A shifting treatment landscape. Bladder cancer is the ninth most common cancer worldwide and ranks 13th in cancer mortality. While BCG immunotherapy has been successful for early-stage disease, the introduction of checkpoint inhibitor drugs has changed the treatment landscape for advanced bladder cancer, yet many patients do not respond to these therapies.
No reliable predictive biomarkers. Despite the clinical success of checkpoint blockade therapies like atezolizumab targeting PD-L1, no single biomarker reliably predicts which patients will benefit. This means many patients receive treatments that are ineffective for them while missing out on alternatives that might work better.
Immune escape as the core problem. As tumors grow, cancer cells develop strategies to evade immune destruction. They recruit inhibitory immune cells such as regulatory T cells (Tregs) and tumor-associated macrophages, and upregulate immune checkpoint proteins like PD-1, PD-L1, and CTLA4, which together suppress immune attack on the tumor.
Transcriptomic profiling as a solution. The maturation of high-throughput RNA sequencing and deconvolution algorithms that can infer immune cell types from gene expression data made it possible to systematically characterize the immune landscape of bladder cancer tumors and potentially predict therapy response from tumor gene expression alone.
Data sources. The study used RNA sequencing data from 406 bladder cancer patients in The Cancer Genome Atlas (TCGA) as the training and primary validation cohort. Three independent external validation cohorts (GSE13507, GSE31684, and GSE32894) totaling 657 additional patients were also used, along with the IMvigor210 cohort of 298 urothelial carcinoma patients treated with the PD-L1 inhibitor atezolizumab.
Differentially expressed gene screening. Starting from 4,946 differentially expressed genes identified between TP53-mutant and TP53-wild-type bladder cancer samples (using FDR less than 0.05), the researchers intersected these with a published immune gene database, yielding 145 candidate immune-related genes with potential prognostic value.
LASSO algorithm for variable selection. The Least Absolute Shrinkage and Selection Operator (LASSO) Cox regression algorithm penalizes models for including too many variables, automatically eliminating genes that do not contribute meaningful prognostic information. Using 1,000 rounds of 10-fold cross-validation, this process narrowed the 145 candidates down to a final set of 13 mRNA genes.
Immune Signature Score calculation. The Immune Signature Score (ISS) for each patient was calculated as a weighted sum of the expression levels of all 13 genes, where the weights were the Cox regression coefficients assigned by the LASSO model. Patients were then divided into high-risk and low-risk groups based on the median ISS.
Strong prognostic performance in training data. In the TCGA training cohort, high-risk patients had significantly worse overall survival compared to low-risk patients, with a hazard ratio of 2.21 (95% CI: 1.57-3.12, P less than 0.001). This means high-risk patients had more than twice the mortality risk of low-risk patients.
Validated in multiple independent datasets. The prognostic value held across all three external validation cohorts: GSE13507 (HR = 1.78, P = 0.014), GSE31684 (HR = 1.89, P = 0.083), and GSE32894 (HR = 5.81, P less than 0.001). Time-dependent ROC curve analysis confirmed the signature accurately predicted both overall survival and progression-free survival.
Independent prognostic factor. Multivariate Cox regression confirmed the immune signature predicted prognosis independently of age, gender, histologic grade, neoadjuvant treatment, and pathologic stage. The ISS carried a hazard ratio of 3.11 (P less than 0.001) in multivariate analysis, making it one of the strongest independent predictors identified.
Works across disease subtypes. Subgroup analyses showed the signature performed well across different tumor stages and grades, and in both TP53-mutant and TP53-wild-type patient subgroups, demonstrating its broad applicability across the heterogeneous landscape of bladder cancer.
Building a clinical prediction tool. To make the prognostic information accessible to clinicians, the researchers constructed a nomogram combining the three independent prognostic factors identified by multivariate analysis: ISS, patient age, and pathologic stage. The nomogram provides a visual score that predicts 1-year, 3-year, and 5-year survival probabilities.
Calibration curve validation. The nomogram's predictions were validated using calibration curves, which compare predicted survival probabilities to actual observed outcomes. The calibration curve closely followed the ideal 45-degree reference line, indicating that predicted and observed survival rates were highly consistent.
Clinical utility confirmed by decision curve analysis. Decision curve analysis (DCA) was used to determine whether using the nomogram to guide clinical decisions would provide net benefit compared to treating all patients or no patients. The results demonstrated that the nomogram offered meaningful clinical benefit across a wide range of threshold probabilities.
Practical implementation. The nomogram allows any clinician to calculate a patient's personalized survival probability using only three inputs -- the ISS from the gene expression signature, the patient's age, and their pathologic stage -- making it practical for real clinical settings.
Dominant immune cell types in bladder cancer. Using the quanTIseq immune deconvolution algorithm on TCGA RNA-seq data, the researchers found that macrophages (both M1 and M2 subtypes), regulatory T cells (Tregs), and neutrophils are the predominant immune cell types infiltrating bladder cancer tumors.
Inhibitory immune cells dominate high-risk tumors. High-risk patients (high ISS) showed significantly higher infiltration of macrophages, monocytes, and Tregs compared to low-risk patients. Both macrophages and Tregs showed strong positive correlations with ISS, while neutrophil infiltration showed a negative correlation, suggesting that a suppressive immune environment defines high-risk disease.
Immune checkpoint upregulation. High-risk patients had significantly higher expression of five key immune checkpoint molecules: PD-1, PD-L1, CTLA4, LAG-3, and TIM-3. These checkpoints collectively suppress cytotoxic immune responses and allow cancer cells to evade elimination.
Pan-cancer relevance. The immune signature showed prognostic value not only in bladder cancer but also in cervical, head and neck, kidney, liver, and ovarian cancers in pan-cancer analysis, suggesting that the underlying immune escape mechanisms captured by this signature are shared across cancer types.
TGF-beta pathway activation. Single-sample gene set enrichment analysis (ssGSEA) revealed that the TGF-beta signaling pathway was significantly activated in high-risk patients. TGF-beta is known to shape the tumor microenvironment by restricting T cell infiltration into tumors, contributing to the immune-excluded phenotype where immune cells cannot penetrate the tumor core despite being present at the margins.
Epithelial-mesenchymal transition enrichment. GSEA analysis showed that the EMT signaling pathway was the most significantly enriched pathway in high-risk patients. EMT is the process by which cancer cells acquire migratory and invasive properties, and it is also linked to immune suppression through upregulation of PD-L1 and TGF-beta in the tumor microenvironment.
Additional activated pathways. GO and KEGG pathway analyses linked the immune signature to extracellular matrix organization, PI3K-Akt signaling, focal adhesion, and ECM-receptor interaction pathways. The G2M checkpoint and mismatch repair pathways were also activated in high-risk patients, indicating alterations in cell cycle regulation and DNA repair.
Stromal infiltration increase. The ESTIMATE algorithm, which uses gene expression patterns to infer stromal cell presence, confirmed that high-risk patients had higher stromal infiltration alongside the elevated immune infiltration. This is consistent with TGF-beta's role in promoting fibrosis and stromal activation as part of tumor immune exclusion.
Testing the signature in a real immunotherapy trial. The IMvigor210 cohort included 298 urothelial carcinoma patients treated with atezolizumab, a PD-L1 inhibitor. In this cohort, 47% of patients had an immune-excluded tumor phenotype, 27% had immune desert tumors, and 26% had immune-inflamed tumors.
An apparent paradox resolved. High-risk patients, despite showing an immune-excluded phenotype with strong immune escape characteristics, actually had higher rates of complete and partial response to PD-L1 inhibitor therapy compared to low-risk patients. This initially paradoxical finding is explained by the fact that high-risk patients express more PD-L1, giving PD-L1 inhibitors more target to block, thereby restoring immune activity more effectively.
ISS predicts treatment response. Patients who achieved complete response (CR) had significantly higher ISS values than those with progressive disease (PD) or stable disease (SD). This confirms that the immune signature not only predicts prognosis but also identifies which patients are most likely to respond to PD-L1 inhibitor therapy.
Combined signature improves prediction accuracy. Combining the ISS with tumor mutational burden (TMB) produced an area under the ROC curve (AUC) of 0.736 for distinguishing responders from non-responders to atezolizumab therapy, outperforming either metric alone and suggesting a multi-biomarker strategy improves treatment selection accuracy.
A comprehensive prognostic and predictive classifier. The 13-mRNA immune signature addresses two major unmet clinical needs simultaneously: predicting overall survival prognosis in bladder cancer patients and identifying who will benefit from PD-L1 checkpoint inhibitor immunotherapy. Most existing biomarkers address only one of these needs.
Characterizing the immune microenvironment at scale. By linking the signature to specific immune cell types, checkpoint expression levels, and signaling pathway activation, the study provides a mechanistic explanation for why high-risk patients behave differently and suggests that the signature captures a biologically meaningful disease subtype rather than a statistical artifact.
Implications for treatment stratification. Patients could be classified into high-risk and low-risk groups based on their ISS, with high-risk patients being the best candidates for PD-L1 inhibitor therapy. Low-risk patients, whose tumors have lower PD-L1 expression and a less suppressive microenvironment, may benefit more from alternative treatment strategies.
Foundation for future research. The identification of TGF-beta as a key activated pathway in high-risk patients points to TGF-beta inhibition as a potential combination strategy with PD-L1 blockade. The pan-cancer validity of the signature also suggests its utility may extend beyond bladder cancer to multiple tumor types with similar immune escape mechanisms.