Base Editing Windows and Bystander Edits, Explained

Two editor classes, four possible changes

A base editor is a Cas9 that has been disabled as a nuclease and fused to a deaminase. It binds where the guide directs it, opens the DNA, and the deaminase chemically modifies a base in the exposed strand. Because there is no double-strand break, you largely avoid the indels that come with ordinary Cas9 editing.

Cytosine base editors (CBEs) carry a cytidine deaminase that converts C to uracil; the cell reads uracil as thymine, giving a C-to-T change. These editors also carry a uracil glycosylase inhibitor to stop base excision repair from simply reverting the intermediate — BE4max is a more efficient descendant of the original BE3 along exactly these lines.

Adenine base editors (ABEs) carry an engineered adenosine deaminase that converts A to inosine, which is read as guanine, giving an A-to-G change. ABE8e is a substantially faster descendant of the ABE7.10 lineage.

Between them that is four available changes, because either editor can act on either strand: a CBE gives you C-to-T reading the forward strand, or G-to-A when the guide targets the reverse strand; an ABE gives you A-to-G or T-to-C the same way. All four are transitions. No classical base editor performs a transversion — you cannot get C-to-G or A-to-T this way — so if that is the change you need, base editing is the wrong technique and prime editing is the usual answer.

Where the window sits

The deaminase is tethered to Cas9, so it can only reach a limited stretch of the opened DNA. That stretch is the activity window, and it is described by position along the 20-nucleotide protospacer, numbered from 1 at the PAM-distal end to 20 immediately adjacent to the PAM.

The widely-cited canonical windows for the common SpCas9 editors, all using an NGG PAM:

• BE3 (CBE, C→T): positions 4 to 8.
• BE4max (CBE, C→T): positions 4 to 8.
• ABE7.10 (ABE, A→G): positions 4 to 7.
• ABE8e (ABE, A→G): positions 3 to 9 — a deliberately wider window, reflecting a faster deaminase that reaches more.

The PAM decides, not you

Here is the constraint that shapes every base-editing design. The window's position is fixed relative to the protospacer, and the protospacer's position is fixed by where an NGG PAM happens to sit in your target. So the set of bases you can reach is determined by the genome, and your only freedom is choosing among whichever NGG PAMs are available nearby.

In practice this means a target base is reachable only if some PAM places it at roughly positions 4 to 8 of that PAM's protospacer. Sometimes several PAMs qualify and you get to pick the one with the cleanest window. Sometimes exactly one does, and you take what you are given. Sometimes none does, and no amount of guide optimisation helps — at that point the options are a Cas9 variant with a different PAM requirement, an editor with a shifted window, or a different technique.

This is also why the useful question is never "can I make this change" in the abstract. It is "is there a PAM that puts this base in a window, and what else does that window contain". Enumerating every candidate guide and looking at each one's window contents is exactly the mechanical work worth handing to a tool.

Bystanders, and why they are not always a problem

The deaminase acts on every eligible base in the window, not the one you had in mind. A CBE window containing three cytosines will tend to edit all three. Those extra edits are bystanders, and they are the main source of unintended changes in base editing.

But the count of bystanders at the nucleotide level is the wrong thing to optimise, because what matters is usually the consequence in protein. Take a concrete case. Say your target region is ATGACCAAGATCAGATGGAATGG, where the first 20 bases are the protospacer and the final TGG is the PAM, and the coding sequence begins at position 1. With a BE4max window of positions 4 to 8, the window covers ACCAA — two cytosines, at positions 5 and 6. There is one bystander no matter which of them you wanted.

Now read it in frame. Positions 4 to 6 are the codon ACC, threonine. Suppose the edit you want is C5 to T, making the codon ATC, isoleucine — a deliberate missense change. The bystander at position 6 fires too, so the codon actually becomes ATT. ATT is also isoleucine. The bystander changed the DNA and did nothing to the protein, because it landed in the wobble position of the codon. This guide is perfectly usable.

Run the same logic the other way and it turns hostile. If the edit you wanted was C6 to T — silent on its own, since ACC and ACT are both threonine — then the bystander at position 5 converts a silent edit into a missense one. Same guide, same two edits, opposite verdict, entirely because of which base you were aiming at.

So the rule is: count bystanders in nucleotides, judge them in amino acids. A tool that flags the bystanders is doing half the job; one that also translates each resulting codon and tells you whether the change is silent, missense, nonsense or a stop loss is answering the question you actually have.

Reducing bystander damage

In rough order of how often it works:

1. Try every available PAM, not the first one you find. A different protospacer shifts the window along the sequence, and a window that captures your target with fewer eligible bases beside it is the single best fix.
2. Read the consequences in frame before rejecting a guide. As above, a bystander in a wobble position is frequently silent, and a guide with one harmless bystander beats a guide with none that is not available.
3. Exploit position within the window. Editing efficiency is generally highest near the middle of the window and falls off at the edges, so a guide that puts your target mid-window and a bystander at the margin is better than the reverse, even with identical counts.
4. Consider a narrowed-window editor variant. Engineered deaminases with tighter windows exist precisely for this, at some cost in efficiency on the intended base.
5. Switch technique. If no guide gives an acceptable outcome, prime editing writes an arbitrary specified edit without depending on a deaminase window, and handles transversions and small insertions and deletions that base editors cannot.

Strand bookkeeping, and what a window tool does not tell you

One source of confusion worth naming: when the guide targets the reverse strand, the deaminase acts on a cytosine in the protospacer, which is a guanine on the forward strand. The change is a C-to-T in the editor's frame of reference and a G-to-A in your reference sequence's frame. Both descriptions are correct and they look like different edits. Any design output should tell you the forward-strand change so it lines up with the coordinates you are actually working in.

It is also worth being clear about what window-based design does and does not cover. Placing edits by an editor's canonical window is deterministic and reliable: given the sequence and the editor, the eligible bases and their coding consequences follow. What it does not do is predict how efficiently a particular guide will edit, or find genome-wide off-target sites — that needs a reference genome and a specificity model. SeqBench's Base Editing Designer is explicit about this boundary: it enumerates every SpCas9 guide whose window places an editable base, flags the bystanders, ranks bystander-free guides first, and translates the consequences when you give it a reading frame. Validate the spacer's genome-wide specificity separately before ordering anything.

Related references

Related tools

Related guides

• Reverse Complement of DNA, Explained →
• Open Reading Frames (ORFs) Explained →
• siRNA, shRNA and ASO Design Rules That Actually Matter →