Simulation-based predictions of binding affinities of matrix metalloproteinase inhibitors

Inhibitory potencies of 28 hydroxamate derivatives to matrix metalloproteinase 9 (MMP-9) have been modeled using the Linear Response method that describes free energy of binding as a linear combination of ensemble averages of van der Waals, electrostatic, and cavity-related terms. Individual terms are differences in the respective quantities between the bound and free ligands, which were determined using two methods: a hybrid Monte Carlo method within the frame of Surface-generalized Born Continuum Solvation Model and Molecular Dynamics simulation. The MD simulations of ligands in the hydrated MMP-9 active site, as well as of free ligands in water were carried out for 200 ps. The time-averaged structures of the ligands bound with the protein and of the free ligands were collected at 5, 10, 25, 50, 100, and 200 ps. The best correlations between experimental and calculated binding energies, obtained using the training set containing 21 structurally diverse compounds, explained about 90% of experimental variance. The predicted binding affinities for the test set of 7 inhibitors were in good agreement with the experimental data (RMSE = 0.944). Similar RMSE values were observed with 200 random selections of the 7-member test set. The loss of polar and nonpolar solvent accessible surface areas upon binding was identified as the main phenomenon contributing to affinity, accompanied by the enhancement of van der Waals interactions upon binding.