Identification of RNA processing-related gene-based bladder cancer subtypes for prognosis and immune landscape assessment

Abstract

Background

Bladder cancer (BLCA) is a clinically complex malignancy characterized by high heterogeneity and recurrence, posing significant challenges for patient management. Growing evidence implicates dysregulated RNA processing, a crucial layer of gene expression control, as a key driver of BLCA pathogenesis and progression, highlighting its potential as a therapeutic vulnerability. This study aims to identify novel molecular subtypes based on RNA processing-related genes (RPRGs) and construct a robust prognostic risk model to improve survival prediction and personalize treatment strategies, particularly in the context of the immune landscape and immunotherapy response.

Methods

We analyzed BLCA data from The Cancer Genome Atlas (TCGA) (training cohort, N=394) and Gene Expression Omnibus (GEO) (validation cohort, GSE32894, N=224). Using RPRGs from the Molecular Signatures Database (MSigDB), we identified prognostic genes via univariate Cox regression and performed molecular subtyping with the non-negative matrix factorization (NMF) algorithm. A prognostic model was constructed using least absolute shrinkage and selection operator (LASSO) Cox regression (“glmnet” package), and its performance was validated using receiver operating characteristic (ROC) analysis (timeROC). Immune infiltration was assessed via single-sample gene set enrichment analysis (ssGSEA), ESTIMATE, and CIBERSORT, while functional enrichment was analyzed using GSEA and clusterProfiler. Drug sensitivity was predicted using the CellMiner and DGIdb databases, along with the “pRRophetic” package.

Results

Using RPRGs, we stratified BLCA into two molecular subtypes and developed an 8-gene prognostic model via LASSO Cox regression. The model effectively stratified patients into high-risk (poor prognosis) and low-risk (favorable prognosis) groups in both TCGA and GEO cohorts [area under the curve (AUC) >0.71]. High-risk group displayed immunosuppressive traits [e.g., elevated Tumor Immune Dysfunction and Exclusion (TIDE) score, M2 macrophage enrichment] and reduced immunotherapy response, while low-risk group showed elevated tumor mutation burden (TMB) and CD8⁺ T cell infiltration. Functional enrichment revealed extracellular matrix pathways in high-risk cases, and drug sensitivity profiling identified signature gene-chemotherapy associations.

Conclusions

Based on RPRGs, this study establishes a novel risk model that effectively stratifies BLCA patients, not only predicting prognosis but also revealing its close association with the tumor microenvironment and immunotherapy response, providing a new tool for personalized treatment.

Highlight box.

Key findings

• This study identified two distinct molecular subtypes of bladder cancer (BLCA) based on RNA processing-related genes (RPRGs).

• An 8-gene prognostic risk model was constructed and validated to accurately predict patient survival.

• This risk model revealed that high-risk patients possess an immunosuppressive tumor microenvironment, whereas low-risk patients exhibit an immune-activated state and high tumor mutational burden.

What is known and what is new?

• BLCA is a highly heterogeneous malignancy where dysregulated gene expression control, including RNA processing, drives pathogenesis and progression.

• This manuscript provides a novel RPRG-based stratification framework that links post-transcriptional RNA processing dysregulation directly to specific immune landscapes and immunotherapy response rates in BLCA.

What is the implication, and what should change now?

• The RPRG-based risk score serves as an independent prognostic factor that guides individualized treatment by identifying patients likely to benefit from immune checkpoint inhibitor therapy; consequently, clinical management should integrate molecular risk stratification to adopt combination therapies for high-risk patients instead of single-agent immunotherapy.

Introduction

Bladder cancer (BLCA) ranks among the most common malignant tumors of the urinary system worldwide, with significant geographic and gender disparities in its incidence and mortality rates (1,2). In recent years, the number of new cases has remained high, and the incidence in men is substantially higher than in women, which is associated with unique risk factor exposures (3). The development of BLCA results from the combined effects of environmental exposures and genetic susceptibility. Smoking is the most significant and modifiable risk factor, accounting for approximately 50% of all BLCA cases (4,5). Occupational exposure to aromatic amines (e.g., in the dye, rubber, and paint industries) is another well-established causative factor (6). Other risk factors include chronic urinary tract infections, schistosomiasis (strongly associated with squamous cell carcinoma), and long-term use of certain drugs such as cyclophosphamide (7). Histopathologically, BLCA is primarily classified into urothelial carcinoma (accounting for about 90%), squamous cell carcinoma, and adenocarcinoma, among others. Urothelial carcinoma is highly heterogeneous (8). Based on the depth of tumor invasion, BLCA are further categorized into non-muscle-invasive BLCA and muscle-invasive BLCA (9). Non-muscle-invasive BLCA is confined to the mucosa or submucosa. Although these tumors have a relatively favorable prognosis, they are characterized by multifocality and a high recurrence rate (10,11). In contrast, muscle-invasive BLCA invades the muscularis propria or beyond. These tumors are highly aggressive, prone to metastasis, and associated with a poor prognosis (12,13). Although the majority of patients are initially diagnosed with non-muscle-invasive disease, which has a high disease-specific survival rate, the high risks of recurrence and progression pose long-term, substantial economic burdens and challenges to quality-of-life management (14). Therefore, BLCA represents a research area of significant complexity in both cancer biology and clinical therapeutics. A deeper understanding of the complex molecular pathogenesis driving tumor initiation, progression, and heterogeneity, as well as the identification of robust biomarkers and new therapeutic vulnerabilities across all subtypes, is crucial for improving the survival and quality of life for BLCA patients.

RNA processing refers to a series of co-transcriptional and post-transcriptional events that precursor messenger RNA (pre-mRNA) undergoes after transcription, including 5' capping, splicing, 3' cleavage and polyadenylation, and RNA editing (15,16). These steps determine the sequence, stability, subcellular localization, and translational potential of the final RNA molecule, thereby exerting a decisive influence on gene expression. With the advancement of high-throughput sequencing and epitranscriptomic technologies, defects in RNA processing have been identified as a significant pathogenic mechanism in various human diseases (17). Aberrant RNA processing is frequently associated with neurological disorders and cancers. PTEN-directed PI3K signaling maintains neuron-specific splicing patterns, and its loss broadly disrupts the splicing of transcripts related to synapses, ion channels, and the cytoskeleton (18). High expression of capping enzymes (such as RNGTT and MCEE) prolongs the half-life of mRNAs like MYC, VEGFA, and EGFR, thereby promoting cell proliferation, angiogenesis, and immune evasion, which is linked to poor prognosis in cancers including BLCA and hepatocellular carcinoma (19). The capping enzyme CMTR1 concurrently regulates the growth and immune evasion of colorectal cancer cells via the “capping-STAT3” axis, leading to shortened survival (20). METTL3-mediated m⁶A modification enhances MYC translation and promotes BLCA progression (21), whereas loss of YTHDF2 destabilizes pro-metastatic MYC/VEGFA transcripts and impairs the survival of glioblastoma stem cells, illustrating the context-dependent oncogenic role of the m⁶A machinery (22). In summary, RNA processing is not merely a “modification step” after transcription but constitutes a core layer of gene expression regulation. Its dysregulation can lead to immune dysregulation and subsequently contribute to diseases like cancer. Therefore, elucidating the association between RNA processing and the prognosis of BLCA may aid in the development of novel cancer therapeutic strategies.

This study aims to comprehensively investigate the prognostic value and molecular role of RNA processing-related genes (RPRGs) in BLCA. We first employed bioinformatics methods to perform molecular subtyping of BLCA patients and constructed a robust prognostic risk model. Subsequently, we conducted an in-depth analysis of the differences in tumor microenvironment (TME) composition, immune activation status, and drug sensitivity among different risk groups to uncover potential biological mechanisms. We present this article in accordance with the TRIPOD reporting checklist (available at https://tau.amegroups.com/article/view/10.21037/tau-2025-aw-780/rc).

Methods

Data collection

From The Cancer Genome Atlas (TCGA) database (https://portal.gdc.cancer.gov/), we downloaded mRNA expression data, mutation data, and clinical data for BLCA. Samples with an overall survival (OS) time of less than 30 days were excluded, and subsequent analysis was performed on the remaining 394 BLCA samples. From the Gene Expression Omnibus (GEO) database (https://www.ncbi.nlm.nih.gov/geo/), download microarray data (dataset GSE32894, N=224 samples). Similarly, samples with an OS time of less than 30 days were removed. The TCGA dataset serves as the training set, and the GEO dataset serves as an independent validation set. Additionally, RPRGs are from the Molecular Signatures Database (MSigDB) (https://www.gsea-msigdb.org/gsea/msigdb) [Gene Ontology (GO): 0006396, Table S1]. The study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments.

Construction of RNA processing-related BLCA subtypes

Leveraging univariate Cox regression (P<0.05), we first screened RPRGs associated with BLCA prognosis. These genes were then used to cluster the BLCA samples into two subtypes (k=2) via the non-negative matrix factorization (NMF) algorithm. The clustering pattern was evaluated through principal component analysis (PCA) and survival analysis. Finally, a comparative analysis using the “limma” package identified differentially expressed genes (DEGs) between the subtypes, with criteria set at |logFC| >1 and adjusted P value <0.05.

Screening prognostic features to construct a prognostic model

Following the identification of DEGs (|logFC| >1, adjusted P value <0.05) between subtypes, univariate Cox regression (P<0.05) selected candidate prognostic genes. Least absolute shrinkage and selection operator (LASSO) regression analysis (via the “glmnet” package) was then employed to penalize and select the most informative genes, preventing overfitting. A multivariate Cox proportional hazards model was built with the selected genes to generate a risk score for each patient. Using the median score as a threshold, patients were divided into high- and low-risk groups. The model’s performance was assessed by plotting survival curves and calculating area under the curve (AUC) values for 1-, 3-, and 5-year OS using the “timeROC” package. Additional visualizations included risk score distribution, survival status distribution, and a gene expression heatmap.

A nomogram for individualized survival prediction

To ascertain the independent prognostic value of the risk model, univariate and multivariate Cox analyses were performed, integrating the risk score with clinical parameters. A predictive nomogram for 1-, 3-, and 5-year survival was then constructed (“rms” package), and its performance was validated through calibration curves and decision curve analysis (DCA).

Immune infiltration analysis

The infiltration levels of various immune cells and related functions were quantified using single-sample gene set enrichment analysis (ssGSEA). The “ESTIMATE” package was used to calculate Immune, Stromal, and ESTIMATE scores along with tumor purity, while the CIBERSORT algorithm provided a detailed deconvolution of immune cell subsets. Furthermore, we compared the expression of immune checkpoint molecules between the high- and low-risk groups. To extend our analysis, we incorporated data from The Cancer Immunome Atlas (TCIA, https://tcia.at) to compare immunogenic potential and classified tumors into six published immune subtypes (C1-C6) based on a previous study (23). The likelihood of immune evasion was assessed by calculating Tumor Immune Dysfunction and Exclusion (TIDE) scores. Finally, the clinical relevance of our risk model was validated using the IMvigor210 cohort treated with anti-PD-L1 therapy. In this cohort (N=348), patients were categorized as responders (R; exhibiting complete or partial response) or non-responders (NR; with stable or progressive disease), and their distribution was compared between the risk groups.

Gene set enrichment analysis (GSEA) and pathway enrichment assessment

Functional enrichment analyses were conducted on the transcriptional profiles of the risk groups. This included GSEA (v4.3.2) for pre-defined pathways and GO/Kyoto Encyclopedia of Genes and Genomes (KEGG) enrichment analysis (via clusterProfiler) for the DEGs (|logFC| >1, adj. P<0.05) identified between the groups.

Drug sensitivity prediction

A multi-database approach was employed to investigate drug-gene interactions and predict clinical chemosensitivity. Drug associations with prognostic genes were screened via the CellMiner database (https://discover.nci.nih.gov/cellminer/), while potential targets and mechanisms were identified using DGIdb database (https://www.dgidb.org/). Subsequently, the “pRRophetic” algorithm was used to profile the predicted half maximal inhibitory concentration (IC₅₀) values of various drugs across the risk groups.

Statistical analysis

All statistical analyses and data visualizations were conducted in R (version 4.4.0). Inter-group comparisons were performed using the Wilcoxon test, associations between variables were assessed with Pearson correlation, and a significance threshold of P<0.05 was applied. The majority of figures were generated using the “ggplot2” package.

Results

Construction of RNA processing-related molecular subtypes based on prognosis-associated genes

Based on univariate Cox regression (P<0.05), 38 prognostic RPRGs were selected for BLCA (Table S2), from which NMF clustering analysis identified k=2 as the optimal cluster number. The clustering matrix for k=2 clearly delineated two distinct clusters. Two disease subtypes were identified: Cluster 1 comprised 191 samples, and Cluster 2 comprised 203 samples (Figure 1A,1B). PCA analysis of the samples showed clear separation into two molecular subtypes (Figure 1C). Kaplan-Meier analysis revealed a significant difference in survival curves between the two subtypes, with Cluster 1 exhibiting a poorer survival outcome (Figure 1D). To further characterize the two subtypes, a differential expression analysis was conducted, which yielded 392 DEGs (Figure 1E, Table S3). The expression patterns of the 38 prognostic genes were then profiled across BLCA and normal tissues, revealing distinct clustering (Figure 1F).

Figure 1.

Differential gene analysis and subtype identification. (A) cophenetic coefficient plot. (B) NMF consensus clustering plot. (C) PCA between subtypes based on prognostic genes. (D) Survival curves of the two molecular subtypes. (E) Differential expression analysis between Cluster 1 and Cluster 2. (F) PCA of tumor group and normal group based on prognostic genes. NMF, non-negative matrix factorization; PCA, principal component analysis.

Immune landscape characterization

Assessment of the TME using the ESTIMATE algorithm revealed significantly higher immune, stromal, and ESTIMATE scores in Cluster 1, whereas Cluster 2 exhibited greater tumor purity (Figure 2A). Further immune deconvolution via CIBERSORT identified a neutrophil- and macrophage-rich environment in Cluster 1, contrasting with an enrichment of adaptive immune cells like plasma cells and Tregs in Cluster 2 (Figure 2B). Cluster 2 demonstrated a higher immunophenoscore (IPS) (Figure 2C). A significantly higher immune infiltration level in Cluster 1 was confirmed by ssGSEA (Figure 2D), alongside elevated expression of immune checkpoint genes (Figure 2E). Finally, a higher TIDE score was observed in Cluster 1, suggesting increased immune evasion potential (Figure 2F). The enrichment analysis of DEGs between subtypes showed significant enrichment in gene sets for collagen-containing extracellular matrix, humoral immune response, granulocyte chemotaxis, and killing of cells of another organism (Figure 2G).

Figure 2.

Immune microenvironment characterization. (A) Differences in ESTIMATE scores between different clusters. (B) CIBERSORT: immune fractions (Cluster 1 vs. Cluster 2). (C) IPS scores (Cluster 1 vs. Cluster 2). (D) Profiling of immune cell infiltration by ssGSEA: Cluster 1 vs. Cluster 2. (E) Immune checkpoint expression (Cluster 1 vs. Cluster 2). (F) TIDE scores of different clusters. (G) Enrichment analysis of DEGs between the two subtypes. ns, P>0.05; *, P<0.05; **, P<0.01; ***, P<0.001; ****, P<0.0001. BP, biological process; CC, cellular component; DEGs, differentially expressed genes; IPS, immunophenoscore; KEGG, Kyoto Encyclopedia of Genes and Genomes; MF, molecular function; ssGSEA, single-sample gene set enrichment analysis; TIDE, Tumor Immune Dysfunction and Exclusion.

Screening prognostic features to construct a prognostic model for BLCA

First, univariate Cox regression on the TCGA-BLCA training cohort identified 115 survival-associated candidate genes (P<0.05; Table S4). Subsequently, LASSO regression refined this list to 19 feature genes (Figure 3A,3B). Finally, multivariate Cox analysis selected 8 key genes from these to build the final prognostic model (Figure 3C):

Figure 3.

Prognostic model development and validation. (A) LASSO coefficient profiles. (B) λ-selection cross-validation curve. (C) Forest plot of multivariate regression analysis. (D) Differences in expression levels of feature genes between tumor and normal groups. (E) Differences in expression levels of feature genes among different clusters. ROC curve (F), survival curve (G), score distribution and survival status plot of risk groups (H) in the training set. ROC curve (I), survival curve (J), score distribution and survival status plot of high- and low-risk groups (K) in the validation set. ns, P>0.05; *, P<0.05; **, P<0.01; ****, P<0.0001. AUC, area under the curve; BLCA, bladder cancer; CI, confidence interval; HR, hazard ratio; LASSO, least absolute shrinkage and selection operator; ROC, receiver operating characteristic; TCGA, The Cancer Genome Atlas.

Riskscore = −0.35974 * WBP1 − 0.13612 * TOX3 + 0.16717 * MYC + 0.05618 * KRT5 − 0.43346 * GNLY + 0.08936 * FN1 − 0.12266 * DEGS2 − 0.16968 * AIM2

The expression of the eight feature genes was analyzed first. Comparison between TCGA-BLCA tumors and normal tissues identified three genes with significant differences: MYC was highly expressed in normal tissues, whereas AIM2 and WBP1 were upregulated in tumors (Figure 3D). Additionally, all eight genes showed significant expression differences between the two molecular subtypes (Figure 3E). Subsequently, patients were stratified into high- and low-risk groups based on the median risk score derived from the model. In the training cohort, the model demonstrated strong predictive accuracy, with 1-, 3-, and 5-year survival AUC values all exceeding 0.72 (Figure 3F). Kaplan-Meier analysis confirmed significantly worse OS in the high-risk group (Figure 3G), a finding further supported by the risk score distribution and survival status plot (Figure 3H). These results were consistently replicated in the independent validation cohort, where the AUC values remained above 0.71 (Figure 3I), survival was significantly poorer in the high-risk group (Figure 3J), and the corresponding distribution plots validated the stratification (Figure 3K).

Construction of a nomogram for independent prognostic analysis

Prognostic evaluation of the BLCA risk model began with univariate Cox analysis, which identified several variables significantly associated with survival, including stage, risk score, and TNM stage (Figure 4A). Subsequent multivariate analysis established the risk score as an independent prognostic factor (Figure 4B). To facilitate clinical translation, a nomogram integrating relevant clinical parameters was constructed to predict individual survival probabilities (Figure 4C). The nomogram’s clinical utility was confirmed by DCA, which showed robust net benefits for predicting 1-, 3-, and 5-year survival (Figure 4D). Furthermore, calibration curves indicated a strong concordance between the nomogram’s predictions and actual observed outcomes (Figure 4E). Supplementary analyses affirmed the prognostic significance of the individual feature genes (Figure S1A) and revealed that risk scores were significantly higher in advanced-stage (III + IV) tumors and in Cluster 1 compared to early-stage (I + II) tumors and Cluster 2, respectively (Figure S1B).

Figure 4.

Nomogram survival model: development and validation. (A) Univariate Cox: clinicopathologic factors & risk score. (B) Multivariate Cox: clinicopathologic factors & risk score. (C) Nomogram predicting BLCA prognosis. (D) Assessment of clinical net benefit by DCA across 1-, 3-, and 5-year time horizons. (E) Comparison of predicted versus observed survival probabilities at 1, 3, and 5 years. BLCA, bladder cancer; CI, confidence interval; DCA, decision curve analysis.

Immune infiltration analysis of high- and low-risk groups

Comprehensive immune profiling revealed distinct microenvironment landscapes between the risk groups. Although the overall immune cell composition was broadly similar, dominated by macrophages and CD8⁺ T cells (Figure 5A), specific subsets differed significantly. The high-risk group was enriched for immunosuppressive cells (M2 macrophages, neutrophils, M0 macrophages) and resting CD4⁺ memory T cells, whereas the low-risk group showed higher levels of cytotoxic CD8⁺ T cells, plasma cells, activated CD4⁺ memory T cells, and Tregs (Figure 5B). This pattern was consistent with ESTIMATE scores, which indicated a more complex stroma-rich microenvironment with lower tumor purity in the high-risk group (Figure 5C). Functional enrichment by ssGSEA further supported a suppressive milieu in the high-risk group (e.g., macrophages), contrasting with enhanced anti-tumor immunity in the low-risk group (e.g., CD8⁺ T cells, NK cells, Type II IFN response) (Figure 5D). The high-risk group also exhibited a higher TIDE score and a lower IPS (Figure 5E,5F), suggesting a greater potential for immune evasion and a weaker inherent immune response. Concordantly, immune checkpoint gene expression was dysregulated (Figure 5G), and the low-risk group demonstrated significantly better response rates (Figure 5H) and survival outcomes in the IMvigor210 anti-PD-L1 cohort (Figure 5I). Finally, an integrated analysis showed that Cluster 1/high-risk associated with immune subtypes C1 (wound healing)/C2 (IFN-γ dominant), while Cluster 2/low-risk linked to subtypes C3 (inflammatory)/C4 (lymphocyte depleted), consolidating the link between molecular subtypes, risk model, and immune classification (Figure 5J).

Figure 5.

Immune microenvironment characterization in high/low risk groups. (A) Immune cell composition in high- and low-risk groups. (B) Differentially infiltrated immune cells between risk groups. (C) Tumor microenvironment scores estimated by the ESTIMATE algorithm. (D) Immune activity enrichment scores quantified by ssGSEA. (E) Assessment of immune evasion potential by TIDE score. (F) Immunophenoscore comparing risk groups. (G) Differential expression of immune checkpoint molecules. (H) Immunotherapy response rates in the IMvigor210 cohort. (I) Survival outcomes of anti-PD-L1 therapy in the IMvigor210 cohort. (J) The interaction between immune subtypes and the two risk groups was depicted using an alluvial diagram. ns, P>0.05; *, P<0.05; **, P<0.01; ***, P<0.001; ****, P<0.0001. ssGSEA, single-sample gene set enrichment analysis; TIDE, Tumor Immune Dysfunction and Exclusion.

Enrichment analysis of risk groups

GSEA revealed that the high-risk group was significantly enriched in arrhythmogenic right ventricular cardiomyopathy (ARVC), ECM-receptor interaction, focal adhesion, and regulation of actin cytoskeleton, while the low-risk group was significantly enriched in antigen processing and presentation, olfactory transduction, peroxisome, and primary immunodeficiency (Figure 6A,6B). Differential analysis between the high- and low-risk groups (|logFC| >1, p.adj <0.05) identified 56 upregulated genes and 53 downregulated genes, which were subsequently subjected to GO and KEGG enrichment analyses. GO enrichment analysis showed that the upregulated genes were primarily associated with biological processes such as the organization and composition of the extracellular matrix, collagen formation, and skin development (Figure 6C). The downregulated genes were mainly involved in the metabolic processes of fatty acids and icosanoids, as well as the response to hormones, and were significantly localized to the apical plasma membrane in terms of cellular components (Figure 6D). KEGG enrichment analysis indicated that the upregulated genes were primarily involved in biological pathways such as the PPAR signaling pathway, arachidonic acid metabolism, and cornified envelope formation (Figure 6E). The downregulated genes were closely related to cellular structure and adhesion, showing significant enrichment in pathways such as the cytoskeleton in muscle cells, focal adhesion, and ECM-receptor interaction (Figure 6F).

Figure 6.

Enrichment analysis for risk groups. (A) GSEA, high-risk group. (B) GSEA, low-risk group. (C) GO terms, upregulated DEGs. (D) GO terms, downregulated DEGs. (E) KEGG pathways, upregulated DEGs. (F) KEGG pathways, downregulated DEGs. BP, biological process; CC, cellular component; DEGs, differentially expressed genes; GO, Gene Ontology; GSEA, gene set enrichment analysis; KEGG, Kyoto Encyclopedia of Genes and Genomes; MF, molecular function.

Genomic and pharmacogenomic analyses

We leveraged the CellMiner database to explore associations between characteristic gene expression and drug sensitivity. Significant positive correlations were identified for AIM2/AEG-40730 (cor=0.389), DEGS2/Ribavirin (cor=0.432), MYC/DMAPT (cor=0.555), and WBP1/XL-147 (cor=0.437). Conversely, significant negative correlations were observed for FN1/Des-fluoro-TAK-960 (cor=−0.483), GNLY/CCT-128930 (cor=−0.394), KRT5/nintedanib (cor=−0.361), and TOX3/6-Thioguanine (cor=−0.418) (all P<0.05; Figure 7A). Analysis of chemotherapeutic agent sensitivity revealed that the high-risk group was associated with higher resistance to 5-fluorouracil (higher IC₅₀), whereas the low-risk group showed higher resistance to bortezomib, cisplatin, docetaxel, and lenalidomide (Figure 7B). Interrogation of the DGIdb database predicted multiple interacting drugs for MYC and FN1, with MYC having a substantially larger number of potential targeted agents (Figure 7C). Finally, assessment of tumor mutation burden (TMB) using TCGA data showed that the low-risk group exhibited a higher TMB than the high-risk group. Waterfall plots of the top 20 mutated genes revealed high mutation frequencies in TP53 and TTN in both groups (Figure 7D,7E).

Figure 7.

Tumor mutation and drug sensitivity analysis. (A) CellMiner-derived drug-gene correlations. (B) Comparative IC₅₀ values of chemotherapeutic agents. (C) Druggable predictions for characteristic genes from DGIdb. (D) Mutational landscape of the top 20 genes. (E) TMB comparison. *, P<0.05; **, P<0.01; ***, P<0.001; ****, P<0.0001. IC₅₀, half maximal inhibitory concentration; TMB, tumor mutation burden.

Discussion

In this study, we systematically investigated the molecular characteristics of RPRGs in BLCA, establishing a novel disease stratification framework. Our integrated analysis revealed two distinct molecular subtypes characterized by significant differences in RPRG expression patterns, which were strongly associated with patient clinical outcomes. The prognostic model we developed was rigorously validated in independent cohorts, providing a robust tool for risk assessment that integrates molecular features with clinical parameters. Importantly, our analyses uncovered fundamental differences in TME composition and immune activation states between risk groups, offering potential mechanisms underlying their distinct clinical behaviors. Collectively, these findings provide new insights into the biological determinants of BLCA progression while identifying potential prognostic biomarkers for therapeutic response.

The 8 signature genes (FN1, KRT5, MYC, AIM2, DEGS2, WBP1, GNLY, and TOX3) establish a robust molecular framework for prognostic assessment in BLCA. Secreted by cancer-associated fibroblasts, FN1 remodels the extracellular matrix and activates the NF-κB/STAT3 signaling pathway via integrin α5β1, upregulating PD-L1 and VEGFA. This enhances the proliferative and invasive capabilities of BLCA cells, fosters an immunosuppressive microenvironment, and ultimately drives disease progression and poor prognosis (24,25). Furthermore, FN1 is closely linked to cancer through its role in RNA processing. The EDA-positive splice variant of FN1 promotes mesenchymal transition and invasive behavior in glioblastoma by activating the integrin α9β1-mediated FAK/Src signaling axis (26). KRT5 serves as an immunohistochemical marker for “basal-type” muscle-invasive BLCA. Tumor cells expressing KRT5 retain a basal/stem cell phenotype, demonstrate increased sensitivity to neoadjuvant cisplatin-gemcitabine chemotherapy, and thus KRT5 expression emerges as an independent favorable factor for predicting pathological response and improved prognosis (27). In BLCA, MYC acts not only as a direct downstream effector of KLF16—driving proliferation, stemness, and cisplatin resistance—but also reciprocally suppresses KLF16 via miR-33b-5p, forming a self-amplifying loop. Intervention in this loop can simultaneously block MYC transcriptional output and restore endogenous tumor-suppressive signaling, offering a novel combination therapy strategy for patients with basal-type or chemotherapy-resistant disease (28). Previous studies have demonstrated that high expression of AIM2 suppresses the PI3K-AKT-mTOR signaling pathway, reduces stemness, and consequently inhibits proliferation (29). In the present study, the observed upregulation of AIM2 in tumor tissues may exert a similar tumor-suppressive function. Therefore, RNA processing of signature genes in BLCA is critical for prognostic assessment and therapeutic strategies by regulating gene expression, promoting tumor progression, and influencing treatment response.

Additionally, to contextualize within a broader clinical framework, the risk model from this study was compared with the well-established luminal/basal molecular subtypes. Notably, KRT5—a classic basal marker—is included among the characteristic genes of our model, indicating a partial overlap between our risk groups and the basal-like subtype with aggressive features (30,31). However, existing luminal/basal classifications primarily reflect the cellular origin and differentiation state (32), whereas our model captures post-transcriptional dysregulation of RNA processing. This mechanistic distinction suggests that the classification in this study complements, rather than merely overlaps with, the existing framework.

This study found that the high-risk BLCA group was significantly enriched in the ECM-receptor interaction, focal adhesion, and regulation of actin cytoskeleton pathways. The activation of the ECM-receptor pathway suggests high invasiveness, immune resistance, and short survival in BLCA, and it can independently predict prognosis and guide FAK/PI3K-targeted therapy (33). Alterations in the ECM microenvironment may be a key driver of malignant progression in BLCA. Dysregulation of the actin cytoskeleton directly affects the motility of tumor cells, which may form the structural basis for the increased propensity of infiltration and metastasis in high-risk patients (34). Research indicates that TNF-α upregulates CLASP2 via METTL3-mediated m⁶A modification, which cooperates with IQGAP1 to drive F-actin remodeling, thereby promoting BLCA cell migration and lung metastasis (35). Notably, our differential expression analysis further revealed that the upregulated genes in the high-risk group were primarily involved in the organization of the extracellular matrix and collagen formation, providing direct molecular evidence for the aforementioned pathway enrichment results. Conversely, the low-risk BLCA group was significantly enriched in antigen processing and presentation, peroxisome, and primary immunodeficiency pathways. Patients with RAG or LIG4 deficiencies have a 3–12-fold increased standardized incidence ratio for cancers such as lymphoma and gastrointestinal cancer, indicating that loss of immune surveillance creates a breeding ground for cancer development (36).

This study found that the high-risk group is characterized primarily by an immunosuppressive microenvironment, marked by M2 macrophage infiltration and T cell dysfunction, leading to poor prognosis and diminished response to immunotherapy. This group showed an abundance of M0 macrophages, which possess strong immunosuppressive capabilities: M0 macrophages readily transform into M2-type tumor-associated macrophages (TAMs) that inhibit T cell function and suppress immune responses (37,38). However, the overall immune score in the high-risk group was higher. This paradox may be explained by the secretion of factors like IL-10 and TGF-β by a large number of immunosuppressive cells (M2 macrophages, neutrophils, Tregs). These factors inflate the overall immune cell count while functionally suppressing effector T cells, thereby forming the core mechanism for immune escape and poor prognosis (39-41). This microenvironment results in insufficient T cell infiltration, reflected in a low IPS score and a high TIDE score, indicating a high potential for immune escape. Consequently, in the IMvigor210 cohort, the high-risk group exhibited a lower response rate to anti-PD-L1 therapy and worse survival outcomes. In contrast, the low-risk group exhibited an immune-activated state, characterized by stronger anti-tumor immunity and a better treatment response. Tumor antigens are cross-presented by dendritic cells (DCs) to activate CD8⁺ T cells. These infiltrating T cells utilize granzymes and IFN-γ to kill tumor cells and establish a pool of TCF1⁺ memory cells, enabling immune activation and long-term immunosurveillance (42).

The TP53 gene shows high mutation frequencies in both the high- and low-risk groups. This finding highlights the universality and central role of TP53 inactivation in the occurrence and development of BLCA. As one of the most critical tumor suppressor genes, TP53, in normal cells, responds to genotoxic damage and other stress signals by inducing cell cycle arrest, DNA repair, or programmed cell death, thereby preventing malignant transformation (43,44). However, when TP53 mutates, it not only loses its normal tumor-suppressive functions, but some mutant p53 proteins can also gain novel oncogenic functions, promoting tumor invasion, metastasis, and therapy resistance (45). In this study, the seemingly paradoxical phenomenon where the high-risk group has a lower TMB but a worse prognosis may be closely related to the TP53 mutation status. Extensive research indicates that TP53 mutation is an early event leading to genomic instability, which can accelerate tumor evolution (46,47). TP53 mutations are often associated with more advanced tumor stage, stronger chemotherapy resistance, and a poorer prognosis, which may constitute the core molecular basis driving the adverse clinical outcomes in the high-risk group patients.

To further clarify the clinical utility of this model, it is integrated into a multimodal treatment decision-making system. Patients in the low-risk group exhibit an “immune-hot” phenotype and high tumor mutational burden, suggesting they are more likely to benefit from immune checkpoint inhibitor (ICI) therapy, consistent with current trends in precision oncology (48,49). In contrast, the high-risk group not only has a poorer prognosis but also demonstrates significant immunosuppressive and stromal activation features (e.g., ECM remodeling), indicating that these patients may be less responsive to single-agent immunotherapy. Future studies should explore combination therapies targeting their RNA processing defects or the stromal microenvironment (50).

Additionally, this study holds significant translational potential when combined with emerging diagnostic technologies such as liquid biopsy. Although this research is based on tissue samples, RNA processing products and their regulatory factors are often released into body fluids (51). Recent studies have highlighted the substantial potential of circulating microRNAs as minimally invasive biomarkers for risk stratification in genitourinary cancers (52). Such liquid biomarkers share a common principle with our model: moving beyond static histopathology to capture dynamic molecular signals. As liquid biopsy technologies for BLCA gradually enter routine clinical practice (53), future efforts could explore the detection of key RPRGs features or their associated splicing variants in urine or blood, thereby enabling non-invasive, real-time monitoring of BLCA patients. This also represents an important direction for integrating our model into future diagnostic pathways.

Despite the significant clinical value of the constructed BLCA prognostic model, certain limitations remain. Our core findings and the developed prognostic model are based entirely on bioinformatics analyses of public databases and lack validation in an independent, large-scale prospective clinical cohort. Secondly, although we have inferred the potential mechanisms of these signature genes in BLCA progression, there is an absence of functional validation through experimental studies (such as cellular or animal models). This limits our ability to directly elucidate the specific molecular mechanisms of the eight signature genes under the regulation of RNA processing. Further experimental research is needed in the future to validate the functions of these genes and their potential as therapeutic targets.

Conclusions

This study successfully constructed a risk assessment model based on RPRGs, effectively stratifying BLCA patients at the molecular level. The model revealed a significant association between differential prognosis and unique TME characteristics and potential response to immunotherapy, offering a novel perspective for prognosis assessment and individualized treatment strategies in BLCA.

Supplementary

The article’s supplementary files as

Ethical Statement: The authors are accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved. The study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments.