Docking of side chains derived from molecules of the ZINC database allowed

AND LEU26 BINDING POCKETS TO BE IDENTIFIED

This section describes the docking of oligobenzamides with a diverse range of side chains using FlexX and the subsequent analysis of side chain properties and docking scores. The aim was to identify the side chain properties likely to increase the affinity of an oligobenzamide binding with its side chains in the sites usually occupied by p53 residues Phe19, Trp23 and Leu26 in the p53-Mdm2 interaction.

An initial oligobenzamide binding position was required for high-throughput screening to identify potential side chains. The O1 scaffold starting compound (Figure 2.1, p102) was docked into Mdm2 from PDB structure 1T4F37 using AutoDock 4.2 as detailed in the methods section (p109). The side chains of the compound resemble the side chains of Phe19, Trp23 and Leu26 from the p53 helix. Comparison of the highest scoring pose obtained (Figure 2.8B) with the position of the p53 helix bound to Mdm2 in structure 1T4F (Figure 2.8A), which played no part in the docking, shows that the side chains have been positioned close to their expected positions (Figure 2.8C). The heavy atom RMSDs for the first side chain, second side chain terminal benzene ring and third side chain relative to the p53 residue side chains are 2.09 Å, 1.24 Å and 1.66 Å respectively.

Molecules from the ZINC source database were cut using ReCore (BioSolveIT) as described in the methods section (p106). These were spliced on to the docked O1 starting molecule (Figure 2.8B) and screened using ReCore. 536, 505 and 508 possible side chains were identified by this high-throughput screening for the first, second and third side chain positions on the scaffold respectively. FlexXc (part of FlexX (BioSolveIT)) was then used to carry out combinatorial docking. This software docks the scaffold and then attaches and tests each member of a library of possible side chains. The side chains are attached in situ and the resulting pose is minimised to complete the docking process.

The side chain attachment positions were treated separately; each of the possible side chains was tested at each attachment position while the other two side chains were left unchanged from those of the initial structure. (For further details see p109.) Poses with an RMSD of greater than 2 Å relative to the high-throughput screening pose generated by ReCore, in which the scaffold position is unchanged from in the Autodock pose, were discarded. Consequently all of the poses resemble that shown in Figure 2.8B where the N-terminal, central and C-terminal side chains are in the pockets usually occupied by p53 residues Phe19, Trp23 and Leu26.

A

B

C

Figure 2.8: The best pose resulting from the docking of an oligobenzamide into Mdm2 from PDB structure 1T4F superimposed on the structure of p53 from the PDB structure. A) Structure 1T4F showing a modified p53 peptide (green) in the p53 binding site of Mdm2 (grey). B) The highest scoring pose generated by docking the oligobenzamide shown in Figure 2.6A (p114) (blue) into the structure of Mdm2 from 1T4F. C) The result of removal of the protein and superimposition of the original p53 structure (as in A) on the docked oligobenzamide structure (as in B). There is excellent agreement between the structures despite the p53 peptide playing no part in the oligobenzamide docking.

The properties of each side chain were calculated and Spearman’s Rank correlation coefficients (rs) were determined for each property to investigate whether its value correlated with the FlexX docking score of the resulting oligobenzamide. Positive Spearman’s rank values indicate that the higher the value of the property is, the greater the docking score; however, the scores are predicted ΔG values so higher scores correspond to weaker binding. Table 2.3 (p131) shows the results. Caution is necessary when interpreting these results for several reasons. Firstly, multiple hypotheses were tested. Secondly, it was assumed that all side chains bound in the expected docking pose (side chain 1 in or near the p53 Phe19 pocket, side chain 2 in or near the Trp23 pocket and side chain 3 in or near the Leu26 pocket) and this assumption might be incorrect. Finally, docking scores are often poor measures of binding affinity161,162.

As shown in Table 2.3, at the middle side chain position, larger side chains lead to better scores. (For the total number of atoms, number of carbons and number of heteroatoms, rs<-0.03, rs<- 0.14 and rs<-0.10 respectively based on the 95% confidence limits shown in the table.)

However, at the side chain position nearest to the amino terminus, small, low molecular weight groups appear to be favoured (because larger side chains result in larger scores and higher scores signify weaker binding). According to the 95% confidence limits, for the number of atoms in, number of carbons in and molecular weight of side chain 1, rs>0.05, rs>0.04 and rs>0.08 respectively. At the carboxyl terminus, the side chain has little room to manoeuvre so escape from the expected binding pocket requires side chain flexibility. This could explain why rigid conjugated systems are unfavourable at this position. (For the number of rotatable bonds, rs>0.25 and for the number of rigid bonds, rs <-0.29 based on the 95% confidence limits.) At all three side chain positions, fragments that lead to a molecule with a high polar surface area (PSA) have good scores (rs<-0.27, rs<-0.30 and rs<-0.47 for side chains 1, 2 and 3 respectively). More polar molecules have lower logP values and, consistent with this, side chains with a lower XlogP (increased water solubility relative to their solubility in octanol) also score favourably (rs>0.33, rs>0.24 and rs<0.29 for side chains 1, 2 and 3 respectively). The presence of hydrogen bond donors (rs<-0.02, rs<-0.11 and rs<-0.13) and acceptors (rs<-0.15, rs<-0.15 and rs<-0.28) appears to increase affinity and in the central and C-terminal side chain position, side chains with a positive charge have good scores (rs<-0.07 and rs<-0.10 respectively).

It is difficult to reconcile the PSA, XlogP and hydrogen bonding results with the overall hydrophobic nature of the p53 binding site of Mdm2253 and the known importance of side chain hydrophobicity for strong binding31. Park and Jeon229, point out that the FlexX force field lacks terms to directly account for ligand and binding site desolvation. These unexpected correlations could be the result of this bias inherent in the FlexX scoring function. Shoichet et al.254 discuss how virtual screening without considering differences in solvation energy between compounds has a tendency to return molecules that are too large or too highly charged.

Table 2.3: How the properties of side chains at the three side chain positions in the molecules docked with FlexX correlated with the docking scores of the compounds.

Property

Side chain 1 Side chain 2 Side chain 3

n Med. value Spearman’s Rank n Med. value Spearman’s Rank n Med. value Spearman’s Rank Low limit High limit Low limit High limit Low limit High limit Number of atoms 365 13 0.05 0.12 0.20 348 13 -0.17 -0.10 -0.03 359 13 -0.13 -0.06 0.01 Number of carbons 365 9 0.04 0.12 0.19 348 9 -0.21 -0.14 -0.08 359 9 -0.12 -0.05 0.02 Number of heteroatoms 365 3 -0.18 -0.11 -0.04 348 3 -0.17 -0.10 -0.03 359 3 -0.29 -0.23 -0.17 Hydrogen bond acceptors 365 1 -0.28 -0.22 -0.15 348 1 -0.28 -0.22 -0.15 359 1 -0.38 -0.33 -0.28

Hydrogen bond donors 365 1 -0.15 -0.08 -0.02 348 1 -0.24 -0.17 -0.11 359 1 -0.25 -0.19 -0.13

Molecular weight 365 190 0.08 0.16 0.23 348 186 -0.10 -0.03 0.04 359 186 -0.07 0.00 0.07

Number of formal charges 365 0 -0.10 -0.03 0.04 348 0 -0.21 -0.14 -0.07 359 0 -0.23 -0.17 -0.10

Sum of formal charges 365 0 -0.06 0.01 0.09 348 0 -0.19 -0.12 -0.05 359 0 -0.19 -0.12 -0.05

Rotatable bonds 365 3 0.10 0.18 0.26 348 3 -0.03 0.05 0.13 359 3 0.25 0.33 0.40

Rigid bonds 365 10 -0.07 0.00 0.07 348 10 0.02 0.10 0.18 359 9 -0.39 -0.34 -0.29

Predicted compound XlogP 278 1.8 0.33 0.42 0.50 262 1.7 0.24 0.33 0.41 273 1.8 0.29 0.38 0.46 Compound 2D PSA* 278 -0.72 -0.39 -0.33 -0.27 262 -0.65 -0.42 -0.36 -0.30 273 -0.67 -0.56 -0.52 -0.47 Each side chain position was investigated separately. For each side chain position, a library of compounds which differed in terms of the side chain at the position in question was docked. The side chains used for this analysis were those obtained by cleavage of molecules from the ZINC source database, filtered such that all of the oligobenzamides generated were theoretically synthetically accessible. For further details see p106. The column labelled “n” indicates the number of side chains used and therefore also the number of data points used in the subsequent regression analysis. The “Med. value” column shows the median value of the side chain property for all of the compounds in the library used to investigate the side chain position in question. Spearman’s rank correlation coefficients were calculated to look for any correlation between the properties of side chains and the relative binding affinities of oligobenzamides containing them. 95% confidence limits (“high limit” and “low limit”) on the values of each Spearman’s rank are shown in the table on

either side of the ranks themselves. For details of how these confidence limits were calculated, see p291. All of the properties shown were calculated using the Open Babel217 Filter

program. *Topological polar surface area. (See p104.) Calculation of the topological PSA and logP estimate (XlogP222) of a molecule requires data for the functional groups within it.

The functional group composition of some side chains meant that their predicted logP and polar surface area could not be calculated, so fewer side chains were investigated when analysing the effect of these two properties on docking score.

2.3.2. LONG SIMULATIONS OF OLIGOBENZAMIDES BASED ON THE N1