Synthesis and structure-activity relationships of a new model of arylpiperazines. 6. Study of the 5-HT1A/α1-adrenergic receptor affinity by classical Hansch analysis, artificial neural networks, and computational simulation of ligand recognition

Article abstract of DOI:10.1021/jm000930t

A classical quantitative structure-activity relationship (Hansch) study and artificial neural networks (ANNs) have been applied to a training set of 32 substituted phenylpiperazines with affinity for 5-HT_1A and α₁-adrenergic receptors, to evaluate the structural requirements that are responsible for 5-HT_1A/α₁ selectivity. The resulting models provide a significant correlation of electronic, steric, and hydrophobic parameters with the biological affinities. Although the derived linear Hansch correlations give good statistics and acceptable predictions, the introduction of nonlinear relationships in the analysis gives more solid models and more accurate predictions. In the ANN models on the basis of the obtained 3D plots, the 5-HT_1A affinity has a nonlinear dependence on F, V_o, V_m, and π_o, although the nonlinear relationship is not far from a planar one. The α₁-adrenergic receptor affinity has a clear nonlinear dependence on F, V_o, V_m, π_o, and π_m. A comparison of both analyses gives an additional understanding for 5-HT_1A/α₁ selectivity: (a) high F values increase the binding affinity for 5-HT_1A receptors and decrease the affinity for α₁ sites; (b) the hydrophobicity at the meta-position has only influence for the α₁-adrenergic receptor; (c) the meta-position seems to be implicated in the 5-HT_1A/α₁ selectivity. While the 5-HT_1A receptor is able to accommodate bulky substituents in the region of its active site, the steric requirements of the α₁-adrenergic receptor at this position are more restricted. This information was used for the design of the new ligand EF-7412 (33) (5-HT_1A: K_{i exptl} = 27 nM, α₁: K_{i exptl} > 1000 nM; 5-HT_1A: K_{i pred ANN} = 36 nM, α₁: K_{i pred ANN} = 2745 nM) which was characterized as an antagonist in vivo in pre- and postsynaptic 5-HT_1AR sites. Computational simulations of the complex between EF-7412 (33) and a 3D model of the transmembrane domain of the 5-HT_1A receptor allowed us to define the molecular details of the ligand-receptor interaction that includes: (i) the ionic interaction between the protonated amine of the ligand and Asp 3.32; (ii) the hydrogen bonds between the m-NHSO₂Et group of the ligand and Asn 7.39; and the hydrogen bonds between the hydantoin moiety of the ligand and (iii) Thr 3.37, (iv) Ser 5.42, and (v) Thr 5.43. These QSAR and ANN results in combination with computational simulations of ligand recognition will be useful for the design of potent selective 5-HT_1A ligands.

Full text of DOI:10.1021/jm000930t

198
J . Med. Chem. 2001, 44, 198-207
Syn th esis a n d Str u ctu r e-Activity Rela tion sh ip s of a New Mod el of
Ar ylp ip er a zin es. 6. Stu d y of th e 5-HT1A/r1-Ad r en er gic Recep tor Affin ity by
Cla ssica l Ha n sch An a lysis, Ar tificia l Neu r a l Netw or k s, a n d Com p u ta tion a l
Sim u la tion of Liga n d Recogn ition †
Mar´ıa L. Lo´pez-Rodr´ıguez,*,§ M. J ose´ Morcillo,‡ Esther Ferna´ndez,§ M. Luisa Rosado,§ Leonardo Pardo, and
Klaus-J u¨rgen Schaper#
Departamento de Quı´mica Orga´nica I, Facultad de Ciencias Qu´ımicas, Universidad Complutense, 28040 Madrid, Spain,
Facultad de Ciencias, Universidad Nacional de Educacio´n a Distancia, 28040 Madrid, Spain, Laboratori de Medicina
Computacional, Unitat de Bioestadı´stica, Facultat de Medicina, Universitat Autonoma de Barcelona, 08193 Bellaterra, Spain,
and Research Center Borstel, D-23845 Borstel, Germany
Received February 23, 2000
A classical quantitative structure-activity relationship (Hansch) study and artificial neural
networks (ANNs) have been applied to a training set of 32 substituted phenylpiperazines with
affinity for 5-HT1A and R1-adrenergic receptors, to evaluate the structural requirements that
are responsible for 5-HT1A/R1 selectivity. The resulting models provide a significant correlation
of electronic, steric, and hydrophobic parameters with the biological affinities. Although the
derived linear Hansch correlations give good statistics and acceptable predictions, the
introduction of nonlinear relationships in the analysis gives more solid models and more
accurate predictions. In the ANN models on the basis of the obtained 3D plots, the 5-HT1A
affinity has a nonlinear dependence on F, Vo, Vm, and πo, although the nonlinear relationship
is not far from a planar one. The R1-adrenergic receptor affinity has a clear nonlinear dependence
on F, Vo, Vm, πo, and πm. A comparison of both analyses gives an additional understanding for
5-HT1A/R1 selectivity: (a) high F values increase the binding affinity for 5-HT1A receptors and
decrease the affinity for R1 sites; (b) the hydrophobicity at the meta-position has only influence
for the R1-adrenergic receptor; (c) the meta-position seems to be implicated in the 5-HT1A/R1
selectivity. While the 5-HT1A receptor is able to accommodate bulky substituents in the region
of its active site, the steric requirements of the R1-adrenergic receptor at this position are more
restricted. This information was used for the design of the new ligand EF-7412 (33) (5-HT1A
:
Ki exptl ) 27 nM, R1: Ki exptl > 1000 nM; 5-HT1A: Ki pred ANN ) 36 nM, R1: Ki pred ANN ) 2745 nM)
which was characterized as an antagonist in vivo in pre- and postsynaptic 5-HT1AR sites.
Computational simulations of the complex between EF-7412 (33) and a 3D model of the
transmembrane domain of the 5-HT1A receptor allowed us to define the molecular details of
the ligand-receptor interaction that includes: (i) the ionic interaction between the protonated
amine of the ligand and Asp 3.32; (ii) the hydrogen bonds between the m-NHSO2Et group of
the ligand and Asn 7.39; and the hydrogen bonds between the hydantoin moiety of the ligand
and (iii) Thr 3.37, (iv) Ser 5.42, and (v) Thr 5.43. These QSAR and ANN results in combination
with computational simulations of ligand recognition will be useful for the design of potent
selective 5-HT1A ligands.
In tr od u ction
chemical parameters, cross-product terms must be
included in the regression analysis. This adds complex-
ity to data analysis and interpretation. During the past
years artificial neural networks (ANNs) have been
applied successfully in the QSAR field.2,3 It has been
demonstrated that this new technique is often superior
to the traditional Hansch approach, providing more
accurate predictions. The advantage of ANNs is that
with the presence of hidden layers, neural networks are
implicitly able to perform nonlinear mapping of the
physicochemical parameters to the corresponding bio-
logical activities.4
The formulation of quantitative structure-activity
relationships (QSAR) is an important tool in the devel-
opment of new agents because it can keep the number
of synthesized and tested compounds to a minimun.
Hansch1 demonstrated that the biological activities of
drug molecules can be correlated by a linear combina-
tion of the physicochemical parameters of the corre-
sponding drug. However in the cases that biological
activities do not vary in a linear manner with physico-
† Dedicated to Professor J ose´ Luis Soto on the occasion of his 70th
birthday.
* To whom correspondence should be addressed. Phone: 34-91-
3944239. Fax: 34-91-3944103. E-mail: mluzlr@eucmax.sim.ucm.es.
§ Universidad Complutense.
The preceding paper in this series describes the
design and synthesis of a test series of 32 arylpiperazine
derivatives with affinity for 5-HT1A and R1-adrenergic
receptors. The wide range of receptor affinities of these
‡ Universidad Nacional de Educacio´n a Distancia.
Universitat Autonoma de Barcelona.
#
Research Center Borstel.
10.1021/jm000930t CCC: $20.00 © 2001 American Chemical Society
Published on Web 12/22/2000

Synthesis and SARs of New Arylpiperazines. 6
J ournal of Medicinal Chemistry, 2001, Vol. 44, No. 2 199
Ta ble 1. 5-HT1A and R1-Adrenergic Receptor Binding Dataa
5-HT1A
Ki (nM)
R1
Ki (nM)
(SEMb
compd
X
m n
R
(SEMb pKic
pKic
1
2
3
4
5
6
7
8
9
-(CH2)3- 0 3 o-OBu
-(CH2)3- 0 3 o-CONHPr
-(CH2)3- 0 3 m-NH2
-(CH2)3- 0 3 m-Br
-(CH2)4- 0 3 o-CH3
-(CH2)4- 0 3 o-COOPr
-(CH2)4- 0 3 o-CN
15 ( 2
808
1182
34 ( 9
220 ( 2
46 ( 6
7.82
6.09
5.93
7.47
6.66
7.34
6.4 ( 4.2 8.19
608
393
6.22
6.41
25 ( 2 7.61
5.7 ( 0.2 8.25
13 ( 1 7.89
8.2 ( 0.8 8.09
5.35
17 ( 1 7.76
12 ( 7 7.93
17 ( 3 7.76
65 ( 12 7.19
-(CH2)4- 0 3 m-NHCOPri 776
-(CH2)3- 1 3 o-CH3 48 ( 2
6.11 4430
7.31
7.26
7.56
F igu r e 1. Compounds 1-32.
10 -(CH2)3- 1 3 o-COOPr
11 -(CH2)3- 1 3 o-CN
55 ( 3
27 ( 5
12 -(CH2)3- 1 3 m-NHCOPri 266 ( 72 6.57 1440
5.84
29 ( 1 7.53
13 ( 15 7.89
compounds, which is due to a rational choice of substit-
uents, makes this material suitable for a QSAR inves-
tigation.
13 -(CH2)4- 1 3 o-OCH3
14 -(CH2)4- 1 3 o-OBu
15 -(CH2)4- 1 3 o-CONHPr 4614
143 ( 8
92 ( 15 7.04
5.34
179 ( 21 6.75
16 ( 2 7.78
4.8 ( 1.4 8.32
4.0 ( 0.8 8.40
6.84
811
6.09
339 ( 3 6.47
7.0 ( 2.1 8.15
41 ( 5 7.39
4.9 ( 1.4 8.31
A classical Hansch regression analysis and ANNs
were employed to sort out the structural requirements
that are responsible for 5-HT1A/R1 selectivity. This type
of analysis leads to the identification and characteriza-
tion of the physicochemical factors of the pharmaco-
phore on which biological affinity and selectivity (5-HT1A
vs R1 affinity) are based. In the present paper we report
the results of this study, which provided information
for the design of the ligand EF-7412 (33) (5-HT1A: Ki
) 27 nM; R1: Ki > 1000 nM).
The 5-HT1A receptor (5-HT1AR) belongs to the family
of G protein-coupled receptors (GPCRs). Recently, the
detailed 3D structure of bovine rhodopsin (RHO) was
determined at 2.8 Å resolution.5 This major break-
through has confirmed that RHO and the RHO-like
family of GPCRs are formed by seven antiparallel
R-helical transmembrane domains connected by hydro-
philic loops. The highly conserved motifs that character-
ize the RHO-like family of GPCRs are all located in the
transmembrane region, which suggests a common trans-
membrane structure. In contrast the extracellular do-
main diverges in the GPCRs family.6 Here we present
the derivation, from the crystal structure of RHO,5 of a
computational model of the complex between EF-7412
(33) and the transmembrane domain of the 5-HT1AR.
The computed model rests on the definition of the
putative residues of the ligand-binding site guided by
published results of modifications of GPCRs using
methods of molecular biology. These insights obtained
from classical Hansch analysis, ANNs, and computa-
tional simulations of ligand recognition can be used to
guide the design of ligands with predetermined affinities
and selectivity.
16 -(CH2)4- 1 3 m-CF3
17 -(CH2)3- 0 4 o-CH3
18 -(CH2)3- 0 4 o-COOPr
19 -(CH2)3- 0 4 o-CN
20 -(CH2)3- 0 4 m-NHCOPri 193 ( 8
6.72 5425
5.27
21 -(CH2)4- 0 4 o-OBu
22 -(CH2)4- 0 4 o-CONHPr
23 -(CH2)4- 0 4 m-NH2
24 -(CH2)4- 0 4 m-Br
25 -(CH2)3- 1 4 o-OCH3
26 -(CH2)3- 1 4 o-OBu
2.2 ( 0.1 8.66
5.0 ( 0.3 8.30
378 ( 106 6.42
68 ( 18 7.17
5.4 ( 1.1 8.27
521
6.28
231 ( 6 6.64
5.9 ( 0.3 8.23
25 ( 1 7.59
12 ( 1
7.92
8.7 ( 1.1 8.06 10.5 ( 8.6 7.98
5.91 1354 5.87
27 -(CH2)3- 1 4 o-CONHPr 1232
28 -(CH2)3- 1 4 m-CF3
29 -(CH2)4- 1 4 o-CH3
30 -(CH2)4- 1 4 o-COOPr
31 -(CH2)4- 1 4 o-CN
5.5 ( 0.6 8.26
72 ( 11 7.14
27 ( 2 7.56
392 ( 33 6.41
30 ( 5 7.52
32 ( 8
28 ( 6
15 ( 1
7.50
7.55
7.82
32 -(CH2)4- 1 4 m-NHCOPri 198 ( 64 6.70 16140
4.79
a
Ki ( SEM values were derived from 2-4 experiments per-
b
formed in triplicate. SEM is indicated when Ki values were
obtained from complete DRCs. c pKi values were calculated from
Ki values in M.
only one point of the curve was available was performed
by the application of a method,7 which consists of a
simultaneous nonlinear regression analysis of all dose-
response curves (DRCs) of the drug series using eq 1:
% SB ) 100(1 - Cib/(IC50j + Cib))
(1)
b
where SB is the specific binding of radioligand, b is the
slope, i ) 1...n (measurements), and j ) 1...m (number
of compounds).
This analysis is performed under the assumption that
all derivatives present the same mechanism of action
within the given test model (i.e., parallel sigmoidal
DRCs, identical Hill coefficients/slopes). This approach
requires that complete DRCs (g3 data pairs) are avail-
able for some analogues. Missing IC50 values are
obtained from “fragmentary” DRCs by a computational
parallel shift of complete DRCs. IC50 values were
converted to Ki values (Table 1) using the Cheng-
Prusoff equation.8
Each chemical structure was described by six phys-
icochemical parameters and three indicator variables.
As electronic descriptors, the field and resonance con-
stants of Swain and Lupton (F , R ) were used. To attend
a better quantification of the electrostatic effects, the
ortho- and meta-position factors calculated by Williams
and Norrington9 were introduced. The van der Waals
volumes (Vo, Vm) calculated with the SYBYL program10
were employed as steric parameters. The hydrophobic
effects exerted by the ortho- and meta-substituents were
measured using Hansch πo and πm constants.11
Ma ter ia ls a n d Meth od s
Da ta Set. The training set of 32 phenylpiperazines
substituted in the ortho- and meta-positions (Figure 1)
was evaluated for in vitro 5-HT1A and R1-adrenergic
receptor affinities by radioligand binding assays. The
compounds were first tested at the fixed dose of 10-6
M, and for those that in this prescreening process
presented high activity (displacement of the radioligand
g55%), the dose-response curves were determined.
However, for the members of the series showing low
activity (displacement < 55%) the binding constants
were not determined. Nevertheless, low-activity com-
pounds of a drug series allow the activity scale to be
expanded. Obviously a broad range of activity data
facilitates the recognition of QSAR relationships. The
estimation of the IC50 values for the compounds where

200 J ournal of Medicinal Chemistry, 2001, Vol. 44, No. 2
Lo´pez-Rodrı´guez et al.
Ta ble 2. Indicator Variables and Physicochemical Parameters
Used in the QSAR Study
The hidden and output neurons received an additional
constant input, the “bias”. For the input layer a linear
transfer function was used, and for the hidden and
output layer the following sigmoid was taken, to enable
modeling of nonlinear relations:
b
b
c
c
compd
R
IA IB In)3
Fa
Ra
Vo
Vm
πo
πm
1
2
3
4
5
6
7
8
9
o-OBu
0
0
0
0
1
1
1
1
0
0
0
0
1
1
1
1
0
0
0
0
1
1
1
1
0
0
0
0
1
1
1
1
0
0
0
0
0
0
0
0
1
1
1
1
1
1
1
1
0
0
0
0
0
0
0
0
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0.513 -0.475 65.3
0.00 1.71 0.00
0.00 -0.09 0.00
o-CONHPr
m-NH2
m-Br
0.703 0.038 69.0
0.036 -0.236 0.00 11.4
0.712 -0.061 0.00 17.2
-0.065 -0.122 16.4
0.689 0.121 69.3
1.057 0.159 15.9
0.00 -1.23
0.00 0.86
outj ) f(netj) ) 1/(1 + e-netj
)
(2)
(3)
o-CH3
0.00 0.56 0.00
0.00 1.17 0.00
0.00 -0.57 0.00
o-COOPr
o-CN
net ) (w out ) + bias
m-NHCOPri
o-CH3
0.461 -0.095 0.00 70.8
0.00 0.11
∑
j
ji
i
i
i
-0.065 -0.122 16.4
0.689 0.121 69.3
1.057 0.159 15.9
0.00 0.56 0.00
0.00 1.17 0.00
0.00 -0.57 0.00
10 o-COOPr
The input to neuron j, netj, is calculated by eq 3 where
outi represents the output values from neurons i con-
nected to j, wji is the connection weight between neurons
i and j, and biasi is the intercept of the linear combina-
tion ∑i(wji outi) which can be regarded as an extra
weight of the neuron with a constant output of 1.
Feedforward calculations and error BP were employed
in this study. The BP algorithm is based on a gradient
descent method to minimize an error function E with
respect to the connection weights (wji) and biases from
the output layer toward the input layer. E is given as
follows:
11 o-CN
12 m-NHCOPri
13 o-OCH3
14 o-OBu
0.461 -0.095 0.00 70.8
0.00 0.11
0.515 -0.431 22.7
0.513 -0.475 65.3
0.703 0.038 69.0
0.00 -0.02 0.00
0.00 1.71 0.00
0.00 -0.09 0.00
15 o-CONHPr
16 m-CF3
17 o-CH3
0.618 0.064 0.00 24.2
0.00 0.88
-0.065 -0.122 16.4
0.689 0.121 69.3
1.057 0.159 15.9
0.461 -0.095 0.00 70.8
0.513 -0.475 65.3
0.703 0.038 69.0
0.036 -0.236 0.00 11.4
0.712 -0.061 0.00 17.2
0.515 -0.431 22.7
0.513 -0.475 65.3
0.703 0.038 69.0
0.00 0.56 0.00
0.00 1.17 0.00
0.00 -0.57 0.00
0.00 0.11
0.00 1.71 0.00
18 o-COOPr
19 o-CN
20 m-NHCOPri
21 o-OBu
22 o-CONHPr
23 m-NH2
24 m-Br
25 o-OCH3
26 o-OBu
27 o-CONHPr
28 m-CF3
29 o-CH3
0.00 -0.09 0.00
0.00 -1.23
0.00 0.86
0.00 -0.02 0.00
0.00 1.71 0.00
0.00 -0.09 0.00
i)N
i)N
0.618 0.064 0.00 24.2
0.00 0.88
[y (obsd) - y (cald)]2
(4)
2
E )
E
)
-0.065 -0.122 16.4
0.689 0.121 69.3
1.057 0.159 15.9
0.00 0.56 0.00
0.00 1.17 0.00
0.00 -0.57 0.00
∑
∑
i
i
i
30 o-COOPr
31 o-CN
i)1
i)1
32 m-NHCOPri
0.461 -0.095 0.00 70.8
0.00 0.11
where N is the number of objects in the training set, Ei
represents the error of the ith training data, yi(cald) is
the calculated output value for the ith training data
from the output neuron, and yi(obsd) is the observed
value. Once the actual error produced by the network
is known, partial derivatives of the error function are
used to change the connection weights from their
initially assigned random values according to the fol-
lowing equation:2
a
b
Data taken from ref 9. Calculated with the SYBYL program.
c Data taken from ref 11.
Ta ble 3. Correlation Matrix for the Indicator Variables and
Physicochemical Descriptors
IA
IB
In)3
F
R
Vo
Vm
πo
πm
IA
IB
In)3
F
R
Vo
Vm
πo
πm
1
0
0
0
0
0
0
0
0
1
0
1
0
0
0
0
0
0
0.075
-0.021
0.048
-0.012
-0.002
0.180
1
l
l(previous)
∆wji ) ηδlj outli-1 + µ∆wji
(5)
0.480
0.243 -0.052
-0.096 0.019 -0.547
-0.316 -0.500 0.562 -0.238
0.366 0.232 -0.110 0.178 -0.048
1
1
1
in which ∆wij denotes the applied adaptation of the
weights from unit i to unit j in the next layer, δlj is the
calculated correction factor of these weights, outi is the
output of unit i, η is the learning rate, and µ is the
momentum term. The learning rate, η, which deter-
mines the speed at which the weights change, is used
to dampen the rate of change of the weights from
iteration to iteration, whereas the momentum, µ, adds
a certain amount of the weight change in the previous
presentation to keep the direction of change from
varying too fast, preventing possible oscillations.
The biological affinity (pKi) data set was scaled to lie
in the range of -0.9 to +0.9, so as to allow margin for
network predictions at the extremes. The input vari-
ables were scaled to lie in the range -0.75 to +0.75.
The error BP was performed after each iteration. The
networks were trained until no further significant
change in the standard deviation, s, between observed
and calculated affinities was observed. The optimal
number of hidden neurons was determined by gradually
eliminating nodes.15 In all calculations random numbers
in the range -0.1 to +0.1 were used to initialize the
network weights. The momentum parameter was set at
0.7, while the learning rate was initially set at 0.1 and
was reduced to 0.01 or 0.005 if oscillations in s were
observed.
1
1
The indicator variables IA, IB and In)3 indicate the
presence or absence of certain structural elements: IA
) 0 or 1 for X ) -(CH2)3- or -(CH2)4-, IB ) 0 or 1 for
m ) 0 or 1, and In)3 distinguishes between compounds
where n ) 3 (In)3 ) 1) and n ) 4 (In)3 ) 0). The values
of the input parameters as well as the intercorrelation
among the physicochemical descriptors and the indica-
tor variables are shown in Tables 2 and 3.
H a n sch Met h od . A classical Hansch multivariate
regression analysis using the least-squares method12
was used to derive QSAR equations for our data set.
The level of significance of each coefficient is judged by
statistical procedures such as the Student t and F
tests.13
To obtain suitable equations two factors must be
taken into account. First, the ratio of compounds to
variables should be greater than 5, and second, a
minimal intercorrelation among the independent vari-
ables should be observed.
ANN. The neural network employed for this modeling
was a fully connected three-layer network (input, hid-
den, output) trained by back-propagation (BP) of error.14

Synthesis and SARs of New Arylpiperazines. 6
J ournal of Medicinal Chemistry, 2001, Vol. 44, No. 2 201
Ta ble 4. Equations for QSAR of the 5-HT1A Serotoninergic Receptor
equationsa
r
SE
F
pKi ) 7.66((0.18; 0.38; 42.5) - 0.251((0.114; 0.237; 2.19)IA - 0.145((0.117; 0.244; 1.23)IB
-
0.944
0.325
19.923
0.762((0.114; 0.237; 6.64)In)3 + 1.62((0.26; 0.54; 6.23)F + 0.078((0.353; 0.733; 0.221)R -
0.0284((0.0034; 0.0071; 8.24)Vo - 0.0195((0.0029; 0.0062; 6.53)Vm + 1.24((0.15; 0.31;
8.23)πo +0.136((0.160; 0.332; 0.848)πm (6)
pKi ) 7.65((0.17; 0.36; 45.0) - 0.251((0.112; 0.232; 2.23)IA - 0.147((0.114; 0.237; 1.28)IB
-
0.944
0.318
23.374
0.762((0.114; 0.232; 6.78)In)3 + 1.63((0.25; 0.52; 6.47)F - 0.0282((0.0033; 0.0068; 8.54)Vo
- 0.0195((0.0029; 0.0060; 6.67)Vm + 1.22((0.13; 0.27; 9.24)πo + 0.143((0.154;
0.318; 0.928)πm (7)
pKi ) 7.60((0.16; 0.34; 47.5) - 0.251((0.112; 0.231; 2.24)IA - 0.128((0.112; 0.231; 1.14)IB
-
0.941
0.938
0.317
0.319
26.743
30.613
0.762((0.112; 0.231; 6.80)In)3 + 1.74((0.21; 0.45; 8.00)F - 0.0292((0.0031; 0.0064; 9.30)Vo
- 0.0192((0.0029; 0.0060; 6.63)Vm + 1.26((0.12; 0.25; 10.08)πo (8)
pKi ) 7.55((0.16; 0.33; 47.2) - 0.251((0.112; 0.232; 2.23)IA - 0.762((0.112; 0.232; 6.76)In)3
1.73((0.21; 0.45; 7.90)F - 0.0292((0.0031; 0.0065; 9.28)Vo - 0.0193((0.0029; 0.0060;
6.61)Vm + 1.26((0.12; 0.26; 10.02)πo (9)
+
a
n ) 32.
Molecu la r Mod el of th e Com p lex betw een EF -
7412 (33) a n d 5-HT1AR. The 3D model of the trans-
membrane domain of the 5-HT1AR was constructed by
computer-aided model building techniques from the
crystal structure of RHO5 (PDB access number 1F88).
Conserved residues Asn 55 (residue number in the PDB
file) or Asn 1.50 (nomenclature of Ballesteros & Wein-
stein),16 Asp 83 or Asp 2.50, Arg 135 or Arg 3.50, Trp
161 or Trp 4.50, Pro 215 or Pro 5.50, Pro 267 or Pro
6.50, and Pro 303 or Pro 7.50 were employed in the
alignment of RHO and human 5-HT1AR transmembrane
sequences. The obtained molecular model was energy
minimized (5000 steps), heated from 0 to 600 K in 30
ps, equilibrated from 30 to 150 ps at 600 K, cooled to
300 K from 150 to 210 ps, and equilibrated from 210 to
300 ps at 300 K. During these processes the CR atoms
and the conserved residues were kept fixed at the
positions originally determined in the crystal structure
of RHO.5 This procedure allows the arbitrarily posi-
tioned amino acid side chains to adopt an energetically
favorable conformation.
The mode of recognition of the ligands was first
determined by ab initio geometry optimization with the
3-21G* basis set. The model system consisted on Asp
3.32 and Asn 7.39 (only the CR atom of the backbone is
included) of the 5-HT1AR and the ligands pindolol (the
indole ring and the terminal -CH(CH3)2 were replaced
by methyl groups) and EF-7412 (33) (the hydantoin
moiety plus the -(CH2)4- chain were replaced by a
methyl group). All free valences were capped with
hydrogen atoms. The CR atoms of Asp 3.32 and Asn 7.39
were positioned and kept fixed at the positions previ-
ously obtained.5 The optimized reduced model of the
ligand-receptor complex was used to position the
complete structure of EF-7412 (33) inside the previously
equilibrated transmembrane domain of the 5-HT1AR.
Subsequently, the complete system was energy mini-
mized (5000 steps), heated (from 0 to 300 K in 15 ps)
and equilibrated (from 15 to 500 ps). The geometries
obtained during the trajectory from 400 to 500 ps (a total
of 20) were averaged and energy minimized. The CR
atoms were kept fixed at the positions originally deter-
mined5 because of the absence of the lipid environment
in the molecular dynamics simulations.
step, and a 13 Å cutoff for nonbonded interactions.
Parameters for EF-7412 (33) were adapted from the
Cornell et al. force field19 using RESP point charges.20
Resu lts a n d Discu ssion
Ha n sch An a lysis. 5-HT1A Ha n sch Mod el: Hansch
equations for the serotoninergic receptor were derived
from multiple regression analysis between the pKi
values at the 5-HT1A sites and the physicochemical
parameters discussed in the Materials and Methods
section. In eqs 6-9 reported in Table 4, n is the number
of compounds, r is the correlation coefficient, SE is the
standard error, and F is the value of the F test. The
significance of each regression coefficient is expressed
by its standard error, the partial correlation coefficient,
and the Student t, all of them in parentheses. Analysis
of the initial eq 6 reveals that IB, R, and πm are not
statistically significant. The stepwise elimination of
these descriptors from the analysis led to the eqs 7-9
in which electronic, steric, and hydrophobic effects are
involved. Equation 9 was adopted as the final model,
first because of its greater simplicity and second due to
its satisfactory overall correlation.
Equation 9 shows that the structural features X )
-(CH2)4- and especially n ) 3 exert a negative effect
on the affinity. High and positive values of F along with
hydrophobic substituents at the ortho-position increase
the 5-HT1A affinity. The above results also indicate that
bulky substituents should be avoided for high affinity.
Among the substituents that keep these requirements,
we have selected o-Br and o-OPh in order to evaluate
the predictive capacity of the model. Two new ligands,
34 and 35, were synthesized. The experimental and
predicted values for both tested compounds are pre-
sented in Table 5, showing an acceptable overall agree-
ment. The fact that the affinity of 35 is greater than
expected for such a bulky substituent as the o-OPh
indicates that the receptor is able to accommodate bulky
substituents in this region of its active site.
r1 Ha n sch Mod el: An approach identical to that
used in the development of the QSAR for 5-HT1AR
ligands was applied to the binding of the same ligands
to the R1-adrenergic receptor, leading to the stepwise
development of eqs 10-14 (Table 6). Following the same
criteria as for the serotoninergic receptor, eq 14 was
selected as the best model. This equation indicates that
the length of the alkyl chain seems to be irrelevant,
while the presence of the hydantoin moiety exerts a
Quantum mechanical calculations were performed
with the GAUSSIAN-98 system of programs.17 Molec-
ular dynamics simulations were run with the Sander
module of AMBER5,18 the all-atom force field,19 SHAKE
bond constraints in all bonds, a 2 fs integration time

202 J ournal of Medicinal Chemistry, 2001, Vol. 44, No. 2
Lo´pez-Rodrı´guez et al.
Ta ble 5. Experimental and Predicted 5-HT1A Ki (eq 9) and R1
Ki (eq 14) Values of Compounds 34 and 35
Using this technique, the weights are adjusted in a way
that their differences are increased, allowing to distin-
guish which of them are less important and, therefore,
to prune the corresponding input neuron. In Table 7 are
shown the connection weights between the input layer
and the hidden layer. As all inputs were scaled to an
input range of -0.75 to +0.75, the magnitude of the
connection weights after training gave a direct indica-
tion of the contribution of each input parameter. Thus
the input parameters with the smallest connection
weights contributed least. To prune a given input
parameter, the connections to the three hidden neurons
should be small. Only two descriptors, R and πm,
fulfilled this condition. The elimination of two input
neurons yields a 7-3-1 network with F ) 1.1. This F
value is still low, and for the final model an architecture
of 7-2-1 was chosen with F ) 1.7. Four sets of
randomly assigned connection weights and biases were
used to estimate the correlation coefficients correspond-
ing to the global minimum of the network. For the final
model r, r2, and s of the training set are 0.983, 0.966,
and 0.149, respectively (Table 8).
5-HT1A
Ki exptl Ki pred Ki exptl
R1
(
(
Ki pred
compd
R
F
πo
Vo SEM (nM) (nM) SEM (nM) (nM)
34
35
o-Br
0.907 0.86 17.2 3.4 ( 0.6
0.4
1.2
6.2 ( 1.8
2.4 ( 0.6
0.4
2.5
o-OPh 0.932 2.08 89.8 0.6 ( 0.2
slightly positive effect at the R1-adrenergic active site.
Electronic and hydrophobic requirements suggest that
high and positive values of F as well as positive
contributions of π at the ortho-position increase the R1-
adrenergic receptor affinity. Equations 14 and 9 show
that electronic and hydrophobic effects play a similar
role in increasing affinity at both R1-adrenergic and
5-HT1A sites. With respect to the steric effects, bulky
substituents decrease the affinity. However, it is inter-
esting to note that negative steric effects are especially
significant at the meta-positon in eq 14, which indicates
an important difference between the binding require-
ments at the 5-HT1A active site and the R1-adrenergic
receptor. Thus, the meta-position seems to be implicated
in the 5-HT1A/R1 selectivity. Again, compounds 34 and
35 were used to investigate the predictive power of the
derived model (Table 5).
To validate the model, the affinities for compounds
34 and 35 were predicted (Table 9). In light of these
results, one can be confident that the neural network
is able to provide reliable predictions of biological
affinities of novel arylpiperazines.
Dep en d en ce of Biologica l Activity on th e P h ys-
icoch em ica l P a r a m eter s: The relative importance of
each parameter in affecting the biological activity was
investigated. It is known that the ANN model is difficult
to interpret because it consists of some hidden nodes
which produce nonlinear outputs for describing the
nonlinear behavior in the data set. However, the varia-
tion of the affinity can be monitored in 3D diagrams by
changing the values of two inputs while keeping the
remaining inputs of the neural network constant at one-
half of their maximum ranges. On the basis of the
obtained plots, the 5-HT1A affinity seems to have a
moderate nonlinear dependence on F, Vo, Vm and πo,
although the nonlinear relationship is not far from the
planar one. This fact explains why the Hansch model
shows an acceptable predictive power. It is interesting
to note that the neural network is consistent with some
of Hansch findings. First, both the neural network
model and linear regression analysis agreed that posi-
tive values of F together with hydrophobic substituents
at the ortho-position give high affinities, and second,
bulky substituents exert a negative effect. The main
discrepancy between the neural network model and the
regression analysis seems to be the effect of IB. While
the size of the ring is irrelevant for the Hansch model,
in the neural network model the highest values of
affinity are observed for IB ) 0. These findings demon-
strate that although the Hansch model is able to
describe some features governing the 5-HT1A affinity,
the ANN model reveals a certain nonlinear tendency,
offering a better insight into the factors that are
resposible for the serotoninergic affinity and shows an
enhanced predictive capacity.
Although the MLR studies conducted for the 5-HT1A
and R1-adrenergic receptors have generated equations
with an acceptable predictive power, it was desirable
to investigate possible nonlinear relationships that may
not be explained by the linear Hansch models.
ANN An a lysis. 5-HT1A ANN Mod el: A three-layer
BP learning network was used to study the set of 32
phenylpiperazines described above. The connection
between the three different layers was complete. Ini-
tially the number of neurons in the input layer was
equal to 9, identical to the number of molecular descrip-
tors and indicator variables, whereas the output layer
had only one neuron. The number of neurons in the
hidden layer was determined by trial and error. The best
correlation coefficient and lower standard deviation was
obtained with three hidden layer neurons (9-3-1). The
number of nodes in the hidden layer is an important
factor determining the network’s performance. It was
found that too many nodes cause the network to
memorize the data set (overfitting). However, networks
with few nodes may be insufficient to use all the
information of the data set (underfitting) and generali-
zation is poor. It is desirable to establish a network
which generalizes the input patterns rather than merely
memorizing them. Previous studies conducted to deter-
mine the appropriate number of hidden units suggest
that F, the ratio of the number of data points to the
number of adjustable weights in the neural network,
should have a value between 1.8 and 2.3.4,21 For a
network with a 9-3-1 configuration, the number of
weights is 34, therefore F ) 0.9, far from the optimal
values. To reduce the complexity of the multilayer
network and improve its ability to generalize, we have
applied the ‘forgetting’ method22 to the 9-3-1 network.
r1 ANN Mod el: The same set of 32 phenylpiperazines
was used to carry out this study. Initially a network
with a configuration 9-3-1 was employed. The applica-
tion of the forgetting effect led to the weights record in

Synthesis and SARs of New Arylpiperazines. 6
J ournal of Medicinal Chemistry, 2001, Vol. 44, No. 2 203
Ta ble 6. Equations for QSAR of the R1-Adrenergic Receptor
equationsa
r
SE
F
pKi ) 7.73((0.24; 0.51; 32.2) - 0.132((0.153; 0.318; 0.861)IA - 0.321((0.157; 0.327; 2.03)IB
+
0.933
0.435
16.528
0.115((0.153; 0.318; 0.749)In)3 + 1.14((0.34; 0.72; 3.26)F - 0.355((0.473; 0.982; 0.751)R -
0.0281((0.0046; 0.0095; 6.09)Vo - 0.0421((0.0040; 0.0083; 10.52)Vm + 1.03((0.20; 0.41;
5.11)πo + 0.30((0.215; 0.446; 1.41)πm (10)
pKi ) 7.79((0.235; 0.486; 33.1) - 0.132((0.152; 0.315; 0.869)IA - 0.321((0.156; 0.323; 2.05)IB
1.14((0.34; 0.71; 3.29)F - 0.355((0.469; 0.971; 0.758)R - 0.0281((0.0045; 0.0094; 6.15)Vo
- 0.0421((0.0039; 0.0082; 10.62)Vm + 1.03((0.20; 0.41; 5.16)πo + 0.304((0.213;
0.440; 1.42)πm (11)
+
0.932
0.431
18.884
pKi ) 7.84((0.22; 0.46; 35.6) - 0.132((0.150; 0.311; 0.877)IA - 0.309((0.154; 0.318; 2.01)IB
1.101((0.33; 0.70; 3.24)F - 0.0288((0.0044; 0.0091; 6.48)Vo - 0.0422((0.0039; 0.0081;
10.72)Vm + 1.09((0.17; 0.36; 6.15)πo + 0.272((0.207; 0.427; 1.31)πm (12)
pKi ) 7.77((0.21; 0.43; 36.7) - 0.309((0.153; 0.315; 2.01)IB + 1.10((0.33; 0.69; 3.26)F -
0.0288((0.0044; 0.0091; 6.51)Vo - 0.0422((0.0039; 0.0080; 10.77)Vm + 1.09((0.17;
0.36; 6.18)πo + 0.272((0.206; 0.424; 1.32)πm (13)
+
0.930
0.927
0.922
0.427
0.425
0.431
21.887
25.644
29.570
pKi ) 7.68((0.19; 0.41; 40.4) - 0.272((0.152; 0.314; 1.78)IB + 1.32((0.29; 0.61; 4.44)F -
0.0306((0.0042; 0.0087; 7.16)Vo - 0.0417((0.0039; 0.0081; 10.54)Vm + 1.17((0.17;
0.35; 6.86)πo (14)
a
n ) 32.
Ta ble 7. Values of the Connection Weights between the Input
and Hidden Layers for the 5-HT1A ANN Model
model, the affinities for compounds 34 and 35 were
calculated (Table 9). A comparison of the observed and
predicted values for both tested compounds shows a
satisfactory agreement.
N1
N2
N3
IA
IB
In)3
F
R
Vo
Vm
πo
πm
bias
0.492
-0.340
0.174
-0.612
0.045
-0.052
0.479
-0.633
-0.280
0.355
-0.744
0.641
0.641
-0.259
-0.044
-0.048
0.229
-0.254
-0.050
1.251
-0.127
-0.549
0.191
-0.041
0.073
-2.340
-0.046
0.937
0.048
2.017
Dep en d en ce of Biologica l Activity on th e P h ys-
icoch em ica l P a r a m eter s: As for the 5-HT1AR, the
variation of the affinity for the R1-adrenergic receptor
was monitored in 3D diagrams by changing the values
of two inputs while keeping the remaining ones at one-
half of their maximum ranges. The obtained plots show
a clear nonlinear dependence of the affinity on F, Vo,
Vm, πo and πm. Moreover, the depicted surfaces do not
appear to be simple mathematical functions. Neverthe-
less, the comparison of the Hansch analysis on this set
of compounds with the ANN model shows certain
similarities. For both models hydrophobic substituents
at the ortho-position result in high activities, while
bulky substituents at the ortho and especially at the
meta-positions exert the opposite effect. The major
difference between both models lies in the polar effects.
In the Hansch correlation eq 14 high values of F
strongly favor an increase in the predicted receptor
affinity. 3D diagrams show clearly that an increase in
the electronic parameter F exerts a detrimental effect
on the affinity. A second difference was found on the
lipophilicity of substituents at the meta-position. While
in the Hansch model the hydrophobic properties of the
groups at this position have no effect, in the ANN model
high values of πm were found to favor an increase in
the predicted affinity. The explanation for these dis-
crepancies between the neural network model and
linear regression analysis seems to be the implicit
nonlinear mapping performed by the neural network.
Indeed the capability of the ANN to develop nonlinear
terms underlines the strength of this approach and the
improvements obtained.
Ta ble 8. ANN Models
nonsignificant
parameters
receptor
architecture
r
r2
s
5-HT1A
R1
R, πm
R
7-2-1
8-2-1
0.983 0.966 0.149
0.991 0.982 0.136
Ta ble 9. Experimental and Predicted Ki Values by ANN
Models of Compounds 34 and 35
5-HT1A
R1
Ki exptl
SEM (nM)
(
Ki pred
(nM)
Ki exptl
SEM (nM)
(
Ki pred
(nM)
compd
34
35
3.4 ( 0.6
0.6 ( 0.2
1.5
1.7
6.2 ( 1.8
2.4 ( 0.6
3.2
7.2
Ta ble 10. Values of the Connection Weights between the
Input and Hidden Layers for the R1 ANN Model
N1
N2
N3
IA
IB
In)3
F
R
Vo
Vm
πo
πm
bias
-0.173
0.049
-0.094
0.328
0.033
0.049
-0.456
0.683
-0.246
0.553
0.609
-0.751
1.867
0.733
-0.049
0.045
0.030
0.103
-0.072
0.044
1.642
0.292
-0.278
-1.526
-1.517
0.405
2.582
0.047
Com p a r ison of th e 5-HT1A a n d r1 ANN Mod els.
A comparison of both analyses gives an additional
understanding for the 5-HT1A/R1 selectivity, leading to
three important differences between both receptors. (a)
Electrostatic factors: the polar factors are much more
important for the 5-HT1AR than for the R1-adrenergic
receptor. The serotoninergic affinity increases with the
increase of F values, while for the R1-adrenergic receptor
the opposite effect was found. (b) Lipophilicity: hydro-
phobic substituents at the ortho-position result in high
-0.994
0.313
Table 10. In this case only one descriptor, R, could be
eliminated, leading to the final architecture 8-2-1,
with 21 connection weights and F ) 1.5. The training
process was repeated five times in order to find the
global minimum of the network. For the final model r,
r2, and s are 0.991, 0.982, and 0.136, respectively (Table
8). To investigate the predictive power of the derived

204 J ournal of Medicinal Chemistry, 2001, Vol. 44, No. 2
Lo´pez-Rodrı´guez et al.
Ta ble 11. Experimental and Predicted Ki Values for EF-7412
(33) According to Hansch and ANN Models at 5-HT1A and
R1-Adrenergic Receptors
Ki pred (nM)
receptor
Ki exptl ( SEM (nM)
Hansch
ANN
5-HT1A
R1
27 ( 8
>1000
98
3158
36
2745
affinities, for both receptors. However, highly lipophilic
groups at the meta-position only potentiate the affinity
for the R1-adrenergic receptor. (c) Steric requirements
at the meta-position: the meta-position seems to be
significantly involved in the 5-HT1A/R1 selectivity. While
the 5-HT1AR is able to accommodate bulky substituents
(about 60 Å3) in the region of its active site, the steric
requirements of the R1-adrenergic receptor at this
position are more restricted (between 0 and 22 Å3).
These results suggest that a good way to improve the
5-HT1A/R1 selectivity would be the synthesis of long-
chain derivatives (n ) 4) bearing a hydantoin moiety,
according to the Hansch and ANN analysis, respec-
tively. Both models reveal that the meta-position seems
to be implicated in the 5-HT1A/R1 selectivity. Thus bulky
substituents with high F values and low π values would
show an important selectivity for the 5-HT1AR. Among
the different groups that fulfill these requirements
F igu r e 2. Ab initio geometry optimization of the antagonist-
binding site of the 5-HT1AR composed of the side chains of Asp
3.32 and Asn 7.39 (nomenclature of Ballesteros & Weinstein)16
and (a) pindolol and (b) EF-7412 (33). Nonpolar hydrogens are
not depicted to offer a better view of the recognition pocket.
essential determinants for recognition, we performed ab
initio geometry optimization of PND and EF-7412 (33)
inside the side chains of Asp 3.32 and Asn 7.39 (see
Materials and Methods for computational details).
Figure 2 presents a detailed view of Asp 3.32 and Asn
7.39 in its interaction with PND (Figure 2a) and EF-
7412 (33) (Figure 2b). The ionic interaction between the
protonated amine of PND and Asp 3.32 occurs through
the two N-H groups pointing toward both Oδ atoms of
Asp at optimized distances between heteroatoms of 2.58
and 2.54 Å (see Figure 2a). Similarly, the unique N-H
group of the protonated amine of EF-7412 (33) interacts
with one of the Oδ atoms of Asp at an optimized distance
between heteroatoms of 2.51 Å (see Figure 2b). The
recognition of PND by Asn 7.3927,28 occurs through
hydrogen bonds between the ether oxygen and the
hydroxyl group of PND and the N-H and CdO groups
of Asn, at optimized distances between heteroatoms of
2.84 and 2.75 Å, respectively. The m-NHSO2Et group
of EF-7412 (33) can fulfill a similar hydrogen bond
network (see Figure 2b). The N-H group of the m-NHSO2-
Et substituent acts as a hydrogen bond donor in the
hydrogen bond interaction with the CdO group of Asn
(2.72 Å), whereas the SdO group acts as a hydrogen
bond acceptor in the hydrogen bond with the N-H group
of Asn (2.89 Å). It is important to remark that although
the ether oxygen and the hydroxyl group of PND and
the m-NHSO2Et group of EF-7412 (33) are structurally
dissimilar, these are interacting in a similar manner
with Asn 7.39 and thus serving an analogous function.
Moreover, the absence of Asn 7.39 in the R1-adrenergic
receptor explains the selectivity of EF-7412 (33) for the
5-HT1AR sites.
m-NHSO2Et was chosen (F ) 0.419, πm ) -0.64, Vm
)
65.31). On these bases, the new ligand EF-7412 (33) (X
) -(CH2)3-, m ) 0, n ) 4, R ) m-NHSO2Et) was
designed and synthesized. This analogue exhibits a high
selectivity for the 5-HT1AR versus R1-adrenergic receptor
(Ki ) 27 nM vs Ki > 1000 nM) and also a satisfactory
affinity. The Hansch and ANN predicted affinities for
EF-7412 (33) are given in Table 11. These values clearly
reveal the greater predictive power of the ANN model
and the importance of the nonlinear relationships
mapped by the neural network.
Molecu la r Mod el of th e Com p lex betw een EF -
7412 (33) a n d 5-HT1AR. In the preceding paper we
reported the pharmacological characterization of EF-
7412 (33) as an antagonist in vivo in pre- and postsyn-
aptic 5-HT1AR sites. Mutagenesis experiments on GPCRs
that bind protonated amine neurotransmitters indicate
that the conserved Asp 3.32 is involved in the interac-
tion with protonated amines of both agonists and
antagonists (see ref 23 and references therein). Simi-
25
larly, mutagenesis experiments on the R224, â2 and
5-HT1A receptors26 have shown that the presence of an
Asn residue at position 7.39 is important in conferring
specificity to a series of ligands such as pindolol (PND)
and propranolol. Moreover, the recognition of PND by
Asn 7.39 occurs through two hydrogen bonds to the
ether oxygen and the hydroxyl group of the ligand.27,28
We can assume, as a working hypothesis, that EF-7412
(33) interacts with the 5-HT1AR sites in a similar
manner throughout: (i) the ionic interaction between
the protonated amine and Asp 3.32 and (ii) the hydrogen
bonds between the m-NHSO2Et substituent and Asn
7.39. To identify the arrangement in space of these
The obtained complex depicted in Figure 2b between
the modeled part of EF-7412 (33) and Asp 3.32 and Asn
7.39 was used to position the complete structure of EF-
7412 (33) inside a 3D model of the transmembrane
domain of the 5-HT1AR (see Materials and Methods for
computational details). Figure 3 shows EF-7412 (33) in
the binding pocket of the 5-HT1AR, in a view perpen-
dicular (Figure 3a) and parallel (Figure 3b) to the
membrane. Figure 3c presents a detailed view of the

Synthesis and SARs of New Arylpiperazines. 6
J ournal of Medicinal Chemistry, 2001, Vol. 44, No. 2 205
predictions. The obtained Hansch and ANN models for
the 5-HT1A and R1-adrenergic receptors led us to design
and synthesize a new antagonist, EF-7412 (33), with a
high selectivity profile for the 5-HT1AR sites. The
computed model of the complex between EF-7412 (33)
and a 3D structure of the transmembrane domain of
the 5-HT1AR has identified the key structural features
of both EF-7412 (33) and the 5-HT1AR that are respon-
sible for complex formation. Classical Hansch analysis,
ANNs, and the detailed 3D model of the ligand-receptor
complex provide an essential tool for the development
of highly desirable,31 selective 5-HT1A antagonists.
Exp er im en ta l Section
Ch em istr y. Melting points (uncorrected) were determined
on a Gallenkamp electrothermal apparatus. Infrared (IR)
spectra were obtained on a Perkin-Elmer 781 infrared spec-
trophotometer. 1H and 13C NMR spectra were recorded on a
Varian VXR-300S or Bruker 250-AM instrument. Chemical
shifts (δ) are expressed in parts per million relative to internal
tetramethylsilane; coupling constants (J ) are in hertz (Hz).
Elemental analyses (C, H, N) were determined within 0.4% of
the theoretical values. Thin-layer chromatography (TLC) was
run on Merck silica gel 60 F-254 plates. For normal pressure
and flash chromatography, Merck silica gel type 60 (size 70-
230 and 230-400 mesh, respectively) was used. Unless stated
otherwise, starting materials used were high-grade commercial
products.
F igu r e 3. Molecular model of the transmembrane helix
bundle of the 5-HT1AR constructed from the crystal structure
of bovine rhodopsin.5 EF-7412 (33) is shown in the binding
pocket (a) in a view perpendicular to the membrane, (b) in a
view parallel to the membrane, and (c) in a detailed view of
the ligand-binding site. The binding mode includes: (i) ionic
interaction between the protonated amine of the ligand and
Asp 3.32; (ii) hydrogen bonds between the m-NHSO2Et group
of the ligand and Asn 7.39; and hydrogen bonds between the
hydantoin moiety of the ligand and (iii) Thr 3.37, (iv) Ser 5.42,
and (v) Thr 5.43 (nomenclature of Ballesteros & Weinstein).16
Nonpolar hydrogens are not depicted to offer a better view of
the recognition pocket. Figure was created using MOL-
SCRIPT.38
The 2-(4-bromobutyl)-1,3-dioxoperhydroimidazo[1,5-a]pyri-
dine was synthesized according to the literature.32 1-Arylpip-
erazines were prepared by reaction of the corresponding
anilines with bis(2-chloroethyl)amine, as previously de-
scribed.33,34
2-[4-[4-(o-Br om op h en yl)p ip er a zin -1-yl]bu tyl]- a n d 2-[4-
[4-(o-(P h en oxy)p h en yl)p ip er a zin -1-yl]b u t yl]-1,3-d ioxo-
p er h yd r oim id a zo[1,5-a ]p yr id in e (34, 35). Gen er a l P r o-
ced u r e. To a suspension of 2-(4-bromobutyl)-1,3-dioxoperhy-
droimidazo[1,5-a]pyridine (9 mmol) and the corresponding
1-arylpiperazine (15 mmol) in acetonitrile (20 mL), was added
2.0 mL of triethylamine (1.5 g, 14.6 mmol). The mixture was
refluxed for 20-24 h. Then, the solvent was evaporated under
reduced pressure and the residue was resuspended in water
and extracted with dichloromethane (3 × 100 mL). The
combined organic layers were washed with water and dried
over MgSO4. After evaporation of the solvent the crude oil was
purified by column chromatography (eluents: ethyl acetate/
ethanol 9:1). Spectral data refer to the free base and then
hydrochloride salts were prepared.
ligand-binding site. The proposed recognition of the
endogenous ligand involves, in addition to the (i) ionic
and (ii) hydrogen bond interactions with Asp 3.32 and
Asn 7.39 (see above), the hydrogen bonds between both
CdO groups of the hydantoin moiety of the ligand and
(iii) Thr 3.37 (3.06 Å), (iv) Ser 5.42 (3.42 Å), and (v) Thr
5.43 (3.70 Å) of the 5-HT1AR. Mutation of Thr 3.37 to
Ala in the R1B receptor29 decreases the affinity of the
antagonist prazosin by a factor of 6. This result indicates
that this locus is approachable by the extracellular
ligand. On the other hand, mutagenesis experiments
have shown that a series of conserved Ser or Thr
residues at positions 5.42 and 5.43 act as hydrogen-
bonding sites for the hydroxyl groups present in the
chemical structure of many neurotransmitters and
synthetic agonists (see ref 23 and references therein).
In particular, Ser 5.42 and Thr 5.43 are important in
the binding of 5-HT to the 5-HT1AR.30 Thus, Asp 3.32,
Ser 5.42, and Thr 5.43 are common anchoring residues
of both agonists and EF-7412 (33). We can conclude that
EF-7412 (33) binds to the same receptor sites than
agonists do and act by blocking the access of agonists
to receptor sites.
2-[4-[4-(o-Br om oph en yl)piper azin -1-yl]bu tyl]-1,3-dioxo-
p er h yd r oim id a zo[1,5-a ]p yr id in e (34): yield 56%; mp 220-
1
222 °C (methanol/ethyl ether); H NMR (CDCl3) δ 1.22-1.69
(m, 8H), 1.92 (dm, J ) 13.5, 1H), 2.14 (dd, J ) 13.2, 3.6, 1H),
2.36 (t, 2H), 2.56 (brs, 4H), 2.76 (td, J ) 12.3, 3.6, 1H), 3.00
(brs, 4H), 3.46 (t, J ) 7.2, 2H), 3.67 (dd, J ) 12.0, 4.2, 1H),
4.10 (dd, J ) 13.5, 4.8, 1H), 6.83 (td, J ) 8.1, 1.5, 1H), 6.98
(dd, J ) 7.8, 1.5, 1H), 7.19 (td, J ) 7.8, 1.5, 1H), 7.48 (dd, J )
8.1, 1.5, 1H); 13C NMR (CDCl3) δ 22.7, 24.0, 24.9, 26.2, 27.8,
38.5, 39.2, 51.6, 53.3, 57.2, 58.0, 119.8, 120.8, 124.2, 128.2,
133.7, 150.6, 154.4, 173.1. Anal.(C21H29BrN4O2‚HCl) C, H, N.
2-[4-[4-(o-(P h en oxy)p h en yl)p ip er a zin -1-yl]b u t yl]-1,3-
d ioxop er h yd r oim id a zo[1,5-a ]p yr id in e (35): yield 64%; mp
171-173 °C (methanol/ethyl ether); 1H NMR (CDCl3) δ 1.23-
1.65 (m, 7H), 1.73 (d, J ) 10.8, 1H), 1.98 (dm, J ) 12.3, 1H),
2.19 (dd, J ) 9.8, 3.8, 1H), 2.38 (tt, J ) 7.5, 2H), 2.49 (brs,
4H), 2.82 (td, J ) 12.6, 3.6, 1H), 3.15 (brs, 4H), 3.50 (t, J )
6.9, 2H), 3.73 (dd, J ) 11.7, 3.9, 1H), 4.15 (dd, J ) 14.7, 3.0,
1H), 6.91-7.12 (m, 7H), 7.28 (dd, J ) 8.4, 7.5, 2H); 13C NMR
(CDCl3) δ 22.7, 23.5, 24.9, 26.0, 27.7, 38.3, 39.2, 49.8, 53.1,
57.2, 57.8, 117.2, 119.1, 121.1, 122.3, 122.7, 124.6, 129.3, 143.6,
148.5, 154.4, 157.3, 173.1. Anal. (C27H34N4O3‚2HCl‚2H2O) C,
H, N.
Con clu sion
A classical QSAR and ANN analysis have been
successfully applied to a series of 32 phenylpiperazines
with affinity for 5-HT1A and R1-adrenergic receptors.
The resulting models provide a significant correlation
of electrostatic, steric, and lipophilic parameters with
biological affinities. Although the derived linear Hansch
models are able to give acceptable predictions, the
introduction of nonlinear relationships in the analysis
clearly gives more solid models and more accurate
Ra d ioliga n d Bin d in g Assa ys. For all receptor binding
assays, male Sprague-Dawley rats (Rattus norvegicus albi-

206 J ournal of Medicinal Chemistry, 2001, Vol. 44, No. 2
Lo´pez-Rodrı´guez et al.
(4) Andrea, T. A.; Kalayeh, H. Applications of Neural Networks in
Quantitative Structure-Activity Relationships of Dihydrofolate
Reductase Inhibitors. J . Med. Chem. 1991, 34, 2824-2836.
(5) Palczewski, K.; Kumasaka, T.; Hori, T.; Behnke, C. A.; Mo-
toshima, H.; Fox, B. A.; Trong, I. L.; Teller, D. C.; Okada, T.;
Stenkamp, R. E.; Yamamoto, M.; Miyano, M. Crystal Structure
of Rhodopsin: A G Protein-coupled Receptor. Science 2000, 289,
739-745.
nus), weighing 180-200 g, were killed by decapitation and the
brains rapidly removed and dissected.
5-HT1A R ecep t or . The receptor binding studies were
performed by a modification of a previously described proce-
dure.35 The cerebral cortex was homogenized in 10 volumes of
ice-cold Tris buffer (50 mM Tris-HCl, pH 7.7 at 25 °C) and
centrifuged at 28000g for 15 min. The membrane pellet was
washed twice by resuspension and centrifugation. After the
second wash the resuspended pellet was incubated at 37 °C
for 10 min. Membranes were then collected by centrifugation
and the final pellet was resuspended in 50 mM Tris-HCl, 5
mM MgSO4, and 0.5 mM EDTA buffer (pH 7.4 at 37 °C).
Fractions of the final membrane suspension (about 1 mg of
protein) were incubated at 37 °C for 15 min with 0.6 nM [3H]-
8-OH-DPAT (8-hydroxy-2-(dipropylamino)tetralin) (133 Ci/
mmol), in the presence or absence of several concentrations
of the competing drug, in a final volume of 1.1 mL of assay
buffer (50 mM Tris-HCl, 10 nM clonidine, 30 nM prazosin, pH
7.4 at 37 °C). Nonspecific binding was determined with 10 µM
5-HT.
r1-Ad r en er gic Recep tor . The radioligand receptor binding
studies were performed according to a previously described
procedure.36 The cerebral cortex was homogenized in 20
volumes of ice-cold buffer (50 mM Tris-HCl, 10 mM MgCl2,
pH 7.4 at 25 °C) and centrifuged at 30000g for 15 min. Pellets
were washed twice by resuspension and centrifugation. Final
pellets were resuspended in the same buffer. Fractions of the
final membrane suspension (about 250 µg of protein) were
incubated at 25 °C for 30 min with 0.2 nM [3H]prazosin (23
Ci/mmol) in the presence or absence of several concentrations
of the competing drug, in a final volume of 2 mL of buffer.
Nonspecific binding was determined with 10 µM phentolamine.
For all binding assays, competing drug, nonspecific, total
and radioligand bindings were defined in triplicate. Incubation
was terminated by rapid vacuum filtration through Whatman
GF/B filters, presoaked in 0.05% poly(ethylenimine), using a
Brandel cell harvester. The filters were then washed with the
assay buffer and dried. The filters were placed in poly-
(ethylene) vials to which were added 4 mL of a scintillation
cocktail (Aquasol), and the radioactivity bound to the filters
was measured by liquid scintillation spectrometry. The data
were analyzed by an iterative curve-fitting procedure (program
Prism, Graph Pad), which provided IC50, Ki, and r2 values for
test compounds, Ki values being calculated from the Cheng
and Prusoff equation.8 The protein concentrations of the rat
cerebral cortex and the rat striatum were determined by the
method of Lowry,37 using bovine serum albumin as the
standard.
(6) Bourne, H. R.; Meng, E. C. Rhodopsin Sees the Light. Science
2000, 289, 733-734.
(7) Schaper, K.-J .; Emig, P.; Engel, J .; Fleischhauer, I.; Kutscher,
B.; Rosado, M. L.; Lo´pez-Rodr´ıguez, M. L. Dose Response
Relationships, Biotest Intercorrelations, QSAR and the Saving
of Animals Experiments. In QSAR and Molecular Modelling:
Concepts, Computational Tools and Biological Applications;
Sanz, F., Giraldo, J ., Manaut, F., Eds.; Prous Science: Barcelona,
1995; pp 73-76.
(8) Cheng, Y. C.; Prusoff, W. H. Relationship between the Inhibition
Constant (Ki) and the Concentration of Inhibitor which Causes
50 Per Cent Inhibition (IC50) of an Enzymatic Reaction. Biochem.
Pharmacol. 1973, 22, 3099-3108.
(9) Williams, S. G.; Norrington, F. E. Determination of Positional
Weighting Factors for the Swain and Lupton Substituent
Constants F and R . J . Am. Chem. Soc. 1976, 98, 508-516.
(10) SYBYL Molecular Modelling System (version 6.0), Tripos As-
sociates, 1699 S. Hanley Rd., St. Louis, MO 63144.
(11) Hansch, C.; Leo, A. J .; Hoekman, D. Exploring QSAR: Hydro-
phobic, Electronic and Steric Constants; ACS Professional Refer-
ence Book, ACS Advisory Board: Washington, D.C., 1995.
(12) Hansch, C. In Structure-Activity Relationships; Cavallito, C. J .,
Ed.; Pergamon Press: Oxford, 1973; Vol. 1, p 50.
(13) Bennet, C. A.; Franklin, N. L. Statistical Analysis in Chemistry
and the Chemical Industry; Wiley: New York, 1963.
(14) Rumelhart, D. E.; Hinton, G. E.; Williams, R. J . Learning
Representations by Back-propagatin Errors. Nature 1986, 323,
533-536.
(15) Schaper, K.-J . Free-Wilson-Type Analysis of Non-Additive Sub-
stituent Effects on THPB Dopamine Receptor Affinity Using
Artificial Neural Networks. Quant. Struct.-Act. Relat. 1999, 18,
354-360.
(16) Ballesteros, J . A.; Weinstein, H. Integrated Methods for the
Construction of Three-Dimensional Models and Computational
Probing of Structure-function Relations in G-Protein Coupled
Receptors. Methods Neurosci. 1995, 25, 366-428.
(17) Frisch, M. J .; Trucks, G. W.; Schlegel, H. B.; Scuseria, G. E.;
Robb, M. A.; Cheeseman, J . R.; Zakrzewski, V. G.; Montgomery,
J . A.;; Keith, T. A.; Petersson, G. A.; Raghavachari, K.; Al-
Laham, A.; Stratmann, R. E.; Burant, J . C.; Dapprich, S.;
Millam, J . M.; Daniels, A. D.; Kudin, K. N.; Strain, M. C.; Farkas,
O.; Tomasi, J .; Barone, V.; Cossi, M.; Cammi, R.; Mennucci, B.;
Pomelli, C.; Adamo, C.; Clifford, S.; Ochterski, J .; Petersson, G.
A.; Ayala, P. Y.; Cui, Q.; Morokuma, K.; Malick, D. K.; Rabuck,
A. D.; Raghavachari, K.; Foresman, J . B.; Cioslowski, J .; Ort´ız,
J . V.; Stefanov, B. B.; Liu, G.; Liashenko, A.; Piskorz, P.;
Komaromi, I.; Gomperts, R.; Mart´ın, R. L.; Fox, D. J .; Keith, T.;
Al-Laham, M. A.; Peng, C. Y.; Nanayakkara, A.; Gonza´lez, C.;
Challacombe, M.; Gill, P. M. W.; J ohnson, B. G.; Chen, W.; Wong,
W.; Andres, J . L.; HeadBGordon, M.; Replogle, E. S.; Pople, J .
A. Gaussian 98; Gaussian Inc., Pittsburgh, PA.
(18) Case, D. A.; Pearlman, D. A.; Caldwell, J . W.; Cheatham, III, T.
E.; Ross, W. S.; Simmerling, C. L.; Darden, T. A.; Merz, K. M.;
Stanton, R. V.; Cheng, A. L.; Vicent, J . J .; Crowley, M.; Ferguson,
D. M.; Radmer, R. J .; Seibel, G. L.; Singh, U. C.; Weiner, P. K.;
Kollman, P. A. AMBER 5; University of California, San Fran-
cisco, 1997.
Ack n ow led gm en t. This work was supported by
grants from DGESIC (PB97-0282), CICYT (SAF99-073),
Comunidad de Madrid (08.5/000466.1/98), and Fundacio´
La Marato´ TV3 (0014/97). Some of the simulations were
run at the Centre de Computacio´ i Comunicacions de
Catalunya. The authors are grateful to UNED for a
predoctoral grant to E.F. and the German Academic
Exchange Service (DAAD) for financial support of
M.L.R. in the form of a 10-month research stipend.
(19) Cornell, W. D.; Cieplak, P.; Bayly, C. I.; Gould, I. R.; Merz, J r.,
K. M.; Ferguson, D. M.; Spellmeyer, D. C.; Fox, T.; Caldwell, J .
W.; Kollman, P. A. A Second Generation Force Field for the
Simulation of Proteins, Nucleic Acids, and Organic Molecules.
J . Am. Chem. Soc. 1995, 117, 5179-5197.
(20) Cieplak, P.; Cornell, W. D.; Bayly, C.; Kollman, P. A. Application
of the Multimolecule and Multiconformational RESP Methodol-
ogy to Biopolymers: Charge Derivation for DNA, RNA and
Proteins. J . Comput. Chem. 1995, 16, 1357-1377.
(21) So, S. S.; Richards, W. G. Application of Neural Networks:
Quantitative Structure-Activity Relationships of the Deriva-
tives of 2,4-Diamino-5-(substituted-benzyl)pyrimidines as DHFR
Inhibitors. J . Med. Chem. 1992, 35, 3201-3207.
(22) Aoyama, T.; Ichikawa, H. Reconstruction of Weight Matrixes in
Neural Networks- a Method of Correlating Outputs with Inputs.
Chem. Pharm. Bull. 1991, 39, 1222-1228.
(23) Rhee, A. M.v.; J acobson, K. A. Molecular Architecture of G
Protein-Coupled Receptors. Drug Dev. Res. 1996, 37, 1-38.
(24) Suryanarayana, S.; Daunt, D. A.; Zastrow, M. V.; Kobilka, B.
K. A Point Mutation in the Seventh Hydrophobic Domain of the
R2-Adrenergic Receptor Increases its Affinity for a Family of
â-Receptor Antagonists. J . Biol. Chem. 1991, 266, 15488-15492.
Su p p or tin g In for m a tion Ava ila ble: 3D diagrams of
ANN models. This information is available free of charge via
the Internet at http://pubs.acs.org.
Refer en ces
(1) (a) Hansch, C.; Fujita, T. F-σ-π Analysis. A Method for the
Correlation of Biological Activity and Chemical Structure. J . Am.
Chem. Soc. 1964, 86, 1616-1626. (b) Kubiny, H. QSAR: Hansch
Analysis and Related Approaches. In Methods and Priciples in
Medicinal Chemistry; VCH: Weinheim, New York, 1993; Vol. I.
(2) Zupan, J .; Gasteiger, J . Neural Networks for Chemists: An
Introduction; VCH: New York, 1993.
(3) Devillers, J ., Ed. Neural Networks in QSAR and Drug Design.
In Principles of QSAR and Drug Design; Academic Press:
London, 1996.

Synthesis and SARs of New Arylpiperazines. 6
J ournal of Medicinal Chemistry, 2001, Vol. 44, No. 2 207
(25) Suryanarayana, S.; Kobilka, B. K. Amino Acid Substitutions at
Position 312 in the Seventh Hydrophobic Segment of the
â2-Adrenergic Receptor Modify Ligand-Binding Specificity. Mol.
Pharmacol. 1993, 44, 111-114.
(26) Guan, X.-M.; Peroutka, S. J .; Kobilka, B. K. Identification of a
Single Amino Acid Residue Responsible for the Binding of a
Class of Beta-Adrenergic Receptor Antagonists to 5-HT1A Recep-
tors. Mol. Pharmacol. 1992, 41, 695-698.
(27) Glennon, R. A.; Dukat, M.; Westkaemper, R. B.; Ismaiel, A. M.;
Izzarelli, D. G.; Parker, E. M. The Binding of Propranolol at
5-Hydroxytryptamine1Dâ T355N Mutant Receptors May Involve
Formation of Two Hydrogen Bonds to Asparagine. Mol. Phar-
macol. 1996, 49, 198-206.
(28) Pardo, L.; Batlle, M.; Dun˜ach, M.; Weinstein, H. The Structure
and Activity of Membrane Receptors: Modeling and Computa-
tional Simulation of Ligand Recognition in a Three-Dimensional
Model of the 5-Hydroxytryptamine1A Receptor. J . Biomed. Sci.
1996, 3, 98-107.
(29) Cavalli, A.; Fanelli, F.; Taddei, C.; De Benedetti, P. G.; Cotecchia,
S. Amino Acids of the Alpha1â-Adrenergic Receptor Involved in
Agonist Binding: Differences in Docking Catecholamines to
Receptor Subtypes. FEBS Lett. 1996, 399, 9-13.
(30) Ho, B. Y.; Karschin, A.; Branchek, T.; Davidson, N.; Lester, H.
A. The Role of Conserved Aspartate and Serine Residues in
Ligand Binding and in Function of the 5-HT1A Receptor: A Site-
directed Mutation Study. FEBS Lett. 1992, 312, 259-262.
(31) Fletcher, A.; Cliffe, I. A.; Dourish, C. T. Silent 5-HT1A Receptor
Antagonists: Utility as Research Tools and Theraupetic Agents.
Trends Pharmacol. Sci. 1993, 14, 441-448.
(32) Lo´pez-Rodr´ıguez, M. L.; Rosado, M. L.; Benhamu´, B.; Morcillo,
M. J .; Sanz, A. M.; Orensanz, L.; Beneytez, M. E.; Fuentes, J .
A.; Manzanares, J . Synthesis and Structure-Activity Relation-
ships of a New Model of Arylpiperazines. 1. 2-[[4-(o-Methox-
yphenyl)piperazin-1-yl]methyl]-1,3-dioxoperhydroimidazo[1,5-
a]pyridine: A Selective 5-HT1A Receptor Agonist. J . Med. Chem.
1996, 39, 4439-4450.
(33) Lyon, R. A.; Titeler, M.; McKenney, J . D.; Magee, P. S.; Glennon,
R. A. Synthesis and Evaluation of Phenyl- and Benzoylpipera-
zines as Potential Serotonergic Agents. J . Med. Chem. 1986, 29,
630-634.
(34) Martin, G. E.; Elgin, R. J ., J r.; Mathiasen, J . R.; Davis, C. B.;
Kesslick, J . M.; Baldy, W. J .; Shank, R. P.; DiStefano, D. L.;
Fedde, C. L.; Scott, M. K. Activity of Aromatic Substituted
Phenylpiperazines Lacking Affinity for Dopamine Binding Sites
in a Preclinical Test of Antipsychotic Efficacy. J . Med. Chem.
1989, 32, 1052-1056.
(35) Clark, R. D.; Weinhardt, K. K.; Berger, J .; Fischer, L. E.; Brown,
C. M.; MacKinnon, A. C.; Kilpatrick, A. T.; Spedding, M. 1,9-
Alkano-bridged 2,3,4,5-Tetrahydro-1H-3-benzazepines with Af-
finity for the R2-Adrenoceptor and the 5-HT1A Receptor. J . Med.
Chem. 1990, 33, 633-641.
(36) Ambrosio, E.; Montero, M. T.; Ferna´ndez, I.; Azuara, M. C.;
Orensanz, L. M. [3H]Prazosin Binding to Central Nervous
System Regions of Male and Female Rats. Neurosci. Lett. 1984,
49, 193-197.
(37) Lowry, O. H.; Rosebrough, N. J .; Farr, A. L.; Randall, R. J .
Protein Measurement with the Folin Phenol Reagent. J . Biol.
Chem. 1951, 193, 265-275.
(38) Kraulis, J . MOLSCRIPT: A Program to Produce Both Detailed
and Schematic Plots of Protein Structure. J . Appl. Crystallogr.
1991, 24, 946-950.
J M000930T