High-Resolution Tuning of Non-Natural and Cyclic Peptide Folding Landscapes against NMR Measurements Using Markov Models and Bayesian Inference of Conformational Populations.

The rational design of stable and highly preorganized non-native and/or cyclic peptides is a challenging task that requires atomically detailed insight into folded and unfolded conformational ensembles. In this work, we demonstrate how Markov models constructed from collections of simulated trajectories using general-purpose force fields can be reweighted against NMR measurements to produce accurate folding landscapes. Here, we model the folding landscapes of 12 linear and cyclic peptide β hairpin mimics studied by the Erdelyi group, with the goal of reproducing the effects of subtle chemical modifications on peptide folding stability. The Bayesian Inference of Conformational Populations (BICePs) algorithm was first used to refine Karplus parameters to obtain an optimal forward model for scalar coupling constants; then, BICePs was used to reweight conformational ensembles against experimental NMR observables (NOE distances, chemical shifts, and scalar couplings). Before reweighting, Markov models of the folding dynamics reasonably capture the key features of the folding landscape. Only after reweighting, however, do we obtain folding landscapes that agree with experimental trends. Compared to previous estimates of folded populations made using the NAMFIS algorithm, BICePs-reweighted landscapes predict that the introduction of a side chain hydrogen- or halogen-bonding group changes the folding stability by no more than 2 kJ mol. The overall agreement between simulated and experimental NMR observables suggests that our approach is highly robust, offering a reliable pathway for designing foldable non-natural and cyclic peptides.