Novel and highly potent histamine H3 receptor ligands. Part 1: Withdrawing of hERG activity

Article abstract of DOI:10.1016/j.bmcl.2011.07.006

Pre-clinical investigation of some aryl-piperidinyl ether histamine H3 receptor antagonists revealed a strong hERG binding. To overcome this issue, we have developed a QSAR model specially dedicated to H3 receptor ligands. This model was designed to be di

Full text of DOI:10.1016/j.bmcl.2011.07.006

Bioorganic & Medicinal Chemistry Letters 21 (2011) 5378–5383
Contents lists available at ScienceDirect
Bioorganic & Medicinal Chemistry Letters
journal homepage: www.elsevier.com/locate/bmcl
Novel and highly potent histamine H3 receptor ligands. Part 1: withdrawing
of hERG activity
⇑
Nicolas Levoin , Olivier Labeeuw, Thierry Calmels, Olivia Poupardin-Olivier, Isabelle Berrebi-Bertrand,
Jeanne-Marie Lecomte, Jean-Charles Schwartz, Marc Capet
Bioprojet-Biotech, 4 rue du Chesnay Beauregard, BP 96205, 35762 Saint Grégoire, France
a r t i c l e i n f o
a b s t r a c t
Article history:
Pre-clinical investigation of some aryl-piperidinyl ether histamine H3 receptor antagonists revealed a
strong hERG binding. To overcome this issue, we have developed a QSAR model specially dedicated to
H3 receptor ligands. This model was designed to be directly applicable in medicinal chemistry with no
need of molecular modeling. The resulting recursive partitioning trees are robust (80–85% accuracy),
but also simple and comprehensible. A novel promising lead emerged from our work and the struc-
ture–activity relationships are presented.
Received 21 June 2011
Revised 4 July 2011
Accepted 5 July 2011
Available online 13 July 2011
Dedicated to Pr. Walter Schunack, in
memoriam
Ó 2011 Elsevier Ltd. All rights reserved.
Keywords:
Histamine
H3R
HERG
QSAR
Histamine H3 receptors (H3Rs) are widely distributed in the
central nervous system where they modulate transmitter release,
particularly in brain regions critical for cognition and sleep regula-
tion. They are located presynaptically and postsynaptically, acting
both as auto- and heteroreceptors. As autoreceptors, H3R nega-
tively regulate the synthesis and release of histamine (negative
feedback mechanism of histamine neurotransmission). As hetero-
receptors, H3R modulate other neurotransmitters, through cholin-
ergic, adrenergic, dopaminergic and serotoninergic pathways.1 Due
to the effect of H3R signalling on multiple neuronal transmitters,
H3R antagonists represent very promising drug candidates for
the treatment of many CNS disorders, such as Alzheimer’s disease,
attention deficit hyperactivity disorder, schizophrenia, sleep dys-
functions and memory impairment.2–4 H3R ligands may also be
useful in the control of pain.5
To help and rationalize the chemical modifications that de-
crease hERG affinity, we built different models by use of published
structure–activity relationships.
An ideal model in medicinal chemistry should be accurate, but
also comprehensible and simple to interpret. So we turned towards
ligand-based approaches. Many hERG models have been described
in the literature,10–12 but, to our knowledge, none deals specifically
with H3R ligands. This renders the applicability of existing phar-
macophore and QSAR models questionable. Actually, these meth-
ods are known to be robust with their parent data, but to quickly
loose their efficacy when chemical series diverge.13,14 To overcome
this inherent limit, we used only H3R ligands to build a QSAR mod-
el for hERG activity, compiling recent data.15
We used 2D descriptors to alleviate conformational problems,16
and we chose recursive partitioning (RP) methodology because of
its intuitive meaning. In a medicinal chemistry perspective, RP is
actually amenable to visual and logical perception, and seems pref-
erable to complex equations and mathematical models. The activ-
Our past work targeting H3R gave rise to many potent chem-
ical series.6–9 From these medicinal chemistry efforts, FUB2.922
(Table 3) emerged as a very interesting lead, combining the
advantages of displaying a strong H3R affinity (Ki = 3 nM) and
being easily synthesizable. However, it turned out to have cardio-
vascular safety concerns. In particular, its strong inhibition of
ity threshold was set to 7 lM for hERG IC50, dividing the dataset
between 112 active and 105 inactive ligands. We obtained 2 RP
models, of high statistical quality (Fig. 1).
hERG channel (70% inhibition at 1
clinically developed.
lM), precluded it from being
The first tree contains 8 nodes and uses 8 descriptors.17 Nodes
of the trees are globally pure (mean of 81%), and the worst node
is 62% pure (Table 1). This tree is a very robust categorization tool
since an accuracy of 86% was calculated.18 The second model is
slightly less robust (accuracy of 81%), but has the advantage of
⇑
Corresponding author.
E-mail address: n.levoin@bioprojet.com (N. Levoin).
0960-894X/$ - see front matter Ó 2011 Elsevier Ltd. All rights reserved.
doi:10.1016/j.bmcl.2011.07.006

N. Levoin et al. / Bioorg. Med. Chem. Lett. 21 (2011) 5378–5383
5379
Fig. 1. The 2 RP models obtained with the training set (217 H3R ligands with hERG IC50 ranging from 4 nM to 300
l
M).15 ‘A’ means that the branch is mainly composed of
active ligands (high hERG affinity). ‘I’ stands for mainly inactive molecules. The threshold was set to 7 l
M. The meaning of descriptors abbreviation is given in.17
Table 1
Descriptive parameters of the 2 RP models
Nodes
Descriptors
Tree depth
Mean purity
Worst purity
Accuracy
Sensitivity
Specificity
Model 1
Model 2
8
5
8
5
5
4
0.81
80
0.62
68
0.86
0.81
0.88
0.83
0.83
0.79
Table 2
Examples of simple chemical modifications of different scaffolds withdrawing hERG activity. References are given in Supplementary data15
Molecule
Fragment
HERG IC50
(lM)
ES Sum aaN
HBD Count
Apol
SI reference
5
O
N
S
N
EX1
0.1
13.7
16,496
N
N
N
S
N
EX2
23
17.5
15,331
5
N
R
O
N
EX3
EX4
1.9
7
0
1
17,980
16,953
10
10
R
N
HO
O
EX5
EX6
3.7
40
0
1
12,295
11,900
1
1
R1
O
R2
R2
R1
N
H
being more interpretable than the precedent, because it is shorter
and does not use topological descriptor (as CIC for the first RP). A
reading of the RP tree is the following: if a molecule has descriptor
Apol < 15473.8 and A log P P 3.1785, then it might have hERG
stance, a pyridinylthiazolopyridine is hERG active (see fragment of
H3 ligand EX1 in Table 2).15 It has 3 aromatic nitrogens and sum
aaN < 13.8465. Replacing pyridinyl by pyrimidinyl (increase in
sum aaN in model 2) reduce greatly hERG binding (EX2). As an-
other example, an ether derivative is hERG active (EX3). Adding a
hydrogen-bond donor (HBD count in model 2) alleviates hERG is-
IC50 < 7
lM (model 1). From a drug design perspective, the tree is
a rational way of ameliorating the selectivity of H3R ligands. For in-

5380
N. Levoin et al. / Bioorg. Med. Chem. Lett. 21 (2011) 5378–5383
Table 3
New derivatives of FUB2 922, with variations of the phenyl A substituent. H3R$ : competition with [125I]-iodoproxifan binding. hERG$ : competition with [3H]-Dofetilide binding
B
FUB2.922
R
A
N
O
N
O
Human recombinant
Compound
R
H3R$
hERG$
Apol
A log P
sum aaN
Log D
HBD Count
Ki (nM)
Ki (nM)
1
2
1.4
5.0
14
12993.4
11653.2
4.4
3.8
0.0
0.0
2.8
2.3
0
0
X
X
S
90
O
O
N
3
4
9.0
4.8
2747
11024.0
11434.0
3.2
2.7
0.0
4.2
1.7
1.6
0
1
X
X
10,000
190
N
H3C
N
N
5
6
1.8
2.5
11947.3
11345.6
3.4
2.6
8.2
3.9
1.8
1.1
0
0
X
N
1194
O
X
N
7
8
3.6
1.6
10,000
230
11030.9
11507.1
2.9
4.4
0.0
0.0
1.3
2.9
0
0
X
X
O
9
2.8
4.9
5.2
8.3
3.6
13
11705.7
11625.7
11537.2
11537.2
13711.7
5.1
4.0
3.7
3.4
4.2
0.0
0.0
0.0
0.0
0.0
3.6
1.7
2.1
1.8
2.7
0
0
0
0
0
X
X
X
X
10
11
12
13
841
1125
2220
5
N
O
O
O
O
X
X
14
15
2.2
1.6
76
13507.7
16475.8
4.4
4.8
0.0
0.0
2.9
3.2
0
0
O
CH3
X
818
O
N

N. Levoin et al. / Bioorg. Med. Chem. Lett. 21 (2011) 5378–5383
Human recombinant
5381
Table 3 (continued)
Compound
R
H3R$
hERG$
Apol
A log P
sum aaN
Log D
HBD Count
Ki (nM)
Ki (nM)
O
CH3
O
16
17
10.4
0.2
8
14988.3
4.7
0.0
3.1
0
0
X
X
CH3
N
CH3
10,000
13165.5
4.6
0.0
1.4
c,d or e
a,b
a,b
c
O
Br
O
CHO
HO
Br
HO
CHO
N
( )3
N
( )3
18
19
N
O
R
O
N
( )3
N
H
N
( )3
4
N
R=
S
O
O
O
N
d
O
CHO
O
MeO
14
13
1
2
10
N
O
( )3
N
( )3
Scheme 1. Synthesis of compounds 1, 2, 10, 13 and 14. Reagents and conditions: (a)
1-bromo-3-chloropropane, K2CO3, DMF, rt, 15 h; (b) piperidine, K2CO3, KI, DMF,
100 °C, 15 h; (c) 2-thiopheneboronic acid (for 1) or 3,4-methylenedioxyphenylbo-
ronic acid (for 13) or 2-methoxyphenylboronic acid (for 14), Pd(PPh3)4, toluene or
THF, 100 °C, 15 h; (d) 2-(tributylstannyl)furan, PdCl2(PPh)2, THF, reflux, 15 h; (e)
piperidine, tBuONa, Pd2(dba)3, BINAP, toluene, 100 °C, 15 h.
6
Scheme 3. Synthesis of compounds 4 and 6. Reagents and conditions: (a) 1-bromo-
3-chloropropane, K2CO3, DMF, rt, 15 h; (b) piperidine, K2CO3, KI, DMF, 110 °C, 3 h;
(c) glyoxal, NH4OH, EtOH/H2O, rt, 15 h; (d) TosMIC, MeOH, K2CO3, reflux, 2 h.
and 14), 2-(tributylstannyl)furan (compound 2) or piperidine
(compound 10). Compound 18 was obtained from commercially
available 4-bromophenol after condensation with 1-bromo-3-chlo-
ropropane and piperidine.
sue (hydroxyl in EX4). Butenone to amide isosterism is another
illustration (EX5/EX6).
For both trees, the most differentiating nodes deal with lipo-
philic character of the molecules (Apol, A log P, log D),17 as well
as aromatic tendency (aaN). This reflects most important aspects
for hERG affinity, at least for H3 ligands. Hence, modulating lipo-
philicity of our lead FUB2.922 (Table 3) should have strong effect
in its hERG profile.
We tested this hypothesis by varying the phenyl A substituents
(Table 3). The synthetic route for the compounds reported herein is
illustrated in Schemes 1–7.
The common intermediate 18 was also used to form the cor-
responding Grignard or lithium reagents. Condensation with
cycloalkylketones or tetrahydropyran-4-one, dehydration and
subsequent hydrogenation allowed to access to 8, 9 and 12.
Imidazole 4 and oxazole 6 were synthesized as shown in scheme
3 from benzaldehyde 19, readily accessible from 4-hydroxybenzal-
dehyde. Pyrazole 5 and oxazoline 7 were prepared from the corre-
sponding acetophenone 20 or carboxylic acid 21 (scheme 4).
The synthesis of the tetrahydrofuran or -pyran moieties in 3 and
11 required a RCM (ring closure metathesis) strategy employing
adequate vinyl or allyl derivatives (scheme 5).
The first compounds ( Scheme 1) were synthesized by palla-
dium-catalyzed cross-coupling reactions of aryl bromide interme-
diate 18 with aryl- or heteroarylboronic acids (compounds 1, 13
OH
c
a,b
O
O
Br
N
( )3
N ( )3
X
18
d
O
O
X
X
N
( )3
N
( )3
8 (X=CH2)
9 (X=CH2CH2)
12 (X=CH2-O)
Scheme 2. Synthesis of compounds 8, 9 and 12. Reagents and conditions: (a) Mg, THF or BuLi, THF ꢀ78 °C to 0 °C, 1 h; (b) cyclopentanone (for 8) or cyclohexanone (for 9) or
tetrahydropyran-4-one (for 12), THF, 0 °C, 2 h; (c) H2SO4, EtOH, rt, 2 h; (d) 1 atm H2, 5% Pd/C, EtOH, rt 15 h.

5382
N. Levoin et al. / Bioorg. Med. Chem. Lett. 21 (2011) 5378–5383
The trans isomer was purified by careful flash chromatography
over silica.
Reducing the aromaticity appears as a general solution to de-
O
c,d
O
a,b
O
HO
N
( )3
crease hERG binding (see furan 2 and tetrahydrofuran 3, oxazoline
6 and oxazole 7), but some heterocycle isosters also permit the
reduction of hERG affinity (oxazoles vs furan or thiophene). By
combining heteroatom introduction and hydrophobicity reduc-
tion, hERG binding can be largely abrogated (imidazole 4). Even
with aliphatic groups, introducing heteroatoms greatly diminishes
hERG affinity (compare cyclohexyl 9 vs piperidine 10 or tetrahyd-
ropyrans 11 and 12). Thus, the derivatives presented here clearly
show the importance of lipophilicity and polarity of the molecules.
In the extreme, the addition of a formal charge is another solution
to attenuate hERG issue: compound 17 is the lowest binder for
hERG, though being the most potent H3R ligand of the series pre-
sented here.
20
N
N
O
N
( )3
5
f,g,h
a,b,e
O
CO2H
HO
CO2Me
N
( )3
21
For our QSAR models, these variations constitute an external
dataset to test the predictive behavior of our QSAR models. Table
4 shows that model 1 and 2 predict rather correctly the hERG class
of the new molecules (89% accuracy), even if they are less specific
than for the training set (67% vs 79% or 83%).
We have previously described a structure-based approach to
deal with hERG issue, using a refined docking/scoring protocol.19
We confronted this homology model to the current datasets of
H3R ligands. We obtained an accuracy of 73% for the training set,
and 65% for the test set using a customized Ludi_2 scoring func-
tion. This suggests that structure-base approach is less precise than
the RP models presented here, but on the other hand, docking is
N
O
O
N
( )3
7
Scheme 4. Synthesis of 5 and 7. Reagents and conditions: (a) 1-bromo-3-chloro-
propane, K2CO3, DMF, rt, 15 h; (b) piperidine, K2CO3, KI, DMF, 80 °C, 6 h; (c) NaH,
THF, then HCO2Me, rt, 2.5 h; (d) MeNH-NH2, THF, rt, 1 h; (e) NaOH, MeOH, reflux,
30 min; (f) SOCl2, rt, 15 h; (g) 2-chloroethylamine, NEt3, CHCl3, rt, 7 h; (h) DBU,
DCM, reflux, 24 h.
c,d
a,b
e,f
O
BnO
CHO
( )
n
BnO
a,b
O
O
HO
O
N
( )3
O
22
O
( )
N ( )3
n
( )
n
O
O
N
c
3 (n=0)
N
( )3
HO
11 (n=1)
17
Scheme 5. Synthesis of compounds
3 and 11. Reagents and conditions: (a)
Scheme 7. Synthesis of compound 17. Reagents and conditions: (a) 1-bromo-3-
chloropropane, K2CO3, DMF, rt, 15 h; (b) piperidine, K2CO3, KI, DMF, 80 °C, 6 h; (c)
Me2NH, NaBH(OAc)3, AcOH, THF, rt, 15 h.
vinylmagnesium bromide (for 3) or allylmagnesium bromide (for 11), THF, THF,
40 °C, 15 min; (b) NaH, DMF, then allyl bromide, rt, 1 h; (c) Grubbs’ catalyst 2nd
generation, CH2Cl2, reflux, 24 h; (d) 3 bar H2, 10% Pd/C, EtOH, rt 15 h; (e) 1-bromo-
3-chloropropane, K2CO3, DMF, rt, 20 h; (f) piperidine, K2CO3, KI, DMF, 60 °C, 5 h.
Table 4
Statistical analysis of the 2 RP models with the test set
Preparation of compounds 16 and 15 relied on the usual intro-
duction of the side chain by alkylation of phenols and the transfor-
mation of an ester into an amide function (scheme 6).
The desired dimethylcyclohexylamine 17 was prepared via a
reductive amination reaction of cyclohexanone 22 (scheme 7).
Accuracy
Sensitivity
Specificity
Model 1
Model 2
0.89
0.89
0.93
0.93
0.67
0.67
a,b
O
HO
CO2Et
N
( )3
CO2Et
16 (p-CO2Et)
c,d,e
O
N
( )3
O
15
N
Scheme 6. Synthesis of compounds 15 and 16. Reagents and conditions: (a) 1-bromo-3-chloropropane, K2CO3, CH3CN, reflux, 18 h; (b) piperidine, K2CO3, KI, CH3CN, reflux,
15 h; (c) NaOH, EtOH, reflux, 3 h; (d) (COCl)2, DCM, DMF, rt, 1 h; (e) piperidine, rt, 1 h.

N. Levoin et al. / Bioorg. Med. Chem. Lett. 21 (2011) 5378–5383
5383
7. Meier, G.; Ligneau, X.; Pertz, H. H.; Ganellin, C. R.; Schwartz, J.-C.; Schunack, W.;
Stark, H. Bioorg. Med. Chem. 2002, 10, 2535.
8. Lazewska, D.; Ligneau, X.; Schwartz, J.-C.; Schunack, W.; Stark, H.; Kiec-
Kononowicz, K. Bioorg. Med. Chem. 2006, 14, 3522.
9. von Coburg, Y.; Kottke, T.; Weizel, L.; Ligneau, X.; Stark, H. Bioorg. Med. Chem.
Lett. 2009, 19, 538.
10. Thai, K.-M.; Ecker, G. F. Bioorg. Med. Chem. 2008, 16, 4107.
11. Su, B.-H.; Shen, M.-Y.; Esposito, E. X.; Hopfinger, A. J.; Tseng, Y. J. J. Chem. Inf.
Model. 2010, 50, 1304.
12. Johnson, S. R.; Yue, H.; Conder, M. L.; Shi, H.; Doweyko, A. M.; Lloyd, J.;
Levesque, P. Bioorg. Med. Chem. 2007, 15, 6182.
13. Kühne, R.; Ebert, R.-U.; Schüürmann, G. J. Chem. Inf. Model. 2006, 46, 636.
14. Hansen, K.; Rathke, F.; Schroeter, T.; Rast, G.; Fox, T.; Kriegl, J. M.; Mika, S. J.
Chem. Inf. Model. 2009, 49, 1486.
much more informative in terms of structural information. So the
gold-standard should combine both approaches.
Our past drug discovery efforts in the H3R resulted in the very
interesting lead FUB2.922. However, it had not been developed due
to its high hERG affinity. To overcome this issue, we build a QSAR
tool specially dedicated to H3R ligands. We chose a model that is
easily understandable and useful for the non-specialist in molecu-
lar modeling, that is, RP. The trees presented in Fig. 1 are thus prac-
tically interpretable, and might be very useful for medicinal
chemists involved in H3 area. They are statistically very robust,
as judged by the accuracy calculated from the training set. An
external test set constituted of close analogs of FUB2.922 demon-
strate their predictive capacity.
Variation of the lipophilicity of the biphenyl had clearly shown
a strong influence on hERG affinity, as revealed by the structure–
activity relationships presented here. This brings to compound
17, the lead of a new chemical series described in the companion
paper.20
15. Supplementary data
16. Ermondi, G.; Visentin, S.; Caron, G. Eur. J. Med. Chem. 2009, 44, 1926.
17. All the descriptors are available in Discovery Studio or Pipeline Pilot (Accelrys,
San Diego, CA). Apol is the sum of atomic polarizabilities, A log P is the log of
octanol-water partition calculated coefficient, Log D is the same partition
coefficient taking into account the ionization state of the molecule, molecular
solubility is the log of calculated solubility (in mol/L), and HBD Count is the
sum of hydrogen-bond donor functions. sum aaN is the electrotopological
state sum of aromatic nitrogens (Hall L., Kier L., J. Chem. Inf. Comput. Sci. 2000,
40, 784–791.). CIC is a graph-theoretical topological descriptor (Bonchev, D.,
Information Theoretic Indices for Characterization of Chemical Structures,
Chemometrics Series, ed. D.D. Bawden, 1983 , 5, New York: Research Studies
Press Ltd.).
Acknowledgments
There is almost a correlation between sum_aaN and the number of aromatic
We thank Isabelle Delimoge, Marie-Noëlle Legave, Stéphanie Le
Meur, Philippe Guibet, Benoît Messager, and Sébastien Nicolas for
their expert technical assistance.
nitrogens, so
a simple estimate of sum aaN may be the following:
sum_aaN = 0.24 + (4.37 ⁄number of aromatic nitrogens).
As an approximation, we can consider that Apol exceed the threshold of
15473.8 if one of the above conditions is established: (i) either the molecule
contains 10 or more double-bonds, (ii) or the molecule contains 8 or more
double-bonds and
nitrogen
1 or more halogen, sulfur, phenolic oxygen or aniline
Supplementary data
18. Quality of the models is evaluated by their accuracy, sensitivity and specificity,
as follow:
Supplementary data associated with this article can be found, in
the online version, at doi:10.1016/j.bmcl.2011.07.006.
TP þ TN
TP
TPþ FN
TN
TN þ FP
Accuracy ¼
; Sensitivity ¼
; Specificity ¼
TP þ TN þ FP þ FN
References and notes
true positives (TP) and true negatives (TN) are hERG binders or not binders
at a 7 M threshold, correctly classified by the model. False negatives (FN)
l
1. Brioni, J. D.; Esbenshade, T.; Garrison, T.; Bitner, S.; Cowart, M. J. Pharmacol. Exp.
Ther. doi: 10.1124/jpet.110.166876.
2. Leurs, R.; Bakker, R. A.; Timmerman, H.; de Esch, I. J. P. Nat. Rev. Drug Discov.
2005, 4, 107.
3. Esbenshade, T. A.; Fox, G. B.; Cowart, M. D. Mol. Interventions 2006, 6, 77.
4. Sander, K.; Kottke, T.; Stark, H. Biol. Pharm. Bull. 2008, 31, 2163.
5. Hough, L.; Rice, F. L. J. Pharmacol. Exp. Ther. doi: 10.1124/jpet.110.171264.
6. Ganellin, C. R.; Leurquin, F.; Piripitsi, A.; Arrang, J.-M.; Garbarg, M.; Ligneau, X.;
Schunack, W.; Schwartz, J.-C. Arch. Pharm. Pharm. Med. Chem. 1998, 331, 395.
and false positives (FP) are hERG binders or not binder misclassified.
19. Levoin, N.; Calmels, T.; Poupardin-Olivier, O.; Labeeuw, O.; Danvy, D.; Robert,
P.; Berrebi-Bertrand, I.; Ganellin, C. R.; Schunack, W.; Stark, H.; Capet, M. Arch.
Pharm. Pharm. Med. Chem. 2008, 341, 610.
20. Labeeuw, O.; Levoin, N.; Poupardin-Olivier, O.; Calmels, T.; Ligneau, X.; Berrebi-
Bertrand, I.; Robert P.; Lecomte, J.-M.; Schwartz, J.-C.; Capet, M. Bioorg. Med.
Chem. Lett. doi:10.1016/j.bmcl.2011.06.102.