An informatics approach to distinguish RNA modifications in nanopore direct RNA sequencing

Abstract

Modifications in RNA can influence their structure, function, and stability and play essential roles in gene expression and regulation. Methods to detect RNA modifications rely on biophysical techniques such as chromatography or mass spectrometry, which are low throughput, or on high throughput short-read sequencing techniques based on selectively reactive chemical probes. Recent studies have utilized nanopore-based fourth-generation sequencing methods to detect modifications by directly sequencing RNA in its native state. However, these approaches are based on modification-associated mismatch errors that are liable to be confounded by SNPs. Also, there is a need to generate matched knockout controls for reference, which is laborious. In this work, we introduce an internal comparison strategy termed “IndoC,” where features such as ‘trace’ and ‘current signal intensity’ of potentially modified sites are compared to similar sequence contexts on the same RNA molecule within the sample, alleviating the need for matched knockout controls. We first show that in an IVT model, ‘trace’ is able to distinguish between artificially generated SNPs and true pseudouridine (Ψ) modifications, both of which display highly similar mismatch profiles. We then apply IndoC on yeast and human ribosomal RNA to demonstrate that previously reported Ψ sites show marked changes in their trace and signal intensity profiles compared with their unmodified counterparts in the same dataset. Finally, we perform direct RNA sequencing of RNA containing Ψ intact with a chemical probe adduct (N-cyclohexyl-N′-β-(4-methylmorpholinium) ethylcarbodiimide [CMC]) and show that CMC reactivity also induces changes in trace and signal intensity distributions in a Ψ specific manner, allowing their separation from high mismatch sites that display SNP-like behavior.