Abstract

   The prognosis for muscle‐invasive bladder cancer (MIBC) remains poor,
   and reliable prognostic markers have yet to be identified. Tertiary
   lymphoid structures (TLS) have been associated with favorable outcomes
   in certain cancers. However, the relationship between TLS and MIBC
   remains unclear. A multi‐omics approach is utilized, leveraging
   single‐cell RNA sequencing, spatial transcriptomics, bulk RNA
   sequencing, and immunohistochemistry, to investigate the roles of B
   cells and TLS in MIBC. These findings indicate that elevated levels of
   B cells and TLS correlate with improved prognoses in patients with
   MIBC, aligning with the robust antitumor immune responses observed in
   the TLS region. From a mechanistic perspective, CXCL13 serves as a
   critical cytokine for TLS formation in MIBC, primarily secreted by
   clonally expanded CXCL13+ T cells. This cytokine interacts with the
   CXCR5 receptor on NR4A2+ B cells, promoting TLS development. Plasma
   cells arising within the TLS microenvironment predominantly produce the
   IGHG antibody, potentially enhancing the phagocytic capabilities of
   C1QC+ macrophages. From an application standpoint, a TLS‐specific gene
   signature is developed that effectively predicts outcomes in MIBC and
   other cancers. This study highlights the prognostic potential of TLS in
   MIBC and reveals immune mechanisms, offering insights for personalized
   treatment strategies.

   Keywords: B cells, CXCL13, muscle‐invasive bladder cancer, tertiary
   lymphoid structures
     __________________________________________________________________

   This study employed multi‐omics approaches to investigate the role of B
   cells and TLS in muscle‐invasive bladder cancer (MIBC) and their
   potential clinical applications. The results indicate that both B cells
   and TLS are associated with improved prognosis and enhanced antitumor
   immunity in MIBC patients. Furthermore, the TLS biomarker developed can
   robustly predict the prognosis of MIBC.

   graphic file with name ADVS-12-2410998-g008.jpg

1. Introduction

   Urothelial carcinoma (UC) is a prevalent and lethal malignancy of the
   urinary system. Muscle‐invasive bladder cancer (MIBC), a clinical
   subtype of UC, is associated with a poor prognosis and limited
   treatment options.^[ [44]^1 ^] The primary treatments for MIBC
   currently include cisplatin‐based neoadjuvant chemotherapy and radical
   cystectomy, yet the overall 5‐year survival rate remains only 60–70%.^[
   [45]^2 ^] Recently, immune checkpoint inhibitors (ICB), such as
   anti‐PD‐1/PD‐L1 antibodies, have demonstrated significant efficacy in
   treating MIBC, particularly in patients with advanced disease. However,
   only a subset of patients exhibited a durable response to
   immunotherapy, underscoring the critical need for new biomarkers to
   more accurately predict therapeutic response.

   The presence of tertiary lymphoid structures (TLS) has been associated
   with improved clinical outcomes and enhanced responses to immunotherapy
   in several cancers.^[ [46]^3 , [47]^4 , [48]^5 ^] TLS are organized
   aggregates of immune cells that form in non‐lymphoid tissues,^[ [49]^6
   ^] with B cells being the predominant cell type. Some studies have
   examined the prognostic significance of TLS in UC cohorts. Research
   suggests that the clustering of highly expressed TLS‐related genes is
   associated with a better prognosis.^[ [50]^7 ^] In one cohort of UC
   patients treated with combined CTLA‐4 and PD‐1 inhibitors, responders
   exhibited higher baseline TLS density.^[ [51]^8 ^] However, in another
   cohort receiving similar treatment, baseline TLS did not significantly
   affect prognosis, likely due to corticosteroid use, which may have
   inhibited TLS formation.^[ [52]^9 ^] In non‐immunotherapy cohorts, TLS
   have been shown to predict both progression‐free survival and overall
   survival in UC patients.^[ [53]^10 , [54]^11 ^] Although several
   studies have investigated the prognostic value of TLS in MIBC, sample
   sizes, and cohort numbers have been relatively small compared to those
   in other cancers,^[ [55]^4 ^] highlighting the need for further
   validation and investigation.

   Beyond its prognostic significance, the complex mechanistic role of TLS
   in tumors is a critical area of tumor immunology research. For
   instance, studies have shown that the presence of TLS and
   tissue‐resident T cells within tumors can promote the efficacy of
   anti‐PD‐1 therapy in gastric cancer.^[ [56]^12 ^] In esophageal
   squamous cell carcinoma, Tfh cells within TLS express SEMA4D, which
   stimulates dendritic cells and enhances their maturation and
   antigen‐presenting capabilities.^[ [57]^13 ^] In hepatocellular
   carcinoma, reduced intra‐tumoral TLS abundance has been linked to mTOR
   signaling activation and dysregulated cell cycle progression, both
   associated with poor prognosis.^[ [58]^14 ^] However, the potential
   anti‐tumor mechanisms of TLS in MIBC remain unclear.

   This study aims to systematically investigate the role of B cells and
   TLS in the MIBC tumor microenvironment through a multi‐omics approach
   by integrating single‐cell RNA sequencing (scRNA‐seq), spatial
   transcriptomics (ST), and bulk RNA‐seq (Figure [59] 1A). We
   investigated the impact of B cell and TLS abundance on patient
   prognosis, elucidating the cellular composition and potential functions
   of TLS within MIBC tumors. Meanwhile, we explored the interactions of
   TLS within the MIBC tumor microenvironment. Finally, we developed a
   specific set of markers for TLS that can accurately predict the
   prognosis of MIBC. Our findings underscore the potential of B cells and
   TLS as biomarkers and enhance our understanding of the potential
   immunoregulatory mechanisms of TLS in MIBC.

Figure 1.

   Figure 1
   [60]Open in a new tab

   Cox regression analysis of cell subpopulations. A) The workflow diagram
   illustrates the data collection process, with in‐house data on the left
   and public data on the right. B) The UMAP plot shows the clustering of
   cells from 13 MIBC samples, grouped into three major categories. The
   accompanying heatmap highlights the expression of characteristic genes
   within these cell types. FC, fold change. C) Subpopulations. D,E)
   Forest plots of univariate and multivariate Cox regression analyses for
   major subpopulations. F,G) Forest plots for smaller subpopulations
   analyzed using univariate and multivariate Cox regression.

2. Results

2.1. Cell Abundance Serves as a Predictive Marker for Prognosis in UC
Patients Undergoing ICB Therapy as well as in MIBC Patients not Receiving ICB
Treatment

   To identify potential cell types relevant to prognosis, we conducted
   scRNA‐seq on fresh tumor samples from 13 MIBC patients treated at our
   hospital, including six cases from previously published datasets.^[
   [61]^15 ^] Through dimensionality reduction and clustering, we first
   categorized the cells into three main groups: epithelial cells, immune
   cells, and stromal cells (Figure [62]1B). Further analysis identified
   distinct subpopulations, including T/NK cells, plasma cells,
   neutrophils, monocytes, mast cells, macrophages, dendritic cells, B
   cells, fibroblasts, endothelial cells, and epithelial cells
   (Figure [63]1C). Differential expression analysis revealed specific
   marker genes for these subpopulations (log2 FC > 3, p < 0.05).

   Subsequently, we categorized the public transcriptomic cohorts based on
   whether ICB therapy had been administered, combining cohort‐2 UC (ICB,
   n = 386) and cohort‐6 MIBC (Non‐ICB, n = 667). Using the MCP‐counter
   tool and identified markers from the above analysis, we next
   quantitatively assessed the abundance of various cell types in cohort‐2
   and cohort‐6. Univariate and multivariate Cox regression analyses of
   major subpopulations (Figure [64]1D,E) demonstrated that the presence
   of immune cells significantly correlated with survival in cohort‐2 and
   cohort‐6, while stromal cells showed an opposite association. Similar
   analysis applied to smaller subpopulations (Figure [65]1F,G) revealed
   that cohort‐2, B cells and T/NK cells may contribute to improved
   prognoses, whereas mast cells were associated with increased risk, with
   B cells displaying the lowest hazard ratio (HR = 0.5), thereby halving
   the risk. Additionally, in cohort‐6, B cells, T/NK cells, and dendritic
   cells demonstrated a favorable prognostic effect, whereas fibroblasts
   were associated with poorer outcomes. These findings suggest that B
   cells may play a critical prognostic role in the tumor microenvironment
   of MIBC, potentially independent of other cell types. Kaplan–Meier
   analysis further confirmed that patients with higher B cell levels had
   improved prognoses (Figure [66]S1A,B, Supporting Information).
   Moreover, the expression levels of the key B cell markers, CD19 and
   CD20, can also predict patient outcomes (Figure [67]S1C, Supporting
   Information). Meanwhile, B cells were more abundant in female patients
   and in patients with high‐grade disease (Figure [68]S1D, Supporting
   Information). To explore the potential contribution from other immune
   cell types, we performed co‐expression analysis and found that B cells
   were most closely correlated with T/NK cell abundance (Figure [69]S1E,
   Supporting Information). Consistently, patients with elevated B cell
   levels showed upregulation of immune‐related pathways, such as pathway
   for “T cell activation” (Figure [70]S1F, Supporting Information).

2.2. The Abundance of TLS can Predict the Prognosis of UC Patients at our
Hospital

   Because the above analyses based on public cohorts suggested the
   prognostic potential of B‐cells, we sought to substantiate this
   conclusion by analyzing samples from patients treated at our hospital.
   We collected paraffin‐embedded specimens from 51 MIBC, 21
   non‐muscle‐invasive Bladder Cancer (NMIBC), and 6 upper tract
   urothelial carcinoma (UTUC) patients. These samples underwent
   hematoxylin and eosin (HE) staining and immunohistochemistry (IHC) for
   CD20, CD3, CD21, and CD68, key immune cell markers. We observed that B
   cells aggregate within certain areas, forming TLS (Figure [71] 2A). We
   validated the presence of TLS using multiplex immunofluorescence (IF)
   (Figure [72]2B), revealing that CD20 was highly concentrated, CD3 was
   dispersed within TLS, and CD68 was present both within and around TLS.
   TLS was abundant in urothelial carcinoma, with higher levels in MIBC
   and UTUC (Figure [73]S2A, Supporting Information). In MIBC, TLS of
   varying maturity were identified (Figure [74]2C), including iTLS
   (immature TLS) and mTLS (mature TLS), with mTLS further classified into
   FL1 (Primary follicle) and FL2 (Secondary follicle). iTLS comprised
   55.9% of the TLS in MIBC, FL1 37.6%, and FL2 6.5% (Figure [75]S2B,
   Supporting Information). Analysis of clinical characteristics revealed
   that TLS density was significantly higher in high‐grade UC
   (Figure [76]S2C, Supporting Information). Based on HE and IHC results,
   sample were categorized into tumor regions and stromal regions
   (Figure [77]S2D, Supporting Information), with TLS more likely to
   accumulate in the stromal region (Figure [78]S2E, Supporting
   Information), accounting for 55.2%. Moreover, TLS in the stromal region
   tended to be more mature (Figure [79]S2F, Supporting Information).
   Finally, we assessed the prognostic value of TLS in UC patients,
   finding that the number or density of TLS could predict patient
   prognosis (Figure [80]2D). Survival analysis based on mTLS density in
   the stromal region suggested that patients with higher stromal TLS
   density tended to have better prognoses, although this was not
   statistically significant (Figure [81]S2G, Supporting Information).

Figure 2.

   Figure 2
   [82]Open in a new tab

   TLS staining and clinical analysis of hospital specimens. A) Panoramic
   view of CD20 and HE staining, illustrating the distribution of TLS in
   MIBC samples and highlighting TLS of varying maturity, including iTLS,
   FL1, and FL2. B)Multicolor  IF staining of MIBC samples, including
   DAPI, CD20, PANCK, CD23, CD68, and CD3. C) Staining results for TLS at
   different maturity stages using HE, CD20, CD3, CD21, and CD68. D)
   Kaplan–Meier survival analysis showing the impact of TLS number and
   density on UC patient prognosis.

2.3. Spatial Transcriptomics Reveals Cellular Heterogeneity Between TLS‐Area
and no TLS‐Area

   To investigate the cellular composition and function of TLS in MIBC, we
   conducted spatial transcriptomics sequencing on seven paraffin‐embedded
   MIBC samples. Among these, P1, P3 to P5 were TLS‐positive (TLS‐pos), P6
   to P7 were TLS‐negative (TLS‐neg), and P2 was a lymph node metastasis
   sample. HE staining of the P1 sample revealed abundant TLS presence
   (Figure [83] 3A), and spatial gene expression analysis identified this
   sample as belonging to the luminal subtype (Figure [84]3B), where KRT19
   was used as a marker for bladder cancer, KRT20 for the luminal subtype,
   KRT5 and KRT14 for the basal subtype,^[ [85]^16 ^] COL1A1 served as a
   marker for fibroblasts, PECAM1 for endothelial cells, CD20 for B cells,
   and CD3 for T cells. Moreover, high expression levels of CD20 and CD3
   genes were observed specifically in the TLS region as expected.
   Subsequent IHC staining of adjacent ST sections for CD20 and CD3
   further validated the capacity of spatial transcriptomics to accurately
   characterize TLS (Figure [86]3C). Conversely, the P6 sample, identified
   as TLS‐neg (Figure [87]3D), showed gene expression consistent with the
   basal subtype and lacked clustered CD20 and CD3 expression
   (Figure [88]3E). IHC staining of the P6 sample also confirmed the
   absence of TLS (Figure [89]3F). Additionally, IHC validation was
   performed on four other samples (Figure [90]S3A–C, Supporting
   Information).

Figure 3.

   Figure 3
   [91]Open in a new tab

   Analysis of cellular heterogeneity between TLS and non‐TLS areas in
   bladder cancer. A) HE staining highlights the TLS‐rich region in the P1
   sample. B) Spatial expression maps of various cell marker genes,
   including KRT19, KRT20, KRT5, KRT14, COL1A1, PECAM1, CD20, and CD3, in
   the P1 sample. C) IHC staining for CD20 and CD3 in the P1 sample. D) HE
   staining of the P6 sample. E) Spatial expression of marker genes in the
   P6 sample. F) IHC staining for CD20 and CD3 in the P6 sample. G)
   Distribution of different cell types in the P1 ST. H) Comparison of
   cell content between TLS‐area and no TLS‐area. I) Multicolor
   immunofluorescence staining of the P1 sample. J) The upper panel shows
   the cell deconvolution results for the P1 sample, while the lower panel
   shows the corresponding staining for cell markers. “I” represents the
   TLS‐area, and “II” represents the no TLS‐area. “****” indicates p
   < 0.0001.

   Using SpaCET combined with single‐cell data from 13 cases, we conducted
   deconvolution analysis on the ST samples (Figure [92]3G). The
   deconvolution results showed significant enrichment of CD4+
   conventional T (CD4conv T), regulatory T cells (Tregs), CD8+ T cells,
   NK cells, B cells, plasma cells, and dendritic cells within TLS areas
   (Figure [93]3H), while monocytes, mast cells, neutrophils, fibroblasts,
   endothelial cells, and epithelial cells were more prevalent in non‐TLS
   areas. Furthermore, multicolor IF on the P1 sample (Figure [94]3I,J)
   revealed staining patterns closely aligned with the deconvolution
   results, reinforcing the validity of our findings. The TLS region
   indeed exhibited higher levels of CD20+ B cells, CD4+ T cells, CD8+ T
   cells, and Treg cells, along with fewer epithelial cells.

2.4. TLS Exhibits Stronger Anti‐Tumor Immune Responses

   To further analyze the function of TLS, we integrated six ST samples of
   primary MIBC. We used the Harmony method to correct for batch effects
   (Figure [95]S4A,B, Supporting Information). The cell subpopulations
   were annotated and classified into seven distinct groups (Figure [96]
   4A; Figure [97]S4D, Supporting Information). The Plasma subgroup
   predominantly expressed high levels of genes characteristic of plasma
   cells (Figure [98]S4C, Supporting Information), while the Myeloid/T
   subgroup displayed gene expression profiles typical of myeloid cells
   and T cells. The B/T/NK subgroup primarily showed co‐expression of
   characteristic genes for T cells, NK cells, and B cells. Analysis of
   cell proportions revealed that in the TLS‐pos group, the proportions of
   B/T/NK cells, and plasma cells were significantly increased, whereas
   endothelial cells were more prevalent in the TLS‐neg group
   (Figure [99]4B). Combining CD21 gene expression with IHC staining, we
   categorized TLS in the ST samples into iTLS and mTLS
   (Figure [100]S4E,F, Supporting Information). UMAP analysis revealed
   that TLS predominantly clustered in the B/T/NK cell region
   (Figure [101]4C), with notable distribution differences. We defined the
   B/T/NK cell region outside TLS as nTLS (non‐TLS, Figure [102]4D) and
   mapped these regions back onto the ST map, finding that nTLS primarily
   surrounded the TLS (Figure [103]4E; Figure [104]S4G, Supporting
   Information). The GSVA analysis of mTLS, iTLS, and nTLS
   (Figure [105]4F) revealed that mTLS was significantly enriched for
   “Type 1 IFN Receptor Binding,” suggesting that mTLS may play a role in
   antitumor immunity. In nTLS, enrichment of lymphangiogenesis and
   angiogenesis pathways suggests that immune cells are being recruited
   within this immune cluster, facilitating its development toward TLS
   formation. To enable additional characterization, we curated markers
   for tumor‐reactive T cells from two studies.^[ [106]^17 , [107]^18 ^]
   Using these markers, we scored the tumor‐reactive T cells within the
   TLS (Figure [108]4G) and found that mTLS had the highest tumor‐reactive
   scores for CD4 and CD8 T cells, while nTLS had the lowest. Further
   analysis of immune signaling pathways (Figure [109]4H) revealed that
   both mTLS and iTLS exhibited significant upregulation of cytotoxic
   scores, exhaust score, chemokine receptor activity, and
   inflammation‐promoting pathways. In mTLS, there was a significant
   upregulation of IFNG, GZMA, GZMH, GZMK, PDCD1, and BTLA, whereas in
   iTLS, expression levels of GNLY, GZMB, and HAVCR2 were elevated. In
   contrast, nTLS exhibited increased expression of CD276 genes
   (Figure [110]4I). Immunofluorescence for CD4, CD8, and PDCD1 on the P1
   sample confirmed the enrichment of PDCD1+ CD4 T cells in the mTLS and
   iTLS region (Figure [111]S4H, Supporting Information).

Figure 4.

   Figure 4
   [112]Open in a new tab

   Functional analysis of TLS. A) UMAP plot integrating six ST samples. B)
   Proportion analysis of subpopulations, categorized by TLS‐pos and
   TLS‐neg groups. C) Mapping the distribution of TLS on the UMAP. D)
   Identification of B/T/NK subpopulations. E) Visualization of mTLS,
   iTLS, and nTLS in the P1 sample. F) Radar chart presenting the GSVA
   results. G) Comparison of tumor‐reactive T cell scores. H) Heatmap
   depicting immune‐related pathway score. I) Heatmap showing the
   expression of cytotoxic and ICB‐related genes. J) A Venn diagram
   illustrating the differentially expressed genes. K) Results of
   univariate Cox analysis for CXCL13, IL32, and CCL19 across eight
   independent cohorts. L) Immunofluorescence staining of CXCL13 and CD20
   in FFPE samples. M) Spatial expression patterns of CXCL13 and CXCR5 in
   ST data. *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001.

   We subsequently explored the underlying mechanisms of TLS formation.
   Data for 189 cytokines were retrieved from the UniProt database
   (Table [113]S1, Supporting Information), and we conducted differential
   analyses between mTLS and nTLS (ST), iTLS and nTLS (ST), as well as
   between TLS‐pos and TLS‐neg samples (Bulk RNA‐seq). By intersecting the
   sets of differentially expressed genes (log2 FC > 0.3), we identified
   consistent upregulation of CXCL13, CCL19, and IL32 within the mTLS,
   iTLS, and TLS‐pos groups (Figure [114]4J). To further investigate the
   therapeutic potential of CXCL13, CCL19, and IL32 for urothelial
   carcinoma, we performed Cox survival analyses in eight distinct
   cohorts. Our results showed that CXCL13 had a hazard ratio (HR) below 1
   across all cohorts, with p‐values less than 0.05 in six of these
   cohorts (Figure [115]4K). Notably, IF staining of CXCL13 verified its
   expression in TLS (Figure [116]4L). Moreover, significant expression of
   CXCL13 and its receptor CXCR5 was observed in TLS within the ST samples
   (Figure [117]4M). CXCL13 is a key chemokine for B cells and plays a
   crucial role in TLS formation.^[ [118]^19 ^] Previous studies have also
   highlighted the prognostic value of CXCL13 in bladder cancer^[ [119]^20
   ^]; however, these studies included only two cohorts. Our study,
   utilizing eight cohorts, demonstrates the potential clinical
   translational capability of CXCL13.

2.5. CXCL13+ T Cells are Upregulated in TLS‐Pos Samples, Correlating with a
Stronger Anti‐Tumor Response

   To further investigate the relationship between TLS and the tumor
   immune microenvironment, we categorized our hospital's scRNA‐seq
   samples into TLS‐pos and TLS‐neg groups based on the presence of TLS in
   the corresponding paraffin‐embedded samples (Figure [120]S5A,
   Supporting Information). Four samples lacked TLS, while the remaining
   samples contained TLS, with at least one mTLS present (Figure [121]S5B,
   Supporting Information). Dimensionality reduction and clustering
   revealed distinct differences in the proportions of cell subpopulations
   between the two groups (Figure [122]S5C, Supporting Information). The
   TLS‐neg group was enriched with mast cells, and neutrophils, whereas
   the TLS‐pos group exhibited higher proportions of plasma cells, and B
   cells (Figure [123]S5D, Supporting Information).

   Since UMAP analysis revealed that CXCL13 was primarily expressed in the
   T/NK subpopulation (Figure [124]S5E, Supporting Information), we
   focused our analysis on T/NK cells. We subsequently performed
   deconvolution analysis of T/NK cells in the ST samples and found a
   higher abundance of these cells in the TLS‐pos group (Figure [125]S6A,
   Supporting Information). This finding was further validated using
   adjacent tissue sections. Drawing from previous studies,^[ [126]^21 ,
   [127]^22 ^] the T/NK subpopulation was further subdivided into six
   groups (Figure [128] 5A,B). Both CXCL13+ CD4conv T cells and CXCL13+
   CD8 T cells were significantly upregulated in the TLS‐pos group
   (Figure [129]5C,D). To analyze T cell evolution, slingshot trajectory
   analysis was performed (Figure [130]S6B, Supporting Information),
   revealing that CXCL13+ T cells were more likely to appear in the later
   stages of cell development. In TLS‐pos samples, T cells tended to
   differentiate into CXCL13+ T cells (Figure [131]5E). To further explore
   the transcriptomic differences between CXCL13⁺ and CXCL13⁻ cells,
   differential expression analysis was performed on CD4conv T and CD8 T
   cells. The results showed that IL7R, KLF2, and ANXA1 were upregulated
   in CXCL13‐ CD4conv T cells, while TIGIT, TOX, RBPJ, and CCL4 were
   upregulated in CXCL13+ CD4conv T cells. Similarly, IL7R, KLF2, and
   ANXA1 were upregulated in CXCL13‐ CD8 T cells, whereas ITGAE, ENTPD1,
   and HAVCR2 were upregulated in CXCL13+ CD8 T cells. IL7R and KLF2
   contribute to maintaining T cell quiescence,^[ [132]^23 , [133]^24 ^]
   while ENTPD1, ITGAE, and PDCD1 are markers of tumor‐reactive T cells.^[
   [134]^18 , [135]^25 , [136]^26 ^] Gene trajectory analysis indicated
   that CXCL13 expression was initially low but significantly increased
   during cell development, whereas IL7R and KLF2 showed opposite
   expression patterns (Figure [137]5H). ITGAE was co‐expressed with
   CXCL13 in CD8 T cells, a pattern also observed in our bulk data (R =
   0.64; Figure [138]S6C, Supporting Information) and in lung cancer
   studies.^[ [139]^27 ^] Research showed that ITGAE+ CD8 T cells were
   upregulated in TLS‐pos gastric cancer.^[ [140]^12 ^] In our samples, ST
   and IF staining revealed that ITGAE+ CD8 T cells were abundant in
   TLS‐pos MIBC (Figure [141]S6D,E, Supporting Information). To further
   investigate the relationship between these markers and the antitumor
   response of T cells, we used TRUST4 to predict TCR clones and
   categorized them into three levels: 1, 2–3, and greater than 3. As
   clonal expansion increased, the positivity rates of CXCL13 and ITGAE
   significantly increased, while those of IL7R and KLF2 decreased
   (Figure [142]5G). Clonal expansion, representing the generation of
   tumor‐reactive T cells, suggests that CXCL13 and ITGAE are potential
   markers for tumor‐reactive T cells in MIBC. To further validate the
   hypothesis that CXCL13⁺ T cells are tumor‐reactive, we scored T cells
   for tumor‐reactive markers and found that CXCL13⁺ T cells exhibited
   significantly higher reactivity compared to CXCL13⁻ T cells
   (Figure [143]5I). Immune‐related pathway analysis revealed that
   check‐point, CCR, and APC‐co‐stimulation pathways were upregulated in
   CXCL13+ T cells, while para‐inflammation was downregulated
   (Figure [144]5J). Finally, using differential genes from four cell
   types for MCP scoring and combining survival data for KM analysis, we
   found that samples with high CXCL13+ CD8 T cells were associated with
   better patient prognosis, while CXCL13‐ CD8 T cells had the opposite
   effect (Figure [145]5K; Figure [146]S6F, Supporting Information).
   However, no obvious differences were observed between CXCL13+ CD4conv T
   and CXCL13‐ CD4conv T cells (Figure [147]S6G, Supporting Information).
   Our results suggest that CXCL13+ CD8 T cells are upregulated and
   undergo clonal expansion in TLS‐pos samples, potentially contributing
   to improved patient outcomes.

Figure 5.

   Figure 5
   [148]Open in a new tab

   Analysis of T/NK cell subpopulations. A) UMAP plot depicting T/NK cell
   subpopulations. B) Expression levels of CD3D, CD8A, FOXP3, GNLY, CD4,
   and CXCL13 in T/NK cells. C) Comparison of the proportion of CXCL13+
   cells between TLS‐pos and TLS‐neg samples. D) Analysis of
   tissue‐specific biases in cell subpopulations. E) Slingshot trajectory
   analysis of T cell development, grouped by CXCL13 expression and TLS
   status. F) Differential gene expression analysis between CXCL13+ and
   CXCL13‐ cells. G) Correlation between TCR clonality and the proportion
   of CXCL13, ITGAE, IL7R, and KLF2 positive cells. H) Gene expression
   trends for CXCL13, IL7R, KLF2, and ITGAE during T cell development. I)
   Tumor reactivity score comparison between CXCL13+ and CXCL13‐ T cells.
   J) Heatmap illustrating immune pathway scores in CXCL13+ and CXCL13‐ T
   cells. K) Kaplan–Meier survival analysis based on CXCL13 expression
   levels in CD8 T cells. ****p < 0.0001.

2.6. CXCL13 is Likely to Target the CXCR5 on NR4A2 B Cells, Thereby
Catalyzing the Formation of TLS in MIBC

   The interaction between CXCL13 and CXCR5 constitutes a critical pathway
   for TLS development.^[ [149]^19 ^] We conducted an analysis of CXCR5
   expression, identifying its primary expression on T/NK and B cells
   (Figure [150]S7A, Supporting Information). Subsequently, we classified
   B cells into distinct subgroups: DUSP4 Atm B (Atypical memory B), ITGB1
   SwBm (Switched memory B), NR4A2 B, PTPRJ B, and TCL1A naive B
   (Figure [151]S7B,C, Supporting Information). Comparative analysis of
   TLS‐neg and TLS‐pos samples revealed a higher prevalence of NR4A2 B
   cells in the TLS‐pos group (Figure [152]S7D, Supporting Information).
   Further, CXCR5 exhibited increased expression specifically within NR4A2
   B cells (Figure [153]S7E, Supporting Information), an upregulation of
   CXCR5 was observed in TLS‐pos samples (Figure [154]S7F, Supporting
   Information). Spatial distribution patterns showed that NR4A2 B cells,
   along with CXCL13+ CD4conv T cells and CXCL13+ CD8 T cells, were
   predominantly found in TLS regions (Figure [155]S7G, Supporting
   Information). Notably, in TLS‐pos samples, co‐localization occurred
   between NR4A2 B cells and either CXCL13+ CD4conv T cells or CXCL13+ CD8
   T cells (Figure [156]S7H, Supporting Information). These findings imply
   that CXCL13 may bind to CXCR5 on NR4A2 B cells in MIBC, promoting the
   formation of TLS.

2.7. Plasma Cells in TLS‐Pos Samples Secrete More IGHG, Correlating with
Improved Prognosis

   Some B cells in mature TLS differentiate into plasma cells, producing
   antibodies that induce antitumor immunity.^[ [157]^28 ^] Thus, we aimed
   to investigate the differences in plasma cells between TLS‐pos and
   TLS‐neg MIBC samples. Both the deconvolution results of plasma cells
   (Figure [158] 6A) and the proportion of plasma cells (Figure [159]6B)
   demonstrate that plasma cells are more abundant in TLS‐pos samples. Our
   analysis identified two distinct plasma cell subgroups: plasmablasts
   and pan‐plasma (Figure [160]6C). In MIBC, plasma cells primarily
   express IGHG, with IGHA being expressed to a lesser extent
   (Figure [161]6D). Differential expression (DE) score analysis of the
   pan‐plasma subgroup revealed that antibody production‐related signaling
   pathways are enriched in TLS‐pos samples, while stress‐related pathways
   are more prevalent in TLS‐neg samples (Figure [162]6E). Therefore, we
   aimed to examine whether there are differences in antibody expression
   between the two groups. The results revealed that IGHG1 and IGHG3 were
   significantly upregulated in TLS‐pos samples, while IGHA1 and IGHA2
   were upregulated in TLS‐neg samples (Figure [163]6F). The number of
   clones is indicative of tumor reactivity, prompting us to examine the
   differences in IGH clone numbers between the two groups. Using TRUST4
   to predict BCR clones, we identified an intriguing pattern: samples
   with higher clonal expansion (>3) exhibited upregulation of IGHG1 and
   IGHG3, while IGHA1 and IGHA2 were downregulated (Figure [164]6G). This
   finding suggests that IGHG antibodies are the primary anti‐tumor
   antibodies produced by plasma cells within the tumor microenvironment.
   Survival analysis of IGH genes in the TCGA‐BLCA cohort further
   demonstrated that high expression of IGHG1 and IGHG3 is associated with
   better prognosis, while high expression of IGHA1 and IGHA2 correlates
   with poorer outcomes (Figure [165]6H). To conclude, we conducted IHC
   staining for IGHG and observed a significant abundance of IGHG
   expression in TLS‐pos samples, while its presence was notably
   diminished in TLS‐neg samples (Figure [166]6I).

Figure 6.

   Figure 6
   [167]Open in a new tab

   Plasma cell analysis. A) Convolution results of plasma cells in ST
   samples. B) Proportional analysis of plasma cells. C) UMAP plot
   depicting plasma cell subgroups. D) Expression levels of key plasma
   cell marker genes (MZB1, MKI67, IGHD, IGHM, IGHG1, IGHG3, IGHA1,
   IGHA2). E) Pathway enrichment analysis of pan‐plasma cells. F)
   Differential gene expression of IGHG and IGHA. G) Expression patterns
   of IGH genes across different BCR clone groups. H) Survival analysis
   based on IGH gene expression in the TCGA‐BLCA cohort. (I) IHC staining
   images of IGHG. ***p < 0.001, ****p < 0.0001.

   Given that antibodies in the tumor microenvironment can mediate
   antibody‐dependent cellular phagocytosis (ADCP) by macrophages, we
   further analyzed macrophage subpopulations.^[ [168]^29 ^] We
   categorized myeloid cells into 11 subgroups, including 4 macrophage
   subpopulations (Figure [169]S8A, Supporting Information). All
   macrophage subpopulations were found to be more enriched in TLS‐pos
   samples (Figure [170]S8B, Supporting Information). CD68 is a marker for
   tumor‐infiltrating macrophages. IHC analysis confirmed that TLS‐pos
   MIBC tumors contain a higher density of macrophages (Figure [171]S8C,D,
   Supporting Information). Phagocytic function scoring revealed that the
   Macro_C1Q1 subgroup exhibited significantly increased phagocytic
   activity in TLS‐pos samples (Figure [172]S8E, Supporting Information).
   Macrophages are functionally classified into the tumor‐suppressive M1
   phenotype and the tumor‐promoting M2 phenotype, with the potential for
   phenotypic switching between the two.^[ [173]^30 ^] Polarization
   analysis of Macro_C1Q1 revealed that these cells are more likely to
   polarize toward the M1 subtype in TLS‐pos samples, whereas they tend to
   polarize toward the M2 subtype in TLS‐neg samples (Figure [174]S8F,
   Supporting Information). Our findings suggest that in TLS‐pos MIBC,
   plasma cells secrete higher levels of IGHG, which may enhance the
   phagocytic activity of C1QC+ macrophages, thereby amplifying the
   antitumor immune response.

2.8. TLS Markers can Predict Prognosis in MIBC and Other Tumor Immunotherapy
Cohorts

   Identifying representative markers for TLS is a central focus in
   research^[ [175]^6 ^]; however, no TLS‐specific markers have been
   identified for MIBC to date. We used samples P1, P4, and P5 as the
   discovery samples, and sample P3 as the validation sample, to identify
   TLS signature markers through differential analysis of TLS‐area and no
   TLS‐area. This differential analysis revealed seven common differential
   genes (p < 0.05, logFC > 0.7) (Figure [176] 7A), including CXCL13,
   CCL19, MS4A1, LTB, CD37, CORO1A, and IKZF1 (Figure [177]7B;
   Figure [178]S9A, Supporting Information), collectively termed RMYY‐TLS.
   Utilizing MCP and RMYY‐TLS for evaluation in ST, we found that RMYY‐TLS
   effectively differentiated between TLS‐area and no TLS‐area in both
   discovery and validation samples (p < 0.001, Figure [179]7C,D). To
   further validate the presence of TLS, we analyzed datasets from renal
   cell carcinoma, hepatocellular carcinoma, and nasopharyngeal carcinoma
   to assess the capacity of RMYY‐TLS to distinguish TLS‐area
   (Figure [180]S9B, Supporting Information). In the ST samples, areas
   with high RMYY‐TLS scores consistently overlapped with TLS regions
   identified through corresponding HE staining (Figure [181]S9C,D,
   Supporting Information).

Figure 7.

   Figure 7
   [182]Open in a new tab

   Construction of the TLS marker RMYY‐TLS and its prognostic value
   evaluation. A) Venn diagram illustrating the overlap of differential
   genes across various ST samples. B) Volcano plot comparing differential
   gene expression between no TLS‐area and TLS‐area. C) The upper panel
   shows the distribution of RMYY‐TLS scores in the discovery samples,
   while the lower panel compares RMYY‐TLS scores between the two areas.
   D) Validation sample results. E) Survival analysis based on RMYY‐TLS
   scores in UC and MIBC patients. F) Survival analysis of RMYY‐TLS scores
   in additional immunotherapy cohorts, including those with melanoma and
   renal cell carcinoma. ***p < 0.001.

   Subsequently, we applied MCP and RMYY‐TLS scoring to two ICB UC cohorts
   and seven no ICB MIBC cohorts, followed by survival analysis using
   clinical data (Figure [183]7E). The results showed that RMYY‐TLS could
   predict tumor prognosis in both ICB cohorts (IMvigor210 and
   [184]GSE176307) (p < 0.05) and in non‐ICB cohorts ([185]GSE31684,
   [186]GSE48075, and [187]GSE32894). Although not statistically
   significant, the remaining four cohorts ([188]GSE32548, [189]GSE13507,
   E_MTAB_1083, TCGA_BLCA) displayed a trend where higher RMYY‐TLS levels
   correlated with better prognosis. When these cohorts were combined,
   patients with high RMYY‐TLS levels had significantly improved prognoses
   (p < 0.001). BCG therapy is a widely used immunotherapy for NMIBC
   cancer. Accordingly, we applied RMYY‐TLS to assess the prognostic
   outcomes in a BCG‐treated cohort. Although the trend did not reach
   statistical significance, patients with high RMYY‐TLS scores still
   exhibited a trend toward improved prognosis (Figure [190]S9E,F,
   Supporting Information). Our findings suggest that RMYY‐TLS possesses
   superior predictive ability in patients receiving immunotherapy,
   prompting further examination across five additional immunotherapy
   cohorts. In melanoma, a representative hot tumor, patients with high
   RMYY‐TLS levels derived significant benefit from immunotherapy in four
   melanoma cohorts (Figure [191]7F). Conversely, in renal cell carcinoma,
   a representative cold tumor, higher RMYY‐TLS levels were associated
   with poorer prognosis (Figure [192]7F). Our results indicate that TLS
   markers have potential clinical application value.

3. Discussion

   This study highlights the essential role of B cells and TLS within the
   tumor microenvironment of MIBC. Our findings indicate that elevated
   levels of B cells and TLS are associated with improved prognoses in
   MIBC patients. Notably, CXCL13 plays a significant role in TLS
   formation in MIBC and has been demonstrated to reduce mortality risk
   across various MIBC cohorts. CXCL13 secreted by CXCL13+ T cells,
   facilitates the clustering of NR4A2 B cells, thus fostering TLS
   development. TLS generates more IGHG subtype plasma cells, enhancing
   macrophage phagocytic function and activating anti‐tumor immunity.
   Moreover, our research has resulted in the development of a novel TLS
   biomarker for MIBC that effectively predicts patient outcomes. Overall,
   our results substantially advance the understanding of TLS's role in
   MIBC (Figure [193] 8 ).

Figure 8.

   Figure 8
   [194]Open in a new tab

   Proposed working model. In TLS‐pos MIBC, CXCL13+ T cells undergo
   expansion, leading to increased secretion of CXCL13. This cytokine
   interacts with the CXCR5 receptor on NR4A2+ B cells, facilitating TLS
   formation. The plasma cells generated in this microenvironment
   predominantly produce the IGHG antibody, which enhances the phagocytic
   activity of C1QC macrophages. In contrast, TLS‐neg MIBC is marked by a
   higher prevalence of CXCL13‐ T cells and IGHA‐subtype plasma cells. The
   presence of TLS induces stronger anti‐tumor immunity, leading to a
   better prognosis in MIBC.

   B cells and TLS are widely recognized as prognostic markers across the
   same cancers.^[ [195]^4 ^] However, large‐scale, cross‐cohort
   validation studies in MIBC remain limited.^[ [196]^10 , [197]^11 ,
   [198]^31 ^] In this study, we used a large no ICB MIBC cohort to
   conduct survival analysis, confirming that elevated B cell levels are
   associated with improved survival outcomes in MIBC patients.
   Additionally, B cells have been significantly upregulated in ICB
   responders in melanoma,^[ [199]^32 ^] renal cell carcinoma,^[ [200]^33
   ^] and sarcoma.^[ [201]^34 ^] Given that ICB only elicits durable
   responses in 20–25% of metastatic UC patients,^[ [202]^1 ^] selecting
   patients using appropriate biomarkers is crucial for optimizing ICB
   therapy. Our results demonstrate that B cells are the strongest
   predictors of response in the ICB‐treated UC cohort. While previous
   studies on the bladder cancer immune microenvironment have primarily
   focused on T cells, myeloid cells, and fibroblasts,^[ [203]^35 ^] our
   findings suggest that research efforts should increasingly prioritize B
   cells in bladder cancer immunology. Furthermore, as B cells are
   predominantly localized within TLS in urothelial carcinoma, TLS
   provides a practical clinical marker for quantifying B cell levels. Our
   hospital cohort further validated that TLS can serve as a prognostic
   marker in urothelial carcinoma patients, highlighting the translational
   potential and broad clinical utility of TLS.

   Spatial transcriptomics is a powerful tool for investigating the
   composition and function of tumor‐associated TLS. However, to date, no
   studies have explored its application in MIBC.^[ [204]^36 ^] In this
   study, we employed ST to characterize the cellular composition of TLS
   in MIBC. Compared to renal cancer,^[ [205]^37 ^] the TLS in MIBC
   exhibited a higher proportion of CD8+ T cells, Tregs, NK cells, DC
   cells, and plasma cells, potentially indicative of immune activation.
   In the TLS regions, there is an observed increase in tumor‐reactive T
   cells and significant activation of immune signaling pathways. This
   phenomenon elucidates the positive correlation between TLS and the ICB
   response in MIBC.^[ [206]^8 ^] Cytokines play a pivotal role in the
   formation of TLS,^[ [207]^38 ^] with CXCL13 emerging as a critical
   factor specifically in MIBC. Furthermore, CXCL13 has shown strong
   prognostic value in cohorts with MIBC and UC. As a B cell chemokine,
   CXCL13 has also been demonstrated in other cancer types to effectively
   induce the formation of TLS.^[ [208]^39 ^] In a mouse ovarian cancer
   model, recombinant CXCL13 protein induced TLS formation and improved
   survival.^[ [209]^40 ^] Similarly, in a colorectal cancer liver
   metastasis model, intra‐tumoral injection of CCL19/CXCL13 promoted TLS
   formation.^[ [210]^41 ^] Future studies could focus on inducing TLS
   formation by enhancing CXCL13 expression as a potential therapeutic
   strategy for MIBC.

   While TLS is known to reshape the immune microenvironment,^[ [211]^5 ^]
   its mechanism of action in MIBC remains unclear. In our study, T cells
   in TLS‐pos MIBC were more likely to differentiate into CXCL13+ T cells.
   Compared to CXCL13‐ T cells, CXCL13+ T cells expressed more
   tumor‐reactive markers and exhibited significant clonal expansion.
   Further analysis revealed that a higher proportion of CXCL13+ CD8+ T
   cells was associated with better prognosis, whereas CXCL13‐ CD8+ T
   cells correlated with poorer outcomes. CXCL13+ T cells have also been
   upregulated in other tumors responsive to immunotherapy^[ [212]^21 ^]
   and are positively correlated with tumor neoantigen reactivity.^[
   [213]^18 ^] These cells secrete CXCL13, which targets CXCR5 on NR4A2+ B
   cells and may promote TLS formation. Our findings showed that plasma
   cells were enriched in TLS‐pos MIBC, producing more IGHG antibodies and
   fewer IGHA antibodies, a pattern linked to improved prognosis. Recent
   pan‐cancer analyses have indicated that plasma cells secrete more IGHG
   in tumors and more IGHA in normal tissues.^[ [214]^42 ^] Antibodies
   play a crucial role in the tumor microenvironment, particularly in
   mediating ADCP by macrophages.^[ [215]^29 ^] Our results demonstrated
   that C1QC+ macrophages were upregulated in TLS‐pos MIBC, exhibited
   enhanced phagocytic function, and polarized toward the M1 phenotype,
   amplifying the antitumor immune response. As a result, the presence of
   TLS is linked to stronger antitumor immunity and better prognosis.

   TLS gene signatures have become a central focus in TLS research, with
   signatures already developed for colorectal cancer,^[ [216]^43 ^]
   breast cancer,^[ [217]^44 ^] gastric cancer,^[ [218]^45 ^] ovarian
   cancer,^[ [219]^46 ^] renal cancer,^[ [220]^37 ^] and other cancers.^[
   [221]^5 ^] However, a defined TLS signature for MIBC has yet to be
   established. In this study, we identified a novel TLS‐specific gene
   signature, RMYY‐TLS, which includes CXCL13, CCL19, MS4A1, LTB, CD37,
   CORO1A, and IKZF1. RMYY‐TLS was successfully validated in two ICB UC
   cohorts and seven no ICB MIBC cohorts, with high RMYY‐TLS scores
   correlating with improved survival outcomes, particularly in the ICB
   cohorts. BCG is a standard immunotherapy for NMIBC,^[ [222]^47 ^] and
   the RMYY‐TLS signature also provides prognostic value for NMIBC
   patients undergoing BCG treatment. To further confirm the broad
   applicability of this signature, we validated its prognostic value in
   immunotherapy‐treated melanoma cohorts and renal cancer cohort. This
   study represents the largest multicenter effort to date in validating
   TLS as a biomarker in MIBC, underscoring the significant clinical
   translational potential of RMYY‐TLS.

   This study underscores the pivotal role of B cells and TLS in MIBC.
   Higher levels of B cells and TLS are linked to a more effective
   antitumor immune response, resulting in significantly improved
   prognoses. Furthermore, the TLS‐specific gene signature identified in
   this study shows strong potential for clinical application. Overall,
   this research provides new insights into MIBC immunotherapy,
   highlighting the importance of TLS as both a biomarker and a
   therapeutic target.

4. Experimental Section

Sample Collection

   This study cohort comprised urothelial carcinoma specimens collected
   from Peking University People's Hospital between 2015 and 2024. The
   clinical follow‐up period for the patients spanned from the date of
   surgery until November 30, 2023. These samples were surgically
   resected, and pathologically confirmed. Fresh tissues intended for
   single‐cell sequencing were preserved in a tissue preservation
   solution, while bulk RNA sequencing samples were immediately stored at
   −80 °C. A portion of the samples was paraffin‐embedded and stored at
   room temperature. Informed consent was obtained from all patients
   before participation in the study. A total of 91 tumor samples were
   collected, and the detailed usage of each sample is provided in
   Table [223]S2 (Supporting Information). This study involves human
   participants and was approved by the Ethics Committee of Peking
   University People's Hospital (ID: 2019PHB133‐01). Written informed
   consent was obtained from all patients and from all relatives of organ
   donors. All research procedures were conducted in accordance with the
   ethical standards of the Institutional Review Board.

Collection of Non‐Immunotherapy Public Data

   Seven bladder cancer cohorts were downloaded and curated from the
   Cancer Genome Atlas (TCGA, [224]https://cancergenome.nih.gov/), the
   Gene Expression Omnibus (GEO, [225]http://www.ncbi.nlm.nih.gov/geo/),
   and ArrayExpress ([226]https://www.ebi.ac.uk/biostudies/arrayexpress).
   After screening, only MIBC samples with complete survival and RNA‐seq
   data were retained. The included cohorts were TCGA‐BLCA (n = 366),
   [227]GSE13507 (n = 61),^[ [228]^48 ^] [229]GSE31684 (n = 66),^[
   [230]^49 ^] [231]GSE32548 (n = 38),^[ [232]^50 ^] [233]GSE32894 (n =
   51),^[ [234]^51 ^] [235]GSE48075 (n = 73),^[ [236]^52 ^] and
   E‐MTAB‐1803 (n = 73).^[ [237]^53 ^] Cohort‐6 refers to the combined
   non‐ICB MIBC datasets, comprising TCGA‐BLCA, [238]GSE31684,
   [239]GSE32548, [240]GSE32894, [241]GSE48075, and E‐MTAB‐1803.
   Furthermore, spatial transcriptomics datasets for renal cell
   carcinoma,^[ [242]^37 ^] hepatocellular carcinoma,^[ [243]^54 ^] and
   nasopharyngeal carcinoma were acquired.^[ [244]^55 ^]

Collection of Immunotherapy Public Data

   IMvigor210 was a clinical cohort study investigating anti‐PD‐L1 therapy
   in patients with metastatic urothelial carcinoma (mUC).^[ [245]^56 ^]
   The data for UC patients were obtained from the R package
   “IMvigor210CoreBiologies,” and 298 samples were retained after
   screening. Additionally, the [246]GSE176307 (n = 88) dataset, which
   includes an MIBC cohort treated with PD‐1/PD‐L1 inhibitors, was
   downloaded from GEO. [247]GSE154261 represents a cohort of NMIBC
   patients treated with BCG, with data sourced from the GEO database.^[
   [248]^57 ^] From the TIGER database
   ([249]http://tiger.canceromics.org/), four melanoma immunotherapy
   cohorts were downloaded: [250]GSE91061 (n = 101),^[ [251]^58 ^]
   phs000452 (n = 153),^[ [252]^59 ^] PRJEB23709 (n = 91),^[ [253]^60 ^]
   and [254]GSE100797 (n = 25).^[ [255]^61 ^] The renal cell carcinoma
   cohort Braun_2020 (n = 311)^[ [256]^62 ^] was also downloaded. Cohort‐2
   refers to the combined ICB‐treated UC cohorts from IMvigor210 and
   [257]GSE176307.

4.1. scRNA‐Seq Workflow

   To begin, single cells were isolated from fresh tissue samples. Using
   the 10x Genomics Chromium platform, single cells were encapsulated with
   barcoded gel beads within a microfluidic chip, creating Gel
   Bead‐in‐Emulsions (GEMs). Inside these GEMs, reverse transcription was
   performed to convert mRNA into cDNA, which was subsequently tagged with
   cell‐specific barcodes and unique molecular identifiers (UMIs). After
   the GEMs were lysed, the cDNA was released and subjected to PCR
   amplification to generate sequencing libraries. These libraries were
   then sequenced using the Illumina platform to obtain high‐throughput
   single‐cell transcriptome data.

scRNA‐Seq Data Analysis

   Single‐cell RNA sequencing data generated by the 10x Genomics platform
   were processed by aligning raw reads to the human genome (GRCh38) using
   Cell Ranger, generating a feature‐barcode matrix. Low‐quality cells
   (those with fewer than 500 genes, more than 5000 genes, or over 20%
   mitochondrial content) were filtered out to ensure data integrity. Data
   normalization was performed using the Seurat package, where the
   “NormalizeData” and “ScaleData” functions were applied to minimize
   technical noise. Harmony was employed to correct batch effects.^[
   [258]^63 ^] Principal component analysis (PCA) was used for
   dimensionality reduction, followed by the “FindNeighbors” and
   “FindClusters” functions for unsupervised clustering. Visualization of
   the clustering results was achieved through UMAP or TSNE. Differential
   gene expression analysis was conducted using the “FindAllMarkers”
   function, and cell populations were annotated using a combination of
   classical marker genes and those from published literature.

Single‐Cell TCR and BCR Prediction

   The TRUST4 algorithm was employed to reconstruct immune receptor
   repertoires from single‐cell RNA‐seq data produced by the 10x Genomics
   Chromium platform.^[ [259]^64 ^] TRUST4 extracts candidate TCR and BCR
   sequences from FASTQ files, prioritizing assembly by abundance for
   increased speed and sensitivity. The resulting single‐cell
   transcriptomic data and VDJ recombination data were integrated into the
   R environment. Using the scRepertoire package,^[ [260]^65 ^] the
   “combineTCR” and “combineBCR” functions aggregated TCR and BCR
   sequences by cell barcode, allowing for the identification of
   clonotypes. These clonotypes were then integrated with scRNA‐seq data
   for further analysis and visualization using Seurat. Clonotypes with a
   frequency of two or more were considered clonal expansions, typically
   associated with tumor‐reactive cells.

Pseudotime Analysis

   To infer lineage and pseudotime from single‐cell RNA‐seq data, the
   Slingshot algorithm was applied.^[ [261]^66 ^] Initially, data reduced
   via UMAP were clustered using a Gaussian mixture model. Slingshot then
   constructed a minimum spanning tree based on the cluster centers to
   identify global lineage structures. Smooth branching curves were fitted
   along each lineage using a principal curve approach, allowing for
   pseudotime inference. Known biological start and end states were
   provided to guide the analysis and ensure biologically meaningful
   trajectory inference.

Spatial Transcriptomics Sequencing

   In order to profile the spatial gene expression within MIBC employed
   spatial transcriptomics technology. RNA was extracted from FFPE tissue
   sections, ensuring that the RNA Integrity Number exceeded 7. The
   sections were affixed to the capture area of a gene expression slide,
   followed by H&E staining and imaging. Tissue permeabilization was
   tailored to bladder cancer‐specific conditions, enabling the capture of
   mRNA from released cells via specific probes located in the microwells
   of the 10x Visium Spatial Gene Expression Kit. Library construction
   followed the 10x Visium protocol, with cDNA synthesis and
   amplification. The resulting cDNA libraries were subjected to
   high‐throughput sequencing on an Illumina platform.

Spatial Transcriptomics Analysis

   The raw Visium ST data were processed using 10x Genomics' Space Ranger
   software. The sequencing reads were aligned to the human reference
   genome (GRCh38), producing a matrix of feature barcodes. Data were
   subsequently scaled and normalized using the “scTransform” function to
   analyze highly variable features. To address batch effects across
   integrated ST samples, the Harmony R package was employed.^[ [262]^63
   ^] Dimensionality reduction was performed through principal component
   analysis, followed by unsupervised clustering and data visualization
   using the “FindNeighbors,” “FindClusters,” and “RunUMAP” functions in
   the Seurat package. Additionally, convolution analysis was performed
   using SpaCET, leveraging in‐house scRNA‐seq data.^[ [263]^67 ^]

Bulk RNA Sequencing

   The tumor tissue was lysed using TRIzol RNA Isolation Reagents, and
   total RNA was extracted. The NEBNext Ultra RNA Library Prep Kit for
   Illumina was employed to construct the sequencing library. Poly(A) mRNA
   was enriched using Oligo d(T) magnetic beads, followed by fragmentation
   of the mRNA. Subsequently, first‐strand and second‐strand cDNA
   synthesis was performed, and the resulting double‐stranded cDNA was
   purified. The cDNA library was then subjected to end repair, ligation
   of Illumina sequencing adapters, and PCR amplification to enrich the
   library DNA. Finally, PCR products were purified using magnetic beads,
   and the library quality was assessed with an Agilent Bioanalyzer to
   ensure a correct and consistent size distribution, with fragment sizes
   of ≈350 bp. The qualified library was sequenced using the Illumina PE
   300 platform.

Bulk RNA Data Analysis

   Raw data quality was first assessed using FastQC to ensure high‐quality
   reads. Sequencing adapters were then trimmed with Cutadapt, and the
   cleaned reads were aligned to the GRCh38 reference genome using the
   STAR aligner. To standardize gene expression, TPM normalization was
   applied. Differential gene expression analysis was performed using
   DESeq2 to identify genes significantly upregulated or downregulated
   under varying conditions.

Enrichment Analysis

   Two enrichment analysis methods were employed: DE‐score and Gene Set
   Variation Analysis (GSVA).^[ [264]^68 ^] DE‐score quantifies gene
   expression enrichment by calculating the difference between the number
   of upregulated and downregulated genes within each enriched term. GSVA,
   a non‐parametric and unsupervised method, assesses pathway activity
   differences between samples based on gene set enrichment scores.

Survival Analysis

   The Kaplan–Meier method was used to evaluate the impact of various
   variables on survival time. The “survminer” package was utilized to
   determine the optimal cutoff for continuous variables, converting them
   into binary variables for more tailored analysis. Additionally,
   univariate and multivariate Cox regression analyses were conducted.
   Univariate Cox regression assessed the effect of individual variables
   on survival, while multivariate Cox regression evaluated the influence
   of multiple variables simultaneously, identifying independent
   predictors of survival.

MCPcounter

   MCPcounter was a transcriptome‐based computational method designed to
   robustly quantify the abundance of eight immune cell types and two
   stromal cell types.^[ [265]^69 ^] For each sample, abundance scores
   were derived from the gene expression matrix, facilitating direct
   comparison of the corresponding cell types across different samples.
   MCPcounter was also commonly used to score other gene sets.^[ [266]^34
   ^] This approach allows for precise quantification of various cell
   populations and pathways within tissue samples, providing valuable
   insights into the tumor microenvironment and diseases such as cancer.

HE Staining, Immunohistochemistry, and Immunofluorescence

   In this study, hematoxylin and eosin (HE) staining,
   immunohistochemistry (IHC), and immunofluorescence (IF) were utilized
   to analyze tissue morphology, specific protein expression, and spatial
   distribution. All tissue samples were paraffin‐embedded. The
   preparation of paraffin sections involved fixation in 10% neutral
   formalin, dehydration through a graded ethanol series, clearing with
   xylene, and embedding in paraffin. Sections were then cut to a
   thickness of 4 µm. The HE staining procedure included
   deparaffinization, hydration, nuclear staining with hematoxylin,
   differentiation, counterstaining with eosin, graded dehydration,
   clearing, and mounting. Under the microscope, cell nuclei appeared
   blue, and the cytoplasm appeared red. IHC was employed to detect
   antigens in tissue sections using specific primary and secondary
   antibodies. The protocol included fixation, sectioning, antigen
   retrieval, blocking, incubation with the primary antibody, incubation
   with the secondary antibody, color development, and mounting. IF was
   used to detect multiple antigens simultaneously, following a procedure
   similar to IHC but utilizing fluorescently labeled secondary
   antibodies, with DAPI used for nuclear counterstaining at the final
   step.

Statistical Analysis

   All statistical analyses were conducted using R (version 4.2.1).
   Comparisons between the two groups in the study were made using
   Wilcoxon rank‐sum tests, while correlation analysis was performed using
   Pearson's correlation. Statistical significance was defined as a
   p‐value of less than 0.05.

Conflict of Interest

   The authors declare no conflict of interest.

Supporting information

   Supporting Information
   [267]ADVS-12-2410998-s001.docx^ (8MB, docx)

   Supporting Information
   [268]ADVS-12-2410998-s002.xlsx^ (17.3KB, xlsx)

Acknowledgements