3.1. Lead optimization procedure
The lead/hit optimization procedure used in this study was previously reported by Ogata
et al [
33] and is only briefly summarized here. The first step consists in extracting the atomic coordinates of the ligand’s heavy atoms (referred to as ‘geometry’) from a high resolution structure of the protein-ligand complex. The geometry is then divided into fragments which are grouped into three partial structures types:
rings,
linkers (defined as the fragments that connect rings), and
terminals (defined as other types of fragments). In addition, all the atoms in the geometry are classified according to their bond order types (sp3, sp2, etc ) and atomic species (CH
3, CH
2, CH, NH
2, NH etc). For example, consider the geometry X···Y···Z, in which X, Y, and Z represent the atoms in the geometry and ‘···’ is a generic representation of the bonds connecting the atoms. Replacing Y with a “=CH–” generates a chemically incomplete compound X=CH–Z for which “=” and “–” indicate a double and a single bond, respectively. Then, X should be assigned to an atomic species linked through a double bond to Y (ex. O= or CH2=). Similarly, Z should be assigned to an atomic species capable of linking to Y through a single bond (ex. –CH
3 or –NH
2). By assembling all possible combinations of these atomic species, four compounds are obtained: O=CH–CH
3, O=CH–NH
2, CH
2=CH–CH
3, and CH
2=CH–NH
2. For the work presented in this paper, eighteen atomic species were used (see
Table 6).
Table 6.
Atomic Chemotypes used in this study.
After assigning all possible combinations of atomic species to the native ligand’s core coordinates, all the partial structures are considered and bond order requirements are satisfied thereby generating the hit compound database. Compounds in the newly generated database have similar core geometries as the native ligand and each atomic position satisfies different chemically meaningful combinations.
In a second step, compounds from the newly generated database are subjected to two filters. The Rishton nonleadlikeness filter (to remove undesirable functional groups, see
Figure 6) [
34] and the Lipinski’s rule of five (compounds with more than five hydrogen-bond donors, more than 10 hydrogen-bond acceptors, molecular mass greater than 500 Da, logP values greater than 5, or more than 10 rotatable bonds are not desirable for orally active drugs) [
35]. From the remaining list of compounds, molecules with ring(s) and condensed ring structures were selected because known hits for the three target kinase proteins contain such structures. These molecules were treated as the final list of lead candidates and ranked based on a scoring function,
Score, evaluated for the protein-ligand complex. The scoring function comprises four empirical energy terms:
where,
Ev is the van deer Waals interaction energy,
Ee the electrostatic interaction energy,
Eh the hydrogen bond energy, and
Es the solvation energy.
Ev and
Ee were obtained using the AMBER force field with the GAFF parameter set. [
36]
Eh was defined as:
where
rH is the distance between the hydrogen and the heavy atom (H
…X, set at 2.0 Å for this study).
Es was computed for the bound and unbound states of the ligand, protein and the complex as:
where
Ai and
σi are the solvent-accessible surface area and the proportionality factor for the solvent-accessible surface area of atom
i, respectively [
37]. The free energy of
and
were calculated in the same manner. This method has been designed to provide a ranked list of compounds with better “drug-type” properties (more stable, druglikeness and synthesizable compounds) than other approaches.
Figure 6.
Nonleadlikness filter. The substituent types were extracted from Rishton work [
34]. The electrophilic functional groups shown here are the most common protein-reactive covalent-acting false positives in biochemical assays. Compounds with substituents shown in this figure were removed from our results.
Figure 6.
Nonleadlikness filter. The substituent types were extracted from Rishton work [
34]. The electrophilic functional groups shown here are the most common protein-reactive covalent-acting false positives in biochemical assays. Compounds with substituents shown in this figure were removed from our results.
3.2. Application to serine/threonine protein kinases
Our approach was tested using three
serine/threonine protein kinases as targets. The X-ray protein-ligand complex structures used in this study were: p38 MAP kinase/ 3-(4-fluorophenyl)-2-pyridin-4-yl-1
H-pyrrolo[3,2-b]pyridine-1-ol (
FPH) complex, p42 MAP Kinase (Erk2)/
N-benzyl-4-[4-(3-chlorophenyl)-1
H-pyrazol-3-yl]-1
H-pyrole-2-carboxamide (
33A) complex, and c-Jun N-terminal kinase 3 (JNK3)/
N-(3,4-dichlorophenyl)-4-hydroxy-1-methyl-2,2-dioxo- 1,2-dihydro-2lamda~6~-thieno[3,2-c][
1,
2]thiazine-3-carboxamide (in house code
Z1208) complex obtained from the Protein Data Bank (PDB) [
31] (see
Table 2). The three structures display different ligand binding modes, and feature differences in the electrostatic potentials at the ATP-binding site [
29,
30,
38,
39,
40]. In addition, inhibitory activity against the target proteins has been reported for series of compounds. These compounds were derived by small modifications (changing or adding substituents) of the native ligand structures and atom types [
29,
30,
38]. We used this data to validate the results of our calculations, which involves potential compound candidates with larger structural and chemical differences than the original authors considered in their study.
In preparing the input structures for our calculations, the following steps were performed (see
Figure 1): 1) all water molecules were removed from the original complexes’ PDB files; 2) In FPH, the hydroxyl group attached to the 5- and 6-membered condensed ring was replaced by a hydrogen atom because the modified compound has a larger number of similar compounds with experimentally demonstrated inhibitory activity than the native compound; 3) In FPH and Z1208, all the fluorine atoms were replaced by chlorine atoms as this replacement made the chemical synthesis easier and; 4) the thioamide group in Z1208 was replaced by an amide group for the same reason.
In JNK3, two ligands were used to create the input structure: the native Z1208 and a derivative of Z1208 that acts as an ATP hydrolysis inhibitor. The experimental data used for this analysis were the in-house X-ray crystal structure (2.1 Å resolution and R-factor= 24.6%, see
Figure 4) and the IC50 values of 22.8 μM and 9.6 μM for native Z1208 and Z1208-derived ATP hydrolysis inhibitor respectively.
3.3. Inhibition assay
To measure the inhibitory activity of the Z1208, we used the following assay system: adenosine triphosphate (ATP), phosphoenolpyruvate (PEP), nicotinamide adenine dinucleotide (NADH), and a solution mixture of pyruvate kinase and L-lactate dehydrogenase (PK-L-LDH) were purchased from Roche Diagnostics. Other reagents were purchased from Sigma-Aldrich. JNK3 was expressed and purified by the method of Xie
et al. [
41]. After a purification step, JNK3 was activated by GST-fused MKK7 and further purified against a glutathione-fixed column. Inhibitory activity was estimated by detecting the inhibition of ATP hydrolysis reaction monitored by the coupled reaction of NADH oxidation; a slightly modified method of Xie
et al. Experimental conditions were: 100 nM JNK3 in 50 mM Hepes, pH 7.6, 10 mM MgCl
2, 1 mM NADH, 90 mg/mL PK, 30 mg/mL L-LDH, 2 mM PEP, 200 mM ATP, and each concentration of compound under 1% DMSO. The conversion of NADH was measured by kinetic monitoring with SpectraMax 190 (Molecular Devices).