Proteolytic processing of precursor proteins liberates bioactive fragments, yet the systematic discovery of such peptides remains challenging. Here, we establish a fragment-mining strategy that combines in silico prediction with chemical tailoring to generate self-assembling antimicrobial peptides directly from protein sequences, including previously unannotated proteins. Self-assembly proved integral to activity: amphiphilic modification promoted nanonet formation and efficient bacterial membrane disruption, while rational glycosylation reprogrammed physicochemical properties to confer selectivity for bacterial phosphatidylglycerol over mammalian phosphatidylcholine, thereby broadening the therapeutic window. The optimized peptide, C16-KA6-Glc, displayed broad-spectrum bactericidal activity, superior biofilm eradication, and negligible resistance development. In murine models of thigh and pneumonia infection, C16-KA6-Glc achieved therapeutic efficacy comparable to that of conventional antibiotics while also attenuating inflammatory responses. These findings demonstrate a generalizable approach to translating latent proteome fragments into de novo self-assembling peptide therapeutics with clinical potential.
Liu et al. (Thu,) studied this question.