Multiomics and machine learning-based analysis of pseudouridine modifications in pan-cancer

Abstract

Abstract Pseudouridine widely affects the stability and function of different types of RNA. However, our knowledge of pseudouridine properties in tumors is incomplete. We systematically analyzed pseudouridine synthases (PUSs) expression, genomic aberrations and prognostic features in 10907 samples from 33 tumors. We found that the pseudouridine-associated pathway showed significant abnormalities in tumors and affected patient prognosis. Dysregulation of the expression pattern of PUSs may arise from CNV mutations and aberrant DNA methylation. Functional enrichment analyses determined that the expression of PUSs was closely associated with the MYC, E2F and MTORC1 signaling pathways. In addition, PUSs are involved in the remodeling of the tumor microenvironment (TME) in solid tumors, such as kidney and lung cancers. Particularly in lung cancer, increased expression of PUSs is accompanied by increased immune checkpoint expression and Treg infiltration. The best signature model based on more than 10 random 112 machine learning combinations has good predictive prognostic ability in ACC, DLBC, GBM, KICH, MESO, THYM, TGCT, and PRAD, and is expected to guide immunotherapy for 19 tumors. In addition, the model was effective in identifying patients with tumors amenable to treatment with etoposide, camptothecin, cisplatin, and bexarotene. In conclusion, our work highlights the dysregulated features of PUSs, and their role in TME and prognosis, providing an initial molecular basis for future exploration of pseudouridine. Studies targeting pseudouridine are expected to develop potential diagnostic strategies, evaluate and improve antitumor therapies.