A generative AI model designed a small peptide from scratch, and the peptide's job was to hold an inflammation sensor shut. Dropped into a gel that sat in the wound bed of diabetic mice, it calmed the chronic inflammation that keeps diabetic sores from closing and sped up healing.
The sensor is STING, short for stimulator of interferon genes, a protein inside cells that fires an alarm when an upstream enzyme called cGAS detects stray DNA. A short burst of that alarm helps fight infection. When it runs constantly, it becomes a source of the low-grade, self-sustaining inflammation that stalls repair. Diabetic foot ulcers ↗, a complication that can lead to amputation, are one place where that alarm is stuck on, and the wound sits in a soup of inflammation and destructive enzymes instead of rebuilding tissue.
Designing the off switch
The peptide, called SCP-1, was not screened out of a library. It was built by a computational pipeline that has become the standard toolkit for designing proteins that did not previously exist: RFDiffusion ↗ to sketch a backbone, ProteinMPNN to choose the amino acids that would fold into it, and AlphaFold2-multimer to check that the result would grip its target. The design goal was specific. STING has to swing into an active shape to fire, and SCP-1 was drawn to wedge the sensor's two halves into their inactive dimer, the resting position, so it cannot flip on.
That is a harder ask than blocking a pocket. Stabilizing a shape, rather than plugging a hole, is the kind of problem the newer design tools handle well and older docking screens handle badly. The work, published August 7 in Advanced Science ↗ by a group led by Datao Li at Shanghai Ninth People's Hospital and Shanghai Jiao Tong University, with collaborators at Zhejiang University and the Georgia Institute of Technology, is the second STING-quieting peptide we have covered in a month. The first ↗ came from a docking screen of nearly sixty thousand existing peptides and was aimed at autoimmune disease. This one was designed rather than found, and aimed at a wound. Turning STING down, after years of everyone trying to turn it up to fight cancer, is becoming its own small field.
Getting it to stay in the wound
A peptide that works in a dish still has to survive a wound, which is full of proteases that chew up loose molecules. The group's answer was delivery. They loaded SCP-1 into a hydrogel that turns from liquid to gel once it is in place and releases the peptide only when it senses MMP-9, an enzyme that runs high in inflamed diabetic wounds. The idea is that the wound's own damage signal doses the drug: more inflammation, more release.
In db/db mice, a standard genetic model of type 2 diabetes, the loaded gel quieted the STING-TBK1-IRF3 signaling chain, lowered inflammatory and oxidative stress, nudged the wound's immune cells from an attacking state toward a repairing one, and encouraged new blood vessels to grow. Full-thickness wounds closed faster, with better regrowth of the surface skin layer and better collagen structure underneath.
The honest limits
This is a mouse result with a designed molecule, and both halves carry caveats. The db/db mouse heals differently from a human diabetic foot, which is older, larger, often infected, and starved of blood flow in ways a cage animal is not. The peptide was delivered locally in a gel, so nothing here speaks to whether SCP-1 would behave if it got loose in the body, and STING is a defense sensor, so any drug that dampens it invites the question of what infection it might let through. There is no human data and no dose in a person.
The design story is also easy to oversell. An AI pipeline that produced a peptide which then worked in mice is a real proof of concept, but these tools spit out many candidates and the paper reports the single winner. The interesting claim is narrower and sturdier than the headline: a molecule built to lock a protein in its off shape, rather than to block its active site, held up through a delivery system and an animal wound. For a target as double-edged as STING, being able to design the brake as precisely as the accelerator is the part worth watching.