A Direct RNA Sequencing Platform to Detect Uridine Modifications in Human Lung Cancer Transcripts

About This Grant

PROJECT SUMMARY/ABSTRACT RNA modifications are critical regulators of RNA stability, structure, and translation, yet their functions in cancer remain poorly understood because few technologies can detect them transcriptome-wide at single-nucleotide resolution. Uridine is the most chemically diverse base, giving rise to multiple modifications including uridine (U), pseudouridine (ψ), and dihydrouridine (D). ψ has been systematically profiled and is already being leveraged in therapeutic pipelines, such as premature stop codon suppression in cystic fibrosis and Hurler syndrome. By contrast, D has only recently been identified in mRNAs, and overexpression of its writer enzyme, DUS2, is associated with poor prognosis in lung adenocarcinoma (LUAD). However, no current technology can reliably distinguish D from ψ or U, leaving a critical gap in understanding how uridine modifications regulate RNA biology in cancer. Our preliminary studies demonstrate that D produces a reproducible U→V basecalling error in nanopore direct RNA sequencing (DRS), suggesting that this platform is uniquely suited to distinguish D from U and ψ. Building on our ModQuant machine learning framework, which integrates basecalling errors and ionic current features, we will develop the first broadly applicable technology for transcriptome-wide mapping of D. In the R61 exploratory phase, we will: (1) generate A549 knockout and overexpression lines for DUS1L and DUS2 and establish a high-confidence atlas of D sites using bootstrapped analyses with matched IVT controls; (2) develop machine learning classifiers using synthetic RNAs processed through our PRECISE-QC/ModQuant pipeline to achieve ≥80% three-way accuracy in distinguishing U, D, and ψ; and (3) apply this framework to LUAD tumors and matched normal tissues from the NCI Cooperative Human Tissue Network, incorporating patient-specific IVT controls, pathology-guided macrodissection, and computational deconvolution to control for tumor heterogeneity. The expected outcomes are the first transcriptome-wide maps of D in human cells and tumors, validated machine learning models for discriminating uridine modifications, and a framework that integrates genetic, synthetic, and clinical controls for rigorous modification detection. The impact of this work is twofold: it will close a major technological gap in the epitranscriptomics field by enabling accurate detection of modifications on the same base, and it will lay the mechanistic foundation for therapeutic strategies targeting uridine modifications in cancer, paralleling the translational trajectory already underway for pseudouridine.