Deep mutational scanning and mobility-based selection for 6 RNA targets

RNA mutations are known to change mobility in native gels. What is not known is if mobility can serve as an effective tool to separate structurally similar (structural homologs) from structurally destabilized variants in a deep-mutation library of an RNA. Here we defined the proportion of a mutant in the native band from native polyacrylamide gel electrophoresis (PAGE) as a native-mobility-fitness score. The fitness scores of single and double mutants allowed a unsupervised, RNA-specific analysis to detect key secondary and tertiary base pairs through covariational signals. Subsequent amplification of these signals and their use as restraints for folding led to not only high-accuracy secondary structures with the F1-score > 0.9, but also quality tertiary-structure models between 3.6 Å and 7.7 Å RMSD from their native structures for the best in top 5 models for 6 RNAs tested including two CASP 15 difficult targets. This MobiSeq method should provide a simple and effective method for inferring 2D and 3D structures and improving mechanistic understanding of all structured RNAs. The RNA-MobiSeq pipeline is a method to predict the 3D structure of a target RNA from its sequence by obtaining artificial homologous sequences and their native-mobility-fitness scores from high-throughput sequencing of mobility-selected RNA mutants, followed by a new RNA-specific covariation analysis as restraints for energy-based structure prediction. Error-prone PCRs (or oligo pools) were first employed to obtain randomly mutated sequences, which were assembled into plasmids and then electro-transformed into DH5α for library constructions. The DNA libraries were further in vitro transcribed into RNAs that were subjected to native PAGE separation according to mobility. The band with the same mobility as the wild-type RNA sequence was excised for reverse transcription. Both pre- and post-selection samples were sent for high-throughput sequencing. The obtained sequences along with their fitness scores estimated from mutants’ fractions in the excised band were then analyzed for co-variation to infer the base pairing structures by an RNA-specific method CODA2 and MC simulated annealing. The resulting base pairing structures were employed as restraints for energy minimization to model the 3D structures of the RNA by using the BRiQ statistical energy function and a folding-tree-based folding algorithm.