Finding. Mitochondrial base editors get precise when you mail them as a short-lived mRNA inside a virus-like particle, not as a lingering protein or a plasmid. Jang, Palczewski and Raguram find that canonical protein-packaging VLPs do not edit mtDNA at a detectable level. Engineered mRNA-packaging particles (mtBE-VLPs) do. Iterating the particle composition lifts bulk-culture editing from below 1% to above 50% at several mitochondrial sites. A trick that damps cargo expression in producer cells but not in transduced cells adds another 7.1-fold for some editors. Against plasmid DNA, on-target rates match and mtDNA-wide off-targets drop 23- to 41-fold. The same particles write a pathogenic mutation into 88% of mouse-embryonic-fibroblast mtDNA, correct a pathogenic mutation in 41% of a human patient-derived line, and, after one subretinal injection, edit 6% to 20% (average 12%) of mouse retinal-pigment-epithelium mtDNA with minimal genome-wide mitochondrial damage.
Why this paper matters
Every mitochondrial base editor has the same original sin. There is no guide-RNA off-switch in the matrix. The longer the protein lives, the more the mitochondrial genome looks like a dartboard. Plasmid and virus make that worse on purpose. Transient mRNA in a VLP is the obvious fix if the particle can be made to work. Protein-packaging failed. mRNA-packaging, after composition work, did not. That is a platform sentence for anyone sitting on a pathogenic mtDNA variant.
The in vivo number is modest and honest: 12% mean RPE after one shot. The specificity number is the one that changes practice. 23- to 41-fold fewer off-targets at matched on-target is why you would pick this vehicle even before you chase a higher tissue percentage.
What they actually measured
Failed protein VLPs, successful mRNA VLPs, titration of composition, a producer-cell expression split, multi-locus culture efficiencies, mtDNA-wide off-target surveys, MEF install, patient-line correction, subretinal mouse RPE. No vision assay.
How to read the score
Mid 90s. Delivery plus specificity plus in vivo plus a patient-line correction. Confidence is high for the editing numbers as stated. It is not a therapy.
Caveats
12% is not a cure. Heteroplasmy outcome (shift versus mixed edited genomes) needs the figures. Systemic tissues are untested here. Do not inject patients from this brief.
What to do with it
If you run mtBEs, stop leading with plasmid when you care about off-targets and try these particles. Pull the 23- to 41-fold off-target table and the producer-cell damping trick. If you model LHON or other RPE/retina mtDNA disease, the subretinal 12% is your first in-animal number to beat.