Our Take
This is a genuine discovery platform win: AI screened millions of fragments, synthesized candidates, and validated them in vivo. But it remains early-stage; no pathway to clinic is established, and whether these peptides function naturally in host defense remains unknown.
Why it matters
Antibiotic resistance is an urgent clinical challenge, and prions have never been systematically mined for antimicrobial activity. This work demonstrates that AI can find therapeutic peptides in protein families previously studied only for their role in disease, expanding where drug hunters should look.
Do this week
Antimicrobial peptide researchers: map your existing AMP screens against the APEX 1.1 model and the prionin dataset (published in Nature Microbiology) to identify overlaps and gaps in your current coverage before planning next-generation in vitro assays.
University of Pennsylvania researchers screen prion proteins for hidden antibiotics
Researchers at the University of Pennsylvania used a deep learning platform called APEX 1.1 to search 19.3 million short protein fragments derived from 2,897 prion and prion-like proteins. The scan identified 1,179 candidate antimicrobial peptides, which the team named "prionins."
The researchers then synthesized 75 of these candidates and tested them against clinically relevant bacterial pathogens, including multidrug-resistant strains. Fifty-nine inhibited at least one pathogen. Forty-two showed potent activity at concentrations of 16 micromolar or lower. Many active prionins worked by damaging bacterial membranes, a common mechanism for antimicrobial peptides. Sixteen active peptides showed no measurable hemolysis or cytotoxicity at the highest concentrations tested (per the published paper in Nature Microbiology).
Two of the strongest candidates were evaluated in a mouse skin-infection model caused by Acinetobacter baumannii, a difficult-to-treat, multidrug-resistant pathogen. A single topical dose of each peptide significantly reduced bacterial burden, with efficacy comparable to polymyxin B in the model tested. No treatment-associated weight loss was observed.
The work expands the University of Pennsylvania's broader effort to discover "encrypted peptides"—hidden sequences within larger proteins that can have biological function when isolated. Previous work from the same lab has mined human proteins, extinct organisms, archaea, microbiomes, and venoms.
Prions were never considered a source for antibiotic discovery until now
Prions have a singular reputation in medicine: they are the misfolded proteins responsible for rare, fatal neurodegenerative diseases like Creutzfeldt-Jakob disease. No systematic search for antimicrobial activity within prion-related proteins had been conducted before this study.
The discovery works because it answers a different question. Instead of asking "where do antibiotics usually come from," the researchers asked "could biology have hidden antimicrobial molecules in protein families we associate with disease?" The answer, supported by in vivo data, is yes.
The finding also raises a speculative but significant possibility: prion and prion-like proteins may contain cryptic antimicrobial sequences that contribute to host defense in specific biological contexts. The current study does not establish that these peptides are naturally released during infection or function physiologically in immune response, but it opens that door.
For antibiotic discovery specifically, this matters because traditional natural-source screening (soil microbes, marine organisms) has been depleted by decades of mining. AI-driven searches across unexpected protein families offer a new search space at a time when resistance to existing antibiotics is accelerating.
Validate prionin candidates in your own infection models before committing resources to optimization
The published dataset includes the full list of 1,179 prionin candidates and their sequences. If you work in antimicrobial peptide development, structure-activity relationship analysis, or translational infection research, download the Nature Microbiology supplementary data and cross-reference your existing pathogen panel against the two validated leads (the ones tested in mice). Check whether your assay conditions (pH, ionic strength, serum concentration) align with the Penn team's in vitro protocols before claiming negative results.
The gap between in vitro screening and in vivo efficacy is substantial. Two candidates out of 59 active in the lab proved efficacious in the animal model. That is a real hit rate for early-stage antimicrobial discovery, but it signals that in vitro potency does not predict in vivo success. If you are designing a follow-up study, build in a dose-response curve and kinetic clearance work before scaling synthesis.