Spatial mapping reveals luminal and basal bladder cancer lineage states with distinct immune contexts and treatment sensitivities, supporting spatially informed, lineage-directed strategies for muscle-invasive bladder cancer.
Abstract
Muscle-invasive bladder cancer (MIBC) is clinically heterogeneous, and current molecular subtyping does not capture the spatial organization of tumor states and microenvironmental context. In this study, we construct a spatial atlas of MIBC by integrating spatial transcriptomics from 22 tumors with matched bulk RNA and whole-exome sequencing data. We identify a continuous, spatially organized luminal-to-basal axis within individual tumors that is associated with greater chromosomal instability and transcriptional plasticity. Luminal tumor cores are enriched for FGFR3 and NECTIN4, whereas basal-like states localize toward invasive margins and are associated with elevated EGFR signaling, epithelial–mesenchymal transition, genomic instability, immune infiltration, and greater chemotherapy sensitivity. Spatial analyses further reveal lineage- and location-associated differences in tertiary lymphoid structure states. Pan-cohort analyses across more than 3,000 tumors confirm conserved FGFR3–EGFR lineage exclusivity and associated immune programs. Together, these findings define a spatial framework for understanding lineage states, immune architecture, and therapeutic vulnerabilities in MIBC.
Significance:
MIBC is clinically heterogeneous, yet current classification lacks spatial context for therapeutic guidance. This study establishes a spatial framework linking luminal and basal lineage states to distinct immune phenotypes and therapeutic vulnerabilities, providing a foundation for biomarker development and more precise, spatially informed, lineage-guided treatment strategies.
Introduction
Bladder cancer is among the 10 most common malignancies worldwide (1, 2), with muscle-invasive bladder cancer (MIBC) accounting for approximately 25% to 30% of new diagnosed cases (3, 4). Defined by invasion of tumor cells into the detrusor muscle (muscularis propria), MIBC is associated with a high risk of progression and metastatic spread. Although standard therapeutic regimens, including platinum-based neoadjuvant chemotherapy [e.g., gemcitabine (Gem)/cisplatin (Cis; GC) or methotrexate/vinblastine/doxorubicin/Cis (MVAC); refs. 5, 6], with or without immune checkpoint blockade (ICT; ref. 7), and the new antibody–drug conjugate (ADC) enfortumab vedotin (EV) plus ICT pembrolizumab (8), followed by radical cystectomy, are widely used, long-term outcomes remain suboptimal, with 5-year survival rates below 60% for most patients. This limited efficacy is partly attributable to the pronounced molecular and microenvironmental heterogeneity of MIBC, encompassing distinct tumor lineage states and variable immune infiltration, highlighting the urgent need for predictive biomarkers and therapeutic strategies tailored to tumor lineage identity and their spatially organized immune microenvironment.
Comprehensive transcriptomic profiling has demonstrated that MIBC comprises distinct molecular subtypes, most prominently luminal and basal-squamous lineages (9–11). Luminal tumors exhibit selective vulnerability to EV, an ADC targeting the luminal-associated surface molecule NECTIN4, which has been approved for patients with advanced bladder cancer (12, 13). In contrast, basal tumors exhibit increased epithelial–mesenchymal transition (EMT) and proliferative signatures (14, 15), features that have been associated with enhanced sensitivity to DNA-damaging agents (10, 11, 16, 17). Nonetheless, therapeutic responses remain heterogeneous even within molecular subtypes, indicating that current classification schemes are insufficient to fully predict treatment sensitivity. Accumulating evidence suggests that therapeutic response in MIBC is shaped not only by intrinsic lineage identity but also by dynamic epithelial plasticity, immune architecture, and microanatomic constraints within the tumor microenvironment (TME; refs. 18–20).
Recent work has overturned the view of luminal and basal subtypes as fixed and mutually exclusive entities, instead revealing the coexistence of distinct tumor cell states in MIBC. In this model, tumor cells can dynamically transition between luminal- and basal-like states in response to therapeutic exposure or environmental cues (21). The luminal-to-basal transition is marked by loss of luminal differentiation markers (e.g., GATA3 and FOXA1), induction of basal keratins (e.g., KRT5 and KRT14), and increased EGFR signaling and EMT-associated programs. Preclinical data demonstrate that the epithelial plasticity can generate mixed lineage states, with tumor cells coexpressing luminal and basal programs, consistent with dynamic interconversion rather than fixed subtype identity (22). Additional studies further show that disruption of luminal lineage regulators, including FOXA1 or epigenetic regulators such as KDM6A, can drive dedifferentiation toward basal-like phenotypes, frequently accompanied by immune evasion and resistance to targeted therapies (23, 24). Collectively, these observations indicate that lineage identity in MIBC is not static, but spatially and temporally regulated within the tumor ecosystem, highlighting the need for spatially resolved approaches to determine where, when, and how lineage reprogramming occurs in the TME.
Spatial transcriptomics (ST) preserves tissue morphology while enabling transcriptome-wide profiling, enabling direct interrogation of lineage topology, immune cell positioning, and tumor–microenvironment interactions (25–27). Spatial approaches have uncovered organized immune niches, invasive fronts, and tertiary lymphoid structures (TLS) across multiple cancer types (28–31), in which they correlate with treatment response and clinical outcome. Yet in bladder cancer, ST studies remain limited and largely confined to very small cohorts (32–36), limiting power to resolve subtype-specific immune organization or spatial determinants of therapeutic sensitivity.
To address this gap, we constructed a ST atlas of MIBC by integrating pretreatment spatial transcriptomes with matched whole-exome sequencing (WES) and bulk RNA sequencing (RNA-seq). This multiomic framework enables systematic characterization of the spatial architecture of tumor cell differentiation and lineage states, clonal architecture, and immune contexture. Using this approach, we investigate how epithelial lineage plasticity and immune spatial organization shape therapeutic response, as well as the location and maturation states of TLSs across distinct tumor compartments. We also utilize pan-cohort analyses across more than 3,000 tumors to confirm conservation of FGFR3–EGFR lineage exclusivity, associated genomic alterations, and immune programs. Collectively, our study reveals lineage-specific vulnerabilities and spatially organized immune niches in MIBC, establishing a spatially resolved framework to inform rational therapeutic strategies and precision oncology in bladder cancer.
Results
Spatial and Single-Cell Profiling Defines the Cellular Architecture of MIBC
We performed high-resolution ST profiling on 22 pretreatment MIBC specimens using the 10x Genomics Visium platform, complemented by matched bulk RNA-seq and WES for 21 tumors (Fig. 1A; Supplementary Table S1). To enable accurate spatial deconvolution, we additionally generated single-cell RNA-seq (scRNA-seq) profiles from 8 additional MIBC tumors, capturing both malignant and nonmalignant cellular states. To provide orthogonal validation at single-cell resolution, we further profiled 5 additional tumors using the Xenium Prime 5K ST platform. All patients subsequently received neoadjuvant chemotherapy. Based on the pathologic response at cystectomy, we classified 5 patients as responders (≤pT1) and 17 as nonresponders. TLSs were histologically identified in 9 tumors. After quality control (see Supplementary Methods), 70,575 high-quality spatially resolved spots were retained for downstream analyses (Fig. 1B).
Figure 1.
Spatial multiomic profiling defines the cellular and architectural landscape of MIBC. A, Schematic of the study design. We performed ST (10x Visium) on 22 pretreatment MIBC tumors (70,575 spots), with matched bulk WES and RNA-seq on 21 tumors. Eight additional tumors underwent scRNA-seq to establish a reference for spatial deconvolution. An additional 5 tumors were profiled using Xenium 5K ST for orthogonal validation at near single-cell resolution. B, Clinical, treatment, and sequencing metadata for each sample, including response status, pathologic stage, presence of TLSs, sequencing modality, and number of captured spatial spots and tumor spots (Tspots). C, Hierarchical tissue segmentation and pathology annotation. Spatial barcodes were manually classified into tumor, normal, or low-quality regions, followed by refinement into 10 major compartments (e.g., tumor cell, tumor stroma, TLS/TIL/LA, muscle, etc.). Sankey diagram summarizes region annotation and spot distribution across samples. N.Region, normal regions; T.Region, tumor regions. D, Representative pathology-informed spatial maps for 2 tumors (BC01 and BC13), illustrating intratumoral heterogeneity and spatial distribution of annotated regions. E, Uniform Manifold Approximation and Projection for Dimension Reduction (UMAP) visualization of single-cell transcriptomic data from 5 independent biopsy samples profiled using the Xenium 5K platform, colored by annotated cell types. F, Dot plot showing the expression of representative marker genes across annotated cell types, with dot size indicating the fraction of expressing cells and color intensity indicating mean expression levels. G, Heatmap showing average gene expression across spatial compartments.
To integrate histologic features with ST data, we implemented a hierarchical tissue annotation framework. Spatial sections were first segmented into tumor regions, normal regions, and low-quality or artifact-prone regions (others), with the latter excluded from further analysis. Fine-grained pathologist-assigned annotations were then consolidated into 10 major tissue compartments, including tumor cells, tumor stroma, immune-rich regions [TLSs/tumor-infiltrating lymphocytes (TIL)/lymphoid aggregates (LA)], vasculature, necrosis, urothelium, normal stroma, muscle, and others. This harmonized annotation layer served as the reference for all spatial analyses. Tumor epithelium and tumor-associated stroma accounted for the majority of spots (23,296 and 18,112, respectively), followed by normal stroma (7,585), necrosis (5,415), and muscle (4,004), whereas TLS-associated regions were less frequent (844 spots) but consistently detected across samples. Sample-level comparisons revealed substantial interpatient heterogeneity in spatial compartment composition, as illustrated by marked differences between representative tissues BC01 and BC13 (Fig. 1C and D; Supplementary Fig. S1).
scRNA-seq profiling identified 15 transcriptionally distinct cell populations, including epithelial cells, fibroblasts [inflammatory cancer-associated fibroblast (iCAF) and myofibroblastic cancer-associated fibroblast (mCAF)], endothelial cells, and diverse immune subsets such as T cells, B cells, NK cells, monocytes, macrophages, dendritic cells (DC), plasmacytoid dendritic cells (pDCs), mast cells, and LAMP3+ DCs (Supplementary Fig. S2A–S2C). Using this reference, we performed cell-type deconvolution of ST data with robust cell type decomposition (RCTD) (see Supplementary Methods; ref. 37), revealing substantial variation in epithelial and immune cell composition across tumors and tissue compartments (Supplementary Fig. S2D). Decomposition accuracy was supported by strong concordance between RCTD-inferred cell-type proportions and independent gene signature scores across individual spots, with epithelial cells, T cells, B cells, and plasma cell populations showing the highest agreement (Supplementary Fig. S2E and S2F).
To further resolve cellular composition at single-cell resolution, we performed Xenium 5K ST profiling and identified major cellular populations within tumor sections (Fig. 1E and F). Uniform Manifold Approximation and Projection for Dimension Reduction embedding of 185,149 cells revealed distinct clusters corresponding to malignant cells, which could be further subdivided into luminal, intermediate, and basal states, as well as multiple nonmalignant cell types, including immune populations (e.g., T cells, macrophages, and plasma cells) and stromal components [e.g., endothelial cells, cancer-associated fibroblast (CAF), and pericytes].
Cell-type composition differed markedly across histologically defined spatial compartments. Tumor regions were enriched for epithelial cells and displayed elevated GATA3 expression, consistent with luminal differentiation (Supplementary Fig. S2G). TLS-associated regions showed high expression of TLS signature genes and increased B- and T-cell fractions. Endothelial cells localized predominantly to vascular compartments, whereas fibroblasts were most abundant within muscle-associated regions. These findings reflect localized cellular specialization across tissue zones. Compartment-level transcriptional profiling further confirmed distinct molecular programs across spatial domains, including GATA3 and PPARG in tumor cells, COL1A1 and FAP in tumor stroma, VWF in vessel regions, CXCL13 in TLS/TIL/LA regions, and MYH11 and ACTG2 in muscle regions (Fig. 1G). Together, these analyses establish a transcriptionally and spatially resolved atlas of the MIBC TME, providing a foundation for dissecting lineage-specific differentiation states and immune organization in subsequent analyses.
Intratumoral Luminal–Basal Plasticity Is Associated with Progressive Copy-Number Variation Accumulation
To investigate intratumoral genomic heterogeneity within malignant compartments, we inferred copy-number variation (CNV) profiles across spatially annotated tumor regions using siCNV algorithm (38), leveraging histologically normal urothelial or nontumor spots as references when available (see Supplementary Methods). Spatial tumor regions were stratified into CNV-defined subclusters, which closely recapitulated expression-based clusters derived from transcriptomic profiling, indicating that CNV states are tightly coupled to distinct transcriptional programs (Fig. 2A–C).
Figure 2.
Spatially resolved CNV subclones define intratumoral lineage divergence in MIBC. A–C, Spatial analysis of 3 representative tumors (BC07, BC04, and BC13). Seurat clusters, inferred CNV states, CNV burden (CNVscore), pseudotime trajectory, differentiation state (CytoTRACE), and basal/luminal signature scores. Corresponding Uniform Manifold Approximation and Projection for Dimension Reduction visualizations (below). Right, CNV patterns across tumor subclones. Top, CNV similarity tree based on inferCNV profiles, reflecting relative similarity among subclones. Middle, CNV gain and loss frequencies across genomic regions calculated based on inferCNV profiles. Bottom, bulk WES-derived copy-number profiles across chromosomes, enabling cross-platform comparison of CNV patterns. D, Comparison of transcriptional signatures between CNV-defined basal and luminal subclones across tumors. CIS, carcinoma in situ; CLDN-low, claudin-low; CSC, cancer stem cell; sq-diff, squamous differentiation. Statistical significance is indicated as follows: **, P < 0.01. E, Gene set enrichment analysis comparing pathway activity in basal over luminal subclones. Node size reflects pathway size; edge thickness indicates gene overlap; color denotes normalized enrichment score (NES). Red, elevated in basal subclones; blue, elevated in luminal subclones. ns, not statistically significant. Oxphos, oxidative phosphorylation; Contr., control; Ref, reference; HCM, hypertrophic cardiomyopathy.
For instance, in tumor BC07, 2 spatial CNV subclusters (C1 and C2) were identified, representing transcriptionally and genomically distinct malignant states. One subcluster (C2) exhibited a luminal-like phenotype, characterized by higher luminal signature activity, lower basal signature activity, less CNVs, and lower aneuploid burden, with lower stemness signature and differentiation potential (estimated by CytoTRACE; ref. 39), and positioning at earlier stages along inferred differentiation trajectory (as determined by pseudotime analysis; ref. 40). In contrast, the other subcluster (C1) displayed elevated basal signature activity, increased aneuploid burden, and features consistent with a less differentiated and more plastic malignant state. Notably, the CNV landscape of C1 encompassed all alterations present in C2 along with additional private events, consistent with a clonal progression from a CNV-low, luminal-like state to a CNV-high, basal-like state (Fig. 2A). This potential evolutionary trajectory was independently corroborated by matched WES, which confirmed concordant CNV events across spatial clusters [Fig. 2A (right)]. These observations support the notion that increasing CNV burden accompanies more aggressive phenotypes, consistent with prior observations across multiple cancer types (41, 42).
Similar CNV-associated luminal-to-basal progression patterns were observed across additional tumors (Fig. 2B and C; Supplementary Fig. S3A and S3B), supporting the generalizability of this differentiation-linked genomic trajectory in MIBC. To evaluate the concordance between spatial RNA-inferred CNVs and WES-derived CNVs, we performed a systematic quantitative analysis across all samples. At the sample level, we computed the average inferCNV signal across spatial spots and compared it with matched bulk WES-derived CNV profiles. Across tumors, we observed consistent positive correlations (Pearson R = 0.42–0.77; median R = 0.63; Supplementary Fig. S4A). At the event level, we compared CNV gain and loss calls between the 2 platforms by using Matthews correlation coefficient (MCC) to quantify agreement and observed strong concordance, with median MCC values of 0.75 for CNV gains and 0.70 for CNV losses (Supplementary Fig. S4B). Importantly, we further compared CNV gain and loss events between luminal-like and basal-like tumor compartments within each tumor. We found that the majority of CNV events observed in luminal compartments were also present in basal compartments, indicating a substantial shared genomic backbone between these states. In contrast, basal-like compartments exhibited additional private CNV alterations that were not observed in luminal regions. Consistently, the proportion of shared CNV events was significantly higher in luminal compartments, whereas basal compartments showed increased numbers of state-specific alterations (Supplementary Fig. S4C).
Given that ST spots represent mixtures of malignant and nonmalignant cells, we next validated tumor-intrinsic CNV differences at single-cell resolution. Analysis of scRNA-seq data from 5 tumors containing both luminal- and basal-like malignant populations revealed a consistent increase in CNV burden in basal-like tumor cells compared with luminal counterparts, despite intertumor variability in absolute CNV levels (Supplementary Fig. S4D). These single-cell data mirror the spatial findings and reinforce a model in which luminal-to-basal transitions in MIBC are accompanied by progressive CNV accumulation and clonal diversification.
To further characterize molecular divergence between luminal and basal subclones, we examined canonical subtype–associated transcriptional programs in spatial regions where both states coexisted (17). Basal subclones exhibited increased activity of basal differentiation, cancer stem cell, EMT, and claudin-low signatures, alongside diminished luminal differentiation programs (Fig. 2D). Conversely, luminal subclones retained strong luminal identity, whereas cell cycle– and carcinoma in situ–associated programs did not differ significantly between states, implying that proliferative activity is largely uncoupled from lineage identity at the subclonal level. Pathway-level analysis further revealed enrichment of inflammatory and immune signaling pathways (e.g., IL6–JAK–STAT3 signaling, antigen presentation, and DNA sensing), extracellular matrix (ECM) interactions (ECM–receptor interaction), and metabolic remodeling in basal subclones (Fig. 2E). In contrast, luminal subclones were enriched for pathways associated with transcriptional and replicative activity (e.g., RNA polymerase complex and DNA replication) and organelle and metabolic homeostasis (e.g., peroxisome and inositol phosphate metabolism), consistent with previously reported luminal and basal phenotypes (10, 16, 17). Together, these results delineate a potential recurrent intratumoral trajectory in which malignant cells undergo luminal–basal divergence coupled to progressive genomic instability and functional reprogramming.
Spatially Resolved Tumor States Define a Continuous Luminal–Basal Differentiation Axis
To delineate spatially and transcriptionally distinct malignant states in MIBC, we clustered all pathology-annotated tumor spots based on their transcriptomic profiles. This analysis identified 4 recurrent spatial tumor states: luminal, intermediate (IM), basal-EMT (B-EMT), and basal-immune suppressive (B-IS), that together capture a spectrum of malignant differentiation programs (Fig. 3A; Supplementary Table S2). Luminal spots expressed canonical urothelial differentiation markers, including GATA3 and UPK2, whereas IM spots showed partial basal features with induction of EGFR, KRT6A, and TP63 (Fig. 3B; Supplementary Fig. S5A and S5B). B-EMT spots were characterized by enrichment of EMT- and stemness-related genes, such as CD44 and SPARC, consistent with a transcriptionally plastic basal state. In contrast, B-IS spots retained basal identity but exhibited concomitant activation of EMT- and immune-related programs, including IFNG, IL10, and CD274 (PD-L1), likely reflecting immune-enriched basal tumor niches.
Figure 3.
Spatially resolved differentiation states define a luminal-to-basal trajectory and highlight targetable lineage programs. A, Uniform Manifold Approximation and Projection for Dimension Reduction (UMAP) embedding of all tumor-designated spatial barcodes reveals 4 major transcriptional states: luminal (Lum), IM, B-EMT, and B-IS, ordered along a luminal-to-basal pseudotime axis. B, Expression of representative marker genes for each state. Luminal states express GATA3 and UPK2; IM regions show EGFR and KRT5 expression; B-EMT and B-IS express CD44 and SPARC. C, Pseudotime heatmap of selected genes across the inferred trajectory, based on pseudotime values computed using Monocle 3. D, Joint density plots of basal and luminal signature Z-scores across the 4 spatial states, illustrating gradual transitions. E, Distribution of tumor states along normalized pseudotime, in which pseudotime values derived from Monocle 3 are scaled to a 0–1 range and used to represent the trajectory axis. F, Spatial proximity matrix shows the topological relationship among tumor states. G, Pie charts of spatial state proportions across all tumors, compared with bulk RNA-seq–based BASE47 subtype classification (top). H, Gene set enrichment analysis highlights lineage-specific pathway activation, including FGFR signaling in luminal and inflammatory/EMT programs in basal states. Statistical significance is indicated as follows: *, P < 0.05. I, Venn diagram of differentially expressed genes across luminal- and basal-enriched tumors identifies potential lineage-specific therapeutic targets. FDA, genes with FDA-approved targeted therapies; SC, DEGs identified from scRNA-seq data; SP, DEGs identified from ST data. J, Representative spatial maps show mutually exclusive expression of EGFR and FGFR3 across subtype-defined tumor regions. K, Representative spatial maps from Xenium datasets (Xenium01) showing cell-type annotation, EGFR expression, and FGFR3 expression, with zoom-in views highlighting local details.
Pseudotime trajectory analysis reconstructed a continuous luminal-to-basal differentiation axis, with progressive loss of luminal markers (e.g., GATA3 and UPK3A), induction of EMT-associated regulators (e.g., TGFB1 and VIM), and subsequent upregulation of immune checkpoint molecules (e.g., CD274 and IFNG; Fig. 3C; Supplementary Fig. S5C). Consistent with this model, luminal signature activity decreased along the trajectory, whereas basal and EMT programs increased (Fig. 3D; Supplementary Fig. S5D; Supplementary Table S3). Mapping tumor states onto the inferred pseudotime trajectory revealed an ordered distribution, with luminal cells enriched at early pseudotime positions, followed sequentially by IM, B-EMT, and B-IS states (Fig. 3E). Spatial proximity analysis further supported this progression, showing that IM regions preferentially neighbored both luminal and B-EMT regions, whereas B-EMT regions were enriched adjacent to B-IS areas (Fig. 3F).
To relate these spatial tumor states to established bulk molecular subtypes, we analyzed matched bulk RNA-seq data from 19 tumors. Tumors were classified as bulk-luminal or bulk-basal subtypes using the BASE47 gene signature as previously described (Fig. 3G; ref. 43). Although bulk profiling captures a dominant luminal or basal identity, spatial deconvolution revealed that these classifications reflect the relative abundance of coexisting tumor states in each tumor rather than uniform lineage identity. For example, tumors with focal or extensive squamous differentiation (e.g., BC03, BC19, and BC22) tended to show higher proportions of B-EMT and B-IS states, whereas tumors classified as pure urothelial carcinoma tended to show relatively higher luminal state proportions (Supplementary Table S4). We next examined association between tumor-state composition and clinical variables using the ST data (Supplementary Fig. S5E). We found that the IM state tended to be more enriched in T2 tumors than in higher-stage tumors (T3+), consistent with a possible shift in tumor-state composition with disease progression, although this trend did not reach statistical significance (P = 0.067). We also observed differences in tumor-state composition by response group, although these did not reach statistical significance in this cohort (P = 0.12). We further demonstrated that compartment-fraction estimates are robust to variations in spot number and spatial sampling density (Supplementary Table S5).
We then validated these spatially defined tumor states using scRNA-seq of malignant epithelial cells. Unsupervised clustering recovered luminal, IM, and B-EMT populations that closely mirrored their spatial counterparts (Supplementary Fig. S6A and S6B; Supplementary Table S6). The B-IS state was not detected in the single-cell dataset, consistent with its dependence on tumor–immune admixture that is likely lost upon epithelial cell isolation. Pseudotime ordering of single cells recapitulated the luminal-to-basal trajectory, with progressive loss of luminal differentiation markers (e.g., GATA3, UPK3A, and KRT20) and gain of basal and EMT-associated programs (e.g., EGFR, KRT5, SPARC, and VIM), paralleling spatial pseudotime results (Supplementary Fig. S6C–S6F).
FGFR3 and EGFR Define Opposing Lineage Programs along the Luminal–Basal Axis
Pathway-level analyses across the 4 tumor states revealed distinct, lineage-associated signaling programs (Fig. 3H). The luminal state exhibited activation of FGFR2/FGFR3 and MET signaling pathways, consistent with epithelial maintenance and differentiated urothelial identity (16). IM states showed increased MAPK and TGFβ activity (44, 45), whereas B-EMT and B-IS states were characterized by enrichment of EMT, stemness, inflammation, and IL6–JAK–STAT3 signaling programs, consistent with dedifferentiated and immunomodulatory malignant phenotype (46). Notably, B-IS state displayed particularly strong immune signaling, further supporting immune enrichment within this basal tumor state.
To define lineage-specific therapeutic dependencies along this differentiation axis, we examined the spatial distribution of clinically actionable receptor tyrosine kinases. FGFR3 expression was selectively enriched in luminal state, whereas EGFR expression increased progressively toward basal and EMT-dominant states (Fig. 3I). Although both receptors belong to the same kinase family, FGFR3 and EGFR marked opposing poles of the luminal–basal differentiation spectrum, with FGFR3 associated with luminal lineage fidelity, more differentiated tumor state, and favorable clinical features, whereas EGFR linked to EMT, stemness, and poor prognosis.
Interestingly, EGFR-high regions further exhibited increased transcriptional entropy, consistent with elevated plasticity and transitional potential (Supplementary Fig. S7A). In parallel, analysis of pharmacogenomic data from the Cancer Cell Line Encyclopedia revealed that EGFR dependency correlated with basal program activity (Supplementary Fig. S7B). Spatial mapping in representative tumors confirmed mutually exclusive expression of FGFR3 and EGFR (Fig. 3J; Supplementary Fig. S7C).
EGFR expression is highest in the IM state and remains elevated in the B-EMT state while being consistently lower in luminal regions (Supplementary Fig. S7D). The B-IS state shows lower EGFR expression, which may seem inconsistent at first glance. However, we note that this state is characterized by significantly lower tumor purity (Supplementary Fig. S7E), consistent with a TME-rich niche in 10x Visium data rather than a purely tumor-intrinsic malignant state. Consistent with this interpretation, we do not detect a corresponding malignant cell population in the scRNA-seq data. To further resolve these observations, we performed orthogonal analysis using single-cell resolved Xenium Prime 5K ST (10x Genomics, 5,000-gene panel) in 5 independent biopsy samples. Consistent with Visium-based observations, tumor cells in the intermediate state exhibited higher EGFR expression compared with luminal cells (Fig. 1F). We next quantified the local TME context surrounding tumor cells. Basal-like tumor cells were preferentially located in regions enriched for nonmalignant cells, with significantly higher fractions of surrounding TME cells compared with luminal cells (Supplementary Fig. S7F). Moreover, tumor cells surrounded by a high proportion of TME cells (≥50% of neighboring cells) display reduced luminal features while retaining EGFR expression levels comparable with basal-like tumor cells (Supplementary Fig. S7G), supporting their basal-like phenotype in a TME-rich niche. Representative Xenium spatial maps further support this interpretation, showing EGFR enrichment in the IM and basal-like tumor regions, whereas FGFR3 expression is largely restricted to luminal regions (Fig. 3K; Supplementary Fig. S7H).
We next investigated the molecular basis of lineage-associated receptor tyrosine kinase (RTK) signaling. Analysis of a published human urothelial dataset (GSE184111; ref. 47) showed that FGFR3 and luminal markers such as UPK3A are enriched in umbrella cell layers, whereas EGFR and basal markers such as KRT13 are preferentially expressed in basal compartments (Supplementary Fig. S8A). This suggests that the FGFR3–luminal and EGFR–basal association may reflect conserved epithelial lineage programs. We also found that lineage-specific transcriptional regulators are strongly associated with these signaling patterns. Luminal transcription factors (GATA3, FOXA1, and PPARG; ref. 48) are enriched in luminal tumor states and positively associated with FGFR3 expression, whereas basal/EMT-associated states show increased expression of regulators such as TP63 (49), which are known to promote basal programs and may suppress luminal differentiation while favoring EGFR-associated signaling activity (Supplementary Fig. S8B). These observations are consistent with lineage-specific transcriptional networks underlying preferential FGFR3 versus EGFR pathway usage. We further validated these associations across multiple independent bulk cohorts (TCGA, ABACUS, and IMvigor), in which luminal subtypes consistently show higher expression of FGFR3 and luminal regulators, whereas basal-like subtypes exhibit higher EGFR expression (Supplementary Fig. S8C). Finally, to assess generalizability, we examined additional epithelial cancers with well-defined luminal–basal or classical–basal–like stratification, including breast (BRCA), pancreatic (PAAD), and prostate (PRAD) cancers from TCGA. Across these tumor types, EGFR expression was consistently enriched in basal-like states, whereas FGFR3 expression showed more context-dependent patterns, consistent with tissue-specific differences in lineage-defining RTK programs (Supplementary Fig. S8D).
Spatial Differentiation States Shape Tumor Architecture, Immune Organization, and Therapeutic Response
To investigate how spatial differentiation states influence tumor architecture and therapeutic response in MIBC, we stratified each tissue section into tumor and normal compartments and further subdivided them into zones based on distance from the tumor–normal boundary using a sliding-window approach (Fig. 4A; Supplementary Fig. S9). Tumor regions were classified as core, intermediate, or boundary, whereas adjacent normal tissue was grouped into distal, juxtatumor, and boundary zones (see Supplementary Methods).
Figure 4.
Spatially resolved lineage states shape tumor architecture and therapy response. A, Schematic of tumor–normal (T/N) compartmentalization and distance-based zoning using a sliding window approach. H&E, hematoxylin and eosin. B, Barplot showing the distribution of spatial tumor states [luminal (Lum), IM, B-EMT, and B-IS] across tumor core, intermediate, and boundary regions. C, Mean coherence of spatial clusters across tumor states in each sample. D, Spatial maps of representative tumors (BC01, BC07, and BC16) showing luminal/basal classification, pseudotime, and distance to tumor–normal boundary. Line plots depict changes in scores and proportions across zones. E, Permutation-based co-localization framework for quantifying tumor–immune spatial interactions. ES, enrichment score; cdf, cumulative distribution function; abs, absolute value. F, Co-localization enrichment scores and counts of immune cell types in luminal and basal regions across 20 tumors. cDC, conventional dendritic cell; pDC, plasmacytoid dendritic cell. G, Representative Xenium spatial maps showing co-localization of basal-like tumor cells with TME components. Insets highlight regions with dense TME infiltration surrounding basal-like tumor cells. H, Venn diagram showing overlap between tumor-upregulated genes and FDA-approved drug targets; 4 luminal-enriched genes (NECTIN4, TACSTD2, BCR, and EPCAM) are identified. I, Spatial distribution of NECTIN4 expression in 2 representative tumors (BC01 and BC16), overlaid with tumor–normal boundaries and subtype classification. J, Schematic and heatmap showing the effect of NECTIN4 overexpression in UM-UC-3 cells on drug response and transcriptomic profile, including upregulation of FGFR3 and suppression of immune/MHC-II genes. Ctr, control; OE, overexpression.
Across patients, a consistent spatial pattern emerged (Fig. 4B); luminal tumor spots were preferentially enriched in tumor cores, whereas basal-like populations (B-EMT and B-IS) accumulated toward the tumor periphery, particularly at the tumor–normal interface. From core to boundary, luminal state frequency gradually declined, whereas basal-like state increased, mirroring the luminal–basal differentiation axis defined in earlier analyses (Fig. 3). Spatial coherence analysis further revealed that the luminal state formed compact, spatially coherent clusters, whereas the basal state was more diffusely distributed, consistent with an infiltrative growth pattern (Fig. 4C). Spatial maps of representative tumors demonstrated a continuous radial gradient from the tumor center to the periphery, luminal scores decreased, basal scores and pseudotime with decreasing luminal scores, increasing basal scores and pseudotime values, and dominance of basal states at invasive edges (Fig. 4D; Supplementary Fig. S10A). We also found that the deconvoluted cell fractions clearly recapitulate the 2 major tumor lineages. As shown in Supplementary Fig. S10B, basal tumor fractions progressively increase along the spatial axis from tumor core to boundary, whereas luminal fractions show the opposite trend. Consistently, region-based comparisons (tumor core vs. boundary) demonstrate significant enrichment of luminal cells in core regions and basal cells at tumor boundaries. Furthermore, representative spatial maps illustrate that luminal tumor cells are predominantly localized in tumor cores, whereas basal tumor cells are enriched toward peripheral and invasive regions, consistent with the spatial organization observed in spot-level analyses (Supplementary Fig. S10C).
We next examined how these spatial tumor states interact with the immune microenvironment. Permutation-based co-localization analysis (see Supplementary Methods) across 20 evaluable tumors revealed significant enrichment of immune and stromal populations within basal-like regions, particularly B-EMT and B-IS states, including iCAFs, mCAFs, T cells, and monocytes (Fig. 4E and F). In contrast, luminal regions were consistently immune-excluded, consistent with the immune-cold phenotype previously associated with luminal MIBC (9–11). These patterns were consistently observed by using Xenium 5K data, supporting a spatially organized association between tumor lineage states and immune microenvironment composition (Fig. 4G). Among immune-enriched basal regions, B-IS state exhibited high expression of immunoregulatory and suppressive ligands, including CD274 (PD-L1), NECTIN3, HAVCR2 (TIM-3) and SIRPA (Supplementary Fig. S10D). Pathway-level analysis further showed that tumor–normal boundary regions were enriched for complement, interferon (IFN)/MHC-II signaling, and ECM–receptor interaction pathways (Supplementary Fig. S10E). Collectively, these results suggest that basal tumor states reside within immune-rich yet immunosuppressive microenvironments, whereas luminal regions remain largely immune-silent.
We next assessed whether spatial tumor states were associated with response to neoadjuvant chemotherapy. Responders exhibited significantly higher proportions of basal-like tumor states, whereas nonresponders showed increased luminal coverage (Supplementary Fig. S10F). This relationship was further supported by matched clinical comparisons, in which responder tumors (e.g., BC01) were dominated by basal subtypes, whereas nonresponders (e.g., BC17) contained extensive luminal regions. These observations suggest that luminal tumors may be intrinsically less sensitive to cytotoxic chemotherapy, whereas basal tumors may retain relative chemosensitivity.
To explore lineage-associated mechanisms of chemoresistance, we analyzed expression of genes linked to resistance to commonly used chemotherapeutic agents (e.g., Gem, doxorubicin, methotrexate, vinblastine, and Cis) and cellular stress response pathways (Supplementary Fig. S10G). We found that luminal tumors displayed markedly higher expression of proteotoxic stress response programs, including proteasome components (PSMA1–7, PSMB1–10, and PSMD1–14) and molecular chaperones (HSPA5 and DNAJB1). In contrast, basal tumors were enriched for replication stress and unfolded protein response sensor pathways, including CHEK1, EIF2AK3, and XBP1. These results support lineage-specific mechanism of therapeutic adaptation in MIBC.
To identify targetable vulnerabilities enriched in luminal tumors, we intersected tumor-upregulated genes with a panel of FDA-approved drug targets and identified 4 candidates, namely NECTIN4, TACSTD2, BCR, and EPCAM (Fig. 4H). Among these, NECTIN4 exhibited the strongest luminal specificity, with high expression in luminal core regions and a sharp decline toward the tumor–normal boundary, paralleling the transition to basal identity (Fig. 4I; Supplementary Fig. S11A and S11B). Quantitative analyses confirmed significantly higher NECTIN4 expression in luminal versus basal spots and in tumor versus normal compartments (Supplementary Fig. S11A and S11B).
To functionally assess the relationship between NECTIN4 expression and therapeutic response, we examined sensitivity to EV, an ADC targeting NECTIN4 and recently approved as part of a first-line therapy (EV plus pembrolizumab) for advanced and MIBC (8, 12, 13). Using the UM-UC-3 cell line, which is NECTIN4-low and EV-resistant, we overexpressed NECTIN4 and observed increased EV sensitivity in vitro (Fig. 4J). Notably, NECTIN4 overexpression induced a coordinated transcriptional shift characterized by upregulation of FGFR3 and suppression of EGFR, MHC-II components, and IFN-stimulated genes (HLA-DRA, IFITM3, and STAT1), consistent with a transition from a basal-like, immune-active state toward a luminal-like, immune-quiescent phenotype. This transcriptional reprogramming mirrors spatial patterns observed in tumor–normal boundary regions, where basal states dominate and immune signaling is heightened.
Pan-cohort Analyses Reveal Conserved FGFR3–EGFR Lineage Exclusivity and NECTIN4 as a Luminal Biomarker
Prior studies have shown that specific genomic alterations are enriched in distinct molecular subtypes of bladder cancer, with TP53 and RB1 mutations typically associated with basal/squamous-like tumors and FGFR3 alterations enriched in luminal tumors (9, 17). To systematically evaluate these relationships, we analyzed 7 independent MIBC cohorts with matched genomic and transcriptomic data, quantifying associations between genomic alterations and subtype identity, including TCGA (n = 407 patients), IMvigor (010, 130, 210, and 211; n = 2,803 patients), ABACUS (n = 148 patients), and an independent in-house bulk RNA-seq cohort (n = 21 patients; Supplementary Fig. S12A).
Across all datasets, FGFR3 alterations were consistently depleted in basal tumors and enriched in luminal tumors, reaching statistical significance in 5 of 7 cohorts. Conversely, TP53 and RB1 mutations were enriched in basal tumors, with additional consistent associations observed for NF2 and NFE2L2 alterations. Examination of transcriptional programs revealed that FGFR3 alterations were associated with lower basal signature activity, whereas TP53, RB1, and NF2 mutations correlated with increased basal program expression across cohorts (Supplementary Fig. S12B). These findings support a conserved model in which FGFR3 alterations align with luminal differentiation, whereas TP53, RB1, and NF2 alterations are associated with basal lineage programs.
To assess the generalizability of the spatially defined tumor states identified in earlier analyses (Figs. 3 and 4), we next projected a spatially derived basal program score onto 7 independent transcriptomic cohorts (n = 3,379 patients; Fig. 5A–F). For each sample, we calculated a basal score reflecting the activity of basal-like transcriptional programs derived from our spatial framework. In our in-house cohort (Fig. 5A; Supplementary Fig. S13A), tumors with low basal scores exhibited higher FGFR3 expression and mutation frequency, whereas tumors with high basal scores were enriched for RB1, TP53, and NF2 alterations, consistent with spatial findings.
Figure 5.
Pan-cohort analysis reveals FGFR3–EGFR exclusivity and NECTIN4 as a luminal biomarker. A–C, Multicohort overview of subtype-defining genomic and transcriptomic features across 7 MIBC datasets: in-house cohort (A), TCGA cohort (B), and ABACUS cohort (C). Subtype annotations, stage, key genomic alterations (e.g., TP53, RB1, FGFR3, and NF2), and expression levels of FGFR3 and EGFR. BC, bladder cancer; Luminal_inf, luminal infiltrated; Luminal_pa, luminal papillary; N, no; Y, yes. UroA, urothelial-like A; UroB, urothelial-like B; GU, genomically unstable; SCCL, squamous cell carcinoma-like. Exp, expression. D, Summary of mutation status, NMF classification, CD8+ T-cell infiltration, and basal signature score across 4 IMvigor cohorts: IMvigor010, IMvigor130, IMvigor210, and IMvigor211. chemo, chemotherapy. Atezo, atezolizumab; mBC, metastatic bladder cancer. E, Boxplots showing basal score distributions stratified by CD8+ T-cell infiltration level (desert, excluded, and inflamed) in each IMvigor cohort. F, Stacked barplots of NECTIN4 expression across all 7 cohorts, comparing tumors with different molecular subtypes. Statistical significance is indicated as follows: *, P < 0.05; ****, P < 0.0001.
In the TCGA cohort (Fig. 5B; Supplementary Fig. S13B), basal scores varied significantly across molecular subtypes, with basal and neuronal tumors showing the highest scores. Tumors with elevated basal scores were also associated with more advanced stages in local stage (P = 0.00054), consistent with spatial enrichment of basal-like states at invasive margins. In the ABACUS cohort (Fig. 5C; Supplementary Fig. S13C), basal scores were significantly lower in luminal tumors defined by the Lund classifier and increased following neoadjuvant immunotherapy (P = 1.7e−07), suggesting that immune activation may be accompanied by basal-like transcriptional reprogramming. Similar patterns were observed across IMvigor010, IMvigor130, IMvigor210, and IMvigor211 cohorts (Fig. 5D and E; Supplementary Fig. S13D), in which tumors with high basal scores were significantly enriched for basal non-negative matrix factorization (NMF) subtypes and CD8+ T-cell inflammation. Conversely, tumors with low basal scores corresponded to immune-deserted, luminal-enriched phenotypes, consistent with the spatial immune organization observed in Fig. 4F.
We next examined luminal-specific therapeutic markers across cohorts. NECTIN4 expression was consistently enriched in tumors with low basal scores across all 7 datasets (Fig. 5F), corresponding to luminal molecular subtypes. This pan-cohort pattern reinforces our spatial findings and supports that NECTIN4 could function as a lineage-specific biomarker and therapeutic target in luminal MIBC.
Across all cohorts, FGFR3 and EGFR expression showed mutual exclusivity, mirroring the opposing spatial distributions observed in Fig. 3J and reinforcing their alignment with luminal and basal lineage programs, respectively. Collectively, these pan-cohort analyses suggest that the basal–luminal differentiation axis uncovered by ST may reflect a broadly conserved biological framework.
Spatially Defined TLS Subtypes Reveal Distinct Maturity States and Associations with Treatment Response
TLSs are ectopic LAs that arise in chronically inflamed tissues, including tumors, in which they have been associated with improved prognosis and enhanced responsiveness to immunotherapy, highlighting their potential as both predictive biomarkers and functional immune niches (50). Interestingly, a few studies also reported that baseline tumor TLS correlated with increased pathologic responses and improved clinical outcomes with neoadjuvant chemotherapy in colorectal and breast cancers (51–54). Despite these findings, the prevalence, spatial distribution, maturation states, and therapeutic relevance of TLSs in MIBC remain incompletely defined, especially in the setting to neoadjuvant chemotherapy.
To systematically characterize TLSs in MIBC, we integrated ST with histology-guided annotation to identify, localize, and characterize TLSs across tumor tissue sections (see Supplementary Methods). As shown in Fig. 6A, we used the iSTAR-TLS model developed in our laboratory to infer TLS activity scores across ST spots and extract TLS-high regions based on peak TLS scores. TLSs were further categorized based on spatial proximity to tumor regions: Peritumoral TLSs (P-TLS) were defined as those located within 6 spatial spots from the tumor boundary, whereas distant TLSs (D-TLS) resided beyond this range. Across the cohort, we identified 21 TLSs, including 12 D-TLSs, 4 P-TLSs from nonresponders [P-TLS(NR)], and 5 P-TLSs from responders [P-TLS(R); Fig. 6B]. iSTAR-derived TLS predictions showed strong concordance with pathologist-annotated TLS regions (Fig. 6C), with comprehensive TLS annotations shown in Supplementary Fig. S14.
Figure 6.
Spatially defined TLS subtypes exhibit distinct immunologic states and response associations. A, Schematic of TLS identification using ST-derived iSTAR-TLS scores and classification into P-TLS and D-TLS based on distance to tumor regions. B, Representative hematoxylin and eosin (H&E) images showing identified TLSs stratified into D-TLS, P-TLS(NR), and P-TLS(R). C, Concordance between iSTAR-TLS annotation and histologic pathology annotation. D, Heatmap of selected immune cell type and functional gene module expression across D-TLS, P-TLS(R), and P-TLS(NR) regions. GCB, germinal center B cells; MemB, memory B cells; Breg, regulatory B cells; PC, plasma cells; Tn, naive T cells; Tcm, central memory T cells; Teff, effector T cells; Tfh, follicular helper T cells; Tisg, interferon-stimulated gene-expressing T cells. TCR, T-cell receptor; Treg, regulatory T cells. E, Spatial gene expression maps for representative TLSs and boxplots comparing marker expression across TLS subtypes. F, Barplot showing counts of TLS subtypes stratified by response status. G, Sankey plot showing TLS subtype distribution in responders (R) and nonresponders (NR). Statistical significance is indicated as follows: *, P < 0.05. H, Dot plot of inferred ligand–receptor interactions between tumor cells and TLS-associated compartments, with dot size indicating statistical significance and color indicating communication probability.
We next compared transcriptional programs across D-TLSs, P-TLS(R), and P-TLS(NR) to define their immunologic states (Fig. 6D). D-TLSs were enriched for gene modules associated with naïve B cells (e.g., FCRL1 and TCL1A), memory B cells (e.g., CCR6 and BANK1), germinal center B cells (e.g., BCL2 and AICDA), and follicular dendritic cells (FDC; e.g., CR2 and FCER2). These features are characteristic of mature, well-organized TLSs and are consistent with prior description of functionally competent lymphoid structures (55, 56). In contrast, P-TLSs exhibited reduced expression of B-cell and FDC markers, indicative of less mature or incompletely organized LAs.
Importantly, P-TLSs from responders and nonresponders exhibited marked transcriptional divergence. P-TLS(R) regions showed enrichment of IFN-stimulated genes (e.g., IFI44 and ISG15) and proliferation markers (e.g., MKI67), consistent with antigen-stimulated, immune-active states (57, 58). In contrast, P-TLS(NR) regions were characterized by expression of regulatory T-cell markers (e.g., FOXP3 and IL2RA), T-cell exhaustion markers (e.g., LAG3, CTLA4, and TIGIT), and cellular stress response genes (e.g., HSPA6 and HSPB1), suggestive of dysfunctional and immunosuppressive TLS states (59, 60).
Representative spatial maps illustrate these TLS subtypes (Fig. 6E). In a D-TLS (BC13.10), expression of B-cell and FDC lineage markers, including CD79B, CD3E, and CR2, was prominent. In a P-TLS(R) (BC01.12), IFN-responsive genes such as IFI44 and ISG15 were highly expressed, whereas a P-TLS(NR) (BC17.8) exhibited elevated expression of cellular stress–associated gene HSPB1 and plasma marker JCHAIN. Quantitative analyses confirmed that these transcriptional differences were significant across TLS subtypes (Fig. 6E).
We next examined TLS frequency and subtype distribution across treatment response groups (Fig. 6F and G). TLSs were detected in 4 of 5 (80%) responders but in only 5 of 17 (29%) nonresponders. Notably, P-TLSs were significantly enriched in responders and were nearly absent in nonresponders (χ2 test; P = 0.05), suggesting a positive association between tumor-proximal TLSs and therapeutic efficacy, consistent with previous studies in multiple cancer types (51–54).
To further define the spatial relationship between TLSs and tumor lineage states, we quantified the distance between TLSs and basal-like versus luminal-like tumor spots (Supplementary Fig. S15A). In nonresponders, TLS proximity did not differ significantly between basal and luminal tumor compartments (P = 0.29). However, TLSs in responders were preferentially located closer to basal-like regions (P = 0.0092; Supplementary Fig. S15A), consistent with earlier observations that basal tumor states localize to invasive margins and are enriched for immune infiltration.
Finally, to explore potential tumor-intrinsic drivers of TLS formation, we analyzed expression of canonical lymphoid-organizing chemokines and cytokines across tumor states (Supplementary Fig. S15B). Key mediators of TLS formation and lymphocyte recruitment, including CXCL13, CCL19, and CXCL10 (61–63), were selectively enriched in basal-like tumor compartments, particularly B-IS and B-EMT states, and were largely absent from luminal tumors. To further examine the relationship between tumor states and TLS-associated microenvironments, we performed cell–cell communication analysis and found that basal-like tumor regions showed stronger CXCL9/CXCL10/CXCL11–CXCR3 interactions with TLS-related compartments (Fig. 6H; Supplementary Fig. S15C), consistent with immune recruitment and activation programs (64). In contrast, luminal-like regions were more strongly associated with interactions involving MIF and NAMPT, which are more consistent with an immune-cold or immunosuppressive microenvironment (65, 66). In addition, TLS niches in nonresponder samples showed more immunosuppressive ligand–receptor interactions with tumor cells, suggesting functional heterogeneity among TLS-associated microenvironments. Finally, analysis of public bulk transcriptomic cohorts showed that CXCL9, CXCL10, and CXCL11 are consistently more highly expressed in basal-like tumors across datasets (Supplementary Fig. S15D).
Reciprocal FGFR3 and EGFR Signaling Modulates Bladder Cancer Lineage Identity and Therapeutic Vulnerabilities
To functionally interrogate how lineage-defining RTKs govern transcriptional identity in bladder cancer, we treated a panel of 6 molecularly annotated cell lines with selective FGFR3 or EGFR inhibition. The panel included 3 luminal-type cell lines (RT4, RT112, and UM-UC-14) and 3 basal-type cell lines (5637, HT1376, and SCaBER). Cells were exposed to the FGFR3 inhibitor (erdafitinib) or the EGFR inhibitor (erlotinib), and transcriptomes were profiled after 5 days of treatment (Fig. 7A and B).
Figure 7.
Lineage-directed therapeutic strategies in bladder cancer. A, Schematic illustrating reciprocal lineage transitions between luminal and basal states mediated by FGFR3 or EGFR inhibition. B, Experimental design: Luminal and basal bladder cancer cell lines were treated with FGFR3 inhibitor (FGFR3i) or EGFR inhibitor (EGFRi) for 5 days. C, Heatmaps showing expression of basal (pink) and luminal (green) marker genes in 3 luminal (RT4, RT112, and UM-UC-14) and 3 basal (5637, HT1376, and SCaBER) cell lines. Right, signature scores quantifying luminal and basal program activity under FGFR3i or EGFRi. Inh, inhibitor-treated. D, Schematic illustration of the relationship among tumor lineage states, TME composition, and potential therapeutic response. Luminal tumor cells (green) predominantly occupy the tumor core and coexpress FGFR3 and NECTIN4. Basal-like tumor cells (pink) localize to the invasive edge and are enriched for EGFR expression and immune infiltration, including T cells, DCs, tumor-associated macrophages (TAM), and TLSs. Chemo, chemotherapy. ICPs, immun checkpoints. E, Bladder cancer cell lines (5637, HT1376, and SCaBER) were treated with Gem + Cis or EV, followed by RNA-seq. F, Signature scores for luminal and basal were quantified in 5637, HT1376, and SCaBER cells following Gem + Cis and EV. Values are shown as fold changes relative to the untreated control. G, Experimental schematic illustrating the sequential drug treatment and synergy assessment workflow. Statictical significance is indicated as follows: *, P < 0.05; **, P < 0.01. H, Bliss independence synergy heatmaps depicting the interaction between Gem/Cis and EV across dose combinations in HT1376, 5637, and SCaBER cells. Bliss scores were calculated based on observed vs. expected additive effects. Color coding indicates antagonistic interactions (Bliss score < −10), additive or near-additive effects (−10 to 10), and synergistic interactions (Bliss score >10). Median Bliss scores across the dose matrix are indicated for each cell line.
FGFR3 inhibition in luminal cell lines resulted in suppression of luminal differentiation markers, including KRT19 and FOXA1, accompanied by induction of basal-associated keratins such as KRT14, KRT16, and KRT6B. Conversely, EGFR inhibition in basal cell lines led to downregulation of basal programs and partial reactivation of luminal-associated transcriptional signatures. Lineage signature scoring confirmed these directional shifts, with progressive loss of luminal identity following FGFR3 blockade and gain of luminal features upon EGFR inhibition (Fig. 7C). Together, these results indicate that FGFR3 and EGFR signaling axes act in a reciprocal manner to stabilize lineage identity in bladder cancer and that pharmacologic perturbation of this axis is sufficient to induce transcriptional lineage plasticity.
Lineage-Dependent Therapeutic Sensitivity to Chemotherapy and NECTIN4-Targeted Therapy
Building on these findings, we developed a schematic model integrating spatial tumor organization within lineage-associated therapeutic vulnerabilities (Fig. 7D). In this framework, luminal-like tumor regions (green) are characterized by enrichment of FGFR3 signaling and a relatively immune-cold microenvironment with limited immune cell infiltration. In contrast, basal-like tumor regions (red) exhibit elevated EGFR signaling and are associated with increased infiltration of immune cells, including T cells, B cells, DCs, tumor-associated macrophages, CAFs, and the formation of TLS. Within basal-like niches, tumor cells and surrounding immune populations engage in active cytokine and ligand–receptor interactions, contributing to an immune-hot microenvironment. These regions may be more responsive to chemotherapy and immune checkpoint blockade. Conversely, luminal-like regions, enriched for FGFR3 signaling and NECTIN4 expression, may show relative resistance to immune-based therapies but may be susceptible to targeted therapies such as FGFR inhibitors or ADCs (e.g., EV).
To evaluate how tumor cell lineage state influences therapeutic response, we examined transcriptional responses to Gem (GC) and EV in bladder cancer cell lines. Cells were treated with GC or EV for 48 hours, followed by RNA-seq to quantify treatment-induced transcriptional changes (Fig. 7E). Relative to untreated controls, GC and EV elicited distinct lineage-biased effects; GC preferentially suppressed basal-associated transcriptional programs, whereas EV more prominently attenuated luminal-associated gene signatures (Fig. 7F), consistent with their respective lineage alignments observed in spatial and pan-cohort analyses.
Because concurrent administration of chemotherapy and EV is limited in clinical practice by substantial toxicities, we next modeled a sequential treatment strategy. Cells were first treated with GC (intended to kill the more aggressive basal component) at IC50 concentrations for 48 hours, followed by drug washout and subsequent EV treatment (intended for consolidation killing of the less aggressive luminal component) at IC50 for an additional 48 hours, after which cell viability was assessed (Fig. 7G). Bliss independence analysis revealed predominantly additive interactions between GC and EV across dose combinations, with limited evidence of strong synergy (Fig. 7H). Bliss scores were consistently centered within the additive range across HT1376, 5637, and SCaBER cells, indicating predominantly additive interactions between GC and EV under a sequential treatment regimen. Rather than reflecting synergistic interactions, these results support a lineage-aligned therapeutic model in which chemotherapy and NECTIN4-targeted ADCs act on distinct tumor cell programs. Chemotherapy preferentially affects basal-like, proliferative tumor states, whereas EV targets luminal tumor populations enriched for NECTIN4 expression.
Discussion
MIBC is characterized by profound clinical heterogeneity and a lack of predictive biomarkers to guide personalized therapy. Here, we present a large-cohort ST atlas of MIBC that integrates spatial transcriptomes, scRNA-seq, bulk RNA-seq, and WES to reconstruct tumor–immune architecture and uncover lineage-specific therapeutic vulnerabilities.
Rather than conforming to a binary luminal or basal classification, our analyses reveal that individual MIBC tumors harbor a spatially ordered and continuous luminal-to-basal differentiation trajectory that is evident both transcriptionally and anatomically. Luminal-like tumor populations enriched for FGFR3 and NECTIN4 preferentially occupy tumor cores and exhibit heightened proteotoxic stress responses and vulnerability to ADCs. In contrast, peripheral tumor regions transition toward basal-like states characterized by EGFR expression, EMT, IFN signaling, and dense immune infiltration. These programs are not merely transcriptional states but are embedded with radially organized, lineage-defined microenvironments. This intratumoral axis recapitulates canonical TCGA and IMvigor subtype features (11, 17) while also uniquely resolving them into spatially coherent and therapeutically distinct tumor compartments.
This spatial dissection provides an adaptive framework for understanding the discordant therapeutic responses historically observed between luminal and basal MIBC subtypes. Although prior clinical trials, including ABACUS (67), PURE-01 (68), and IMvigor (010, 130, 210, and 211; ref. 11), have linked ICT response to immune infiltration and molecular subtype, these studies lacked spatial context. Our findings clarify these associations by demonstrating that basal programs are enriched at the tumor–normal interface, where they co-localize with T cells, CAFs, and myeloid populations. These immune-rich peripheral niches exhibit elevated IFN signaling, PD-L1 expression, and immunosuppressive ligands (e.g., CD276 and HAVCR2), suggesting a microenvironment poised for both immune activation and exhaustion. In contrast, luminal tumor cores are largely immune-deserted and exhibit high proteostasis burden, features that may underlie their relative resistance to ICT.
These observations underscore the unique value of ST in resolving structured intratumoral heterogeneity that remains invisible to traditional bulk and dissociated approaches. Although bulk RNA-seq assigns tumors to dominant molecular subtypes, spatial profiling reveals that luminal- and basal-like programs frequently coexist within individual tumors in highly organized spatial architectures. scRNA-seq captures cellular heterogeneity but lacks anatomic context; spatial profiling demonstrates that lineage states are not randomly intermixed but radially ordered, with luminal cells concentrated centrally and basal-like cells positioned peripherally. This reproducible spatial organization across patients provides a unifying explanation for divergent therapeutic responses and highlights tumor architecture, beyond cellular composition alone, as a critical determinant of treatment sensitivity.
Our data further establish FGFR3 and EGFR as opposing lineage-defining RTKs that actively regulate transcriptional identity in bladder cancer. Pharmacologic perturbation experiments demonstrated that inhibition of FGFR3 in luminal cell lines resulted in coordinated suppression of luminal differentiation programs accompanied by induction of basal-associated transcriptional features, whereas EGFR inhibition in basal cell lines partially restores luminal identity. These reciprocal responses indicate that lineage states in bladder cancer are dynamically maintained by RTK signaling and can be modulated through targeted pathway interference. Importantly, these functional observations are reinforced by convergent evidence from large-scale patient cohorts in which FGFR3 alterations and expression are consistently enriched in luminal tumors and associated with reduced basal program activity, whereas EGFR expression aligns with basal states. The mutual exclusivity of FGFR3 and EGFR across cohorts mirrors their opposing spatial distribution, supporting a conserved basal–luminal regulatory axis that integrates genomic alterations, transcriptional programs, and microenvironmental context. This plastic regulatory architecture provides a mechanistic basis for subtype interconversion during disease progression and treatment.
In parallel, our study nominates NECTIN4 as a luminal-restricted therapeutic vulnerability. NECTIN4 expression is spatially confined to luminal tumor cores, sharply downregulated at basal boundaries, and predictive of response to NECTIN4-directed ADCs such as EV. Functional experiments further show that NECTIN4 overexpression promotes luminal reprogramming and EV sensitivity in vitro, suggesting a potential lineage-stabilizing role in addition to its function as a therapeutic target. These findings help explain the limited efficacy of EV in basal tumors and suggest potential strategies to expand its benefit (69–71). Consistent with this framework, our pharmacologic perturbation studies indicate that chemotherapy and EV exert lineage-aligned but largely independent cytotoxic effects rather than strong synergy. Transcriptomic profiling following single-agent treatment revealed preferential attenuation of basal- or luminal-associated programs by chemotherapy and EV, respectively, whereas sequential drug administration resulted in predominantly additive cytotoxic effects across multiple cell line models. This compartmentalized mode of action suggests that treatment sequencing and spatial targeting, rather than indiscriminate combination treatment strategies, may be critical for maximizing clinical benefit.
Spatially, basal-associated tumor cell states frequently coexist with immune-enriched niches and TLS-positive regions, whereas luminal states are more often associated with relatively immune-excluded microenvironments, suggesting that therapeutic heterogeneity in MIBC cannot be attributed solely to tumor cell–intrinsic states but also reflects remodeling of the tumor-induced immune microenvironment. Multiple studies across cancer types, including breast (51, 52) and colorectal cancers (53, 54), have shown that TLS presence or TLS-associated signatures are associated with improved response to neoadjuvant chemotherapy with or without immunotherapy and more favorable clinical outcomes. Given the established role of TLS in organizing local immune responses and shaping antitumor immunity, differences in TLS presence and immune context may further modulate treatment sensitivity across MIBC subtypes.
Despite the breadth of our study, certain limitations remain. The static nature of ST profiling precludes direct observation of temporal lineage transitions or immune dynamics. In addition, posttreatment spatial profiling will be essential to fully capture therapy-induced architectural remodeling. Another limitation of this study is that spatial distances were calculated based on two-dimensional tissue sections, which may not fully capture the three-dimensional (3D) organization of TMEs. Future studies integrating high-resolution 3D spatial profiling will further refine our understanding of TLS organization within tumor tissues. Although ST profiling using Visium provides valuable insights into tissue architecture, its spot-level resolution inherently captures mixed cell populations, which may introduce ambiguity in assigning gene expression signals to specific cell types. Although we applied RCTD-based deconvolution to mitigate this limitation, residual cellular admixture may still influence interpretation. To address this, we incorporated orthogonal validation using higher-resolution ST data, which confirmed the major spatial patterns observed. Nevertheless, future studies leveraging single cell–resolved spatial platforms will be important to further refine lineage relationships and spatial interactions. Clinically, our findings support the incorporation of spatially informed biomarkers into therapeutic stratification: not only determining whom to treat, but also where to target. This framework enables anatomically informed precision strategies, such as deploying NECTIN4-directed ADCs against immune-excluded luminal cores while using chemotherapy or immunotherapy to target inflamed basal margins, ultimately advancing precision oncology for patients with MIBC.
Methods
Collection of Patient Samples
Twenty-two patients with primary MIBC (T2–T4) were evaluated at the University of Texas MD Anderson Cancer Center (MDACC) and underwent diagnostic transurethral resection of bladder tumor. All patients received no treatments at the time of surgery. All patients subsequently received neoadjuvant chemotherapy, with pathologic response at cystectomy with either GC, Gem/taxane/doxorubicin (GTA), or MVAC. The cohort included 10 T2-stage, 9 T3-stage, and 3 T4-stage tumors. Based on the clinical response, we classified 5 patients as responders (≤pT1) and 17 as nonresponders. TLSs were histologically identified in 9 tumors. All formalin-fixed, paraffin-embedded (FFPE) blocks in this study were obtained from Pathology Department under a waiver of consent from banked tissues approved by MDACC Institutional Review Board (IRB) protocols. Eight fresh biopsy tissues were collected under institutional protocol PA17-0577 with patients’ written informed consent to research and approved by the IRB at MDACC. All patient data were deidentified according to institutional and federal regulations.
Sample Processing and Histology Assessment
FFPE tissue blocks were sectioned at 5 μm thickness using a microtome and stained with hematoxylin and eosin (H&E) following standard pathology protocols. All slides underwent independent review by 3 dedicated genitourinary pathologists (Rossana, Ziba, and Diana) to confirm urothelial carcinoma histology, assess tumor cellularity, exclude necrotic or artifact-prone regions, and delineate tumor-enriched areas suitable for ST.
For ST profiling, regions of interest were selected exclusively within viable, well-preserved tumor areas, avoiding cautery artifact, necrosis, crush artifact, and regions with insufficient tumor cellularity. The annotated tumor regions guided subsequent placement of tissue sections on Visium Spatial Gene Expression slides.
ST Library Preparation and Sequencing
ST analysis on FFPE slides were performed with the Visium spatial technology from 10x Genomics. Two to three consecutive tissue sections of 5 μm thickness were collected for RNA extraction with the Qiagen RNeasy FFPE Kit to assess the RNA quality of the tissue. The purified RNA was immediately processed to calculate the percentage of total RNA fragments >200 nucleotides (DV200) using the Agilent 4200 TapeStation system and High Sensitivity RNA ScreenTape Analysis (Agilent Technologies). Based on DV200 evaluation, blocks with DV200 >30% were selected for proceeding with sectioning. The area of interest (11 × 11 mm) on section was carefully placed within the allowable area to ensure compatibility with the Visium CytAssist instrument. The tissues were then deparaffinized, stained, and decross-linked, followed by probe hybridization, ligation, CytAssist-enabled RNA digestion and oligo capture, release, and extension. The Visium spatial gene expression FFPE libraries were constructed using the Visium CytAssist Spatial Gene Expression for FFPE Human Transcriptome Probe Kit (PN-1000444) following the manufacturer’s guidance. Constructed libraries were sequenced on the Illumina NovaSeq 6000 platforms to achieve a depth of at least 75,000 mean read pairs per spot. H&E images with fiducial frames were acquired at 40× resolution using a high-definition digital pathology scanner (Vectra Polaris, Akoya Biosciences) and registered to the capture areas.
ST Spot-Level Histopathologic Annotation
Spot-wise histologic annotation was performed on H&E images paired with Visium sections by experienced genitourinary pathologists (Rossana Lazcano, Ziba Rahbar, and D. Shamsutdinova) using Loupe Browser (v6.1.0, 10x Genomics), which enables direct alignment of histologic images with spatially barcoded Visium spots. Pathologists manually reviewed the H&E images and assigned fine-grained histologic labels to individual spots based on established morphologic features, including tumor cell architecture, stromal composition, immune cell aggregation (e.g., TLS, TIL, and lymphoid aggregates), vascular structures, necrosis, urothelium, and muscle. To ensure annotation reliability, only spots with ≥50% cellularity, as determined by histologic assessment, were retained. Spots corresponding to acellular or low-cellularity regions (e.g., collagen-rich stroma or empty areas lacking visible cells) were excluded from both annotation and downstream analyses. Fine-grained annotations were subsequently consolidated into 10 major tissue compartments, including tumor cells, tumor stroma, immune-rich regions (including TLS/TIL/LA), vasculature, necrosis, urothelium, normal stroma, muscle, and others.
scRNA-seq, Bulk RNA-seq, and WES Processing
WES Library Preparation and Sequencing
Genomic DNA for WES was isolated from 21 bladder cancer tumor specimens (Fig. 1B) by the MD Anderson Biospecimen Extraction Resource, with adjacent normal tissue–derived DNA serving as the germline control. For each sample, 500 ng of DNA was prepared into sequencing libraries using the Agilent SureSelect XT HS and XT Low Input Reagent kits. Whole-exome capture employed Agilent SureSelect Human All Exon V4 biotinylated probes per the manufacturer’s instructions, and libraries were sequenced on the Illumina NovaSeq 6000 platform.
Bulk RNA-seq Library Preparation and Sequencing
Total RNA was isolated from the same 21 bladder cancer tumor specimens that underwent WES. RNA purity was assessed for quality and then fragmented and reverse-transcribed into first- and second-strand cDNA followed by adapter ligation. A measure of 200 ng of the prepared library was hybridized with biotinylated probes (SureSelect XT HS Human All Exon V8 probes) and captured using streptavidin-coated magnetic beads. Libraries were amplified via PCR, after which the final libraries were normalized and pooled to a 6 nmol/L concentration and sequenced on the Illumina NovaSeq 6000 system. Sequencing runs were done using dual-indexed paired-end read format with a total of 220 cycles.
scRNA-seq Library Preparation and Sequencing
A total of 8 biopsy specimens were minced to <1-mm fragments, placed in 2-mL Eppendorf tubes containing Liberase TM (0.1 mg/mL) and DNase I (0.2 mg/mL; MilliporeSigma) in RPMI-1640/0.1% BSA, and incubated at 37°C for 15 minutes with gentle rotation. The resulting suspension was passed through 70-μm strainers, pelleted, and treated with ACK red-blood-cell lysis buffer (Thermo Fisher Scientific, #A1049201). Cell viability was assessed by trypan blue exclusion, and only samples with >60% viability proceeded to Chromium iX (10x Genomics) for generation of 5′ v2 gel beads-in-emulsions. The barcoded cDNA was then purified, amplified, and converted into 10x scRNA-seq libraries, quantified on a Bioanalyzer 2100 (Agilent), and sequenced on NovaSeq 6000 at ∼50,000 reads per cell following the manufacturer’s guidelines.
Chemicals, Cell Lines, and In Vitro Treatments
Erdafitinib and erlotinib were purchased from MedChemExpress. Gem and Cis were obtained from SelleckChem. EV was collected from the clinical pharmacy at the MDACC. Human bladder cancer cell line HT1376 was purchased from the American Type Culture Collection (ATCC; cat. #CRL-1472; RRID: CVCL_1282). ScaBER (RRID: CVCL_1787) and UM-UC-14 (RRID: CVCL_2747) were obtained from the University of Texas MDACC Bladder SPORE Tissue Bank. All 3 cell lines were cultured in minimum essential medium supplemented with 10% FBS, 2 mmol/L L-glutamine, and 0.1 mmol/L nonessential amino acids. RT112 (Creative Bioarray, cat. #CSC-C0268; RRID: CVCL_1670) and 5637 (ATCC, cat. #HTB-9; RRID: CVCL_0126) cells were cultured in RPMI-1640 supplemented with 10% FBS. The RT4 cell line (ATCC, cat. #HTB-2; RRID: CVCL_0036) was maintained in McCoy’s 5A medium containing 10% FBS. All cells were cultured at 37°C in a humidified incubator with 5% CO2. Cells were passaged every 2 to 3 days upon reaching 80% to 90% confluence using trypsin-ethylenediaminetetraacetic acid (trypsin-EDTA) and subcultured at a 1:3 split ratio. All cell lines were routinely tested for Mycoplasma contamination by PCR, and only Mycoplasma-free cultures were used in experiments. Luminal-type cell lines (RT4, RT112, and UM-UC-14) were treated with erdafitinib, and basal-type cell lines (5637, HT1376, and SCaBER) were treated with the erlotinib for 3 or 5 days. For chemotherapy/EV treatment, only basal-type cell lines (5637, HT1376, and SCaBER) were treated with Gem/Cis or EV for 48 hours.
Cell RNA Extraction, Library Construction, and Sequencing
At the end of treatment, cell samples were collected and total RNA was extracted with the Total RNA purification Kit (Norgen Biotek). The library construction for RNA-seq was performed with VAHTS Universal V10 RNA-seq Library Prep Kit (Vazyme). Briefly, 1 μg of total RNA underwent mRNA enrichment, fragmentation, cDNA synthesis, Illumina adapter ligation, amplification, and purification. The quality of constructed libraries was determined using TapeStation 4200 (Agilent) and sequenced at 75-bp paired-end format (60 million reads per sample) with an Element AVITI Sequencer at the Texas Scientific LLC NGS Lab.
Supplementary Material
Supplementary Methods file contains extended methods, including the processing and analysis of multiple sequencing datasets.
Supplementary Tables file contains key information and supporting datasets. Table S1 | Clinicopathological features of the 22 MIBC patients. Table S2 | The top 50 most significantly differentially expressed genes between the 15 single cell clusters. Table S3 | GSEA pathway enrichment results for 4 spatial clusters. Table S4 | Clinical characteristics and spatial tumor state composition across samples. Table S5 | Bootstrap-based robustness analysis of spatial tumor state proportions. Table S6 | The top 50 most significantly differentially expressed genes between the 4 spatial clusters.
Supplementary Figures file contains Supplementary Figures S1–S15. Figure S1 | Pathology-guided spatial annotation across 22 MIBC tumors. Figure S2 | Validation of spatial deconvolution accuracy and cell-type compartmentalization. Figure S3 | Additional tumors demonstrating spatial CNV-based subclones and transcriptional divergence. Figure S4 | Cross-platform concordance of CNV profiles inferred from spatial transcriptomics and WES. Figure S5 | Transcriptional and functional features across the luminal-to-basal differentiation continuum. Figure S6 | Single-cell analysis recapitulates the spatial luminal-to-basal differentiation continuum. Figure S7 | Spatial validation of lineage-associated EGFR and FGFR3 expression patterns. Figure S8 | Lineage-associated expression of FGFR3–EGFR signaling and transcriptional regulators across normal tissue, tumor states, and independent cohorts. Figure S9 | Pathology-guided spatial annotation across all MIBC tumors. Figure S10 | Spatial tumor states associate with chemotherapy response and stress-related gene expression. Figure S11 | NECTIN4 expression is enriched in luminal tumors and declines at the tumor-normal boundary. Figure S12 | Subtype-associated genomic alterations and their relationship with basal program activity. Figure S13 | Associations between basal signature scores and clinical or molecular features across MIBC cohorts. Figure S14 | Pathologist-guided TLS annotation confirms spatial transcriptomic predictions. Figure S15 | Lineage-associated chemokine programs and tumor–TLS interactions.
Acknowledgments
L. Wang was partially supported by NIH/National Cancer Institute (NCI) grants (R01CA266280 and U01CA294518), the Cancer Prevention Research Institute of Texas (CPRIT) grant (RP200385), University of Texas (UT) MD Anderson Strategic Research Programs, and Break Through Cancer. L. Wang is a member of the James P. Allison Institute, the Institute for Data Science in Oncology, and the Center for Cellular Language Intelligence at UT MD Anderson and receives research funding from all three. A.Y. Shah and M.T. Campbell were supported by the NCI training grant (T32217789). J. Gao was supported in part by the Khalifa Physician Scientist Award, the David H. Koch Center for Applied Research of Genitourinary Cancers, Joan and Herb Kelleher Charitable Foundation, Kidney Cancer Association (KCA) Advanced Discovery Award, the Williams TNT Fund, the V Foundation Translational Award, the Department of Defense (DOD) KCRP Translational Research Partnership Award, and NIH/NCI grants (R01CA254988, R01CA269489, and R01 CA282282). P. Msaouel was supported by the NCI grants (R37CA288448 and R01CA285454), Gateway for Cancer Research, a Translational Research Partnership Award (KC240237P1), an Idea Development Award (RA230062) by the DOD, an Advanced Discovery Award by the KCA, and a Translational Research Award by the V Foundation. This work was also supported in part by the MD Anderson’s Prometheus informatics system and by the Department of Genitourinary Medical Oncology’s Eckstein and Alexander Laboratories. G. Genovese was supported by the NIH grants (R01CA258226 and R01CA258886), DOD (HT942525-1-0047), the CPRIT (RP260133), the Koch Center for Applied Research of Genitourinary Cancers, and the MD Anderson Moonshot Program. The authors used a generative artificial intelligence tool to assist with language refinement. All scientific content and conclusions were reviewed and validated by the authors.
Footnotes
Note: Supplementary data for this article are available at Cancer Discovery Online (http://cancerdiscovery.aacrjournals.org/).
Contributor Information
Jianjun Gao, Email: JGao1@mdanderson.org.
Linghua Wang, Email: Lwang22@mdanderson.org.
Data Availability
Publicly available bulk transcriptomic, copy-number, mutation, and clinical data from TCGA Bladder Cancer (BLCA) cohort were obtained from the UCSC Xena data portal (https://xenabrowser.net/datapages/). RNA-seq data from the IMvigor clinical trials were accessed under a Genentech Data Access Agreement and are available through the European Genome-phenome Archive (EGA) under controlled access (EGAD50000001100). RNA-seq data from the ABACUS trial were obtained through the EGA under controlled access (EGAD00001006205). Processed human single-cell and ST data in this study are available in the Gene Expression Omnibus (datasets GSE319536, GSE326225 and GSE326226). Raw cell line sequencing data are available in Sequence Read Archive (dataset PRJNA1443383). Any additional information required to reanalyze the data reported in this article is available from the lead contact upon request.
Code Availability
All data analysis was performed using publicly available software, packages, and tools, as described in the “Methods” section. The codes used to produce the results of this article are available at GitHub (https://github.com/VividKai/SpatialBC.git), including specific code and functions used in the study, and are provided to ensure transparency and reproducibility of the results.
Authors’ Disclosures
E. Hasanov reports personal fees from Pfizer, Eisai, and Telix pharmaceuticals outside the submitted work. L. Eberlin reports a patent for Mass Spectrometry Imaging Technologies issued, licensed, and with royalties paid from Waters Corporation. K. Bree reports personal fees from Johnson & Johnson and Aptitude Health outside the submitted work. A.M. Kamat reports grants from FKD Therapies, Patient-Centered Outcomes Research Institute (PCORI), Seagen, and SWOG, grants and personal fees from Photocure and enGene, personal fees from Astellas Pharma, Atonco Pharma, Bristol Myers Squibb, CG Oncology, Cystotech, Ferring, Genentech, ImmunityBio, Janssen, Medac, Merck, Nonagen Bioscience, Pfizer, Protara Therapeutics, Theralase, UroGen Pharma, and Valar Labs, other support from International Bladder Cancer Group (IBCG), European Urology Oncology, Journal of Urology, UroToday, and World Bladder Cancer Patient Coalition, and nonfinancial support and other support from American Urological Association outside the submitted work, as well as a patent for CyPRIT (Cytokine Predictors of Response to Intravesical Therapy) issued to A.M. Kamat/MDACC. P. Sharma reports other support from Achelois, Adaptive Biotechnologies, Akoya Biosciences, Apricity, Asher Bio, BioAtla LLC, BioNTech, Candel Therapeutics, Catalio, Dragonfly Therapeutics, Earli Inc., Glympse, Gyges, ImaginAb, JSL Health, LAVA Therapeutics, Lytix Biopharma, Matrisome, Neuvogen, NTx, Oncolytics, Osteologic, Phenomic AI, PyrOjas, Inc., Soley Therpeutics, Sporos, Time BioVentures, Two Bear Capital, Vironexis, and Xilis, Inc. outside the submitted work. A.Y. Shah reports other support from EMD Serono during the conduct of the study. O. Alhalabi reports grants from Rondo Therapeutics, Arcus Biosciences, and Ikena Therapeutics, grants and personal fees from AstraZeneca, and personal fees from Curio Science, IDEOlogy Health, and Seagen outside the submitted work. P. Msaouel reports honoraria for service on scientific advisory boards for Mirati Therapeutics, Bristol Myers Squibb, and Exelixis; consulting for Axiom Healthcare Strategies; nonbranded educational programs supported by DAVA Oncology, Exelixis and Pfizer; and research funding for clinical trials from Regeneron Pharmaceuticals, Summit Therapeutics, Merck, Takeda, Bristol Myers Squibb, Mirati Therapeutics, Gateway for Cancer Research, and the University of Texas MDACC. M.T. Campbell reports grants and personal fees from Pfizer, Exelixis, and Astellas Pharma, grants from AstraZeneca, and personal fees from Eisai, Curio Science, and MJH Life Sciences outside the submitted work. L. Wang reports grants from Break Through Cancer and personal fees from SELLAS Life Sciences outside the submitted work. No disclosures were reported by the other authors.
Authors’ Contributions
K. Yu: Resources, data curation, formal analysis, investigation, visualization, methodology, writing–original draft, writing–review and editing. J. Chen: Resources, data curation, validation, methodology, writing–original draft, writing–review and editing. Y.-Y. Chu: Validation, methodology, writing–original draft, writing–review and editing. S. Nair: Validation, methodology, writing–original draft, writing–review and editing. E. Crupi: Data curation, visualization, methodology, writing–original draft, writing–review and editing. E. Hasanov: Resources, data curation, writing–review and editing. Y. Mei: Validation, writing–review and editing. X. Han: Formal analysis, investigation, writing–review and editing. Yang Liu: Formal analysis, investigation, writing–review and editing. Yunhe Liu: Formal analysis, visualization, writing–review and editing. G. Pei: Investigation, visualization, methodology. F. Peng: Investigation, methodology, writing–review and editing. J. Liao: Investigation, writing–review and editing. E. Dai: Investigation, writing–review and editing. T. Chu: Investigation, visualization, writing–review and editing. K.S. Cho: Investigation, methodology, writing–review and editing. J. Jiang: Investigation, methodology, writing–review and editing. X. Yan: Investigation, writing–review and editing. Y. Dai: Investigation, visualization, writing–review and editing. J. Wang: Investigation, writing–review and editing. B. Zhang: Visualization, writing–original draft, writing–review and editing. G. Genovese: Data curation, methodology, writing–review and editing. N. Putluri: Methodology, writing–review and editing. L. Eberlin: Methodology, writing–review and editing. L. Yang: Resources, methodology, writing–review and editing. C. Lin: Resources, methodology, writing–review and editing. L.N. Kwong: Resources, methodology, writing–review and editing. D. Shamsutdinova: Resources, methodology, writing–original draft. K.M. Wani: Methodology, writing–review and editing. B.A. Czerniak: Investigation, writing–review and editing. W. Yu: Investigation, writing–review and editing. J. Zhao: Methodology, writing–review and editing. C.C. Guo: Resources, writing–review and editing. D. Hansel: Resources, writing–review and editing. C. Pettaway: Resources, writing–review and editing. L. Pisters: Resources, writing–review and editing. K. Bree: Resources, writing–review and editing. N. Navai: Resources, writing–review and editing. B.H. Lee: Resources, writing–review and editing. A.M. Kamat: Resources, writing–review and editing. P. Sharma: Investigation, writing–review and editing. P. Corn: Resources, writing–review and editing. C.Y. Jiang: Resources, writing–review and editing. A.Y. Shah: Resources, writing–review and editing. O. Alhalabi: Resources, writing–review and editing. P. Msaouel: Investigation, writing–review and editing. A.O. Siefker-Radtke: Investigation, writing–review and editing. C.J. Logothetis: Investigation, writing–review and editing. M.T. Campbell: Investigation, writing–review and editing. A.J. Lazar: Investigation, writing–review and editing. J. Gao: Conceptualization, resources, supervision, investigation, writing–original draft, project administration, writing–review and editing. L. Wang: Conceptualization, resources, supervision, funding acquisition, writing–original draft, project administration, writing–review and editing.
References
- 1. Bray F, Laversanne M, Sung H, Ferlay J, Siegel RL, Soerjomataram I, et al. Global cancer statistics 2022: GLOBOCAN estimates of incidence and mortality worldwide for 36 cancers in 185 countries. CA Cancer J Clin 2024;74:229–63. 10.3322/caac.21834. [DOI] [PubMed] [Google Scholar]
- 2. Richters A, Aben KKH, Kiemeney LALM. The global burden of urinary bladder cancer: an update. World J Urol 2020;38:1895–904. 10.1007/s00345-019-02984-4. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 3. Burger M, Catto JWF, Dalbagni G, Grossman HB, Herr H, Karakiewicz P, et al. Epidemiology and risk factors of urothelial bladder cancer. Eur Urol 2013;63:234–41. 10.1016/j.eururo.2012.07.033. [DOI] [PubMed] [Google Scholar]
- 4. Stein JP, Lieskovsky G, Cote R, Groshen S, Feng AC, Boyd S, et al. Radical cystectomy in the treatment of invasive bladder cancer: long-term results in 1,054 patients. J Clin Oncol 2001;19:666–75. 10.1200/JCO.2001.19.3.666. [DOI] [PubMed] [Google Scholar]
- 5. Grossman HB, Natale RB, Tangen CM, Speights VO, Vogelzang NJ, Trump DL, et al. Neoadjuvant chemotherapy plus cystectomy compared with cystectomy alone for locally advanced bladder cancer. N Engl J Med 2003;349:859–66. 10.1056/NEJMoa022148. [DOI] [PubMed] [Google Scholar]
- 6. von der Maase H, Hansen SW, Roberts JT, Dogliotti L, Oliver T, Moore MJ, et al. Gemcitabine and cisplatin versus methotrexate, vinblastine, doxorubicin, and cisplatin in advanced or metastatic bladder cancer: results of a large, randomized, multinational, multicenter, phase III study. J Clin Oncol 2000;18:3068–77. 10.1200/JCO.2000.18.17.3068. [DOI] [PubMed] [Google Scholar]
- 7. Powles T, Catto JWF, Galsky MD, Al-Ahmadie H, Meeks JJ, Nishiyama H, et al. Perioperative durvalumab with neoadjuvant chemotherapy in operable bladder cancer. N Engl J Med 2024;391:1773–86. 10.1056/NEJMoa2408154. [DOI] [PubMed] [Google Scholar]
- 8. Vulsteke C, Adra N, Danchaivijitr P, Sabadash M, Rodriguez-Vida A, Zhang Z, et al. Perioperative enfortumab vedotin and pembrolizumab in bladder cancer. N Engl J Med 2026;394:1257–69. 10.1056/NEJMoa2511674. [DOI] [PubMed] [Google Scholar]
- 9. Kamoun A, de Reyniès A, Allory Y, Sjödahl G, Robertson AG, Seiler R, et al. A consensus molecular classification of muscle-invasive bladder cancer. Eur Urol 2020;77:420–33. 10.1016/j.eururo.2019.09.006. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10. Cancer Genome Atlas Research Network . Comprehensive molecular characterization of urothelial bladder carcinoma. Nature 2014;507:315–22. 10.1038/nature12965. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11. Hamidi H, Senbabaoglu Y, Beig N, Roels J, Manuel C, Guan X, et al. Molecular heterogeneity in urothelial carcinoma and determinants of clinical benefit to PD-L1 blockade. Cancer Cell 2024;42:2098–112.e4. 10.1016/j.ccell.2024.10.016. [DOI] [PubMed] [Google Scholar]
- 12. Powles T, Rosenberg JE, Sonpavde GP, Loriot Y, Durán I, Lee J-L, et al. Enfortumab vedotin in previously treated advanced urothelial carcinoma. N Engl J Med 2021;384:1125–35. 10.1056/NEJMoa2035807. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 13. Powles T, Valderrama BP, Gupta S, Bedke J, Kikuchi E, Hoffman-Censits J, et al. Enfortumab vedotin and pembrolizumab in untreated advanced urothelial cancer. N Engl J Med 2024;390:875–88. 10.1056/NEJMoa2312117. [DOI] [PubMed] [Google Scholar]
- 14. McConkey DJ, Choi W, Ochoa A, Dinney CPN. Intrinsic subtypes and bladder cancer metastasis. Asian J Urol 2016;3:260–7. 10.1016/j.ajur.2016.09.009. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15. Tan J, Wang Y, Sun L, Xu S, Li C, Jin X. The origin and evolution of bladder cancer stem cells. Front Cell Dev Biol 2022;10:950241. 10.3389/fcell.2022.950241. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 16. Choi W, Porten S, Kim S, Willis D, Plimack ER, Hoffman-Censits J, et al. Identification of distinct basal and luminal subtypes of muscle-invasive bladder cancer with different sensitivities to frontline chemotherapy. Cancer Cell 2014;25:152–65. 10.1016/j.ccr.2014.01.009. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 17. Robertson AG, Kim J, Al-Ahmadie H, Bellmunt J, Guo G, Cherniack AD, et al. Comprehensive molecular characterization of muscle-invasive bladder cancer. Cell 2017;171:540–56.e25. 10.1016/j.cell.2017.09.007. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 18. Garg M. Urothelial cancer stem cells and epithelial plasticity: current concepts and therapeutic implications in bladder cancer. Cancer Metastasis Rev 2015;34:691–701. 10.1007/s10555-015-9589-6. [DOI] [PubMed] [Google Scholar]
- 19. Pérez-González A, Bévant K, Blanpain C. Cancer cell plasticity during tumor progression, metastasis and response to therapy. Nat Cancer 2023;4:1063–82. 10.1038/s43018-023-00595-y. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 20. Lavallee E, Sfakianos JP, Mulholland DJ. Tumor heterogeneity and consequences for bladder cancer treatment. Cancers (Basel) 2021;13:5297. 10.3390/cancers13215297. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 21. Wu K, Liu X, Zhang J, Wang X, Li X, Chen C. Mechanisms and implications of epithelial cell plasticity in the bladder. Nat Rev Urol 2026;23:70–88. 10.1038/s41585-025-01066-y. [DOI] [PubMed] [Google Scholar]
- 22. Sfakianos JP, Daza J, Hu Y, Anastos H, Bryant G, Bareja R, et al. Epithelial plasticity can generate multi-lineage phenotypes in human and murine bladder cancers. Nat Commun 2020;11:2540. 10.1038/s41467-020-16162-3. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 23. Warrick JI, Hu W, Yamashita H, Walter V, Shuman L, Craig JM, et al. FOXA1 repression drives lineage plasticity and immune heterogeneity in bladder cancers with squamous differentiation. Nat Commun 2022;13:6575. 10.1038/s41467-022-34251-3. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 24. Barrows D, Feng L, Carroll TS, Allis CD. Loss of UTX/KDM6A and the activation of FGFR3 converge to regulate differentiation gene-expression programs in bladder cancer. Proc Natl Acad Sci U S A 2020;117:25732–41. 10.1073/pnas.2008017117. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 25. Lewis SM, Asselin-Labat M-L, Nguyen Q, Berthelet J, Tan X, Wimmer VC, et al. Spatial omics and multiplexed imaging to explore cancer biology. Nat Methods 2021;18:997–1012. 10.1038/s41592-021-01203-6. [DOI] [PubMed] [Google Scholar]
- 26. Rao A, Barkley D, França GS, Yanai I. Exploring tissue architecture using spatial transcriptomics. Nature 2021;596:211–20. 10.1038/s41586-021-03634-9. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 27. Longo SK, Guo MG, Ji AL, Khavari PA. Integrating single-cell and spatial transcriptomics to elucidate intercellular tissue dynamics. Nat Rev Genet 2021;22:627–44. 10.1038/s41576-021-00370-8. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 28. Chang J, Lu J, Liu Q, Xiang T, Zhang S, Yi Y, et al. Single-cell multi-stage spatial evolutional map of esophageal carcinogenesis. Cancer Cell 2025;43:380–97.e7. 10.1016/j.ccell.2025.02.009. [DOI] [PubMed] [Google Scholar]
- 29. Zheng Y, Carrillo-Perez F, Pizurica M, Heiland DH, Gevaert O. Spatial cellular architecture predicts prognosis in glioblastoma. Nat Commun 2023;14:4122. 10.1038/s41467-023-39933-0. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 30. Yan Y, Sun D, Hu J, Chen Y, Sun L, Yu H, et al. Multi-omic profiling highlights factors associated with resistance to immuno-chemotherapy in non-small-cell lung cancer. Nat Genet 2025;57:126–39. 10.1038/s41588-024-01998-y. [DOI] [PubMed] [Google Scholar]
- 31. Chen H, Deng C, Gao J, Wang J, Fu F, Wang Y, et al. Integrative spatial analysis reveals tumor heterogeneity and immune colony niche related to clinical outcomes in small cell lung cancer. Cancer Cell 2025;43:519–36.e5. 10.1016/j.ccell.2025.01.012. [DOI] [PubMed] [Google Scholar]
- 32. Shi ZD, Sun Z, Zhu ZB, Liu X, Chen JZ, Hao L, et al. Integrated single-cell and spatial transcriptomic profiling reveals higher intratumour heterogeneity and epithelial–fibroblast interactions in recurrent bladder cancer. Clin Transl Med 2023;13:e1338. 10.1002/ctm2.1338. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 33. Long F, Wang W, Li S, Wang B, Hu X, Wang J, et al. The potential crosstalk between tumor and plasma cells and its association with clinical outcome and immunotherapy response in bladder cancer. J Transl Med 2023;21:298. 10.1186/s12967-023-04151-1. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 34. Lindskrog SV, Schmøkel SS, Nordentoft I, Lamy P, Knudsen M, Prip F, et al. Single-nucleus and spatially resolved intratumor subtype heterogeneity in bladder cancer. Eur Urol Open Sci 2023;51:78–88. 10.1016/j.euros.2023.03.006. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 35. Biswas A, Sahoo S, Riedlinger GM, Ghodoussipour S, Jolly MK, De S. Transcriptional state dynamics lead to heterogeneity and adaptive tumor evolution in urothelial bladder carcinoma. Commun Biol 2023;6:1292. 10.1038/s42003-023-05668-3. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 36. Gouin KH 3rd, Ing N, Plummer JT, Rosser CJ, Ben Cheikh B, Oh C, et al. An N-Cadherin 2 expressing epithelial cell subpopulation predicts response to surgery, chemotherapy and immunotherapy in bladder cancer. Nat Commun 2021;12:4906. 10.1038/s41467-021-25103-7. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 37. Cable DM, Murray E, Zou LS, Goeva A, Macosko EZ, Chen F, et al. Robust decomposition of cell type mixtures in spatial transcriptomics. Nat Biotechnol 2022;40:517–26. 10.1038/s41587-021-00830-w. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 38. Erickson A, He M, Berglund E, Marklund M, Mirzazadeh R, Schultz N, et al. Spatially resolved clonal copy number alterations in benign and malignant tissue. Nature 2022;608:360–7. 10.1038/s41586-022-05023-2. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 39. Gulati GS, Sikandar SS, Wesche DJ, Manjunath A, Bharadwaj A, Berger MJ, et al. Single-cell transcriptional diversity is a hallmark of developmental potential. Science 2020;367:405–11. 10.1126/science.aax0249. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 40. Cao J, Spielmann M, Qiu X, Huang X, Ibrahim DM, Hill AJ, et al. The single-cell transcriptional landscape of mammalian organogenesis. Nature 2019;566:496–502. 10.1038/s41586-019-0969-x. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 41. Walker LC, Krause L, Spurdle AB, Waddell N; kConFab Investigators . Germline copy number variants are not associated with globally acquired copy number changes in familial breast tumours. Breast Cancer Res Treat 2012;134:1005–11. 10.1007/s10549-012-2024-6. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 42. Detlefsen S, Boldt HB, Burton M, Thomsen MM, Rasmussen LG, Ørbeck SV, et al. High overall copy number variation burden by genome-wide methylation profiling holds negative prognostic value in surgically treated pancreatic ductal adenocarcinoma. Hum Pathol 2023;142:68–80. 10.1016/j.humpath.2023.11.002. [DOI] [PubMed] [Google Scholar]
- 43. Damrauer JS, Hoadley KA, Chism DD, Fan C, Tiganelli CJ, Wobker SE, et al. Intrinsic subtypes of high-grade bladder cancer reflect the hallmarks of breast cancer biology. Proc Natl Acad Sci U S A 2014;111:3110–5. 10.1073/pnas.1318376111. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 44. Shi X, Yang J, Deng S, Xu H, Wu D, Zeng Q, et al. TGF-β signaling in the tumor metabolic microenvironment and targeted therapies. J Hematol Oncol 2022;15:135. 10.1186/s13045-022-01349-6. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 45. Bahar ME, Kim HJ, Kim DR. Targeting the RAS/RAF/MAPK pathway for cancer therapy: from mechanism to clinical studies. Signal Transduct Target Ther 2023;8:455. 10.1038/s41392-023-01705-z. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 46. Johnson DE, O’Keefe RA, Grandis JR. Targeting the IL-6/JAK/STAT3 signalling axis in cancer. Nat Rev Clin Oncol 2018;15:234–48. 10.1038/nrclinonc.2018.8. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 47. Fink EE, Sona S, Tran U, Desprez P-E, Bradley M, Qiu H, et al. Single-cell and spatial mapping identify cell types and signaling networks in the human ureter. Dev Cell 2022;57:1899–916.e6. 10.1016/j.devcel.2022.07.004. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 48. Warrick JI, Walter V, Yamashita H, Chung E, Shuman L, Amponsa VO, et al. FOXA1, GATA3 and PPARɣ cooperate to drive luminal subtype in bladder cancer: a molecular analysis of established human cell lines. Sci Rep 2016;6:38531. 10.1038/srep38531. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 49. Jiang Y, Jiang Y-Y, Xie J-J, Mayakonda A, Hazawa M, Chen L, et al. Co-activation of super-enhancer-driven CCAT1 by TP63 and SOX2 promotes squamous cancer progression. Nat Commun 2018;9:3619. 10.1038/s41467-018-06081-9. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 50. Teillaud J-L, Houel A, Panouillot M, Riffard C, Dieu-Nosjean M-C. Tertiary lymphoid structures in anticancer immunity. Nat Rev Cancer 2024;24:629–46. 10.1038/s41568-024-00728-0. [DOI] [PubMed] [Google Scholar]
- 51. Wang Q, Yu Y, Wang C, Jiang Z, Li J, Li X, et al. Heterogeneity of tertiary lymphoid structures predicts the response to neoadjuvant therapy and immune microenvironment characteristics in triple-negative breast cancer. Br J Cancer 2025;132:295–310. 10.1038/s41416-024-02917-y. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 52. Bertucci F, De Nonneville A, Finetti P, Mamessier E. Predictive power of tertiary lymphoid structure signature for neoadjuvant chemotherapy response and immunotherapy benefit in HER2-negative breast cancer. Cancer Commun (Lond) 2023;43:943–6. 10.1002/cac2.12447. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 53. Wang Q, Zhong W, Shen X, Hao Z, Wan M, Yang X, et al. Tertiary lymphoid structures predict survival and response to neoadjuvant therapy in locally advanced rectal cancer. NPJ Precis Oncol 2024;8:61. 10.1038/s41698-024-00533-w. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 54. Hosoi N, Komine C, Dorjkhorloo G, Shiraishi T, Okada T, Sano A, et al. Mature tertiary lymphoid structures indicate good chemotherapy response and prognosis in advanced colorectal cancer. Ann Gastroenterol Surg 2025;10:710–21. 10.1002/ags3.70142. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 55. Siliņa K, Soltermann A, Attar FM, Casanova R, Uckeley ZM, Thut H, et al. Germinal centers determine the prognostic relevance of tertiary lymphoid structures and are impaired by corticosteroids in lung squamous cell carcinoma. Cancer Res 2018;78:1308–20. 10.1158/0008-5472.CAN-17-1987. [DOI] [PubMed] [Google Scholar]
- 56. Posch F, Silina K, Leibl S, Mündlein A, Moch H, Siebenhüner A, et al. Maturation of tertiary lymphoid structures and recurrence of stage II and III colorectal cancer. Oncoimmunology 2018;7:e1378844. 10.1080/2162402X.2017.1378844. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 57. Villarreal DO, Wise MC, Siefert RJ, Yan J, Wood LM, Weiner DB. Ubiquitin-like molecule ISG15 acts as an immune adjuvant to enhance antigen-specific CD8 T-cell tumor immunity. Mol Ther 2015;23:1653–62. 10.1038/mt.2015.120. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 58. Kinker GS, Vitiello GAF, Diniz AB, Cabral-Piccin MP, Pereira PHB, Carvalho MLR, et al. Mature tertiary lymphoid structures are key niches of tumour-specific immune responses in pancreatic ductal adenocarcinomas. Gut 2023;72:1927–41. 10.1136/gutjnl-2022-328697. [DOI] [PubMed] [Google Scholar]
- 59. Budimir N, Thomas GD, Dolina JS, Salek-Ardakani S. Reversing T-cell exhaustion in cancer: lessons learned from PD-1/PD-L1 immune checkpoint blockade. Cancer Immunol Res 2022;10:146–53. 10.1158/2326-6066.CIR-21-0515. [DOI] [PubMed] [Google Scholar]
- 60. Chu Y, Dai E, Li Y, Han G, Pei G, Ingram DR, et al. Pan-cancer T cell atlas links a cellular stress response state to immunotherapy resistance. Nat Med 2023;29:1550–62. 10.1038/s41591-023-02371-y. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 61. Wengner AM, Höpken UE, Petrow PK, Hartmann S, Schurigt U, Bräuer R, et al. CXCR5- and CCR7-dependent lymphoid neogenesis in a murine model of chronic antigen-induced arthritis. Arthritis Rheum 2007;56:3271–83. 10.1002/art.22939. [DOI] [PubMed] [Google Scholar]
- 62. Fleige H, Ravens S, Moschovakis GL, Bölter J, Willenzon S, Sutter G, et al. IL-17–induced CXCL12 recruits B cells and induces follicle formation in BALT in the absence of differentiated FDCs. J Exp Med 2014;211:643–51. 10.1084/jem.20131737. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 63. Rouanne M, Arpaia N, Marabelle A. CXCL13 shapes tertiary lymphoid structures and promotes response to immunotherapy in bladder cancer. Eur J Cancer 2021;151:245–8. 10.1016/j.ejca.2021.03.054. [DOI] [PubMed] [Google Scholar]
- 64. Tokunaga R, Zhang WU, Naseem M, Puccini A, Berger MD, Soni S, et al. CXCL9, CXCL10, CXCL11/CXCR3 axis for immune activation–a target for novel cancer therapy. Cancer Treat Rev 2018;63:40–7. 10.1016/j.ctrv.2017.11.007. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 65. Noe JT, Mitchell RA. MIF-dependent control of tumor immunity. Front Immunol 2020;11:609948. 10.3389/fimmu.2020.609948. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 66. Hong SM, Lee A-Y, Kim B-J, Lee J-E, Seon S-Y, Ha Y-J, et al. NAMPT-driven M2 polarization of tumor-associated macrophages leads to an immunosuppressive microenvironment in colorectal cancer. Adv Sci (Weinh) 2024;11:2303177. 10.1002/advs.202303177. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 67. Powles T, Kockx M, Rodriguez-Vida A, Duran I, Crabb SJ, Van Der Heijden MS, et al. Clinical efficacy and biomarker analysis of neoadjuvant atezolizumab in operable urothelial carcinoma in the ABACUS trial. Nat Med 2019;25:1706–14. 10.1038/s41591-019-0628-7. [DOI] [PubMed] [Google Scholar]
- 68. Necchi A, Anichini A, Raggi D, Briganti A, Massa S, Lucianò R, et al. Pembrolizumab as neoadjuvant therapy before radical cystectomy in patients with muscle-invasive urothelial bladder carcinoma (PURE-01): an open-label, single-arm, phase II study. J Clin Oncol 2018;36:3353–60. 10.1200/JCO.18.01148. [DOI] [PubMed] [Google Scholar]
- 69. Challita-Eid PM, Satpayev D, Yang P, An Z, Morrison K, Shostak Y, et al. Enfortumab vedotin antibody–drug conjugate targeting nectin-4 is a highly potent therapeutic agent in multiple preclinical cancer models. Cancer Res 2016;76:3003–13. 10.1158/0008-5472.CAN-15-1313. [DOI] [PubMed] [Google Scholar]
- 70. Chu CE, Sjöström M, Egusa EA, Gibb EA, Badura ML, Zhu J, et al. Heterogeneity in NECTIN4 expression across molecular subtypes of urothelial cancer mediates sensitivity to enfortumab vedotin. Clin Cancer Res 2021;27:5123–30. 10.1158/1078-0432.CCR-20-4175. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 71. Jiang CY, Hwang H, Epstein IY, Bakaloudi DR, Talukder R, Taylor AK, et al. Efficacy of erdafitinib before or after enfortumab vedotin in FGFR3-altered advanced urothelial cancer: analysis of the UNITE collaborative study. Oncologist 2025;30:oyaf342. 10.1093/oncolo/oyaf342. [DOI] [PMC free article] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
Supplementary Methods file contains extended methods, including the processing and analysis of multiple sequencing datasets.
Supplementary Tables file contains key information and supporting datasets. Table S1 | Clinicopathological features of the 22 MIBC patients. Table S2 | The top 50 most significantly differentially expressed genes between the 15 single cell clusters. Table S3 | GSEA pathway enrichment results for 4 spatial clusters. Table S4 | Clinical characteristics and spatial tumor state composition across samples. Table S5 | Bootstrap-based robustness analysis of spatial tumor state proportions. Table S6 | The top 50 most significantly differentially expressed genes between the 4 spatial clusters.
Supplementary Figures file contains Supplementary Figures S1–S15. Figure S1 | Pathology-guided spatial annotation across 22 MIBC tumors. Figure S2 | Validation of spatial deconvolution accuracy and cell-type compartmentalization. Figure S3 | Additional tumors demonstrating spatial CNV-based subclones and transcriptional divergence. Figure S4 | Cross-platform concordance of CNV profiles inferred from spatial transcriptomics and WES. Figure S5 | Transcriptional and functional features across the luminal-to-basal differentiation continuum. Figure S6 | Single-cell analysis recapitulates the spatial luminal-to-basal differentiation continuum. Figure S7 | Spatial validation of lineage-associated EGFR and FGFR3 expression patterns. Figure S8 | Lineage-associated expression of FGFR3–EGFR signaling and transcriptional regulators across normal tissue, tumor states, and independent cohorts. Figure S9 | Pathology-guided spatial annotation across all MIBC tumors. Figure S10 | Spatial tumor states associate with chemotherapy response and stress-related gene expression. Figure S11 | NECTIN4 expression is enriched in luminal tumors and declines at the tumor-normal boundary. Figure S12 | Subtype-associated genomic alterations and their relationship with basal program activity. Figure S13 | Associations between basal signature scores and clinical or molecular features across MIBC cohorts. Figure S14 | Pathologist-guided TLS annotation confirms spatial transcriptomic predictions. Figure S15 | Lineage-associated chemokine programs and tumor–TLS interactions.
Data Availability Statement
Publicly available bulk transcriptomic, copy-number, mutation, and clinical data from TCGA Bladder Cancer (BLCA) cohort were obtained from the UCSC Xena data portal (https://xenabrowser.net/datapages/). RNA-seq data from the IMvigor clinical trials were accessed under a Genentech Data Access Agreement and are available through the European Genome-phenome Archive (EGA) under controlled access (EGAD50000001100). RNA-seq data from the ABACUS trial were obtained through the EGA under controlled access (EGAD00001006205). Processed human single-cell and ST data in this study are available in the Gene Expression Omnibus (datasets GSE319536, GSE326225 and GSE326226). Raw cell line sequencing data are available in Sequence Read Archive (dataset PRJNA1443383). Any additional information required to reanalyze the data reported in this article is available from the lead contact upon request.