In this short communication, I for the first time experimentally (in vitro and in vivo) tested the method suggested by Alex T Müller et al. 12 to generate novel peptides with presumable antimicrobial activity. I slightly modified the input data for the neural network to use all 3,100 antimicrobial peptides (with a length of more than 7 amino acid residues) from the APD3 database (https://aps.unmc.edu/) 13. The neural network generated 198 sequences that I ran through various in silico screening systems, after which, based on my own subjective opinion, I selected 5 peptides for synthesis (PEP-36, PEP-38, PEP-136, PEP-137 and PEP-174).
The first stage of the experimental screening was the study of the minimum inhibitory concentrations of these peptides by the method of serial dilutions. PEP-38 was found to be active against K. aerogenes and K. pneumoniae. PEP-137 was more effective against Klebsiella spp., as well as against P. aeruginosa. The PEP-36, PEP-136 and PEP-174 peptides were not active against these bacteria. It should be noted, however, that one of the limitations of this work was that I did not study the effectiveness of concentrations above 8 µg/ml.
The data obtained prompted me to test the effectiveness of the PEP-38 and PEP-137 peptides in the simplest experimental model of sepsis in mice. Moreover, as an additional control, I randomly selected PEP-36 which is not active in vitro. It should be noted that I used a simple and easily reproducible sepsis model with a single injection of the studied peptides at a dose of 100 µg/mouse.
Surprisingly, the PEP-36 peptide was found to be the most effective – the animal survival rate was 66.7%. PEP-137 showed a survival rate of 50%. PEP-38 was proven to be ineffective.
In their work on LL-37, Guangshun Wang et al. using nuclear magnetic resonance structural analysis identified a short three-turn amphipathic helix rich in positively charged side chains, which helps to effectively compete for anionic phosphatidylglycerols in bacterial membranes. 23
The modeled spatial structures of the novel peptides are very similar to the LL-37 molecule. It seems likely that the helix-rich structure of the PEP-36, PEP-38 and PEP-137 peptides may be an important contributor to the demonstrated antimicrobial effect. It remains unclear why PEP-38 was active in vitro, but did not affect mortality in experimental sepsis; it may be necessary to investigate larger doses of this peptide and other routes of administration.
It is also unclear, on the other hand, why PEP-36 did not have antimicrobial activity in vitro, but was most effective in vivo. There can be two explanations: after entering a living organism, PEP-36 becomes somehow modified and acquires antimicrobial activity; or this peptide has some kind of immunomodulatory effect. To clarify the nature of the data obtained, more research is required. Earlier, in the P. aeruginosa-induced sepsis model, epinecidin-1 (Epi-1; by Epinephelus coioides) due to its antimicrobial and pronounced immunomodulatory effect has been shown to reduce mortality in mice 19. Epi-1 enhances the production of IgG antibodies by activating the Th2-cell response 24, reduces the level of tumor necrosis factor alpha by reducing the level of endotoxins 19.
The data obtained in this work require a wide range of additional experiments: the study of pharmacodynamics and pharmacokinetics, toxicity, as well as the study of the combined action of the obtained peptides with conventional antibiotics.
Thus, I conducted a simple and easily reproducible study with an experimental assessment of the feasibility of using the generative long-term memory recurrent neural network to generate novel peptides that demonstrate antimicrobial activity in vitro and reduce mortality in experimental sepsis in vivo. The peptides I obtained can be used to develop new antibacterial drugs for the treatment of infections caused by carbapenem-resistant gram-negative bacteria.