A single-sample mRNA molecular classification of bladder cancer predicting prognosis and response to immunotherapy

BACKGROUND: As an immunogenic cancer, crosstalk between cancer cells and immune cells has been gradually recognized in bladder cancer (BC). Several studies have emphasized the clinical significance of the molecular stratification of BC without highlighting the role of the immune microenvironment. Although immunotherapy acted as a prospective treatment, more precise molecular stratification should be established to select those sensitive to immunotherapy. METHODS: To select specific immune genes forming subtypes indicating disparate prognoses, we performed bioinformatic analysis using BC transcriptomic profiles from six published datasets, with 408 BC samples in The Cancer Genome Atlas (TCGA) database and 295 individuals in International Cancer Genome Consortium (ICGC) database. Survival analyses were conducted using Kaplan-Meier curves, while Kruskal-Wallis tests were applied to test the differences among groups. Except for unsupervised clustering based on the differential expression of genes, we additionally performed binomial logistic regression, focusing on the mRNA level of a single sample. RESULTS: Unsupervised clustering showed that 4 clusters captured the best segmentation. After validation with survival data and simplification using binomial logistic regression, we found that cluster B and cluster D showed worse survival outcomes (P=0.012). Considering the similar survival outcomes of these two clusters, we recombined and performed another survival analysis, which also showed significant survival differences (P=0.0041). Bonding with clinical data, a greater proportion of risk factors were assigned to the worse prognosis subtype, especially showing higher grades in the subtype (P<0.001). In addition, immune cell infiltration, single nucleotide polymorphism (SNP) and copy number variation (CNV) all showed differences between clusters, indicating changes in the immune microenvironment and mutation burden. Through phenotypical analysis, we found metabolism and proliferation phenotypes associated with the immune clusters and mutually exclusive in BC, of which proliferation contributed to worse outcomes. Using the tumor immune dysfunction and exclusion (TIDE) score, a worse immunotherapy benefit was predicted in clusters B&D, defined as the worse prognosis subtype. CONCLUSIONS: With this novel clustering criterion based on immune-related genes, we provide a better understanding of the immune microenvironment, further guiding the use of immunotherapy.