Efficient free energy and enhanced sampling calculations of protein-protein interactions

Modern molecular life sciences place a large emphasis on the complex interactions between proteins and corresponding networks. In the pharmaceutical sciences, the focus is shifted from small-molecule drugs to so-called biologicals, which may be complex protein systems. Computational descriptions of such interactions lead to insight at the molecular level and predictions of affinities between proteins open the way to the rational design of novel therapeutics. An accurate description of protein-protein interactions and the relevant free energy differences crucially depend on appropriate sampling of all relevant conformational states, both in the bound as well as in the unbound state of the binding partners. While relatively efficient computational tools have been described for the interactions between proteins and small molecules, the protein-protein interactions pose additional challenges due to the large diversity in amino acid sequences and the intrinsic flexibility of protein structures. Another challenge is posed by proteins of which parts seem to be intrinsically disordered. In the current proposal, an international, interdisciplinary team of researchers suggests to develop efficient free energy methods and enhanced sampling tools to compute the binding free energy for complex protein systems. As a model system, the 14-3-3 family of proteins and their interaction with tyrosine hydroxylase is selected. In the unbound state, the relevant region in tyrosine hydroxylase is intrinsically disordered, and the affinity for many different sequences is to be evaluated. NMR experiments will be supported by Hamiltonian and Temperature Replica Exchange Molecular Dynamics simulations to describe the conformational ensemble of the partner protein, while the third-power fitting / one-step perturbation method will be extended to develop an universal model to compute the free energy differences between amino acids, allowing for an efficient prediction of binding affinities. The binding process itself will be described using Hamiltonian replica exchange calculations, in combination with distance field distance restraints. Overall, the developed methods will be applicable for a wide variety of protein-protein interactions and the enhanced sampling tools will allow for the calculation of complex potential of mean force profiles to describe the interaction between very flexible molecules.

The project aimed at the development of several computational and experimental methods to describe protein-protein interactions. The Austrian subproject focused on the development of computational methods, while the Czech subproject mostly focused on the experimental side, in particular NMR experiments. The computational methodology was developed and applied to a number of relevant model compounds, such as (1) the oligopeptide binding protein A, (2) a complex of ubiquitin with a ubiquitin-binding domain of human DNA polymerase and (3) a series of small peptides. In addition to these, and in particular in cooperation with the Czech team, the methods were applied to 14-3-3 proteins and their complexes with selected phosphorylated binding partners. Phosphorylation binding sites involved in the interaction with 14-3-3s are mostly located within intrinsically disordered regions of proteins. Within the Austrian subproject we have mainly used two computational approaches. In the first approach we assessed the strength of interactions between two proteins or a protein and a peptide directly, by computing the work that needs to be done to separate the two binding partners. This involved molecular dynamics simulations to describe the entire binding and unbinding pathway, and it was crucial that this was done in a complete and reversible way. We have applied the distancefield methodology that was developed in our group earlier, and further established a new approach to ensure complete sampling of the binding process. The computed binding affinities were in remarkable agreement with experimentally measured values. In the second approach, the effects of mutations on proteins and their binding partners were addressed by computing the free energy of mutating a specific amino acid within the protein. This allowed us to predict the stability of the 14-3-3 protein as well as the dimerization tendency of a suggested mutant. Furthermore, using this method, we could predict the relative binding strength of a number of small peptides to the oligopeptide binding protein. In this protein, water molecules play a particular role to facilitate the binding. We have benchmarked a number of methods for their ability to deal with the appearance or disappearance of water molecules, during such calculations. Finally, we have established a method to efficiently compute the effect of many mutations at a single position of the protein, using a combination of the one-step perturbation and third-power fitting methods. Remarkable agreement with much more time consuming simulations was obtained, suggesting that this method can be used in future applications of computational saturation mutagenesis.