GLG1 in cancer-associated fibroblasts promotes bladder cancer growth while limiting tumor apoptosis

Abstract

Cancer-associated fibroblasts (CAFs) shape bladder cancer progression, but functionally relevant CAF-associated drivers remain insufficiently characterized. This study aimed to identify CAF-enriched genes with translational relevance and to determine whether GLG1 mediates tumor–stromal crosstalk in bladder cancer. Single-cell RNA sequencing from bladder cancer tissues was integrated with Mendelian randomization, Bayesian colocalization, correlation analyses, immunostaining of clinical specimens, CAF-conditioned medium assays, GLG1 knockdown/overexpression, TGF-β1/Smad2/3 assessment, and xenograft modeling. Single-cell profiling of nine bladder cancer samples retained 60,561 cells and identified fibroblast-enriched transcriptional programs. Among 960 fibroblast-enriched genes, Mendelian randomization identified 22 genes whose genetically predicted expression was associated with bladder cancer risk, and colocalization prioritized GLG1, HES4, and HSP90B1. GLG1 was selected for validation because it showed fibroblast-enriched expression, genetic support for bladder cancer association, and shared cis-eQTL/GWAS signals. GLG1 expression was higher in tumor tissues than in adjacent normal tissues, consistent with a stromal localization pattern, and GLG1 positivity was associated with tumor invasiveness. Tumor-derived TGF-β1/Smad2/3 signaling contributed to GLG1 upregulation in fibroblasts. CAF-conditioned medium enhanced EJ-1 cell viability, migration, and Matrigel invasion; these effects were weakened by GLG1 silencing and strengthened by GLG1 overexpression in CAFs. In xenografts, CAFs-shGLG1 reduced tumor growth and Ki-67 positivity and increased TUNEL-positive tumor cells relative to CAFs-shNC, whereas CAFs-OE-GLG1 showed the opposite Ki-67 and TUNEL staining pattern. CAF-associated GLG1 promotes bladder cancer cell viability, migration, Matrigel invasion, and xenograft growth while limiting tumor-cell apoptosis, likely by reinforcing tumor–stromal crosstalk. These findings support GLG1 as a stromal factor involved in bladder cancer progression and provide a basis for further evaluation of its biomarker potential and therapeutic relevance.

Supplementary information

The online version contains supplementary material available at https://doi.org/10.1007/s10495-026-02439-3.

Introduction

Bladder cancer is a common urinary tract malignancy with clinical heterogeneity, ranging from recurrent non-muscle-invasive tumors to muscle-invasive and metastatic disease associated with high morbidity and mortality [1–3]. Clinical decision-making relies heavily on pathological grade, invasion status, and risk stratification; however, these variables do not fully explain the mechanisms that contribute to cancer progression. Cancer progression is shaped by tumor cell alterations and by interactions among tumor microenvironment (TME) cells and extracellular matrix (ECM) components [4–9].

Cancer-associated fibroblasts (CAFs) are heterogeneous fibroblast-lineage cell populations that can acquire myofibroblast-like features, express markers such as α-SMA and FAP, and regulate paracrine signaling, tumor cell migration, angiogenesis, immune evasion, and therapeutic resistance [4–8, 10]. Their phenotypes and functions are influenced by inflammatory factors, tumor-derived signals, and cell–ECM interactions, including integrin signaling. In bladder cancer, single-cell transcriptomic studies have identified distinct fibroblast and stromal cell populations, including inflammatory CAFs expressing cytokines and chemokines, myofibroblast-like CAFs, and fibroblasts enriched in recurrent bladder cancer and expressing collagen genes, including COL1A2 and COL6A3. In addition, CDH12⁺ epithelial cells interact with fibroblasts through pathways involving TGFBR1, CD44, and several integrins [11–13]. Nevertheless, the mechanisms by which CAF-associated genes contribute to bladder cancer cell viability, proliferation, and migration remain unclear.

Single-cell RNA sequencing (scRNA-seq) provides the resolution needed to assign candidate gene expression to specific cell populations. However, cell-type expression patterns alone cannot establish whether a gene is relevant to disease activity. Mendelian randomization (MR) and Bayesian colocalization analyses can complement single-cell analysis by assessing the relationship between genetically predicted gene expression and cancer risk and whether expression quantitative trait locus (eQTL) signals and GWAS signals are consistent with a shared causal variant [14–18]. Therefore, combining single-cell transcriptomics, genetic analysis, clinical validation, and in vitro and in vivo assays may provide a stronger framework for identifying CAF-associated factors with biological and translational relevance.

Golgi-complex-localized glycoprotein 1 (GLG1), also known as MG-160 or E-selectin ligand-1, is a conserved membrane sialoglycoprotein that has been linked to fibroblast growth factor binding and E-selectin interactions [19–23]. Nonetheless, its role in bladder cancer remains incompletely understood. Our preliminary analyses identified GLG1 as a candidate fibroblast-associated gene because it was expressed in fibroblast/CAF clusters in the single-cell analysis, showed MR evidence of an association with bladder cancer risk, showed eQTL–GWAS colocalization, and was upregulated in fibroblasts exposed to bladder cancer cell-derived stimuli in vitro.

Transforming growth factor-β1 (TGF-β1) signaling is a major regulator of fibroblast activation and CAF-like phenotypes in the TME. Through Smad signaling, TGF-β can modulate fibroblast gene expression, promote ECM remodeling, and create a stromal microenvironment that facilitates tumor cell migration and invasion [24–26]. Based on the central role of TGF-β signaling in CAF biology and GLG1 expression in fibroblast/CAF clusters, we hypothesized that tumor TGF-β1/Smad signaling may modulate GLG1 during fibroblast activation and that GLG1 expression in CAFs may support malignant bladder cancer phenotypes.

This study aimed to determine whether GLG1 is associated with fibroblasts/CAFs in bladder cancer, whether tumor-derived TGF-β1/Smad2/3 signaling is implicated in GLG1 upregulation during fibroblast activation, and whether GLG1 expression in CAFs alters bladder cancer cell viability, invasion, migration, and xenograft tumor growth.

Methods

Study design

This study combined scRNA-seq, MR and Bayesian colocalization analysis, tissue validation, fibroblast stimulation experiments, gain- and loss-of-function experiments, cellular assays, and xenograft experiments. The in vitro and in vivo experimental arms were designed as sequential and complementary tests of the same CAF-GLG1 hypothesis. EJ-1 cells were used as recipient tumor cells in CAF-conditioned-medium assays to evaluate soluble-factor-dependent changes in viable metabolic activity, Matrigel invasion, and wound closure under controlled conditions. T24 cells were used in the subcutaneous xenograft model to assess the integrated effects of GLG1-modified CAFs on tumor growth, proliferative activity, and apoptosis. The study workflow is shown in Fig. 1.

Fig. 1.

Study design and working model. The study workflow included single-cell transcriptomics, MR and colocalization analyses, clinical validation, mechanistic validation, functional validation, and the construction of a working model. The proposed model suggests that tumor TGF-β1 activates fibroblasts through Smad2/3-related signaling, upregulates GLG1 in fibroblasts, and promotes CAF-mediated paracrine effects on bladder cancer cell viability, migration, invasion, proliferation, apoptosis, and xenograft growth. Abbreviations: CAF, cancer-associated fibroblasts; GLG1, Golgi glycoprotein 1; MR, Mendelian randomization; TGF-β1, transforming growth factor beta 1

Single-cell RNA-seq processing and annotation

Bladder cancer scRNA-seq data were obtained from the GEO dataset GSE222315 [27]. Nine bladder cancer samples were analyzed: GSM6919778, GSM6919780, GSM6919782, GSM6919784, GSM6919786, and GSM7898117–GSM7898120. The original gene expression matrices were processed using the Seurat R package [28]. For each sample, quality control was based on three metrics: the total UMI count per cell (nCount_RNA), the number of genes detected per cell (nFeature_RNA), and the proportion of reads assigned to mitochondrial genes (percent.mt). Quality control was performed using a median absolute deviation (MAD)-based filtering strategy. Cells that met the following criteria were retained: nFeature_RNA > 200, percent.mt ≤ sample-specific median + 3 MAD, nFeature_RNA ≤ sample-specific median + 3 MAD, and nCount_RNA ≤ sample-specific median + 3 MAD. The percentage of reads mapping to ribosomal genes was calculated and included as a covariate during scaling because ribosomal gene content can vary by cell type, tissue, platform, and biological state; therefore, no fixed threshold was applied. Using DoubletFinder v2.0.4, we identified and removed the doublets. After quality control filtering, doublet removal, and dataset integration, 60,561 high-quality cells were retained for analysis.

UMI counts were normalized with the NormalizeData function, followed by identification of highly variable genes (HVGs) using FindVariableFeatures. The 2000 most variable genes were then used for principal component analysis. Data were scaled using the ScaleData function. CellCycleScoring was used to assign each cell to a predicted cell-cycle phase based on S- and G2/M-phase marker expression and to assess cell-cycle phase-specific differences across clusters. Batch effects across samples were corrected using Harmony [29]. Cells were clustered using the Louvain algorithm at a resolution of 0.3, and clusters were visualized using Uniform Manifold Approximation and Projection (UMAP).

Cell types were annotated using SingleR based on the expression of cell-type-specific genes retrieved from the CellMarker and PanglaoDB databases and published studies, enabling the identification of major cell types. Fibroblast subpopulations were annotated based on the expression of ECM-associated fibroblast markers, including COL1A1, COL1A2, COL3A1, LUM, and DCN, and myofibroblast markers, including TAGLN and ACTA2 [30].

Dimensionality reduction plots, violin plots, scatter plots, dot plots, bar charts, pseudotime trajectories, and heatmaps were generated using R version 4.0, Seurat v4.3.0, GraphPad Prism 8.0, Monocle v2.30.0, DDRTree v0.1.5, and AUCell v1.24.0, as appropriate. Cell–cell interactions were visualized using CellChat.

Differential gene expression and gene screening

Differentially expressed genes (DEGs) were detected using the FindAllMarkers function with thresholds set at |log₂ fold change|> 0.25 and Benjamini–Hochberg-adjusted P < 0.05. DEGs between cell types were identified using the FindMarkers function with default parameters [28]. A total of 960 candidate DEGs were identified in fibroblast clusters and selected for genetic and functional analyses.

MR and Bayesian colocalization analysis

Genes in the fibroblast-associated DEG pool were evaluated using two-sample MR and Bayesian colocalization analysis to identify candidate genes with evidence of an association between genetically predicted gene expression and bladder cancer risk, as well as support for a shared genetic signal. Summary-level eQTL data were sourced from the eQTLGen consortium, whereas bladder cancer GWAS data were retrieved from the IEU OpenGWAS dataset ieu-b-4874, comprising 1279 cases and 372,016 controls of European ancestry.

SNPs associated with candidate gene expression at a locus-wide significance threshold of P < 1 × 10⁻⁵ were selected as instrumental variables. Linkage disequilibrium (LD) clumping was performed with an r² threshold of < 0.001 within a 10,000-kb window. Instrument strength was assessed using F statistics, and SNPs with F ≤ 10 were excluded to reduce weak-instrument bias.

The primary MR estimate was obtained using the inverse-variance weighted method. MR-Egger regression, weighted median, and weighted mode methods were used as complementary methods when sufficient SNP instruments were available [14–18]. For gene exposures with only one available SNP instrument, the Wald ratio method was used. MR estimates were generated on the log-odds scale and exponentiated for reporting as odds ratios (ORs) with 95% confidence intervals. OR < 1 indicated that higher genetically predicted gene expression was associated with lower odds of bladder cancer, whereas OR > 1 indicated higher odds of bladder cancer.

The robustness of MR estimates was assessed using leave-one-out sensitivity analysis. Cochran’s Q test was used to examine heterogeneity across instrumental variables, whereas horizontal pleiotropy was evaluated using the MR-Egger intercept test.

Bayesian colocalization analysis was performed to assess whether gene expression and bladder cancer risk shared the same causal variant within the genomic region of interest, rather than reflecting correlation through LD. Colocalization was tested within a 100-kb region around the index SNP. A posterior probability for H4 greater than 0.9 was considered supportive of colocalization [16].

Genes were prioritized for subsequent analysis if they showed MR evidence of association with bladder cancer and Bayesian colocalization support between the corresponding eQTL and bladder cancer GWAS signals. GLG1 was selected for experimental validation because it met three selection criteria: enriched expression in fibroblasts, evidence of an association between genetically predicted GLG1 expression and bladder cancer odds, and colocalization support between GLG1 expression and bladder cancer risk.

Cell–cell communication, pathway enrichment, correlation analysis, and pseudotime analysis

CellChat is used to infer cell–cell communication [31, 32]. Communication networks were summarized based on the number and strength of interactions, pathway-level signaling, and ligand–receptor interactions, with emphasis on fibroblast-centered incoming and outgoing signals.

Pathway analyses were performed using gene set enrichment analysis (GSEA) and gene set variation analysis (GSVA), as described previously [33, 34]. Associations between GLG1 expression and pathways or gene sets related to stromal remodeling, ECM remodeling, focal adhesion, TGF-β signaling, PI3K/AKT signaling, epithelial–mesenchymal transition, invasion, and immune signaling were examined. Correlation analyses were performed to examine associations between GLG1, HES4, and HSP90B1 expression and immunoregulatory gene expression. Pseudotime analysis was performed to evaluate expression patterns across fibroblast cell states [35].

Human tissue specimens and clinicopathological characteristics

The use of human bladder cancer tissues was reviewed and approved by the Institutional Review Board of Beijing Tiantan Hospital, Capital Medical University (Approval No. KY2024-005-01; approval date: January 25, 2024). All procedures involving human tissue specimens were conducted in accordance with the Declaration of Helsinki, and written informed consent was obtained from all participants. Of 35 tumor specimens, 30 had complete data on invasion status and histological grade and were classified as non-invasive (n = 13), invasive (n = 17), low-grade (n = 9), and high-grade (n = 21).

Immunohistochemistry (IHC)

GLG1 and TGF-β1 expression in bladder cancer tissues, including non-invasive and invasive tumors, and adjacent normal tissues, was evaluated by IHC, as described previously. IHC images were analyzed using ImageJ. Immunostaining was assessed using H-scores based on staining intensity and the percentage of positive cells. Staining intensity was scored on a four-point scale: 0, no staining; 1, weak staining; 2, moderate staining; and 3, strong staining. The percentages of GLG1-positive and TGF-β1-positive cells were recorded from 0 to 100%, and H-scores were calculated by multiplying the staining intensity score by the percentage of positive cells, yielding scores from 0 to 300 [36].

Immunofluorescence

Double immunofluorescence staining for GLG1 and α-SMA was performed to assess the spatial overlap between GLG1-positive cells and α-SMA-positive fibroblasts. Briefly, tissue sections were fixed in 4% formaldehyde for 20 min, permeabilized with 0.25% Triton X-100 for 10 min, and washed three times with PBS for 5 min each. Sections were blocked with 5% bovine serum albumin in PBS for 1 h at room temperature and incubated overnight at 4 °C with primary antibodies against GLG1 and α-SMA. After washing twice with PBS, sections were incubated with the appropriate fluorophore-conjugated secondary antibodies for 1 h at room temperature and washed twice with PBS. Nuclei were counterstained with DAPI. Sections were mounted with an anti-fade mounting medium and imaged using a fluorescence microscope. Images were analyzed using ImageJ. GLG1/α-SMA co-staining was assessed in merged images by quantifying yellow fluorescence intensity [37].

Stimulation of normal fibroblasts by cancer cell-conditioned medium (CM)

To evaluate whether cancer cell-derived soluble factors induced GLG1 expression in normal fibroblasts, normal fibroblasts were exposed to CM from BIU-87 or EJ-1 cells, and GLG1 protein expression was evaluated by western blotting at 12, 24, and 48 h. To assess whether TGF-β1 contributed to CM-induced GLG1 expression and fibroblast activation, normal fibroblasts were treated with BIU-87-CM or EJ-1-CM with or without anti-TGF-β1 antibody. After treatment, the expression of GLG1, FAP, α-SMA, p-Smad2/3, and Smad2/3 was measured by western blotting [38].

GLG1 transfection in CAFs and real-time quantitative polymerase chain reaction (RT-qPCR)

CAFs were seeded in 6-well plates at a density of 4.0 × 10⁵ cells/well 24 h before transfection. GLG1 knockdown was performed using three synthetic siRNA duplexes targeting human GLG1, designated si-GLG1-1, si-GLG1-2, and si-GLG1-3. A scrambled siRNA was used as the siRNA negative control. For GLG1 overexpression, the full-length human GLG1 coding sequence was cloned into the pcDNA3.1 vector and verified by sequencing. The resulting pcDNA3.1-GLG1 plasmid was used for overexpression, and the empty pcDNA3.1 vector was used as the overexpression negative control.

Transfection was performed using Lipo6000™ Transfection Reagent (Beyotime Biotechnology, C0526) according to the manufacturer’s six-well-plate protocol. Before transfection, each well was supplied with 2 mL of fresh serum-containing, antibiotic-free medium. For siRNA transfection, 100 pmol siRNA and 5 μL Lipo6000™ reagent were separately diluted in 125 μL serum-free medium. For plasmid transfection, 2.5 μg pcDNA3.1-GLG1 or empty pcDNA3.1 vector and 5 μL Lipo6000™ reagent were separately diluted in 125 μL serum-free medium. The diluted nucleic acid and Lipo6000™ solutions were each incubated at room temperature for 5 min, combined, gently mixed, and incubated for an additional 5 min. A total of 250 μL transfection complex was then added dropwise to each well. After 5–6 h, the transfection medium was replaced with fresh complete medium, and cells were harvested 48 h after transfection.

For RT-qPCR validation, the experimental groups included Control, si-GLG1 NC, si-GLG1-1, si-GLG1-2, si-GLG1-3, and OE-GLG1. Each group included three independent biological replicates. GLG1 knockdown and overexpression efficiencies were confirmed at the mRNA level by RT-qPCR. Among the three GLG1-targeting siRNAs, si-GLG1-1 produced the greatest reduction in GLG1 mRNA expression and was selected for subsequent functional assays. For functional assays, CAFs-si-GLG1 and CAFs-OE-GLG1 were compared with their corresponding negative-control groups, CAFs-si-GLG1 NC and CAFs-OE-GLG1 NC.

The siRNA duplex sequences were as follows: si-GLG1-1, sense 5′-UUGGUUAUCAUCUGUUUGCAU-3′ and antisense 5′-GCAAACAGAUGAUAACCAAGC-3′; si-GLG1-2, sense 5′-UCAUUGAGGCGCUUUUUGCAC-3′ and antisense 5′-GCAAAAAGCGCCUCAAUGACC-3′; and si-GLG1-3, sense 5′-AUUCAAAGCUCGAUCAAUGCG-3′ and antisense 5′-CAUUGAUCGAGCUUUGAAUGA-3′. The negative-control siRNA sequences were sense 5′-ACGUGACACGUUCGGAGAA-3′ and antisense 5′-UUCUCCGAACGUGUCACGU-3′.

Stable shRNA-mediated GLG1 knockdown in CAFs for xenograft experiments

For the xenograft loss-of-function experiment, stable GLG1-knockdown CAFs were generated using the pLKO.1-puro lentiviral vector (Addgene plasmid #8453). Three GLG1-targeting shRNAs were evaluated, and shGLG1-1, targeting 5′-TTGGTTATCATCTGTTTGCAT-3′, was selected because it produced the greatest reduction in GLG1 mRNA. Lentiviral particles were generated in HEK293T cells by cotransfecting the shRNA plasmid with psPAX2 (Addgene plasmid #12,260) and pMD2.G (Addgene plasmid #12,259) at a ratio of 6:3:1 using Lipofectamine 3000. Viral supernatants were collected at 48 and 72 h after transfection, passed through a 0.45-μm filter, and used to transduce CAFs at a multiplicity of infection of approximately 5–10 in the presence of 8 μg/mL polybrene. After 24 h, the viral medium was replaced with fresh medium. Cells were subsequently selected with 1.5 μg/mL puromycin for 5–7 days, with medium replacement every 2–3 days. A non-targeting shRNA was used as the negative control (shNC). Puromycin-resistant CAF populations showing at least a 70% reduction in GLG1 mRNA relative to shNC by RT-qPCR were used for the xenograft experiments.

Real-time quantitative polymerase chain reaction (RT-qPCR)

Total RNA was extracted from transfected CAFs using a total RNA extraction kit and TRnaZol reagent. RNA quantity and purity were assessed using a BioMATE 160 UV spectrophotometer. Genomic DNA was removed using 4× gDNA wiper Mix at 42 °C for 2 min. Reverse transcription was performed using HiScript II Reverse Transcriptase First Strand cDNA Synthesis Kit, with the reaction performed at 50 °C for 15 min and 85 °C for 5 s.

RT-qPCR was performed using ChamQ SYBR qPCR Master Mix on an ABI 7500 real-time PCR system. Each 20 μl reaction contained 10.0 μl 2× ChamQ SYBR qPCR Master Mix, 0.4 μl forward primer, 0.4 μl reverse primer, 2.0 μl cDNA template, and 7.2 μl ddH₂O. The cycling program was 95 °C for 30 s, followed by 40 cycles of 95 °C for 3–10 s and 60 °C for 10–30 s. Melting-curve analysis was performed using the default instrument program. ACTB was used as the internal reference gene, and relative GLG1 expression was calculated using the 2^−ΔΔCt method [39]. Each group included three biological replicates.

The primer sequences were as follows: ACTB forward, 5′-CATGTACGTTGCTATCCAGGC-3′; ACTB reverse, 5′-CTCCTTAATGTCACGCACGAT-3′; GLG1 forward, 5′-ATTACCAAGATGACGGCCAT-3′; and GLG1 reverse, 5′-ATCATCTCGGCAAGCAAAAT-3′.

Western blotting

Protein lysates from tumor tissues, adjacent normal tissues, and cultured cells were analyzed by western blotting. Frozen tissues were thawed, minced in 1.5-mL tubes, and lysed in 300 μL of cell lysis buffer (Beyotime, P0013) on ice for 45 min with intermittent mixing. Cultured cells were lysed using the same lysis buffer and processed under the same conditions. Lysates were centrifuged at 12,000 rpm for 15 min at 4 °C, and protein concentrations were measured using the Enhanced BCA Protein Assay Kit (Beyotime, P0010). Samples were mixed with 5× loading buffer and heated at 100 °C for 5 min. Proteins were separated by 12% SDS-PAGE, with 50 μg of protein per lane, and transferred to PVDF membranes (Millipore, IPVH00010) at 4 °C. Membranes were blocked with 5% skim milk in TBST for 2 h at room temperature and incubated overnight at 4 °C with primary antibodies against GLG1 (Abcam, Ab271182; 1:1000), TGF-β1, FAP, α-SMA, p-Smad2/3, Smad2/3, or β-actin (Proteintech, 66,009–1-Ig; 1:10,000). After washing with TBST, membranes were incubated with HRP-conjugated goat anti-rabbit IgG (H + L) (Beyotime, A0208; 1:1000) or goat anti-mouse IgG (H + L) for 2 h at 37 °C. Immunoreactive bands were detected using High-sig ECL Western Blotting Substrate (Tanon Science & Technology, ECL-180-501) and visualized using a chemiluminescence imaging system (Tanon 5260 Multi). Band intensities were quantified using ImageJ and normalized to β-actin, which served as the loading control.

CAF-derived conditioned medium and in vitro functional assays

CAF-derived conditioned medium (CAF-CM) was used to evaluate whether soluble factors from GLG1-silenced or GLG1-overexpressing CAFs affected EJ-1 bladder cancer cells. EJ-1 cells were treated with control medium, CAF-CM, or conditioned medium from CAFs-si-GLG1 NC, CAFs-si-GLG1, CAFs-OE-GLG1 NC, or CAFs-OE-GLG1. For the Transwell invasion assay, EJ-1 cells were seeded in the upper chamber of Matrigel-coated 8.0-μm Transwell inserts, and group-specific CAF-derived conditioned medium was added to the lower chamber.

CAF-CM

CAFs, CAFs-si-GLG1 NC, CAFs-si-GLG1, CAFs-OE-GLG1 NC, and CAFs-OE-GLG1 were seeded in 6-well plates at a density of 5.0 × 10⁵ cells/well and cultured in RPMI 1640 medium. After 48 h of conditioning, the culture media were collected and centrifuged at 2000 rpm for 10 min to remove suspended cells and cellular debris. The resulting supernatants were used as group-specific conditioned media for downstream assays. EJ-1 cells were treated with control medium, CAF-CM, or conditioned medium derived from CAFs-si-GLG1 NC, CAFs-si-GLG1, CAFs-OE-GLG1 NC, or CAFs-OE-GLG1.

CCK-8 assays

Cell viability was evaluated using the CCK-8 assay. EJ-1 cells were cultured in RPMI 1640 medium containing 10% fetal bovine serum and penicillin/streptomycin at 37 °C with 5% CO₂. At 80% confluence, cells were digested with 0.25% trypsin–EDTA, centrifuged, and resuspended to 4.0 × 10⁴ cells/mL. Cells were seeded at 100 μL/well into 96-well plates for 18 h. The medium was then replaced with CM, and cells were incubated for 24 h. CCK-8 reagent (Dojindo, CK04) was added at 10 μL per 100 μL of culture, and cells were incubated for 2–4 h at 37 °C with 5% CO₂. Absorbance was measured at 450 nm using a microplate reader (Multiskan FC, Thermo Fischer Scientific).

Transwell assays

Transwell invasion assays were performed using 24-well plates and 6.5-mm Transwell inserts with 8.0 μm polycarbonate membranes (Costar, 3422). Matrigel basement membrane matrix (BD, 356,234) was diluted to 1 mg/mL in serum-free medium and added to the upper chamber. EJ-1 cells were seeded into the upper chamber at 1.5 × 10⁴ cells/well in 200 μL of serum-free medium, and CM (800 μL) was added to the lower chamber. After 24 h, cells remaining on the upper surface of the membrane were removed using a sterile cotton swab. EJ-1 cells that traversed the membrane and attached to the lower surface of the membrane were fixed with methanol or 4% paraformaldehyde, stained with Giemsa or 2% crystal violet, imaged using an inverted microscope, and counted.

Wound-healing assays

To assess cell migration, EJ-1 cells were seeded in 6-well plates and cultured until approximately 90% confluent. A scratch was generated using a 1000-μL Eppendorf pipette tip. Cells were washed 1–2 times with PBS to remove detached cells and then treated with the corresponding CAF-derived conditioned medium according to the experimental grouping. Images were acquired at 0 h and 24 h using an inverted microscope (Olympus IX73). Wound closure was quantified as: migration rate = [(wound area at 0 h − wound area at 24 h) / wound area at 0 h] × 100%. Each group included three biological replicates.

Xenograft model

Animal experiments were reviewed and approved by the Beijing Maide Kangna Laboratory Animal Welfare and Ethics Committee (Approval No. MDKN-2024-125). All animal procedures were conducted in accordance with the approved animal-use protocol, institutional guidelines for laboratory animal care and use, the Guide for the Care and Use of Laboratory Animals, and the principles of replacement, reduction, and refinement. Male BALB/c nude mice aged 5–6 weeks were used to establish subcutaneous xenografts with T24 bladder cancer cells. The animals were divided into five groups (n = 6 per group) to receive T24 cells alone (control group), T24 cells mixed with sh-GLG1 CAFs, T24 cells mixed with sh-GLG1 NC CAFs, T24 cells mixed with GLG1-OE CAFs, or T24 cells mixed with GLG1-OE NC CAFs. Each mouse received 2.0 × 10⁶ T24 cells in 100 µL of serum-free medium or 2.0 × 10⁶ T24 cells together with 2.0 × 10⁶ CAFs in 100 µL of serum-free medium without Matrigel. The cell suspension was injected subcutaneously into the left flank. Tumors were harvested for gross imaging, tumor volume and weight measurements, as well as hematoxylin–eosin, Ki-67, and TUNEL staining. Tumor volume was measured every 3 days. One mouse in each group died before endpoint collection; therefore, terminal analyses included five mice per group.

Statistical analysis

Normally distributed continuous variables were compared using an independent two-sample t-test and were expressed as mean ± standard deviation. Non-normally distributed continuous variables were compared using the Mann–Whitney U test and were expressed as median and interquartile range. Data are presented as mean ± standard deviation unless otherwise indicated. Differences between three or more groups were evaluated using one-way or two-way analysis of variance followed by post hoc tests. P < 0.05 was considered statistically significant.

Results

Integrated analysis identified GLG1 for experimental validation

The study workflow is shown in Fig. 1. scRNA-seq analysis was used to characterize the bladder cancer microenvironment and identify fibroblast-associated genes for genetic analysis and functional validation.

Single-cell profiling localizes GLG1 to stromal fibroblasts/CAFs

Single-cell analysis retained 60,561 cells from nine bladder cancer samples after quality control, doublet removal, and data integration (Fig. 2A). UMAP visualization showed seven major cell types: epithelial cells, fibroblasts, endothelial cells, macrophages, T cells, B cells and mast cells (Fig. 2B). Cell-type annotations were supported by the expression of canonical marker genes (Fig. 2C).

Fig. 2.

Single-cell RNA-seq analysis identifies fibroblast-associated GLG1 expression in bladder cancer. A Single-cell RNA-seq data processing and quality control workflow for nine bladder cancer samples; 60,561 cells were retained after quality control, doublet removal, and data integration. B UMAP plot of major annotated cell types. C Canonical marker genes used for cell-type annotation. Dot size indicates the percentage of cells expressing the marker, and color indicates average expression. D UMAP plot of GLG1-expressing cells. E Expression of HES4, HSP90B1, and GLG1 across major cell types. F Top DEGs in fibroblasts, ranked by average log₂ fold change. Dot size indicates the fraction of fibroblasts expressing each gene (pct.1). The table shows the three candidate DEGs selected for Mendelian randomization and colocalization analyses, with their average log₂ fold change and expression fractions in fibroblasts (pct.1) and non-fibroblast cells (pct.2). Abbreviations: DEGs, differentially expressed genes; GLG1, Golgi glycoprotein 1; UMAP, Uniform Manifold Approximation and Projection

UMAP plots showed that GLG1 was expressed in multiple cell types, including fibroblasts (Fig. 2D). Dot plot analysis of GLG1, HES4 and HSP90B1 showed that GLG1 had higher average expression in fibroblasts than in most other major cell types (Fig. 2E). Fibroblast-enriched DEGs included ECM-associated fibroblast markers, such as COL1A1, COL1A2, COL3A1, LUM, and DCN, and myofibroblast markers, including TAGLN, ACTA2, MYL9, CALD1, and TPM2 (Fig. 2F). GLG1 was detectable in approximately 50% of fibroblasts in the displayed single-cell summary. These results established the basis for selecting GLG1 as a candidate fibroblast-associated gene.

MR and colocalization identify GLG1, HES4 and HSP90B1 among fibroblast-enriched genes

An analytical workflow was used to prioritize fibroblast-associated genes. Differential expression analysis identified 960 fibroblast-associated DEGs. After filtering based on fibroblast enrichment, effect size, eQTL availability, and quality control criteria, MR analysis identified 22 candidate DEGs with evidence of association between genetically predicted gene expression and bladder cancer risk. Among these, GLG1, HES4, and HSP90B1 showed evidence of colocalization between gene expression and bladder cancer risk and were selected for subsequent analysis (Fig. 3A). MR effect estimates for the 22 candidate DEGs are shown as odds ratios with 95% confidence intervals in Fig. 3B, with GLG1, HES4, and HSP90B1 highlighted as the genes selected for subsequent validation and functional analyses (Fig. 3B).

Fig. 3.

MR and colocalization analyses of fibroblast-enriched differentially expressed genes (DEGs). A Workflow for the identification of fibroblast-enriched DEGs, filtering genes for genetic analysis, followed by two-sample MR analysis based on cis-eQTL and GWAS summary statistics. B Forest plots of inverse-variance weighted estimates for 22 fibroblast-enriched DEGs associated with bladder cancer risk. GLG1, HES4 and HSP90B1 are highlighted. C Summary of inverse-variance weighted and weighted mean estimates for GLG1, HES4 and HSP90B1, including leave-one-out, heterogeneity, pleiotropy, and colocalization analysis. D MR analysis of GLG1. E Bayesian colocalization analysis at the GLG1 locus, showing SNP-level GWAS and eQTL signals and regional association plots; rs16943885 highlighted. Abbreviations: eQTL, expression quantitative trait locus; GLG1, Golgi glycoprotein 1; GWAS, genome-wide association study; MR, Mendelian randomization; OR, odds ratio; SNP, single-nucleotide polymorphism

For these three genes, summary panels showed MR estimates together with sensitivity, heterogeneity, pleiotropy, colocalization, and reverse MR analyses (Fig. 3C). In the MR analysis, genetically predicted GLG1 and HES4 expression was associated with lower bladder cancer risk, whereas genetically predicted HSP90B1 expression was associated with higher bladder cancer risk. For GLG1, sensitivity analyses supported the robustness of the main estimate, and the leave-one-out analysis showed that the IVW estimate was not driven by any single SNP (Fig. 3D). A colocalization plot is shown in Fig. 3E. In this plot, GWAS −log10(P) values indicate SNP-level association with bladder cancer risk, whereas eQTL −log10(P) values indicate SNP-level association with GLG1 expression. At the GLG1 locus on chromosome 16, rs16943885 was highlighted in both the GLG1 eQTL plot and the bladder cancer GWAS plot, indicating that this genetic variant was associated with both GLG1 expression and bladder cancer risk at this locus.

GLG1 participates in fibroblast-centered communication, pathway signatures, immune regulation, and fibroblast state dynamics

CellChat analysis revealed extensive communication among cell types (Fig. 4A). Fibroblasts contributed to both the number and strength of intercellular interactions, supporting their active role in the bladder cancer communication network (Fig. 4A). Fibroblast-centered outgoing ligand–receptor analysis identified multiple fibroblast-to-recipient-cell communication pairs, including PTN-, MDK-, MIF-, VEGFB/PGF-, NAMPT-, ANGPT2-, GRN-, and ANXA1-related interactions (Fig. 4B). For fibroblast-to-endothelial-cell communication, the outgoing ligand–receptor plot showed several candidate axes, including VEGFB–VEGFR1, PGF–VEGFR1, NAMPT–INSR, NAMPT–(ITGA5+ITGB1), MIF–ACKR3, MIF–(CD74+CXCR4), MDK–NCL, MDK–(ITGA6+ITGB1), ANGPT2–TEK, and ANGPT2–(ITGA5+ITGB1). MIF–(CD74+CXCR4) was also detected in fibroblast interactions with T cells, macrophages, and B cells, whereas MIF–(CD74+CD44) was detected in fibroblast interactions with T cells, macrophages, and B cells. The GAS signaling pathway network shown in Fig. 4B represents a pathway-level CellChat summary involving fibroblasts and other cell types, including endothelial and epithelial compartments. Therefore, we interpreted the GAS result as pathway-level evidence of fibroblast-associated signaling rather than direct evidence that fibroblast–endothelial communication was mediated by a single GAS-related ligand–receptor pair.

Fig. 4.

Fibroblast-centered cell–cell communication, GLG1-associated pathway signatures, immune cell abundance, and fibroblast state dynamics. A Cell–cell communication among major cell types. B Fibroblast outgoing ligand–receptor pairs and GAS signaling pathway network. C GLG1-associated pathway analyses, including GSEA curves, gene set scores in between GLG1-high and GLG1-low groups, and pathway-level gene network visualization. D CIBERSORT-based relative abundance of immune cell subsets in normal and tumor samples and a heatmap showing correlations between GLG1, HES4, and HSP90B1 expression and immune cell abundance. E Fibroblast pseudotime trajectories colored by pseudotime and state. F GLG1 expression across pseudotime and state. Abbreviations: CIBERSORT, Cell-type Identification By Estimating Relative Subsets Of RNA Transcripts; GAS, growth arrest-specific; GLG1, Golgi glycoprotein 1; GSEA, gene set enrichment analysis; GSVA, gene set variation analysis

Pathway enrichment analysis showed that GLG1 expression was associated with chemokine signaling, JAK–STAT signaling, and PI3K–Akt signaling (Fig. 4C). GSVA showed that GLG1-high samples had higher scores for multiple tumor–stromal and cancer-related programs, including epithelial–mesenchymal transition, complement, apical junction, coagulation, TGF-β signaling, IL2–STAT5 signaling, KRAS signaling, hypoxia, inflammatory response, angiogenesis and PI3K–Akt–mTOR signaling. In contrast, GLG1-low samples showed higher scores for peroxisome, oxidative phosphorylation, fatty acid metabolism and xenobiotic metabolism-related programs (Fig. 4C). These results indicate that GLG1 expression is associated with stromal, inflammatory, and oncogenic pathway signatures.

CIBERSORT analysis showed differences in the estimated relative abundance of several immune cell subsets between normal and tumor samples (Fig. 4D). Tumor samples showed higher estimated abundance of M0 and M1 macrophages and lower estimated abundance of naïve B cells, resting mast cells, resting NK cells, resting CD4 memory T cells, and γδ T cells (Fig. 4D).

Correlation analysis showed associations between GLG1, HES4, and HSP90B1 expression and the estimated abundance of immune cells. GLG1 expression was positively correlated with M0, M1 and M2 macrophages and resting CD4 memory T cells and negatively correlated with memory B cells, activated dendritic cells, plasma cells, naïve CD4 T cells, CD8 T cells, and regulatory T cells. HES4 was positively correlated with regulatory T cells and negatively correlated with resting CD4 memory T cells. HSP90B1 was positively correlated with M0 and M1 macrophages, resting NK cells, and activated CD4 memory T cells, and negatively correlated with B-cell subsets, activated dendritic cells, resting mast cells, resting CD4 memory T cells, and regulatory T cells (Fig. 4D).

Pseudotime analysis showed that fibroblasts occupied distinct trajectory states across pseudotime (Fig. 4E). GLG1 expression displayed a state-associated distribution with modest variation along pseudotime (Fig. 4F).

Immunostaining analysis supports GLG1 enrichment in bladder cancer stroma

Bladder cancer tissues were classified according to clinicopathological characteristics (invasion status and histological grade) (Fig. 5A). Among 35 specimens, 30 had complete information on invasion status and histological grade and were classified as non-invasive (n = 13), invasive (n = 17), low-grade (n = 9) or high-grade (n = 21) (Fig. 5A).

Fig. 5.

Tissue validation of GLG1 and TGF-β1 expression and GLG1/α-SMA co-staining in bladder cancer tissues. A Classification of bladder cancer tissues according to invasion status and histological grade. B Immunohistochemical staining of GLG1 in tumor and adjacent normal tissues. C Percentage of GLG1-positive cells in tumor and adjacent normal tissues. D Immunohistochemical staining of TGF-β1 in tumor and normal adjacent tissues. E Percentage of TGF-β1-positive cells in tumor and adjacent normal tissues. F Immunofluorescence staining of GLG1 and α-SMA in tumor and adjacent normal tissues. G Quantification of GLG1/α-SMA co-staining. H Western blot analysis and quantification of GLG1 protein expression in tumor and adjacent normal tissues. Abbreviations: GLG1, Golgi glycoprotein 1; TGF-β1, transforming growth factor beta 1

IHC showed weaker GLG1 staining in normal tissues and stronger staining in cancer tissues, particularly invasive tumors (Fig. 5B). The percentage of GLG1-positive cells was higher in invasive tumors than in adjacent normal tissues and non-invasive tumors (Fig. 5C). IHC showed stronger TGF-β1 staining in tumor tissues than in adjacent normal tissues. The percentage of TGF-β1-positive cells was significantly higher in tumor tissues than in normal tissues (Fig. 5D, E).

Double immunofluorescence showed the absence of GLG1-positive cells in normal tissues. GLG1-positive cells were found in the stroma of non-invasive tumor tissues and co-stained with α-SMA-positive cells in invasive tumor tissues (Fig. 5F, G). Consistent with this finding, western blots showed higher GLG1 protein expression in tumor tissues than in adjacent normal tissues (Fig. 5H). These results suggest that GLG1 is associated with tumor invasiveness.

Tumor cell-CM induces fibroblast activation and GLG1 upregulation partly through TGF-β1/Smad2/3 signaling

We next tested whether bladder cancer cell-derived soluble factors induced fibroblast activation and GLG1 upregulation (Fig. 6A). Western blots of cell lysates showed relatively higher TGF-β1 protein expression in BIU-87 and EJ-1 cells than in CAFs and NFs, supporting the use of BIU-87- and EJ-1-CM for fibroblast stimulation (Fig. 6B). BIU-87- and EJ-1-conditioned medium-induced GLG1 protein expression in fibroblasts in a time-dependent manner (Fig. 6C).

Fig. 6.

TGF-β1 neutralization attenuates GLG1 upregulation and fibroblast activation induced by bladder cancer cell-conditioned medium. A Schematic overview of bladder cancer cell-conditioned medium exposure, TGF-β1 neutralization, Smad2/3 signaling, fibroblast activation, and GLG1 upregulation. B Western blot analysis of TGF-β1 protein expression in BIU-87, EJ-1, HBMEC-2, RT-112, SW1170, CAFs, and NFs. C Western blot analysis and quantification of GLG1 expression in NFs exposed to BIU-87- or EJ-1-conditioned medium for 12, 24, and 48 h. D Anti-TGF-β1 antibody attenuated BIU-87- or EJ-1-conditioned medium-induced GLG1 expression in NFs. E Western blot analysis and quantification of FAP, α-SMA, p-Smad2/3, and Smad2/3 in NFs exposed to BIU-87- or EJ-1-conditioned medium with or without anti-TGF-β1 antibody. F Model showing that TGF-β1 neutralization partially attenuates cancer cell-conditioned medium-induced GLG1 upregulation and fibroblast activation. Abbreviations: CAFs, cancer-associated fibroblasts; FAP, fibroblast activation protein; GLG1, Golgi glycoprotein 1; NFs, normal fibroblasts; p-Smad2/3, phosphorylated Smad2/3; TGF-β1, transforming growth factor beta 1; α-SMA, alpha-smooth muscle actin

Anti-TGF-β1 antibody partially attenuated BIU-87- and EJ-1-CM-induced GLG1 upregulation, with stronger attenuation at 5 μg/mL than at 1 μg/mL (Fig. 6D). Western blots showed that tumor cell-CM increased FAP, α-SMA, and p-Smad2/3 expression, whereas total Smad2/3 remained unchanged. Anti-TGF-β1 treatment attenuated the expression of FAP, α-SMA, and p-Smad2/3 (Fig. 6E). These results suggest that tumor TGF-β1/Smad2/3 signaling contributes to fibroblast activation and GLG1 upregulation (Fig. 6F).

CAF-associated GLG1 promotes EJ-1 cell viability, Matrigel invasion, and wound closure in vitro

RT-qPCR confirmed GLG1 knockdown and overexpression in CAFs. Among the three siRNAs, si-GLG1-1 produced the strongest knockdown and was used in functional assays, whereas GLG1 overexpression markedly increased GLG1 mRNA expression (Fig. 7A). CAF-CM was then used to evaluate the effects of secreted factors on EJ-1 cell viability, invasion, and migration (Fig. 7B).

Fig. 7.

CAF-associated GLG1 promotes EJ-1 cell viability, Matrigel invasion, and wound closure in vitro. A RT-qPCR analysis of the relative GLG1 mRNA expression in GLG1-silenced and GLG1-overexpressing CAFs. B Schematic of the Matrigel-coated 8.0 μm Transwell invasion assay using EJ-1 cells and CAF-conditioned medium. C EJ-1 cell viability after treatment with control medium, CAF-conditioned medium, or conditioned medium from GLG1-silenced or GLG1-overexpressing CAFs and respective negative controls. D, E Representative crystal violet-stained images and quantification of EJ-1 cells collected from the lower chamber after incubation with control medium, CAF-conditioned medium, or conditioned medium from GLG1-silenced, GLG1-overexpressing, or corresponding control CAFs. F, G Representative images at 0 h and 24 h and quantification of cell migration. Abbreviations: CAFs, cancer-associated fibroblasts; CCK-8, Cell Counting Kit-8; GLG1, Golgi glycoprotein 1; OE, overexpression; RT-qPCR, reverse-transcription quantitative polymerase chain reaction

CCK-8 assays showed that CAF-CM increased EJ-1 cell viability compared with control medium. CM from GLG1-silenced CAFs reduced this effect relative to CAFs-siGLG1 NC, whereas CM from GLG1-OE CAFs increased cell viability relative to CAFs-OE-GLG1 NC (Fig. 7C).

In Transwell invasion assays, EJ-1 cell invasion was significantly greater after incubation with CAF-CM than with control medium. GLG1 silencing decreased invasion, whereas GLG1 overexpression increased invasion compared with their respective controls (Fig. 7D, E). CAF-CM increased wound closure, GLG1 silencing attenuated this effect, and GLG1 overexpression enhanced this effect (Fig. 7F, G). These results suggest that GLG1 mediates the effects of CAF-derived soluble factors on EJ-1 cell viability, invasion, and migration.

Having shown that GLG1 manipulation in CAFs altered conditioned-medium effects on EJ-1 cells in vitro, we next evaluated whether GLG1-modified CAFs affected tumor growth, proliferation, and apoptosis-associated TUNEL positivity in the T24 xenograft model.

CAF-associated GLG1 promotes xenograft tumor growth in vivo

In vivo experiments were performed to evaluate the effects of GLG1-silenced or GLG1-OE CAFs on xenograft tumor growth. Xenograft groups, T24 cell/CAF cell ratios, and endpoints are shown in Fig. 8A.

Fig. 8.

CAF-associated GLG1 promotes xenograft tumor growth in vivo. A Schematic of the experiments, including the xenograft groups, T24 cell/CAF ratios, group allocation, tumor monitoring, and endpoint analyses. B Gross images of tumors. C Tumor growth measured every 3 days. D Tumor weights. E Representative hematoxylin–eosin staining of xenograft tumors. F Ki-67 staining and quantification of Ki-67-positive cells. G TUNEL staining and quantification of TUNEL-positive cells. H Working model in which tumor TGF-β1/Smad2/3 signaling contributes to fibroblast activation and GLG1 upregulation in CAFs, and GLG1-expressing CAFs promote bladder cancer cell proliferation, invasion, and migration and reduce tumor cell apoptosis

Tumors were smaller in the CAFs-shGLG1 group than in the CAFs-shGLG1 NC group (Fig. 8B). Tumor volume was lower in the CAFs-shGLG1 group than in the CAFs-shGLG1 NC group and higher in the CAFs-OE-GLG1 group than in the CAFs-OE-GLG1 NC group (Fig. 8C). Tumor weight was lower in the CAFs-shGLG1 group and higher in the CAFs-OE-GLG1 group compared with the respective NCs (Fig. 8D). Hematoxylin and eosin staining showed representative tumor morphology across groups (Fig. 8E). Ki-67 staining and the percentage of Ki-67-positive cells were lower in the CAFs-shGLG1 group and higher in the CAFs-OE-GLG1 group compared with the respective NCs (Fig. 8F). TUNEL staining and the percentage of TUNEL-positive cells were higher in the CAFs-shGLG1 group and lower in the CAFs-OE-GLG1 group compared with the respective NCs (Fig. 8G). Based on these results, we propose a model in which TGF-β1/Smad2/3 signaling contributes to fibroblast activation and GLG1 upregulation in CAFs, and GLG1-expressing CAFs promote bladder cancer cell proliferation, invasion, and migration, and reduce tumor cell apoptosis through CAF-derived soluble factors (Fig. 8H).

Discussion

Bladder cancer progression is strongly influenced by tumor–stroma interactions. Nonetheless, CAF-associated factors that mediate these interactions remain incompletely characterized. This study integrated single-cell transcriptomics, MR and colocalization analyses, tissue staining, functional assays, and xenograft modeling to evaluate GLG1 as a CAF-associated factor in bladder cancer.

scRNA-seq analysis showed that GLG1 was expressed in several cell types within the bladder cancer TME, with higher average expression in fibroblasts than in other annotated cell types. Differential expression analysis identified 960 fibroblast-associated DEGs. Of these, MR analysis identified 22 genes with evidence of association between genetically predicted expression and bladder cancer risk, and colocalization analysis further identified GLG1, HES4, and HSP90B1 as genes with evidence of a shared genetic effect between gene expression and bladder cancer risk. GLG1 was selected for tissue and functional validation because it combined fibroblast-enriched expression with evidence of a relationship between genetically predicted GLG1 expression and bladder cancer risk and colocalization between GLG1 expression and bladder cancer risk.

Previous studies have shown that GLG1 is involved in ligand binding and fibroblast growth factor signaling [19–21]. Nonetheless, the role of GLG1 in bladder cancer tumor–stroma interactions remains unclear. This study extends this evidence by showing that GLG1 is enriched in the bladder cancer stroma and that altered GLG1 expression in CAFs affects bladder cancer cell phenotypes.

CellChat analysis showed that fibroblasts interact with several other cell types in the TME. GLG1 expression was associated with chemokine, JAK–STAT, and PI3K–Akt signaling. Correlation analyses showed that GLG1 was associated with macrophages (M0, M1, and M2) and resting memory CD4+ T cells. GLG1 was also positively correlated with immune checkpoint and regulatory genes, including TGFB1, PDCD1LG2, IL10, and CSF1R. IHC and immunofluorescence analyses supported the tissue-level findings by showing that GLG1 expression was higher in tumor tissues than in normal tissues, with spatial overlap between GLG1-positive cells and α-SMA-positive fibroblasts. Together, these findings support GLG1 as a CAF-associated factor linked to bladder cancer progression.

In vitro, BIU-87- and EJ-1-CM increased GLG1 expression in fibroblasts, and this increase was attenuated by anti-TGF-β1 antibody treatment, suggesting that TGF-β1 contributes to GLG1 upregulation during fibroblast activation. Functional assays showed that GLG1 expression in CAFs affects bladder cancer cell viability, invasion, migration, proliferation, and apoptosis.

This study has several limitations. First, the number of tissue samples was small, limiting subgroup analysis by invasion status and histological grade. Second, the scRNA-seq analysis was based on a single public dataset, and GLG1 expression in fibroblasts/CAFs was not validated in an independent single-cell dataset. Third, in vitro assays used selected bladder cancer cell lines, and the subcutaneous xenograft model used only one bladder cancer cell line. Fourth, the scope of the computational analyses was broader than that of the experimental validation. CellChat and immune-correlation analyses suggested fibroblast-associated communication with endothelial and immune compartments; however, these findings are hypothesis-generating. The functional experiments were confined to the CAF–tumor-cell axis and did not directly examine endothelial activation, angiogenesis, macrophage polarization, or T-cell function. Therefore, the endothelial- and immune-related associations should not be interpreted as experimentally validated GLG1-dependent mechanisms. Future studies integrating independent single-cell cohorts, multicellular functional systems, and in vivo models that more fully retain vascular and immune components will be needed to determine whether GLG1-modified CAFs directly regulate these compartments in bladder cancer.

Conclusions

CAF-associated GLG1 supports bladder cancer progression in the models studied by enhancing tumor-cell viability, migration, Matrigel invasion, and xenograft growth while limiting tumor-cell apoptosis. Tumor-derived TGF-β1/Smad2/3 signaling contributes to fibroblast activation and GLG1 upregulation. These findings provide a basis for further evaluation of GLG1 as a stromal biomarker and for assessing its therapeutic relevance in bladder cancer.

Supplementary information

Below is the link to the electronic supplementary material.

Abbreviations

α-SMA

Alpha-smooth muscle actin

CCK-8

Cell counting kit-8

CAF

Cancer-associated fibroblast

CI

Confidence interval

CM

Conditioned medium

DEGs

Differentially expressed genes

ECM

Extracellular matrix

eQTL

Expression quantitative trait locus

FAP

Fibroblast activation protein

GLG1

Golgi glycoprotein 1

GSEA

Gene set enrichment analysis

GSVA

Gene set variation analysis

GWAS

Genome-wide association study

IHC

Immunohistochemistry

LD

Linkage disequilibrium

MR

Mendelian randomization

NC

Negative control

NF

Normal fibroblast

OE

Overexpressing

OR

Odds ratio

RT-qPCR

Reverse-transcription quantitative polymerase chain reaction

scRNA-seq

Single-cell RNA sequencing

shRNA

Short hairpin RNA

siRNA

Small interfering RNA

SNP

Single-nucleotide polymorphism

TGF-β1

Transforming growth factor-β1

TME

Tumor microenvironment

TUNEL

Terminal deoxynucleotidyl transferase dUTP nick-end labeling

UMAP

Uniform manifold approximation and projection

Author contributions

Conceptualization: Jingpeng Liu and Yong Zhang; Methodology: Jingpeng Liu, Baining Zhang, and Zhongbao Zhou; Formal analysis and investigation: Jingpeng Liu, Zhongbao Zhou, Baining Zhang, and Runze Liu; Data curation: Jingpeng Liu and Zhongbao Zhou; Visualization: Jingpeng Liu and Baining Zhang; Writing—original draft preparation: Jingpeng Liu and Baining Zhang; Writing—review and editing: Zhongbao Zhou, Baining Zhang, Runze Liu, and Yong Zhang; Resources: Runze Liu and Yong Zhang; Funding acquisition: Yong Zhang; Supervision: Yong Zhang. All authors read and approved the final manuscript.

Funding

This work was supported by the Beijing Tiantan Hospital, Capital Medical University Key Clinical Research Project (No. 2025-B21).

Data availability

The public scRNA-seq dataset used in this study can be accessed through the Gene Expression Omnibus under accession number GSE222315. Numerical data underlying the figures are included in the Supplementary Materials. Further data supporting the findings of this study can be obtained from the corresponding author upon reasonable request, in accordance with institutional and ethical requirements.

Declarations

Conflict of interest

The authors declare that they have no competing interests.

Ethics approval and consent to participate

The use of human bladder cancer tissue specimens was reviewed and approved by the Institutional Review Board of Beijing Tiantan Hospital, Capital Medical University (Approval No. KY2024-005-01; approval date: January 25, 2024). The human tissue component of this study was conducted in accordance with the Declaration of Helsinki. Written informed consent was obtained from all participants. All animal experiments were reviewed and approved by the Beijing Maide Kangna Laboratory Animal Welfare and Ethics Committee (Approval No. MDKN-2024-125). All animal procedures were conducted in accordance with the approved animal-use protocol, institutional guidelines for laboratory animal care and use, the Guide for the Care and Use of Laboratory Animals, and the principles of replacement, reduction, and refinement.

Consent for publication

Not applicable because the manuscript does not contain identifiable individual data or images.