OptiPrime selects RNA instructions for DNA editing based on cellular repair
OptiPrime selects RNA instructions for DNA editing based on cellular repair
On August 12, Nature Biotechnology published a study of OptiPrime, a model that selects RNA instructions for prime editing, a method for precisely replacing short DNA sequences. The authors trained the model on 297 962 measurements and tested the selected instructions in cells and mice.
A prime editor nicks one DNA strand and copies the intended change from a pegRNA instruction. The cell then repairs the site. At this stage, the original and edited versions of the sequence are both present. The mismatch repair system (MMR) can restore the original version, causing the edit to disappear.
Making a single change requires selecting the target site in the DNA, the design of the RNA instruction, and nearby silent substitutions. These substitutions preserve the protein sequence but alter how cellular repair machinery recognizes the site. The number of possible designs quickly reaches the hundreds. The authors assembled large libraries of instructions and tested them in cells with impaired MMR and in cells with functional MMR. This comparison separated the effect of MMR from other reasons why one instruction works better than another.
OptiPrime evaluates each stage of editing separately. Different parts of the model predict whether the editor will find its target, whether it will copy the intended change, and how cellular repair machinery will process the resulting site. The model then combines these estimates and ranks the instructions. Prediction accuracy declined when the authors removed components from this framework or randomly swapped their positions.
In May, David Liu’s team redesigned the prime editor protein so that more active enzyme would remain in the cell. OptiPrime selects an RNA instruction for this editor while accounting for DNA repair.
When correcting the CFTR F508del variant, the top eight OptiPrime candidates achieved up to 22% editing, compared with 11% for the manual strategy used in the same study.
In the Kif1a mouse model, OptiPrime first proposed eight combinations of silent substitutions. The researchers then tested seven additional RNA segment length variants for the best combination. In the next two experiments, they selected a separate guide RNA for nicking and a version of the editor protein. The four experiments took four weeks. After the selected system was delivered to newborn mice, the mean proportion of corrected Kif1a sequence in the cerebral cortex exceeded 40% at four weeks.
The model gives laboratories a short list of instructions and indicates which stage is limiting editing efficiency. Instead of screening hundreds of designs, researchers can test specific hypotheses about how cellular repair machinery will process the edited DNA sequence.