AlphaFold3 helped ContactSeek identify substitutions that reduce off-target DNA base editing
AlphaFold3 helped ContactSeek identify substitutions that reduce off-target DNA base editing
On July 22, Haowei Meng's team published a Nature paper on ContactSeek. The method combines AlphaFold3 predictions of protein, RNA, and DNA complexes with sequencing data to propose amino acid substitutions for more selective editors.
A base editor changes a single DNA letter without creating a double-strand break. A guide RNA directs it to the intended sequence, a CRISPR protein holds the complex on the DNA, and an attached enzyme performs the chemical conversion. Similar sequences can sometimes retain the complex as well. When this happens, the editor changes letters outside the intended target.
Engineers usually test amino acid substitutions in the protein and measure the results. ContactSeek begins with the original editor's error profile. The authors identified sites across the genome where the editor bound together with the guide RNA, then supplied AlphaFold3 with the sequences of the target and these off-target sites. The model generated possible structures of the protein, RNA, and DNA complex.
AlphaFold3 provides a map of probable contacts within the complex. ContactSeek compares differences between the target and off-target sites with the sequencing signal. This allows the program to identify amino acids whose contacts with RNA or DNA vary with the frequency of off-target editing, then propose substitutions for testing in cells.
The authors tested this workflow with a Cas9 adenine base editor and a Cas12a cytosine base editor in HEK293T cells. For the ABE8e-DD variant, the total signal from guide RNA-dependent off-target editing was 99,2% lower than for the original ABE8e. At the ABEsite16 target, the comparison included 270 off-target sites. For the Cas12a cytosine base editor variant, the signal decreased by 82,1–95,1% across four guide RNAs. The authors also measured on-target editing, off-target RNA changes, and guide RNA-independent editing.
Base editors have already been used in a patient. In KJ Muldoon's personalized therapy, lipid nanoparticles delivered the editor to the liver. One year later, the child showed clinical improvement without serious adverse effects. ContactSeek has so far been tested only in HEK293T cells, so its variants still require evaluation after delivery into tissues and during long-term follow-up.
ProteinMPNN helped generate prime editor variants that accumulated more effectively in cells. ContactSeek addresses a different problem. It narrows the list of sites where an editor may act outside its intended target. The contact map turns a broad search through amino acid substitutions into specific hypotheses that can be tested experimentally.