Modified Potential Functions Result in Enhanced Predictions of a Protein Complex by All-Atom MD Simulations, Confirming a Step-wise Association Process for Native PPIs

Abstract

Native protein-protein interactions (PPIs) are the cornerstone for understanding the structure, dynamics and mechanisms of function of protein complexes. In this study, we investigate the association of the SAM domains of the EphA2 receptor and SHIP2 enzyme by performing a combined total of 48 s all-atom molecular dynamics (MD) simulations. While the native SAM heterodimer is only predicted at a low rate of 6.7% with the original CHARMM36 force field, the yield is increased to 16.7% and to 18.3% by scaling the vdW solute-solvent interactions (better fitting the solvation free energy of amino acid side chain analogues) and by an increase of vdW radius of guanidinium interactions, and thus a dramatic reduction of electrostatic interaction between Arg and Glu/Asn in CHARMM36m, respectively. These modifications effectively improve the overly sticky association of proteins, such as ubiquitin, using the original potential function. By analyzing the 25 native SAM complexes formed in the simulations, we find that their formation involves a pre-orientation guided by electrostatic interaction, consistent with an electrostatic steering mechanism. The complex could then transform to the native protein interaction surfaces directly from a well pre-orientated position ({Delta}interface-RMSD < 5[A]). In other cases, modest (< 90{degrees}) orientational and/or translational adjustments are needed (5 [A] <{Delta}i-RMSD <10 [A]) to the native complex. Although the tendency for non-native complexes to dissociate has nearly doubled with the modified potential functions, a re-association to the correct complex structure is still rare. Instead a most non-native complexes are undergoing configurational changes/surface searching, which do not lead to native structures on a timescale of 250 ns. These observations provide a rich picture on mechanisms of protein-protein complex formation, and suggest that computational predictions of native complex protein-protein interactions could be improved further.