All docking and scoring was performed with the software eHiTS (electronic High Throughput Screening) Lightning (Version 8.0.rc2.4) by
SimBioSys Inc. running on a Sony PlayStation 3. eHiTS docks small molecules into a protein structure by first dividing up the small
molecules into rigid fragments and flexible chains. The rigid fragments are docked independently into the receptor site, generating
multiple poses which are stored in DockTable, an SQL database so that common molecular fragments that occur in multiple small molecules
can be reused. A graph matching algorithm enumerates all compatible fragment pose combinations and then flexible chains are fitted
between the rigid fragment poses to satisfy steric criteria imposed by the fragments and the receptor site. Finally, local minimization
is performed using a modified Powell's method on the reconstructed structures to obtain the final poses. An empirical scoring
function is used several times during the algorithm including during evaluation of rigid poses, selection of best graph matching
solutions, flexible chain fitting, and final local optimization.
To handle the problem of protein flexibility, eHiTS provides a “soft” representation of the protein structure in
three respects. The eHiTS scoring function utilizes the temperature factor information provided in the PDB files to attempt in its
gauging of the interaction as well as considering the probability of the atom positions to create a derived empirical scoring function.
eHiTS rotates the hydroxyl groups of the serine, threonine and tyrosine residues of the protein and also the -NH3+
group of lysine. Thus, the interaction flexibility of these is considered even though eHiTS does not move the heavy atoms of the main or
side chains during this process. Furthermore, the steric clash, or van der Waals potential, is considered with a “soft”
quadratic potential as opposed to the harder 6‐12 potential often employed in force fields. The top scoring thirty‐two
orientations for each compound successfully docked are saved and the compounds are ranked by the top scoring pose calculated for each
structure.
Two sets of compounds were assembled using data from the Binding Database
[
10,
11]. One set of compounds
contained active compounds which were reported as having an inhibition constant K
i of less than 10 nM for ERα
[
12,
13,
14,
15,
16].
The other set of compounds was a set of decoy molecules which were found to show no activity even at concentrations greater than 100000
nM [
17]. Ten PDB structures were chosen which were able to accurately rank the
majority of actives from the decoy molecules ‐‐ 1R5K, 1SJ0, 1XP1, 1XP6, 1XP9, 1XPC, 1YIM, 1YIN, 2OUZ, and 3ERT. To
validate this, the best (i.e. minimum score) a molecule received across any of the above structures was taken and a Student's
t‐test was performed comparing the scores between active and decoy molecules. A
p‐value of less than
0.0001 suggested that there was a significant difference between the scores ctives received and the scores decoys received with the
active molecules having better scores.
A library of triphenylphosphonium salts and cations was assembled through a substructure search of the Sigma‐Aldrich
catalogue. Anions were removed from the salts and redundant compounds were removed to form a library of 315 compounds. 3D coordinates
for the structures were generated using the Molconvert utility from ChemAxon [
18].
Compounds were docked into the ten X‐ray structures with standard settings in eHiTS. PyMol from DeLano Scientific was used for
visual inspection of results and graphical presentations. After observing an interesting orientation of one of the higher scoring
phosphonium cations, a structure which combined the phosphonium cation with the co‐crystalized ligand was also tested using the
same settings.
This article has been 