Antibiotic candidates in an unlikely place: AI finds 'prionins' hidden inside prion proteins

Prions are best known as misfolded proteins behind fatal brain diseases. So it was a surprise this summer when a University of Pennsylvania team reported that the same family of proteins hides short fragments able to kill bacteria ¹. The researchers call these fragments "prionins." The results so far come from lab dishes and a single mouse skin-infection model, not from people, and the lab presents the work as a new source for antibiotic discovery, not a treatment ².

Why look for antibiotics inside other proteins

Antimicrobial peptides (AMPs) are short chains of amino acids that many living things use as a first line of defence. Most carry a positive charge that draws them to the negatively charged membranes of bacteria, which they can then disrupt ⁵. Reviewers point to a key appeal: bacteria seem to develop resistance to them less readily than to many conventional antibiotics ⁵.

The idea is not only theoretical. Daptomycin, a cyclic lipopeptide antibiotic, is approved for complicated skin infections caused by certain Gram-positive bacteria ⁶. It works by inserting into the bacterial membrane and disrupting its electrical charge, which kills the cell quickly ⁶. Polymyxins such as colistin remain last-resort options against some multidrug-resistant infections ⁵.

The Penn group, led by César de la Fuente, uses AI to mine protein databases for possible antibiotics. Their deep-learning tools read protein sequences and predict which short stretches might behave like an AMP ¹⁴.

What the prionin study found

Using a model called APEX 1.1, the team scanned 2,897 prion and prion-like proteins, breaking them into 19.3 million short fragments ¹. The model flagged 1,179 as likely antimicrobial ². The researchers then made 75 of the top candidates and tested them against 11 bacterial pathogens, including drug-resistant strains ¹³.

The hit rate was striking:

• 59 of the 75 peptides slowed the growth of at least one pathogen ¹.
• 42 were strongly active at low concentrations ³.
• 16 active peptides did not measurably damage red blood cells or human cells, even at the top concentrations tested in the lab ¹.

Follow-up experiments suggested many of the active prionins work by damaging bacterial membranes, the classic AMP strategy ¹.

The team then moved two of the strongest candidates into mice. One came from a fungus and one from a roundworm ³. In a standard skin infection caused by Acinetobacter baumannii, a hard-to-treat pathogen, a single topical dose of each lowered bacterial levels, with effects comparable to the older peptide antibiotic polymyxin B in that model ¹³. The mice showed no treatment-related weight loss ¹.

Part of a bigger push

This study fits a pattern. Earlier this year the same lab reported ApexGo, a generative AI system that suggests changes to existing peptides to make them better at killing bacteria ⁴. Starting from ten peptides found in extinct organisms, it designed 100 variants. 86 killed at least one type of bacteria, and 68% killed bacteria better than the peptide they were based on ⁴. In mice infected with a resistant strain, two optimised peptides did better than their two originals ⁴. That work, too, is so far limited to lab and animal tests ⁴.

For researchers, the appeal is focus: in the prionin study, the model narrowed 19.3 million fragments to a shortlist, and 59 of the 75 peptides made from it showed activity in the lab ¹.

Keeping it in perspective

The authors are careful about what the study does not show. It does not prove that prionins are naturally released during infection, or that prion proteins act as antibiotics in the body ¹. It also does not change what is known about the harmful role of misfolded prions in brain disease ¹.

There are also known hurdles for the whole AMP field. The same features that let these peptides break bacterial membranes can harm human cells, and peptides are often broken down quickly in the body ⁵. Modified, more stable peptides can be expensive to make ⁵. Several AMPs that reached clinical trials have failed or been suspended ⁵. Prionins will need to clear the same bars.

What to watch

• More animal models: whether prionins work beyond a single topical skin-infection model ³.
• Safety testing: broader toxicity studies to confirm the early signs of selectivity for bacteria over human cells ².
• AI-designed versions: whether optimisation tools like ApexGo, so far applied to other peptides, could be used to improve prionins ⁴.

← All articles · News updates