Computational methods

The crystal structures were taken from the Protein Data Bank for FKBP (pdb entry: lfkf [6]), and stromelysin (pdb entry: lsln [7]). The proteins were prepared for docking by adding hydrogen atoms and optimizing their positions in the presence of the co-crystal ligand. All titratable residues were assigned protonation states consistent with neutral pH. Water molecules in the co-crystal structures that exposed more than 1 A2 of surface area to other water molecules or to solvent accessible regions were removed. This left five and eight water molecules in the FKBP and stromelysin structures, respectively, which were retained in the docking simulations.

Using the co-crystal structure of the protein-ligand complex, a binding site was defined for each of the complexes by an orthorhombic cell. The dimensions of the cell were initially calculated to include the entire co-crystal ligand within the cell. A 5-A buffer was also added to each dimension of the cell. These binding sites encompassed significant portions of the proteins given the large size of the co-crystal ligands. Docking was confined to these binding sites. Making the assumption that the protein remains fixed at its initial coordinates and the force field is pairwise decomposable, the field of the protein can be pre-calculated on a collection of three-dimensional grids which completely cover the binding site. A full description of the grid calculations

Figure 3. Stromelysin inhibitors used in the docking study.

and the method for extracting the energy for an arbitrary molecule located anywhere in the binding site can be found elsewhere [8].

The intermolecular energy was calculated using the Amber 94 force field plus a desolvation term. The desolvation term is based upon the simple Gaussian functional form proposed by Stouten et al. [9]. Each atom is assigned a solvation parameter based on the affinity of the atom for solvent and a fragmental volume based on the amount of water the atom excludes from solvating the surrounding atoms. Here we assumed that the solvation parameter was proportional to the square of the charge of the atom plus a small negative constant that is related to the hydrophobicity of a neutral atom. The fragmental volume was estimated as the volume of a sphere with a radius equal to the minimum in the van der Waals interaction potential for the given atom interacting with an atom identical to itself. This function was optimized by comparing desolvation energies calculated with this method for a large number of protein-ligand complexes to desolvation energies calculated using a Finite-Difference Poisson-Boltzmann method [10]. The intramolecular energy of the ligand was estimated using the Dreiding generic force field [11]. This force field describes torsions using generic atom types, which is useful when screening a database that contains functional groups not found in most protein force fields. Other internal degrees of freedom were not sampled for the ligand. The total energy was the sum of the Amber intermolecular energy, the solvation energy, and the Dreiding intramolecular energy without any scaling of any individual term.

The charges for the docked ligands were calculated using the MNDO [12] Hamiltonian within MOPAC 7.0 [13] and were scaled by the factor 1.18 if the ligand was neutral. The atomic charges for the protein atoms were taken from the Amber 94 [ 14] force field, except for the charges in the stromelysin active site which contains a zinc. MNDO charges were obtained for the stromelysin active site from a smaller zinc-centered model that consisted of the metal ion and three 5-methyl imidizole molecules. Lennard-Jones parameters for all atoms were taken from the Amber 94 [14] parameter set.

The docking was performed on a potential surface of the protein which had been smoothed by using the soft core technique suggested by Zacharias et al. [15] and Beutler et al. [16] in a slightly different context. The smoothing of the potential surface removes high interconversion barriers and allows slight overlap of protein and ligand atoms. This improves search performance. The functional form used was similar to that suggested by Beutler et al. [16] except a constant value of 2.75 Á was used for softening the Lennard-Jones potential and a constant value of 1.75 Á, was used for softening the electrostatic potential. After docking, each structure was relaxed to a local minimum, using the original (not softened) potential function.

All docking simulations were performed using the AGDOCK program [17]. The program uses an evolutionary programming algorithm to optimize the score of a population of proposed protein-ligand complexes over a number of generations. Position, orientation, and the internal torsion angles of the ligand are sampled with the algorithm.

The resulting docked structures were scored using the High Throughput Screening (HTS) program developed at Agouron. This program uses an empirical energy function that has been parameterized using experimental data (see the Appendix). Although this function rapidly estimates the binding affinity of a protein-ligand complex, it is not used in the docking study, because it is not as computationally efficient as the soft-core potential. Throughout the remainder of this manuscript, the energetics and structures given for the docked compounds are the orientations that were scored to give the most favorable binding free energy using HTS .

Four sets of docking simulations were performed on each protein. For each simulation, the ligand was docked 200 times starting from a random conformation, orientation, and position in the binding site. First the ligands that were shown to bind in the primary site of the FKBP and the stromelysin were docked to their respective proteins. The ligands that were known to bind to the secondary site were docked in both the absence and the presence of the lowest energy structure of the primary ligands of FKBP and stromelysin. Finally, the composite inhibitors created from the linked lead compounds were also docked to their respective proteins.

0 0

Post a comment