Bladder Cancer Gene Signature Holds Up Across Two Large Cohorts

Bladder urothelial carcinoma develops inside an immune-active, metabolically stressed, and tissue-remodeling microenvironment, and a new secondary analysis of public transcriptomic datasets has now mapped how reproducible the tumor-associated transcriptional changes in that environment really are. The study, published in Frontiers in Cell and Developmental Biology, analyzed two independent bulk-expression cohorts, TCGA-BLCA and GSE13507, and found that tumor-versus-normal expression effects were positively correlated across 16,759 matched genes, with a Spearman correlation of 0.58 and a P value below 1 × 10−300. In total, 4,102 genes were statistically significant in both cohorts with the same direction of change, a level of directional concordance that the authors present as evidence of a cross-cohort reproducible tumor-associated transcriptional signal rather than a dataset-specific artifact.

The two bulk cohorts differed substantially in technology and composition, which is precisely why the comparison was informative. TCGA-BLCA contributed 412 primary tumors and 19 solid-tissue normal samples, while GSE13507 contributed 165 primary bladder cancer samples and 9 normal bladder mucosa controls, with one sample annotated as a multipotent progenitor-cell sample excluded from the control group. TCGA transcript-per-million values were log2-transformed, GSE13507 series-matrix values were already normalized on a log2 scale, and because the cohorts were generated on different platforms, absolute expression values were never pooled. Differential expression was estimated independently in each cohort using two-sided Welch t tests, with Benjamini–Hochberg adjustment and a prespecified combined threshold of FDR below 0.05 and an absolute log2 fold change of at least 0.5.

The scale of the tumor-associated remodeling was broad in both datasets. In TCGA-BLCA, 5,811 genes met the combined threshold, and the differential signal was distributed across a wide transcriptional space rather than concentrated in a handful of markers. Among the genes with the most negative effects were PI16, with a log2 fold change of −6.54, and DES, at −6.51. In GSE13507, 7,708 genes reached FDR significance and 3,379 also met the effect-size threshold. The authors emphasize that directional concordance across two sampling contexts supports reproducibility but does not prove that every concordant gene is urothelial-carcinoma-specific or tumor-cell autonomous, since the reference tissues differ biologically in stromal content, inflammatory exposure, and epithelial composition.

From the TCGA discovery set, the researchers derived a compact 15-gene signature built exclusively from genes with FDR below 0.05 and a positive log2 fold change greater than 0.5, ranked by adjusted P value. The score, calculated as the unweighted mean log-transformed expression of available signature genes, averaged 1.311 in tumors versus 0.155 in normal tissues in TCGA. Critically, the signature was then evaluated in GSE13507 without any gene re-selection or refitting, and its mean value was 8.029 in primary tumors versus 7.870 in controls. Although the external difference was smaller than in TCGA and absolute values were not directly comparable across platforms, the directional transfer without re-selection supported the reproducibility of the underlying tumor-associated signal.

The signature also varied across reported pathologic stages in TCGA, with mean scores of 1.00 for stage I, 1.03 for stage II, 1.42 for stage III, and 1.45 for stage IV among 403 patients with stage information after patient-level sample selection. The authors are careful to note that only two stage I tumors remained after that selection, so the pattern should not be read as a smooth biological progression across stages or as a clinically useful stage classifier. The analysis is presented as a description of biological distribution rather than a diagnostic or prognostic task, and the authors state that independent cohorts with clinically comparable stage distributions, prespecified thresholds, and harmonized endpoints would be required before any classification or prognostic use could be considered.

To place the bulk-derived signals in a cellular context, the team turned to an exploratory single-cell RNA-sequencing component using five pure-urothelial-carcinoma tumor regions from three patients in dataset GSE293189. After quality filtering, 1,512 cells were retained, and Leiden clustering resolved 13 clusters that were assigned by marker overlap to six identities: epithelial/tumor-associated cells, myeloid cells, T cells, B cells, NK cells, and fibroblasts. The composition varied dramatically by region. In the HG and SG tumor regions, epithelial/tumor-associated fractions ranged from 94.2 percent to 100 percent, whereas the P12049 region had an epithelial fraction of just 8.1 percent and contained the broadest mixture of immune and stromal identities. Because inferred copy-number profiles were not used to establish malignant status, epithelial assignments are described as epithelial/tumor-associated rather than definitively malignant.

Within these three patients, the highest mean luminal score occurred in epithelial/tumor cells at 0.589, the highest EMT score in fibroblasts at 1.028, and the highest immune score in NK cells at 1.229. Among assigned epithelial/tumor cells, basal scores ranged from 0.000 to 1.359 and luminal scores from 0.000 to 2.619, with EMT varying across the same continuum. Basal and EMT scores were positively correlated, while luminal and EMT scores were negatively correlated. The authors stress that this marked region dependence cautions against interpreting cell fractions as prevalence estimates for bladder cancer generally, and illustrates how bulk-expression differences can arise partly from cellular admixture rather than from within-cell transcriptional regulation alone.

Finally, five candidate genes were taken into the laboratory for quantitative reverse-transcription PCR as an orthogonal test of expression direction. Using the immortalized normal urothelial line SV-HUC-1 as the nonmalignant control and the bladder-cancer cell lines T24, 5637, and UM-UC-3, all five candidates—ULBP1, IL37, COL10A1, COL11A1, and IGF2BP3—showed directionally concordant upregulation. Relative to SV-HUC-1, fold changes in UM-UC-3 cells were 3.12 for ULBP1, 2.08 for IL37, 5.42 for COL10A1, 4.21 for COL11A1, and 3.01 for IGF2BP3, with all comparisons statistically significant across three independent biological replicates. COL10A1 and COL11A1 showed the largest relative increases, while IL37 displayed a more modest but still consistent rise.

The authors frame apoptosis, necroptosis, pyroptosis, and ferroptosis as a restrained interpretive framework rather than measured activity states, and they are explicit that neither transcriptomic nor qRT-PCR data alone establish functional execution of any of these death programs. Transcript abundance is an indirect readout that can reflect susceptibility, resistance, cell composition, or stress adaptation rather than active death signaling, and no modality-specific protein-level execution marker or functional death assay was measured. The study also identifies substantial limitations: the TCGA normal group was small, the 15-gene score was not developed through nested cross-validation, the single-cell component comprised only three patients and five regions with low detection fractions for several candidates, cluster identities were assigned by marker overlap without orthogonal protein validation, and no formal tumor-purity deconvolution was applied.

What the study delivers, then, is a layered but deliberately bounded result: a broad, directionally reproducible tumor-associated transcriptional landscape across two independent cohorts; a compact 15-gene signature whose distribution varies across reported pathologic stages; exploratory single-cell localization showing that these signals arise within markedly heterogeneous epithelial, immune, and stromal compartments; and concordant qRT-PCR upregulation of five selected candidates in controlled cell models. The authors propose that future work should test the candidates in independent cohorts with biologically comparable reference tissues, evaluate signature stability by resampling, quantify the influence of tumor purity and immune and stromal fractions, and use pathway-appropriate protein and biochemical readouts to test whether the qRT-PCR-supported candidates causally influence any specific regulated cell-death modality.

Subject of Research: Cross-cohort reproducibility of tumor-associated transcriptomic signatures in bladder urothelial carcinoma, with exploratory single-cell localization and in vitro qRT-PCR validation of selected candidate genes.

Article Title: Inflammation- and regulated-cell-death–relevant transcriptomic signatures in bladder urothelial carcinoma: cross-cohort reproducibility, exploratory single-cell localization, and in vitro qRT-PCR validation

Article References: Wang, W., Zhang, X., Li, Z., Wang, D., & Cai, H. (2026). Inflammation- and regulated-cell-death–relevant transcriptomic signatures in bladder urothelial carcinoma: cross-cohort reproducibility, exploratory single-cell localization, and in vitro qRT-PCR validation. Frontiers in Cell and Developmental Biology, 14, Article 1986148. https://doi.org/10.3389/fcell.2026.1986148

Image Credits: AI Generated

DOI: 10.3389/fcell.2026.1986148

Keywords: bladder urothelial carcinoma, transcriptomic signature, TCGA-BLCA, GSE13507, single-cell RNA sequencing, qRT-PCR, tumor microenvironment, regulated cell death, inflammation, epithelial–mesenchymal transition, cross-cohort reproducibility, differential expression