Page 241 - Read Online

Raevsky et al. Neuroimmunol Neuroinflammation 2018;5:33 I http://dx.doi.org/10.20517/2347-8659.2018.34 Page 3 of 10

In this study, we describe an in silico model developed for identification of pharmacophores capable of in-
creasing the biological activity of Au(I)-based drug-like compounds. Previously, it was shown that changes
in steric and electronic properties of phosphine derivatives led to increased affinity of these compounds
[31]
toward cathepsin B . The aim of the present study was to develop and test novel structural modifications
of cathepsin B inhibitors, which could be used for synthesis of new, potentially more effective anti-neuro-
inflammatory drugs. To achieve this goal, we first developed an in silico docking model of the cathepsin
B enzymatic pocket. Subsequently, a series of novel cathepsin B inhibitors were designed and synthesized,
based on their calculated binding affinity to the enzymatic pocket and docking scores. An in vitro testing of
selected compounds as possible cathepsin B inhibitors was also performed.

METHODS
In silico modeling
All manipulations with protein-ligand structures and generation of structural models of cathepsin-ligand
complexes were performed with Sybyl-X software (Tripos Inc., St. Louis, MO, USA). The three-dimensional
structure of the triethylphosphine was generated by the CONCORD version 3.0 software (CONCORD, St
Louis, MO, USA). Ligand topologies for molecular dynamics (MD) studies were calculated using the ante-
[32]
chamber module of AmberTools version 12 . The protein structures of cathepsins B and K (Protein Data
Bank Identifiers (PDB IDs): 1HUC and 2ATO) were obtained from the Research Collaboratory for Structural
[33]
Bioinformatics (RCSB, www.rcsb.org) . Structures of the complexes formed by cathepsin B interacting with
triethylphosphine, as well as cathepsin K with myochrysine (Au-thiomalate), were modeled in the experi-
ment. The subsequent steps of the relaxation strategy for the structural optimization of active site geometry
[34]
were carried out with Gromacs (version 4.5) software with implementation of Amber99 force field .
Due to the absence of defined atom-type and coordination bonding parameters for Au in both MD and
docking software, we utilized fragment-based docking methods for the model development. Replacement
of the Au atom with a CF group, which has four heavy atoms, allowed us to preserve both geometry and
3
distances between the ligand and C-alpha of the catalytic residue Cys29 (C29) in the cathepsin B molecule
[Figure 1]. As a result of this manipulation, a “joint-like” metal-based binding of Au was replaced with the
“anchor-like” binding mode of CF . Additional mutation of the catalytic cysteine (C29) to alanine (C29A)
3
provided more flexibility for the CF -containing ligands and avoided artificial interaction between the sulfur
3
atom of C29 and the CF group.
3
In order to remove any unfavorable steric clashes between the proteases and the inhibitor molecules, 1000
steps of steepest descent and 5000 steps of conjugate-gradient energy minimization were employed. The
complex between cathepsin B and triethylphosphine was solvated with a 12 Å radius water box centered at
the C29A. A center of mass (COM) pulling mode was applied during all steps of MD calculations to improve
the geometry of the P-CF -C-alpha (CA, of alanine) bridge. The optimized and equilibrated system was
3
used as the starting configuration for MD simulations spanning 500 ps. Such fast molecular dynamics was
enough to stabilize the group of constrained P-CF -CA atoms. Average root-mean-square deviation (RMSD)
3
values for the whole structure, including the active site and ligand atoms, were calculated from each of the
five picoseconds frames of the simulation. RMSD values, together with the structure alignment, confirmed
that conformation of the ligand and the active site of protein had not changed dramatically and were consid-
ered stable enough to continue modeling.

For the docking procedure, a specific binding site representation (namely Protomol) was generated based on
the relaxed protein-ligand complex. Protomol is an interaction map, implemented in the Surflex-Dock mod-
[35]
ule, which is based on the probing method used . Protomol model is very sensitive to input parameters;
therefore, to avoid mistakes in predictions, it was parameterized based on scores obtained through valida-
tion docking procedures. To keep the amino acid environment flexible during each docking run, the most