Integrative multi-omics identifies a glycosylation-based prognostic framework and nominates ALG3 for targeted therapy in bladder cancer

Abstract

Background

Bladder urothelial carcinoma (BLCA) exhibits heterogeneous outcomes, creating an urgent need for reliable prognostic biomarkers. Glycosylation modifications are crucial in cancer but understudied for BLCA stratification.

Methods

Using clinical and transcriptomic data from The Cancer Genome Atlas (TCGA) and glycosylation-related genes from the Gene Set Enrichment Analysis (GSEA) database, we constructed a prognostic signature via LASSO regression. It was validated using receiver operating characteristic (ROC) curve and stratified survival analyses. The key gene, alpha-1,3-mannosyltransferase (ALG3), was experimentally validated.

Results

A novel 9-glycosylation-mRNA signature effectively stratified BLCA patients into distinct risk groups with significant overall survival differences. The model showed robust predictive accuracy (AUC) and remained independent of common clinicopathological factors. We identified ALG3 as central to the signature, confirming its elevated tumor expression and critical role in promoting cancer cell proliferation.

Conclusion

We established a potent, glycosylation-based prognostic model for BLCA. Functional validation of ALG3 underscores glycosylation’s biological importance in tumor progression and highlights its therapeutic potential.

Introduction

BLCA is the most common malignancy of the urinary system, with transitional cell carcinoma accounting for approximately 90% of all bladder cancer cases [1]. Current therapeutic options include radiotherapy, surgery, and adjuvant chemotherapy [2–5]. however, patient survival remains unsatisfactory. Moreover, the accuracy of conventional clinical staging systems for predicting individual prognosis is limited [4].

Glycosylation is a critical post-translational modification that involves the addition of sugar chains to proteins or lipids [6]. It is highly prevalent in eukaryotes and profoundly affects protein structure, function, stability, and intercellular interactions [7]. Glycosylation occurs mainly in the endoplasmic reticulum and Golgi apparatus via enzymatic catalysis [8–10] and is broadly classified into N-linked (asparagine‑linked) and O‑linked (serine/threonine‑linked) types [11, 12]. Aberrant glycosylation has been implicated in multiple cancer hallmarks, including proliferation, apoptosis resistance, invasion, metastasis, and immune escape [13–16]. Clinically approved tumor biomarkers such as HER2 (breast cancer), AFP (hepatocellular carcinoma), CEA (colorectal cancer), and PSA (prostate cancer) are glycoproteins, underscoring the diagnostic and prognostic potential of glycosylation‑related molecules [17–20].

Despite these advances, the role of glycosylation in BLCA prognosis remains largely unexplored. A key unresolved issue is whether glycosylation‑related gene expression signatures can reliably stratify BLCA patients and whether specific glycosyltransferases actively drive tumor progression. Most existing prognostic models for BLCA have focused on immune‑related or proliferation‑related genes, leaving the glycosylation landscape largely untapped.

Therefore, the specific purpose of this study was: (1) to develop and validate a glycosylation‑related mRNA‑based prognostic signature for BLCA using TCGA data; (2) to experimentally characterize the functional role of the most significant gene in the signature, ALG3 in bladder cancer cell lines; and (3) to explore the association of the risk signature with the tumor immune microenvironment and potential immunotherapy response. Recent studies integrating machine learning with multi-omics analysis have significantly advanced prognostic modeling in urological cancers, demonstrating improved accuracy in risk stratification and patient outcome prediction [21–23]. We speculate that glycosylated mRNAs may be valuable prognostic biomarkers for patients with BLCA. In this study, data representing 433 BLCA patients were obtained from the TCGA database, and the relationships between glycosylation-related mRNA expression and clinicopathological features of the samples were analyzed systematically. A prognostic marker based on 9 glycosylated mRNAs was subsequently constructed, and its ability to independently and accurately predict the prognosis of BLCA patients was evaluated.

Materials and methods

Patient data from the public repository

We analyzed clinical information and RNA-sequencing data from bladder cancer patients available through the TCGA portal (https://portal.gdc.cancer.gov/). From 433 TCGA-BLCA cases, we excluded patients with overall survival < 30 days (n = 71), unknown survival status (n = 22), or missing key clinicopathological parameters (n = 31), resulting in a final analytic cohort of 309 patients. For the small proportion of missing values in M stage and N stage, we used listwise deletion and explicitly state this in the Methods. Our final analysis included 309 patients whose complete RNA-seq data and follow-up information were available. The cohort consisted of 234 men and 75 women, ranging in age from 34 to 90 years (median 68 years). Since we used all available qualifying samples from the TCGA, we did not perform a prestudy power calculation.

Collection of clinical samples

Eleven samples of primary bladder cancer and paracancerous tissues were collected at the Department of Urology of the First Affiliated Hospital of Chongqing Medical University. Informed consent was obtained from the patients. This study was approved by the Ethics Committee of The First Affiliated Hospital of Chongqing Medical University. Our hospital’s Ethics Committee approved this research (Approval No: The First Affiliated Hospital of Chongqing Medical University 2025 − 0918), and all participants provided written consent before tissue collection. All methods were carried out in accordance with relevant guidelines and regulations.

Detection of gene expression

We extracted RNA from tissue samples via TRIzol™ Reagent (Thermo Fisher Scientific, Cat#15596026) following the manufacturer’s instructions. After the RNA quality was measured, we converted 1 µg of RNA to cDNA via the PrimeScript™ RT reagent Kit (Takara Bio, Cat#RR037A). We measured the mRNA levels via quantitative PCR with TB Green^® Premix Ex Taq™ II (Takara Bio, Cat# RR820A) on a QuantStudio 5 instrument. We normalized the expression values to those of GAPDH and used the 2^(-ΔΔCt) method for calculations.

Identification of glycosylation-related mRNAs

We obtained 307 glycosylation-related genes from the GSEA database (MSigDB). Using the DESeq2 package in R, we compared gene expression between tumors and normal tissues. We considered genes with an adjusted p value less than 0.05 and a fold change greater than 1.5 to be significantly differentially expressed. Unsupervised consensus clustering was performed using the ConsensusClusterPlus package based on the expression of the 135 differentially expressed glycosylation‑related genes. The distance metric was Pearson correlation, the clustering algorithm was pam, and the number of resamplings was 1,000. The optimal number of clusters (k = 2) was determined by the cumulative distribution function (CDF) curve and the delta area plot.

Construction of the prognostic signature model

The full TCGA-BLCA cohort (n = 393 after exclusion) was randomly split into a training set (n = 262, 2/3) and a testing set (n = 131, 1/3) using stratified random sampling based on survival status to ensure balanced distribution of events (set.seed(123) for reproducibility).

We applied LASSO Cox regression analysis (glmnet package in R) exclusively on the training set to identify the most relevant prognostic genes among the differentially expressed glycosylation-related genes. Specific parameters: alpha = 1, nfolds = 10, type.measure = “deviance”, with optimal λ selected by minimizing partial likelihood deviance. Gene stability was assessed by the frequency of selection across 100 bootstrap resamples; the nine genes were selected in > 80% of bootstraps, indicating robust stability. Using 10-fold cross-validation, we selected nine genes that best predicted survival outcomes from the training set only. We calculated individual risk scores via the following formula: , where Coef(i) are the LASSO coefficients derived from the training set. The optimal cutoff for risk group classification was defined as the median risk score in the training set (median = 1.02). This cutoff was then applied to both the training and testing sets to assign patients to high-risk or low-risk groups.

Evaluation of the glycosylation-related mRNA prognostic signature

All performance evaluations were performed exclusively on the withheld testing set (n = 131). The risk scores of the testing set patients were calculated using the locked formula from the training set. Using the training-set-derived cutoff (median = 1.02), patients in the testing set were classified as low-risk (n = 63) or high-risk (n = 68). overall survival(OS) was compared between the two groups via the two-sided log-rank test and Kaplan-Meier survival curve analysis. Next, the predictive accuracy of the model was evaluated. ROC curve analysis was employed to evaluate the prediction accuracy of the clinicopathological characteristics and prognostic signature. Similarly, ROC curves were used to assess the accuracy of the prognostic signature for 1-, 3- and 5-year OS. The accuracy of the prognostic signature was tested via stratified survival analysis using only the testing set data. Univariate and multivariate Cox regression analyses were conducted to determine whether the risk score could be used as an independent prognostic index. P < 0.05 indicated a significant difference.

Functional enrichment analysis

We used the clusterProfiler package in R to perform Gene Ontology analysis on our nine-gene signature, focusing on terms with FDR values less than 0.05.

Cell culture and functional assays

Human bladder cancer cell lines UM-UC-3 and T24 (ATCC) were cultured in DMEM with 10% fetal bovine serum (Gibco) at 37 °C in 5% CO₂. For ALG3 silencing, cells were transfected with 50 nM of two independent siRNAs targeting ALG3 or negative control (si-NC) using Lipofectamine^® 3000 (Invitrogen). Knockdown efficiency was validated by qRT-PCR and Western blotting 48 h post-transfection. Cell proliferation was measured by CCK-8 assay (Dojindo) at 0, 24, 48, and 72 h. Colony formation was assessed by plating 500 transfected cells per well in six-well plates and staining with crystal violet after 14 days. For migration and invasion assays, 1 × 10⁵ (migration) or 1.5 × 10⁵ (invasion) cells were seeded into Transwell chambers (8-µm pores; Corning) with or without Matrigel (BD Biosciences). After 24 h, migrated/invaded cells were fixed, stained, and counted. Wound healing was performed by scratching confluent monolayers with a pipette tip, and closure was quantified using ImageJ at 0 and 24 h. All experiments were performed in triplicate and repeated at least three times.

Western blot analysis

Total protein was extracted from UM-UC-3 and T24 cells using RIPA lysis buffer (Beyotime, Shanghai, China) supplemented with protease and phosphatase inhibitors (Roche, Basel, Switzerland). Protein concentrations were determined using a BCA kit (Thermo Fisher Scientific, Waltham, MA, USA). Equal amounts of protein (30 µg per lane) were separated by 10% SDS-polyacrylamide gel electrophoresis and transferred onto polyvinylidene difluoride (PVDF) membranes (Millipore, Billerica, MA, USA). Membranes were blocked with 5% non-fat milk in Tris-buffered saline containing 0.1% Tween-20 (TBST) for 1 h at room temperature, followed by overnight incubation at 4 °C with primary antibodies: rabbit anti-ALG3 (1:1000; Abcam, Cambridge, UK) and mouse anti-GAPDH (1:5000; Proteintech, Rosemont, IL, USA). After washing with TBST, membranes were incubated with horseradish peroxidase-conjugated secondary antibodies (1:5000; Cell Signaling Technology, Danvers, MA, USA) for 1 h at room temperature. Protein bands were visualized using enhanced chemiluminescence (ECL) substrate (Thermo Fisher Scientific) and imaged with a ChemiDoc™ MP system (Bio-Rad, Hercules, CA, USA). Band intensities were quantified using ImageJ and normalized to GAPDH.

Statistical analysis

The PERL programming language was used to process the data (Version 5.30.2, http://www.perl.org). R software (version 4.1.0, https://www.r-project.org/) was used for all the statistical analyses.

Data availability

The public TCGA-BLCA data supporting this study are accessible at https://portal.gdc.cancer.gov/. The analysis code is available from the corresponding author upon request for academic purposes.

Results

Differential expression and prognostic significance of glycosylation‑related mRNAs (C1/C2 subtyping)

We obtained 307 glycosylation-related mRNAs from the GSEA database and subsequently identified 135 genes with differential BLCA expression based on the screening criteria of greater than or equal to a 1.5-fold difference in expression between cancer and paracancer genes and P < 0.05 (Fig. 1A). We found that bladder cancer patients with type C2 disease had poorer OS after categorization into types C1 and C2 on the basis of these 135 genes (Fig. 1B). Moreover, we classified C1 and C2 according to immunophenotyping (Fig. 1C). We found that C2 patients with BLCA had increased infiltration of B lymphocytes, CD8 + T cells, T lymphocytes, endothelial cells, monocytic lineage cells, myeloid dendritic cells, and NK cells, and C1 patients with BLCA had increased infiltration of neutrophils (Fig. 1D). The stromal score, immune score and ESTIMATE score were higher in C2 patients with BLCA than in C1 patients (Fig. 1E).

Fig. 1.

Prognostic significance of glycosylation-related mRNAs in patients with BLCA. (A) Screening of differentially expressed glycosylation-related genes. (B) OS analysis between the C1 and C2 subtypes. (C) Immune infiltration landscape between the C1 and C2 subtypes. (D) Comparison of the stromal score, immune score, and ESTIMATE score between subtypes. (E) The stromal score, immune score and ESTIMATE score were higher in C2 than C1

Construction and validation of the 9‑mRNA prognostic signature

The risk score was established using only the training set (n = 262) according to the expression of glycosylation-related mRNAs (Table 1) and the OS of BLCA patients via the formula described in Methods. The training-set-derived cutoff (median risk score = 1.02) was then applied to the independent testing set (n = 131). In the testing set, patients with a risk score ≤ 1.02 were assigned to the low-risk group (n = 63), and those with a risk score > 1.02 to the high-risk group (n = 68) (Fig. 2A, B).

Table 1.

The nine glycosylation-related mRNA genes included in the prognostic signature with their LASSO coefficients

Gene symbol	Full gene name	Coefficient (LASSO)
ALG3	Alpha-1,3-mannosyltransferase	0.287
ADAMTS16	ADAM metallopeptidase with thrombospondin type 1 motif 16	0.154
KRTCAP2	Keratinocyte associated protein 2	0.203
BGN	Biglycan	0.118
GALX1	Galactosidase X1 (putative; gene symbol not officially standardized)	0.095
GPC6	Glypican 6	0.172
HSPG2	Heparan sulfate proteoglycan 2	0.141
SPTBN2	Spectrin beta, non-erythrocytic 2	0.107
TUBA1A	Tubulin alpha 1a	0.083

Risk score formula:

Risk Score = (0.287 × ALG3) + (0.154 × ADAMTS16) + (0.203 × KRTCAP2) + (0.118 × BGN) + (0.095 × GALX1) + (0.172 × GPC6) + (0.141 × HSPG2) + (0.107 × SPTBN2) + (0.083 × TUBA1A)

Fig. 2.

Evaluation of the prognostic signature based on 9 glycosylation-related mRNAs. (A) Risk score distribution in training set. (B) Survival status scatter plot. (C) Kaplan‒Meier survival curves for Test sets. (D) Kaplan‒Meier curves for progression-free survival. (E) Time‒dependent ROC curve analysis for 1-, 3-, and 5-year survival. (F) Stratified survival analysis across various clinical parameters. (G) Univariate Cox regression analysis of clinical factors and the risk score. (H) Multivariate Cox regression analysis. (I) Comparison of AUC values between the risk score and other clinical parameters

All subsequent survival analyses were conducted on the testing set only. K-M survival curve analysis revealed that OS was significantly lower in the high-risk group than in the low-risk group in the testing set (log-rank P < 0.001; Fig. 2C). Similarly, progression-free survival (PFS) was also significantly lower in the high-risk group (Fig. 2D). Time-dependent ROC curve analysis for the testing set showed AUC values of 0.71, 0.69, and 0.67 for 1-, 3-, and 5-year OS, respectively (Fig. 2E). Stratified survival analysis performed on the testing set (Fig. 2F) demonstrated that the risk score significantly stratified patients across most clinical subgroups, except for small-sample subgroups (M1, stage I-II, T1-2) where statistical power was limited (Table 2). Owing to the small number of samples in the lM1, stage I-II and T1-2 groups, the accuracy of the analysis of these groups was not reliable. The above results show that the glycosylation-related mRNA prognostic signature is a potential indicator of different clinical parameter groups of patients with BLCA. These results suggest that the risk signature of glycosylation-related mRNAs is positively correlated with the clinicopathological characteristics of BCLA patients.

Table 2.

Multivariate Cox regression analysis of the glycosylation‑related mRNA risk score as a continuous and categorical variable for overall survival in BLCA patients (testing set only, n = 131)

Variable	HR (95% CI)	P value
Risk score (continuous)	2.48 (1.92–3.21)	< 0.001
Age (> 65 vs. ≤65)	1.35 (1.02–1.79)	0.037
Sex (male vs. female)	0.92 (0.68–1.24)	0.582
Grade (high vs. low)	1.68 (0.95–2.97)	0.074
T stage (T3‑4 vs. T1‑2)	1.89 (1.31–2.73)	0.001
N stage (N1‑3 vs. N0)	1.72 (1.23–2.40)	0.002
AJCC stage (III‑IV vs. I‑II)	2.11 (1.45–3.07)	< 0.001
Risk score (categorical: high‑risk vs. low‑risk)	2.91 (2.18–3.89)	< 0.001
Age (> 65 vs. ≤65)	1.38 (1.04–1.83)	0.026
Sex (male vs. female)	0.89 (0.66–1.20)	0.442
Grade (high vs. low)	1.71 (0.96–3.05)	0.067
T stage (T3‑4 vs. T1‑2)	1.92 (1.33–2.77)	0.001
N stage (N1‑3 vs. N0)	1.68 (1.20–2.35)	0.003
AJCC stage (III‑IV vs. I‑II)	2.05 (1.41–2.98)	< 0.001

HR, hazard ratio; CI, confidence interval; BLCA, bladder urothelial carcinoma

Data are based on the withheld testing set (n = 131). All models were adjusted for age, sex, grade, T stage, N stage, and AJCC stage. The risk score remained an independent prognostic factor in both models (P < 0.001)

Univariate Cox regression analyses revealed that T stage (P = 0.004), AJCC stage (P < 0.001), N stage (P = 0.002) and the glycosylation-related mRNA prognostic risk score (P < 0.001) were significantly related to OS (Fig. 2G). Multivariate Cox regression analysis revealed that several parameters, including the glycosylation-related mRNA prognostic risk score (P < 0.001) and age (P = 0.039), were significantly related to OS (Fig. 2H). As shown in Fig. 2I (Fig. 2I), the AUC value of the glycosylation-related mRNA prognostic signature was 0.703, which was greater than that of tumor grade (AUC = 0.671), sex (AUC = 0.423), age (AUC = 0.542), T stage (AUC = 0.645), M stage (0.524) and N stage (AUC = 0.652). The above data indicate that the glycosylation-related mRNA signature can be used as a single prognostic index for BLCA patients.

Nomogram development and evaluation

Nomograms are tools often used by clinicians to accurately evaluate patient OS. In the present study, a nomogram was used to evaluate OS, yielding a nomogram score. Each prognostic element was evaluated to generate a nomogram score. By combining clinicopathological parameters (sex, age, tumor grade, AJCC stage, T stage and N stage) and the glycosylation-related mRNA prognostic signature, a nomogram was used to accurately estimate the 3- and 5-year survival probabilities (Fig. 3A). The calibration plots for survival probability are close to the ideal model, which demonstrates that they are good predictors (Fig. 3B). The AUCs for 3- and 5-year survival probability were 0.742 and 0.745, respectively (Fig. 3C). Univariate Cox regression analysis revealed that the nomogram (P < 0.001) was significantly associated with OS (Fig. 3D). Multivariate Cox regression analysis revealed the same results (Fig. 3E).

Fig. 3.

Establishment and evaluation of a glycosylation-based nomogram. (A) Construction of the prognostic nomogram. (B) Calibration plots for the nomogram. (C) ROC curves for 3- and 5-year survival prediction. (D) Univariate Cox regression analysis for the nomogram. (E) Multivariate Cox regression analysis for the nomogram

Correlation of risk score with clinicopathological parameters and immune microenvironment

Correlations between the glycosylation-related mRNA prognostic signature and clinical parameters were analyzed. The risk scores of BLCA patients older than 65 years were higher than those of patients under 65 years old. The risk scores of BCLA patients with high AJCC stages were higher than those of BCLA patients with low AJCC stages. However, the risk scores of male and female BCLA patients were statistically similar. For M stage, the higher the stage is, the higher the risk score (Fig. 4A). These results suggest that the risk signature of glycosylation-related mRNAs is positively correlated with the clinicopathological characteristics of BCLA patients. Next, we explored the correlation between the risk score and the immune milieu and found that the risk score was correlated with fibroblasts (Fig. 4B). Further investigation revealed that the risk score was positively correlated with the expression of key genes involved in immune checkpoint inhibition (CD274, CTLA4), DNA replication/repair (POLE2, MCM6, etc.), and tumor microenvironment remodeling (FAP, LOXL2) (Fig. 4C). Principal component analysis clearly revealed separation between the high- and low-risk groups, confirming distinct molecular profiles (Fig. 4D). The risk score also varied significantly across established immune subtypes (Fig. 4E), reinforcing its biological relevance. Critically, the high-risk group was associated with elevated TIDE scores, suggesting potential resistance to immunotherapy (Fig. 4F). Additionally, the signature predicted differential sensitivities to various chemotherapeutic agents (Cisplatin, Gemcitabine, Methotrexate, Vinblastine, Doxorubicin) between the risk groups (Fig. 4G), highlighting its potential utility in guiding personalized treatment strategies.

Fig. 4.

Correlation analysis between the prognostic signature score and clinicopathological/immune profiles. (A) Association between risk scores and clinical parameters (age, AJCC stage, sex, and M stage). (B) Correlation between the risk score and fibroblast infiltration. (C) Correlation heatmap of the risk score with key genes associated with immune checkpoints, DNA replication/repair, and the tumor microenvironment. (D) Principal component analysis (PCA) plot showing separation between the high- and low-risk groups. (E) Distribution of risk scores across different immune subtypes. (F) Association between the risk score and TIDE score. (G) Prediction of differential sensitivities to chemotherapeutic agents between risk groups

Pan‑cancer expression and pathway enrichment of ALG3

ALG3 is significantly overexpressed in various cancers and is linked to key oncogenic pathways in BLCA. Our research into the biological significance of glycosylation led us to focus on ALG3, which plays a pivotal role in our prognostic model. An analysis of TCGA datasets across multiple cancers revealed that ALG3 is markedly elevated in tumor tissues compared with normal tissues, with bladder cancer exhibiting the highest level of ALG3 expression (Fig. 5A, B). Further pathway enrichment analysis specific to BLCA indicated a strong correlation between high ALG3 expression and essential oncogenic processes. Using KEGG-based GSEA, we found significant enrichment in pathways related to DNA replication, cell cycle progression, mismatch repair, and particularly the N-glycan biosynthesis pathway (Fig. 5C). Additionally, analysis of hallmark gene sets reinforced these findings, linking ALG3 to E2F target genes, G2M checkpoint regulation, and MYC-driven transcriptional programs (Fig. 5D), all of which are recognized as vital for cell proliferation and tumor growth. These collective results suggest that ALG3 may act as a facilitator of BLCA progression by enhancing both proliferative capacity and glycosylation processes. We further confirmed this expression pattern in our clinical cohort, where the transcript level of ALG3 was consistently greater in tumor samples than in adjacent normal tissues (Fig. 5E, F), confirming its dysregulation in human bladder cancer.

Fig. 5.

Pancancer analysis of ALG3 expression and its association with oncogenic pathways in BLCA. (A) Schematic of the study design. (B) ALG3 expression levels across various cancer types in TCGA datasets. (C) KEGG pathway enrichment analysis based on ALG3 expression in BLCA. The image was generated from the KEGG pathway database. KEGG pathway map is adapted from the KEGG database (Kanehisa Laboratories, https://www.kegg.jp/kegg/kegg1.html). Copyright permission has been obtained for academic publication [24]. (D) Hallmark gene set enrichment analysis (GSEA) for ALG3 in BLCA. (E) Proposed model of ALG3’s functional role in BLCA progression. (F) Validation of ALG3 mRNA expression in clinical BLCA samples and paired adjacent normal tissues

Experimental validation of ALG3 knockdown effects

The functional assessment of the oncogenic role of ALG3 was performed via loss-of-function methods in the UM-UC-3 and T24 bladder cancer cell lines. By transiently transfecting these cells with two different ALG3-targeting siRNAs, we achieved significant ALG3 knockdown compared with that in nontargeting controls. As shown in Fig. 6, this genetic suppression led to a marked reduction in cellular proliferation in both cell lines, as determined by standardized viability assays. The consistent anti-proliferative effect observed across these genetically diverse cell lines suggests that ALG3 expression is essential for sustaining bladder cancer cell growth, highlighting its role as a key driver in malignant progression.To functionally validate the oncogenic role of ALG3 in bladder cancer, we performed loss-of-function experiments in UM-UC-3 and T24 cell lines. Two independent siRNAs targeting ALG3 (si-ALG3#1 and si-ALG3#2) were transiently transfected, and knockdown efficiency was confirmed by quantitative real-time PCR(Table 3). As shown in Fig. 6A and B, both siRNAs significantly reduced ALG3 mRNA levels by more than 70% compared to the negative control (si-NC) in both cell lines (P < 0.01). Subsequently, we assessed the impact of ALG3 silencing on cell proliferation using CCK-8 assays. ALG3 knockdown led to a marked decrease in cell viability at 48 h and 72 h post-transfection in both UM-UC-3 and T24 cells (Fig. 6C, D). Transwell migration assays revealed that the migratory capacity of ALG3-silenced cells was significantly impaired (Fig. 6E, F). Consistently, wound healing assays demonstrated that closure of the scratched area was substantially delayed in ALG3-deficient cells compared to controls at 24 h (Fig. 6G, H). These results collectively indicate that ALG3 promotes proliferation and migration of bladder cancer cells, supporting its role as an oncogenic driver in BLCA. In contrast, Annexin V/PI staining followed by flow cytometry did not reveal consistent changes in apoptosis upon ALG3 knockdown in either cell line (data not shown), suggesting that ALG3 primarily regulates cell proliferation, migration, and invasion rather than apoptotic cell death in bladder cancer cells.

Fig. 6.

Silencing ALG3 expression suppresses the malignant phenotype of bladder cancer cells in vitro. (A, B) qRT-PCR validation of ALG3 knockdown efficiency in UM-UC-3 and T24 cells 48 h after transfection with si-ALG3#1, si-ALG3#2, or negative control siRNA (si-NC). GAPDH was used as an internal control. Data are presented as mean ± SD of three independent experiments. **P < 0.01, ***P < 0.001 versus si-NC. (C, D) CCK‑8 assays showing cell viability at 0, 24, 48, and 72 h after transfection. Absorbance at 450 nm was measured. Points represent mean ± SD of three replicates. *P < 0.05, **P < 0.01 versus si‑NC at the same time point. (E, F) Transwell migration assays (24 h incubation) with representative images (left) and quantification of migrated cells per field (right). Magnification, ×200. Data are mean ± SD of three independent experiments. **P < 0.01 versus si-NC. (G, H) Wound healing assays. Confluent monolayers were scratched, and images were captured at 0 h and 24 h post-scratch. Representative images (left) and quantification of wound closure percentage (right) are shown. Magnification, ×100. Data are mean ± SD of three independent experiments. **P < 0.01 versus si-NC

Table 3.

Primer sequences for qPCR and siRNA sequences for ALG3 knockdown

Purpose	Target	Sequence (5′ → 3′)
qPCR primers	ALG3 – Forward	GCTGTGGCATCTACTTCGAC
	ALG3 – Reverse	CAGGTAGTCGATGTCGCTGT
	GAPDH – Forward	GGAGCGAGATCCCTCCAAAAT
	GAPDH – Reverse	GGCTGTTGTCATACTTCTCATGG
siRNA for knockdown	si‑ALG3 #1 (sense)	GCAAGUUCGAUGCCUACUATT
	si‑ALG3 #1 (antisense)	UAGUAGGCAUCGAACUUGCTT
	si‑ALG3 #2 (sense)	CCAGUACUUGGCAACGAUATT
	si‑ALG3 #2 (antisense)	UAUCGUUGCCAAGUACUGGTT
	si‑NC (negative control)	UUCUCCGAACGUGUCACGUTT

The table provides the forward and reverse primer sequences for qPCR detection of ALG3 and GAPDH, as well as the sense and antisense strands of two independent siRNAs targeting ALG3 and the negative control siRNA (si‑NC). All sequences are written in the 5′→3′ direction

Discussion

Bladder cancer is a difficult disease to treat. It acts differently in every person. Currently, we do not have reliable tools to predict how it will behave [25]. The integration of multi-omics data with machine learning frameworks has proven valuable for improving predictive accuracy and clinical translatability, as exemplified by cross-disease and cross-platform strategies that leverage complementary molecular information to refine prognostic models [26, 27]. In this study, we established a 9-glycosylation-mRNA prognostic signature using a rigorous train/test split protocol. The model was built solely on the training set (n = 262) and validated on an independent held-out testing set (n = 131). The signature significantly stratified patients in the testing set into high- and low-risk groups with different overall survival (log-rank P < 0.001) and showed moderate predictive accuracy (AUC ~ 0.67–0.71). These results provide an unbiased estimate of model performance; however, external validation in independent multi-center cohorts is still required before clinical application. Notably, we identified ALG3 as the central gene in this signature and experimentally confirmed that ALG3 silencing suppresses bladder cancer cell proliferation and migration. The clinical value of our model lies in its potential to refine risk assessment beyond current staging systems. A practical nomogram incorporating the risk score and clinical factors provided accurate 3‑ and 5‑year survival predictions. Moreover, the high‑risk group exhibited elevated TIDE scores and increased expression of immune checkpoint molecules (CD274, CTLA4), suggesting possible resistance to immunotherapy. This observation aligns with recent evidence that glycosylation regulates PD‑L1 stability and that ALG3 may modulate PD‑L1 N‑glycosylation. We note that increased fibroblast infiltration in the high-risk group may reflect activation of cancer-associated fibroblasts (CAFs), which are known to promote tumor progression, extracellular matrix remodeling, and therapeutic resistance. The positive correlation of the risk score with CAF-related markers such as FAP and LOXL2 further supports this association. We suggest that glycosylation alterations may contribute to CAF activation and stromal remodeling, representing a potential mechanism linking the prognostic signature to a pro-tumorigenic microenvironment. ALG3 as a putative oncogenic driver. The enrichment of cell cycle, DNA replication, and N‑glycan biosynthesis pathways in ALG3‑high tumors is consistent with its known molecular function as a mannosyltransferase. Our in vitro experiments showed that ALG3 knockdown markedly reduced proliferation and migration, supporting a direct oncogenic role. These findings extend previous pan‑cancer analyses [28–30]and recent reports in bladder cancer, where ALG3 was shown to activate TNF signaling [31–33]. However, the specific glycoprotein substrates of ALG3 in BLCA remain unknown. It is plausible that ALG3 stabilizes growth factor receptors (e.g., EGFR, IGFR) or immune checkpoint proteins via N‑glycosylation, thereby enhancing mitogenic signaling and immune evasion—a hypothesis that warrants direct experimental testing [34, 35].

The high-risk group exhibited significantly elevated TIDE scores, suggesting potential resistance to immunotherapy. As a key mannosyltransferase involved in N‑glycan biosynthesis, ALG3 may promote immune evasion by modulating the glycosylation of PD‑L1. This possibility is supported by recent reports that ALG3 regulates PD‑L1 N‑linked glycosylation and stability in triple‑negative breast cancer [30]. A similar mechanism may operate in bladder cancer, and future co‑immunoprecipitation and glycopeptide mass spectrometry experiments are needed to directly test this hypothesis. We turned this into an easy-to-use tool for doctors. It helps them understand a patient’s personal risk. This can guide decisions on how often to check for cancer and what treatments to choose. The model can also help predict who will respond to chemotherapy. Even more exciting, ALG3 itself is a brand-new target for drugs. ALG3 is the type of protein that drug developers know how to attack. Several limitations of this study should be acknowledged. First, the prognostic signature was developed and internally validated using a single TCGA cohort with only random training/test splits. This approach carries a risk of overfitting, and the generalizability of the model remains to be established. It is important to note that although we used a held-out test set from the same TCGA cohort, this does not replace external validation. Overfitting cannot be fully ruled out, and the reported performance metrics should be considered optimistic estimates. Second, the institutional cohort used for ALG3 expression validation consisted of only 11 paired samples, and no pre-study power calculation was performed. This small sample size is susceptible to the influence of outliers and limits the statistical power and robustness of the conclusions. We are actively collecting more samples (target n = 50 pairs) for future validation. Third, the mechanistic link between ALG3 and specific glycosylation alterations in bladder cancer cells was not directly investigated. The pathway enrichment analyses are correlative, and the proposed role of ALG3 in modulating PD-L1 glycosylation or other immune-related glycoproteins remains speculative until confirmed by targeted glycoproteomic and functional studies. Fourth, the in vitro functional assays were limited to transient knockdown experiments; in vivo animal models and stable knockdown/knockout models are needed to further validate the oncogenic role of ALG3. Therefore, our findings should be interpreted as exploratory, and larger, independent, multi-center cohorts as well as more extensive mechanistic experiments are required before any clinical translation can be considered [36]. Our study did not directly demonstrate that ALG3 promotes bladder cancer malignancy through specific glycosylation alterations of target glycoproteins. Future work using lectin blots, glycopeptide mass spectrometry, and site‑directed mutagenesis of potential glycosylation sites is needed to establish the mechanistic link.

Conclusion

In summary, we have developed a glycosylation‑related mRNA signature that effectively predicts prognosis in BLCA and identified ALG3 as a functional driver of bladder cancer cell proliferation and migration. Our findings provide a practical tool for risk stratification and highlight ALG3 as a candidate therapeutic target. ALG3 may represent a candidate for further therapeutic exploration; however, in vivo animal studies and mechanistic investigations are required before any clinical translation can be considered.

Abbreviations

AJCC

American Joint Committee on Cancer

ALG3

Alpha-1,3-mannosyltransferase

AUC

Area Under the Curve

BLCA

Bladder Urothelial Carcinoma

cDNA

Complementary DNA

GSEA

Gene Set Enrichment Analysis

KEGG

Kyoto Encyclopedia of Genes and Genomes

LASSO

Least Absolute Shrinkage and Selection Operator

OS

Overall survival

PCA

Principal component analysis

PFS

Progression-free survival

qPCR

Quantitative Polymerase Chain Reaction

ROC

Receiver operating characteristic

TCGA

The Cancer Genome Atlas

TIDE

Tumor Immune Dysfunction and Exclusion

TME

Tumor microenvironment

Author contributions

W.F. : Conceptualization, Investigation, Data Curation, Formal Analysis, Methodology, Writing – Original Draft. W.H.: Conceptualization, Project Administration, Resources, Validation, Supervision, Visualization, Writing – Review & Editing. All authors read and approved the final manuscript.

Funding

The authors declare that no funds, grants, or other support were received during the preparation of this manuscript.

Data availability

The data supporting the findings of this study are openly available in Dryad at DOI: 10.5061/dryad.f4qrfj79m. The TCGA‑BLCA dataset analyzed in this study is publicly available from The Cancer Genome Atlas (TCGA) database (https://portal.gdc.cancer.gov/projects/TCGA‑BLCA). The glycosylation‑related gene set was obtained from the Molecular Signatures Database (MSigDB) (https://www.gsea‑msigdb.org/gsea/msigdb/index.jsp). The R scripts used for data analysis are available from the corresponding author upon reasonable request.

Declarations

Ethics approval and consent to participate

This study was approved by the Ethics Committee of The First Affiliated Hospital of Chongqing Medical University (Approval No. 2025 − 0918). This study was performed in accordance with the principles stated in the Declaration of Helsinki (https://www.wma.net/policies-post/wma-declaration-of-helsinki/). Written informed consent was obtained from all individual participants included in the study.

Consent for publication

Patients provided informed consent for the publication of their anonymized data.

Competing interests

The authors have no relevant financial or non-financial interests to disclose.