Multi-omics integration identifies HSPA4 as a therapeutic target and reveals DON-induced endoplasmic reticulum stress in bladder cancer

Introduction

Bladder cancer is a highly prevalent urological malignancy, distinguished by high recurrence rates and a propensity for progression to muscle-invasive disease (1,2). Although advances have been made in standard therapies, including surgical resection, chemotherapy, and immunotherapy, their overall efficacy is constrained by tumor heterogeneity, drug resistance, and a paucity of precise therapeutic targets (3,4). Consequently, a primary research focus in this field is the systematic identification of key driver genes integral to bladder cancer pathogenesis and, based on this foundation, the exploration of potential targeted therapeutics (5).

The rapid advancement of bioinformatics technologies has established high-throughput screening of public gene expression databases as a crucial approach for identifying disease-associated genes (6). Methods such as differential expression analysis and weighted gene co-expression network analysis (WGCNA) can systematically identify phenotype-related gene modules at the transcriptomic level (7). Furthermore, Mendelian randomization, which utilizes genetic variants as instrumental variables, can mitigate confounding biases inherent in traditional observational studies, thereby offering a more robust analytical framework for inferring potential causal relationships between genes and diseases (8,9). However, individual analytical methods are often limited. Consequently, a key challenge in the field is the effective integration of multidimensional data and diverse bioinformatics strategies to systematically screen and validate therapeutic targets with clinical translational potential (10).

To address this, the present study was designed to systematically screen for potential key genes in bladder cancer by integrating differential expression analysis, WGCNA, and Mendelian randomization. Candidate genes are validated using independent datasets. Subsequently, compounds targeting these key genes will be predicted through drug-target database screening. The binding affinity and therapeutic potential of prioritized candidates are then evaluated preliminarily through molecular docking, molecular dynamics (MD) simulations, and in vitro assays. Overall, this work establishes a systematic research workflow from computational screening to experimental validation, thereby providing novel target insights and a drug selection rationale for the targeted therapy of bladder cancer (Figure 1). We present this article in accordance with the MDAR reporting checklist (available at https://tau.amegroups.com/article/view/10.21037/tau-2026-0330/rc).

Figure 1 Research flowchart. This study employed a multi-omics strategy to systematically identify novel therapeutic targets and candidate drugs for bladder cancer. First, differential expression analysis and WGCNA were performed using the GEO dataset GSE3167. Concurrently, Mendelian randomization analysis (SMR/HEIDI) was conducted utilizing blood tissue eQTL data and pooled bladder cancer GWAS data. Core candidate genes were identified by intersecting the results from these three analytical approaches. Subsequently, potential targeting drugs were predicted via the DSigDB database and computationally validated through molecular docking, MD simulations, and ADMET analysis. Finally, in vitro functional experiments were conducted to validate the key target, HSPA4, and the candidate drug, DON. **, P<0.01; ***, P<0.001. ADMET, absorption, distribution, metabolism, excretion, and toxicity; DEG, differentially expressed gene; DON, 6-diazo-5-oxo-L-norleucine; DSigDB, Drug Signature Database; eQTL, expression quantitative trait loci; GEO, Gene Expression Omnibus; GMFA, gene mutation frequency analysis; GWAS, genome-wide association study; HEIDI, heterogeneity in dependent instrument; IC₅₀, half-maximal inhibitory concentration; KEGG, Kyoto Encyclopedia of Genes and Genomes; MD, molecular dynamics; ns, nanosecond; Rg, radius of gyration; RMSD, root-mean-square deviation; RMSF, root-mean-square fluctuation; SASA, solvent-accessible surface area; SMR, summary data-based Mendelian randomization; WGCNA, weighted gene co-expression network analysis.

Methods

Differential expression analysis and weighted gene co-expression network construction

To systematically identify transcriptomic alterations associated with bladder cancer, the gene expression dataset GSE3167 was obtained from the Gene Expression Omnibus (GEO) database (11). This dataset comprises expression profiles from both bladder cancer tissues and adjacent normal tissues.

First, differential expression analysis was performed on dataset GSE3167 using the limma package in R. Following data standardization and log₂-transformation, genes meeting the thresholds of an adjusted P<0.05 and an absolute log₂ fold change (|log₂FC|) > 1 were identified as significantly differentially expressed in bladder cancer tissue. These genes were subsequently defined as differentially expressed genes (DEGs) and categorized as either up- or down-regulated.

Subsequently, WGCNA was performed on all gene expression data from GSE3167 using the R package WGCNA to explore synergistic expression relationships and identify gene modules highly correlated with the bladder cancer phenotype (12). First, sample clustering was examined to identify and exclude outliers. A scale-free co-expression network was then constructed by selecting a soft threshold (β) that achieved a scale-free topological fit index close to 0.9. Gene modules were identified using dynamic tree pruning. The module eigengenes (MEs) were correlated with the bladder cancer phenotype to select key modules demonstrating both the highest correlation with disease status (|correlation| >0.6) and statistical significance (P<0.05). Finally, genes within the key modules were extracted as a candidate bladder cancer-associated gene set for subsequent analyses. The study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments.

Summary data-based Mendelian randomization (SMR) analysis

To systematically uncover causal gene variants driving bladder cancer risk, we implemented SMR analytical workflows (8). This statistical framework leverages genetic variants as instrumental variables to reduce confounding bias existing in traditional observational research, and quantifies the pleiotropic connections between gene expression levels and bladder cancer susceptibility.

We acquired complete summary statistics for bladder cancer genome-wide association studies from the GWAS Catalog database (https://www.ebi.ac.uk/gwas/home) (13). For expression quantitative trait loci (eQTL) datasets, we adopted GTEx_V8 whole-blood tissue data provided by the SMR software official platform, which records the correlation between germline variants and transcriptional expression across diverse human tissues (14).

Single nucleotide polymorphisms (SNPs) reaching genome-wide significance threshold (P<5×10⁻⁸) were screened as valid instrumental variables to infer causal effects of gene expression on bladder cancer risk.

All SMR statistical calculations and heterogeneity in dependent instruments (HEIDI) linkage differentiation tests were run via the official downloadable SMR tool. Genes with significant SMR P values (P<0.05) alongside non-significant HEIDI statistics (P>0.05) were marked as high-confidence candidate targets for subsequent multi-omics validation.

The analytical outputs of this work offer novel clues to decode the genetic architecture of bladder cancer and screen promising therapeutic targets for further functional verification.

Identification and cross-validation of core candidate genes

To enhance the robustness of the key gene selection, a multi-tiered validation strategy was employed, integrating computational cross-validation with subsequent experimental assays. First, the high-confidence gene set identified in Section “Differential expression analysis and weighted gene co-expression network construction”—comprising the intersection of DEGs and key WGCNA module genes—was cross-validated against the list of significant genes derived from the SMR analysis in Section “Summary data-based Mendelian randomization (SMR) analysis”. The overlapping genes from these three analytical sources (DEGs, key WGCNA module genes, and SMR-significant genes) were defined as “core candidate genes”, provided they demonstrated consistent directional change (i.e., all co-upregulated or all co-downregulated).

Subsequently, the expression patterns of these core candidate genes were validated externally using three independent bladder cancer gene expression datasets (GSE32894, GSE13507, and GSE37816). This analysis determined whether the differential expression of these candidate genes between cancerous and normal tissues was both consistent and statistically significant across the validation cohorts (15).

Drug prediction and molecular docking

Evaluating the interactions between proteins and small molecules is critical for assessing the therapeutic potential of a target protein. To this end, a drug prediction analysis was performed using the Drug Signature Database (DSigDB, http://dsigdb.tanlab.org/DSigDBv1.0/). Specifically, the genetic identifiers for the target proteins identified in the cross-validation step were submitted to DSigDB. This enabled the prediction of candidate compounds with potential binding affinity for the target genes, thereby identifying leads for targeted therapy (16).

To investigate the molecular interactions between the target proteins and predicted therapeutic compounds, molecular docking was performed using the CB-Dock2 platform. CB-Dock2 is an advanced online tool that integrates cavity detection, molecular docking, and homology-based template fitting to improve the accuracy of binding site prediction and ligand docking (17). The platform achieved a success rate of 85.9% on its benchmark test dataset.

The three-dimensional structures of the target proteins were retrieved from the Protein Data Bank (PDB) (18). Prior to docking, the CB-Dock2 system automatically removed water molecules, ions, and other non-essential atoms from the protein structures (17). Ligand structures were uploaded in MOL2, MOL, SDF, or PDB format. Files in MOL or SDF format were converted to PDB format using Open Babel. For each docking run, the protein and ligand structure files were uploaded to the CB-Dock2 server, specifying the number of cavities for docking (default: 5). The docking process comprised three main steps: (I) cavity detection using CurPocket, which identifies potential binding sites on the protein surface via curvature analysis; (II) molecular docking with AutoDock Vina, which generates and ranks binding conformations based on docking scores; and (III) homology template fitting via FitDock, which refines the results by aligning the ligand with templates of high topological similarity (19).

Upon completion, the docking system provides detailed visual outputs, including the composition of the binding pocket, geometric parameters, the optimal ligand binding conformation, predicted binding affinity, and molecular interaction patterns. All result files can be downloaded for further analysis.

To evaluate the drug-like properties and safety profiles of the candidate compounds, a systematic predictive analysis of their absorption, distribution, metabolism, excretion, and toxicity (ADMET) characteristics was conducted.

Specifically, the canonical SMILES structures of the candidate compounds were submitted to computational platforms such as ADMETlab 2.0 to predict key ADMET parameters (20). These parameters included oral bioavailability, blood-brain barrier permeability, P-glycoprotein substrate potential, cytochrome P450 (CYP450) enzyme metabolism profiles, drug half-life, and toxicity risks (e.g., mutagenicity, cardiotoxicity). Compounds exhibiting favorable ADMET profiles and high drug-likeness were selected for further mechanistic studies based on established standard thresholds.

MD simulations

MD simulations were conducted using GROMACS 2022. The protein-ligand complex was parameterized with the AMBER14SB force field for the protein, the General AMBER Force Field (GAFF) for the ligand, and the transferable intermolecular potential with 3 points (TIP3P) model for water. Electrostatic interactions were calculated using the Particle Mesh Ewald (PME) method with a 1.2 nm cutoff, while van der Waals interactions were truncated at 1.0 nm. The system was equilibrated for 100 ps under NVT (constant particle number, volume, and temperature) and NPT (constant particle number, pressure, and temperature) ensembles, with the temperature and pressure maintained at 298 K and 1 bar using the V-rescale and Berendsen coupling methods, respectively. Finally, a 100 ns production simulation was performed with a 2fs integration timestep, saving trajectory coordinates every 10 ps. The simulated trajectories were visualized and analyzed using VMD and PyMOL. The protein-ligand binding free energy was calculated using the molecular mechanics Poisson-Boltzmann surface area (MMPBSA) method with the g_mmpbsa tool (21).

Phenotypic experiments

To investigate the function of HSPA4 in bladder cancer and the potential inhibitory effects of 6-diazo-5-oxo-L-norleucine (DON), the following in vitro phenotypic assays were conducted.

Cell transfection and drug treatment

To modulate HSPA4 expression, an HSPA4 overexpression plasmid (Gemma Gene, Cat No. GG-HSPA4-01, RRID: SCR_021568, Shanghai, China) was transfected into the human bladder cancer cell lines T24 and UMUC-3. The T24 and UMUC-3 cell lines (Species: Homo sapiens, Strain: Human bladder cancer, ATCC, Cat No. T24: HTB-4; UMUC-3: CRL-1749, RRID: T24: CVCL_0063; UMUC-3: CVCL_0087) were used in this experiment. Transfection was performed using Lipofectamine™ 2000 reagent (Thermo Fisher Scientific, Cat No. 11668019, RRID: AB_2536236, Waltham, USA). Cells were seeded into 6-well plates (Corning, Cat No. 3516, RRID: AB_2814850, Corning, USA) prior to transfection. At 70–80% confluence, the transfection complex was prepared by mixing 4 µg of plasmid DNA with 10 µL of Lipofectamine™ 2000 in 500 µL of Opti-MEM medium (Thermo Fisher Scientific, Cat No. 31985070, RRID: AB_2534826), following the manufacturer’s protocol. Following a 20-minute incubation at room temperature, the complex was added to the wells. After 6 hours, the medium was replaced with complete growth medium, and the cells were cultured for an additional 48 hours before further analysis.

To assess the effect of DON (Sigma-Aldrich, Cat No. D2141, RRID: AB_259828, Saint Louis, USA) on cell viability, a Cell Counting Kit-8 (CCK-8) assay was performed using a CCK-8 kit (Beyotime Biotechnology, Cat No. C0038, RRID: AB_2827882, Shanghai, China). T24 and UMUC-3 cells were seeded into 96-well plates (Corning, Cat No. 3599, RRID: AB_2814851) and treated with a range of DON concentrations (0.01–100 µM) for 48 hours, with five replicate wells per condition. Based on the viability results, subsequent experiments included an untreated control group and two DON treatment groups at selected concentrations.

Cell invasion assay

Cell invasion capacity was assessed using a Transwell assay. Transwell inserts (8 µm pore size, Corning, Cat No. 3422, RRID: AB_2814852) were used in this experiment. First, 50 µL of Matrigel (Corning, Cat No. 356234, RRID: AB_2814853) diluted 1:8 in serum-free medium was used to pre-coat the upper chamber of the Transwell insert and allowed to polymerize at 37 °C for 2 hours. Subsequently, 5×10⁵ cells were resuspended in 200 µL of serum-free medium and seeded into the upper chamber. The lower chamber was filled with 600 µL of complete medium supplemented with 10% fetal bovine serum (FBS) (Gibco, Cat No. 10099141, RRID: AB_2524026, Grand Island, USA) as a chemotactic attractant. Following a 24-hour incubation, non-invaded cells on the upper surface of the membrane were removed with a cotton swab (Puritan, Cat No. 806-WB, RRID: SCR_019058, Guilford, USA). Cells that had invaded through the membrane were fixed with 4% paraformaldehyde (Beyotime Biotechnology, Cat No. P0099, RRID: AB_2827883) for 30 minutes and stained with 0.25% crystal violet (Sigma-Aldrich, Cat No. C3886, RRID: AB_259868) for 20 minutes. Five randomly selected fields of view were imaged under a microscope (Olympus, Model: BX53, RRID: SCR_018681, Tokyo, Japan) for cell counting and analysis.

Colony formation assay

To assess long-term proliferative capacity, a colony formation assay was performed. T24 and UMUC-3 cells were seeded in 6-well plates (Corning, Cat No. 3516, RRID: AB_2814850) at a density of 2×10³ cells per well and treated with designated concentrations of DON (Sigma-Aldrich, Cat No. D2141, RRID: AB_259828). The cells were cultured for 10–14 days, with the medium refreshed every 3 days. Upon the formation of visible colonies, the medium was aspirated, and the cells were washed twice with phosphate-buffered saline (PBS) (Beyotime Biotechnology, Cat No. P0014, RRID: AB_2827884), fixed with 4% paraformaldehyde (Beyotime Biotechnology, Cat No. P0099, RRID: AB_2827883) for 30 minutes, and stained with 0.25% crystal violet (Sigma-Aldrich, Cat No. C3886, RRID: AB_259868) for 30 minutes. Following staining, the plates were washed with PBS, air-dried, scanned, and colonies containing more than 50 cells were counted.

GeneMANIA analysis and experimental validation

A protein-protein interaction (PPI) network was constructed using the GeneMANIA online platform (22). The network analysis incorporated the three core target genes identified from previous screening along with 20 additional genes predicted by the algorithm to have potential functional associations. Functional enrichment analysis was subsequently performed on the complete gene set within the network to systematically elucidate the relevant biological processes and signaling pathways.

To validate the expression of relevant molecules and the activity of key proteins within the endoplasmic reticulum stress (ERS) pathway at the protein level, a Western blot analysis was performed. Total protein was extracted from T24 and UMUC-3 cells using RIPA Lysis Buffer (Beyotime Biotechnology, Cat No. P0013B, RRID: AB_2827878) supplemented with phenylmethylsulfonyl fluoride (PMSF, Beyotime Biotechnology, Cat No. ST506, RRID: AB_2827880). Protein concentration was quantified using a Bicinchoninic Acid (BCA) Assay Kit (Beyotime Biotechnology, Cat No. P0010, RRID: AB_2827879). Equal amounts of protein (25 µg per lane) were resolved by 10% sodium dodecyl sulfate-polyacrylamide gel electrophoresis (SDS-PAGE) using electrophoresis buffer (Beyotime Biotechnology, Cat No. P0012A) and subsequently transferred to Polyvinylidene Fluoride (PVDF) membranes (Millipore, Cat No. IPVH00010, RRID: AB_10001342, Burlington, USA).

The membranes were blocked with 5% non-fat milk at room temperature for 1 h and then incubated overnight at 4 °C with the following primary antibodies: rabbit anti-HSPA4 (Abcam, Cat No. ab13492, RRID: AB_2079756, Cambridge, UK), anti-XBP1 (CST, Cat No. 12782S, RRID: AB_2798520, Danvers, USA), anti-IRE1α (CST, Cat No. 3294S, RRID: AB_2295073), and anti-phospho-IRE1α (p-IRE1α, CST, Cat No. 3294T, RRID: AB_2798521). After washing with Tris-buffered saline containing 0.1% Tween 20 (TBST, Beyotime Biotechnology, Cat No. P0023B), the membranes were incubated with HRP-conjugated goat anti-rabbit IgG secondary antibody (CST, Cat No. 7074S, RRID: AB_2099233) at room temperature for 1 h. Finally, protein bands were visualized using an enhanced chemiluminescence (ECL) substrate (Beyotime Biotechnology, Cat No. P0018FS, RRID: AB_2827881), and signals were detected with a digital imaging system (Tanon, Model: 5200Multi).

Statistical analysis

Multiple statistical methods were employed for data analysis. For the bioinformatics analyses, DEGs were identified using linear models, with multiple testing correction applied via the false discovery rate (FDR) method. In the WGCNA, the association between gene modules and phenotypes was assessed using Pearson correlation. Results from the Mendelian randomization analysis were also corrected for multiple comparisons using the FDR method.

For molecular docking and MD simulations, binding free energy and the structural stability of the protein-ligand complex served as the primary evaluation metrics. In the experimental validation, quantitative data from quantitative reverse transcription polymerase chain reaction (qRT-PCR), Western blotting, and cellular functional assays are presented as the mean ± standard deviation. Comparisons between two groups were performed using unpaired Student’s t-tests. Comparisons among multiple groups were conducted using one-way analysis of variance (ANOVA), followed by appropriate post-hoc tests. All statistical analyses were performed using R software (version 4.2.0) and GraphPad Prism (version 9.0).

Results

Multi-omics integrated screening identifies key driver genes in bladder cancer

To systematically identify core driver genes associated with bladder cancer pathogenesis, differential expression analysis and WGCNA were performed on the GEO dataset GSE3167. These results were integrated with summary-based Mendelian randomization (SMR/HEIDI) analysis for causal inference using pooled data. Differential expression analysis identified 2,156 significantly DEGs (|log₂FC| >1, adjusted P<0.05), consisting of 1,248 upregulated and 908 downregulated genes (Figure 2A). WGCNA constructed a scale-free co-expression network using a soft threshold power (β) of 6 (Figure 2B). Multiple modules were identified via dynamic tree cutting (Figure 2C). Module-trait relationship analysis revealed that the blue module exhibited the strongest positive correlation with bladder cancer status (tumor vs. normal) (correlation r=0.71, P=8×10⁻⁹) (Figure 2D). Within this module, the module membership (MM) of genes was highly correlated with their gene significance (GS) for the phenotype (Figure 2E).

Figure 2 Identification of key driver genes in bladder cancer. (A) Volcano plot of DEGs in the GSE3167 dataset. The x-axis represents log₂ FC, and the y-axis represents −log₁₀(P). Red and blue points denote significantly upregulated and downregulated genes, respectively (|log₂FC| >1, adjusted P<0.05). (B) Selection of the soft-thresholding power for WGCNA. The horizontal axis shows candidate power values (β), while the left and right vertical axes represent the scale-free topology model fit index (R²) and the mean connectivity, respectively. The minimum power (β=6) was chosen to achieve a network approximating scale-free topology (R²>0.85) while preserving high connectivity. (C) Gene hierarchical clustering dendrogram and module assignment. The dendrogram was constructed using the selected soft-thresholding power, and co-expression modules were identified via dynamic tree cutting. Distinct colors represent different modules. (D) Heatmap depicting module-trait correlations. Each cell shows the correlation coefficient (top) and the corresponding P value (bottom) between a ME and the tumor/normal phenotype. The blue module exhibited the strongest positive correlation with the cancer phenotype and was selected as the key module. (E) Scatterplot of MM versus GS for genes in the key module. GS is defined as the absolute correlation between gene expression and the tumor/normal phenotype. The strong positive Pearson correlation (r=0.63, P<1×10⁻²⁰) indicates that highly connected intramodular genes are also strongly associated with bladder cancer status. BC, breast cancer; DEGs, differentially expressed genes; FC, fold change; GS, gene significance; ME, module eigengene; MM, module membership; WGCNA, weighted gene co-expression network analysis.

Multi-omics cross-validation and experimental verification of core candidate genes

In participants of European ancestry, our SMR analysis initially identified several putative causal genes associated with bladder cancer (Figure 3A and table available at https://cdn.amegroups.cn/static/public/tau-2026-0330-1.xlsx). To enhance the reliability of the screening results and identify core candidate genes, we performed triple cross-validation by intersecting these SMR-identified putative causal genes (with HEIDI P>0.05), DEGs, and genes from the key WGCNA module.

Figure 3 Multi-omics integration identifies core driver genes in bladder cancer. (A) Volcano plot of SMR analysis results. The x-axis represents the SMR effect size (beta SMR), and the y-axis represents the negative logarithm of the SMR P value [−log₁₀(P SMR)]. Genes with significant SMR associations (P SMR <0.05) and non-significant HEIDI test results (P HEIDI >0.05) are highlighted, with the size of the dots proportional to −log₁₀(P SMR). (B,C) Venn diagrams showing the triple intersection of gene sets to identify core candidate genes. (B) Intersection of upregulated DEGs (DEG-up), genes from the key WGCNA module (WGCN A), and upregulated putative causal genes identified by SMR analysis (SMR-up). (C) Intersection of downregulated DEGs (DEG-down), genes from the key WGCNA module (WGCN A), and downregulated putative causal genes identified by SMR analysis (SMR-down). (D-F) Regional SMR plots prioritizing three core candidate genes with pleiotropic associations with bladder cancer in participants of European ancestry, using GTEx Whole_Blood eQTL data. For each gene locus: the top panel shows the GWAS association results for bladder cancer [−log₁₀(P GWAS)], the middle panel shows the eQTL association results [−log₁₀(P eQTL)], and the bottom panel displays the gene structure in the region. (D) TGOLN2 locus; (E) HSPA4 locus; (F) SYNM locus. The red diamond in each GWAS plot marks the lead SNP for the corresponding gene, with the red dashed line indicating the genome-wide significance threshold (P GWAS =0.05). DEG, differentially expressed gene; eQTL, expression quantitative trait loci; GWAS, genome-wide association study; HEIDI, heterogeneity in dependent instrument; SMR, summary data-based Mendelian randomization; WGCNA, weighted gene co-expression network analysis.

Venn diagram analysis of the three gene sets identified three core candidate genes at their intersection, including two co-upregulated genes (HSPA4 and TGOLN2) and one co-downregulated gene (SYNM) (Figure 3B,3C). Specifically, using GTEx eQTL data from Whole_Blood tissue, the SMR analysis results for these three core genes were as follows: HSPA4 [SMR analysis P value (PSMR) =0.000984524; HEIDI test P value (PHEIDI) =0.2275759; FDR =0.963772554959785], TGOLN2 (PSMR =0.00230376; PHEIDI =0.4699494; FDR =0.963772554959785), and SYNM (PSMR =0.02135751; PHEIDI =0.939058; FDR =0.963772554959785).

Overall, our study revealed that the expression changes of these three core genes may play a causal role in the pathogenesis of bladder cancer, which is consistent with the results of our SMR analysis and FDR correction (Figure 3D-3F). External validation using independent datasets further supported these findings (Figure S1).

Computational prediction and molecular docking of candidate drugs

For the core target HSPA4, the DSigDB database was used for reverse pharmacophore screening to predict potential binding compounds. To evaluate binding potential, four candidate molecules were selected for molecular docking. These included traditional chemotherapeutic agents (perillyl alcohol, ifosfamide, cisplatin) and the glutamine antagonist DON, which served as the primary focus of this study. Docking with CB-Dock2 indicated that all ligands bound within a potential active site of HSPA4 (Figure 4A-4D). DON demonstrated the strongest predicted binding affinity (−7.5 kcal/mol), which was substantially lower (i.e., more favorable) than that of the other candidates (Figure 4D). Analysis of the docked conformation revealed that DON forms stable hydrogen bonds and hydrophobic interactions with several key amino acid residues in the HSPA4 binding pocket, suggesting a high degree of target specificity. A comprehensive computational analysis of the physicochemical and pharmacokinetic properties of DON was conducted. DON has the molecular formula C₆H₉N₃O₅, a molecular weight of 171.15 g/mol, and a topological polar surface area (TPSA) of 117.78 Ų. It possesses 6 hydrogen bond acceptors, 2 hydrogen bond donors, and 5 rotatable bonds, indicating moderate structural flexibility. Key molecular properties are visualized in a radar plot, highlighting significant molecular size and polarity, along with moderate water solubility and lipophilicity. The predicted octanol-water partition coefficient (Log P_o/w) ranged from −3.80 to 0.61, with a consensus value of −1.86, confirming its hydrophilic nature and high-water solubility. Consistent with this, both the ESOL and Ali models predict high solubility (approximately 4.37×10¹ to 4.15×10¹ mmol/L), as does the SILICOS-IT model (1.45×10⁰ mmol/L).

Figure 4 Molecular docking analysis of candidate compounds with HSPA4. (A) Predicted binding pose of perillyl alcohol within the HSPA4 binding site (binding energy =−5.5 kcal/mol). (B) Predicted binding pose of ifosfamide within the HSPA4 binding site (binding energy =−6.2 kcal/mol). (C) Predicted binding pose of cisplatin within the HSPA4 binding site (binding energy =−6.6 kcal/mol). (D) Predicted binding pose of DON within the HSPA4 binding site (binding energy =−7.5 kcal/mol). DON, 6-diazo-5-oxo-L-norleucine.

Pharmacokinetic predictions indicate that this compound exhibits high gastrointestinal absorption but cannot cross the blood-brain barrier. It is not a P-glycoprotein substrate and does not inhibit drug-metabolizing enzymes such as CYP1A2, CYP2C19, CYP2C9, CYP2D6, or CYP3A4, suggesting a low risk of drug interactions. However, its skin permeability is poor [Log permeability coefficient (KP) =−9.84 cm/s]. Druggability assessment reveals that DON fully complies with Lipinski, Veber, and Egan rules, with only two violations each in the Ghose and Muegge rules. Overall druggability is favorable, and its bioavailability score of 0.55 indicates potential drug-like properties. Notably, the compound contains one Pan-Assay Interference Compounds (PAINS) warning structure (azo_A). Brenk analysis also indicates the presence of three potentially hazardous groups: diazo, imine, and quaternary nitrogen. Subsequent studies should focus on its potential toxicity and chemical reactivity. Regarding synthetic accessibility and lead potential, DON scored 2.27 for synthetic accessibility, indicating moderate chemical synthesis difficulty. It complies with the Leadlikeness rule (violating only the molecular weight <250 criterion), establishing a foundation for subsequent structural optimization as a lead compound (Figure S2).

MD simulation of the HSPA4-DON complex

To evaluate the stability of the HSPA4-DON complex under physiological conditions, a 100 ns MD simulation was performed. The root-mean-square deviation (RMSD) of the complex, the protein backbone, and the ligand all reached stable plateaus after approximately 20 ns, indicating that the system achieved a state of dynamic equilibrium (Figure 5A). The radius of gyration (Rg) of the protein remained stable around 28 Šduring the entire trajectory, suggesting that the overall folded structure of HSPA4 was not substantially perturbed by DON binding (Figure 5B). Furthermore, root-mean-square fluctuation (RMSF) analysis showed reduced flexibility in amino acid residues within the binding pocket, confirming its structural rigidity upon ligand interaction (Figure 5C). Furthermore, the distance between the centroid of the DON ligand and its initial binding site remained consistently within ~1.5 Š(Figure 5D), while the buried solvent-accessible surface area (SASA) stabilized at ~350 Ų (Figure 5E). These findings indicate the formation of a stable and tightly bound interaction interface. Energy decomposition analysis identified van der Waals interactions as the primary driving force for the binding (Figure 5F). Collectively, the MD simulations provide strong evidence for the high stability and binding specificity of the HSPA4-DON complex.

Figure 5 MD simulation analysis of the HSPA4-DON complex (100 ns). (A) RMSD over time for the protein backbone (blue), the ligand DON (orange), and the entire complex (gray). The system reaches equilibrium after ~20 ns. (B) Time evolution of the Rg for the protein. The stable Rg value (~28 Å) indicates maintenance of a compact, folded conformation. (C) Per-residue RMSF. Peaks correspond to flexible regions (e.g., termini, loops). Residues in the binding site show lower fluctuation, indicating a rigid pocket. (D) Distance between the centroid of DON and the centroid of its initial binding site. The stable distance (~1.5 Å) suggests minimal ligand displacement. (E) Buried SASA at the protein-ligand interface. The stable value (~350 Ų) reflects a persistent, extensive interaction surface. (F) Decomposition of the interaction free energy into VDW and ELE components. VDWs interactions provide the major stabilizing contribution. DON, 6-diazo-5-oxo-L-norleucine; ELE, electrostatic; MD, molecular dynamics; Rg, radius of gyration; RMSD, root-mean-square deviation; RMSF, root-mean-square fluctuation; SASA, solvent-accessible surface area; VDW, van der Waals.

DON inhibits malignant phenotypes of bladder cancer cells by targeting HSPA4

To investigate the biological function of DON, we first evaluated its anti-proliferative activity in T24 bladder cancer cells via CCK-8 assay. DON exerted dose-dependent growth suppression on T24 cells, with a half-maximal inhibitory concentration (IC₅₀) value of 2.07 µM (Figure 6A). Subsequent colony formation assays revealed that DON markedly impaired the long-term proliferative capacity of T24 cells (Figure 6B), and Transwell assays further verified that DON obviously restrained the migratory potential of T24 cells (Figure 6C).

Figure 6 DON inhibits bladder cancer cell proliferation and migration, effects partially reversed by HSPA4-OE. (A,D) Viability of T24 (A) and UM-UC-3 (D) cells treated with DON for 48 h (CCK-8 assay). DON inhibited proliferation in a dose-dependent manner, with IC₅₀ values of 2.07 μM (T24) and 5.23 μM (UM-UC-3). (B,E) Colony formation assays in T24 (B) and UM-UC-3 (E) cells. DON treatment significantly reduced colony formation. This inhibition was partially rescued in cells HSPA4-OE. (C,F) Transwell migration assays in T24 (C) and UM-UC-3 (F) cells. DON significantly reduced cell migration, an effect that was similarly partially restored by HSPA4-OE. Staining: crystal violet; magnification: ×100. Data are presented as mean ± SD (n=3). *, P<0.05; **, P<0.01; ***, P<0.001; ****, P<0.0001 versus the NC group. CCK-8, Cell Counting Kit-8; DON, 6-diazo-5-oxo-L-norleucine; DON + OE, co-treatment group; HSPA4-OE, HSPA4 overexpression group; IC₅₀, half-maximal inhibitory concentration; NC, negative control; SD, standard deviation.

Consistent inhibitory phenotypes were observed in UM-UC-3 cells. The CCK-8 assay showed that DON inhibited UM-UC-3 cell proliferation with an IC₅₀ of 5.23 µM (Figure 6D). DON also dramatically decreased colony formation (Figure 6E) and cell migration (Figure 6F) in UM-UC-3 cell line.

To further verify the regulatory role of HSPA4 in DON-induced anti-tumor effects, we constructed the HSPA4 overexpression (HSPA4-OE) cell model. Ectopic HSPA4-OE partially rescued the DON-mediated suppression of colony formation and migration in both T24 (Figure 6B,6C) and UM-UC-3 cells (Figure 6E,6F). This phenotypic rescue experiment strongly proves that HSPA4 acts as the core functional target of DON. Taken together, these data demonstrate that DON exerts anti-bladder cancer activity mainly by downregulating HSPA4.

HSPA4 mediates the antitumor effects of DON by regulating the ERS pathway

To elucidate the molecular mechanism of DON, a PPI network for HSPA4 was constructed using GeneMANIA. Network analysis indicated that HSPA4 closely interacts with numerous proteins associated with ATPase activity, chaperone function, and endoplasmic reticulum (ER) protein folding, positioning it as a central hub within cellular homeostasis networks (Figure 7A). Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway enrichment analysis showed that HSPA4-interacting proteins were significantly enriched in pathways including “Protein processing in ER” and “Endocytosis” (FDR <0.05) (Figure 7B), strongly implicating HSPA4 in ERS regulation. To test this hypothesis, key components of the ERS IRE1α-XBP1 axis were analyzed by Western blot. DON treatment significantly reduced HSPA4 protein levels in both T24 and UMUC-3 cells. Concurrently, DON induced phosphorylation of IRE1α (p-IRE1α) and upregulated the expression of its downstream target, the spliced form of XBP1 (XBP1s), indicating activation of the ERS pathway (Figure 7C,7D). Notably, overexpression of HSPA4 (HSPA4-OE) effectively reversed the DON-induced upregulation of p-IRE1α and XBP1s, partially restoring pathway activity toward baseline levels.

Figure 7 HSPA4 interaction network and its role in regulating the endoplasmic reticulum stress pathway. (A) PPI network of HSPA4 generated by GeneMANIA. Nodes represent proteins; edges represent interactions. Colors indicate interaction types (e.g., physical interaction, co-expression) and major functional annotations. (B) KEGG pathway enrichment analysis for HSPA4-interacting proteins. Significantly enriched pathways (FDR <0.05) include “Protein processing in endoplasmic reticulum”, “Endocytosis”, and “Longevity regulating pathway”, implicating HSPA4 in endoplasmic reticulum homeostasis and stress response. (C,D) Western blot analysis of key ERS pathway proteins in T24 (C) and UM-UC-3 (D) cells. DON treatment downregulated HSPA4 and activated the IRE1α-XBP1 axis, as shown by increased p-IRE1α and XBP1s. HSPA4-OE attenuated this activation, and the effect was partially reversed in the co-treatment group (DON + OE). GAPDH was used as a loading control. CON, control; DON, 6-diazo-5-oxo-L-norleucine; DON + OE, DON + HSPA4-OE co-treatment; ERS, endoplasmic reticulum stress; FDR, false discovery rate; GAPDH, glyceraldehyde-3-phosphate dehydrogenase; HSPA4-OE, HSPA4 overexpression; KEGG, Kyoto Encyclopedia of Genes and Genomes; p-IRE1α, phosphorylation of IRE1α; PPI, protein-protein interaction; XBP1s, spliced form of XBP1.

In summary, these findings suggest that DON targets HSPA4, disrupting its role in maintaining ER homeostasis. This leads to excessive activation of the IRE1α-XBP1 stress pathway, which ultimately inhibits the proliferation and migration of bladder cancer cells.

Discussion

Through a multi-omics integration strategy, this study systematically identified HSPA4 as a key driver gene that is causally associated with and highly expressed in bladder cancer. Further computational drug prediction and in vitro functional validation demonstrated that DON, a glutamine antagonist, specifically targets HSPA4, significantly inhibiting proliferation, colony formation, and migration of T24 and UMUC-3 bladder cancer cells. Mechanistically, DON downregulates HSPA4 expression, inducing IRE1α phosphorylation and XBP1s, thereby activating the ERS pathway. Conversely, overexpression of HSPA4 partially reversed the antitumor effects of DON, confirming HSPA4 as its functional target.

These findings indicate that HSPA4 maintains ER homeostasis in bladder cancer cells, thus exerting a protective role. When HSPA4 function is suppressed, cells fail to adequately respond to protein folding stress, leading to sustained ERS and subsequent inhibition of malignant phenotypes. Hence, targeting HSPA4 attenuates the adaptive survival capacity of bladder cancer cells. Notably, DON displayed a significantly lower IC₅₀ in T24 cells (2.07 µM) than in UMUC-3 cells (5.23 µM), indicating potential cell-specific differences in HSPA4 dependency. This observation offers insights for stratified therapeutic approaches.

Within the heat shock protein (HSP) family, HSPA4 (HSP110), a core member of the HSP110 subfamily, has recently been recognized as an oncogene in various solid tumors. In hepatocellular carcinoma (HCC), HSPA4 is among the most highly active super-enhancer-associated genes. Huang et al. initially identified HSPA4 as such by integrating super-enhancer maps with transcriptomic data. Using CRISPR-Cas9 to specifically delete this super-enhancer region, they further demonstrated that its ablation reversed the aberrant overexpression of HSPA4, significantly suppressing both in vitro proliferation of HCC cells and their tumorigenicity in nude mice. These findings confirmed that aberrant activation of the super-enhancer region constitutes a critical event driving HSPA4 transcriptional upregulation and sustaining the malignant phenotype of HCC (23). In gastric cancer, HSPA4 has been established as a functional oncogene. Suo et al., using immunohistochemistry and multivariate Cox regression, reported that HSPA4 protein levels were significantly elevated in tumor tissues and independently correlated with poorer overall survival. Functionally, short hairpin RNA (shRNA)-mediated silencing of HSPA4 markedly impaired colony formation and subcutaneous xenograft growth of gastric cancer cells. This work further uncovered a novel immune evasion mechanism involving the ALKBH5/CD58 axis, thereby reinforcing HSPA4’s role as a functional oncogene (24). In head and neck squamous cell carcinoma, HSPA4 is not only an independent prognostic marker of poor outcome but also a promising therapeutic target (25). In breast cancer, Gu et al. used a mouse model of spontaneous lymph node metastasis to demonstrate that tumor-educated B cells activate ITGB5 and downstream Src/NF-κB pathways in tumor cells by secreting pathogenic IgG targeting the glycosylated membrane protein HSPA4, thereby mediating lymph node metastasis via the CXCR4/SDF1α axis. This work was the first to link HSPA4-targeted humoral immune responses to pre-metastatic niche formation (26). In colorectal cancer, Zhang et al. showed that shRNA-mediated knockdown of HSPA4 effectively suppressed tumor progression by inducing cell cycle arrest and apoptosis (27).

However, the expression pattern, causal role, and druggability of HSPA4 in bladder cancer have not been explored. This study is the first to establish a causal association between HSPA4 expression and bladder cancer risk using SMR and to systematically validate its functional role as an “endoplasmic reticulum gatekeeper”, thereby addressing this critical knowledge gap. Combined with the causal results from SMR analysis, we further interpreted the role of HSPA4 in bladder cancer. Genetic variants modulate the transcription level of HSPA4, leading to its sustained overexpression in bladder tissues. Abnormally high HSPA4 maintains ERS homeostasis of tumor cells, which ultimately drives the malignant progression of bladder cancer. Previous pan-cancer research has systematically analyzed the prognostic and immunological phenotypes of HSPA4 (28), so our work mainly focuses on its causal regulatory function in bladder cancer rather than repeating pan-tumor descriptive analysis.

The molecular chaperone network constitutes a core defense mechanism that enables tumor cells to withstand ERS by facilitating proper protein folding and preventing toxic aggregation, thereby maintaining proteostasis (29-31). Classic chaperones such as HSP90 and HSP70 have been extensively validated as anticancer targets (32,33). However, broad-spectrum HSP inhibitors (e.g., geldanamycin derivatives) are limited by significant toxicity and suboptimal efficacy due to their disruption of protein homeostasis in normal tissues (33). ERS and the unfolded protein response (UPR) exert context-dependent dual functions in tumors: moderate UPR activation promotes survival by restoring homeostasis, whereas sustained, unresolved stress shifts the signaling balance from adaptation to apoptosis (34,35). This duality underpins two therapeutic strategies: direct induction of exogenous stress to overwhelm tumor cells, or inhibition of adaptive UPR components to abrogate stress defense mechanisms (36). Early approaches predominantly adopted the former, exemplified by potent ERS inducers such as toxocarotene and euprolide. However, their clinical translation was hampered by poor tumor selectivity and pronounced systemic toxicity (37). More recently, selective small-molecule inhibitors targeting UPR sensors, including GRP78, IRE1α, and PERK, have been developed. Nonetheless, their efficacy as monotherapies remains modest, and concerns over disrupting normal tissue homeostasis persist (37,38).

The present study proposes an alternative strategy: targeting HSPA4—a tumor-dependent protective chaperone—to indirectly disrupt ER homeostasis, thereby inducing synthetic lethality by exploiting the inherent stress vulnerability of cancer cells. Unlike direct ERS modulation, this approach does not actively impose stress but instead disables the cellular stress buffer, leading to self-destruction under basal stress conditions. Consequently, this strategy offers enhanced therapeutic selectivity.

Mechanistically, this study uncovers a non-canonical role for HSPA4 that extends beyond its conventional molecular chaperone function. Traditional views characterize molecular chaperones primarily as “protein folding assistants” (31,39); however, accumulating evidence indicates their direct functional integration with the UPR pathway. Certain HSPs do not merely respond passively to stress but instead act as active regulatory nodes within UPR signaling. In a systematic review of BiP/GRP78, Lee et al. demonstrated that, under nonstress conditions, the ER chaperone BiP (HSPA5) binds to the luminal domains of the three UPR sensors—PERK, IRE1α, and ATF6—thereby maintaining them in inactive conformations. Upon accumulation of unfolded proteins in the ER lumen, BiP preferentially binds exposed hydrophobic residues, dissociates from the sensors, and relieves UPR inhibition, thereby initiating adaptive responses (40). Furthermore, Hetz et al. highlighted that P58ᴵᴾᴷ (DNAJC3), an HSP40 family member, participates in a negative‑feedback loop that curbs UPR signaling. As a UPR target gene, P58ᴵᴾᴷ expression is upregulated during ERS; its protein product directly binds the PERK kinase domain, inhibiting PERK autophosphorylation and kinase activity, thereby selectively attenuating the proapoptotic PERK-eIF2α-ATF4-CHOP axis (41).

The present study extends this paradigm by demonstrating that HSPA4 serves as a dose-sensitive regulator and signal brake for ER homeostasis, modulating IRE1α phosphorylation and XBP1s. Under normal conditions, HSPA4 facilitates the clearance of misfolded proteins, buffers ERS, and sustains adaptive UPR outputs. Conversely, upon HSPA4 inhibition, unfolded proteins persistently accumulate, converting intermittent IRE1α-XBP1 signaling pulses into sustained activation that ultimately shifts cells from adaptive survival to apoptosis. This finding redefines HSPA4 from a passive gatekeeper to an active modulator of ER homeostasis, offering a novel entry point for understanding the regulatory switch mechanisms governing ERS signaling. ERS activates three canonical UPR branches: IRE1α, PERK and ATF6. As a glutamine antagonist, DON disturbs intracellular glutamine homeostasis. A high-impact study targeting glutamine competition demonstrated that glutamine deficiency predominantly induces activation of the IRE1α-XBP1s signaling cascade (42). For this reason, we selected this core axis as our primary detection target in current in vitro assays. All protein samples used for Western blot have been exhausted, so supplementary detection of PERK and ATF cannot be completed in this work. When we obtain adequate research funds and experimental resources in follow-up research, we will examine all classic ERS markers to comprehensively evaluate the overall UPR triggered by DON treatment.

DON is an established glutamine antagonist that exerts broad-spectrum antitumor activity through irreversible inhibition of glutaminase and multiple transaminases, thereby blocking both the carbon and nitrogen metabolic pathways of glutamine. However, its clinical application has long been limited by dose-dependent gastrointestinal toxicity and neurotoxicity (43,44). In recent years, prodrug derivatization has substantially expanded the therapeutic window of DON. High-quality preclinical studies have confirmed its antitumor activity and combination potential in multiple solid tumors. In pancreatic ductal adenocarcinoma, the DON prodrug DRP-104 induces metabolic crisis in pancreatic cancer cells and significantly inhibits tumor growth in vivo (43). Nevertheless, tumors can acquire adaptive resistance via activation of the MEK/ERK pathway; combining DRP-104 with the MEK inhibitor trametinib significantly prolonged survival in syngeneic xenograft models. A more notable advance is the application of DRP-104 in combination with immunotherapy. Yokoyama et al. systematically evaluated the synergy between DRP-104 and immune checkpoint inhibitors. In the CT26 colon cancer model, DRP-104 monotherapy inhibited tumor growth by 90%; combination with an anti-programmed cell death protein 1 (anti-PD-1) antibody increased inhibition to 94% and extended median survival from 17.5 to 56 days. Notably, all nine mice in the combination therapy group remained tumor‑free at the experimental endpoint (day 77) and exhibited complete resistance to syngeneic tumor rechallenge, suggesting that DRP-104 induces durable immune memory (45).

From a clinical translational perspective, this study holds substantial value. First, HSPA4 may serve as a potential biomarker: its consistently elevated expression across multiple independent cohorts and cell lines suggests utility in identifying high‑risk bladder cancer patients (e.g., those with high recurrence risk or invasive subtypes), thereby informing personalized therapeutic strategies. Second, the repurposing potential of DON markedly shortens the drug development timeline. As a glutamine antagonist, DON has been evaluated in multiple Phase I/II clinical trials for solid tumors such as glioblastoma and pancreatic cancer and possesses well‑characterized pharmacokinetic and safety profiles. Validation of its efficacy in preclinical bladder cancer models would therefore facilitate rapid clinical translation. Finally, our findings provide a theoretical rationale for combination therapy strategies. Given the frequent metabolic reprogramming in bladder cancer (e.g., heightened glutamine dependency), co‑administration of DON with immune checkpoint inhibitors (e.g., anti-PD-1 antibodies) or DNA-damaging agents (e.g., cisplatin) may synergistically augment cellular stress and immunogenic cell death, thereby potentiating antitumor efficacy. This combinatorial approach warrants systematic evaluation in future studies. As a multi-target glutamine antagonist, DON exerts two independent yet synergistic anti-tumor mechanisms. Its well-documented canonical function is inhibiting glutaminase to disrupt tumor glutamine metabolism. Distinct from this metabolic suppression effect, our study identifies a novel action mode: DON directly targets HSPA4 and activates ERS signaling. These two pathways operate independently to restrain bladder cancer malignant phenotypes and work together to strengthen anti-tumor efficacy. Such multi-target properties of DON offer novel theoretical support for developing combinatorial therapeutic regimens against bladder cancer.

The present study has certain limitations. First, all functional experiments were conducted in vitro using bladder cancer cell lines; the in vivo efficacy and toxicity of DON have not yet been validated in animal models [e.g., patient-derived xenografts (PDX) or orthotopic models]. Due to the limitation of research funds and experimental facilities, we cannot conduct xenograft tumor experiments in the present study. A previous study has confirmed the in vivo anti-tumor efficacy of DON derivatives in bladder cancer animal models, but its research core is immune regulation rather than ERS pathway (46). We will complete relevant in vivo verification to detect HSPA4 and full ERS markers once conditions permit in follow-up work. Due to the restriction of clinical specimen collection and ethical procedures, we did not detect self-owned bladder cancer tissues in this work. Nevertheless, accumulated studies on HCC and lung adenocarcinoma have proved that high HSPA4 expression is closely associated with advanced tumor stage and unfavorable prognosis (47,48). We further analyzed the expression difference and survival significance of HSPA4 using three independent public bladder cancer transcriptome datasets, which supports its potential as a clinical therapeutic target. Second, although molecular docking and kinetic simulations suggest direct binding between DON and HSPA4, their physical interaction remains to be confirmed by orthogonal biophysical assays, such as surface plasmon resonance (SPR) or cellular thermal shift assay (CETSA). Finally, we did not systematically assess the cytotoxicity of DON toward normal urinary tract epithelial cells—an evaluation critical for its clinical application.

Conclusions

Using a multi-omics cross-validation strategy, this study identifies HSPA4 as a key therapeutic target in bladder cancer. We demonstrate that the glutamine antagonist DON exerts antitumor effects by targeting HSPA4, leading to activation of the IRE1α-XBP1 ERS pathway. These findings provide not only a novel targeting strategy for bladder cancer but also a rationale for the clinical repurposing of DON. Future studies should focus on validating DON’s efficacy and safety in relevant animal models and exploring its potential in combination with standard chemotherapy or immunotherapy.

Acknowledgments

We thank all the participants involved in this study. We thank the Gene Expression Omnibus (GEO) database, GWAS Catalog database, DSigDB database, Protein Data Bank (PDB), and ADMETlab 2.0 platform for allowing us to use their data and resources.

Footnote

Reporting Checklist: The authors have completed the MDAR reporting checklist. Available at https://tau.amegroups.com/article/view/10.21037/tau-2026-0330/rc

Data Sharing Statement: Available at https://tau.amegroups.com/article/view/10.21037/tau-2026-0330/dss

Peer Review File: Available at https://tau.amegroups.com/article/view/10.21037/tau-2026-0330/prf

Funding: This study was supported by the “Yaodong Shenzhou-Pharmacy Scientific Research Capacity Building Fund” of the Bethune Public Welfare Foundation (grant No. Z04JKM2023E040).

Conflicts of Interest: All authors have completed the ICMJE uniform disclosure form (available at https://tau.amegroups.com/article/view/10.21037/tau-2026-0330/coif). The authors have no conflicts of interest to declare.

Ethical Statement: The authors are accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved. The study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments.

Open Access Statement: This is an Open Access article distributed in accordance with the Creative Commons Attribution-NonCommercial-NoDerivs 4.0 International License (CC BY-NC-ND 4.0), which permits the non-commercial replication and distribution of the article with the strict proviso that no changes or edits are made and the original work is properly cited (including links to both the formal publication through the relevant DOI and the license). See: https://creativecommons.org/licenses/by-nc-nd/4.0/.

References

1. van Hoogstraten LMC, Vrieling A, van der Heijden AG, et al. Global trends in the epidemiology of bladder cancer: challenges for public health and clinical practice. Nat Rev Clin Oncol 2023;20:287-304. [Crossref] [PubMed]
2. Sung H, Ferlay J, Siegel RL, et al. Global Cancer Statistics 2020: GLOBOCAN Estimates of Incidence and Mortality Worldwide for 36 Cancers in 185 Countries. CA Cancer J Clin 2021;71:209-49. [Crossref] [PubMed]
3. Liu S, Feng C, Tan L, et al. Single-cell dissection of multifocal bladder cancer reveals malignant and immune cells variation between primary and recurrent tumor lesions. Commun Biol 2024;7:1659. [Crossref] [PubMed]
4. Li F, Zhang H, Huang Y, et al. Single-cell transcriptome analysis reveals the association between histone lactylation and cisplatin resistance in bladder cancer. Drug Resist Updat 2024;73:101059. [Crossref] [PubMed]
5. Liu S, Chen X, Lin T. Lymphatic metastasis of bladder cancer: Molecular mechanisms, diagnosis and targeted therapy. Cancer Lett 2021;505:13-23. [Crossref] [PubMed]
6. Galesloot TE, Grotenhuis AJ, Kolev D, et al. Genome-wide Meta-analysis Identifies Novel Genes Associated with Recurrence and Progression in Non-muscle-invasive Bladder Cancer. Eur Urol Oncol 2022;5:70-83. [Crossref] [PubMed]
7. Wang Y, Zhu H, Ren J, et al. Integrative machine learning models predict prostate cancer diagnosis and biochemical recurrence risk: Advancing precision oncology. NPJ Digit Med 2025;8:524. [Crossref] [PubMed]
8. Richmond RC, Davey Smith G. Mendelian Randomization: Concepts and Scope. Cold Spring Harb Perspect Med 2022;12:a040501. [Crossref] [PubMed]
9. Lovegrove CE, Howles SA, Furniss D, et al. Causal inference in health and disease: a review of the principles and applications of Mendelian randomization. J Bone Miner Res 2024;39:1539-52. [Crossref] [PubMed]
10. Brandi A, Majello B, Simeone I, et al. MiDNE a tool for Multi-omics genes and drugs interactions discovery. Comput Struct Biotechnol J 2025;27:4469-75. [Crossref] [PubMed]
11. Huang W, Zhu L, Huang H, et al. IGF2BP3 overexpression predicts poor prognosis and correlates with immune infiltration in bladder cancer. BMC Cancer 2023;23:116. [Crossref] [PubMed]
12. Xiao Y, Xu D, Bao E, et al. Identification of inflammation related gene signatures for bladder cancer prognosis prediction. Sci Rep 2024;14:28867. [Crossref] [PubMed]
13. Koyama S, Ito K, Terao C, et al. Population-specific and trans-ancestry genome-wide analyses identify distinct and shared genetic risk loci for coronary artery disease. Nat Genet 2020;52:1169-77. [Crossref] [PubMed]
14. The GTEx Consortium atlas of genetic regulatory effects across human tissues. Science 2020;369:1318-30.
15. Cheng Q, Duan W, He S, et al. Multi-Omics Data Integration Analysis of an Immune-Related Gene Signature in LGG Patients With Epilepsy. Front Cell Dev Biol 2021;9:686909. [Crossref] [PubMed]
16. Kuleshov MV, Jones MR, Rouillard AD, et al. Enrichr: a comprehensive gene set enrichment analysis web server 2016 update. Nucleic Acids Res 2016;44:W90-7. [Crossref] [PubMed]
17. Liu Y, Yang X, Gan J, et al. CB-Dock2: improved protein-ligand blind docking by integrating cavity detection, docking and homologous template fitting. Nucleic Acids Res 2022;50:W159-64. [Crossref] [PubMed]
18. Protein Data Bank. the single global archive for 3D macromolecular structure data. Nucleic Acids Res 2019;47:D520-8.
19. Yang X, Liu Y, Gan J, et al. FitDock: protein-ligand docking by template fitting. Brief Bioinform 2022;23:bbac087. [Crossref] [PubMed]
20. Xiong G, Wu Z, Yi J, et al. ADMETlab 2.0: an integrated online platform for accurate and comprehensive predictions of ADMET properties. Nucleic Acids Res 2021;49:W5-W14. [Crossref] [PubMed]
21. Valdés-Tresanco MS, Valdés-Tresanco ME, Valiente PA, et al. gmx_MMPBSA: A New Tool to Perform End-State Free Energy Calculations with GROMACS. J Chem Theory Comput 2021;17:6281-91. [Crossref] [PubMed]
22. Noll LK, Giuliani NR, Beauchamp KG, et al. Behavioral and neural correlates of parenting self-evaluation in mothers of young children. Soc Cogn Affect Neurosci 2018;13:535-45. [Crossref] [PubMed]
23. Huang J, Xu L, Chen Y, et al. Redistribution of super-enhancers promotes malignancy in human hepatocellular carcinoma. J Adv Res 2026;82:803-19. [Crossref] [PubMed]
24. Suo D, Gao X, Chen Q, et al. HSPA4 upregulation induces immune evasion via ALKBH5/CD58 axis in gastric cancer. J Exp Clin Cancer Res 2024;43:106. [Crossref] [PubMed]
25. Fan G, Tu Y, Wu N, et al. The expression profiles and prognostic values of HSPs family members in Head and neck cancer. Cancer Cell Int 2020;20:220. [Crossref] [PubMed]
26. Gu Y, Liu Y, Fu L, et al. Tumor-educated B cells selectively promote breast cancer lymph node metastasis by HSPA4-targeting IgG. Nat Med 2019;25:312-22. [Crossref] [PubMed]
27. Zhang M, Dai W, Li Z, et al. HSPA4 Knockdown Retarded Progression and Development of Colorectal Cancer. Cancer Manag Res 2021;13:4679-90. [Crossref] [PubMed]
28. Yang J, Wu X, You J. Unveiling the potential of HSPA4: a comprehensive pan-cancer analysis of HSPA4 in diagnosis, prognosis, and immunotherapy. Aging (Albany NY) 2024;16:2517-41. [Crossref] [PubMed]
29. Twito G, Abu Abayed FA, Gilad A, et al. Non-genetic inactivation of caspase-3 and P53 increases cancer cell fitness by PDIA4 redistribution. Oncogene 2025;44:4565-75. [Crossref] [PubMed]
30. Zhang C, Li J, Tang Q, et al. Targeting proteostasis for cancer therapy: current advances, challenges, and future perspectives. Mol Cancer 2025;24:265. [Crossref] [PubMed]
31. Kuzu OF, Granerud LJT, Saatcioglu F. Navigating the landscape of protein folding and proteostasis: from molecular chaperones to therapeutic innovations. Signal Transduct Target Ther 2025;10:358. [Crossref] [PubMed]
32. Cai RD, Lin MJ, Ye QM. Overcoming cancer drug resistance through small-molecule targeting of HSP90 and HSP70. Cancer Drug Resist 2025;8:60. [Crossref] [PubMed]
33. Abdullah O, Omran Z. Geldanamycins: Potent Hsp90 Inhibitors with Significant Potential in Cancer Therapy. Int J Mol Sci 2024;25:11293. [Crossref] [PubMed]
34. Chen P, Sharma A, Weiher H, et al. Biological mechanisms and clinical significance of endoplasmic reticulum oxidoreductase 1 alpha (ERO1α) in human cancer. J Exp Clin Cancer Res 2024;43:71. [Crossref] [PubMed]
35. Ilieva E, Avnet S, Baldini N, et al. The Unfolded Protein Response in Sarcomas: From Proteostasis to Therapy Resistance. Cancers (Basel) 2025;17:3489. [Crossref] [PubMed]
36. Marfatiya S, Mubariz F, Pal A, et al. Targeting the unfolded protein response in cancer: exploiting endoplasmic reticulum stress for therapeutic intervention. Biochem Pharmacol 2025;242:117221. [Crossref] [PubMed]
37. Agalakova NI. Modulation of Endoplasmic Reticulum Stress in Experimental Anti-Cancer Therapy. Int J Mol Sci 2025;26:6407. [Crossref] [PubMed]
38. Khan JT, Sood S, Piedra V, et al. The hidden hand of endoplasmic reticulum stress in anticancer drug resistance. Crit Rev Oncol Hematol 2026;217:104998. [Crossref] [PubMed]
39. Morán Luengo T, Mayer MP, Rüdiger SGD. The Hsp70-Hsp90 Chaperone Cascade in Protein Folding. Trends Cell Biol 2019;29:164-77. [Crossref] [PubMed]
40. Lee AS. Stress-induced translocation of the endoplasmic reticulum chaperone GRP78/BiP and its impact on human disease and therapy. Proc Natl Acad Sci U S A 2025;122:e2412246122. [Crossref] [PubMed]
41. Hetz C, Zhang K, Kaufman RJ. Mechanisms, regulation and functions of the unfolded protein response. Nat Rev Mol Cell Biol 2020;21:421-38. [Crossref] [PubMed]
42. Yang Y, Pei T, Liu C, et al. Glutamine metabolic competition drives immunosuppressive reprogramming of intratumour GPR109A(+) myeloid cells to promote liver cancer progression. Gut 2025;74:255-69. [Crossref] [PubMed]
43. Encarnación-Rosado J, Sohn ASW, Biancur DE, et al. Targeting pancreatic cancer metabolic dependencies through glutamine antagonism. Nat Cancer 2024;5:85-99. [Crossref] [PubMed]
44. Lemberg KM, Vornov JJ, Rais R, et al. We’re Not “DON” Yet: Optimal Dosing and Prodrug Delivery of 6-Diazo-5-oxo-L-norleucine. Mol Cancer Ther 2018;17:1824-32. [Crossref] [PubMed]
45. Yokoyama Y, Estok TM, Wild R. Sirpiglenastat (DRP-104) Induces Antitumor Efficacy through Direct, Broad Antagonism of Glutamine Metabolism and Stimulation of the Innate and Adaptive Immune Systems. Mol Cancer Ther 2022;21:1561-72. [Crossref] [PubMed]
46. Ma G, Jia H, Li Z, et al. Gefitinib Reverses PD-L1-Mediated Immunosuppression Induced by Long-term Glutamine Blockade in Bladder Cancer. Cancer Immunol Res 2025;13:66-83. [Crossref] [PubMed]
47. Li R, Wang K, Hou N, et al. Activating transcription factor 4-mediated upregulation of Heat Shock Protein Family A Member 4 promotes hepatocellular carcinoma progression through activation of Wnt/β-catenin/EMT signaling pathway. Int J Biol Macromol 2025;333:148850. [Crossref] [PubMed]
48. Wu X, Yang SY, Zhang YH, et al. Prognostic and immunological roles of heat shock protein A4 in lung adenocarcinoma. World J Clin Oncol 2024;15:45-61. [Crossref] [PubMed]