Journal of Molecular Modeling

, Volume 16, Issue 12, pp 1919–1929

Modeling the binding modes of stilbene analogs to cyclooxygenase-2: a molecular docking study

Authors

    • Laboratoire de Physico-Chimie Théorique et Chimie Informatique, Faculté de Chimie
  • Amel Toumi-Maouche
    • Laboratoire de Physico-Chimie Théorique et Chimie Informatique, Faculté de Chimie
  • Boubekeur Maouche
    • Laboratoire de Physico-Chimie Théorique et Chimie Informatique, Faculté de Chimie
  • Safia Taïri-Kellou
    • Laboratoire de Physico-Chimie Théorique et Chimie Informatique, Faculté de Chimie
Original Paper

DOI: 10.1007/s00894-010-0679-7

Cite this article as:
Bouaziz-Terrachet, S., Toumi-Maouche, A., Maouche, B. et al. J Mol Model (2010) 16: 1919. doi:10.1007/s00894-010-0679-7

Abstract

Stilbene analogs are a new class of anti-inflammatory compounds that effectively inhibit COX-2, which is the major target in the treatment of inflammation and pain. In this study, docking simulations were conducted using AutoDock 4 software that focused on the binding of this class of compounds to COX-2 protein. Our aim was to better understand the structural and chemical features responsible for the recognition mechanism of these compounds, and to explore their binding modes of interaction at the active site by comparing them with COX-2 co-crystallized with SC-558. The docking results allowed us to provide a plausible explanation for the different binding affinities observed experimentally. These results show that important conserved residues, in particular Arg513, Phe518, Trp387, Leu352, Leu531 and Arg120, could be essential for the binding of the ligands to COX-2 protein. The quality of the docking model was estimated based on the binding energies of the studied compounds. A good correlation was obtained between experimental logAr values and the predicted binding energies of the studied compounds.

https://static-content.springer.com/image/art%3A10.1007%2Fs00894-010-0679-7/MediaObjects/894_2010_679_Figa_HTML.gif
Figure

The interaction of thirty stilbene analogs to COX-2 protein was studied using the automated docking simulation. The results of this study showed that the analogs can adopt two different binding modes in the active site of the COX-2 protein and that their predicted binding energies are in agreement with their experimental logAr values.

Keywords

Cyclooxygenase-2COX-2Molecular dockingAutoDockStilbenesNSAIDsHydrophobic interactions

Introduction

Cyclooxygenase (COX), also known as prostaglandin endoperoxide H synthase (PGHS), catalyzes the conversion of arachidonic acid (AA), its natural substrate, to prostaglandin (PGs). This enzyme exists in two isoforms: PGHS-1 (or COX-1) and PGHS-2 (or COX-2), which perform the same enzymatic activities [13]. These enzymes are approximately 60% identical in terms of amino acid composition, and their catalytic regions are widely conserved [46]. Furthermore, the two active sites of these isoforms differ by only two amino acids, at positions 523 (Ile for COX-1 and Val for COX-2) and 513 (His for COX-1 and Arg for COX-2) [7]. COX-1 is homeostatic. It is expressed in many tissues, including the gastrointestinal tract, kidney, and platelets, whereas COX-2 is expressed at sites of inflammation, the hippocampus, female reproductive tissue, and in many cancers [811]. COX exhibits activity towards both cyclooxygenase and peroxidase. The first involves the conversion of AA to prostaglandin G2 (PGG2), while the second catalyzes the transformation of PGG2 to prostaglandin H2 (PGH2), from which PGs, prostacyclin, and thromboxanes are derived.

Nonsteroidal anti-inflammatory drugs (NSAIDs) block the production of PGs by inhibiting both COX-1 and COX-2. Most of these drugs are associated with well-known side effects at the gastrointestinal level (mucosal damage, bleeding), and less frequently at the renal level. These drugs appear to produce at least some of their beneficial effects by inhibiting COX-2, and their deleterious side effects by inhibiting COX-1 [12]. Thus, selective inhibition of the induced enzyme, without affecting the homeostatic one, might avoid the side effects of currently available NSAIDs [12]. The differences between the active sites of COX-2 and COX-1 enzymes, which are described as hydrophobic channels, are the basis for the design of COX-2 selective inhibitors (coxibs). Since coxibs (celecoxib, rofecoxib, valdecoxib, etoricoxib, lumeracoxib) fail to inhibit the constitutive COX-1 isoform, they have no adverse gastrointestinal effects. However, recent publications have suggested that some coxibs on the market have been associated with adverse cardiovascular side effects [13]. Thus, there is a need to design new compounds with optimum COX-1/COX-2 selectivity. Indeed, there are already several classes of compounds that are known to display selectivity towards COX-2, such as oxazoles, pyrazoles, pyrroles, and imidazoles [1416]. The physiology and pathophysiology of COX continue to be topics of great interest [1720]. A large number of research studies aimed at finding selective COX–2 inhibitors have been reported [2127].

One of the research techniques employed to provide evidence of ligand binding sites on biological macromolecules is the computational docking method. This technique has been applied previously to the binding of several classes of inhibitors to the X-ray structure of PGHS [2830]. In this context, we are interested in the ligand–protein binding—which is important not only as an essential molecular recognition process but also for the discovery of new COX-2 inhibitors—between a series of stilbene analogs and COX-2 protein.

In the present study, we use the AutoDock 4 program to perform the molecular docking of stilbene analogs with COX-2 protein. Our objective is to search for favorable binding modes and to gain insight into the interaction between these ligands and COX-2 protein. As the result, the correlation between the experimental binding affinities of the studied compounds and the AutoDock binding energies is investigated.

Materials and methods

Dataset preparation for AutoDock

The dataset used for this study contains 30 stilbene analogs, selected in order to define a manageable dataset characterized by adequate biological and structural diversity [3134]. This dataset contains compounds that selectively inhibit COX-2, compounds that have binding affinity for both COX-1 and COX-2, and compounds that are inactive towards both enzymes. In order to simplify our analysis, we organized the 30 stilbene analogs into five groups according to the nature of their R1, R2, R3, and R4 substituents (Table 1). As their binding data were not obtained under the same experimental conditions, Table 1 provides the logarithms of the relative values, denoted Ar, which are calculated with respect to the same reference molecule: celecoxib, which is featured in all series arising from different sources [3134]. The binding affinities, measured in vitro, are given in terms of IC50, the drug concentration that inhibits 50% of the enzymatic activity of COX-2. Therefore, for each molecule, Ar = (IC50 of the molecule)/(IC50 of the celecoxib). The IC50 values of celecoxib according to different sources are listed in Table 2. The 2D structures of these compounds were drawn with standard geometric parameters using the molecular builder in ISIS/Draw as implemented in the ISIS 2.4 package [35], and they were converted to 3D format (.pdb) using ChemOffice 2003 [36], where geometrical optimization was performed with the MM2 method. The structures of the studied compounds, along with their IC50 and logAr values, are shown in Table 1 and Scheme 1.
https://static-content.springer.com/image/art%3A10.1007%2Fs00894-010-0679-7/MediaObjects/894_2010_679_Sch1_HTML.gif
Scheme 1

General structure of the stilbene analogs studied in this work

Table 1

Structures (see also Scheme 1) of the thirty stilbene analogs studied, the original IC50 values (in μM) and logAr

Group

Molecule

R1

R2

R3

R4

IC50

LogAr

1

C1

Ph

4-NHSO2Me

H

n-C6H13

0.030

-0.368

C2

Ph

4-NHSO2Me

H

n-C4H9

0.320

0.660

C3

Ph

4-NHSO2Me

H

Et

1.800

1.410

C4

Ph

4-NHSO2Me

H

Me

33.100

2.675

C5

Ph

4-NHSO2Me

H

Cyclohexyl

>100

>3.155

2

C6

Ph

4-(N3)

H

n-C6H13

0.110

0.196

C7

Ph

4-(N3)

H

n-C4H9

1.000

1.155

C8

Ph

4-(N3)

H

Et

0.280

0.602

3

C9

n-C4H9

4-SO2Me

H

Ph

0.014

-0.609

C10

n-C4H9

H

H

Ph

7.900

2.053

C11

n-C5H11

4-SO2Me

H

Ph

2.000

1.545

C12

n-C6H13

4-SO2Me

H

Ph

0.031

-0.264

C13

Et

4-SO2Me

H

Ph

1.250

1.341

C14

Me

4-SO2Me

H

Ph

0.630

1.043

C15

H

4-SO2Me

H

Ph

1.800

1.410

C16

Et

4-SO2Me

4-OH

Ph

1.900

1.434

4

C17

H

4-SO2Me

H

CO2H

3.000

1.632

C18

H

4-SO2Me

3-Br

CO2H

0.310

0.646

C19

H

4-SO2Me

4-Br

CO2H

3.600

1.711

C20

H

4-SO2Me

4-F

CO2H

36.000

2.711

C21

H

4-SO2Me

4-OH

CO2H

5.300

1.879

C22

H

4-SO2Me

4-OMe

CO2H

1.900

1.434

C23

H

4-SO2Me

4-OAc

CO2H

2.900

1.617

C24

H

4-SO2Me

4-NHAc

CO2H

2.500

1.553

5

C25

H

4-SO2Me

4-(2,4-F2–C6H3)

CO2H

0.320

0.660

C26

H

4-SO2Me

4-(4-i-PrO–C6H4)

CO2H

0.320

0.660

C27

H

4-SO2Me

3-(2,4-F2–C6H3)

CO2H

>100

>3.155

C28

H

4-SO2Me

3-(4-i-PrO–C6H4)

CO2H

>100

>3.155

C29

H

4-SO2Me

4-(4-MeO2S–C6H4)

CO2H

0.320

0.660

C30

H

4-SO2Me

3-(4-MeO2S–C6H4)

CO2H

2.100

1.477

Table 2

Values of IC50 (μM) for the reference molecule, celecoxib, according to different sources

Reference

Molecule

IC50 celecoxib

[31]

C1–C8

0.070

[32]

C9, C11, C12, C13, C14

0.057

[33]

C10, C15, C16

0.070

[34]

C17–C30

0.070

Receptor preparation

The protein coordinates used in this study correspond to the X-ray structure of COX-2 bound with SC-558 (code 6cox determined at 2.80 Å resolution), which was downloaded from the Protein Data Bank [6]. The ligand and water molecules were removed, and all missing hydrogens were added. The active site radius was taken to be 8 Å from the centroid of the ligand, so that the amino acid residues (namely Arg120, Gln192, His90, Leu352, Leu359, Phe518, Trp387, Ser353, Tyr355, Tyr385, Arg513, Val523 and Val349) could be included. After assigning Gasteiger charges, nonpolar hydrogens were merged with their corresponding carbons. These preparations were done for each ligand using the Chimera [37] and AutoDockTools [38] packages.

Molecular docking using AutoDock

The thirty stilbene analogs were employed as the ligands while the docking process was performed. A two-step docking protocol was employed. In the first step, each ligand, placed in a random position, was flexibly docked into the active site using the default parameter means of the AutoDock 4 program [39]. This first-step approach served only to accommodate the different ligands into the binding site. The lowest-energy conformations obtained for each compound were then used as input for the second docking step.

In the second step, a grid map with 48 × 44 × 58 points around each of the modeled domains was used with a grid-point spacing of 0.375 Å. As a result, 30 grid parameters were created.

The docking procedure involved the AutoDock 4 program and its Lamarckian genetic algorithm (LGA), a new and promising hybrid search technique that implements an adaptive global optimizer with local search [40]. For all dockings, 50 independent runs were used with step sizes of 0.01 Å for translations and 1.2° for orientations and torsions, 250,000 energy evaluations, a maximum number of generations of 27,000, an elitism value of 1, a mutation rate of 0.02, and a crossover rate of 0.8. The resulting positions were clustered according to the root mean square deviation (RMSD) criterion of 2.0 Å. Both Autogrid and AutoDock computations were performed on the Cygwin and Linux (Ubunto) platforms.

Results and discussion

Selection of the most relevant conformation

Automated docking of the 30 studied compounds to the COX-2 active site resulted in several plausible docking orientations and conformations for each ligand. The resulting conformations were clustered into conformation families according to the RMSD criterion of lower than 2 Å. For most molecules there were one or two of the most densely populated clusters. These were sorted by energy (from lowest to highest), and their heights were proportional to the number of conformations in each cluster. The selection of the most relevant conformation among the conformations of the most populated cluster was done according to hydrogen bonding, binding energy and low RMSD. When the clusters had similar populations, we chose the conformations similar to those of other molecules.

Comparison between the X-ray structure and the docking structure

The structure of SC-558, extracted from the crystallographic complex 6cox, was used to set the optimal docking parameters for this study. The resulting conformation was close to the crystal structure (as shown in Fig. 1), since the RMSD between the two conformations was only 0.89 Å, which is quite satisfactory. Therefore, the studied compounds docked correctly to the COX-2 active site, with only a slightly different conformation from the crystal conformation.
https://static-content.springer.com/image/art%3A10.1007%2Fs00894-010-0679-7/MediaObjects/894_2010_679_Fig1_HTML.gif
Fig. 1

Comparison between the X-ray structure and the docking structure for SC-558 (blue: docking, red: crystal structure). Magenta ribbon shows the protein backbone; only the amino acids within 5 Å of the docked ligand are presented. For the sake of clarity, two different viewing angles are provided

Binding modes of the studied compounds

The docking results revealed that almost all of the compounds bind at the same binding site in the active site of the COX-2 protein (as determined experimentally). Molecular surface visualization shows that the active site of COX-2 is composed mostly of hydrophobic residues. There are three major pockets (S1, S2, S3) and one shallow pocket (S4) on the COX-2 active site (Figs. 2a and 2b).
  1. Pocket S1:

    consists of the amino acid residues Ala516, Arg513, His90, Gln192 and the backbone of Phe518

     
  2. Pocket S2:

    localized at the top of the channel, and consists of several hydrophobic amino acid residues: Tyr348, Tyt385, Phe518, Met522 and Trp387

     
  3. Pocket S3:

    localized in the mouth of the active site, and is delimited by Leu359, Tyr355, Val116 and Arg120.

     
  4. Pocket S4:

    shallow pocket localized in a hydrophobic area close to the hydrophobic pocket S2; it has little residues, including Leu534, Leu531 and Ile345.

     
https://static-content.springer.com/image/art%3A10.1007%2Fs00894-010-0679-7/MediaObjects/894_2010_679_Fig2_HTML.gif
Fig. 2

a View of the surface of the binding site of COX-2. b View of the most favorable conformation of each compound (shown as a stick) docked into the COX-2 active site. The protein surface is colored according to electrostatic potential. Some residues are presented as lines to help visualize the COX-2 active site

Stilbene analogs can adopt two major conformations (A and B) in the COX-2 active site, which differ mainly in the 3D dispositions of the R1, R3 and R4 groups and the olefinic C=C in the COX-2 active site. Comparison of the binding modes of the studied compounds with the X-ray structure of SC-558 shows that mode A is indeed most closely related to that observed in the crystallographic complex 6cox (Fig. 3).
https://static-content.springer.com/image/art%3A10.1007%2Fs00894-010-0679-7/MediaObjects/894_2010_679_Fig3_HTML.gif
Fig. 3

Superposition of the most favorable binding mode of C1 (blue) and the X-ray structure of SC-558 (red). Magenta ribbon shows the protein backbone

For both binding modes A and B, the automated docking suggests that the C-1 and C-2 phenyl groups are located in the vicinity of residues Val523, Phe518, Trp387 and Leu352, leading to π–π stabilizing interactions with the aromatic side chains of Phe518 and Trp387. Moreover, the backbone CO groups of Leu352 and Phe518 are linked to the ligand via hydrogen bonding with the polar group of the C-1 phenyl group (R2 group).

In mode A, as shown in Fig. 4a, the olefinic C=C interacts with the hydrophobic residues Leu352 and Ala527. The residues Ser353 and Val523 are linked to the ligand via hydrophobic interactions with the C-1 phenyl group. The C-2 phenyl group is surrounded by Trp387, Tyr385, Phe518 and Val349, leading to hydrophobic interactions with these residues.
https://static-content.springer.com/image/art%3A10.1007%2Fs00894-010-0679-7/MediaObjects/894_2010_679_Fig4_HTML.gif
Fig. 4a–b

View of the two possible binding modes for C1 obtained by molecular docking: a binding mode A; b binding mode B. Hydrogen bonds and π–π interactions are shown as dotted black lines and transparent cylinders for C1 in the COX-2 active site. Magenta ribbon shows the protein backbone

In mode B, as shown in Fig. 4b, the olefinic C=C interacts with Leu352 and Val523, resulting in favorable hydrophobic interactions. The C-1 phenyl group is surrounded by Phe518, Leu352 and Val523 and it interacts hydrophobically with these residues. Leu531 and Ala527 are linked to the ligand via hydrophobic interactions with the C-2 phenyl group.

Representations of the selected binding modes of the studied compounds are given in Figs. 3, 4, 5 and 6. In all figures, for reasons of clarity, only the interacting residues of the active site are highlighted, while the hydrogen atoms are omitted. More detail regarding the binding modes is given in the following sections.
https://static-content.springer.com/image/art%3A10.1007%2Fs00894-010-0679-7/MediaObjects/894_2010_679_Fig5_HTML.gif
Fig. 5a–f

Superpositions of selected conformations of compounds: a C6 (red) and C8 (cyan); b C6 (red) and C7 (green); c C9 (blue) and C2 (red); d C18 (red) and C15 (green); e C25 (blue) and C26 (red); f C25 (blue) and C29 (red). Magenta ribbon shows the protein backbone

https://static-content.springer.com/image/art%3A10.1007%2Fs00894-010-0679-7/MediaObjects/894_2010_679_Fig6_HTML.gif
Fig. 6a–b

Views of the most favorable binding modes for a C9 and b C25 obtained by molecular docking. Hydrogen bonds are shown as dotted black lines for C9 and C25 in the COX-2 active site. π–π interactions are shown as transparent cylinders for C25 in the COX-2 active site. Magenta ribbon shows the protein backbone

Docked conformations of group 1 compounds

Analysis of the data for this group, which have a MeSO2NH at the para position on the C-1 phenyl ring, leads to top conformations, all of which are in binding mode A. In this binding mode, the NH of the p-MeSO2NH group of compound C1 is hydrogen bonded to the backbone CO of Leu352 (d = 1.971 Å). One of the O atoms of p-MeSO2 is hydrogen bonded to Gln192 (d = 2.252 Å) and the NH backbone of Ile517 (d = 3.159 Å), while the other O atom of p-MeSO2 is hydrogen bonded to the backbone CO of Phe518 (d = 2.870 Å). The C-2 n-hexyl group is close to Leu531, Val349, Ile345 and Leu359, as shown in Fig. 4a, resulting in aliphatic hydrophobic interactions with them. The C-1 phenyl ring, which is cis to the C-2 n-hexyl group, is oriented toward the mouth of the COX-2 binding site close to Tyr355 and Arg120, and is possibly involved with a cation–π interaction with the NH2 group of Arg120 (d = 4.023 Å). This interaction may have important COX-2 selectivity implications since it disrupts the salt bridge between Arg120 and Glu524 at the mouth of the COX-2 binding site [6, 41]. Figure 4b illustrates the hydrogen bond interactions of C1 in its binding mode B at the COX-2 active site. This figure shows that the NH of C1 forms one hydrogen bond with the backbone CO of Phe518, while the O atoms of p-MeSO2 form two hydrogen bonds with Gln192 and the backbone CO of Leu352.

When the alkyl chain length was decreased, the inhibitory potency of COX-2 decreased, as indicated by C2 (R4 = n-butyl), C3 (R4 = ethyl) and C4 (R4 = methyl). The selected conformations for these compounds are characterized by an absence of interactions with Leu359, Leu531, Val349 and Ile345. It is worth noting that the alkyl chain length is an important influence on binding affinity. The cyclohexane ring in C5 exerts more of a steric effect than the aliphatic groups in the pocket S4, and this pocket is required in new inhibitors.

Docked conformations of group 2 compounds

The replacement of MeSO2NH by its bioisostere, the linear dipolar p-azido group, as in compounds C6, C7 and C8, leads to binding mode A. In the selected conformations of these compounds, the p-azido group is surrounded by Arg513, Val523 and His90. The distance between the terminal nitrogen of the p-N3 group in these three conformations and the NH2 of Arg513 is less than 5 Å. This result is in agreement with a model previously proposed [42] for another class of COX-2 inhibitors that have a p-azido group in which the N3 group is involved in an electrostatic interaction with Arg513. The decrease in the binding potency of C6 compared to that of C1 can be explained by the lack of strong hydrogen bonds with Gln192 and the backbone CO of Phe518. The n-hexyl group of C6 is pointed toward Val344, Tyr348 and Val349. On the other hand, in C7 (R4 = n-butyl) and C8 (R4 = ethyl), the lack of hydrophobic interactions with Val344 and Tyr348, as shown in Figs. 5a and 5b, could explain the moderate activities of these compounds observed experimentally.

Replacing the MeSO2NH or N3 group by a MeSO2 group, as in C9C30, leads to the isopopulated binding modes A and B. Mode A is predominant in all compounds, except in those with a phenyl group at the para position on the C-2 phenyl group (C26C30).

Docked conformations of group 3 compounds

Experimental binding affinity is strongly increased by converting the MeSO2NH of C2 into MeSO2 and interchanging the R1 and R4 groups, as in C9. Figure 5c illustrates the comparative modes of docking of C9 and C2 into the COX-2 active site. However, the selected conformation for C9, the most potent COX-2 inhibitor among the selected compounds, is characterized by hydrogen bond interactions between the O atoms of p-MeSO2 and Arg513 (d = 3.005 Å), the NH of His90 (d = 2.143 Å) and Gln192 (d = 2.780 Å), as shown in Fig. 6a. Arg513 has been previously shown to be involved in a key hydrogen bond interaction with most selective COX-2 inhibitors [43]. Additionally, site-directed mutagenesis experiments identify this residue as being a critical determinant of COX-2 selectivity towards lipoamino acid N-arachidonylglycine (NAGly), which has been shown to suppress tonic inflammatory pain [44]. The C-1 n-butyl group is oriented towards Arg120 and Tyr355 and is involved in favorable hydrophobic interactions with these residues. The C-2 phenyl group which is trans to the C-1 p-MeSO2 phenyl group is surrounded by the hydrophobic residues Val349 and Leu531. The distance between the center of this ring and Arg120 is 5.8 Å. The higher affinity of the MeSO2, which is only capable of hydrogen bonding, of C9 in comparison to the NHSO2Me of C2 and the N3 of C7 may be attributed to the strong hydrogen bonding between an O atom of SO2Me and the NH of His90. It is interesting to note that replacing the SO2Me group by another group capable of hydrogen bonding (SO2) and electrostatic interactions (N3) with residues lining pocket S1 may increase the binding affinity of C9.

The decreased binding affinity of C10 compared to that of C9 may be due to the absence of the p-MeSO2 group in the C-1 phenyl ring of C10. Comparing C11 with C9, the n-pentyl group of C11 has a greater steric effect on Arg120 than the n-butyl group does, and so this compound has a weaker affinity for COX-2. On the other hand, the last methyl of the n-hexyl group of C12 is pointed toward Val116 and is involved in an aliphatic hydrophobic interaction with it. The lack of interactions between the R4 group and Leu531 and Val116 may explain the low experimental activities of C13 (R4 = ethyl), C14 (R4 = methyl) and C15 (R4 = H) towards COX-2 protein. The insertion of the OH group at the para position on the C-2 phenyl group of C13 decreases the enzyme-inhibiting potency, as seen in C16. However, the predicted bioactive conformation lacks the key interactions with residues Arg513 and Gln192. The lack of this hydrogen bond might explain the loss of activity observed experimentally, even if a hydrogen bond interaction between the backbone CO of Met522 and the OH group at the para position on the C-2 phenyl group is established.

Docked conformations of group 4 compounds

Replacing the unsubstituted C-2 phenyl group of C15 by COOH, as in C17, leads to a decrease in the binding affinity. Previous studies have shown the importance of ionic interactions between the NSAID COOH group and Arg120 and its critical role in COX-1 inhibition [41, 45]. On the other hand, it is interesting to note that the COOH group undergoes both electrostatic and hydrogen-bonding interactions with polar amino acid residues [34]. Figure 5d illustrates the comparative docking modes of C18 and C15 at the COX-2 active site. However, in the selected conformation of C18, the distance between the OH group of COOH and the NH of Arg120 is 3.96 Å, and that between the carbonyl CO group of COOH and the NH2 of Arg120 is 3.134 Å. Additionally, the bromo group at the meta position on the C-2 phenyl ring interacts hydrophobically with Met522 and Val523. The lack of these interactions when a bromo group is introduced at the para position, as in C19, decreases the binding affinity of this compound, even if an interaction between this group and Trp387 is established (d = 3.84 Å).

Replacing the bromine in C19 with fluorine, as in C20, leads to strong decrease in the experimental binding affinity. We speculate that the fluorine atom, which is less lipophilic than the bromine atom, exhibits decreased hydrophobic interaction in the pocket S2, which allows us to explain the decrease in the affinity of this compound to COX-2. This interaction is disfavored by the longer C=F bond compared to the C=Br bond (1.323 Å and 1.892 Å, respectively).

The introduction of an OH, OMe, OAc or NHAc group at the para position on the C-2 phenyl group of C17, as in compounds C21, C22, C23 and C24, decreases the experimental binding affinity. In the selected conformation of each compound, the OH, OMe, OAc or NHAc group occupies the top hydrophobic pocket S2. The methyl of the p-OMe group of C22 is an electron-donating group that increases the electron density of the oxygen atom so that it can form a hydrogen bond with Tyr385. As a result, the methyl of the p-OMe group of C22 appears to be important for the inhibitory activity.

Docked conformations of group 5 compounds

The introduction of a phenyl group at the para or meta position on the C-2 phenyl group of C17 leads to isopopulated binding modes A and B. The most populated binding mode corresponds to mode B. In this binding mode, the p-phenyl group of the C-2 phenyl ring is surrounded by Val349, Leu359, Val116, Tyr355 and Arg120.

The introduction of fluorine at the para and meta positions or an isopropyl group at the para position on the p-phenyl group in the C-2 phenyl ring, as in C25 and C26, leads to the same experimental binding affinity. From Fig. 5e, we can see that the p-fluorine and p-isopropyl groups are in the vicinity of Leu93, Ser119, Val116 and Tyr115. Figure 6b illustrates the hydrogen bond interactions of C25 in its most favorable binding mode in the COX-2 active site. This figure shows that the O atoms of p-MeSO2 form three hydrogen bonds with Arg513, Leu352 and Ser353, while COOH forms three hydrogen bonds with Ser530, Gly526 and Tyr385. This result is in agreement with mutagenesis experimental studies, which suggest that arachidonic acid and diclofenac bind to COX-2 in an inverted conformation with their carboxylate groups hydrogen bonded to Tyr385 and Ser530, leading to enzyme inactivity [46, 47]. Additionally, visualization of the selected binding conformation for C25 shows a lack of π–π interactions with Trp387 and Phe518.

Moving the isopropylphenyl and difluorophenyl groups of C25 and C26, respectively, to the meta position on the C-2 phenyl ring (as in C27 and C28) led to inactive inhibitors of the COX-2 enzyme (IC50 > 100 μM). The strong decreases in the experimental binding affinities of these two compounds in comparison with those of C25 and C26 may be attributed to the unfavorable interaction between the m-phenyl group and Arg120 (Fig. 5f). However, the replacement of the isopropyl group in C26 by MeSO2, as in C29, leads to the same binding affinity; the main reason is that substitution at the para position of the p-phenyl group (R3 group) does not influence the binding affinities of this series of compounds.

Moving the p-MeSO2phenyl group on the C-2 phenyl ring of C29 to the meta position on the C-2 phenyl ring (as in C30) leads to a hydrogen bonding interaction between the O atoms of p-MeSO2 and the backbone CO of Phe357. The distance between the S atom of Met113 and an O atom of p-MeSO2 in C30 is 2.786 Å.

Correlation between experimental affinities and predicted binding energies

The predicted binding energies (ΔGbinding, obtained by the AutoDock program) of the studied compounds and the corresponding experimental logAr values are listed in Table 3. The values in the table show that binding mode A is predominant compared to binding mode B. The logAr values of ligands can be grouped into two classes, active or inactive, based on whether the logAr value is less than 1 or more than 1, respectively. The docking energies of the inhibitors can be grouped into four classes: true positive (TP) or false positive (FP) when an inactive compound is predicted to be positive, and true negative (TN) or false negative (FN) when an active compound is predicted to be inactive. As a result, the active molecules are 90% correctly classified, while the inactive molecules are 80% correctly classified (Fig. 7). This indicates that there is a good correlation between the docking energies and the observed activities of the stilbene analogs.
Table 3

Predicted binding modes, docking energies ΔGbinding (kcal mol−1), and logAr values of the studied compounds

Molecule

Binding mode

ΔGbinding

logAr

C1

Mode A

−15.80

−0.368

C 2

Mode A

−11.37

0.660

C 3

Mode A

−10.09

1.410

C 4

Mode A

−10.22

2.675

C5

Mode A

−8.42

>3.155

C6

Mode A

−11.92

0.196

C7

Mode A

−11.18

1.155

C8

Mode A

−11.36

0.602

C9

Mode A

−14.53

−0.609

C10

Mode A

−8.75

2.053

C11

Mode A

−11.32

1.545

C12

Mode A

−12.92

−0.264

C13

Mode A

−10.90

1.341

C14

Mode A

−11.40

1.043

C15

Mode A

−9.63

1.410

C16

Mode A

−9.91

1.434

C17

Mode A

−9.69

1.632

C18

Mode A

−12.76

0.646

C19

Mode A

−11.43

1.711

C20

Mode A

−7.49

2.711

C21

Mode A

−10.56

1.879

C22

Mode A

−10.73

1.434

C23

Mode A

−8.65

1.617

C24

Mode A

−11.46

1.553

C25

Mode B

−12.48

0.660

C26

Mode B

−10.43

0.660

C27

Mode B

−9.19

>3.155

C28

Mode B

−8.59

>3.155

C29

Mode B

−12.29

0.660

C30

Mode B

−10.95

1.477

https://static-content.springer.com/image/art%3A10.1007%2Fs00894-010-0679-7/MediaObjects/894_2010_679_Fig7_HTML.gif
Fig. 7

Classification of the predictions made by the docking model for the thirty studied compounds

The correlation between the calculated ΔGbinding values of the studied compounds and the experimental logAr values is plotted in Fig. 8, and a correlation coefficient R = 0.835 was obtained. The straight line was obtained using the least squares method. Although this comparison represents the best correlation, considering the limitations of the available docking program and the fact that there are many unsolved areas, a value of 0.835 is not too discouraging. The ligand–receptor interaction is mediated by other factors, such as solvation effects and the flexibility and conformational changes of the receptor during binding with the ligand [48, 49]. Moreover, in the version of AutoDock used here, the receptor was treated as semi-flexible by selecting a few degrees of freedom to represent the protein flexibility [39]. The degrees of freedom chosen usually correspond to rotations around single bonds of some residues in the active site. However, the binding site still remains essentially rigid because the conformational changes of the protein are limited to a few terminal bonds.
https://static-content.springer.com/image/art%3A10.1007%2Fs00894-010-0679-7/MediaObjects/894_2010_679_Fig8_HTML.gif
Fig. 8

Plot of the experimental logAr values versus the predicted binding energies ΔGbinding (kcal mol−1) of the studied compounds (R = 0.835)

Conclusions

In the present study, an automated docking procedure was applied in order to determine the optimal positions and orientations of thirty stilbene analogs in the binding site of COX-2. The quality of the docking model was estimated based on their binding energies. A good correlation (R = 0.835) was observed between the experimental logAr values and the predicted binding energies of the studied compounds.

Docking results have allowed us to provide a plausible explanation of the different binding affinities observed by highlighting the most important interactions with COX-2 protein. These results are in agreement with previous site-directed mutagenesis studies on arachidonic acid and some COX-2 inhibitors [44, 46]. Most of the studied compounds have a very similar binding mode to that of the crystallographic complex 6cox. Two different binding modes, A and B, were predicted for this class of compounds within the active site of COX-2, which consists of three main pockets (S1, S2 and S3) and one shallow pocket (S4). Pocket S1 is composed mainly of lipophilic amino acids, while pockets S2, S3 and S4 are composed mostly of hydrophobic amino acids. It was found that the residues lining pocket S1, in particular His90, Gln192 and Arg513, may be essential for the binding of the ligand. Additionally, the backbone CO groups of Phe518 and Leu352 are predicted to be involved in hydrogen bonding interactions with the ligand. Some aromatic residues, in particular Phe518 and Trp387, may be essential for ligand recognition through π–π interactions. Arg120 was found to be essential for the ligand binding due to electrostatic interactions with the unsubstituted phenyl rings of the studied compounds.

This study shows that important conserved residues, in particular His90, Gln192, Arg513, Phe518, Leu352, Leu531, Trp387 and Arg120, could be specifically targeted in order to develop potential anti-inflammatory compounds. This study also opens up new perspectives for rational design in the development of new potential COX-2 inhibitors.

Acknowledgments

Souhila Bouaziz-Terrachet wishes to warmly thank Lehtihet Abdelhalim El Amine and Redouane Terrachet of the EMP (Ecole Militaire Polytechnique) of Algeria for their help.

Copyright information

© Springer-Verlag 2010