13 of 69 designed protein cutters work

Researchers designed protein molecules from scratch that snip a selected link in another protein's chain. Of 69 purified candidates they tested at the bench, 13 cut their intended target (Nature). Cutting a protein at a spot you choose is how cells control which proteins are active, so a tool that does it on demand would let biologists intervene in that system rather than just watch it.

Why this bond is so hard to break

A protein is a chain of amino acids held together by peptide bonds. Those bonds are remarkably reluctant to come apart on their own: drop a protein in neutral water and the links can hold for centuries (Radzicka and Wolfenden).

Cells break them constantly anyway, using enzymes called proteases. An enzyme is a protein that speeds up a chemical reaction, and proteases are the ones specialised in cutting other proteins. Their trick is positioning. The reactive chemical groups must sit at a precise distance and angle from the exact bond being cut, held there long enough for the reaction to happen. Get the geometry wrong by a fraction of a nanometre and nothing happens at all.

That is the real challenge the team took on. They did not invent new chemistry. They borrowed the reactive combination that natural cysteine proteases already use, a sulfur-bearing cysteine amino acid activated by a neighbouring histidine, and tried to rebuild a whole new protein around it that would hold everything in place (Nature).

Designing a shape to hold the chemistry

Here is where the computational part comes in. The team used AI tools to invent protein shapes, the three-dimensional form a chain of amino acids folds into, then to pick amino-acid sequences likely to produce those shapes, and finally to predict whether each sequence would actually fold the way they wanted (ProteinMPNN paper). Software narrowed an enormous range of possibilities down to something a lab could physically make and test. The bench decided what worked.

What makes the result more than copying is how unfamiliar the finished proteins look. The surrounding architecture does not resemble natural proteases (Nature). The chemistry is borrowed; the scaffold holding it is not. The team confirmed this by growing crystals of several designs and reading their atomic structures, which matched what the computers had predicted.

From test tube to living cell

Lab assays showed the designs cutting their targets, and the team went a step further by testing them inside mammalian cells. They introduced the genetic instructions for both the designed protease and a target protein, then looked for evidence of cutting (preprint).

That step matters because a test tube is a forgiving place. A cell is crowded, full of competing molecules, and liable to misfold or destroy a protein it does not recognise. A designed enzyme that survives that environment and still does its job has cleared a real hurdle.

It is worth being precise about what 13 out of 69 means. These were not 69 random guesses; they were candidates already filtered by software, then purified and tested (Nature). So the figure describes how often a carefully selected design works, which is the number that matters if you want to run the process again. Most attempts still failed, and the designs are less active than the natural enzymes whose chemistry they copy (Ars Technica). The researchers think their reactive groups are not yet positioned as precisely as evolution manages.

The appeal of the idea is obvious enough. Many diseases involve a protein doing something harmful, and a tool that could cut a chosen protein would be a way to switch it off (GeekWire). The experiments show that familiar protein-cutting chemistry can work in newly designed protein shapes. Medical use remains untested.