N. Levoin et al. / Bioorg. Med. Chem. 21 (2013) 4526–4529
4527
below. The scoring function PMF04 was used for the sorting of the
poses. The AUC was calculated at a Ki threshold of 0.66 nM, divid-
ing the dataset in 2 near equipopulated subclasses.
acidic residues of the extracellular loops (Glu175, Glu191). In fact,
4 distinct binding modes can be drawn among the 10 models we
found published (Fig. 1).11–21 Binding mode
a
is parallel to the
membrane plane, bridging Asp114 to Glu206. At the middle of
2.3. Pharmacophore design
the site, Tyr115 is engaged in p-stacking with aromatic group of
the ligand. Trp371 is also in contact with the ligand.15 By contrast,
binding mode b is perpendicular to the membrane plane. In this
site, agonists bridge Asp114 to Asp80, whereas antagonists are
thought to project deeper into the receptor, towards Ile125.17 Bind-
The pharmacophore was built in two steps. First, we considered
only low-energy conformers of ligands, generated by the FAST
method of Catalyst (implemented in Discovery Studio 3.0, Accelrys,
CA). This resulted in a mean of 70 conformers per ligand, taken as
an input for 3D QSAR Pharmacophore Generation tool, with solely
positive ionizable function and hydrophobic group selected. The
human H3R Ki was used as the activity property. Afterwards, this
simple pharmacophore hypothesis was completed by customizing
the hydrophobic function, in order to include carbon atoms of
piperidines. The conformational space of ligands was also extended
to higher-energies, by collecting frames of molecular dynamics ran
at 300, 600 and 800 K, in NVT conditions and a time step of 1 fs.
MMFF force field was used, with an implicit distance-dependant
dielectrics of 4, and a non bond cut off distance of 18 Å. Hence
an ensemble of 135,000 ligand conformers was used. The ROC
AUC was calculated at a threshold of 0.7 nM (18 actives, 18 inac-
tives), using the pharmacophore fitting score (Fitvalue).
ing mode
supposed interaction between Asp114 and the amine of the ligand.
As for the binding mode , Tyr189 is involved in ligand recogni-
c was obtained after docking with constraints based on a
a
tion.18 Binding mode d is oriented as b, but ligands are in contact
with Glu206, Tyr115, Tyr374, Phe198, Glu175 or Glu191.[19] Fur-
thermore, even if some models agree in the positioning of the li-
gands, head-to-tail predicted orientations are not uncommon.
This difficulty to establish ligand orientation for histamine recep-
tors was again recently discussed.22 If different receptor conforma-
tion states or some induced-fit could indeed explain these
discordances, it would be very interesting to determine whether
a universal binding mode could be obtained to support drug dis-
covery and medicinal chemistry efforts.
In order to compare the different receptor models and ligand
binding modes, we have collected dibasic H3R ligands from the lit-
erature. Dibasic compounds were chosen because if both cationic
groups form salt bridges, the determination of interacting residues
would be facilitated. As will be seen below, the SAR actually sup-
port this hypothesis. We selected simple and often symmetrical
ligands to define precisely the interbasic distance, and also in-
cluded branched molecules to sterically refine the model (Fig. 2).
The SAR was extended by preparing longer and rigidified cong-
eners of the scaffold A and B series (Fig. 3). Existing SAR demon-
strated that H3R is very tolerant in the region distal from the
first cationic group. But in the dibasic series, the 2nd cationic group
is very important for the binding affinity, as outlined by scaffold A
and B. Homologous monobasic molecules showed indeed a drastic
drop of affinity (cf compound 11 vs 15, 12 vs 16, 13 vs 17, and 27 vs
21).26
3. Results and discussion
Although the H3R models described in the literature have a
majority of common traits, they show some alternatives in ami-
no-acid conformation, rendering ligand binding modes quite dif-
ferent. In a sense, small differences in receptor models (side
chain rotamers) have great consequences for ligand positioning.
This is particularly striking for the orientation of ligands in the
receptor, as well as for the definition of residues interacting with
basic groups of the ligands. Known H3R ligands possess either 1
or 2 positive ionizable groups (imidazole, alkylamine) which,
according to the models, are supposed to interact with Asp80
(transmembrane helix 2, TM2), Asp114 (TM3), Glu206 (TM5), or
This dataset was used as a discriminative tool by docking in the
five H3R models described below. Model A corresponds to that we
previously described.19 Although this model and its binding mode
d were statistically validated, most aminergic GPCRs solved so far
by X-ray crystallography show a ligand orientation similar to bind-
ing mode a, rendering it a priori the most probable. So we tested
whether model A could be reconciled with this binding mode.
We previously observed that model A was incompatible with bind-
ing mode
a because of the rotameric state of Tyr115, located
(
)
p
N
N
O
(
)
q
p = 0-3
q = 1-5
Figure 1. Left: acidic amino-acids in human H3R possibly interacting with dibasic
ligands. Right: schematic representation of the different ligand orientations
suggested in the literature. Binding mode
a c in
is in blue12–16 b in yellow,17
black18 and d in red.19
Figure 3. Newly synthetised derivatives of scaffold A and B.26
(
)
k
scaffold B
scaffold A
N
N
scaffold C
N
(
)
m
N
R2
N
O
(
)
(
)
o
N
n
R1
k, l = 1-2
N
N
(
)
l
m = 1-2
R1, R2 = small alkyl
n, o = 2-4
Figure 2. Dibasic antagonists used in the study: scaffold A,14,23 scaffold B,24 and scaffold C.25
Levoin, Nicolas
