AI Redesigns Minimal CRISPR Nucleases

A study from Jennifer Doudna's lab addresses a longstanding challenge in protein engineering: designing complex, multi-domain enzymes whose activity depends on coordinated structural rearrangements. Rather than relying solely on sequence-based language models, the authors used the ESM Inverse Folding model together with information from natural sequence conservation and co-evolution between TnpB and its RNA and DNA substrates. This approach constrained only residues predicted to be functionally critical while allowing extensive redesign elsewhere.

High-throughput bacterial screening identified hundreds of active synthetic variants, with several outperforming wild-type TnpB. The most successful variants achieved genome-editing efficiencies comparable to or greater than the native enzyme in HEK293T cells and Arabidopsis protoplasts, while maintaining the expected target recognition sequence. Cryo-electron microscopy of one highly divergent variant showed that AI-generated amino acid substitutions created new networks of interactions stabilising the RNA-DNA interface during different conformational states, including a previously unobserved DNA-bound intermediate.

These findings demonstrate that CRISPR-related nucleases can be extensively redesigned while preserving function, expanding the range of programmable genome-editing proteins beyond those produced by natural evolution.

The study was led by Petr Skopintsev, Isabel Esain-Garcia, Evan DeTurk and Jennifer Doudna, with researchers from the Innovative Genomics Institute at UC Berkeley. It was published in Science on 16 July 2026.