Machine learning-based characterization of stemness features and construction of a stemness subtype classifier for bladder cancer.

BACKGROUND

Bladder cancer (BLCA) is a highly heterogeneous disease that presents challenges in predicting prognosis and treatment response. Cancer stem cells are key drivers of tumor development, progression, metastasis, and treatment resistance. The features and prognostic significance of stemness in bladder cancer need further investigation.

METHODS

We used bladder cancer datasets from the TCGA and GEO databases, based on stemness gene sets from the StemChecker database, to identify stemness subtypes using the consensus clustering algorithm. We calculated the mRNA expression-based stemness index (mRNAsi) using the OCLR algorithm. We compared the differences in overall survival, genomic characteristics, tumor microenvironment, and treatment response between the stemness subtypes. We constructed the stemness subtype classifier using machine learning algorithms such as LASSO regression, random forest, and multivariate logistic regression. The function of the classifier gene was validated through experiments.

RESULTS

We divided bladder cancer patients into two subtypes based on the enrichment scores of stemness gene sets. Patients within subtype 1 have a higher mRNAsi score, a better survival rate, an antitumor microenvironment, and higher sensitivity to immunotherapy, while patients within subtype 2 show higher aneuploidy scores, greater homologous recombination defects, an elevated tumor mutation burden, and increased chemotherapy sensitivity. We constructed a stemness subtype classifier based on six differentially expressed genes between the two subtypes. The classifier demonstrated good performance in predicting prognosis on three additional datasets from the GEO database and two non-muscle invasive bladder cancer datasets. Through tumor sphere formation experiments and western blotting, we found that TNFAIP6, out of the six classifier genes, maintains stemness. TNFAIP6 silencing also facilitated the chemotherapy response of cisplatin, docetaxel, and paclitaxel on bladder cancer cells. Furthermore, decreasing TNFAIP6 expression caused the immune checkpoint gene PD-L1 to downregulate.

CONCLUSION

This study provided valuable insights into the heterogeneity of BLCA stemness, and the stemness subtype classifier may facilitate molecular classification and personalized treatment selection for BLCA patients. Besides, TNFAIP6 may serve as a potential future stemness target guiding bladder cancer chemotherapy and immunotherapy.