Computational structure prediction of lanthipeptides with NMR data reveals underappreciated peptide flexibility.

Abstract

Lanthipeptides are a class of thioether-containing ribosomally synthesized and post-translationally modified peptides, which often have antibiotic activity. As a potential starting point for therapeutics, interest in engineering lanthipeptides is growing. Our inability to computationally model and design lanthipeptides in molecular modeling and design software such as Rosetta limits our ability to rationally design lanthipeptides for drug discovery campaigns. We propose that implementing support for the lanthionine rings and dehydrated amino acids found in lanthipeptides will enable accurate lanthipeptide modeling with Rosetta. We find that when compared to the ensembles of lanthipeptides with NMR-determined structures in the PDB, lanthipeptide ensembles generated with Rosetta have similar experimental agreement, lower Rosetta energy scores, and greater flexibility. Our use of ensemble-averaged NOE distances instead of requiring individual structures to satisfy all NOE restraints was key for revealing the flexibility of these peptides. Our Rosetta lanthipeptide ensembles show increased flexibility in non-cyclized peptide regions as well as increased lanthionine ring flexibility when internal hydrogen bonds are absent and glycine residues are present. Support for lanthipeptides in Rosetta enables the design and modeling of lanthipeptides in Rosetta for therapeutic development.