How to Design a pegRNA: PBS, RTT and the PE3 Nick

What prime editing actually is

Anzalone and colleagues described it in 2019 (Nature 576:149-157, doi:10.1038/s41586-019-1711-4): a catalytically impaired Cas9 fused to an engineered reverse transcriptase, programmed by a prime editing guide RNA that both specifies the target site and encodes the desired edit. No double-strand break, no donor DNA.

The scope claim is what made it interesting. That paper reported more than 175 edits in human cells covering targeted insertions, deletions and all 12 types of point mutation, with higher or similar efficiency and fewer byproducts than homology-directed repair, and much lower off-target editing than Cas9 nuclease at known Cas9 off-target sites. Its headline estimate is that prime editing could in principle correct up to 89% of known disease-associated genetic variants.

The cost of that generality is that the guide RNA now carries the edit, so the design has more moving parts than an sgRNA.

The four parts, and what each one is for

• Spacer — the 20 nt protospacer, exactly as in an sgRNA. It decides where the nick goes, and the nick has to be close enough to the edit for the rest to work.
• PBS (primer binding site) — the 3'-most part of the extension. After nicking, the released genomic 3' end anneals here, and the reverse transcriptase starts from it. Too short and it does not hold; too long and it competes with the genomic duplex.
• RTT (reverse transcriptase template) — read backwards by the RT to synthesise the new 3' flap. It encodes your edit plus enough sequence matching the genome 3' of the edit that the new flap has somewhere to land.
• 3' extension — the actual oligo tail you order, which is RTT then PBS, appended after the sgRNA scaffold. Note the order is the reverse of the order in which they are used, which is a persistent source of mangled orders.
• PE3 nicking sgRNA (optional) — a separate ordinary sgRNA that nicks the other strand, biasing repair toward keeping the edited strand. It is a second oligo, not part of the pegRNA.

The PBS is a melting temperature, not a length

Most protocols quote a PBS length — 13 nt is the usual number — and that is a proxy for what actually matters, which is how tightly the nicked genomic 3' end binds. A designer should sweep it, and the right answer moves with GC content.

From one substitution designed on a 200 bp target, three of the top candidates picked three different PBS lengths, all landing on much the same melting temperature:

• A 10 nt PBS, ACATTCAGTT, 30% GC — 32.7 °C.
• A 9 nt PBS, TATTCCCCG, 56% GC — 33.2 °C.
• An 8 nt PBS, AGGGCCAT, 63% GC — 31.5 °C.

Reading the sweep

Three lengths, one target temperature of about 30 °C. Had you fixed the length at 13 nt across all three, the Tm values would have been 43.8, 57.6 and 61.4 °C — the same nominal design giving wildly different binding, which is precisely the situation where one construct works and its neighbour does not for no visible reason.

This is why the sweep is worth returning in full rather than a single recommendation. If your first construct fails, the next thing to try is a neighbouring PBS length, and having the whole curve in front of you tells you which direction moves the Tm and by how much.

The RTT has a similar logic but a wider tolerance: it needs your edit plus homology to the genome 3' of it, and 10 to 16 nt of that homology is the normal working range. Insertions need more, because the flap has to carry the insert as well as enough matching sequence to anneal.

Rank candidates by whether the edit destroys the PAM

This is the single most useful sorting criterion and it is easy to miss. Prime editing does not stop after one round. If the edited allele still presents an intact PAM next to an intact protospacer, the editor can bind and nick it again, which gives the cell more chances to revert your edit than to keep it.

When the edit itself falls inside the PAM or the seed, the edited allele stops being a substrate. That design is strictly better than an otherwise-identical one, and it should be ranked first.

It is not always available. In the worked example, 12 candidate pegRNAs came back for a single substitution and only 3 of them disrupted their own PAM. When none do, the standard fallback is a silent PAM-blocking mutation encoded in the RTT alongside your real edit — which is why a designer that knows your reading frame can tell you whether such a mutation would be silent.

PE3, PE3b, and the distance that matters

PE3 adds a nick on the non-edited strand to bias mismatch repair in your favour. The parameter is the distance between the two nicks, and roughly 40 to 90 bp is the usual working window — close enough to help, far enough that you are not effectively making a double-strand break.

Both extremes exist in real output. Across the 12 designs above, 72 candidate nicking guides came back with nick-to-nick distances from 2 to 85 bp, of which 32 fell inside the recommended range. A designer that only returned guides inside the window would be hiding the fact that for some spacers there is no good option.

PE3b is the safer variant: a nicking guide whose spacer only matches after the edit is installed, so the second nick cannot happen on an unedited allele and the window in which both strands are cut simultaneously closes. Forty of those 72 were PE3b-compatible.

One coordinate subtlety worth knowing, because it will otherwise look like a bug. A PE3b guide exists only in the edited sequence, so its coordinates are reported in the edited sequence's frame, not the frame of the sequence you submitted. For a length-changing edit the two frames diverge downstream of the edit — which is why each nicking guide should carry an explicit note of which frame its coordinates are in, and why you should read that field before pasting a position into anything else.

When the RTT is too small: twinPE

A single pegRNA's flap can carry a point mutation, a small insertion or a small deletion. It cannot carry a kilobase.

Twin prime editing (Anzalone et al., Nat Biotechnol 2022;40:731-740, doi:10.1038/s41587-021-01133-w) uses two pegRNAs that nick opposite strands either side of the region being replaced. Each templates a new 3' flap, the two flaps are complementary to each other across a shared overlap in the middle of the new sequence, and they anneal — so the new sequence is written from both ends inwards, with no donor DNA and no double-strand break. That paper combined twinPE with a site-specific serine recombinase to integrate gene-sized plasmids over 5,000 bp and to invert 40 kb of human genomic sequence.

As a design, it is the single-pegRNA problem twice plus one new parameter: the overlap length where the two flaps meet. Everything else — spacer, PBS sweep, PAM disruption — is the same on each side.

The efficiency levers that are not pegRNA sequence

If a design looks right and still edits poorly, two of the biggest known improvements are not sequence decisions you can make in a designer, and it is worth knowing that before you redesign the spacer for the fifth time.

Engineered pegRNAs (epegRNAs). Nelson and colleagues (Nat Biotechnol 2022;40:402-410, doi:10.1038/s41587-021-01039-7) showed that degradation of the pegRNA's own 3' region — the RTT and PBS — poisons the system, and that adding a structured RNA motif at the 3' terminus prevents it. That change improved prime editing efficiency 3- to 4-fold in HeLa, U2OS and K562 cells and in primary human fibroblasts, without increasing off-target editing. They also released pegLIT, a tool for choosing a linker between the pegRNA and the motif that does not interfere with it. Adding a 3' motif is a modification to the oligo you order, not a different design.

PE4 and PE5. These co-express a dominant-negative mismatch repair protein, MLH1dn, alongside the same editor and the same guides. The pegRNA and nicking-guide sequences are identical; it is an extra plasmid in the transfection, a reagent choice rather than a design one.

Both are worth trying before you conclude the target site is refractory.

What no sequence designer can tell you

Off-target activity. Scoring a spacer's genomic specificity needs the reference genome, which a browser tool does not have. Every spacer you take forward should go through a genome-wide off-target search separately, and this is not optional for anything heading toward cells you care about.

Whether the locus is editable at all. Chromatin state, expression of the editor, and repair context vary between loci and cell lines by more than any design parameter, and the 2019 paper itself reported four human cell lines supporting prime editing with varying efficiencies.

How you will detect the edit. Design the genotyping before the pegRNA, not after — a substitution that neither creates nor destroys a restriction site and sits in the middle of a 700 bp amplicon is a design you will struggle to score.

A working order of operations

1. Define the edit as coordinates plus replacement bases, not as prose. Most design errors are off-by-one errors in the edit window.
2. Generate every candidate for every usable PAM rather than the first that fits, and sort so that designs where the edit destroys the pegRNA's own PAM come first.
3. If none disrupts the PAM, consider a silent PAM-blocking mutation in the RTT — and check it really is silent in your reading frame.
4. Take the whole PBS sweep, not the single recommendation. Order the recommended one first and keep its neighbours as the second experiment.
5. Pick a PE3 nicking guide inside the 40-90 bp window, and prefer a PE3b-compatible one when it exists.
6. Check the coordinate frame on any PE3b guide before using its position anywhere else.
7. Add a 3' structural motif to the pegRNA oligo, and consider co-expressing MLH1dn — the two changes that cost nothing in design and are documented to help.
8. Run the spacer through a genome-wide off-target search before ordering.

The tool

SeqBench's Prime Editing Studio takes a target sequence and an edit — substitution, insertion, deletion or small replacement — and returns a complete design for every usable NGG PAM: spacer with its PAM, nick position, a full PBS length sweep with the melting temperature of each, the RTT encoding the edit with a configurable homology length, the assembled 3' extension to append after the scaffold, and PE3 nicking-sgRNA candidates with their nick-to-nick distances, an in-range flag and a PE3b-compatible flag. Designs that destroy their own PAM are ranked first. A twinPE mode designs the dual-pegRNA pair for a replacement too large for one RTT.

It is explicit about the two things it does not do: off-target activity is not evaluated, and every coordinate is in the submitted pre-edit sequence's frame except the edit-dependent PE3b guides, which say so per entry. It runs in a browser with no install and no account, and the same design is available over the REST API and MCP for batch work.

Related references

Related tools

Related guides

• CRISPR Knockout Workflow: gRNA to Verified Edit →
• Base Editing Windows and Bystander Edits, Explained →
• From a Primer List to a Runnable 96-Well PCR Plate →