Deep Learning Uncovers 1,179 Prionin Antimicrobial Peptides, Reducing Bacterial Infection in Mice

Traditionally, prion and prion-like proteins are associated with protein misfolding diseases and neurodegeneration. However, amyloidogenic sequences, which are characteristic features of prions, also play a crucial role in host defense mechanisms, forming functional amyloids that combat pathogens. The challenge lies in systematically identifying these beneficial antimicrobial sequences hidden within the vast landscape of prion-related proteins. This study addresses this gap by leveraging advanced computational methods to discover novel antimicrobial peptides (AMPs) from an unexpected source.

Researchers employed a deep learning approach to screen 19.3 million peptide fragments derived from 2,897 curated prion-related proteins. This computational screen aimed to identify sequences with high potential for antimicrobial activity, which they termed prionins. From the predicted candidates, 75 prionins were selected for synthesis and subsequent experimental validation. These synthesized peptides were tested for their ability to inhibit bacterial pathogens, perturb bacterial membranes, and, for a subset, reduce infection burden in vivo in a mouse model of Acinetobacter baumannii infection.

The deep learning model successfully identified 1,179 candidate antimicrobial peptides, or prionins, from the vast dataset of prion-related protein fragments. Experimental validation of 75 synthesized prionins revealed significant antimicrobial activity: 59 of these peptides effectively inhibited various bacterial pathogens. Furthermore, 53 of the tested prionins demonstrated membrane-perturbing capabilities, a common mechanism for many antimicrobial peptides. The most compelling finding came from in vivo testing: > 2 of the synthesized prionins significantly reduced the Acinetobacter baumannii infection burden in mice, providing direct evidence of their therapeutic potential in a living system.