Carborane clusters in computational drug design: a comparative docking evaluation using AutoDock, FlexX, Glide, and Surflex

Abstract

Compounds containing boron atoms play increasingly important roles in the therapy and diagnosis of various diseases, particularly cancer. However, computational drug design of boron-containing therapeutics and diagnostics is hampered by the fact that many software packages used for this purpose lack parameters for all or part of the various types of boron atoms. In the present paper, we describe simple and efficient strategies to overcome this problem, which are based on the replacement of boron atom types with carbon atom types. The developed methods were validated by docking closo- and nido-carboranyl antifolates into the active site of a human dihydrofolate reductase (hDHFR) using AutoDock, Glide, FlexX, and Surflex and comparing the obtained docking poses with the poses of their counterparts in the original hDHFR-carboranyl antifolate crystal structures. Under optimized conditions, AutoDock and Glide were equally good in docking of the closo-carboranyl antifolates followed by Surflex and FlexX, whereas Autodock, Glide, and Surflex proved to be comparably efficient in the docking of nido-carboranyl antifolates followed by FlexX. Differences in geometries and partial atom charges in the structures of the carboranyl antifolates resulting from different data sources and/or optimization methods did not impact the docking performances of AutoDock or Glide significantly. Binding energies predicted by all four programs were in accordance with experimental data.

Figure 1

Trimethoprim and carboranyl antifolates 1 and 2. The positioning of the “extra hydrogen” between B9-B10 corresponds to that in 2 N-crys (see Ligand Construction and Optimization for further information)

Figure 2

Binding of 1C-prot with various active site amino acid residues of hDHFR. The 2N-prot enantiomers have similar binding interactions in the hDHFR active site.

Figure 3

(A) Overlap of the crystal structures poses of 1C-prot (red) and 2 N-prot (green). (B) Overlap of the docked pose (AutoDock) of 1C-con (magenta) with the corresponding pose of 1C-prot (cyan) in the original crystal structure (avg. RMSD: 0.666 Å) (C) Overlap of the docked pose (AutoDock, major cluster with 68 solutions) of 2 N-con (magenta) with the corresponding pose of 2 N-prot (cyan) in the original crystal structure (avg. RMSD: 0.499 Å) (D) Overlap of the docked pose (AutoDock) of 2 N-con (magenta) with the corresponding pose of 2 N-prot (cyan) in the original crystal structure. The avg. RMSD of the minor cluster (32 poses) is 5.22 Å.

Figure 4

Average total binding energies of docked 1 C-crys (black bar) and 2 N-crys (grey bar) with hDHFR. For AutoDock and Glide, Mulliken charges at AM1 level were considered for 1C-crys and 2 N-crys. For Surflex, FlexX and Glide, the avg. binding energies for top 25% poses was considered. For AutoDock, the avg. binding energies of all poses were considered.

Figure 5

Comparative docking evaluation of 1 C-crys (A) and 2 N-crys (B). The black bars indicate the percentage of the total poses with RMSD below 2Å and the grey bars indicate the percentage of the total poses with RMSDs below 4 Å. Data used for this presentation are based on Mulliken charges obtained at HF/6-31+G* level for AutoDock and Glide, rigid docking mode for Glide, and BFP of PP with PA1 for FlexX. For Surflex docking, default settings without fragment placement were used for 1 C-crys and PP placement combined with multistart 5 option for 2 N-crys.