Background

In a variety of organismal and developmental contexts, the localization of specific RNAs to specific subcellular locations promotes cellular and organismal function. For example, mate type switching in yeast [1] and developmental patterning in Drosophila [2,3] are controlled in part through the trafficking of specific mRNAs to precise subcellular locations. Mutations in RNA-binding proteins involved in RNA transport are associated with a variety of human neurological diseases [4,5]. However, for most localized RNAs, the mechanisms that govern their transport remain unknown. This includes the cis-elements within transcripts that mark them as RNAs to be localized, as well as the RNA-binding proteins (RBPs) that mediate the process.

A common experimental technique in the study of RNA localization is the characterization of RNA populations at subcellular locations. Historically, this has often been achieved through cellular fractionation, either by biochemical or mechanical means, followed by RNA isolation and characterization from the resulting fractions [6–12]. Although these techniques have led to many insights, they also have disadvantages. Mechanical cellular fractionations often require that the interrogated cell type have specific morphologies. For example, neurons are commonly separated into soma and neurite fractions, but the technique for doing so relies on cells having long, thin projections [7]. Biochemical fractionations, although more flexible regarding cell shape and morphology, have their own challenges. The association between an RNA and a subcellular location or organelle may be weak or transient and therefore lost during purification, leading to false negatives. Similarly, RNA molecules in subcellular environments that are spatially distinct within intact cells may nevertheless have similar biochemical properties and therefore copurify, leading to false positives.

Newer techniques that rely on proximity labeling remove many of these concerns. In these approaches, bait proteins are first localized to the subcellular location of interest [13–16]. Upon a chemical or light-based stimulus, RNAs in close proximity to the bait protein (usually 20–100 nm) are specifically labeled. In the majority of the techniques reported to date, these labels result in the biotinylation of bait-proximal RNAs, facilitating their downstream isolation with streptavidin and characterization.

OINC-seq builds on one of these RNA proximity labeling methods, Halo-seq [13,17]. In Halo-seq, a HaloTag-fused protein is specifically localized to the subcellular location of interest [18]. HaloTag domains covalently bind to Halo ligands. Upon addition of Halo-DBF ligand, a dibromofluorescein (DBF) molecule therefore assumes the same subcellular distribution as the Halo-tagged bait protein. When irradiated with green light, DBF emits singlet oxygen radicals that oxidize nearby RNAs. Because guanosine residues have the lowest redox potential of any of the four bases, they are preferentially oxidized to form 8-oxoguanosine as well as further oxidized hydantoins.

In Halo-seq, these oxidized guanosine residues are then alkynylated, which makes them substrates for biotinylation through Click chemistry. With OINC-seq, however, we sought to streamline the process to detect oxidized guanosines directly without the need for alkynylation, biotinylation, and purification. Instead, following the light-induced oxidation of bait-proximal RNA, total RNA is isolated and reverse transcribed. Reverse transcriptase predictably misinterprets oxidized guanosine [19], leading to G to T, G to C, and G deletion mutations in the resulting cDNA. Quantification of these mutations via high-throughput sequencing therefore provides a readout of the relative proximity of each RNA species to the Halo-tagged bait protein. In order to identify and quantify these mutations in RNAseq data, we developed the PIGPEN software. PIGPEN takes in RNAseq reads from an OINC-seq experiment and returns tables of mutation rates for each gene detected in the sample.

In general, OINC-seq experiments can take one of two forms. They can either be designed to assay the entire transcriptome, or they can be focused on one RNA through the interrogation of an amplicon. Although limited in focus to one RNA species, amplicon-based experiments typically have extensive read depth for the RNA of interest, leading to more accurate results. In contrast, although transcriptome-wide experiments interrogate thousands of RNA species at once, the reduced read depth assigned to any one RNA species can lead to noisier results.

Oxidation-induced mutations are relatively rare. This rarity requires that other, non-oxidation-derived sources of mutations be suppressed as much as possible. For example, Illumina sequencers miscall approximately 1 out of 1,000 bases in a typical sequencing reaction. In principle, it is impossible to distinguish between such a miscall and an oxidation-induced mutation. The usage of paired-end reads can mitigate this by requiring that any given mutation be seen in both reads of a mate pair in order to be classified as a true mutation and not a product of sequencer error. Specific options in PIGPEN implement this functionality.

RNA in close proximity to the bait protein may be expected to have multiple oxidation events and, therefore, multiple mutations along the same molecule. Random mutations, on the other hand, may be expected to be more evenly distributed across the entire RNA population. The appearance of multiple mutations within the same read might then give additional confidence in the localization of an RNA. Options in PIGPEN also support this functionality.

Finally, because of the rarity of oxidation-induced mutations, the fidelity of the polymerases used in the making of RNAseq libraries also matters. During the development of OINC-seq, we tested multiple library preparation kits and found that Quantseq kits from Lexogen consistently had the lowest background mutation rate in untreated samples, making them ideal for use with OINC-seq. However, these kits only interrogate the 3’ ends of molecules with poly-A tails. RNAs without these tails are therefore not assayable using this kit. It is possible that other library preparation methods would have suitable rates of background mutations, but they have not yet been tested. Amplicon-based experiments afford more flexibility in polymerase choice. We have found that the use of SuperScript IV for reverse transcription and NEB Q5 polymerase for PCR consistently leads to acceptably low levels of background mutations.