🎉 Baker Lab just reported an exciting new advance in de novo enzyme design!
🧬 Using RFdiffusion3, the team built a site-selective ubiquitin ligase from scratch. On a PCNA trimer containing 45 competing lysines, it directed ubiquitin to K164 and achieved ~90% PCNA ubiquitination.
This is a new wet-lab application of RFdiffusion3—not a recap of the model’s earlier release. The preprint shows that all-atom generative protein design can build a complex enzyme that controls not only whether ubiquitin is transferred, but where it goes.
🔬 A strong first experimental screen
The team designed ligases around a cysteine-based catalytic triad. Of 48 designs tested, 35 transferred ubiquitin to a model peptide.
🎯 Selectivity on a difficult protein substrate
PCNA is a homotrimer with many possible lysine acceptors. Rather than modifying the most accessible residue, the selected designs preferentially placed ubiquitin at K164—a key regulatory site in DNA-damage tolerance.
⚡ ~90% conversion with unactivated ubiquitin
After adding an activity-enhancing mutation and coupling the system to E1 activation, the best design reached ~90% PCNA ubiquitination using unactivated ubiquitin. It also showed better site selectivity than the engineered E2 control.
⛓️ Programmable polyubiquitin installation
The same cascade installed polyubiquitin chains of defined length onto K164 through native isopeptide bonds. That is a meaningful step beyond one-off protein labeling.
🧱 The designed fold was experimentally realized
An apo crystal structure reached 1.6 Å resolution and closely matched the computational design, with a Cα RMSD of 0.91 Å.
⚠️ Important boundary: this is a bioRxiv preprint. The experiments use in-vitro activated donor chemistry or E1 + MESNa—not a demonstrated intracellular E1–E2–E3 ubiquitination pathway. The manuscript does not provide a verified official code repository or PDB accession.
📄 Preprint:
#
BakerLab# #
RFdiffusion3# #
ProteinDesign# #
EnzymeDesign# #
Ubiquitin# #
AIforScience#