Protein–protein interactions are involved in a wide array of biological processes and diseases, which makes peptides an attractive therapeutic prospect. In order to rigidify the peptide backbone and attain better binding affinity and specificity, backbone cyclization is often employed. Cyclic peptides present themselves as logical candidates for mimicking "hot loop" interfaces of protein–protein interactions. We propose a novel protocol to identify the most promising candidates for cyclic peptide mimicry, and determine optimal cyclic peptide sizes for this purpose. The proposed approach relies on molecular dynamics (MD) simulations of polyglycine model systems. An all-glycine backbone maximizes flexibility and is as such most likely to sample a variety of conformations accessible to cyclic peptides. We performed conventional and enhanced-sampling MD simulations of these systems and compared the values of the backbone dihedral angles φ and ψ in the trajectories to the values of these dihedrals in 193 best-candidate, all-trans hot loops. This approach identified several “obvious” candidate hot loops. For a real-world demonstration, we chose the hot loop with the sequence EASESLCG, associated with the TACE–TIMP-3 interaction, with the last two residues truncated due to a ΔΔG of zero. The optimal cyclic peptide mimic length was found to be 9, and simulations of cyclo-EASESLGGG were subsequently performed. The resulting trajectories were subjected to dihedral principal component analysis and clustered based on local density peaks. The desired geometry was observed in 1.98% of frames, an increase from all-glycine simulations. Modifying the linker sequence from GGG to aGG and aGd further increased the population. Moreover, in order to more effectively analyze the trajectory data and allow for the inclusion of kinetic information in the clustering process, we propose an alternative clustering algorithm, which uses change point detection in the pre-processing stage and treats trajectory segments as distributions to be clustered using an inter-distribution distance metric. The application of the novel algorithm to simple model systems, as well as cyclic peptides, yields similar structural states as the state of the art, but with more physical boundaries and fewer false transitions between states. In union with the screening process described herein, this approach makes for a streamlined, intuitive, and effective inhibitor design and evaluation process.

Computational screening of cyclic peptide inhibitor designs using model system molecular dynamics simulations

ACS Spring Meeting 2021

The official archive and repository for all ACS Meetings & Expositions abstracts and select posters or presentations.