´
K. Martınez-Mayorga et al. / Bioorg. Med. Chem. 16 (2008) 5932–5938
5933
known opiates may help in the design of new opioid
ligands.
ilar to the query set (q1 and q2 in this case) than
molecules represented by data points at the lower left
of the diagram (low average and low maximum combo
score similarity). Closed squares represent molecules
with IC50 values 6502 nM. The corresponding com-
pound number is indicated in the plot. Open squares de-
note molecules with IC50 > 502 nM (circles represent a
second set of BCGs discussed below). As discussed
above, molecules at the top right of the plot, for exam-
ple, 6, 3, and 13 are the most structurally similar to the
queries. The relative high combo score values suggest that
these molecules could adopt a similar binding mode as the
active molecules 1 and/or 2. Finding active molecules (for
example molecules 9 and 11) in the region between 1.1 and
1.3 of combo score is more challenging but could be of
interest because compounds with different binding modes
may potentially be identified. Finally, a region that can be
called ‘inactive’ for this particular analysis, is located at
low average and low maximum combo score, apparently
molecules in this region of the multi-fusion similarity
map do not have the pharmacophoric features that drive
activity of the reference compounds. It is important to
note that nine out of the 12 most active molecules (com-
pounds 1–7, 10, and 12–14) have maximum combo score
greater than 1.4. Thus 75% of the active molecules in the
BGCs data set in Table 1 are similar (maximum combo
score greater than 1.4) to the shape and chemistry of mol-
ecules 1 or 2 depicted in Figure 1. However, active mole-
cules 8, 9, and 11 showed a lower 3D similarity to q1
and q2 (combo score less than 1.3 on maximum and aver-
age Figure 2).
Ligand-based computational methods have been shown
to be useful tools for exploring binding modes.16 Rapid
overlay of chemical structures (ROCS) is a 3D shape-
based method used to superimpose conformers of a candi-
date molecule with a query molecule. ROCS maximizes
the shared volume between each conformer in a database
against the query.17 Taking into account the chemical nat-
ure of the molecule (position of heteroatoms) enhances
the results. When the conformation of a compound in
the binding site is known, obtained by X-ray crystallogra-
phy or NMR for instance, it may be used as the reference
conformation.18 In certain cases a low energy conforma-
tion is considered as a starting point.17 However, there
are several examples, as in the case of this study, where
the bioactive conformation is unknown. Here, we em-
ployed ROCS of BCGs to derive a molecular binding
model. Structural modifications based on the 3D shape
comparison with known opiates are also suggested.
2. Results and discussion
2.1. 3D similarity analysis
Chemical structures and the corresponding IC50 values
for the BCGs used to develop the binding model are
summarized in Table 1. These molecules were identified
from a PS-SCL.11 A short description of the combinato-
rial and synthetic methodology as well as the biological
assay employed is described in Section 4.2. Generation
of the conformer distribution was performed using
OMEGA.19 For 3D shape comparisons the most active
molecules (1, 2 in Table 1) were selected as queries, and
will be called q1 and q2, respectively. The first step con-
sisted in selecting the conformations of the queries. In-
stead of choosing conformations at random or the
ones with lowest energy, the conformation of the query
that showed the best correlation between similarity
(combo score: shape + chemical nature score similarity)
and activity was chosen. The selected conformations are
depicted in Figure 1. The conformation of molecule 2
obtained when molecule 1 was the query was the same
as the one obtained when molecule 2 was the query,
and vice versa. In other words, the ‘Y’ shape conforma-
tion shown in Figure 1 was mutually obtained for 1 and
2.
Recently, a new set of BCGs have been published.13
These molecules were selected based on a computational
deconvolution method. The method involved the calcu-
lation of a predicted inhibitory capacity for all 102,459
compounds in the library. Out of this large collection
of compounds the computational deconvolution method
retrieved five new BCGs with IC50 values below 500 nM,
as well as correctly predicting several inactive
(IC50 > 10,000 nM) BCGs. To test the model proposed
in this work we computed the conformation and 3D
similarity analysis of those molecules with respect to
q1 and q2. Table 2 summarizes the second set of mole-
cules evaluated here. Mean and maximum similarity of
this second set of BCGs to the queries q1 and q2 are also
included in Table 2 and are represented as circles in Fig-
ure 2, closed circles denote molecules with
IC50 < 500 nM and open circles denote molecules with
IC50 > 10,000 nM. It is of interest that all the active mol-
ecules from the second set of BCGs have combo score
values above 1.3. This shows that these active molecules
from an external set can be predicted to bind in a similar
manner than the reference queries.
Once the optimal conformation of the queries was se-
lected from the two most active molecules in the data
set, the similarity of the remaining molecules in the data-
base to q1 and q2 was analyzed. Results are summarized
in Table 1. Figure 2 shows the corresponding multi-fu-
sion similarity map. In this plot, the X-axis represents
the mean combo score similarity of a given molecule
to q1 and q2 while the Y-axis represents the maximum
combo score similarity of a given molecule to either q1
or q2. A detailed description of the multi-fusion similar-
ity maps is described elsewhere.20 Data points at the top
right of the plot (high average and high maximum simi-
larity) indicate molecules that are more structurally sim-
2.2. Structural requirements for activity
Most of the active molecules with IC50 6 500 nM have a
4-methoxybenzyl group at R2 position, and the orienta-
tion of this group seems to be important for binding.
One example is molecule 14 which is the enantiomer of
the most active BCG of the series (1). Molecule 14 is
able to adopt the ‘Y’ shape conformation, however the
Martinez-Mayorga, Karina
