Nucleic Acids Res. 2026 Apr 23. pii: gkag411. [Epub ahead of print]54(8):
While nanopore direct RNA sequencing has substantially advanced transcriptomics, its detection of RNA modifications remains primarily focused on abundant biological base modifications. However, therapeutic RNAs employ a diverse catalog of modifications, including base, sugar, and backbone modifications, to enhance stability and pharmacological properties. To address this gap, we systematically evaluated a set of therapeutically relevant modifications [phosphorothioate (PS)], sugar [2'-O-methylation (2'OMe), 2'-Fluoro (2'F), locked nucleic acid (LNA), 2'-O-(2'-methoxyethyl) (2'MOE)], and base [N1-methylpseudouridine (m1Ψ), 5-methylcytidine (m5C), 5-methoxyuridine (5moU), and 5-iodocytidine (5iodoC)] using direct RNA nanopore sequencing. Modifications were systematically analyzed using basecall errors, raw current signals, and modification-aware basecalling models. Ribose modifications, m1Ψ, and 5moU induced significant error rate increases and noticeable current alterations, whereas 2'OMe and 2'MOE affected dwell time adjacent to the pore. In contrast, PS linkages produced only slight current alterations without increasing basecalling errors. We further evaluated modification-aware basecallers for 2'OMe and m5C. While these tools can distinguish modification types, they are limited by poor quantification accuracy and high local error rates, especially for 2'OMe. This study establishes a critical performance baseline, clarifying the current capability and limitations of nanopore technology for the analysis of therapeutically relevant RNA modifications.