How a short dose of a mitochondrial DNA editor lowers its edits at the wrong sites

In Short. A mitochondrial DNA editor delivered in a short dose that then fades still makes the intended edits and makes far fewer at the wrong sites, in cultured cells and in mouse eyes.

Leber hereditary optic neuropathy takes the central vision of young adults, usually in one eye and then the other within weeks. A 2015 count of adults in northeast England found it among the most common inherited diseases traced to the cell's energy parts, at about 3.65 affected adults per 100,000. Forty-three carried the same letter change in one gene.

That change sits in mitochondrial DNA, the small separate loop of DNA held inside the parts of a cell that make energy. It is not the long genome in the nucleus, and a cell carries thousands of copies rather than two, one reason correcting it is hard.

The tool for the job is a base editor, a protein that changes one DNA letter into another without cutting the strand. The mitochondrial kind came out of a bacterial toxin. Beverly Mok and colleagues reported in Nature in 2020 that this toxin deaminates cytidines inside double-stranded DNA, and that two inactive halves of it, each fused to a programmable DNA-binding protein, make an editor that works inside mitochondria with no guide RNA. Those cytosine editors are called DdCBEs, and a later class, TALEDs, makes the other conversion, A to G.

The difficulty has been precision. An off-target edit is a change made at a site other than the one aimed at, and mitochondrial base editors make many of them elsewhere on the loop. They pile up the longer the editor keeps being made inside a cell. Delivering the editor's gene to mice inside a virus, which keeps producing it for weeks, has sometimes caused severe harm and death.

So Hyewon Jang and Aditya Raguram of the Whitehead Institute, with Carolline Rodrigues Menezes and Krzysztof Palczewski of the University of California, Irvine, tried a delivery method that stops itself. A virus-like particle is a hollow protein shell built from the scaffolding a virus uses, holding no virus genes. What is placed inside is used once.

Their first attempt placed the finished editor protein inside, standard practice for these particles, and produced no detectable editing. The likely reason is the tag that routes the editor into the mitochondrion. It must be exposed at the front while the protein is still being built, so an editor that arrives already folded is never recognized.

The answer was to pack the instructions instead. The particles were loaded with messenger RNA for the editor, so the receiving cell makes the protein itself and the tag is read as it emerges. The editor is made in a short dose that fades instead of being made for days, transient rather than sustained expression. Three adjustments made it work. The ratio of the two scaffold proteins mattered most, with 60 percent of the ordinary version best. Lowering the amount of the surface protein that lets particles enter cells helped again. And the A-to-G editors interfered with their own manufacture, so the team added a tag that lets a small molecule destroy the editor protein in the producing cells while sparing its RNA, lifting efficiency 7.1-fold. Across the targets tested, ND1, ND4 and ATP8 in human cells and Nd4 and Nd5 in mouse cells, editing in bulk cultured cells went from under 1 percent to over 50 percent.

Precision is the claim. In human cells in a dish, set against the older practice of delivery on a plasmid, the particles reached equal or better on-target editing with average off-target editing across the loop 41-fold lower at ND1 and 23-fold lower at ND4, measured against the higher plasmid dose. Against the lower dose the margins were 5.4-fold and 3.5-fold.

The particles installed the mouse counterpart of the Leber mutation in mouse embryonic fibroblasts, up to 88 percent. In a lymphoblastoid cell line from a patient carrying the human version, m.11778G>A in ND4, cells in a dish, they corrected up to 41 percent. A single injection under the retina of adult mice produced 6 to 20 percent editing, averaging 12 percent, in the retinal pigment epithelium seven days later, with off-target editing averaging 0.06 percent across the loop. The work is a preprint and has not been peer reviewed.

What it shows is narrow. Delivery that fades can hold on-target editing where it was and lower off-target editing across the mitochondrial genome. What it does not show is a treatment. Nothing here was done in a person. The patient mutation was corrected in cells in a dish, the editing in the living mouse eye was modest, and the animals were healthy mice, not a disease model.