A single-cell atlas of bladder cancer unveils dynamic cellular composition and endothelial functional shifts during progression

Abstract

Purpose

Bladder cancer (BC) is characterized by high heterogeneity, with non-muscle-invasive (NMIBC) and muscle-invasive (MIBC) stages differing significantly in clinical behavior and outcomes. The transition from NMIBC to MIBC involves extensive tumor microenvironment (TME) remodeling, particularly in endothelial cells (ECs), which drive angiogenesis and modulate immune and extracellular matrix (ECM) interactions. However, the precise roles of ECs in this progression remain poorly defined.

Methods

Public single-cell RNA sequencing (scRNA-seq) datasets from 47 BC patients were analyzed to characterize endothelial cell heterogeneity and functional states across NMIBC and MIBC. Computational tools such as CellChat were applied to reconstruct cell–cell communication networks, focusing on pathways related to angiogenesis, immune crosstalk, and ECM remodeling.

Results

Twelve major cell types were identified, with endothelial cells exhibiting distinct transcriptional profiles between NMIBC and MIBC. NMIBC-associated ECs promoted adhesion and migration through HMGB1 and CXCL12 signaling. In contrast, MIBC was enriched in an ADAM10⁺ endothelial subset associated with vascular remodeling and activation of Wnt signaling via CTNNB1. Key ligand-receptor interactions highlighted the dynamic roles of ECs in TME modulation during BC progression.

Conclusions

This study reveals stage-specific endothelial cell phenotypes and signaling networks in BC. The identification of an MIBC-specific ADAM10⁺ endothelial subset underscores its potential role in driving tumor progression and highlights opportunities for stage-adapted vascular-targeted therapies. These findings advance our understanding of BC pathogenesis and provide the foundation for novel therapeutic strategies.

Supplementary Information

The online version contains supplementary material available at 10.1007/s12672-025-02297-6.

Introduction

Bladder cancer (BC) ranks as the tenth most common malignancy globally and represents a common urological malignancy, primarily presenting as urothelial carcinoma. It is characterized by high recurrence rates and a diverse spectrum of clinical outcomes. Current diagnostic methods including urine tests (such as cytology and biomarker assays), cystoscopy, and imaging techniques (e.g., CT, MRI), are critical for accurate staging and treatment decision-making. The prognosis and therapeutic approach for BC are largely dependent on the tumor’s stage, grade, and molecular characteristics. Notably, non-muscle invasive bladder cancer (NMIBC) constitutes approximately 75% of cases at initial diagnosis [1–4]. While Although NMIBC is typically confined to the mucosa or submucosa and, generally carries a favorable prognosis, with a 5-year survival rate of 90–95%, its clinical management remains challenging due to high recurrence rates (50–70%) and a 10–20% risk of progression to muscle-invasive bladder cancer (MIBC). In contrast, MIBC is characterized by aggressive clinical behavior, high metastatic potential, and poor survival outcomes, with 5-year survival rates typically around 50–60%. Treatment strategies for NMIBC typically involves transurethral resection and intravesical therapies, whereas MIBC requires more aggressive interventions such as radical cystectomy, chemotherapy, and occasionally radiotherapy. Despite significant progress in treatment, several clinical needs remain, including the need for non-invasive, sensitive biomarkers for early detection, strategies to reduce recurrence, and improved therapies for advanced and metastatic BC. The high recurrence and progression rates of NMIBC, alongside the therapeutic difficulties associated with MIBC and metastatic disease, underscore the urgent need for novel treatment strategies, particularly in the form of immunotherapy and targeted therapies. Therefore, a deeper understanding of the molecular and cellular mechanisms underlying the transition from NMIBC to MIBC is essential for improving patient stratification and developing effective therapeutic strategies [5, 6].

The progression from NMIBC to MIBC is driven by dynamic remodeling of the tumor microenvironment (TME), an intricate ecosystem comprising cancer cells, stromal cells, and vasculature. Among TME components, endothelial cells (ECs) play a central role in tumor angiogenesis, supporting vascular remodeling, immune cells recruitment, and extracellular matrix (ECM) interactions. This study identified distinct endothelial phenotypes associated with NMIBC and MIBC progression, including a terminally differentiated ADAM10⁺ EC subtype enriched in MIBC. This specialized EC population was found to contribute to vascular remodeling and pro-tumorigenic signaling via VEGF and ECM pathways, underscoring their critical role in driving tumor progression [7].

Angiogenesis, the formation of new blood vessels, emerges as a key event during BC progression. Dysregulated VEGF signaling and altered vascular permeability have been implicated in tumor growth and metastasis. However, the precise mechanisms by which ECs mediate NMIBC-MIBC transition remain poorly understood. In this study, scRNA-seq analysis revealed that ECs not only exhibited stage-specific functional adaptations but also actively shaped tumor evolution through ligand-receptor interactions. For instance, NMIBC-associated ECs promoted tumor adhesion and migration via HMGB1 and CXCL12 signaling, while MIBC-associated ADAM10⁺ ECs enhanced Wnt pathway activation through CTNNB1 interactions, highlighting their dynamic roles across disease stages [8, 9].

Recent advances in single-cell RNA sequencing (scRNA-seq) have enabled unprecedented resolution in dissecting TME heterogeneity. This study leveraged publicly curated scRNA-seq datasets to systematically analyze EC transcriptional landscapes in BC. Computational tools such as CellChat were employed to map intercellular communication networks, unveiling critical pathways that govern EC-tumor crosstalk and angiogenic remodeling [10].

By defining stage-specific EC phenotypes and signaling networks, this work provides novel insights into the vascular mechanisms underlying BC progression. These findings not only enhance our understanding of BC biology but also lay the groundwork for developing stage-adapted, vascular-targeted therapies aimed at improving patient outcomes in this highly heterogeneous disease.

Methods

Collection of scRNA sequencing cohort

Single-cell RNA sequencing data from 41 BC samples and 6 normal tissue samples of BC patients were obtained from three independent GEO (https://www.ncbi.nlm.nih.gov/geo/) cohorts, namely, GSE135337 (n = 8), GSE169379 (n = 26) and GSE222315 (n = 13) [11–13]. The information of samples was shown in Supplementary Table 1.

ScRNA-seq data processing

The gene expression matrix of each dataset was transformed into a Seurat object via the ‘CreateSeuratObject’ function of the Seurat package (v4.4.0) in R (v4.4.0) [14]. All Seurat objects with their sample labels were merged, resulting in a collective of 239,262 single cells. The Seurat object was converted into AnnData format to facilitate data manipulation using the Scanpy package (v1.10.2) within a Python environment (v3.9.19) [15]. The ‘normalize_total’ and ‘log1p’ functions from the preprocessing module were employed to transform the gene expression data, ensuring that the data was scaled appropriately for downstream analysis. Principal component analysis (PCA) was conducted to diminish the dimensionality of the dataset after finding the top 2000 highly variable genes. We integrated the data from different datasets by using harmony algorithm to remove batch effects. The main reasons for choosing Harmony included its ability to handle large datasets, its relatively short computational runtime, and its effectiveness in preserving cell type purity while correcting for batch effects. The batch effect was mitigated through the application of the 'harmony_integrate' function from the scanpy.external package. The ‘neighbors’ function with default parameters was executed to define neighbors, followed by the application of the Leiden algorithm for cellular clustering. For visualization and further analysis, Uniform Manifold Approximation and Projection (UMAP) was implemented to map the high-dimensional scRNA-seq data onto a two-dimensional plane. By leveraging the gene expression patterns of markers specific to different cell types on the UMAP visualization, the identities of each individual cell were ascertained. Epithelial cells from normal tissue were annotated as normal epithelial cells.

Differential gene expression analysis and functional enrichment analysis

Differential gene expression analysis was conducted by ‘rank_genes_groups’ function with ‘Wilcoxon’ method from tools module of the Scanpy package. The established criteria for identifying differentially expressed genes (DEGs) were as follows: a p-value threshold below 0.05, an absolute log2 fold change (log2 FC) value exceeding 0.585, and the base mean gene expression within the group above 0.5.

Functional enrichment analysis of the top 200 DEGs with the largest log2 FC was conducted via the Metascape platform [16]. The analysis encompassed multiple pathway categories, including GO molecular functions, GO biological processes, canonical pathways, reactome gene sets, KEGG pathway, BioCarta gene sets, and hallmark gene sets. A p value cutoff was established at 0.01 for the purpose of enrichment significance.

Transcription factor analysis

The analysis of transcription factors (TFs) was conducted using pySCENIC (v0.12.1) [17]. The required TF list and database files were sourced from the Aerts Lab’s resource page. We followed the SCENIC pipeline, which consisted of three key stages: first, GENIE3 was used to build a co-expression network linking TFs with their target genes; next, RcisTarget was applied to validate potential regulatory interactions between TFs and genes through DNA motif analysis; and finally, the activity of regulons in each cell was quantified by calculating the Area Under the Curve (AUC) with AUCell. To pinpoint TFs specific to certain cell types, we calculated the regulatory similarity score across different cell types using the ‘calcRSS’ function. The visualization of these cell type-specific TFs was facilitated by the ‘plotRSS’ function, with the ‘zThreshold’ parameter set to 1.5 and the ‘thr’ parameter to 0.1.

Gene set score calculation

Gene sets associated with metabolism processes were identified based on prior scientific publication [18]. The ‘AddModuleScore’ function of Seurat package was employed to assess the expression level of metabolic gene set. This method evaluated the gene set activity on a single-cell basis by averaging the expression of the genes within each set and then subtracting the average expression of a set of control genes. The ‘AddModuleScore’ function was configured with the ‘ctrl’ parameter set to 100, which corresponded to the random selection of 100 control genes for each bin. The resulting scores were then incorporated into the Seurat object’s metadata, allowing for subsequent visualization.

Cell communication

Cell communication is characterized as the interaction between known ligands produced by sender cells and receptors located on receiver cells, as analyzed utilizing the CellChat package (v1.6.1) and NicheNetR package (v2.2.0) within the R programming environment [19, 20].

CellChat object was created based on the expression matrix of scRNA-seq data. We identified over-expressed genes and interactions, and projected them onto a protein–protein interaction network. Subsequently, we computed communication probabilities and aggregated the network to prepare for further analysis, saving the CellChat object at various stages. When comparing the cell communication results of MIBC stage and NMIBC stage, the two interaction objects were initially combined using the ‘mergeCellChat’ function. Differential expression analysis of ligands and receptors was performed using the ‘identifyOverExpressedGenes’ function, and the regulated communications were selected by ‘subsetCommunication’ function with the threshold of log2 FC set to 0.585.

In the context of NicheNet analysis, a regulatory network that outlined the connections between receptors and their targets was established to forecast ligand-target interactions. The necessary datasets, including the ligand-receptor network, ligand-target matrix, and gene regulatory network, were available for download at https://zenodo.org/record/7074291/files. DEGs in the sender cells were selected as ligands, based on criteria such as a log2 FC greater than 0.585 and an adjusted p-value below 0.05. The ‘predict_ligand_activities’ function was utilized to identify ligands with the highest activities. The ‘get_weighted_ligand_target_links’ function was then applied to determine ligand-target associations, and the ‘prepare_ligand_target_visualization’ function was used to filter potential targets, setting a threshold at 0.6 for visualization.

Endothelial cell subtype identification

In the process of identifying endothelial cell subtypes, the batch effect was mitigated through the application of the 'harmony_integrate' function from the scanpy.external package. The Leiden algorithm effectively enabled the segregation of clusters. The subtypes were designated according to the signature marker genes of each cluster. Marker genes indicative of each subtype were corroborated with those documented in the literature [21].

Pseudotime trajectory inference

To characterize the potential phenotype shift process of the endothelial cell subtypes, pseudotime trajectory analysis was conducted utilizing Monocle2 (v2.32.0) [22]. The Monocle2 analysis was based on the DEGs across the groups, selected with criteria of log2 FC > 0.5 and adjusted P value < 0.05. Trajectories were constructed utilizing the ‘reduceDimension’ function through the ‘DDRTree’ method. The final trajectory path was visualized using the ‘plot_cell_trajectory’ function. The ‘BEAM’ (Branched Expression Analysis Modeling) function was used to identify genes whose expression patterns changed at specific branch points in cellular developmental trajectories. Top 50 significant genes were visualized by the ‘plot_genes_branched_heatmap’ function. The ‘plot_genes_in_pseudotime’ function was used to display the expression level of selected gene over pseudotime.

Survival analysis

Survival curve was plotted at the provided URL (http://gepia2.cancer-pku.cn/#survival). Dataset was selected as BLCA and the cutoff for group assignment was determined based on the quartile values. We conducted both univariate and multivariate Cox regression analyses using the TCGA-BLCA cohort to determine the association between ADAM10⁺ ECs and overall survival. We computed the expression ratio of ADAM10-positive EC marker genes to those of other EC subtypes and examined its correlation with overall survival. The Cox regression analysis was conducted by ‘coxph’ function of R package survival (github.com/therneau/survival).

Multiplexed immunofluorescence (MIF) analysis

BC tissue samples in FFPE blocks (2 × 2 × 1 cm, a 2 cm-wide zone centered on the tumor border) were obtained from 26 patients who had undergone radical cystectomy and were pathologically diagnosed with BC. These samples were included in the validation cohort. The paraffin blocks were sectioned into 4-μm thick slices for further analysis. Multiplex immunofluorescence staining was performed using a PANO 6-color IF kit (Panovue, Beijing, China). After dewaxing with xylene and rehydrating through a series of alcohol gradients, antigen retrieval was carried out by immersing the slides in Tris–EDTA buffer (pH 8.0) and heating in a microwave. The slides were then cooled to room temperature. To block endogenous antigens, 1% bovine serum albumin (BSA) was applied for 30 min at room temperature. The primary antibody was incubated for 1 h, followed by a 30-min incubation with the secondary antibody. Antibodies used are listed in Supplementary Table 2. Tyramide signal amplification was performed with fluorescent reagents (PPD 520, PPD 570, PPD 620, PPD 650; Panovue, Beijing; 1:100) for 10 min at room temperature. After staining, nuclear counterstaining was carried out using DAPI (1:100) for 5 min at room temperature. After each incubation step, slides were washed three times with TBST for 2 min. The stained slides were mounted using an anti-fade medium (P36971, ThermoFisher Scientific) and stored at 4℃ until imaging. Confocal microscopy (ZEISS, Germany) was used for scanning, and InForm software (PerkinElmer) was employed to eliminate autofluorescence and analyze the multispectral images.

Statistical analysis and data visualization

Matplotlib package (v3.8.4) and ggplot2 package (v3.5.1) were used to visualize the gene expression heatmaps and gene set score boxplot [23, 24]. Then we applied the ‘t_test’ function from the rstatix package (v0.7.2) to perform two-sample t-tests [25]. This function automatically corrected the p-value to control the type I error rate caused by multiple comparisons. Graph generation was performed in R (v4.4.0) and Python (v3.9.19).

Results

A single-cell atlas of bladder cancer revealed dynamic cellular composition during progression

To better understand the cellular diversity within BC and identify dominant cell types, we analyzed single-cell RNA sequencing data derived from the publicly available datasets GSE135337 (n = 8), GSE169379 (n = 26), and GSE222315 (n = 13). In this compositive dataset, forty-one BC patients at NMIBC or MIBC stages and 6 normal tissues were included. After removing batch effects, we analyzed 239,262 cells, with the dimensionality reduction technique and unsupervised clustering. 12 major cell types were annotated with classical markers, which included normal urothelium cells, basal/luminal tumor cells, endothelial cells (ECs), immune cells (such as T cells, macrophages, and B cells), fibroblasts, and other stromal components (Fig. 1a–d, Supplementary Fig. 1a).

Fig. 1.

Single-cell RNA sequencing and annotation displayed significant differences in the proportion of endothelial cells between NMIBC and MIBC. a Schematic overview of the design and workflow. The single-cell RNA sequencing (scRNA-seq, 47 samples) data were acquired from GSE135337, GSE169379, and GSE222315 datasets. These datasets were analyzed to reveal the differences in endothelial cells (ECs) between MIBC and NMIBC, and the communication signaling sent from ECs to tumor cells and microenvironment cells. (Created in BioRender. https://BioRender.com/z67r722). b Unified Manifold Approximation and Projection (UMAP) visualization of the scRNA-seq data from 47 samples of 3 GEO datasets. c UMAP visualization of the scRNA-seq data from these 47 samples showing the diverse cell types present in the BC tissue. d Dot plot illustrating the expression patterns of canonical marker genes of each identified cell type in the scRNA-seq data. e UMAP visualization of cell type distribution among normal tissues, NMIBC and MIBC. f Tumor-stage proportions of scRNA-seq data across all cell types, highlighting the difference in the proportion of endothelial cells between NMIBC and MIBC

To further assess the cellular and functional heterogeneity among different pathological stages, we performed cell abundance analysis and identified immune cells and ECs were significantly enriched in NMIBC compared to normal and MIBC tissues (Fig. 1e, f, Supplementary Fig. 1b).

Heterogeneity of endothelial cells in normal tissues and bladder cancer

To explore the cellular heterogeneity within ECs in normal tissues and BC, we extracted and analyzed the gene expression profiles from the datasets mentioned above. After performing quality control and normalization, we observed distinct expression profiles among ECs in normal tissues and various stages of BC. ECs in normal tissues exhibited a characteristic gene expression signature, with elevated expression levels of SNTG2, LIFR, ZBTB16, and FKBP5. In contrast, as the disease progressed to NMIBC, we identified significant upregulation of genes such as MRPS24, FKBP1 A, NME2, and MIF (Fig. 2a), reflecting the early molecular changes associated with tumorigenesis. Interestingly, in MIBC, the gene expression profile of ECs resembled that in normal tissues, with the addition of markers such as EYA2, FAM84A, and CYP4F8. This suggests an evolutionary path from normal tissues to NMIBC, and ultimately to MIBC, characterized by both shared and distinct molecular features (Fig. 2a). Furthermore, to investigate the progression of angiogenesis-related features, we analyzed the expression of VEGFA, PDGFD, ERG, and ANGPT2, key markers associated with tumor vascularization [26]. In normal tissues, these markers were expressed at low levels. However, in NMIBC, both ERG and ANGPT2 showed significant upregulation, and in MIBC, all four markers were highly expressed, indicating a progressive activation of angiogenic pathways as the cancer advanced (Fig. 2b).

Fig. 2.

Differences in gene expression and functional characteristics were found among ECs in normal tissues, NMIBC, and MIBC. a Dot plot illustrating the expression patterns of differentially expressed genes in ECs from the scRNA-seq data for normal tissues, NMIBC and MIBC. b Violin plot showing the expression of genes regulating angiogenesis and vascular permeability in ECs from normal tissues, NMIBC and MIBC. c Pathway enrichment analysis of ECs in normal tissues, NMIBC and MIBC. d Analysis of transcription factor (TF) activity of ECs in normal tissues, NMIBC and MIBC, quantified using the Regulon Specificity Score (RSS) and Z-score. e–h Heatmaps showing the expression of metabolism-related genes in ECs from normal tissues, NMIBC and MIBC. i Cumulative box plots showing the distribution of metabolism scores of ECs across normal tissues, NMIBC and MIBC. Stars indicate statistical significance based on t-tests. (****P ≤ 0.0001)

Next, we conducted pathway enrichment analysis, which revealed distinct pathway enrichments for ECs in each stage. ECs in normal tissues were predominantly enriched in pathways related to cell morphogenesis and the regulation of the MAPK cascade, indicating processes involved in cellular development and response to external stimuli. In NMIBC, ECs were enriched in pathways such as cytosolic tRNA aminoacylation, nucleosomal DNA binding, and the electron transport chain (OXPHOS) system in mitochondria, reflecting metabolic and transcriptional reprogramming. ECs in MIBC, on the other hand, showed enrichment in pathways related to cell adhesion molecule binding, β-catenin binding, neuronal system processes, and cell junction assembly, suggesting a shift toward cell migration, invasion, and enhanced cellular communication (Fig. 2c). These subtypes were driven by the activation of distinct transcription factors (TFs). In normal tissues, the key TFs included MEIS1 and GTF2IRD. In NMIBC, the predominant TFs were ATF4 and ZNF580, while in MIBC, HNF4G and THRB were the primary activators (Fig. 2d).

Given the highly active metabolic pathways observed in ECs in NMIBC, it is likely that metabolic reprogramming plays a crucial role in the transition from normal tissues to BC. To further investigate this, we performed a metabolic score analysis across ECs. As anticipated, NMIBC exhibited the highest metabolic score, while MIBC showed the lowest. Notably, SLC3A2 and BSG were identified as the most active transporters in NMIBC. In terms of cholesterol metabolism, FDFT1, IDI1, and SQLE were highly enriched in NMIBC. For cAMP signaling, RAMP2 and CALCRL showed elevated expression in NMIBC. Additionally, in lipid metabolism, MGLL and FABP5 were activated in NMIBC, with all of these markers exhibiting a significant decrease in MIBC (Fig. 2e–i).

Cellular interaction analysis revealed crosstalk between cancer cells and ECs promoted bladder cancer progression

We performed cell–cell communication analysis using CellChat to investigate the interactions among all identified cell types in BC. In the NMIBC stage, luminal tumor cells, as the predominant subtype, exhibited intensive interactions with fibroblasts and ECs. In contrast, in MIBC, basal tumor cells dominated and interacted primarily with myocytes, indicating distinct tumor microenvironment (TME) remodeling mechanisms between NMIBC and MIBC (Fig. 3a, b).

Fig. 3.

Comparison of cell communication in NMIBC and MIBC. a, b CellChat analysis illustrating the number and strength of interactions between different cell types in NMIBC and MIBC. c, d Heatmaps showing the differences in the number and strength of cell–cell interactions when comparing NMIBC to MIBC. e Dot plot illustrating the communication probability of up-regulated signaling when comparing NMIBC to MIBC. Ligands were expressed by ECs. Receptors were expressed by luminal tumor cells or mast cells. f, g Heatmap showing the PDGF and ANGPT signaling pathway network across all cell types in MIBC. h Dot plot illustrating the communication probability of PDGF and ANGPT signaling in MIBC. Ligands were expressed by ECs. Receptors were expressed by ECs, fibroblasts or myocytes

Specifically, during the early stages of BC progression, luminal tumor cells engaged in active crosstalk with stromal cells, including fibroblasts, as well as immune-related cells such as macrophages and mast cells. Interestingly, ECs emerged as key players, exhibiting markedly higher interaction strength and numbers of signaling pathways compared to other cell types in NMIBC (Fig. 3c, d). ECs not only maintained vascular functions but also played a crucial role in modulating immune activity within the TME.

To explore the molecular mechanisms underpinning EC-mediated modulation, we examined ligand-receptor pairs that were upregulated in NMIBC and MIBC. In NMIBC, prominent interactions were mediated by HSPG2 and HMGB1 ligands. HSPG2, as part of the extracellular matrix, interacts with integrins such as ITGB1 through its heparan sulfate chains, regulating cell adhesion and migration. Meanwhile, HMGB1 binds to receptors like SDC1 on tumor cells, exhibiting chemotactic properties linked to tumor initiation, growth, invasion, and lymphatic metastasis (Fig. 3e) [27]. Furthermore, ECs in NMIBC demonstrated increased interactions with mast cells, potentially influencing mast cell recruitment and polarization, thereby promoting BC progression and metastasis [28].

In contrast, EC interactions in MIBC were dominated by angiogenesis-related signaling pathways, particularly PDGF and ANGPT. PDGFD, a specific ligand for PDGFRβ, was notably enriched in EC-fibroblast and EC-myocyte communication, facilitating fibroblast and myocyte proliferation and migration [29]. Additionally, ANGPT2 interacted with integrins to modulate actin cytoskeleton dynamics and intercellular tight junctions, potentially influencing cell morphology, migration, and vascular stability (Fig. 3f–h) [30].

These results underscore the dynamic role of ECs in promoting BC progression, particularly in shaping the invasive potential of the tumor microenvironment during the transition from NMIBC to MIBC.

Subtype analysis of endothelial cells reveals the functional specialization of ADAM10⁺Endothelial subtype in MIBC

To gain deeper insights into the EC compartment in BC, we performed a detailed subtype analysis, uncovering striking differences between NMIBC and MIBC (Fig. 4a–c). In NMIBC, the endothelial population predominantly comprised classical subtypes, including arterial, capillary, and venous ECs, with characteristic gene expression patterns reflective of vascular homeostasis, such as PLVAP [31]. In contrast, MIBC was enriched with a distinct endothelial subtype marked by the high expression of ADAM10, a metalloprotease implicated in angiogenesis and vascular remodeling (Fig. 4a–f). The findings were also validated through the independent bulk RNA-seq data from TCGA-BLCA cohort (Supplementary Fig. 1c). Previous studies have demonstrated that ADAM10 is essential for processes such as Notch signaling modulation and the cleavage of key membrane-bound substrates, both of which are critical for endothelial cell function and vascular integrity [32].

Fig. 4.

Heterogeneity of ECs and the enrichment of ADAM10⁺ ECs in MIBC. a–c UMAP visualization of all EC clusters profiled by scRNA-seq across all samples categorized by subtypes and tumor stages. d Dot plot illustrating the expression patterns of differentially expressed genes from the scRNA-seq data for different EC subtypes. e, f Tumor-stage proportions of scRNA-seq data across different EC subtypes, highlighting the difference in the proportion of ADAM10⁺ ECs between NMIBC and MIBC. g Pseudo-temporal trajectory analysis of different EC subtypes. The root of the trajectory is positioned at the right top. h Branched Expression Analysis Modeling (BEAM) results showing the top 50 significant genes whose expression patterns change along the two differentiation trajectories of ECs. i The change of relative expression of FLT1 along the pseudo-temporal trajectory path

Further analysis revealed that ADAM10⁺ ECs occupied a terminal position in the differentiation trajectory, indicating their role as specialized, functionally mature endothelial cells (Fig. 4g). During the progression from NMIBC to MIBC, the differentiation trajectory of ECs was accompanied by dynamic changes in the expression of FLT1 (VEGFR1), a key receptor in vascular endothelial growth factor (VEGF) signaling (Fig. 4h, i). FLT1 plays a dual role in angiogenesis, mediating VEGF-induced vascular growth while also acting as a decoy receptor to regulate VEGF bioavailability [33]. The pronounced alteration of FLT1 expression during EC differentiation suggests that ADAM10⁺ ECs are deeply involved in the fine-tuning of angiogenic responses within the TME.

Taken together, our findings underscore the critical involvement of ADAM10⁺ ECs in advanced BC. By coupling angiogenesis with vascular remodeling, this endothelial subtype likely contributes to the establishment of a pro-tumorigenic vascular niche, which supports tumor growth, metastasis, and immune evasion. This unique EC population represents a promising target for therapeutic intervention aimed at disrupting angiogenic and vascular remodeling processes in MIBC.

Endothelial ligand–tumor target gene interactions highlight distinct mechanisms in NMIBC and MIBC progression

To directly investigate the role of ECs in tumor progression, we analyzed the ligand-target interactions between EC-derived ligands and tumor-associated target genes. This analysis revealed distinct regulatory mechanisms in NMIBC and MIBC, reflecting the functional diversity of endothelial subtypes during BC progression.

In NMIBC, capillary ECs exhibited high expression of the chemotactic ligand HMGB1 [34], while arterial ECs prominently expressed CXCL12, a member of the chemokine family [35]. These ligands were predicted to target tumor genes CCL2, CXCL8, and CXCR4, which are known to regulate tumor adhesion and migration (Fig. 5a) [36]. The heightened chemotactic interactions between ECs and tumor cells suggest a pivotal role for ECs in shaping the TME during early-stage disease, supporting tumor cell infiltration and immune modulation. These findings are consistent with earlier observations (Fig. 3), further substantiating the pro-tumorigenic influence of NMIBC-associated endothelial cells.

Fig. 5.

Cell communications between EC subtypes and tumor cells and the prognosis of ADAM10⁺ EC in BC. a, b Cell–cell communication analysis between EC subtypes as sender cells and tumor cells as receiver cells. The left panel shows a dot plot illustrating the expression patterns of ligands in each EC subtype. The heatmap analyzes the regulatory potential of prioritized ligands from EC subtypes and predicted target genes in tumor cells. The bottom panel shows a dot plot analyzing the expression levels of target genes in tumor cells. c, d Validation of spatial TME of ECs and cancer cells by mIF in the validation cohort. EPCAM stains tumor cells (green), while CD31, PLVAP and ADAM10 stain ECs (orange, red and purple), with DAPI staining nuclei (blue). Scale bar, 50 μm. e Kaplan–Meier survival analysis illustrating the association between ADAM10 expression level and patient survival outcomes. f Univariate Cox regression analyses using the TCGA-BLCA cohort to assess the association between the expression ratio of ADAM10-positive EC marker genes relative to those of other EC subtypes and overall survival. g Multivariate Cox regression analyses using the TCGA-BLCA cohort to evaluate the association between the expression ratio of ADAM10-positive EC marker genes relative to those of other EC subtypes, age, gender and overall survival

In contrast, in MIBC, the unique ADAM10⁺ EC subtype exhibited regulatory interactions targeting CTNNB1, a key mediator of the Wnt signaling pathway (Fig. 5b). CTNNB1 plays a critical role in maintaining β-catenin stability and functionality, which, in turn, governs the activation state of Wnt signaling [37]. This pathway is well-documented for its involvement in tumor growth, invasion, and metastasis. Interestingly, while the direct EC-tumor interactions in MIBC were relatively diminished compared to NMIBC, endothelial cells in MIBC appeared to shift their role toward promoting angiogenesis and indirectly regulating stromal cells such as fibroblasts and myocytes. This functional shift underscores the evolving role of ECs from chemotactic modulation in NMIBC to vascular remodeling and stromal regulation in MIBC.

The biological significance of these findings was validated by multiplex immunofluorescence staining, which confirmed the expression of ADAM10⁺ ECs in MIBC tissues (Fig. 5c, d). Additionally, high ADAM10 expression in endothelial cells was associated with poorer patient survival, emphasizing the clinical relevance of this endothelial subtype (Fig. 5e–g). Together, these results provide compelling evidence for the dual role of ECs in BC progression—acting as chemotactic modulators in NMIBC and as vascular remodelers in MIBC.

Discussion

BC progression is accompanied by intricate alterations in cellular composition and function within the tumor microenvironment (TME). By leveraging single-cell RNA sequencing data, this study provides a comprehensive atlas of cellular heterogeneity across non-muscle-invasive bladder cancer (NMIBC) and muscle-invasive bladder cancer (MIBC). Our findings highlight the dynamic interplay between epithelial and stromal cells, with endothelial cells (ECs) emerging as key regulators of tumor progression [38–40].

The results illustrate a stepwise evolution of gene expression profiles from normal urothelium to NMIBC and MIBC, marked by metabolic reprogramming and activation of tumor-promoting pathways. ECs in NMIBC exhibit enhanced metabolic activity, while ECs in MIBC adopt features associated with invasion and migration, aligning with the distinct clinical behaviors of these stages. These observations emphasize the molecular heterogeneity within bladder cancer, shedding light on potential therapeutic vulnerabilities in metabolic and adhesive pathways [7, 41].

ECs exhibited stage-specific functional adaptations, with classical arterial, venous, and capillary subtypes predominating in NMIBC and a distinct ADAM10⁺ EC subtype enriched in MIBC. ADAM10⁺ ECs were identified as terminally differentiated cells, actively remodeling the vascular niche via extracellular matrix organization and angiogenic signaling. These cells also demonstrated an upregulation of FLT1, highlighting their role in the fine-tuning of VEGF signaling. This specialized endothelial population represents a novel target for therapies aimed at disrupting vascular remodeling and tumor progression.

The ligand-receptor interactions uncovered in this study provide mechanistic insights into endothelial-tumor crosstalk. In NMIBC, ECs engaged with tumor cells through chemotactic interactions involving HMGB1 and CXCL12, promoting adhesion and migration. In contrast, MIBC-associated ADAM10⁺ ECs targeted CTNNB1, influencing Wnt signaling and underscoring their role in regulating stromal and angiogenic processes. The functional shift from direct chemotactic modulation in NMIBC to vascular remodeling in MIBC illustrates the adaptive role of ECs during tumor evolution.

The clinical relevance of these findings is underscored by the correlation between ADAM10⁺ EC abundance and poor patient survival. Targeting this endothelial subtype may provide a dual benefit of impairing angiogenesis and disrupting pro-tumorigenic signaling pathways. Future studies should explore the therapeutic potential of modulating EC subtypes and their interactions with tumor and stromal cells to arrest bladder cancer progression effectively.

Although the integration of multiple scRNA-seq datasets offers valuable insights, it is important to acknowledge the potential limitations associated with combining datasets from different sources. One key issue is the variation in sequencing depth across datasets, which could introduce biases in gene expression measurements. To address this, we applied normalization methods to standardize gene expression levels and reduce depth-related discrepancies. Additionally, differences in sequencing platforms could contribute to platform-specific biases. To minimize these effects, we employed preprocessing steps such as filtering low-quality cells and applying batch effect correction techniques. Lastly, variations in preprocessing pipelines, such as cell filtering and gene selection, were carefully considered and standardized across datasets to ensure consistency. Despite these efforts, it is important to note that dataset integration still carries inherent challenges, and residual heterogeneity may persist. Future work could focus on developing more refined methods to address these issues, further improving the accuracy and robustness of data integration in scRNA-seq studies.

Conclusion

This study establishes a high-resolution cellular atlas of bladder cancer, revealing dynamic transitions in endothelial specialization and their evolving roles in the TME. The insights gained here not only enhance our understanding of bladder cancer biology but also pave the way for novel therapeutic strategies targeting cellular and molecular drivers of disease progression.

Limitations

This study, while leveraging comprehensive public single-cell RNA sequencing (scRNA-seq) datasets, encounters several limitations. First, the availability and diversity of scRNA-seq data for bladder cancer are limited, potentially restricting the breadth of endothelial cell (EC) phenotypic variations that could be explored. The reliance on public datasets also introduces variability due to differences in sample collection, processing, and sequencing protocols, which may lead to potential batch effects despite our efforts to normalize and integrate the data. Additionally, while computational tools like CellChat were instrumental in reconstructing cell–cell communication networks, the findings largely remain predictive and would benefit from further validation through functional assays. The lack of in vitro or in vivo functional experiments to confirm the roles of identified EC subsets and signaling pathways limits the direct translation of our findings to clinical settings. Future research incorporating more diverse datasets and experimental validations is necessary to fully elucidate the functional implications of EC heterogeneity in bladder cancer progression.

Author contributions

All authors contributed extensively to the work presented in this paper. H.S., G.X., and Z.T. conceived and designed the study. W.X., Z.T, and Z.W. oversaw the overall direction of the project. H.S., Y.L. and X.H. collected public dataset. G.X. and Z.Y. analyzed dataset. H.S., Y.L., X.H., and Y.Z. obtained clinical pathological tissues. Q.S., H.L., Z.L., and Z.Y. optimized antibody staining for multiplex immunofluorescence and analyzed the resulting data. H.S., G.X., and Z.T. authored the manuscript, with all authors reviewing, revising, and approving the final version.

Funding

This work was supported by the National Key Research and Development Program of China (2021YFB3801000) and received additional funding from the National Natural Science Foundation of China (82171999), the Key Research and Development Program of Heilongjiang Province (2022ZX06C04, 2022ZXJ03C07), Spring Swallow Support Program for Young Talents from Heilongjiang Provincial Department of Science and Technology (CYQN24038), the Nn10 project at Harbin Medical University Cancer Hospital (Nn102024-01), the China Postdoctoral Science Foundation (2021M693828), the Heilongjiang Provincial Department of Human Resources and Social Security (LBH-Z22030), and the Harbin Medical University Cancer Hospital Haiyan Foundation (JJZD2024-24). Collectively, these funding sources enabled the comprehensive execution of this study.

Data availability

The datasets generated during and/or analysed during the current study are available GEO Database (https://www.ncbi.nlm.nih.gov/geo/) with GSE135337 (n = 8), GSE169379 (n = 26) and GSE222315 (n = 13), and the TCGA-BLCA cohort. The codes during the current study are available from the corresponding author on reasonable request.

Declarations

Ethics approval and consent to participate

The study was approved by the Ethics Review Committee of Harbin Medical University (Approval Number: KY2024-21).

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.