Two kinds of designed protein scissors: what UW’s new Nature studies actually tested

The shared task is a difficult bond to cut

Two Nature papers published on October 14 describe computationally designed proteases, enzymes that cut proteins at selected sites. The teams at UW Medicine’s Institute for Protein Design approached the chemistry in different ways: an activated cysteine site in one study and a zinc-based site in the other.

UW’s announcement describes tests on purified targets and, in one study, living human cells cultured in the laboratory. These are experiments in biochemical and cell settings, not demonstrated treatments in patients. The examination here draws on the publicly accessible Nature abstracts and the institution’s report; no paywall or institutional-access restriction was bypassed.

The common engineering problem is not simply to make a protein attach to another protein. It is to arrange a catalytic site and its target so that a chosen bond is cut. That makes the local chemistry and positioning of the substrate central to the result.

The cysteine study: active designs and a remaining gap

The cysteine paper reports activity in 13 of 69 designs, a 19% success fraction in that screen. Its public abstract also reports six crystal structures close to the design models and says the resulting folds differ substantially from proteases found in nature. The authors describe cleavage in mammalian cells as well as biochemical activity.

Those are different pieces of evidence. Activity addresses whether a design performs the intended chemistry under the test conditions. A structure close to the computational model addresses whether the molecule takes the anticipated arrangement. A design can be informative on both dimensions without matching the most active natural enzyme.

The authors explicitly discuss that remaining activity gap, pointing to imperfections in active-site preorganization and substrate positioning. That observation is useful because it locates a next design problem rather than treating every active enzyme as a finished optimum.

For a builder of biological tools, the distinction between an active design and an optimized design is practical. It separates proving that an approach works from deciding which features need refinement for a particular use.

The zinc study: a different catalytic arrangement

The metalloprotease paper describes zinc-based, water-activating sites placed beside a peptide-binding pocket. Of 135 tested designs, 36% catalysed cleavage of their intended peptide substrates at the targeted bond, according to the public abstract.

The same abstract reports designed enzymes for targets including TDP-43, amyloid-beta and serum amyloid A, and describes additional work on molecules that can be selectively activated by the proteases. These are reported experimental capabilities, not approvals or commercial products.

The approach provides a second way to organize the cutting chemistry. Instead of treating all successful protease design as one technique, the two papers show that different catalytic arrangements can be engineered and examined. That broadens the set of design choices while leaving each approach’s requirements to be established experimentally.

The substrate pocket is part of that choice. A molecule needs not only a potentially active site but an arrangement that brings the intended target into the useful position. This connection between local chemistry and target handling is more informative than a claim that a model produced a convincing three-dimensional picture.

Nineteen percent and 36 percent are not a league table

The two activity fractions describe different studies and different sets of designs. They should not be used as a simple ranking of which catalytic approach is better. The comparison would need equivalent targets, conditions and acceptance criteria before a percentage could answer that question.

The papers also report large rate enhancements relative to uncatalysed reactions. That denominator is a chemical background reaction, not an existing medicine, a treatment outcome or the speed of a patient's recovery. Keeping the comparison attached to its experimental meaning prevents a striking number from being converted into a clinical promise.

The study-specific evidence offers a better comparison: what was designed, what was tested, what structural support was obtained and which limitations the authors identify. Those questions preserve the engineering content without inventing a head-to-head experiment.

The advance connects a design to an active molecule

Taken together, the publications demonstrate two routes from computational design to measured protein-cutting activity. They also show why the field needs several kinds of evidence: an intended structure, an active catalytic site and a target handled in the right way are related, but not identical achievements.

For future research or product development, the useful starting point is now more concrete. A team can examine a catalytic approach and the particular evidence supporting it, then choose the next experiment for its intended application. That is the significance of these studies as an AI-enabled deeptech advance: they supply tested molecular designs and identifiable questions for further engineering, rather than a generic assertion that biological computation has solved a disease.