Two teams at the University of Washington’s Institute for Protein Design published papers in Nature on October 14, 2026, reporting proteases designed from scratch with computational methods that hydrolyse the polypeptide backbone at chosen sequences, including sites in the human disease-associated proteins TDP-43, amyloid-β and serum amyloid A.
The amide bonds linking amino acids along the peptide backbone are stable for hundreds of years in neutral aqueous solution because of the high energy barrier to hydrolysis, and success in de novo enzyme design had previously been largely limited to low-energy-barrier model reactions, the cysteine protease paper states. One study produced zinc-based metalloproteases; the other produced proteases that operate through an activated cysteine nucleophile. Both papers list David Baker as a corresponding author, and Donald Hilvert of ETH Zurich is a corresponding author on both.
A Zinc-Based Metalloprotease Design Pipeline
In Computational design of metalloproteases, co-first authors Anqi Chen and Kejia Wu and their colleagues describe using RFdiffusion methods to design proteins carrying water-activating zinc-based active sites adjacent to a peptide-substrate binding pocket. Of 135 designs tested, 36% catalysed cleavage of their cognate peptide substrates at the targeted scissile bond, the paper reports. The best initial design accelerated peptide-bond hydrolysis by more than 10⁸-fold relative to the uncatalysed reaction, and four substitutions increased that acceleration to more than 10¹⁰-fold by improving active-site preorganization, according to the paper.
The generation process combined one round of initial motif-scaffolded backbone generation with sequence design and AlphaFold 3 filtering, followed by two rounds of partial diffusion refinement with further sequence design and filtering, and a final iteration between sequence design and AlphaFold 3 filtering. To optimize a design named Zn45, the team screened 378 single-site saturation variants across 21 residue positions and 148 combinatorial variants across six positions, yielding an improved variant, Zn45_v2, that was characterized with total-turnover-number and Michaelis–Menten kinetic measurements.
The researchers determined crystal structures of designs including Zn5, TDPr3, ZnO7 and Zn48 and deposited them in the Protein Data Bank under accession codes 11CE, 11CM, 11CP, 11CR, 11CU, 11CY and 11DU. Machine-learned force-field calculations examined activation energies for the nucleophilic-attack step of the natural zinc protease astacin and of the designs ZnO7 and Zn45. The study’s data and code are available at Zenodo.
Selective Cleavage of Disease-Associated Proteins
The metalloproteases paper reports designs that selectively cleave the human disease-associated proteins TDP-43, amyloid-β and serum amyloid A, with rate enhancements of up to 10⁹-fold. Named examples include TDPr3, directed at TDP-43; SAAc8, directed at serum amyloid A at the sequence SSRSFFSFLG; and AbetaF3, directed at amyloid-β at the sequence KGAIIGLMVG.
For TDP-43, the authors redesigned an existing scaffold to cut within the protein’s disordered C-terminal domain, which governs liquid–liquid phase separation and oligomerization, rather than degrading the full-length protein. The paper presents this strategy as a way to modulate that phase behavior and oligomerization without ablating the essential nuclear functions of the protein’s structured domains.
The team further designed caged cytokines and receptor antagonists that the designed proteases selectively activate, which the paper describes as enabling bioorthogonal control of cell state. The authors conclude that de novo design can yield highly active, programmable proteases with potential applications in therapeutics, biotechnology and bioremediation.
A competing-interests disclosure states that provisional patent applications 64/047,504 and 63/921,483 have been filed related to aspects of the methods and designs, listing Chen, Wu, Choi, Preetham Venkatesh, Samuel J. Pellock, Hilvert, Asim K. Bera and Baker as inventors or contributors.
A Second Protease Family Built on Cysteine
The second study, De novo design of cysteine proteases, with Hojae Choi as first author, used a new deep learning method called RFD2-MI — developed by team members Brian Coventry and Magnus S. Bauer — to design enzymes that use an activated cysteine nucleophile to hydrolyse the polypeptide backbone in a sequence-dependent manner. The paper reports a success rate of 13 active designs out of 69 tested, or 19%, with rate enhancements over the background reaction of up to 3 × 10⁷.
The designed proteases adopt folds very different from those of natural proteases, with a TM score below 0.50, and six crystal structures match the design models to within a Cα root mean squared deere deposited in the Protein Data Bank under accession codes 9YNL, 9YNM, 9YOX, 9YOY, 9YOZ and 9YP0. The design pipeline cycled RFD2-MI backbone generation with ProteinMPNN sequence design and AlphaFold 3 filtering
In one redesign round applied to a protease named Dokki, 24 of 89 generated variants outperformed the parent design, with rate improvements ranging from 1.2- to 31-fold, and the most improved variant was designated Dokki-15. The designed proteases efficiently cleaved their targets in mammalian cells, the paper reports, shown in HeLa cells through a reporter assay tracking enzyme-dependent substrate relocalization, with catalytically inactive mutants as controls. The authors’ experimental and computational analyses attribute the remaining activity gap relative to the most active native cysteine proteases to imperfections in active-site preorganization and substrate positioning.
Publication History, Funding and Code
Nature received both papers on October 14, 2025; the cysteine protease paper was accepted on September 9, 2026, and the metalloproteases paper on September 15, 2026. Earlier versions appeared as bioRxiv preprints in November 2025, with the cysteine protease study posted on November 22, 2025, and the metalloproteases study first posted on November 21, 2025, and revised through a fourth version on September 27, 2026.
Funding statements list the Howard Hughes Medical Institute, the Bill and Melinda Gates Foundation, the Defense Threat Reduction Agency, the National Science Foundation, the National Institutes of Health’s National Institute of Allergy and Infectious Diseases and National Institute on Aging, the Advanced Research Projects Agency for Health’s APECx program, The Audacious Project at the Institute for Protein Design and the Open Philanthropy Project, among other funders. Design and analysis code for the cysteine protease study is publicly available in two GitHub repositories, and the cysteine paper’s authors declared no competing interests.
