(602a) Energy and Sequence Landscapes in Peptide Binding

Peptides have been targeted as major therapeutic platforms due to their high binding selectivity, low toxicity, and ability to be produced using both synthetic and recombinant technologies. Moreover, sophisticated display technologies have enabled systematic screening and identification of peptide inhibitors to specific protein targets, with now over some 500 peptide drug candidates in clinical trials or on the market. However, a molecular-level understanding of many of the basic mechanisms and driving forces involved in the binding of these peptides remains incomplete. Unlike rigid, small-molecule drugs and larger antibodies, peptide binding often involves substantial conformational rearrangement with a complex interplay of coupled association and refolding events and driving forces.

Here, we describe a simple lattice model that is able to address these basic interactions. Due to its simplicity, we are able to perform ?virtual screens? in silico that systematically consider all possible peptide sequences interacting with a given target protein binding site. This model also permits calculation of the free energy of association in a manner that takes into account both peptide refolding and direct interactions with the binding site. The model shows the emergence of several unexpected trends. For instance, binding sites with a low content of hydrophobic groups show an anticorrelation between binding affinity and specificity?good peptide binders also tend to be promiscuous?while highly hydrophobic sites show a correlation. Moreover, peptides that are good folders, i.e. fold to unique structures, tend to have lower affinity. However, the inclusion of very strong polar interactions (e.g., ion pairing) can invert these relationships. Our model provides a thermodynamic interpretation of the characteristics of interaction sites in known peptide-protein structures and the emergence of hydrophobic sequence motifs in peptide display screening efforts.

We also describe recent work in combining simple energy landscape models with all-atom folding simulations to predict bound structures of designed peptides. Using replica exchange molecular dynamics techniques with an accurate atomic force field, we generate an ensemble of peptide structures that are each docked to a target protein. Iterations of docking and structural relaxation using all-atom structures produce a large library of putative holo structures. We then project these onto a simple model of a binding landscape characterized by a single funnel and a prescribed amount of ?roughness.? By smoothing the landscape in this way, the model indicates which structure is closest to the global minimum, even if its energy is not the lowest among those generated.