NEW YORK – New algorithms developed at Spain's Center for Genomic Regulation, or CRG, enable the direct detection of two additional RNA modification classes using nanopore sequencing. Moreover, the researchers have devised a way to predict whether a specific RNA molecule was modified, not just a position across several RNA strands, enabling quantitative analysis of RNA modification dynamics.
"We find that [current intensity] information alone is insufficient to predict whether some reads are modified or not," Eva Maria Novoa, group leader for epitranscriptomics and RNA dynamics at CRG, said in an email.
Her international team explored combinations of additional features available from the Oxford Nanopore Technologies platform, including dwell time — a measure of translocation through the pore — and trace — a metric related to base calling probability. In a paper published this month in Nature Biotechnology, they showed that using their software, called nanoRMS, to analyze both current intensity and trace led to accurate estimates of modification stoichiometry, even at low levels of modification.
The researchers used their method to benchmark detection of two types of RNA modifications, pseudouridine and 2'-O-methylation (Nm). "However, it is in principle applicable to any type of RNA modification," Novoa said.
"These are two of the most common, most important RNA modifications, and it is critical to add these to the nanopore toolbox for the field to better understand the impact of these modifications on disease," said Chris Mason, a researcher at Weill Cornell Medicine who has collaborated with Novoa on previous work in this vein, but who was not involved in the new paper. "This work opens up exciting possibilities for single-molecule, direct RNA sequencing and the field of the epitranscriptome. Specifically, it lets researchers phase the coincidence of the modifications on the same molecule, which can reveal new regulatory layers of RNA function, stability, and folding."
There are over 100 different RNA modifications, Mason said, but solid methods to detect them with RNA sequencing have only recently begun to emerge. Direct detection of DNA modifications was first demonstrated in 2013, by teams from the University of Washington and the University of California, Santa Cruz, respectively.
Most RNA sequencing methods convert RNA to cDNA, so direct detection of RNA modifications has been more challenging. Amplification also introduces hurdles to detecting RNA modifications. However, Oxford Nanopore demonstrated the ability to directly detect RNA in general in 2016, and at the same time showed proof of principle of directly detecting N6-methyladenosine (m6A) methylation in an engineered sample.
Because detecting modifications is so closely tied to base calling on the Oxford Nanopore platform, many of the advances have been made on the computational side. Novoa's lab developed EpiNano, a software that helped demonstrate m6A direct detection in native RNA sequences, and Mason was a coauthor of the 2019 Nature Communications paper that introduced the software.
By analyzing the "errors" in base calling from electric current as modified bases are passing through the nanopore, EpiNano was able to detect several common types of methylation, including 5-methylcytosine, a widespread modification.
EpiNano could predict whether a position across similar RNAs was modified "but could not predict whether a specific read was modified," Novoa said. In the new method, the software identifies the features of a modification and uses machine learning algorithms to classify the reads into two bins, modified and unmodified. "This means that we can now use nanopore sequencing to study RNA modification dynamics in a quantitative manner, and in a transcriptome-wide fashion," she said.
In proof-of-concept studies, the researchers used RNA from Saccharomyces cerevisiae to show that pseudouridine modifications could be detected from uridine-to-cytosine base calling mismatches. They also were able to predict pseudouridine modifications in yeast mitochondrial ribosomal RNAs as well as messenger RNAs — which are modified at much lower levels than rRNAs, as validated with their new methods.
"Validation on more varied sample types and tissue sources would have been a nice addition, but the existing data is already quite strong," Mason said. The authors cautioned that not all RNA modifications lead to base calling errors with single-nucleotide resolution, and that detecting them can be partly dependent on sequence context.
But having single-molecule resolution, along with the new ability to detect more modification types, will open up new lines of research to Oxford Nanopore users. Researchers may be able to quantify the dynamics of RNA modifications across different conditions, for example, such as in disease or across developmental stages, or study the interplay between RNA modifications and polyadenylation status, or even different modifications on the same molecule.
"Of note, the latest RNA vaccines for COVID-19 have extensive presence of pseudouridines, so this might help validation of future RNA vaccines, too," Mason said.
All software used in the study is available through GitHub.