What Gene Editing Can Actually Do Now

One gene editing therapy has full regulatory approval. It is called Casgevy, it cuts DNA in cells taken out of the body, and as of February 2026 — more than two years after approval — roughly sixty people worldwide had received it. Everything else is a trial or a press release. That is not a disappointment. It is a map of where the constraints sit, and they no longer sit in the editing chemistry.

First in a series on the biology of ageing: gene editing, telomeres, epigenetics, the Blue Zones, and cellular reprogramming.

Three ways to change one letter of DNA

Gene editing is a ladder of three techniques, each rung doing what the one below cannot. The differences look like laboratory detail; they decide which diseases are treatable.

The first rung cuts. CRISPR-Cas9 is a programmable nuclease: a protein that severs DNA wherever a guide RNA — a short piece of RNA whose sequence matches the target — points it. It finds that stretch and cuts through both strands.

Then it stops. Cas9 repairs nothing; the cell does. The main repair route, end-joining, chews back a few bases from the broken ends before gluing them together — fast, sloppy, and the sloppiness is the whole story.

A cut is a demolition tool. Lose a few bases from the middle of a gene and the reading frame shifts: everything downstream is nonsense and the gene stops working — often exactly what you wanted.

Ask that same cut to restore one missing letter and you are relying on a repair process never told the right answer. A nuclease destroys sequence well and corrects it badly. That much is established.

The second rung converts one letter into another, chemically. Komor and colleagues removed the break, fusing a cytidine deaminase — an enzyme that strips a chemical group off a single DNA letter — to a Cas9 crippled to nick one strand rather than cut both.

The deaminase turns cytosine into uracil, a letter DNA does not normally carry, anywhere inside a five-base window at the target. The cell reads that uracil as thymine, sees the strands disagree, and rewrites the other strand to match. C has become T, with no break.

That first editor worked in 37% of cultured human cells, with unwanted insertions or deletions at or below 1%. Adenine base editors followed, doing the same in the other direction: A to G. Each family performs one fixed conversion and nothing else — hence the third rung.

The third rung writes new text from a template. Prime editing fuses a nickase Cas9 — again, one strand only — to a reverse transcriptase, the enzyme that copies RNA into DNA. Its guide RNA does double duty: it finds the site and carries the replacement text.

The editor nicks; the reverse transcriptase copies the replacement out of the guide into the genome. Anzalone and colleagues reported over 175 edits in human cells this way — all twelve single-letter substitutions, plus small insertions and deletions — at 20–50% efficiency in one immortalised cell line.

An immortalised cell line is human cells kept growing indefinitely in a dish, a far more forgiving place than a person. Cell-line efficiency is not tissue efficiency.

The same base pair, three tools: the cut disables the gene; only the lower two put the right letter back. Mechanisms per Komor (2016) and Anzalone (2019); the target is the CPS1 nonsense variant discussed below.

What has actually been approved

One product. Casgevy received its initial US approval in 2023 — sickle cell disease in December, transfusion-dependent beta thalassaemia the following January. Now labelled from age two, it was the first CRISPR-based medicine any major regulator approved.

The obvious guess is wrong: Casgevy does not repair the sickle mutation; it never goes near it.

Fetal haemoglobin is the version you make before birth; in adults a repressor protein called BCL11A keeps it switched off. Casgevy cuts a binding site in that gene’s erythroid enhancer — the stretch of DNA that switches the repressor on in the cells that become red blood cells. Break the site, the repressor is not made there, and fetal gamma-globin returns and dilutes the sickling.

So the first approved gene editing therapy works by demolition, because demolition is what a nuclease does well. The lowest rung got to market first.

Casgevy is also ex vivo: the editing happens outside the body.

The patient’s blood stem cells are collected by apheresis — blood is drawn, the wanted cells separated out, the rest returned — and edited at a facility.
The patient is given myeloablative busulfan conditioning: chemotherapy that clears the bone marrow the edited cells have to repopulate.
The edited cells are infused back.

The trial numbers show where that sequence breaks. In the pivotal sickle cell trial 63 people enrolled, 58 started mobilisation, 44 received an infusion. Six — 10% — never got the product, because too few cells could be collected.

Of 31 evaluable patients, 29 (93.5%) then went twelve consecutive months without a severe vaso-occlusive crisis: an episode of blocked vessels and acute pain, the event the trial was counting.

A strong result. Also a single-arm trial: everyone received the drug, with no randomised comparator group to measure it against.

The therapy built for one baby’s genome

In 2025 a team at the Children’s Hospital of Philadelphia and the University of Pennsylvania published something structurally different.

An infant, KJ, was born with severe CPS1 deficiency, a urea-cycle disorder: the body cannot dispose of the nitrogen left over from digesting protein. It affects about 1 in 1,300,000 people, with an estimated 50% mortality in early infancy. KJ carried two truncating variants, Q335X and E714X — each a premature stop signal that cuts the protein short.

The team built an adenine base editor and a guide RNA for his Q335X variant — not for CPS1 deficiency in general, for his particular spelling of it — packaged in a lipid nanoparticle, a fat droplet that carries the machinery to the liver.

The timeline is the part to read twice.

Months one and two: cell line work and editor screening.
Month three: a mouse carrying his variant.
Month five: primate toxicology.
Month six: clinical batch, off-target analyses, and the IND filing — the application that allows an experimental drug to be given to a person. Authorised a week later.

He was first infused at seven months old, at 0.1 mg per kilogram. The second dose came 22 days later, at 0.3.

Over the seven weeks that followed he tolerated more dietary protein and his nitrogen-scavenger dose was halved, with no serious adverse events.

The honesty starts here.

n is one. One patient, no control group.
Follow-up is seven weeks. Nobody knows what year two looks like.
A liver biopsy was judged too risky and deferred, so nobody knows what fraction of his liver cells was corrected.
The one editing-efficiency number on offer — 42% across the whole liver — comes from mice carrying his variant. Mice are not people.

The structural point survives all of that. Every therapy before this one was built for a disease; this one, for a genome. Fix the editor, the nanoparticle and the safety assays, and only the guide RNA changes from patient to patient — cost stops scaling with how many people share your variant.

That is what has kept ultra-rare disease uneconomic: when the cost is almost all fixed and very few people carry your exact variant, the numbers never work. The authors reckon hundreds of hepatic metabolic disorders — inherited faults in the liver’s chemistry — could follow. That is a projection, not a result.

Each gridline is ten times the one before, so the bars are further apart than they look. Population and prices from BioPharma Dive; infusion numbers from the Casgevy US label; ~60 treated worldwide is STAT’s February 2026 figure.

Delivery is the bottleneck, not editing

Lipid nanoparticles go to the liver. Not by design — by physics. A blood protein, apolipoprotein E, sticks to the particle’s surface, and liver cells, hepatocytes, carry LDL receptors that bind it and pull the whole particle in.

A 2025 review states the consequence plainly: that preference is a significant barrier to treating anything else. The therapeutic map follows the delivery map.

Liver diseases get in vivo editing — the drug is injected and works inside you. Hence a urea-cycle disorder going first.
Blood diseases get ex vivo editing: blood stem cells cannot be reached where they live in the marrow, so they must be taken out.
Everything else waits.

The trial funnel is that problem in a table: 63 enrolled, 44 infused, six who could never supply enough cells. Not bad luck in patient selection.

Prime editing’s first-in-human data, published in December 2025, treated two patients with chronic granulomatous disease the same way: cells out, edited, back in. Promising, preliminary at an n of two.

How anyone knows where else the editor acted

An editor aimed at one site may also act elsewhere, typically at a sequence resembling the target. Those unintended edits are called off-targets.

The obvious approach — sequence the treated cells’ whole genome and look — fails on arithmetic. Whole-genome sequencing reads each position a handful of times and makes mistakes of its own at a low rate, while an off-target edit may sit in a tiny fraction of the cells. At that level a real rare edit and a sequencing error look the same.

So the field does something else: nominate, then verify.

Nominate. Draw up a shortlist of sites where this guide could plausibly act, using computational prediction and biochemical assays.
Verify. Sequence that shortlist deeply — many reads per site, enough that a rare edit stands clear of the error rate.

For KJ’s editor the team ran two nomination assays and checked 21 high-priority sites, in a cell line and in primary human hepatocytes — liver cells from donors rather than cells grown indefinitely in a dish. They found low-level editing in one place: an intron of ATP7B, a stretch inside the gene spliced out before the protein is made, in the cell line but not the hepatocytes.

Two limits here are structural.

The first: nomination depends on the genome you nominate against. Genomes differ, and in KJ’s case one assay had to run on his father’s DNA, because not enough of his own could be obtained.

The second is simpler and worse: you only find what you thought to look for. Casgevy’s label says it plainly — off-target editing in an individual’s cells cannot be ruled out, because people differ genetically, and its significance is unknown.

The germline line

Everything above is somatic: the edits sit in the treated person’s body cells and die with them. KJ’s authors could not evaluate germline editing, though a different nanoparticle editing drug produced no edits in primate sperm.

The germline line is governance, not technique: editing an embryo would work, which is exactly why it is prohibited. The changes would be inherited by people who cannot consent, and errors would enter the population permanently. US law bars the FDA from even receiving such an application; the field’s professional body calls implanting edited embryos unacceptable.

What to take away

Approved: one CRISPR-based product, two indications, roughly sixty people treated in two years at $2.2 million each. In trials: in vivo base editing in one infant’s liver, and prime editing in harvested stem cells. Press release: everything else, including every claim that this treats ageing — which is not a single-locus disease in a hepatocyte.

The editing chemistry is close to solved. Three other things are not: getting the machine into the right cells, proving it did nothing else, and paying for it. Those, not the enzyme, decide the next decade.