteins and CD44 antibody was applied to estimate whether CD44 was present among these N-glycosylated proteins. The results confirmed that proteins enriched by GNA-coated agarose beads included CD44, and an increased N-glycosylation level of CD44 following ALG3 overexpression was observed (Fig. 10D). These findings demonstrated that CD44 was an α−1,3 N-glycosylated target of ALG3, highlighting the functional relationship between ALG3 and CD44 in modulating glycosylation.

Fig. 10

CD44 was a potential target protein for ALG3. A Molecular docking analysis evaluating the interaction between ALG3 and CD44. B Visualization of predicted glycosites in CD44, obtained from the UniProt database. C Immunoprecipitation using an anti-CD44 antibody to enrich CD44 proteins in cells, followed by immunoblot analysis to detect changes in CD44 N-glycosylation after ALG3 overexpression. D Lectin pull down assays using agarose-bound GNA, followed by immunoblot analysis to detect alterations in CD44 N-glycosylation after ALG3 upregulation

ALG3 was a novel target of miR-142-5p

Our previous findings established ALG3 as an oncogenic glycobiomarker in bladder cancer, with its overexpression significantly correlating with poor clinical outcomes. However, the upstream regulatory mechanisms controlling ALG3 expression, particularly post-transcriptional regulation by miRNAs, remained unexplored. MicroRNAs (miRNAs) have emerged as pivotal regulators in tumorigenesis, with growing evidence highlighting their roles in cancer initiation, progression, and therapy resistance [19,20,21,22]. To identify ALG3-associated miRNAs, we performed microRNA-seq analysis following ALG3 siRNA knockdown, revealing 64 up-regulated and 61 downregulated miRNAs (Fig. 11A). Bioinformatics intersection of the 64 ALG3-suppressed miRNAs with TargetScan-predicted ALG3-targeting miRNAs (n = 11) identified miR-142-5p as the sole overlapping candidate (Fig. 11B), with a conserved binding sites in the ALG3 3′-UTR (Fig. 11C). Consistent with its tumor-suppressive potential, analysis of GSE211692 data set demonstrated significant downregulation of miR-142-5p in bladder cancer (Fig. 11D), showing an inverse correlation with ALG3 expression.

Fig. 11

Identification of ALG3 as a novel target of miR-142-5p. A microRNA-seq was performed to identify ALG3-associated miRNAs following ALG3 knockdown. B Venn diagram analysis intersecting up-regulated miRNAs from microRNA-seq (n = 64) with TargetScan-predicted ALG3-targeting miRNAs (n = 11). C Predicted binding sites between miR-142-5p and ALG3 3’-UTR from TargetScan. D Validation of miR-142-5p downregulation in bladder cancer from GSE211692 data set. E, F CCK8 assays were utlized to evaluate the effects of overexpression of miR-142-5p (E) and downregulation of miR-142-5p (F) on bladder cancer cell proliferation. G, H Colony formation assays were performed to assess the impacts of miR-142-5p mimics (G) and miR-142-5p inhibitor (H) on the colony forming abilities of bladder cancer cells. I, J Transwell migration and invasion assays were conducted to estimate the effects of overexpression and downregulation of miR-142-5p on cell metastatic capabilities. K Dual-gene luciferase activity assay was utilized to evaluate the direct binding between miR-142-5p and ALG3 3’-UTR. L, M qRT-PCR was conducted to assess the up-regulate effect of miR-142-5p mimics (L) and down-regulate effect of miR-142-5p inhibitor (M) on miR-142-5p expression. N–Q qRT-PCR (N, O) and western blot (P, Q) was conducted to evaluate the ALG3 expression after miR-142-5p mimics (N, P) and miR-142-5p inhibitor (O, Q) transfection for 48 h. **P < 0.01, ***P < 0.001

Functional characterization revealed that miR-142-5p overexpression inhibited cellular proliferation (CCK-8 assays) (Fig. 11E, F), reduced clonogenic capacity (colony formation assay) (Fig. 11G, H), suppressed migration and invasion capabilities (transwell assays) (Fig. 11I, J), while miR-142-5p inhibition produced opposite effects, collectively establishing its tumor-suppressive role in bladder cancer progression.

Having established the tumor-suppressive function of miR-142-5p, we proceeded to elucidate its molecular mechanism through comprehensive target validation. Dual-luciferase reporter assays demonstrated that co-transfection of miR-142-5p mimics and wild-type ALG3 3'-UTR (WT) resulted in a significant reduction in luciferase activity, while mutation of the predicted binding sites (MUT) abolished this suppression, confirming the sequence-specific nature of this interaction (Fig. 11K). To further characterize this regulatory relationship, we performed qRT-PCR and Western blot analysis under conditions of miR-142-5p overexpression (Fig. 11L) and inhibition (Fig. 11M). Consistent with the luciferase assay results, forced expression of miR-142-5p led to marked downregulation of both ALG3 mRNA (Fig. 11N) and protein levels (Fig. 11P). Conversely, miR-142-5p inhibition produced the opposite effect, significantly elevating ALG3 expression at both mRNA (Fig. 11O) and protein levels (Fig. 11Q). These findings suggested that ALG3 was a direct functional target of miR-142-5p in bladder cancer cells.

Screening of potential small molecule compounds targeting ALG3

Using the GSCA database, we screened for potential small molecule drugs targeting ALG3, leveraging data from both the CTRP and GDSC databases. Five small molecule compounds were identified as potential ALG3 inhibitors (Fig. 12A), and their binding patterns with ALG3 were visualized through molecular docking. Trametinib exhibited strong binding to ALG3 (binding energy: −7.441 kcal/mol) with eight hydrophobic interactions involving residues PHE, LEU, ALA and GLN (Fig. 12B). Selumetinib bound to ALG3 (binding energy: − 5.073 kcal/mol) through three hydrophobic interactions involving residues LEU and PRO (Fig. 12C). Olaparib demonstrated tight binding to ALG3 (binding energy: − 6.439 kcal/mol) via seven hydrophobic interactions involving residues ALA, PRO, LEU, GLN and VAL (Fig. 12D). Etoposide bound to ALG3 (binding energy: − 4.009 kcal/mol) through two hydrogen bonds and three hydrophobic interactions (Fig. 12E), although its binding affinity was weaker compared to the other compounds. SN-38 showed strong binding to ALG3 (binding energy: − 5.406 kcal/mol) through one hydrogen bond and four hydrophobic interactions (Fig. 12F) (Supplementary Table 1). These findings highlighted the potential of these small molecule compounds as ALG3 inhibitors, with trametinib, olaparib, and SN-38 exhibiting particularly strong binding affinities.

Fig. 12

Screening of potential small molecule compounds targeting ALG3. A Venn analysis of small molecule compounds targeting ALG3, as predicted by the CTRP and GDSC databases. B–F Visualization of the binding petterns between ALG3 and trametinib (B), selumetinib (C), olaparib (D), etoposide (E), and SN-38 (F) using molecular docking. G 3D structure of selumetinib obtained from the PubChem database. H, I CCK8 assays to assess cell viability after treatment with varying concentrations of selumetinib, with IC₅₀ values calculated. J Western blot analysis of ALG3 expression levels following stimulation with different concentrations of selumetinib (0, 40, 80, 120 μM) for 48 h. K, L CCK8 assays to evaluate changes in cell viability after treatment with selumetinib in combination with ALG3 siRNA transfection in T24 (K) and UMUC3 (L) cells

Among the five compounds identified, we focused on selumetinib due to its high binding affinity for ALG3. The 3D structure of selumetinib was shown (Fig. 12G). To evaluate its therapeutic potential, bladder cancer cells were treated with varying concentrations of selumetinib, and changes in cell viability were measured. The sensitivity of the two cell lines to selumetinib differed, with IC₅₀ values of 85.54 ± 15.50 μM for T24 (Fig. 12H) and 92.0 ± 10.12 μM for UMUC3 cells (Fig. 12I). To further investigated the effects of selumetinib, bladder cancer cells were treated with concentrations of 0, 40, 80, 120 μM for 48 h, followed by protein extraction and Western blot analysis. The results revealed a concentration-dependent decrease in ALG3 expression (Fig. 12J), indicating the inhibitory effect of selumetinib on ALG3. In addition, ALG3 knockdown enhanced cell sensitivity to selumetinib. In T24 cells, the IC₅₀ value decreased from 108.0 μM to 70.29 μM after ALG3 siRNA transfection (Fig. 12K), while in UMUC3 cells, it decreased from 114.3 μM to 78.26 μM (Fig. 12L). These findings provided valuable insights for optimizing therapy selection and advancing precision medicine for bladder cancer patients, as well as facilitating personalized diagnosis and treatment in clinical practice.

Discussion

Currently, bioinformatics analysis based on large data sets represents a cutting-edge approach for elucidating the mechanism underlying disease occurrence and progression, as well as aiding in clinical diagnosis, treatment selection, and outcome evaluation. As one of the most significant diseases impacting global human health, understanding the pathogenesis of cancer and exploring effective treatments are key breakthroughs necessary to improve the current landscape of cancer management. In recent years, machine learning algorithm have emerged as a powerful tool in bioinformatics analysis tool. These algorithms enable real-time processing of clinical data and enhance analytical accuracy by handling massive data sets. With their assistance, several clinical challenges are expected to be addressed, such as the development of diagnostic models to aid in clinical detection or prognostic models to evaluate patient survival rates following treatment. For example, Shen H et al. utilized machine learning techniques to identify optimal diagnostic markers using genome-wide DNA methylation and RNA-seq data, constructing colorectal cancer diagnostic models that support the development of non-invasive, blood-based diagnostic tools [23]. Similarly, Li et al. employed large-scale data sets and machine learning algorithms to establish a three-gene fibroblast-related genes index, which accurately predicted clinical outcomes and immune therapeutic responses in BLCA patients after surgery [24]. Zhang J et al. developed an optimal prognostic model comprising 11 genes (NOP16, YIPF1, HMMR, NDC80, DYNLL1, CDC34, NLN, KHDRBS3, MED8, SLC35G2, RAB3B) based on microvascular invasion-related signature genes created by integrating scRNA-seq analysis with 101 machine learning algorithms, offering novel therapeutic insights for liver cancer [25]. In the current study, we screened 51 differentially expressed GTs and established a six-GTs diagnostic model. The diagnostic value of this model was validated using ROC curves. We observed that ALG3, as a single biomarker, exhibited superior performance with an AUC of 0.939, sensitivity of 84.1% and specificity of 88.9%, outperforming the other five GTs. Furthermore, when incorporated into a combined model, the diagnostic performance improved significantly, with an AUC of 0.966, sensitivity of 88.5% and specificity of 92.6%. These results were further confirmed using the GSE188715 data set, demonstrating the clinical applicability of the model. The GTs-based diagnostic model provided a novel option and theoretical foundation for the clinical detection of bladder cancer.

As critical members of the mannosyltransferase family, the elevated expression of ALG3 and POMT2 in most cancer types suggested their potential roles as tumor-promoting factors. This observation prompted us to evaluate their utility as glycobiomarkers for clinical diagnosis and outcome assessment. Our analysis revealed that ALG3 outperformed POMT2 as a diagnostic biomarker for bladder cancer, based on higher AUC values, sensitivity, and specificity. These findings were consistent across the TCGA–BLCA and GSE13507 data sets, leading us to select ALG3 for further investigation. ALG3 is an α1,3-mannosyltransferase responsible for synthesizing high-mannose N-glycans and is localized in the endoplasmic reticulum and Golgi apparatus. Dysregulated expression of ALG3 has been implicated in cancer progression. For instance, Wu et al. reported that ALG3 inhibited the infiltration of CD8 + T cells by suppressing chemokine secretion, and its inhibition enhanced the responsiveness of breast cancer cells to 5-fluorouracil treatment [18]. Liu et al. uncovered a previously unrecognized role of ALG3 in regulating tumor immunogenicity, proposing a potential therapeutic strategy to enhance cancer immunotherapy [26]. In addition, Luo B et al. identified elevated ALG3 as a potential biomarker for poor prognosis in triple-negative breast cancer, suggesting that it might reduce immunotherapy efficacy by modulating the tumor microenvironment and glycosylation of PD-L1 [27]. These studies underscored the multifaceted roles of ALG3 in cancer malignant transformation, therapy efficacy evaluation, and prognosis assessment. In this study, we further explored the impact of ALG3 on the malignant behaviors of bladder cancer cells. We found that elevated ALG3 expression facilitated the proliferation, migration and invasion capabilities of these cells. A key discovery in this study was the identification of CD44 as a target of ALG3. CD44 is a well-known glycoprotein and a potent therapeutic target in multiple cancer types [28,29,30]. For example, Zhang J et al. demonstrated that GALNT1 overexpression promoted the Wnt/β-catenin signaling pathway via abnormal O-glycosylation of CD44, enhancing the malignant phenotype of gastric cancer [31]. Similarly, Cheng Q et al. revealed that N-glycosylation of CD44 at Asn 57, Asn100 and Asn 110 promoted metastasis and invasion in hepatocellular cancer [32]. In this study, we found that ALG3 overexpression increased the N-glycosylation level of CD44, suggesting that enhanced malignant transformation of bladder cancer cells was driven by abnormal α−1,3 glycosylation of CD44 catalyzed by ALG3. We also identified key signaling pathways involved in ALG3-mediated oncogenic signaling, including the TNF signaling pathway, MAPK signaling pathway, IL-17 signaling pathway and NF-kappa B signaling pathway. Among these, the TNF signaling pathway was of particular interest, due to its involvement in various cancers, such as glioblastoma [33], lung cancer [34], and leukemia [35], etc. Our findings indicated that ALG3 promoted the malignant transformation of bladder cancer cells through the TNF signaling pathway. This was validated by Western blot analysis, which showed changes in the expression levels of TNF-α, TNFR1, TRADD, MMP3 and MMP9. In summary, this study unveiled the tumor-promoting role of ALG3 in bladder cancer, mediated through CD44 N-glycosylation and activation of the TNF signaling pathway. These findings provided novel insights into the molecular mechanism underlying bladder cancer progression and highlighted ALG3 as a potential therapeutic target.

Given its tumorigenic role, targeted drug screening for ALG3 was conducted, identifying four potential small molecule compounds, including the FDA-approved drugs trametinib, selumetinib, olaparib and SN-38. Among these, selumetinib, an FDA-approved MEK1/2 inhibitor, is currently used to treat several malignancies. It has been shown to restore the sensitivity of chemotherapy-resistant CRC cells to 5-FU by inhibiting ITGA2 expression [36]. In addition, improvements in progression-free survival have been observed with the combination of selumetinib and paclitaxel in metastatic uveal melanoma [37]. Furthermore, the combination of selumetinib and the cAMP elevator GPR68 has been proposed as a potential treatment for cutaneous neurofibromas [38]. Gross AM et al. evaluated the efficacy of selumetinib in adult patients with neurofibromatosis type 1 and inoperable plexiform neurofibromas, demonstrating sustained reductions in tumor volume, improvements in pain, and a manageable AE profile [39]. Through molecular docking, the binding interaction between selumetinib and ALG3 was elucidated. Subsequent experiments revealed that selumetinib significantly inhibited cell viability in a dose-dependent manner. Moreover, selumetinib treatment led to a marked suppression of ALG3 expression, and knockdown of ALG3 increased the sensitivity of bladder cancer cells to selumetinib, as evidenced by a decreased IC₅₀ value. These findings collectively suggested that ALG3 inhibition represented a promising therapeutic strategy for bladder cancer.

In summary, the diagnostic model established in this study demonstrated significant value in aiding the detection of bladder cancer. In addition, ALG3, targeted by miR-142-5p, was found to play a critical role in bladder cancer progression via modulating CD44 N-glycosylation and activating the TNF signaling pathway. These findings highlighted the potential of ALG3 as a serum glycobiomarker for clinical detection, an independent biomarker for prognosis evaluation, and a novel therapeutic target for intervention strategies.