CRISPR is no longer a paper. It is a medicine.

Casgevy is the first CRISPR medicine a haematologist can prescribe, and I still find that sentence slightly outrageous. The international name is exagamglogene autotemcel — a mouthful, so the trade name on the UK and US labels is the one you'll actually see. Vertex Pharmaceuticals and CRISPR Therapeutics spent the 2010s turning Cas9, a bacterial immune protein, into an autologous cell product: a patient's own CD34-positive haematopoietic stem and progenitor cells, electroporated with a Cas9–guide RNA ribonucleoprotein that cuts an erythroid-specific enhancer of BCL11A, then returned after myeloablative busulfan. The licensed indications are sickle-cell disease with recurrent vaso-occlusive crises, and transfusion-dependent β-thalassaemia, first in people twelve and older and later expanded. The molecular trick isn't a corrected β-globin gene. It's a developmental switch. Disrupt the enhancer that keeps foetal haemoglobin off in adult red cells, and γ-globin comes back. Adult haemoglobin is α2β2. Foetal haemoglobin is α2γ2. The second tetramer doesn't sickle. Nature already ran the experiment — hereditary persistence of foetal haemoglobin — and the clinic spent forty years wishing it had a way to do that on purpose. In November 2023 it got one.

In short. Casgevy edits a patient's own blood stem cells so childhood haemoglobin comes back. You use that switch to treat sickle-cell disease and severe β-thalassaemia.

The dates still startle me, because the field had been trained not to believe them. The Medicines and Healthcare products Regulatory Agency authorised Casgevy on 16 November 2023, for sickle-cell disease and transfusion-dependent β-thalassaemia in patients twelve years and older. It was the first CRISPR-based medicine licensed anywhere. The US Food and Drug Administration followed on 8 December 2023 for sickle-cell disease, and on 16 January 2024 for transfusion-dependent β-thalassaemia. In July 2026 the FDA extended both indications to children as young as two. Those sentences aren't a press-office cadence. They're the moment a programmable nuclease left a Nature cover and sat in a pharmacy fridge under a marketing authorisation number. I'm still not casual about that last clause. A scissors that a bacterium used to cut phage DNA is now a named product, with a SmPC, with a fifteen-year follow-up protocol, with a dose expressed as CD34-positive cells per kilogram. CRISPR, on that Thursday in November, stopped being a paper.

In short. Britain licensed Casgevy in November 2023, first in the world. CRISPR left the journal page and got a pharmacy label.

The intellectually interesting move is the one the press still under-describes. You could, in principle, try to correct the Glu6Val in HBB, or add a spare anti-sickling β-chain from a lentivirus, and one licensed neighbour does the second of those. Casgevy does neither. It takes a gene that's doing its job — BCL11A, a zinc-finger repressor that shuts the γ-globin genes around birth — and breaks an enhancer that's used only in the red-cell lineage. BCL11A protein falls in erythroblasts. It doesn't have to fall in B cells or in cortical neurons, which is why the field spent a decade mapping that enhancer rather than knocking out the coding sequence. γ-globin rises. Foetal haemoglobin fills a useful fraction of the red cell. Sickle polymer is diluted below the concentration at which fibres take over the cytosol. In β-thalassaemia the missing β-chain is replaced, in stoichiometric spirit, by a γ-chain the locus still knows how to make. A developmental programme, reopened. That's not a full gene replacement. That distinction is the whole of the molecular sentence, and it's why this piece is a chromatin story as much as a haematology one.

In short. The medicine doesn't fix the broken adult globin gene. It turns a childhood globin gene back on by breaking a red-cell switch.

Haydar Frangoul, at the Sarah Cannon centre in Nashville, treated the first sickle-cell patient with what was then called CTX001 in 2019. Victoria Gray's name left the protocol and entered the public record because she chose to put it there. The first two patients, one with sickle-cell disease and one with transfusion-dependent β-thalassaemia, were reported by Frangoul, Locatelli and colleagues in the New England Journal of Medicine in 2021: CRISPR-Cas9 Gene Editing for Sickle Cell Disease and β-Thalassemia. Both showed durable foetal-haemoglobin induction and clinical quiet. We read that paper the way we read a protocol paper, not a film still. Electroporation of a ribonucleoprotein, not a virus. An enhancer, not a coding sequence. Autologous cells, not a donor. Engraftment. Allelic editing that persisted in the marrow. We had been telling students for a decade that CRISPR would reach the clinic. Then it did, and the remaining work was the unglamorous half: pivotal trials, manufacturing slots, busulfan consent forms, off-target maps, a licence, a price. This piece is that half, and the globin biology that made the half possible.

In short. So the first patients were treated in 2019. A 2021 paper showed the edit held in the marrow and the crises stopped.

Adult haemoglobin, HbA, is a tetramer of two α-chains and two β-chains. The β-chain is the product of HBB on the short arm of chromosome 11, at 11p15.4, in a cluster that also holds the embryonic, foetal and delta genes. Sickle-cell disease is, in the overwhelming majority of patients, a single base substitution in HBB: GAG to GTG, glutamate to valine at position 6 of the mature β-chain, written Glu6Val or E6V. Linus Pauling described the electrophoresis in 1949; Vernon Ingram put the substitution on the peptide in 1956, the first disease to be a protein sequence. Deoxygenated HbS polymerises. The fibre distorts the erythrocyte into the sickle that gave the disease its name, rigidifies the cell, and jams the microvasculature. Vaso-occlusive crisis is that jam, felt as bone and chest and abdomen, scored by patients who already knew the events by heart long before a trial statistician named them. Haemolysis, nitric-oxide scavenging, pulmonary hypertension, stroke, avascular necrosis of the femoral head: the rest of the invoice. A Glu6Val. A polymer. A childhood, then an adulthood, spent negotiating that polymer. I want you to keep that on the table.

In short. So sickle-cell disease is one DNA-letter change in adult haemoglobin. The altered protein clumps when oxygen is low and blocks small blood vessels.

β-thalassaemia is a different lesion on the same gene. Dozens of mutations, some in the coding sequence, many in splice sites and in the promoter, collapse β-chain output. The α-chains, still being made, have nothing to partner. They precipitate. The erythroblast dies in the marrow, or the red cell dies in the periphery, and the patient lives on transfusions. Transfusion-dependent β-thalassaemia, TDT in the protocols, is that life: a unit every few weeks, iron arriving with the heme, chelation as a daily discipline, liver and heart as the organs that keep the score. Splenectomy, extramedullary masses, osteoporosis, the endocrinopathy of iron. A matched sibling allogeneic transplant can cure it, and has done since the 1980s, if you have the sibling and if you accept graft-versus-host disease as the price. Most patients don't have the sibling. The globin cluster is still there. The γ-genes are still there. They were switched off on a developmental clock, not deleted. That's the sentence a repressor story is allowed to start from. Insufficient β-chain is a manufacturing problem the locus already solved once, in foetal life, with a different subunit.

In short. In severe β-thalassaemia the adult chain isn't made in enough quantity. Patients live on transfusions. The childhood chain is still in the genome, switched off.

Foetal haemoglobin, HbF, is α2γ2. The γ-chains come from two genes in the same cluster, HBG2 (Gγ) and HBG1 (Aγ), which differ by one residue. HbF doesn't polymerise with HbS in the way HbS polymerises with itself. A red cell that's a mixture of HbS and HbF sickles later, or not at all, depending on the fraction and on whether the HbF is distributed across all cells or parked in a minority of F-cells. Hereditary persistence of foetal haemoglobin, HPFH, is the natural experiment: deletions that take out HBB and HBD, or point mutations in the γ-promoters, leave adults with 10 to 30 percent HbF and a sickle genotype that behaves like a much milder disease. Clinicians had been staring at those families since the 1960s. Hydroxycarbamide, the one oral disease-modifying drug that actually moved sickle-cell outcomes before gene therapy, works in part by raising HbF, modestly, unevenly, with a myelosuppressive tax. If you could put 30 to 40 percent HbF into most of the red cells, and keep it there, the polymer would lose.

In short. Foetal haemoglobin doesn't sickle. Some families keep it on into adult life and have milder disease. That natural experiment is what Casgevy copies.

The β-globin locus is a teaching diagram that earned its keep. From 5' to 3' on the plus strand: a locus control region of five DNase-hypersensitive sites, then HBE1 (embryonic ε), then HBG2 and HBG1 (the γ-genes), then HBD (δ, the minor adult chain of HbA2), then HBB. Expression follows that order in developmental time. Yolk sac makes ε. Foetal liver makes γ. Adult marrow makes β, with a little δ. The locus control region loops onto whichever promoter the stage has selected. The promoters that aren't selected aren't merely quiet; they are occupied by repressors and packed. A 70-kilobase cluster, two metres of DNA in the nucleus as a whole, a handful of kilobases doing the work that a red cell lives on. Scale isn't a decoration here. An enhancer of BCL11A, sitting in an intron of a gene on chromosome 2, talks to this cluster on chromosome 11. The conversation is a transcription-factor occupancy at a GATA1 motif, and then a repressor sitting on a γ-promoter a chromosome away. Break the motif and the conversation stops. That's the geometry the cut is aimed at.

In short. So the globin genes sit in a row and come on in developmental order. An enhancer on another chromosome tells the childhood genes to go quiet after birth.

The switch itself is a committee, which is why knocking out one coding sequence was never going to be the drug. BCL11A is the repressor the genetics wouldn't shut up about. LRF, also called ZBTB7A, is another. KLF1 sits upstream of both and is itself a globin-promoter factor. MYB, from the HBS1L-MYB interval that GWAS kept recovering, is a third microphone. Chromatin conformation at the locus — the LCR looping to γ in the foetus and to β in the adult — is the geometry those proteins argue inside. Around birth, in the weeks when a neonate's haemoglobin electrophoresis is still a mixture, BCL11A rises in the erythroid lineage and the γ-promoters lose. The committee is why hydroxycarbamide is a blunt instrument and why an erythroid-specific enhancer of one repressor was such a gift. You don't need to win every argument on the committee. You need to take BCL11A down in erythroblasts far enough that γ-globin is allowed to be written again, without taking it down in the B-cell and neuronal compartments that also read the gene. Lineage-specific cis-regulatory DNA is how you get that. The next heading is that DNA.

In short. So several proteins shut childhood globin off at birth. Casgevy only has to quiet one of them, and only in red cells, which is why an enhancer was the target.

BCL11A entered globin biology from a genome-wide association study, which is the honest origin story and a useful warning against the idea that the interesting genes are the ones you already knew. Menzel, Thein, and then Uda, Sankaran, Orkin: common variants in BCL11A associated with HbF level in healthy people and in patients with sickle-cell disease and β-thalassaemia. Sankaran, Xu, Orkin, Science 2008, and the Nature Genetics papers around it, put the protein on the γ-genes as a repressor. Knock it down in adult erythroid cells and HbF rises. The mouse genetics followed, with the usual caveat that a mouse globin cluster isn't a human one and that a conditional allele is the only allele you're allowed to use, because a constitutive Bcl11a knockout is a B-cell and neuronal disaster. Stuart Orkin's laboratory spent the next decade making that sentence into a map. The protein is a C2H2 zinc-finger transcription factor, named first from B-cell lymphoma. It binds γ-globin promoters and the locus control region, and it recruits NuRD. It's doing a job. The job is the switch. The disease isn't a BCL11A mutation. The disease is a β-chain the switch leaves you alone with.

In short. A genome study found BCL11A as a control on foetal haemoglobin. The protein turns the childhood genes off. Sickle-cell disease isn't a mutation in this gene.

You don't knock out BCL11A in a haematopoietic stem cell and call it a medicine. The gene is required in B-cell development; the knockout is a lymphopoiesis phenotype. It's required in the central nervous system; the constitutive mouse is lethal or near it, with neuronal migration defects. A guide RNA aimed at an exon, electroporated into CD34-positive cells that will reconstitute a whole marrow, would edit the gene in every lineage those cells make. Erythroblasts would raise HbF, which is the point. B cells would be a different experiment, and not one a regulator would license. That's why the enhancer hunt was the actual programme. If you can find a cis-element that erythroid transcription factors use and that B cells and neurons do not, you can cut that element, collapse BCL11A only where GATA1 is running the show, and leave the coding sequence intact for the lineages that still need it. Lineage-specific enhancers aren't a catchphrase from a chromatin lecture. They are the difference between a globin drug and an immunodeficiency. If you write 'we CRISPR'd BCL11A' without saying enhancer, you haven't yet started.

In short. You can't delete the BCL11A gene itself in a stem cell. B cells and neurons need it. The red-cell-only enhancer is what makes a medicine possible.

Daniel Bauer, in the Orkin laboratory, mapped DNase-hypersensitive sites in the erythroid-specific intron of BCL11A and found three, at +55, +58 and +62 kilobases from the transcription start, of which the +58 was the adult-erythroid one that mattered. Bauer, Orkin, Science 2013: An erythroid enhancer of BCL11A subject to genetic variation associated with fetal hemoglobin. The title is the result. Common HbF-associated SNPs sit in that enhancer. Deleting it in erythroid cells drops BCL11A and raises γ-globin. Deleting it doesn't recapitulate a B-cell knockout. The element is a few hundred base pairs of composite motif, GATA1 and TAL1 and KLF1 among the binders, sitting in the second intron of a gene that's otherwise broadly expressed. Chromatin packing is the whole of the specificity. In an erythroblast the intron is open and looped. In a B cell it's not. A nuclease that cuts only what it's guided to will, if you guide it to this intron, cut an erythroid switch and leave a B-cell gene standing. That's a rare kind of gift in human genetics, and the field knew it at the time. The remaining question was which motif inside the enhancer to break.

In short. A team found a short stretch of DNA, used only in red cells, that turns BCL11A on in that lineage. Break that stretch and the repressor falls in erythroblasts only.

Matthew Canver, Bauer, Orkin, Nature 2015, tiled the enhancer with CRISPR-Cas9 and asked which nucleotides, when scarred, gave the HbF phenotype. The answer was a GATA1-binding motif inside the +58 element. Destroy that motif and GATA1 can't sit. Without GATA1 the enhancer doesn't work. Without the enhancer, erythroid BCL11A transcription falls. Without the repressor, γ-globin is written. The cascade is ugly when you write it as a protocol and beautiful when you write it as a logic gate. Casgevy's guide RNA is aimed at that motif. The Cas9 protein, delivered as protein, not as a gene, makes a double-strand break. Non-homologous end joining, the default repair in a resting stem cell, inserts or deletes a few bases. The motif is gone in a large fraction of alleles. There is no donor DNA. There is no homology-directed repair to wait on. There is no viral integration. The product specification is an allelic-editing percentage, a CD34-positive cell dose, a viability, a sterility, a karyotype. The GATA1 motif is the entire targeting rationale. Everything downstream is haematology.

In short. Here, a 2015 paper found the exact GATA1 spelling inside that enhancer. Casgevy's guide RNA aims at that spelling and lets the cell's own repair scramble it.

Hereditary persistence of foetal haemoglobin remains the control I want to keep in the room. Some HPFH alleles are large deletions that remove HBB and HBD and leave the γ-genes under the LCR; those people make HbF because there is no adult gene left to compete. Some are point mutations in the γ-promoters that destroy a repressor motif or create an activator motif; those people make HbF because the promoter no longer listens. Some are BCL11A haploinsufficiency, rare and informative. Casgevy is closest, in spirit, to the last of those, done only in erythroblasts, and to the promoter-mutant class, in the sense that a repressor is being told to stand down. It's not a deletion of HBB. Patients still make HbS, or still fail to make β, depending on the disease. They make HbF on top. The clinical bet, which the natural experiment had already placed, is that 30 to 40 percent HbF, pancellular enough, is a different disease. Hydroxycarbamide never reliably got you there. An enhancer cut, in a stem cell that will feed the marrow for years, might. The trials were how you find out whether 'might' was a medicine.

In short. So: families who naturally keep foetal haemoglobin on already showed that 30 to 40 percent is enough to change the disease. The trials asked whether an enhancer cut could do the same.

The cut is a double-strand break, and the enzyme is Streptococcus pyogenes Cas9, the same protein Doudna and Charpentier described as a programmable scissors in 2012. In the manufacturing suite it arrives as a ribonucleoprotein: recombinant Cas9 protein, complexed in vitro with a chemically synthesised single-guide RNA, electroporated into the CD34-positive apheresis product. No Cas9 gene is written into the genome. The protein cuts, and is then diluted out as the cells sit and as they later divide. The guide RNA is twenty bases plus a scaffold, chosen to place the break in the GATA1 motif of the +58 BCL11A enhancer, with a neighbouring NGG protospacer-adjacent motif that Cas9 requires. Off-target mapping — GUIDE-seq, CIRCLE-seq, CHANGE-seq, one-off targeted amplicon sequencing of nominated sites — is the adult work that a 2012 paper did not have to do and that a 2023 licence did. Yen, Altshuler, the New England Journal of Medicine in 2026: Specificity of CRISPR-Cas9 editing in exagamglogene autotemcel. The on-target scar is the drug. The off-target map is the reason you're allowed to say so.

In short. Look: cas9 protein, not a Cas9 gene, is put into the cells with a guide RNA and makes one cut at the red-cell switch. The protein then disappears.

Repair is non-homologous end joining in the great majority of alleles, because a haematopoietic stem cell sitting in a bag isn't in S phase waiting to do homology-directed repair, and because no donor template is supplied. Insertions and deletions of one to a few tens of bases destroy the GATA1 motif. Some alleles take a larger deletion. The product is therefore a population, not a clone: a distribution of scars at one address, an allelic-editing fraction that manufacturing has to hit, typically the majority of BCL11A enhancer alleles in the CD34-positive dose. When those cells engraft, the marrow is a mosaic of edited and unedited stem cells. Erythroblasts that came from edited stem cells make less BCL11A and more γ-globin. The red cells they produce carry HbF. If the edited fraction of long-term stem cells is high enough, and if those cells aren't outcompeted, the HbF stays. Persistence of allelic editing in marrow and blood, years out, is the observation that says you edited the stem cell and not only a progenitor. The 2024 and 2025 follow-up posters have been about that persistence. It's the least glamorous successful result in the file, and the one that matters.

In short. So the cell stitches the cut without a template, scrambling the switch. If enough true stem cells were edited, foetal haemoglobin stays up for years.

What rises, chemically, is γ-globin mRNA and then HbF as a fraction of total haemoglobin. In the pivotal sickle-cell cohort, mean HbF sat in the neighbourhood of 40 percent of total haemoglobin, with total haemoglobin itself climbing into a near-normal range as haemolysis quieted. F-cells — red cells with detectable HbF — approached pancellular, which is the distribution you want, because a minority of F-cells in a sea of HbS cells is how hydroxycarbamide sometimes fails. In transfusion-dependent β-thalassaemia the same HbF rise replaces the missing β-chain well enough that most patients stop transfusions, with a total haemoglobin that a haematologist would call transfusion-independent rather than normal-in-every-red-cell. The numbers are in Frangoul, New England Journal of Medicine, 2024, 390: 1649–1662, and in Locatelli, the companion paper, 390: 1663–1674. We keep the trial names and the journal in the paragraph because the names are how you find the data. A press release that says 'crises stopped' without an HbF fraction and an F-cell plot is a press release. The biochemistry is a tetramer count. I want you to keep that on the table.

In short. So foetal haemoglobin rises to about two-fifths of the total in treated sickle-cell patients, in most of their red cells, which is the level that stops the polymer winning.

Glu6Val is still in the genome. That sentence is the one a gene-replacement cartoon can't survive contact with. The patient's HBB alleles are the alleles they were born with. In sickle-cell disease they still encode HbS. In β-thalassaemia they still fail to encode enough β. What has changed is the rest of the haemoglobin mix in the same cell. A polymer needs a concentration. Dilute HbS with HbF, and the delay time to nucleation lengthens past the time the cell spends in the capillary. Eaton and Hofrichter spent careers on that delay time; the clinic is now a comment on their kinetics. Residual sickle haemoglobin is why this is not, even at its best, a molecular eraser of the disease genotype. It's a phenotypic rescue, durable if the stem cells hold, incomplete if they do not, and always a mix. Patients who were told they had been 'gene edited' sometimes hear 'the sickle gene is gone'. It's not gone. The childhood gene is back. Those are different claims, and only the second one is true. We'll keep saying so, because the first one is the sentence that travels.

In short. So the sickle mutation is still in the DNA. Childhood haemoglobin is added on top, which is enough to stop most crises if it stays high.

The manufacturing story starts with mobilisation, because you can't edit a stem cell you haven't collected. In sickle-cell disease, G-CSF is largely avoided; the molecule can provoke crises. Plerixafor, a CXCR4 antagonist, is the mobilising agent that pulls CD34-positive cells into the blood so that apheresis can take them. In β-thalassaemia, G-CSF plus plerixafor is the more common pair. Apheresis is a day, or several days, on a machine that patients with these diseases already know from red-cell exchange. The product is a CD34-enriched bag, shipped to a specialised manufacturing suite, where the cells are cultured briefly, electroporated with the Cas9 ribonucleoprotein, recovered, formulated, and cryopreserved. Release testing is a list: identity, allelic editing at the BCL11A enhancer, off-target nominated sites, viability, sterility, endotoxin, mycoplasma, potency as HbF in an erythroid differentiation assay, cell dose. The patient, meanwhile, isn't yet ready. The bag in the freezer isn't a medicine until the marrow has been emptied to make room.

In short. Stem cells are collected from the blood, sent to a specialised lab, cut with Cas9, frozen, and tested. The patient isn't treated until that bag exists.

Electroporation is the delivery, and it's worth naming because 'CRISPR'd cells' hides a machine. A square-wave or exponential pulse opens transient pores in the plasma membrane. A protein–RNA complex of a few hundred kilodaltons goes in. The pores close. Some cells die, which is why the input dose has to be larger than the output dose, and why a manufacturing run can fail on viability. There is no lentivirus. There is no AAV. There is no Cas9 transcription unit to silence or to integrate. The non-viral character is a regulatory object as much as a scientific one: insertional mutagenesis, the shadow that followed first-generation gene therapy, isn't the risk class here. The risk class is a nuclease — off-target cutting, large deletions, rearrangements, p53 activation in the cells that saw a break. Different shadow, different assays. MaxCyte-style electroporators, GMP Cas9 from a named supplier, HPLC-characterised guide RNA: those are the objects in the batch record. If you can't name the pulse and the protein, you haven't yet described the product.

In short. Here, an electric pulse opens the cell membrane so Cas9 protein can enter. No virus is used. Some cells die in the pulse, which is why the starting dose has to be large.

Dose on the label is a CD34-positive cell number per kilogram, in a specified range, with a minimum below which you don't infuse. The SmPC describes a dispersion for infusion at 4–13 × 10^6 cells per millilitre; the clinical dose is on the order of millions of CD34-positive cells per kilogram, a number haematopoietic-transplant teams already think in. Neutrophil engraftment and platelet engraftment are the first clinical events after infusion, and in the pivotal studies they occurred in every patient who received product, on a timescale of weeks that will be familiar to anyone who has stood on a transplant ward. Failure to engraft would be the catastrophic manufacturing-plus-conditioning outcome; it's also the event the trials were watching for and did not see. A specialised centre isn't a luxury around this product. It's a transplant unit that can collect, condition, infuse, and manage two to four weeks of aplasia, mucositis, infection risk, and, in a subset, veno-occlusive disease of the liver. Casgevy is a gene-edited autograft. The autograft half of that sentence is the half that decides where it can be given.

In short. So the dose is millions of stem cells per kilogram, given like a bone-marrow transplant. Every treated trial patient engrafted. This only happens in a specialist centre.

Myeloablative busulfan is the conditioning, and it's the part of the protocol a gene-editing cartoon never wants to discuss. You empty the marrow so that the edited cells have a niche. Busulfan, an alkylating agent older than CRISPR by several decades, is how most of these protocols empty it. The tax is the tax of myeloablation: pancytopenia, mucositis, infection, infertility that must be counselled, veno-occlusive disease of the liver that can kill. In the 2026 paediatric New England Journal of Medicine paper, Frangoul and colleagues reported two children with transfusion-dependent β-thalassaemia who had severe veno-occlusive liver disease assessed as busulfan-related, one of whom died. That sentence belongs in any essay that's about to call this a beautiful rewrite. The edit is beautiful. The conditioning is chemotherapy. A licensed CRISPR medicine isn't a licence to skip the consent form. Reduced-intensity and non-myeloablative conditioning for edited autografts is a live research programme, because everyone in the field can see the tax. It's not yet the licensed programme. Until it is, Casgevy's safety profile is, in the FDA's own phrasing, generally consistent with myeloablative busulfan and autologous transplantation. Believe that phrasing. It's doing a lot of work.

In short. So chemotherapy is used to empty the marrow so the edited cells can settle. That step, not the CRISPR cut, is what makes the treatment brutal.

Infusion is a bag, thawed, given intravenously, after the busulfan has done its work and after a washout that the protocol specifies. The patient then waits, in protective isolation or its modern equivalent, for neutrophils to appear. Platelets follow. Red cells, in a sickle-cell patient, have been supported by transfusion through the process; in a thalassaemia patient they always were. HbF doesn't peak on day one. The edited stem cells have to engraft, make progenitors, make erythroblasts, make reticulocytes. The haemoglobin electrophoresis shifts over months. Crises, if they are going to stop, stop after that shift, which is why the primary endpoints were scored as twelve consecutive months free of severe vaso-occlusive crises, or twelve consecutive months of transfusion independence, starting after a defined window. A one-time treatment with a delayed readout. Nobody swallowed a pill. They got a transplant of their own edited cells, and then they waited to see whether the childhood haemoglobin would hold. In the large majority of evaluable patients, it held.

In short. So the edited cells are dripped back in after chemotherapy. Foetal haemoglobin rises over months, and the trial scored a full year without crises or transfusions.

CLIMB-121, later written as CLIMB SCD-121, was the pivotal sickle-cell study: phase 1/2/3, single-group, open-label, patients twelve to thirty-five years of age with at least two severe vaso-occlusive crises in each of the two years before screening. CD34-positive cells were edited, busulfan was given, product was infused. The primary endpoint was freedom from severe vaso-occlusive crises for at least twelve consecutive months, VF12 in the statistical analysis plan. Frangoul, New England Journal of Medicine, 24 April 2024, 390: 1649–1662: forty-four patients received exa-cel, median follow-up 19.3 months. Neutrophils and platelets engrafted in each of them. Of thirty patients with sufficient follow-up to be evaluated, twenty-nine — 97 percent, 95 percent confidence interval 83 to 100 — were free of vaso-occlusive crises for at least twelve consecutive months, and all thirty were free of hospitalisation for those crises over the same window. No cancers. The safety profile was the busulfan-plus-autograft profile. Those are the clinical events patients already knew by heart, not surrogate endpoints dreamed up for a press release. We keep the trial name in the paragraph because the name is how you find the data. I want you to keep that on the table.

In short. In short, the main sickle-cell trial, 29 of 30 patients with enough follow-up went a year without a severe crisis. All 30 went a year without a crisis admission.

The US label, at first approval, quoted a closely related snapshot: 29 of 31, 93.5 percent, for VF12 in the efficacy-evaluable set the FDA preferred. Later data cuts, including the ASH 2024 update and the CLIMB-131 long-term follow-up, have sat in the same neighbourhood: 39 of 42, 93 percent, free of vaso-occlusive crises for twelve consecutive months, with mean duration of that freedom past thirty months in the patients who had the time. Mean HbF held near 40 percent. Allelic editing in marrow and blood was stable, which is the stem-cell sentence again. A few patients haven't met VF12 and still show substantial clinical benefit; a trialist won't let you collapse those people into a failure bin without reading the narratives. Open-label, single-arm, historical-crisis-rate as the comparator: those are the design caveats a careful reader is required to say. Randomising a patient with recurrent severe crises to busulfan and an unedited autograft isn't a trial anyone was going to run. The fair reading is that the effect size, against that person's own previous two years, is large, and that the follow-up that will matter is the fifteen-year one still running.

In short. Look: later snapshots stay near 93 percent crisis-free at a year, with foetal haemoglobin holding. Follow-up is still measured in years, not in decades.

CLIMB-111, CLIMB THAL-111, is the transfusion-dependent β-thalassaemia companion, Franco Locatelli the corresponding voice, New England Journal of Medicine 2024, 390: 1663–1674. Same product, same enhancer, same busulfan, a different clinical event: the unit. The primary endpoint was transfusion independence for at least twelve consecutive months, with a weighted average haemoglobin of at least 9 g/dL, TI12. In the label-adjacent snapshot, 32 of 35, 91.4 percent, met that endpoint in the adolescent-and-adult efficacy set. Locatelli's paper, in the evaluable-at-the-time cut, was in the mid-to-high nineties. Patients who achieved independence have remained independent in the follow-up that exists, with median duration of independence past four years in the UK SmPC's later cut. Iron chelation, the daily discipline, is a separate conversation that starts to end when the units stop. The disease isn't a crisis; it's a logistics of blood and iron. Stopping that logistics is the event. Same caveats as 121: single-arm, open-label, a comparison against the patient's own transfusion diary. Same honest reading: the diary stopped, in the large majority, after one autograft of edited cells.

In short. In short, the main thalassaemia trial, about nine in ten patients stopped transfusions for at least a year, with haemoglobin high enough to live on.

The 2021 New England Journal paper, two patients, is still worth reading even after the pivotal cohorts, because it's the protocol in human form. One woman with transfusion-dependent β-thalassaemia, one woman with sickle-cell disease. Both had HbF induction that looked, on an electrophoresis, like a mild HPFH. Both had allelic editing that persisted. The sickle-cell patient stopped having crises. The thalassaemia patient stopped having units. CTX001 was the code. Vertex and CRISPR Therapeutics were the sponsors. Frangoul, Corbacioglu, Grupp, Locatelli: the investigator list was a transplant list, not a gene-therapy-startup list, which was information. A CRISPR medicine was always going to be given by people who already knew how to get a patient through aplasia. The 2021 paper is also the document that made Victoria Gray's clinical course discussable in a journal rather than only in a broadcast. We're not above being moved by a single-patient electrophoresis. We're also not going to pretend two patients are a licence. The licence came from 121 and 111, and from a regulator who had been shown the off-target map.

In short. So the first two patients, published in 2021, already showed the pattern: foetal haemoglobin up, crises or transfusions down, the edit still there in the marrow.

CLIMB-131 is the fifteen-year follow-up that every person who talks about 'cure' should have to name. Patients who complete 111 or 121 are rolled over. The UK SmPC already quotes follow-up measured in additional years on top of the two-year parent studies. What 131 is for: durability of HbF, durability of allelic editing, myeloid neoplasms, solid tumours, unexpected cytopenias, clonal haematopoiesis that a nuclease might have given a push. No cancers in the 2024 New England Journal cuts. That's not the same sentence as no cancers in fifteen years. The 2026 paediatric paper — CLIMB THAL-141 and CLIMB SCD-151, children, New England Journal of Medicine, Frangoul again — is the other document I want us to hold. Fifteen children with transfusion-dependent β-thalassaemia and eleven with sickle-cell disease received exa-cel. Those with enough follow-up met the same clinical events as the adolescents. Two children with thalassaemia had severe busulfan-related veno-occlusive liver disease; one died. Efficacy in a child isn't a smaller adult. Conditioning in a child isn't a smaller busulfan. The FDA's July 2026 expansion to age two sits on those data, with the death in the file. Both things are true.

In short. Here, a fifteen-year follow-up is still running. A 2026 children's study worked, and also reported a busulfan-related death. Efficacy and that tax have to be held together.

The MHRA authorisation of 16 November 2023 is the date I won't let a CRISPR story skip. It was a conditional marketing authorisation in Great Britain, for patients twelve years and older with sickle-cell disease and recurrent vaso-occlusive crises, or with transfusion-dependent β-thalassaemia, at specialised centres. First in the world. The European Medicines Agency followed. The FDA's 8 December 2023 decision for sickle-cell disease made Casgevy the first CRISPR medicine available in the United States; Bluebird's lovotibeglogene autotemcel, Lyfgenia, a lentiviral anti-sickling β-globin addition, was licensed the same week, which is a fact the CRISPR-first headlines compressed. January 2024 added transfusion-dependent β-thalassaemia in the US. July 2026 added children down to two years, the Commissioner's National Priority Voucher sitting on that file. A pharmacy label, a J-code, an SmPC that describes a genetically modified autologous CD34-positive population edited at the erythroid-specific enhancer of BCL11A. We've copied that phrasing because it's the regulator's molecular sentence, and it's a better sentence than most of the coverage.

In short. So Britain licensed Casgevy first, on 16 November 2023. The US followed in December. A second, non-CRISPR gene therapy for sickle-cell disease was licensed the same week.

Specialised centres are the actual distribution, and they are few. A hospital that can mobilise, apherese, send a product across a border or across a country, admit for busulfan, manage aplasia, and report into a follow-up registry isn't every hospital that treats sickle-cell disease. Manufacturing slots are a rate. Vein-to-vein time is measured in months. Some collection runs fail to yield enough CD34-positive cells, particularly in patients whose marrow has had a life. Some manufacturing runs fail release. The licensed product is a bespoke autograft, not a vial in a wholesaler's warehouse. NICE, in the UK, had to decide whether the National Health Service would pay, and for whom, which is a different question from whether the electrophoresis moved. Saudi Arabia, Italy, Germany, the list of reimbursing systems is a live document. A CRISPR medicine that exists on a label and a CRISPR medicine that a given patient can be referred into are two objects. We'll not pretend the first is the second. The first is still the scientific event. The second is the health-system event, and it's slower.

In short. Only specialist transplant centres can give Casgevy, and manufacturing slots are limited. A licence isn't the same thing as a given patient being treated next month.

Price, at US launch, sat at 2.2 million dollars per patient, a predictable output of a bespoke autologous manufacturing process with a tiny denominator. The UK price, under the schemes NICE and NHS England actually use, is a confidential discount on a list that was never going to be the American one. Outcomes-based rebates, instalments, 'pay for a year without crises': the contracting language is trying to make a one-time cost look like the chronic-care cost it replaces. A unit of blood, a hospital night for a crisis, a lifetime of chelation, a stroke: those have prices too, and they are paid by people who don't get a press cycle when they are paid. None of that arithmetic licenses the 2.2 million as a natural fact of physics. It licenses it as a fact of how this product is made in 2026. Allogeneic edited cells, in vivo editors that skip the autograft, a cheaper electroporation at a regional centre: those are the industrial questions that will decide whether Casgevy is a one-off miracle for a few hundred people or a template. The molecular template already works. The industrial template is the remaining invention.

In short. So the US list price was 2.2 million dollars, because each dose is made from one patient's cells. Whether that can become a common treatment is an industry problem, not a biology one.

Casgevy is the first licensed CRISPR medicine. It's not the first licensed gene therapy, and collapsing those two firsts is how a timeline gets cheap. Alipogene tiparvovec was withdrawn. Strimvelis, an ADA-SCID autograft, has been given to a handful of children. Voretigene neparvovec, Luxturna, is an AAV in an eye. Onasemnogene abeparvovec, Zolgensma, is an AAV in a liver, for SMA, a product whose price taught the world the 2-million-dollar sentence before Casgevy reused it. Betibeglogene autotemcel, Zynteglo, is a lentiviral β-globin addition for thalassaemia, Bluebird again, a neighbour of Lyfgenia. The 2023 week in December put two sickle-cell genetic medicines on the US label: one a lentivirus writing an anti-sickling β-chain, one a nuclease breaking a BCL11A enhancer. Different tools, same autograft chassis, same busulfan. CRISPR's first is real. It's a first among nucleases, not a first among genetic medicines. If you can't name Luxturna and Zolgensma in this paragraph is selling a platform, not a history. We would rather have the history.

In short. So Casgevy is the first licensed CRISPR medicine, not the first gene therapy. Eye, spinal and other gene medicines were already on the market, made with viruses.

Infertility counselling isn't optional, and it's not a footnote. Myeloablative busulfan, at the doses these protocols use, is expected to destroy gonadal function in a large fraction of patients. Sperm banking, oocyte cryopreservation, a conversation with a twelve-year-old and their parents about a future pregnancy that the disease already complicated: that's the week before admission, not a pamphlet in the discharge bundle. Veno-occlusive disease, mucositis, bloodstream infection, the weeks of platelet transfusion, the central line: a transplant ward, in other words. Secondary malignancy is the long shadow of alkylators, independent of whether a nuclease was in the bag. The 2026 paediatric death from busulfan-related veno-occlusive disease is the sentence that belongs in the same paragraph as the 97 percent. A beautiful rewrite that you have to survive chemotherapy to receive is a beautiful rewrite with a denominator. Reduced-intensity conditioning, if it can be made to let edited cells engraft without the full tax, would change the denominator. Until then, eligible means eligible for a myeloablative autograft, which is a smaller set than eligible for a CRISPR medicine in the abstract.

In short. So busulfan often causes infertility and can cause lethal liver injury. Eligible for Casgevy means eligible for a full-intensity stem-cell transplant, which is a smaller group.

Off-target cutting is the nuclease's own shadow, and it's the one the 2012 papers were honest about and the 2023 licence had to quantify. Cas9 will cut some genomic sites that aren't the BCL11A enhancer, at rates that depend on the guide, the cell type, and the dose of ribonucleoprotein. Nominated off-target sites are sequenced to a depth a trial can stand on. Large deletions and translocations at the on-target site, and between the on-target site and an off-target, are the rearrangements a karyotype and a specialised assay have to look for. p53-mediated loss of cells that saw too many breaks is a selection pressure in the bag; what engrafts may be the cells that tolerated the cut, which is usually what you wanted and occasionally a clone you will wish you had marked. Yen and colleagues, 2026, put the exa-cel specificity map into the New England Journal of Medicine. The on-target scar is consistent and the off-target map is clean enough to license, not empty. Nucleases have off-target biology. That's why you measure it.

In short. Cas9 can cut the wrong place, rarely. The licensed product was mapped for those cuts. Clean enough to license isn't the same as a scissors that never slips.

Clonal haematopoiesis is the adult pharmacovigilance question that a fifteen-year protocol exists to answer. A haematopoietic stem cell that took a cut, repaired it, and engrafted will found a clone. If a rare off-target or a rare rearrangement confers a growth advantage, that clone can expand. Myeloid neoplasms after autologous transplant are an old problem; myeloid neoplasms after a nuclease-edited autograft would be a new chapter of it. Insertional mutagenesis was the lentiviral chapter, and Lyfgenia's US label carries a boxed warning about haematologic malignancy that Casgevy's does not, which is information and not a verdict on decade-scale risk. Absence of cancers in a 19-month median follow-up is a real observation. It's not a lifetime. A clean 'we edited the genome' pitch without those caveats is selling a press release. A claim that the trials did not work is 2012 talking. We would rather have the caveats and the working enzyme, and a registry that still writes to us in 2038.

In short. Those edited stem cells found clones in the marrow. Whether a rare wrong cut ever starts a leukaemia is why patients are followed for fifteen years.

Delivery into other tissues is a different engineering problem, which is why a licensed ex-vivo haematopoietic product isn't a licence to edit a hepatocyte, a myocyte, a neuron or a photoreceptor with the same sentence. Casgevy works because you can take the target cell out, because you can electroporate it, because you can freeze it, because you can empty the niche, and because a blood cell is a renewable tissue whose product — the red cell — has a hundred-day life and no genome of its own to worry about once the nucleus has been extruded. Liver, muscle, brain and eye don't offer that workflow. In vivo base editors and prime editors, lipid nanoparticles, AAV capsids, the next decade's delivery papers, are the answer to those organs, and they aren't Casgevy. The cousinhood of the enzyme is real. The cousinhood of the product is not. A Dallas mammoth programme and a Boston sickle-cell ward are using relatives of the same nuclease. We find that cousinhood more interesting than most of the headlines. We also find it a reason to keep the headlines from treating every CRISPR sentence as the same medicine.

In short. Casgevy works because blood stem cells can be taken out, edited, and put back. Liver, muscle and brain can't be treated that way, and need a different delivery problem solved.

What 'approved' still costs, then, is a list, and the list is the adult one. Myeloablation. Infertility. A specialised centre. A manufacturing slot measured in months. A price that a health system has to invent a contract for. An off-target map that's clean enough, not empty. A fifteen-year registry. A paediatric death in the conditioning file. A residual Glu6Val that did not go anywhere. HbF at 40 percent, not at 100. A product that's one patient's cells, unusable in the next patient. None of those costs make the 97 percent untrue. They make it a 97 percent in a defined population who got through a transplant, at a defined list of hospitals, with a defined follow-up. Translational science, when you're in it, looks like this: a first with asterisks. The asterisks are the work. We would rather have the asterisks than the TED talk that skipped them. The licence is real. The fridge is real. The busulfan is real. Hold all three, or you're not yet talking about the medicine.

In short. A licence doesn't erase chemotherapy, cost, rare wrong cuts, or the need for long follow-up. The trial result is real. So are those costs.

Lovotibeglogene autotemcel, Lyfgenia, is the cousin that's not a nuclease, and it belongs in this piece so that CRISPR doesn't get to own a week it shared. Bluebird's product is a lentivirus writing a modified β-globin, HbA with a T87Q substitution, into the same autologous CD34-positive cells, after the same flavour of myeloablation. The anti-sickling chain is an addition, not a repressor knockdown. The US licence arrived in December 2023, the same week as Casgevy's sickle-cell licence. The label carries a boxed warning about haematologic malignancy, a lentiviral shadow Casgevy wasn't asked to carry. Comparative tables — 88 versus 93 percent on neighbouring but not identical crisis endpoints, different follow-ups, different manufacturing failures — are already in the haematology reviews. Two genetic medicines, one week, one disease, two mechanisms. A patient and a centre now have a choice that 2018 did not offer. That choice is the industrial fact. The molecular fact is that foetal-haemoglobin induction and anti-sickling β-addition both beat a polymer, and that neither erases the allele the patient was born with.

In short. Here, a second sickle-cell gene therapy, licensed the same week, adds a modified adult chain with a virus instead of turning childhood haemoglobin back on.

NTLA-2001, Intellia's in-vivo CRISPR for transthyretin amyloidosis, is the other cousin a nuclease story has to name, because it's the one you inject. Lipid nanoparticles, carrying Cas9 mRNA and a guide, taken up by hepatocytes, cutting TTR in the liver of a living person, a 2021 New England Journal of Medicine paper by Gillmore and colleagues that the field passed around like gossip. No apheresis. No busulfan. No autograft. A different risk class: you can't recall the nuclease from a liver, you can't karyotype the organ before you put it back, and the off-target conversation is now an in-vivo conversation. ATTR is a secreted-protein disease, which is why a hepatocyte edit can be a whole-body treatment. Sickle-cell disease is a stem-cell-and-red-cell disease, which is why an in-vivo haematopoietic editor is a harder delivery problem and not yet a licensed one. Casgevy and NTLA-2001 share an enzyme family and almost no product architecture. Saying 'CRISPR is in the clinic' without saying ex vivo versus in vivo is saying a catchphrase. The catchphrase is true. The architecture is the interesting bit.

In short. A different CRISPR medicine is injected into the bloodstream and edits the liver in place. Casgevy isn't that. It's cells out, cut, cells back.

Base editors and prime editors are the pencils that David Liu's laboratory spent a decade making from the scissors, and they are the tools a 2030 sickle-cell protocol might actually prefer. A cytosine or adenine base editor could, in principle, correct Glu6Val, or destroy a GATA1 motif without a double-strand break. A prime editor could write a longer specified change from an RNA template. The 2019 Nature prime-editing paper read like a joke the field had been waiting to tell. The 2020s have been delivery, and doing that in a CD34-positive cell at a frequency a manufacturing suite can release. Beam Therapeutics and others have been on the HbF-induction and the direct-correction paths. None of that's Casgevy. Casgevy is a scissors and an NHEJ scar at an enhancer, licensed, sitting in a fridge. The neighbouring essay in this section is the pencil family. The scissors got there first because a break at a motif is a simpler chemistry than a correction of a codon, and because simpler chemistry is what you can take through a pivotal trial in a single decade.

In short. Newer tools can change a single DNA letter without cutting both strands. They may one day correct the sickle mutation itself. Casgevy is the older scissors, and it's the one with a licence.

A Dallas freezer and a London ward are using cousins of the same enzyme, and that sentence is why this piece sits in a frontier section that also holds a mammoth. Colossal's elephant fibroblasts, the dire-wolf edits, the woolly mouse: programmable nucleases writing a palaeogenomic parts list into a living relative. Casgevy's CD34-positive cells: a programmable nuclease writing a scar into an enhancer a developmental biologist had mapped. The ethical payloads couldn't be more different. The enzyme doesn't care. That's the unsettling elegance of a platform technology, and we're living inside it. What the two programmes share is the papers that would have been science fiction when most of us learned haemoglobin. What they don't share is a regulator, a consent form, a disease, or a right to the other programme's headlines. We wish more coverage would say that second sentence. The mammoth calf, if it arrives around 2028, won't have been licensed by the MHRA. Victoria Gray's marrow was. Both can be true at once. Only one of them is a medicine.

In short. The same family of DNA-cutting enzymes is being used to edit elephants and to treat sickle-cell disease. Only the second is a licensed medicine. The enzyme doesn't care. We should.

The node is conserved in the only way that matters here: every mammal that makes a foetal haemoglobin has to switch it off, and a zinc-finger repressor sitting on a γ-promoter is how humans do that. BCL11A has orthologues. The globin cluster has a history you can read in goats and in mice, with the usual warning that a mouse isn't a patient. What isn't conserved is the licence. A yeast paper on Cas9, a 2012 scissors, a 2013 enhancer, a 2015 motif, a 2019 first patient, a 2023 MHRA date: that's a mammalian translational timeline, and it's short by the standards of the diseases it treats. Conservation isn't a licence to treat a mouse HbF induction as a human protocol. It's a licence to take the enhancer seriously enough to electroporate it, in the organism you have, with the off-target map the drains of a genome require. The popular story got loud because a scissors reached a fridge. The work got hard for the same reason. A developmental switch is a simple idea. Making it a batch record is not.

In short. So mammals already knew how to switch childhood haemoglobin off. Casgevy is that switch, run backwards, through a decade of mapping, trials and manufacturing.

The public papers are the reading list, and they are short enough to actually read. Sankaran, Orkin, Science 2008, BCL11A as the HbF repressor. Bauer, Orkin, Science 2013, the erythroid enhancer. Canver, Bauer, Orkin, Nature 2015, the GATA1 motif inside it. Frangoul, New England Journal of Medicine 2021, two patients, the methods in human form. Frangoul, 2024, 390: 1649–1662, CLIMB-121. Locatelli, 2024, 390: 1663–1674, CLIMB-111. Gillmore, 2021, if you want the in-vivo cousin. Yen, 2026, if you want the specificity map. The 16 November 2023 MHRA authorisation, the 8 December 2023 FDA letter, the July 2026 paediatric expansion. Eaton on delay time, if you want to know why 40 percent HbF wins. Ingram, 1956, if you want to remember that this disease was a protein sequence before it was a CRISPR target. That's a fortnight of evenings, not a guru. The headlines will still be there when you come back, and they will look smaller, and the SmPC will look larger.

In short. Here, a short stack of papers covers the repressor, the enhancer, the motif, the first patients, the two pivotal trials and the licence dates. Read those before any headline.

What you should leave with is a map, not a catchphrase. Casgevy is exagamglogene autotemcel, an autologous CD34-positive product edited at a GATA1 motif in the +58 erythroid enhancer of BCL11A. The cut is Cas9 ribonucleoprotein, electroporated, repaired by non-homologous end joining. BCL11A falls in erythroblasts. γ-globin rises. HbF at about 40 percent, pancellular enough, dilutes HbS or replaces missing β. CLIMB-121: 29 of 30 evaluable sickle-cell patients free of severe vaso-occlusive crises for twelve months. CLIMB-111: most transfusion-dependent β-thalassaemia patients transfusion-independent for twelve months. MHRA, 16 November 2023. Conditioning is myeloablative busulfan, with a paediatric death in the 2026 file. Price is a manufacturing fact. Off-targets were mapped. Follow-up is fifteen years and not finished. The Glu6Val is still there. In vivo editors are a different product. Lyfgenia is a different mechanism licensed the same week. A Dallas freezer is a different ethical payload. If your sentence can't hold that list, it's not yet a sentence about this medicine. If it can, you're done, and you can go and read the electrophoresis.

In short. So leave with the map: cells out, a red-cell switch cut, childhood haemoglobin back, crises or transfusions mostly gone, chemotherapy as the tax, follow-up still running.

A licensed CRISPR medicine exists. It rewrites a patient's stem cells. It doesn't mean every genome is now a text editor you can open at home. The gap between Casgevy and a Saturday-night edit is the entire delivery-and-safety stack: apheresis, a GMP suite, an off-target map, a busulfan consent, a transplant ward, a registry, a regulator who can say no. That stack is why we still have jobs, and why the 2012 papers were a beginning rather than a kit. Anyone offering you a guide RNA and a promise, without that stack, is offering you a fantasy the MHRA doesn't recognise. A claim that the stack means CRISPR never arrived is ignoring a label that a pharmacist can already read. We'll stay with the label. Exagamglogene autotemcel. An erythroid-specific enhancer of BCL11A. A childhood haemoglobin, on purpose, in an adult red cell. The sci-fi crossed the pharmacy counter. The counter is still a counter. The patient still had to be eligible for a transplant. Those three sentences are the whole essay, and we're not going to tidy them into one.

In short. A licensed CRISPR medicine exists, and it's not a home kit. The gap is manufacturing, chemotherapy, mapping of wrong cuts, and a regulator who can refuse.

The remaining questions are the ones a licence doesn't close. Can reduced-intensity conditioning let edited cells engraft without infertility and veno-occlusive disease. Can an in-vivo haematopoietic editor skip the autograft. Can a base editor correct Glu6Val at a manufacturing frequency, and would you even want that if 40 percent HbF already wins. What does a myeloid neoplasm look like, if one arrives, in a nuclease-edited marrow versus an alkylator-only marrow. Who, globally, actually receives the product: the epidemiology of sickle-cell disease isn't the geography of specialised centres. The largest burden is in sub-Saharan Africa and in India, where a 2.2-million-dollar autograft isn't a health-system object. A CRISPR medicine that exists for a few hundred patients in Boston and London and Rome is a scientific event. A CRISPR medicine that exists for a child in Lagos or Nagpur is a different invention, and it hasn't been licensed yet. We'll update this when the conditioning gets kinder, when the follow-up hits ten years, and when the geography of the fridge matches the geography of the disease. Until then, the dates. MHRA, 16 November 2023. A scissors, on a licence, sitting in a fridge.

In short. The open questions: kinder chemotherapy, editing the mutation itself, long-term cancer risk, and whether this can ever reach the countries where sickle-cell disease is common.