Transfer RNAs, or tRNAs, play a central role in building proteins. Each tRNA carries a specific amino acid and delivers it to the ribosome, where proteins are assembled. For this process to work correctly, the tRNA must be properly modified and correctly attached to its matching amino acid, a step known as aminoacylation.

Studying tRNAs has always been technically challenging. These molecules contain many chemical modifications, and researchers have struggled to measure three key features at the same time, the tRNA sequence, its chemical modifications, and whether it is actually carrying an amino acid. Until now, most methods could only look at one or two of these properties separately.

Researchers from the University of Colorado Anschutz Medical Campus and colleagues developed a method called aa tRNA seq. This approach allows scientists to directly sequence intact, amino acid charged tRNAs using nanopore sequencing technology.

The method works by chemically linking the amino acid to an adaptor molecule in a way that preserves the connection between the amino acid and the tRNA. The intact complex is then passed through a nanopore sequencer. As the molecule moves through the pore, it produces electrical signals. These signals change depending on the identity of the amino acid and the structure of the tRNA.

Model training and classification of aminoacylated tRNA reads from nanopore signal

A Schematic illustrating ground truth libraries for model training. Library (i) was prepared by only chemical ligation of the 3′ adapter to biological aminoacylated tRNAs, followed by gel purification and enzymatic ligation of the 5′ adapter. Library (ii) was prepared by enzymatic ligation of both adapters to deacylated biological tRNAs. B Strategy for dataset preparation and neural network training of a Remora model to identify charged and uncharged tRNA reads from nanopore current signals. C Schematic depicting reference-anchored signal from libraries (i) and (ii), with the 6-nt training window enclosed within the yellow box, and the position of the amino acid via dashed line. Mean current signals in picoamps for charged and uncharged tRNA reads are indicated by solid lines within each colored trace, with corresponding shading spanning the standard deviation. D Density plot showing the distribution of aminoacylation prediction scores (modification likelihood or “ML” score, 0–255 scale) assigned by Remora model in validation libraries (95,859 deacylated tRNA reads, 33,494 aminoacylated tRNA reads). E Confusion matrix and key metrics illustrating the model’s performance at classifying a second set of replicates prepared as in (A), using a cutoff of ML > 199 for identification of aminoacylated reads from nanopore signal.

To interpret these complex signals, the team built machine learning models. These models were trained to recognize subtle distortions in the nanopore signal that are caused by the presence of different amino acids attached to the tRNA backbone. This allowed the researchers to identify, at the single molecule level, which amino acid was attached to each tRNA in a sample.

The researchers applied aa tRNA seq to cells lacking specific tRNA modification enzymes. They confirmed that loss of certain modifications can make specific tRNAs unstable. They also discovered additional candidate tRNAs that may depend on particular modifications for stability and proper function.

This work provides a powerful new way to study how efficiently and accurately tRNAs are charged with amino acids. Because protein synthesis depends on precise matching between tRNAs and amino acids, understanding these processes is essential for studying cellular stress, disease states, and how cells respond to environmental changes.

By combining chemical ligation, nanopore sequencing, and machine learning, aa tRNA seq opens the door to exploring how tRNA sequence, modification patterns, and environmental conditions influence the fidelity of protein production.