Background: Given the rising number of multidrug-resistant (MDR) bacteria, there is a need to design synthetic antimicrobial peptides (AMPs) that are highly active, non-hemolytic, and highly soluble. Machine learning tools allow the straightforward in silico identification of non-hemolytic antimicrobial peptides.
Methods: Here, we utilized a number of these tools to rank the best peptides from two libraries comprised of: 1) a total of 8192 peptides with sequence bhxxbhbGAL, where b is the basic amino acid R or K, h is a hydrophobic amino acid, i.e. G, A, L, F, I, V, Y, or W and x is Q, S, A, or V; and 2) a total of 512 peptides with sequence RWhxbhRGWL, where b and h are as for the first library and x is Q, S, A, or G. The top 100 sequences from each library, as well as 10 peptides predicted to be active but hemolytic (for a total of 220 peptides), were SPOT synthesized and their IC50 values were determined against S. aureus USA 300 (MRSA).
Results: Of these, 6 AMPs with low IC50's were characterized further in terms of: MICs against MRSA, E. faecalis, K. pneumoniae, E.coli and P. aeruginosa; RBC lysis; secondary structure in mammalian and bacterial model membranes; and activity against cancer cell lines HepG2, CHO, and PC-3.
Conclusion: Overall, the approach yielded a large family of active antimicrobial peptides with high solubility and low red blood cell toxicity. It also provides a framework for future designs and improved machine learning tools.
Keywords: Arg/Trp-rich peptides; antimicrobial peptides; host defense peptides; machine learning; peptide sequence design.
Copyright © 2024 The Author(s). Published by Elsevier Ltd.. All rights reserved.