Stepwise identification of potent antimicrobial peptides from human genome.

The increasing incidence of hospital acquired infections caused by antibiotic resistant pathogens has led to an increase in morbidity and mortality, finding alternative antibiotics unaffected by resistance mechanisms is fundamentally important for treating this problem. Naturally occurring proteins usually carry short peptide fragments that exhibit noticeable biological activity against a wide variety of microorganisms such as bacteria, fungi and protozoa. Traditional discovery of such antimicrobially active fragments (i.e. antimicrobial peptides, AMPs) from protein repertoire is either random or led by chance. Here, we report the use of a rational protocol that combines in silico prediction and in vitro assay to identify potential AMPs with high activity and low toxicity from the entire human genome. In the procedure, a three-step inference strategy is first proposed to perform genome-wide analysis to infer AMPs in a high-throughput manner. By employing this strategy we are able to screen more than one million peptide candidates generated from various human proteins, from which we identify four highly promising samples, and subsequently their antibacterial activity on five strains as well as cytotoxicity on human myoblasts are tested experimentally. As a consequence, two high-activity, low-toxicity peptides are discovered, which could be used as the structural basis to further develop new antibiotics. In addition, from 1491 known AMPs we also derive a quantitative measure called antibacterial propensity index (API) for 20 naturally occurring amino acids, which shows a significant allometric correlation with the theoretical minimal inhibitory concentration of putative peptides against Gram-positive and Gram-negative bacteria. This study may provide a proof-of-concept paradigm for the genome-wide discovery of novel antimicrobial peptides by using a combination of in silico and in vitro analyses.