Gene Editing Reaches the Clinic: From CRISPR Scissors to Base Editors

In late 2023, regulators approved a therapy that edits a patient's own genome to treat sickle cell disease. It was the first approval of a CRISPR-based medicine, and it marked a transition that had been promised since the technology's description a decade earlier. But the approval also revealed the shape of the field's remaining difficulty, because the reason this particular disease came first has less to do with genetics than with logistics.

The mechanism, briefly and accurately

CRISPR systems originate in bacterial adaptive immunity. Bacteria store fragments of viral genomes and use them to recognise returning invaders. Cas9, an enzyme from that system, can be programmed with a short guide RNA: wherever the guide's sequence matches genomic DNA — adjacent to a required short motif called a PAM — Cas9 binds and cuts both strands.

The cut itself is not the therapy. What matters is how the cell repairs it.

Non-homologous end joining, the dominant repair pathway, rejoins broken ends imprecisely, often inserting or deleting a few bases. Within a protein-coding sequence, that usually shifts the reading frame and destroys the gene product. This makes Cas9 an excellent tool for disabling a gene — reliable, efficient, and the basis of most early clinical successes.

Homology-directed repair can instead copy a supplied template into the break, enabling precise correction. But it operates mainly during specific phases of cell division, is far less efficient, and competes with end joining. In non-dividing cells — neurons, cardiac muscle — it is largely unavailable. This asymmetry is why "correcting a mutation" proved much harder than "knocking out a gene".

The clinically decisive question is rarely whether an edit can be made in a dish. It is whether enough of the right cells in a living body can be reached, edited correctly, and left otherwise unharmed.

Editing without cutting

Two later platforms address the precision problem by avoiding double-strand breaks.

Base editors fuse a catalytically weakened Cas protein — one that binds and unwinds DNA but does not cut both strands — to a deaminase enzyme. Cytosine base editors chemically convert C to U, which the cell reads and resolves as T, producing a C-to-T change. Adenine base editors convert A to inosine, read as G, producing A-to-G. Since a large share of known pathogenic single-nucleotide variants fall into categories these transitions can reverse, the clinical reach is significant. And because no double break occurs, the risk of large deletions and chromosomal rearrangements drops substantially.

Base editing has its own error modes. Bystander editing changes other susceptible bases within the editing window. Early cytosine editors caused off-target RNA deamination, since the deaminase acts on RNA as well; engineered variants reduced this. And the platform cannot make transversions or insert sequence.

Prime editors go further. They combine a nickase Cas9 with a reverse transcriptase and an extended guide that carries both the target sequence and a template for the desired edit. The system nicks one strand and writes the new sequence directly from the RNA template. In principle this permits any substitution plus small insertions and deletions. In practice, efficiency has historically been lower than base editing and the molecular components are larger, which complicates delivery — though successive engineering generations have narrowed both gaps considerably.

A related approach, epigenome editing, does not change DNA sequence at all. A programmable binding protein is fused to a domain that adds or removes chemical marks on DNA or histones, turning a gene's expression up or down. This is reversible and avoids permanent sequence change, which is attractive for conditions where you want to silence a gene rather than destroy it.

Why sickle cell disease came first

Sickle cell disease results from a single well-characterised mutation in the beta-globin gene, producing haemoglobin that polymerises when deoxygenated and deforms red blood cells. The consequences — vaso-occlusive crises, chronic anaemia, cumulative organ damage — are severe and lifelong.

The approved therapy does not repair the mutation. It exploits a developmental switch. Fetal haemoglobin, expressed before birth, does not sickle; a transcriptional repressor called BCL11A shuts it off in infancy. Disrupting the erythroid-specific enhancer of BCL11A in blood stem cells lifts that repression, restoring fetal haemoglobin production in the patient's red cells. This is a gene knockout — precisely the operation Cas9 does best.

The decisive advantage, though, is anatomical. Haematopoietic stem cells can be mobilised, collected from blood, edited in a laboratory where conditions are controllable and editing can be verified, and returned to the patient. This ex vivo route sidesteps the entire problem of delivering editing machinery to cells inside a living body.

The cost of that route is the conditioning regimen. For edited cells to engraft, existing marrow must be cleared with high-dose chemotherapy — a procedure with real toxicity, including infertility risk and prolonged immune suppression, requiring weeks of specialist inpatient care. Non-genotoxic conditioning using targeted antibodies is under active investigation and would materially widen who can be treated.

Delivery is the field's central problem

For diseases where the relevant cells cannot be removed and returned, editing must happen in vivo, and the delivery vehicle determines feasibility.

Lipid nanoparticles encapsulate messenger RNA encoding the editor plus the guide RNA. They are manufacturable at scale, and because the mRNA is transient the editing machinery disappears within days, limiting off-target exposure. Their limitation is tropism: intravenously administered lipid nanoparticles accumulate predominantly in the liver. That is fortunate for hepatic targets — an in vivo therapy for transthyretin amyloidosis, which knocks out a liver-expressed gene, has shown durable protein reduction in trials — and a genuine obstacle elsewhere. Engineering particles that reach lung, muscle, bone marrow, or immune cells is one of the most consequential open problems in the field.

Adeno-associated viral vectors transduce many tissue types efficiently, including retina, muscle, liver, and central nervous system with appropriate serotypes. Their drawbacks are a small cargo capacity that struggles to accommodate larger editors, pre-existing neutralising immunity in a substantial fraction of the population, dose-dependent toxicity at the high doses systemic delivery requires, and persistent expression of the nuclease — undesirable when the edit only needs to happen once.

Local administration avoids systemic distribution entirely. Subretinal or intravitreal injection for inherited retinal disease, direct injection into muscle, or delivery to the cornea permits lower doses and confines exposure. The eye's relative immune privilege and small target volume have made ophthalmology an early proving ground.

Virus-like particles and engineered delivery vehicles aim to combine viral efficiency with transient, non-integrating cargo, packaging editor protein and guide rather than DNA. This is promising, early, and not yet manufactured at scale.

Safety, assessed carefully

Genome editing carries risks that require specific evaluation rather than general reassurance.

Off-target editing occurs where a guide partially matches an unintended site. Computational prediction is imperfect, so unbiased genome-wide assays are used to map real cut sites, and high-fidelity engineered nucleases reduce tolerance for mismatches.

On-target structural consequences have received less public attention but matter more in some contexts: double-strand breaks can cause large deletions, chromosomal translocations, and in rare cases loss of an entire chromosome arm. Non-cutting editors substantially reduce this class of risk.

Mosaicism is intrinsic to in vivo editing — not every target cell is edited. For some diseases, partial correction suffices because edited cells have a selective advantage or because a threshold of functional protein is enough. For others it does not, and that biology determines viability.

Immune response to bacterial Cas proteins is plausible, since many people have prior exposure to the source bacteria, and to the delivery vehicle itself.

And the germline distinction remains the field's firmest boundary. All therapies discussed here edit somatic cells; changes are not inherited. Heritable germline editing raises consent problems that cannot be resolved by better technology — the edited individual cannot consent, and errors propagate to descendants. The 2018 case of embryos edited without adequate oversight prompted broad international condemnation and criminal prosecution, and the resulting scientific consensus treats clinical germline editing as unacceptable under current knowledge.

The economics nobody has solved

Approved gene therapies carry list prices in the low millions of dollars per patient. The rationale offered is that a single administration replaces decades of costly chronic care, and for a genuinely durable cure that arithmetic can hold. But it collides with the epidemiology: sickle cell disease is most prevalent in sub-Saharan Africa and South Asia, and a therapy requiring apheresis, marrow conditioning, and specialist transplant infrastructure is unavailable to the large majority of people who have the disease.

Several efforts target this gap directly — in vivo editing of blood stem cells that would remove the need for cell collection and conditioning, simplified manufacturing, and non-genotoxic conditioning. These are not incidental refinements. They determine whether genome editing becomes a broadly available class of medicine or a narrow one.

What comes next

Realistic near-term expansion follows delivery, not editing chemistry. Liver-directed therapies for metabolic and amyloid diseases will progress fastest because lipid nanoparticles already reach the liver. Retinal, muscular, and central nervous system programmes advance where local administration or a suitable viral serotype exists. Base and prime editing will move from first trials toward broader use as efficiency improves and delivery vehicles shrink to accommodate them. Multiplexed editing of immune cells will continue expanding cell therapy, including off-the-shelf allogeneic products.

The decade-old promise that CRISPR would let us correct genetic disease has, in a narrow and hard-won sense, been kept. What the first approvals actually demonstrated is that the molecular biology was the tractable part. Getting the machinery into the right cells, in the right people, at a price a health system will pay, is the work that remains.