AI Rewrites Gene-Editing Scissors: Artificially Designed TnpB Demonstrates Editing Ability in Cells

Gene-editing tools do not necessarily have to be found only in nature. Using artificial intelligence, researchers redesigned TnpB, a miniature CRISPR-associated “gene-editing scissor,” creating synthetic nucleases with sequences that diverge substantially from natural proteins yet can still cut DNA according to RNA guidance. The study suggests that protein engineering may be able to move beyond the existing boundaries left by evolution, adding new editing components for medicine, agriculture, and basic research.

Published in *Science*, the study lists Jennifer Doudna as senior author and Petr Skopintsev, Isabel Esain-Garcia, and Evan C. DeTurk as co-first authors. The team focused on TnpB—a class of compact RNA-guided nucleases considered evolutionary precursors of Cas12—and named the newly designed proteins SynTnpB.

The design process did not allow a generative model to freely rewrite sequences. The researchers first used evolutionary information from the TnpB family to fix amino acids important to structure or function, then applied an “inverse protein-folding” model to work backward from a predetermined three-dimensional structure and infer sequences that might form it. The DNA-binding interface and guide RNA-binding interface were designed separately, with about 50 candidate versions selected for each. These were recombined and screened in *Escherichia coli*.

Candidate proteins that passed the initial screening then underwent genome-editing tests in plant and human cells. The study showed that the editing activity of some SynTnpBs could match or exceed that of natural TnpB; the most active candidate shared only 77% sequence similarity with the natural protein, indicating that function does not necessarily depend on reproducing a natural sequence almost exactly. However, the current findings remain a proof of concept at the cellular and laboratory stages and cannot be directly equated with a technology ready for use in patients or field crops.

Cryo-electron microscopy further captured a conformation of TnpB while it was bound to a target-adjacent motif. This state had previously been proposed by models but had not been directly observed. The structural data therefore not only confirmed that the synthetic protein could still function, but also provided clues explaining how the nuclease recognizes its target and initiates cutting.

Efficiency is not the only challenge. Independent experts noted that some of the more active variants also showed more editing at non-target sites, highlighting the trade-off between activity and specificity; protein stability, intracellular interactions, and manufacturing costs have also not yet been comprehensively assessed. In addition, the current process still cannot rapidly reengineer the enzyme into versions that recognize different DNA or RNA sequence motifs. What this achievement truly opens, therefore, is not an immediately usable pair of universal scissors, but a new path for AI design jointly constrained by structure, evolution, and experimental screening.