AlphaFold gene editing safety method cuts off-target edits in tests

Researchers have used AlphaFold to improve gene editing safety by finding how CRISPR proteins tolerate mismatched DNA targets, according to a study published in Nature. The work matters because gene-editing therapies often must alter many cells, so even uncommon mistakes can become a safety problem.

The team, based at several institutions in China, focused on off-target edits: changes made at DNA sequences that resemble the intended target but are not the correct site. Existing design methods can reduce that risk by choosing guide RNAs with fewer close matches in the genome, but Cas proteins can still bind some imperfect matches.

How does AlphaFold make gene editing safer?

AlphaFold can predict how proteins and nucleic acids sit near each other. The researchers used those predictions to compare CRISPR complexes bound to intended DNA targets with complexes bound to off-target sequences, then looked for amino acids in Cas proteins that changed their contacts when mismatches were present.

CRISPR gene editing relies on several parts working together. A guide RNA pairs with the target DNA sequence, a Cas protein such as Cas9 helps recognize that pairing, and another enzyme or protein activity changes the DNA once the complex is in place.

In principle, a guide sequence about 18 bases long should be rare in a random genome. The Nature paper notes that such a sequence would be expected roughly once in 70 billion bases, while the human genome has about 3 billion bases. The complication is that Cas9 can still bind when a small number of bases do not match, depending on where those mismatches occur.

To study that behavior, the researchers first built a large collection of off-target sites. They used a modified CRISPR system that converts the DNA base adenine into inosine, then isolated DNA fragments carrying that chemical mark. They repeated the process with 10 guide RNAs to gather many examples of unintended targets.

The team initially tried feeding AlphaFold the DNA target, guide RNA, Cas9 and a base-modifying enzyme attached to Cas9. That model placed one protein in an obviously wrong position, according to the report. The researchers then simplified the input to DNA, guide RNA and Cas9, which produced structures consistent with experimentally determined CRISPR structures.

Comparing those models showed two patterns. About two-thirds of off-target sites led Cas9 to take on a slightly altered shape, the researchers reported. More than 95 percent changed which Cas9 amino acids contacted the RNA, suggesting that flexible local contacts can help the protein accept mismatched DNA-RNA pairings.

The group turned that comparison into a computational method called ContactSeek. It uses AlphaFold’s contact probability output, which estimates whether two molecular components sit within a short distance, to identify amino acids whose contacts differ between matched and mismatched targets.

ContactSeek produced many candidate amino acids, so the researchers concentrated on clusters within Cas9 that appeared to adjust around mismatches. They then changed individual amino acids at those sites and tested the resulting proteins.

Across 23 amino-acid swaps at 10 positions, the team found a Cas9 variant that kept activity similar to ordinary Cas9 at the intended target while cutting off-target activity from 28 percent to 5 percent. The researchers also reported similar results with other guide RNAs and showed that the strategy could be applied to a related system using Cas12.

Other groups have developed more accurate Cas9 variants through methods such as directed evolution. In tests described in Nature, the AlphaFold-guided designs generally performed similarly or somewhat better on activity and specificity, though the newly identified changes may be more tailored to particular guide RNA and mismatch combinations.

The authors said the approach could help produce gene-editing systems designed around known off-target risks. They also suggested the same strategy may be useful for tuning other protein-DNA interactions beyond CRISPR-based editing.

This story draws on original reporting from Ars Technica.