Clinical information and transcriptional profiles from the TCGA-BLCA dataset were retrieved through the UCSC Xena platform (https://xenabrowser.net/datapages/). We obtained key information on signatures related to immunotherapy and other therapies from the link at the end of the article published in 2021 (http://www.thno.org/v11p3089s4.zip/) (14). Moreover, we retrieved the gene list of 133 immunomodulators from the article (15). To validate these molecular insights, we incorporated clinical data from the immunotherapeutic advanced urothelial cancer cohort (IMvigor210), a landmark study investigating immunotherapeutic interventions in advanced urothelial carcinoma, which was used to corroborate our findings, from http://research-pub.gene.com/IMvigor210CoreBiologies/ (16).

The human BLCA cell lines 5637 (Cat.TCH-C104, HyCyte, Suzhou, China) and T24 (Cat.TCH-C352, HyCyte, Suzhou, China) were purchased from HyCyte (https://www.cas9x.com/). Specialized media for 5637 cells (Cat.TCH-C104, HyCyte, Suzhou, China) and T24 cells (Cat.TCH-C352, HyCyte, Suzhou, China) were used to maintain the BLCA cells. All cells were added with 10% FBS at 37°C in 5% CO₂.

The BLCA tissue microarray (TMA, HBlaU079Su01) utilized in this study was procured from Outdo Biotech (Shanghai, China), comprising 63 neoplastic specimens and 16 matched adjacent normal tissue controls. The experimental protocols involving these histological samples obtained formal ethical clearance from the Institutional Review Board (Shanghai, China) of Outdo Biotech, in compliance with international biomedical research standards.

The STR was independently scored by two pathologists. All pathological sections were obtained from the website: https://cancer.digitalslidearchive.org/. For the assessment of different tumor cells, the stroma–tumor ratio was scored as follows: 0 (0%–50%) and 1 (50%–100%) (7).

The prognostic value of the stromal score for overall survival (OS) and disease-free survival (DFS) was analyzed using the bioinformatics website (http://www.sxdyc.com/) (17). The outcome of the Kaplan–Meier (KM) curve was presented with clinical data from the TCGA-BLCA and in-house cohorts (18).

Stromal score, tumor purity, immune score, and ESTIMATE score were calculated using the ESTIMATE R package (19). This is why we divided all the samples equally into the high-stromal score group and the low-stromal score group according to the value of the stromal score (20). We estimated the immunological characteristics of the TME for each patient. Information on 133 immunomodulators, such as chemokines, immunoinhibitors, immunostimulators, Major Histocompatibility Complex (MHC), and receptors, was collected from previous studies (21). We also collected well-known effector genes of tumor-infiltrating immune cells (TIICs) to finish our bioinformatics analysis (22). To mitigate the potential miscalculations arising from various algorithms when estimating the levels of TIICs, we conducted a comprehensive analysis of their relative abundance utilizing the following algorithms: TIMER (23), EPIC (24), MCP-counter (25), and quanTIseq (26).

To graphically delineate oncogenic signaling cascades within the inflammatory tumor microenvironment, along with therapeutic responses to molecularly targeted agents and immune checkpoint modulation, a pathway enrichment analysis was conducted based on established computational frameworks from our prior investigations (27). Hypoxia (28), angiogenesis (29), and Epithelial - Mesenchymal Transition Score (EMT score) scores (30) were all calculated using the R package “GSVA.” All the gene sets can be found on the website: https://www.gsea-msigdb.org/. All enrichment scores of these signatures were calculated by the “GSVA” algorithm (31).

The analysis results of heatmap plots were generated using the CNSknowall platform (https://cnsknowall.com/). Furthermore, we also applied the R version 4.3.3 to finish several bioinformatics plots, such as violin plots and triangle plots. In addition, bioinformatics analyses were performed using tools in Hiplot Pro (https://hiplot.com.cn/), a web-based tool for bioinformatics analysis and visualization (32).

Cell viability was quantitatively assessed using the CCK-8 methodology with 96-well microplates. The experimental cohorts included the 5637 and T24 urothelial carcinoma cell lines, with the test groups receiving collagen treatment, and the control groups maintaining negative control (NC) conditions under standardized culture protocols (33). We prepared the acetic acid solution (68.8 μL of glacial acetic acid + 40 mL of double-distilled water) and then the collagen stock solution (100 μg/mL) (collagen: acetic acid solution = 1:30). After incubation at 37°C for 24 h, 10 µL of the CCK-8 reagent was added to each well, and this was cultured continuously for 1 h. Optical density (OD) was detected by a microplate reader Thermo Fisher Scientific Multiskan™ FC Microplate Photometer(Cat.1410101, Shanghai, China).

The 5637 and T24 cells were seeded in a six-well cell culture plate. Once the cells reached confluence, sterile tips were employed to create scratches in the cell monolayer within the six-well plate (34). We prepared the acetic acid solution (68.8 μL of glacial acetic acid + 40 mL of double-distilled water) and then the collagen stock solution (100 μg/mL) (collagen:acetic acid solution = 1:30). Following a 24-h incubation under physiological conditions (37°C), cellular migration dynamics were systematically monitored through time-lapse imaging at 0- and 12-h intervals. Quantitative analysis of wound closure was performed by calculating the mean migration distance across three distinct microscopic fields per scratch boundary using phase-contrast microscopy (Mshot M152-N, Guangzhou, China) with a magnification of ×100.

Cell invasion potential was evaluated through a modified Boyden chamber assay utilizing 24-well inserts pre-coated with growth factor-reduced Matrigel matrix (30 μL, 1:1 dilution in serum-free medium, Cat. 356234, Corning, Bedford, MA, USA). The Transwell apparatus contained polycarbonate membranes with 8.0-μm porosity (Corning, USA), following standardized in vitro invasion protocols (35). We prepared the acetic acid solution (68.8 μL of glacial acetic acid + 40 mL of double-distilled water) and then the collagen stock solution (100 μg/mL) (collagen:acetic acid solution = 1:30). After a standard 24-h culture under physiological conditions (37°C, 5% CO₂), residual cells remaining on the apical aspect of the polycarbonate membranes were mechanically removed with sterile cytology swabs. The migratory cell populations that successfully traversed the matrix barrier underwent methanol fixation followed by morphological visualization using 0.1% crystal violet solution (in 2% ethanol). Quantitative assessment was achieved through the systematic enumeration of stained cells across five predefined microscopic fields per chamber using phase-contrast microscopy (Mshot M152-N, Guangzhou, China), with data normalized to control conditions.

Quantitative analyses were implemented through GraphPad Prism 8.0 with rigorous statistical validation. Intergroup variations were determined via parametric one-way ANOVA, complemented by Bonferroni-corrected post-hoc examinations to address type I error inflation in multigroup comparative analyses. An unpaired, two-tailed Student’s t-test was used for the comparison of two groups. Correlation analyses were performed using Pearson’s correlation coefficient. Kaplan–Meier (KM) analyses were performed using the log-rank test. For all analyses, a two-tailed P <0.05 was considered statistically significant, unless otherwise indicated, unless otherwise indicated. Statistical significance was defined as *P < 0.05, **P < 0.01, ***P < 0.001, and ****P < 0.0001 (36).

As one of the most important components of the TME, the tumor stroma influences tumor biology by promoting tumor initiation, progression, metastasis, and treatment resistance (37). The tumor stroma manifests distinct spatiotemporal dynamics, architectural heterogeneity, and context-dependent molecular signatures. This pathologically specialized niche comprises two distinct compartments: (1) acellular constituents featuring remodeled extracellular matrix (ECM) architectures and aberrant neovascular networks, and (2) cellular constituents encompassing activated fibroblastic lineages (CAFs), multipotent mesenchymal progenitors, perivascular support cells, and immunomodulatory elements (38). The stromal score was derived using the ESTIMATE algorithm, which computes tumor stromal content.

Using the median stromal score as a stratification threshold, BLCA patients were categorized into high-stromal score and low-stromal score groups. A comparative analysis of clinicopathological parameters revealed significant intergroup disparities in survival status across both the TCGA-BLCA dataset and our in-house cohort. This stratification methodology effectively differentiated patient populations with distinct prognostic profiles ( Table 1 ). The Kaplan–Meier overall survival analysis based on the optimal cutoff value showed that the low-stromal score group showed better OS compared with the high-stromal score group, which represented better prognosis outcomes ( Figure 1A ) (17, 39). The DFS Kaplan–Meier plot based on the optimal cutoff value demonstrated that the high-stromal score group showed a poorer survival probability than the low-stromal score group, also supporting the association with worse prognosis outcomes ( Figure 1B ) (17, 39). Next, we stratified patients into subgroups based on T stage (T0–2 to T3–4), N stage (N0 to N1–3), M stage (M0 to M1), and AJCC stage (stages I–II to stages III–IV). According to the violin plots, STRs varied significantly across these subgroups, suggesting a strong correlation between a high stromal score and a poor prognosis. Specifically, higher stromal scores were observed in the T3–4, N1–3, M1, and stage III–IV subgroups ( Figures 1C–F ). We calculated the number of tumor samples in the high-stromal score group and the low-stromal score group according to the different stages, including T stage, N stage, M stage, and AJCC stage ( Figures 1G–J ). The histograms of the different stages showed that the high-stromal score group tended to manifest worse stages, such as T3–T4 stage, M1 stage, N1–N3 stage, and stage III–IV, which are usually associated with a worse prognosis. In addition, hypoxia score (28), angiogenesis score (29), and EMT score (30) are usually associated with poor prognosis in tumors. We also made a correlation plot between the stromal score and the hypoxia score ( Figure 1K ), angiogenesis score ( Figure 1L ), and EMT score ( Figure 1M ), which indicated that a high stromal score was correlated with a worse prognosis.

We performed pathological section scoring to assess the stroma–tumor ratio. All pathological sections were obtained from the Cancer Digital Slide Archive (https://cancer.digitalslidearchive.org/). We found that a high-stromal score sample was usually related to a high STR sample under the t-test result ( Supplementary Figure S2A ). The average stromal score in the high-STR group was approximately 404.8044, while in the low-STR group, it was only −1,166.3234. The stroma–tumor ratio was scored as follows: 0 (0%–50%) and 1 (50%–100%). Those samples with high STRs were usually scored as 1, such as the TCGA-DK-A3IQ ( Supplementary Figure S1A ), TCGA-FD-A5BS ( Supplementary Figure S1B ), TCGA-XF-AAME ( Supplementary Figure S1C ), and TCGA-FD-A5BT samples ( Supplementary Figure S1D ). On the contrary, samples with low STRs were scored as 0, as seen with the TCGA-ZF-AA4X ( Supplementary Figure S1E ), TCGA-ZF-A9RM ( Supplementary Figure S1F ), TCGA-E7-A5KF ( Supplementary Figure S1G ), and TCGA-ZF-AA4U samples ( Supplementary Figure S1H ). In summary, a high stroma was associated with poor survival prognosis outcomes.

Because the stroma is so important to the TME, we conducted research on the function of tumor stroma in shaping the BLCA immune microenvironment. Comprehensive phenotypic profiling of eight functionally distinct T-cell subsets (quiescent, regulatory, proliferating, helper, cytotoxic, progenitor-exhausted, terminally exhausted, and senescent populations) was performed across stromal score-stratified subgroups. Comparative analysis demonstrated significantly elevated activation scores across all T-cell compartments in high-stromal score cohorts, with particularly pronounced enrichment observed in both progenitor-exhausted and terminally exhausted subsets. These differential activation patterns suggest potential mechanistic links between a high stroma and T-cell exhaustion progression ( Figures 2A–H ). As we all know, the T cell is an important immune cell in the TME. We hypothesized that a richer tumor stroma could exhaust T cells, thus resulting in a worse prognosis (9). We searched for information and found that several genes are related to T-cell exhaustion, such as CD3E, PDCD1, CTLA4, HAVCR2, LAG-3, and TIGIT (40, 43). Bioinformatics analysis revealed significantly elevated expression levels of these candidate genes in the high-stromal score cohort ( Figure 2I ). Subsequent correlation analysis demonstrated consistent positive associations between stromal scores and the specified T-cell exhaustion marker. Importantly, these patterns suggest that a high stroma may drive immunosuppressive microenvironment remodeling through T-cell exhaustion ( Figures 2J–O ) (41).

Given the high potential correlation between the stroma and several immune factors in BLCA (42), we investigated the relationship between stromal score and 133 immunomodulators, including chemokines, immunoinhibitors, immunostimulators, MHCs, and receptors. We found that the expression levels of immunomodulators were upregulated in the high-stromal score group, suggesting an important role of a high stroma in shaping the immunosuppressive TME ( Figure 3A ) (15). Next, the violin plots showed that the expression levels of immunomodulators were upregulated in the high-stromal score group ( Figure 3B ). Since the stromal score was calculated by the R package “ESTIMATE,” we also made a violin plot comparing the stromal score and other scores from the R package “ESTIMATE,” where we found out that the high-stromal score group exhibited a higher ESTIMATE score and immune score, accompanied by lower tumor purity ( Figure 3C ) (5). This indicates that tumors with a higher stroma are surrounded by more immune cells. A heatmap was generated to visualize the relationship between stromal score and gene markers of immune cells. The result indicates that tumors with a higher stroma tend to be infiltrated by more immune cells ( Figure 3D ). To further validate these findings, we applied several algorithms to estimate the infiltration levels of TIICs. The results show that the majority of TIIC infiltration levels were highly elevated in the samples in the high-stromal score group ( Figure 3E ). The triangle heatmap plot showed a high correlation between stromal score and common inhibitory immune checkpoints, including CD274, KLRC1, LAIR1, and CD47 ( Figure 3F ) (43). This result shows that the stromal score is closely correlated with the immune checkpoints. Overall, a high stroma was found to shape the immunosuppressive TME.

Given that a higher stroma is associated with higher expression of PD-L1, we hypothesized that patients with a higher stroma would have a better response to immunotherapy. We quoted the IMvigor210 cohort, which provided abundant data on PD-L1 expression in immune cells (ICs) and tumor cells (TCs), immunotypes of tumors, and response to immunotherapy, to explore the relationship between stromal score and immunotherapy (44). The results show that the stromal score was higher in the IC2 group (immune cells with the highest PD-L1 expression) and the TC2 group (tumor cells with the highest PD-L1 expression) ( Figures 4A, B ). We also analyzed the stromal score in different tumor immunotypes (excluding immunotypes, desert immunotypes, and inflamed immunotypes) and found that the stromal score was higher in the excluded group ( Figure 4C ) (45).

Additionally, we calculated the percentage of these three immunotypes and found that the excluded immunotype was the most prevalent in both the high- and low-stromal score groups. Importantly, the percentage of the excluded immunotype was higher in the high-stromal score group, which also validated our previous conclusion ( Figure 4D ). Since the stromal score was higher in the IC2 and TC2 groups, in order to better study the relationship between the two, we introduced the IMvigor210 cohort. We also identified a positive correlation between the stromal score and the T-cell-inflamed score gene list in both the TCGA cohort and the IMvigor210 cohort, including CD27, CD276, and CD274 ( Figures 4E, F ) (46). Heatmap analysis revealed significant associations between stromal indices and the candidate gene cluster. Notably, clinically validated immunotherapeutic targets including CD19 (B-cell marker), PDCD1 (PD-1), and CD274 (PD-L1) demonstrated consistently elevated expression patterns in the high-stromal score subgroups across both the TCGA and IMvigor210 cohorts ( Figures 4G, H ). More importantly, the stromal score demonstrated significant positive correlations with enrichment scores of the majority of immunotherapy-positive gene signatures, suggesting potential predictive value for treatment response stratification ( Figures 4I, J ). Thus, we demonstrated that the stromal score can effectively predict the immune phenotype in BLCA.

Extensive genomic profiling research has established that basal-subclassified BLCA exhibits two distinct therapy advantages, namely, maximized intratumoral lymphocyte infiltration and superior objective response rates to PD-1 inhibitor therapies, including pembrolizumab (47). Notably, molecular taxonomy consensus frameworks further validate the enhanced clinical sensitivity of this molecular subtype to immune checkpoint inhibition strategies (48). As a result, we want to know whether the stroma could predict the molecular subtypes in BLCA. First, we evaluated the stroma and the clinicopathological features of BLCA ( Figure 5A ). The reason why we need to establish molecular subtypes is that they can predict the clinical response to chemotherapy, immunotherapy, and other therapies (14). According to the result shown in Figure 5B , we found that the stromal score was closely related to these subtypes, including the UNC, Consensus, CIT, Lund, MDA, Baylor, and TCGA subtypes ( Figure 5B ). To complement these findings, we systematically employed 10 computational frameworks alongside bladder cancer-specific molecular signatures to delineate stromal heterogeneity across stratification groups. In addition, we also made a heatmap correlation between the stromal score and the response to other therapies. The results from the DrugBank database (https://go.drugbank.com/) demonstrated enhanced efficacy of chemotherapeutic agents, EGFR-targeted therapies, and immunomodulators in the high-stromal score cohorts, whereas ERBB2/4 inhibitors and angiogenesis-blocking regimens exhibited diminished clinical benefit in this population ( Figure 5C ).

Utilizing the pRRophetic algorithm for pharmacogenomic prediction, we conducted systematic profiling of chemotherapeutic response disparities between stromal score-stratified cohorts. Heatmap analysis revealed that patients in the high-stromal score group demonstrated significantly enhanced therapeutic sensitivity in IC50 across multiple anticancer drugs, including cetuximab and vinblastine. These differential response patterns suggest that stromal score profiling may serve as a predictive biomarker for precision chemotherapy regimens ( Figure 5D ). Molecular classification systems demonstrated predictive capacity for multimodal therapeutic outcomes, particularly in neoadjuvant settings, encompassing cytotoxic chemotherapy, radiation protocols, and molecular-targeted interventions. Mechanistic analyses revealed that basal-phenotype malignancies exhibit biological predisposition to NAC responsiveness. Basal subtype tumors were more likely to respond to neoadjuvant chemotherapy (48, 49). We found that the mutation rates of TP53 and RB1 were significantly higher in the high-stromal score group ( Figure 5E ) and lower in the low-stromal score group ( Figure 5F ). The whole heatmap also showed that TP53 and RB1 were significantly higher ( Figure 5G ). Overall, the stroma is a novel classifier for the subtype of BLCA, meaning that patients with a high stroma tend to be sensitive to more therapy options.

Logically, a high stroma TME was correlated with a worse prognosis, which was proven in our previous research. However, we also found that the stromal score was higher in the IC2 and TC2 groups, which is beneficial in the prognosis. We first made the violin plot to describe the tumor purity in both the TCGA-BLCA and IMvigor210 cohorts (50). The results show that the high-stromal score group showed lower tumor purity ( Supplementary Figure S2B , Figure 3C ). Why did the high-stromal score group show worse clinical outcome despite high immune cell infiltration and low tumor purity? To explain the logic between these two, we speculated that it may be related to immune overdrive (51). We focused on profiling IC gene expression patterns, based on the rationale that neoplastic cells exploit inhibitory checkpoint signaling pathways to attenuate T-cell responses and consequently shape an immunosuppressive TME (52). Notably, the high-stromal score group had the highest expression levels of ICs in both the TCGA-BLCA and IMvigor210 datasets ( Supplementary Figures S2C, D ). In addition, we found that the high-stromal score group showed higher TGF-β signaling in both the TCGA-BLCA and IMvigor210 cohorts ( Supplementary Figures S2E, F ), which has been confirmed as one of the mechanisms of immune evasion in previous studies (53). Analytical results demonstrated that patients with elevated stromal scores exhibited distinct TME characteristics marked by immunological hyperactivation. This phenotype is manifested through three hallmark features: pronounced leukocyte infiltration, diminished neoplastic cellularity, and elevated immune checkpoint expression. Our findings collectively suggest that stromal-enriched microenvironments may drive the TME toward an immune-overstimulated state, ultimately fostering a unique ecological niche characterized by lymphocyte abundance, tumor suppression, and upregulated expression of both immunoregulatory molecules and TGF-β.

As we all know, collagen is one of the most important components of the tumor microenvironment. Therefore, we used collagen to imitate the high-STR group (54, 55). In order to validate our conclusion, we performed the CCK-8 assay ( Figure 6A ). Furthermore, the CCK-8 results showed that the tumor’s relative cell viability in the collagen group was higher than in the NC group in both the 5637 and T24 cell lines. We also conducted the wound healing assay ( Figures 6B–D ) and the invasion assay ( Figures 6E–H ) in both the 5637 and T24 cell lines, illustrating that in the high-STR group, tumor cells tended to show higher migration and invasion, thus confirming our previous result. Both the 5637 and T24 cells showed in 12 h that the collagen group cells migrated more than the NC group cells in the wound healing assay. Moreover, 5637 (22 h) and T24 cells (20 h) showed that the collagen group cells were more invasive than the NC group cells in the invasion assay. The reason why a richer tumor stroma showed a worse prognosis remains to be determined.

In our previous study, we found that high STR was associated with high PD-L1. In order to verify our previous conclusion, we also obtained an in-house cohort to validate our conclusion, which included 61 BLCA samples. The Kaplan–Meier OS analysis based on the optimal cutoff value showed that the low-STR group survived longer than the high-STR group ( Figure 7A ). Furthermore, we calculated all the samples’ PD-L1 expression and real STR. The correlation plot between PD-L1 expression and STR showed that they were related ( Figure 7B ). Moreover, the violin plot of different subtypes showed that a higher grade, such as T3–T4 stage ( Figure 7C ), N1 stage ( Figure 7D ), and stage III–IV ( Figure 7E ), tended to have a higher STR. In addition, the current BLCA cohort was classified into low- and high-STR expression groups based on the median level of STR expression. The immunohistochemistry (IHC) results showed that the infiltrating level of PD-L1 was higher in the high-STR group ( Figure 7F ) and that PD-L1 expression was higher in the high-STR group ( Figure 7G ). In conclusion, a high stroma is closely associated with PD-L1expression levels and predicts the clinical phenotypes.