Prime editing

Prime editing installs a chosen DNA sequence at a chosen site by nicking one strand of the double helix and then writing new bases onto the exposed end from an RNA template carried by the guide itself. Described in 2019, it removes two constraints at once: it needs no double-strand break, and unlike Base editing it is not restricted to particular chemical conversions.¹ Its practical difficulty is the reverse of that generality — the machine is large, the design space for each target is wide, and efficiency varies from excellent to negligible depending on the site.

§How it works

Three things have to happen in order, and each can fail independently: the editor must find and nick the target, the reverse transcriptase must copy the intended sequence onto the exposed strand, and the cell must keep the new sequence rather than discard it. The third step is the one the designer controls least.

§The pegRNA

The editor is a Cas9 nickase, disabled in one of its two nuclease domains so that it cuts only the strand containing the PAM, fused to an engineered reverse transcriptase. The guide is extended into a prime editing guide RNA, or pegRNA, which carries three functional parts: a spacer that specifies the target, a primer binding site complementary to the nicked strand, and a reverse transcription template encoding the desired new sequence.

After the nick, the freed 3' end of the DNA anneals to the primer binding site. The reverse transcriptase then extends that end, copying the template into DNA. The result is a flap of newly written DNA carrying the edit, competing with the original flap for reincorporation into the duplex.

§Resolving the flap

Whether the edit survives depends on how the cell resolves that competition. Excision of the original flap and ligation of the new one installs the change on one strand; the resulting mismatch must then be resolved in the edit's favour rather than corrected back. Mismatch repair frequently reverses prime edits, which is why an early and effective improvement was transient suppression of that pathway.² A second strategy nicks the unedited strand as well, marking it as the one to be rewritten. That raises efficiency but reintroduces a modest risk of indels, since two nearby nicks approximate a double-strand break.

1. 2019
   Prime editing described
   Anzalone and colleagues report search-and-replace editing without double-strand breaks or donor DNA, demonstrating all twelve base substitutions and small insertions and deletions in human cells.
2. 2021
   Mismatch repair identified as the brake
   Transient suppression of mismatch repair substantially raises editing efficiency, defining the PE4 and PE5 systems.
3. 2022
   Guide RNA stabilised
   Structured RNA motifs added to the pegRNA 3' end protect it from degradation, and paired-pegRNA designs enable larger replacements and integrase-mediated insertions.
4. 2023–2024
   Compact and stabilised editors
   Evolved reverse transcriptases shrink the construct, and fusion of an RNA-binding protein that recognises the pegRNA improves performance across cell types.
5. 2025
   First clinical data
   Initial results are reported from the first prime editing trial, an ex vivo therapy for chronic granulomatous disease.

§Development history

The original report defined a numbered progression that the field still uses. PE1 paired the nickase with an unmodified viral reverse transcriptase and worked poorly. PE2 used an engineered enzyme and worked well enough to be usable. PE3 added the second nick. Later systems addressed mismatch repair, pegRNA stability, and the size of the reverse transcriptase itself, with each generation typically improving efficiency several-fold in the cell types where it had previously been weakest.

Two extensions widened the scope beyond small edits. Paired pegRNAs that write complementary flaps allow replacement of a whole segment rather than a few bases. Coupling prime editing to a serine integrase — using the editor to install a short recombinase landing site and the integrase to insert a payload into it — permits insertions of many kilobases, closing part of the gap between editing and gene addition.³ That capability connects prime editing to the larger-scale ambitions described in Synthetic genomes and Genetic code expansion and recoding, where the goal is to rewrite rather than repair.

§Current state

Prime editing is a standard research tool for installing precise variants in cell lines, Induced pluripotent stem cells and Organoids, where the ability to write an exact patient mutation makes disease modelling far cleaner than nuclease-based approaches allow. In animals, prime editing has corrected the causal mutation in mouse models of metabolic liver disease and several other monogenic conditions, in each case in mice rather than in humans.⁴

Clinical translation began recently. The first prime editing therapy to reach patients treats chronic granulomatous disease by correcting a mutation in blood stem cells outside the body; initial results reported in 2025 indicated restoration of the missing immune-cell function in a treated patient. The company behind it subsequently restructured and redirected effort toward liver targets, a reminder that in this field the binding constraint is often capital rather than biology. No prime editing product is approved anywhere as of 2026.

Terminology

"Search and replace" is the phrase from the original paper and is apt for the mechanism, but it invites a word-processor analogy that overstates reliability. There is no undo, no confirmation step, and no guarantee the replacement is installed in every cell.

§Limitations

Delivery is the acknowledged bottleneck. Fusing a reverse transcriptase to a Cas9 nickase makes a protein substantially larger than Cas9 alone, and larger again than a base editor; with its pegRNA the construct sits well beyond the packaging capacity of a single adeno-associated virus. Workarounds include splitting the protein across two vectors and reassembling it inside the cell, packaging messenger RNA in Lipid nanoparticles, and delivering preformed protein in engineered virus-like particles. Each adds manufacturing complexity, and none has yet been shown to work efficiently in tissues outside the liver and blood.

Efficiency is also site-dependent in ways that remain partly empirical. The length of the primer binding site and of the template, the local chromatin state, and the cell's mismatch repair activity all matter, so optimising a new target still involves screening dozens of pegRNA designs. Reported efficiencies in primary human cells commonly sit well below those in immortalised cell lines, and papers do not always make the distinction obvious.

§Risks

Because installing an edit requires three separate base-pairing events — spacer, primer binding site and template — prime editing is intrinsically more discriminating than a single-nick or single-cut system, and measured off-target editing at guide-similar sites is generally lower than for CRISPR–Cas9. The characteristic unintended products are different in kind: fragments of the pegRNA scaffold copied into the target site, tandem duplications, and small indels where the second nick was used.

For Somatic gene therapy, the risk profile is favourable enough that the main open questions concern delivery vectors rather than the editor. For heritable applications, prime editing is sometimes advanced as the tool that would make Human germline editing technically defensible. That argument runs into the same objections as its predecessors: efficiency in embryos is not the issue, mosaicism and verification are, and the international frameworks catalogued in Governance of human genome editing are not written in terms of which enzyme is used — a posture that has not moved since the He Jiankui affair. The practical alternative for most heritable disease risk remains embryo selection rather than editing.

§Outlook

Prime editing's ceiling is set by a design problem as much as a chemical one. Whether pegRNA design becomes predictable depends on whether rules learned from large screening datasets transfer to primary cells and to loci nobody has tested. If they do, retargeting turns into a lookup and the platform scales; if they do not, every new mutation stays a small research project, which is a poor foundation for a technology whose stated purpose is treating rare mutations one at a time.

The comparison that will settle the technology's role is with the simpler editors rather than with nucleases. Where a transition substitution would fix the mutation, a base editor is smaller, better characterised in humans, and further along in trials; prime editing earns its complexity only at targets the simpler chemistries cannot reach. Compared with Casgevy and the other approved editing therapies, its repertoire is wider and the clinical evidence behind it far thinner. A secondary question, shared with Gene therapy for aging and with the ambitions of equitable access generally, is whether a platform whose selling point is one bespoke correction per patient can ever be manufactured and approved at a cost that matches the size of each patient population.

§See also

• Base editing
• CRISPR–Cas9
• Epigenome editing
• Off-target effects in genome editing
• AAV vectors
• Somatic gene therapy
• Synthetic genomes
• Human germline editing

§References