Predicting protein ligand binding motions with the conformation explorer

BMC Bioinformatics Predicting protein ligand binding motions with the conformation explorer Samuel C Flores 0 Mark B Gerstein 1 2 0 Department of Cell and Molecular Biology, Uppsala University , BMC Box 596, Uppsala, 75124 , Sweden 1 Department of Computer Science, Yale University , PO Box 208114 MBB, New Haven, CT, 06520 , USA 2 Department of Molecular Biophysics and Biochemistry, Yale University , PO Box 208114 MBB, New Haven, CT, 06520 , USA Background: Knowledge of the structure of proteins bound to known or potential ligands is crucial for biological understanding and drug design. Often the 3D structure of the protein is available in some conformation, but binding the ligand of interest may involve a large scale conformational change which is difficult to predict with existing methods. Results: We describe how to generate ligand binding conformations of proteins that move by hinge bending, the largest class of motions. First, we predict the location of the hinge between domains. Second, we apply an Euler rotation to one of the domains about the hinge point. Third, we compute a short-time dynamical trajectory using Molecular Dynamics to equilibrate the protein and ligand and correct unnatural atomic positions. Fourth, we score the generated structures using a novel fitness function which favors closed or holo structures. By iterating the second through fourth steps we systematically minimize the fitness function, thus predicting the conformational change required for small ligand binding for five well studied proteins. Conclusions: We demonstrate that the method in most cases successfully predicts the holo conformation given only an apo structure. - Background Conformational changes in proteins can take place in a wide variety of ways, not all of which have been formally classified. One important class of motions is shear, in which stacked side chains of the protein can slide without losing contact. In this work we focus on the largest class, domain hinge bending, in which one structural domain of the protein moves relative to another domain about a hinge which connects the two [1,2]. Such motions typically involve the slowest degrees of freedom of that protein and so are difficult to predict by existing methods. The prediction of ligand binding motions of the protein receptor has considerable potential applications in protein-protein and protein-ligand docking. Many methods can predict the side chain rearrangements required for docking [3,4] but these assume that the large scale conformation is already nearly correct. Thus there is a need for a method that will put the receptor in the correct large scale conformation which can be a productive starting point [5]. Much work has been done in this area. Molecular Dynamics (MD) [6-9] explicitly computes the dynamical trajectory of molecules modeled as point masses connected by linear and nonlinear springs and can be used to predict conformational change, but usually only small- or moderate-scale domain motions can be reproduced [10] with many biologically relevant motions remaining out of reach [11]. Accordingly several methods used MD to account for the fast fluctuations of proteins in drug docking by first computing the protein trajectory using MD [4,12,13]. One limitation of such techniques is that they may not escape the vicinity of an initial conformation, even in a time span experimentally known to be sufficient for conformational change [14]. Althaus et al created a combinatorial tree of side-chain rotamers which they explored using a branch-and-cut algorithm, [15] without varying the backbone conformation. Sandak et al. created a flexible-receptor docking code which articulates the protein at a hinge point, but leaves the two resulting domains rigid [16]. This method suffered from the opposite problem: it could generate large scale protein motions, but had no way of dealing with even small side chain rearrangements, a weakness leading to failure [15]. The described methods are good at either treating the side-chain flexibility, or the large scale conformational changes, but not both simultaneously. Conformation Explorer uses Sandak et al.s idea of moving domains about a hinge point to generate large scale conformational change, but also includes equilibration steps which permit relaxation and adjustment of all atoms. Normal modes have also been used by many authors to predict the conformational changes of proteins [17]. Comparison of the atomic coordinates of homologous pairs of proteins shows that the lowest order modes are most involved in conformational change, [18,19] but also that multiple modes are needed to accurately represent the motion [20]. It is possible to determine the correct combination of normal modes that will reproduce a desired motion, but this requires knowledge of at least a few interatomic distance constraints for the final structure [21]. In a different approach, a docked protein-ligand complex was displaced along the lowest-frequency normal mode directions to minimize non-bonded energy terms in an MD force field [22-24]. However a normal mode expansion assumes a quadratic potential and so is accurate only for small fluctuations about an equilibrium structure; therefore the method cannot be used to predict larger scale conformational changes such as we treat in this work. The method of Lindahl et al. gains improvements of 0.3 to 3.2 for several proteins; [22] our method recapitulates much larger conformational changes as we will show. Maiorov and Abagyan [25] rigidified all protein bonds except those in the interdomain linker and interface using Internal Coordinate Modeling, and then used the Biased Probability Monte Carlo protocol to generate potential alternate conformations of the protein. The method succeeded in generating a large number of alternate conformations, and some of these were somewhat similar to alternate conformations known crystallographically. However without referring to the known alternate conformations, it was impossible to determine which of the many predicted structures was thermodynamically plausible. Further, many energy evaluations and minimizations were expended in evaluating generated conformers which were later discarded. Lastly, it was not easy to know how long a thorough exploration of conformation space would take, and no clear way to restrict the search to a given region of interest. Our method is similar in several ways to Maiorov et al.s, but also addresses these limitations. In more recent work, de Groot et al. [26] showed they could find the holo conformations of several ligandbinding proteins. The method relies on tCONCOORD, [5] which determines flexible regions by analyzing hydrogen bonding networks. Once these are known, an ensemble of plausible structures is generated. An interative process involving docking, MD refinement, and filtering by radius of gyration then generates holo structures. However the radius of gyration must be pro (...truncated)