Three-dimensional quantitative structure-activity studies of octopaminergic agonists responsible for the inhibition of sex-pheromone production in Hercoverpa armigera

Article abstract of DOI:10.1016/S0968-0896(99)00189-3

The quantitative structure-activity relationship (QSAR) of octopaminergic agonists responsible for the inhibition of sex-pheromone production in Hercoverpa armigera, was analyzed using physicochemical parameters, molecular shape analysis (MSA), molecular field analysis (MFA), and receptor surface model (RSM), respectively. The dose-response studies were performed in vitro analyzing the effect of these compounds on intracellular cAMP production in the presence of pheromone biosynthesis activating neuropeptide (PBAN) at 1 pmol/intersegment. Six active derivatives were identified in the order of decreasing pheromonostatic activity: 2-(2,6- dimethylanilino)imidazolide (6) > 2-(2-methyl-4-chloroanilino)oxazolidine (1) > clonidine (5) > 2-(2,6-diethylanilino)thiazolidine (8) > 2-(3,5- dichlorobenzylamino)-2-oxazoline (4) > tolazoline (10) which were all active in the nanomolar range in inhibition of cAMP production by 1 pmol PBAN/intersegment. Four other compounds were less active having K(i) in the micromolar range. An MSA was tried to obtain QSAR equation that incorporates spatial molecular similarity data of those compounds. MFA on the training set of those compounds evaluated effectively the energy between a probe and a molecular model at a series of points defined by a rectangular or spherical grid. An RSM was generated using some subset of the most active structures. Three-dimensional energetics descriptors were calculated from RSM/ligand interaction and these three-dimensional descriptors were used in QSAR analysis. These results indicate that these derivatives could provide useful information in the characterization and differentiation of octopaminergic receptor types and subtypes. (C) 1999 Elsevier Science Ltd.

Full text of DOI:10.1016/S0968-0896(99)00189-3

Bioorganic & Medicinal Chemistry 7 (1999) 2621±2628
Three-Dimensional Quantitative Structure±Activity Studies of
Octopaminergic Agonists Responsible for the Inhibition of
Sex-Pheromone Production in Hercoverpa armigera
Akinori Hirashima, a,* Ada Rafaeli, b Carina Gileadi b and Eiichi Kuwano a
aDivision of Bioresource and Bioenvironmental Sciences, Graduate School, Kyushu University, Fukuoka 812-8581, Japan
bDepartment of Stored Products, Pheromone Research Laboratory, Volcani Centre, PO Box 6, Bet Dagan 50250, Israel
Received 28 April 1999; accepted 8 July 1999
AbstractÐThe quantitative structure±activity relationship (QSAR) of octopaminergic agonists responsible for the inhibition of sex-
pheromone production in Hercoverpa armigera, was analyzed using physicochemical parameters, molecular shape analysis (MSA),
molecular ®eld analysis (MFA), and receptor surface model (RSM), respectively. The dose-response studies were performed in vitro
analyzing the e€ect of these compounds on intracellular cAMP production in the presence of pheromone biosynthesis activating
neuropeptide (PBAN) at 1 pmol/intersegment. Six active derivatives were identi®ed in the order of decreasing pheromonostatic
activity: 2-(2,6-dimethylanilino)imidazolide (6) >2-(2-methyl-4-chloroanilino)oxazolidine (1) >clonidine (5) >2-(2,6-diethylanili-
no)thiazolidine (8) >2-(3,5-dichlorobenzylamino)-2-oxazoline (4) >tolazoline (10) which were all active in the nanomolar range in
inhibition of cAMP production by 1 pmol PBAN/intersegment. Four other compounds were less active having Ki in the micro-
molar range. An MSA was tried to obtain QSAR equation that incorporates spatial molecular similarity data of those compounds.
MFA on the training set of those compounds evaluated e€ectively the energy between a probe and a molecular model at a series of
points de®ned by a rectangular or spherical grid. An RSM was generated using some subset of the most active structures. Three-
dimensional energetics descriptors were calculated from RSM/ligand interaction and these three-dimensional descriptors were used
in QSAR analysis. These results indicate that these derivatives could provide useful information in the characterization and di€er-
entiation of octopaminergic receptor types and subtypes. # 1999 Elsevier Science Ltd. All rights reserved.
Introduction
Production of the pheromone blend is under the reg-
ulation of a neuropeptide termed pheromone biosynth-
esis activating neuropeptide (PBAN).12±19 The direct
action of PBAN has been demonstrated by studies in
vitro20±25 showing stimulation of pheromone produc-
tion in the presence of synthetic peptide by isolated
pheromone gland tissue. The exact tissue involved was
delineated as the intersegmental membrane which is
situated between the 8th and 9th abdominal seg-
ments.26,27 In Helicoverpa armigera, we have shown that
OA, tyramine, and clonidine signi®cantly inhibit the
pheromonotropic action due to PBAN in intact moths
and decapitated moths, as well as pheromone gland
incubations in vitro.26±28 The inhibition was also re¯ec-
ted in a signi®cant inhibitory e€ect on intracellular
cAMP levels which were stimulated in the presence of
PBAN. This inhibitory action is a result of a receptor
(separate from the PBAN-receptor) which can be
inhibited by pertussis toxin.28 This provided evidence
that this speci®c pheromonostatic±aminergic receptor is
linked to a G-inhibitory protein. The pheromonostatic
receptor, acting in a neuromodulatory role, represents a
Octopamine (OA), which has been found to present in
high concentrations in various insect tissues, is the
monohydroxyllic analogue of the vertebrate hormone
noradrenaline. It has been well established that OA acts
as neurohormone, neurotransmitter, and neuromodu-
lator in insects.1 Three di€erent receptor classes OAR1,
OAR2A, and OAR2B had been distinguished from
non-neuronal tissues.2 In the nervous system of migra-
tory locust Locusta migratoria L., a particular receptor
class was characterized and established as a new class
OAR3 by pharmacological investigations of the [3H]OA
binding site using various agonists and antagonists.3±7
Structure±activity studies of various types of OA-agonists
and antagonists were also reported using the nervous
tissue of L. migratoria.8±11
Key words: Hercoverpa armigera; octopaminergic agonist; receptor
surface model; molecular ®eld analysis; molecular shape analysis.
* Corresponding author. Tel.: +81-92-642-2856; fax: +81-92-642-
2864/642-2804; e-mail: ahirasim@agr.kyushu-u.ac.jp
0968-0896/99/$ - see front matter # 1999 Elsevier Science Ltd. All rights reserved.
PII: S0968-0896(99)00189-3

2622
A. Hirashima et al. / Bioorg. Med. Chem. 7 (1999) 2621±2628
novel type of octopaminergic receptor which induces an
inhibitory and not a stimulatory action on adenylate
cyclase. It is therefore of critical importance to provide
information on the pharmacological properties of this
OA receptor types and subtypes. This report deals with
the three-dimensional quantitative structure±activity
relationship (QSAR) of octopaminergic agonists
responsible for the inhibition of sex-pheromone pro-
duction in H. armigera, using physicochemical para-
meters, molecular shape analysis (MSA), molecular ®eld
analysis (MFA), and receptor surface model (RSM),
respectively.
Results and Discussion
Six active derivatives were identi®ed in order of
decreasing pheromonostatic activity: 2-(2,6-dimethyl-
anilino)imidazolide (6) > 2-(2-methyl-4-chloroanilino)-
oxazolidine (1) > clonidine (5) > 2-(2,6-diethylanilino)-
thiazolidine (8) > 2-(3,5-dichlorobenzylamino)-2-oxa-
zoline (4) >tolazoline (10) which were all active in the
nanomolar range in OA-agonist activity responsible for
the inhibition PBAN-stimulated cAMP production in
H. armigera (Table 1). Pheromonostatic activity of these
OA agonists showed a good correlation with an inhibi-
tion of PBAN-stimulated cAMP production (unpub-
lished data). Four other compounds were less active
having Ki in the micromolar range. Ki is the concentra-
tion of OA agonist necessary for half-maximal inhibi-
tion of PBAN binding at 1 pmol/intersegment. In order
to quantitatively understand the dependence of biologi-
cal activities on physicoparameters of OA agonists,
regression analysis was applied to representative com-
pounds listed in Figure 1a and Table 1, leading to eq
(1). Preliminary tests indicate that if the genetic function
algorithm (GFA) method is used, non-linear terms must
be included. Hence, GFA was used with non-linear
terms for all following QSAR calculations.
Figure 1. Structures of OA agonists used for regression analysis in (A)
training and (B) test sets.
where n=10, r2=1.000, F=5681.367, CV r2=0.998,
and Bsr2=1.000‹0.000: r-squared (r2), the square of
the correlation coecient; cross-validated r2 (CV r2), a
squared correlation coecient generated during a vali-
dation procedure using the equation; bootstrap r2
(Bsr2), the average squared correlation coecient cal-
culated during the validation procedure; n, the number
of data points; r, the correlation coecient; F, the value
of the F-test. The r2 is used to describe the goodness of
®t of the data in the study table to the QSAR model. A
CV r2 is usually smaller than the overall r2 for a QSAR
equation. It is used as a diagnostic tool to evaluate the
predictive power of an equation generated using the
multiple linear regression or partial least squares
method. A Bsr2 is computed from the subset of vari-
ables used one-at-a-time for the validation procedure. It
can be used more than one time in computing the r2
statistic. Table 1 depicts structures of compounds, their
experimental Ki values, their pKi values, calculated pKi
pKi ˆ 3:08501 0:081986LowEne 14:8188SXZF
2
‡ 8:81828JX ‡ 0:004254ꢀMR 57:7808†
ꢀ1†
Table 1. Regression analysis of structure-OA agonist activity using conventional descriptors
Compound
Ki (nM)
pKi
LowEneb
SXZFc
JXd
MRe
No.
R
Observed
Calculateda
Deviation
1
2
3
4
5
6
7
8
AIO 2-CH3,4-Cl
AIO 2-CH3,6-CH2CH3
BAO 2-F
6.15
6150
100,000
69.2
8.97
3.85
76,900
61.5
5380
76.9
8.21
5.21
4.00
7.16
8.05
8.41
4.11
7.21
5.27
7.11
8.20
5.23
3.99
7.17
8.00
8.43
4.11
7.25
5.24
7.11
0.01
0.02
0.01
0.01
0.05
0.02
0.00
0.04
0.03
0.00
8.848
63.708
49.976
5.690
52.652
37.667
73.526
88.279
106.179
13.907
0.688
0.871
0.782
0.880
0.714
0.795
0.689
0.671
0.777
0.861
2.183
2.296
1.859
1.886
2.247
2.292
1.921
2.329
2.410
1.997
55.6
60.4
50.8
60.2
57.3
57.8
66.6
71.5
74.0
49.1
BAO 3,5-Cl2
AII 2,6-Cl2
AII 2,6-(CH3)2
CBAT
DAT
EPIT
Tolazoline
9
10
a
Calculated by eq (1).
b
c
Conformational descriptor LowEne is the energy of the most stable conformation in the set of conformations belonging to each molecular model.
The descriptor SXZF is the fraction of area of molecular shadow in the XZ plane over area of enclosing rectangle.
The descriptor JX is a highly discriminating descriptor, whose values do not substantially increase with the molecule size and the number of rings
d
present.
e
Thermodynamic descriptor MR index of a substituent is a combined measure of its size and polarizability which was calculated using the atom
typing scheme of Ghose et al.30

A. Hirashima et al. / Bioorg. Med. Chem. 7 (1999) 2621±2628
2623
values using eq (1), di€erence between observed and
calculated pKi values, and descriptors used here to
obtain eq (1). Values of pKi, the logs of the reciprocals
of Ki, were used as OA-agonist activity index. Con-
formational descriptor LowEne is the energy of the most
stable conformation in the set of conformations
belonging to each molecular model. Topological indices
including Balaban index JX are 2-D descriptors based
on graph theory concepts. These indices have been
widely used in QSAR studies. They help to di€erentiate
the molecules according mostly to their size, degree of
branching, ¯exibility, and overall shape. Balaban index
JX is a highly discriminating descriptor, whose values
do not substantially increase with the molecule size and
the number of rings present. The geometric and spatial
descriptor of shadow index helps to characterize the
shape of the molecules. The descriptors are calculated
by projecting the molecular surface on three mutually
perpendicular planes, XY, YZ, and XZ.29 These
descriptors depend not only on conformation but also
on the orientation of the molecule. To calculate them,
the molecules are ®rst rotated to align the principal
moments of inertia with the X, Y, and Z axes. Fraction
of area of molecular shadow in the XZ plane over area
of enclosing rectangle (SXZF). Thermodynamic
descriptor molecular refractivity (MR) index of a sub-
stituent is a combined measure of its size and polariz-
ability which was calculated using the atom typing
scheme of Ghose et al.30
[the sum of all (nonzero) p(k), the number of ways of
choosing k non-adjacent edges from the graph], and the
Molecular Connectivity Index; spatial descriptors such
as, radius of gyration, Jurs charged partial surface area
descriptors, molecular surface area, density, principal
moment of inertia, and molecular volume; structural
descriptors such as, MW (molecular weight), Rotlbonds
(number of rotatable bonds), Hbond acceptor (number
of hydrogen bond acceptors), Hbond donor (number of
hydrogen bond donors); thermodynamic descriptors
such as, AlogP (log of the partition coecient), Fh2o
(desolvation free energy for water), Foct (desolvation
free energy for octanol), and Hf (heat of formation).
MSA
The goal of MSA31,32 is to generate a QSAR equation
that incorporates spatial molecular similarity data. The
process has been described by Hop®nger and Burke.33
The outcome of the MSA process is an optimized
QSAR that can be used for activity estimation and
ligand evaluation. The set of choices available for each
task is employed to generate trial QSARs. The QSAR
that corresponds to the best ®t between observed activ-
ities and computed molecular descriptors de®nes the
speci®c requirements for each MSA task. The ability to
minimize a structure against a model allows one to
¯exibility ®t a structure into the model. The minimiza-
tion can also be used as a shape- (and not atom-)based
alignment technique. Applying the ®tting process to a
set of molecules will force all the molecules to adopt a
shape that is consistent with the model, leading to eq
(2).
According to eq (1), the positive JX, and the negative
LowEne and SXZF terms mean that the larger JX
values, and the smaller SXZF and conformational
energy LowEne, the greater the activity. The positive
MR2 term indicates that the OA-agonist activity is
mimimun when MR value is 57.7808. The use of other
parameters instead of or the addition of other para-
meters to eq (1) did not improve the correlation, e.g.
fragment constant descriptors that relate the e€ect of
substituents on a ``reaction center'' from one type of
process to another such as, sm and sp (electronic e€ect
sigma meta and sigma para, respectively), F and R
[decompositions of sigma para constant into an induc-
tive (polar) part (F) and a resonance part (R) for the
case when the substituent is conjugated with the reac-
tion center producing through-resonance e€ects], pi
(hydrophobic character), HA (hydrogen bond accep-
tor), HB (hydrogen bond donor), Sterimol-L (steric
length parameter), Sterimol-B1 through B4 (steric dis-
tances perpendicular to the bond axis), and Sterimol-B5
(the overall maximum steric distance perpendicular to
the bond axis); conformational descriptors such as,
Energy (the energy of the currently selected conforma-
tion in the study) and EPenalty (the di€erence between
Energy and LowEne); electronic descriptors such as,
Charge (sum of partial charges), Fcharge (sum of for-
mal charges), Apol (sum of atomic polarizabilities),
Dipole (dipole moment), HOMO (highest occupied
molecular orbital energy), LUMO (lowest unoccupied
molecular orbital energy), and Sr (superdelocal-
izability); topological descriptors (2-D descriptors based
on graph theory concepts) such as, Zagreb Index (the
sum of the squares of vertex valencies), Hosoya Index
2
pKi ˆ 39:1401 0:960171ꢀIC†
6:57924RG
2
0:05538ꢀ152:765 COSV†
ꢀ2†
where n ˆ 10; r2 ˆ 0:984; F ˆ 119:504; CV r2 ˆ 0:961,
and Bsr2 ˆ 0:984 Æ 0. The descriptors generated for
MSA, the activities of the compounds, and the predic-
tions using their top model are shown in Table 2. Com-
mon overlap steric volume (COSV) is the common
volume between each individual molecule and the
molecule selected as the reference compound. This is a
measure of how similar in steric shape the analogues are
to the shape reference. Spatial descriptor radius of
gyration (RG) is calculated using the following equa-
p
tion: RG ˆ Æꢀxi ‡ yi ‡ zi†=N, where N is the number
of atoms and x, y, and z are the atomic coordinates
relative to the center of mass. In the approach of infor-
mation-content (IC) descriptors, molecules are viewed
as structures which can be partitioned into subsets of
elements that are in some sense equivalent. The notion
of equivalence depends on the particular descriptor.
According to eq (2), the negative (COSV)2 term indi-
cates that the OA-agonist activity is maximun when
COSV value is 152.765.
MFA
MFA evaluates the energy between a probe and a mole-
cular model at a series of points de®ned by a rectangular

2624
A. Hirashima et al. / Bioorg. Med. Chem. 7 (1999) 2621±2628
Table 2. MSA descriptors and prediction using top model
pKi
372 and 298, respectively. The descriptor CH3/164 is the
energy between methyl probe and the molecule at the
rectangular point 164. According to eq (3), the negative
(H+/372)2 term indicates that the OA-agonist activity
is maximum when H+/372 value is zero.
No.
ICa
RGb COSVc Observed Calculatedd Deviation
1
2
3
4
5
6
7
8
9
10
3.236 3.212 152.765
3.507 3.340 170.173
8.21
5.21
4.00
7.16
8.05
8.41
4.11
7.21
5.27
7.11
7.95
5.36
3.95
7.16
8.01
8.84
4.31
6.89
5.22
7.06
0.26
0.15
0.05
0.00
0.04
0.43
0.20
0.32
0.05
0.05
Figure 2 shows the most potential OA agonist 6 and the
least potential OA agonist 3 embedded in a MFA gen-
erated from training set of OA agonists and the points
used in the resulting QSAR model. Only points con-
tributing to the equation are represented (a green sign
stands for a positive contribution). Note here that all
the points map in the variable part of the molecule 6,
where structural modi®cations have been tested. The dark
blue methyl probe represents for highly active compound
6 depicted by light blue cylinder, while red methyl probe
represents for the least active compound 3 depicted by
dark red cylinder. The most active compound 6 has an
imidazolidine ring overlapping on the methyl probe
(blue region), which is favorable for OA-agonist activ-
ity. Meanwhile, the least active compound 3 has an
oxazoline ring embedded in the methyl probe (red
region), which is less favorable for OA-agonist activity.
3.182 3.399
96.693
3.006 3.177 109.328
3.093 3.228 140.051
3.093 3.210 182.187
3.323 3.374 116.155
3.250 3.360 170.221
3.500 3.248 138.431
3.022 3.044
93.439
a
In the approach of IC descriptors, molecules are viewed as struc-
tures which can be partitioned into subsets of elements that are in
some sense equivalent.
Spatial descriptor RG is calculated using the following equation:
p
b
RG ˆ Æꢀxi ‡ yi ‡ zi†=N, where N is the number of atoms and x, y,
and z are the atomic coordinates relative to the center of mass.
COSV is the common volume between each individual molecule and
the molecule selected as the reference compound.
c
d
Calculated by eq (2).
or spherical grid. These energies were added to the study
table to form new columns headed according to the
probe type. The new columns were used as independent
X variables in the generation of QSAR, leading to eq
(3).
RSM
The energies of interaction between the RSM and each
molecular model were added to the study table as new
columns, which were used for generating QSARs.
Instead of one total number which is the sum of the
interactions evaluated between each point on the sur-
face and each molecular model, leading to one extra
column in the study table, the energies at each surface
point are available. Depending on the size of the drug
2
pKi ˆ7:20883 0:00301ꢀH ‡ =372† ‡ 0:062649
ꢀ3†
ꢀH ‡ =298† ‡ 0:02519CH3=164
where n ˆ 10, r2 ˆ 0:995, F ˆ 407:918, CV r2 ˆ
0:986, and Bsr2 ˆ 0:995 Æ 0. The descriptors generated
for MFA, the activities of the compounds, and the pre-
dictions using their top model are shown in Table 3. The
surface is generated from a ``Shape Field''. The atomic
coordinates of the contributing models are used to
compute ®eld values on each point of a 3-D grid. The
descriptors H+/372 and H+/298 are the energy between
proton probe and the molecule at the rectangular points
Table 3. MFA descriptors and prediction using top model
pKi
No. H+/372a H+/298a CH3/164b Observed Calculatedc Deviation
1
2
3
4
5
6
7
8
9
10
28.25
13.657
1.637
10.681
0.895
0.936
30.000
19.321
2.039
7.760
0.038
0.908
30.000
0.492
0.671
30.000
1.039
2.269
0.614
17.229
0.482
8.21
5.21
4.00
7.16
8.05
8.41
4.11
7.21
5.27
7.11
8.02
5.15
3.82
7.22
8.01
8.46
4.22
7.22
5.42
7.19
0.19
0.06
0.18
0.06
0.04
0.05
0.11
0.01
0.15
0.08
30.000
30.000
2.939
2.272
14.115
24.675
5.764
30.000
1.380
Figure 2. The most potential OA agonist 6 and the least potential OA
agonist 3 embedded in a MFA generated from OA agonist training set
and the points used in the resulting QSAR model. The blue methyl
probe represents for highly active compound 6 depicted by light blue
cylinder, while red methyl probe represents for the least active com-
pound 3 depicted by dark red cylinder. Only points contributing to the
equation are represented (a green sign stands for a positive contribu-
tion). Note here that all the points map in the variable part of the
a
The descriptors H+/372 and H+/298 are the energy between pro-
ton probe and the molecule at the rectangular points 372 and 298,
respectively.
b
The descriptor CH3/164 is the energy between methyl probe and the
molecule at the rectangular point 164.
c
Calculated by eq (3).

A. Hirashima et al. / Bioorg. Med. Chem. 7 (1999) 2621±2628
2625
molecules, this is potentially a great number of surface
points. After adding the receptor surface point energies
to the study table, a QSAR was calculated using the
receptor surface energies and biological activities, lead-
ing to eq (4).
produces a surface that entirely encloses the molecules
over which it is generated. The surface has no holes and
is known as a closed model. RSMs are best constructed
from a set of the most active analogues that are chosen
to cover the variety of steric and electrostatic variations
likely to appear in the test data. The approach we took
was to automatically build a set of di€erent RSMs from
di€erent combinations of the most active analogues,
and then use a variable-selection technique such as
GFA to discover the RSM whose descriptors yield the
best QSARs of the full training set. GFA allows the
discovery and use of nonlinear descriptors by using
spline-based terms. A RSM was generated (Fig. 3) using
some subset of the most active structures (1, 5, and 6).
The rationale underlying this model is that the most
active structures tend to explore the best spatial and
electronic interactions with receptor, while the least
active do not and tend to have unfavorable steric or
electronic interactions. The best model generated using
the descriptors from the closed RSM is given in eq (4).
Figure 3 shows the top three molecules embedded in the
RSM generated from them and the points used in the
resulting QSAR model colored by electrostatic poten-
tial. Only points contributing to the equation are repre-
sented (a green sign stands for a positive contribution
and a red sign stands for a negative one). Note here that
all the points map in the variable part of the molecules Ð
the part where structural modi®cations have been tested.
pKi ˆ 8:2137 ‡ 2:7633ELE=137 1:52881
2
 ꢀELE=546†
0:060721VDW=780
ꢀ4†
where n ˆ 10, r2 ˆ 0:989, F ˆ 184:362, CV r2 ˆ
0:977, and Bsr2 ˆ 0:989 Æ 0. Once the desired RSM has
been constructed, all the structures in the training and
test sets were evaluated against the model. The evalua-
tion consists of computing several energetic descriptors
that are based upon the interactions between ligand and
model. The two functions used to calculate the ®eld
values are a van der Waals function and a Wyvill soft
object function. For this work, three descriptors were
generated. The descriptors ELE/137 and ELE/546 are
the electrostatic interaction energy of the molecule with
the receptor at points 137 and 546, respectively. The
descriptor VDW/780 is the van der Waals interaction
energy of the molecule with the receptor at point 780.
According to eq (4), the negative (ELE/546)2 term indi-
cates that the OA-agonist activity is maximun when
ELE/546 value is zero. By using receptor data to
develop a QSAR model, the goodness of ®t can be
evaluated between a candidate structure and a postu-
lated pseudo-receptor.
In drug discovery, it is common to have measured
activity data for a set of compounds acting upon a par-
ticular protein but not to have knowledge of the three-
dimensional structure of the protein active site. In the
absence of such three-dimensional information, one can
attempt to build a hypothetical model of the receptor
site that can provide insight about receptor site char-
acteristics. Such a model is known as a RSM, which
The descriptors generated for RSM, the activities of the
compounds, and the predictions using their top model
are shown in Table 4. An RSM represents essential
information about the hypothetical receptor site as a
three-dimensional surface with associated properties
mapped onto the surface model. The location and shape
of the surface represent information about the steric
nature of the receptor site: the associated properties
represent other information of interest, such as hydro-
phobicity, partial charge, electrostatic potential, and
hydrogen-bonding propensity. The isosurface procedure
Table 4. RSM descriptors and prediction using top model
pKi
No. ELE/137a ELE/546a VDW/780b Observed Calculatedc Deviation
1
2
3
4
5
6
7
8
9
10
0.191
0.530
0.365
1.018
0.132
0.153
0.002
0.142
0.643
0.000
0.525
1.239
1.427
1.591
0.219
0.483
1.673
0.164
0.013
0.741
0.017
0.073
2.116
0.239
0.017
0.195
0.149
11.012
18.098
0.102
8.21
5.21
4.00
7.16
8.05
8.41
4.11
7.21
5.27
7.11
8.32
5.45
3.96
7.17
7.78
8.29
3.95
7.11
5.34
7.38
0.11
0.24
0.04
0.01
0.27
0.12
0.16
0.10
0.07
0.27
Figure 3. The three most active compounds 1 (green), 5 (purple), and
6 (red) embedded in a closed RSM generated from them and the
points used in the resulting QSAR model colored by electrostatic
potential. A purple sign stands for a negative contribution of electro-
static potential. Only points contributing to the equation are repre-
sented (a green sign stands for a positive contribution and a red sign
stands for a negative one). Note here that all the points map in the
variable part of the moleculesÐthe part where structural modi®ca-
tions have been tested.
a
The descriptors ELE/137 and ELE/546 are the electrostatic interac-
tion energy of the molecule with the receptor at points 137 and 546,
respectively.
b
The descriptor VDW/780 is the van der Waals interaction energy of
the molecule with the receptor at point 780.
c
Calculated by eq (4).

2626
A. Hirashima et al. / Bioorg. Med. Chem. 7 (1999) 2621±2628
provides compact and quantitative descriptors which
capture three-dimensional information about a putative
receptor site. This model ought to be predictive and
suciently reliable to guide the medicinal chemist in the
design of novel compounds. These descriptors were used
for predictive QSAR models. This approach is e€ective
for the analysis of data sets where activity information
is available but the structure of the receptor site is
unknown. RSM attempts to postulate and represent the
essential features of a receptor site itself, rather than the
common features of the molecules that bind to it. RSMs
di€er from pharmacophore models in that the former
tries to capture essential information about the receptor,
while the latter captures information about the com-
monality of compounds that bind. Pharmacophore
models tend to be geometrically underconstrained
(while topologically overconstrained); this steric under-
constraint leads to false positives, that is, compounds
that are deemed active by the model but which are
inactive when tested. They postulate a 3-D arrangement
of atoms recognizable by the active site in terms of the
similarity of functional groups common to the set of
binding molecules. RSMs, on the other hand, tend to be
geometrically overconstrained (and topologically neu-
tral) since, in the absence of steric variation in a region,
they assume the tightest steric surface which ®ts all
training compounds. RSMs do not contain atoms, but
try to directly represent the essential features of an
active site by assuming complementarity between the
shape and properties of the receptor site and the set of
binding compounds. The RSM application uses 3-D
surfaces that de®ne the shape of the receptor site by
enclosing the most active members (after appropriate
alignment) of a series of compounds.
Table 5. Predicted activity of OA agonists from eqs (1)±(4) in the test
set
Compound
No.
Predicted activity (pKi)
R
eq (1)
eq (2)
eq (3)
eq (4)
11
12
13
14
15
16
17a
CAO
AII H
AII 2,4,5-Cl3
CAT
Naphazoline
Synephrine
FBAO
8.768
7.216
9.383
7.925
7.638
9.350
3.232
7.234
9.317
8.254
7.319
7.739
8.175
1.335
9.050
8.593
8.616
9.039
6.994
7.267
7.217
7.922
8.259
8.012
8.261
8.219
7.892
7.552
a
The experimental activity (pKi) of 17 was less than 4.000.
Conclusion
QSAR modeling is an area of research pioneered by
Hansch and Fujita. QSAR attempts to model the activity
of a series of compounds using measured or computed
properties of the compounds as shown in eq (1). In this
report, QSAR has been extended by including the ana-
lysis of three-dimensional information about the series
by three-dimensional shape descriptors as illustrated by
the MSA approach in eq (2), by ®eld descriptors as
illustrated by MFA approach in eq (3), and by receptor
descriptors as illustrated by RSM approach in eq (4),
respectively. Some models were statistically signi®cant
and were used to correctly predict the activities of a set
of training molecules ranging over 4 orders of magni-
tude (max pKi 8.41 and min pKi 4), indicating that these
models could be useful tools to design active OA ago-
nists. This set included a variety of types of molecules
and for this types of training set, the use of these meth-
ods was appropriate. These methods generate multiple
models that can be checked easily for validity.
The predictive character of the QSARs was further
assessed using molecules, whose structures are shown in
Figure 1b, outside of the training set. The best statisti-
cally signi®cant eqs (1)±(4) were applied to access some
OA agonists. The predicted values of these molecules
are listed in Table 5. Some OA agonists of 2-(arylimi-
no)oxazolidine (AIO) 11, 2-(arylimino)imidazolidine
(AII) 12 and 13, 2-(4-chloroanilino)thiazolidine (CAT)
14, naphazoline 15, and synephrine 16 were active
according to eqs (1)±(4) in inhibition of intracellular
cAMP production by 1 pmol PBAN/intersegment. The
most active compounds are found to be di€erent from
di€erent equations, e.g. eq (1) predicts 13 and 16 as the
most active, eq (2) 12 as the most active, eq (3) 11 and
14 as the most active, and eq (4) none so active. A dis-
tinguishing characteristic of the eqs (3) and (4) is that
they overestimated the OA-agonist activity of 1 (experi-
mental pKi: less than 4, estimated pKi: 7.217 and 7.552
for eqs (3) and (4), respectively). This result may imply
that the process of calculating an MFA and RSM does
not treat 17-like structures reasonably. Thus, the above
data suggest that phenyl ring substitution requirements
for AIO and NC derivatives active as octopaminergic
agonists di€er substantially from each other and other
various types of OA agonists could be potent inhibitors
in sex-pheromone production of female moths, although
the number of compounds tested here is still limited to
draw any conclusions.
OA is not likely to penetrate either the cuticle or the
central nervous system of insects e€ectively, since it is
fully ionized at physiological pH. Derivatization of the
polar groups would be one possible solution to this
problem in trying to develop potential pest-control
agents. The above QSAR, MSA, MFA, and RSM stu-
dies show that agonists with certain substituents can be
potential ligands to OA receptors, although the number
of compounds tested here is still limited to draw any
conclusions. They may help to point the way towards
developing extremely potent and relatively speci®c OA
agonists, leading to potential inhibitors in sex-pher-
omone production of moths. In order to optimize the
activities of these compounds as OA-agonists, more
detailed experiments are in progress.
Experimental
Synthesis of OA agonists
1-(2,6-Diethylphenyl)imidazolidine-2-thione (EPIT) 9 was
synthesized by the cyclization of monoethanolamine
hydrogen sulfate with 2,6-diethylphenylisothiocyanate
in the presence of sodium hydroxide as described in the

A. Hirashima et al. / Bioorg. Med. Chem. 7 (1999) 2621±2628
2627
previous report.34 AIOs 1 and 2, 2-(2,3,4,5-tetrachloro-
anilino)oxazolidine (CAO) 11, 2-(substituted benzyl-
amino)-2-oxazolines (BAOs) 3 and 4, and 2-(3-tri-
¯uoromethylbenzylamino)-2-oxazoline (FBAO) 1 were
obtained by cyclodesulfurizing the corresponding
thiourea with yellow mercuric oxide.35 2-(2,3-Dichloro-
benzylamino)-2-thiazoline (CBAT) 7, 2-(2,6-diethyl-
anilino)thiazolidine (DAT) 8, and CAT 14 were
synthesized by cyclization of the corresponding thiourea
with concd hydrogen chloride.36 AII 6, 12, and 13 were
prepared according to a reported method by re¯uxing
the corresponding aniline and 1-acetyl-2-imidazolidone
in phosphoryl chloride followed by hydrolysis.37 The
(Burlington, MA) on a Silicon Graphics O2, running
under the IRIX 6.3 operating system. Due to the large
number of points used as independent variables, GFA
was used to derive the QSAR models. The GFA algo-
rithm was initially conceived by taking inspiration from
two seemingly disparate algorithms: Holland's genetic
algorithm40 and Friedman's41 multivariate adaptive
regression splines (MARS) algorithm. Genetic algo-
rithms are derived from an analogy with the evolution
of DNA. In this analogy, individuals are represented
by a one-dimensional string of bits. An initial popula-
tion of individuals is created, usually with random
initial bits. A ®tness function is used to estimate the
quality of an individual, so that the best individuals
receive the best ®tness scores. Individuals with the best
scores are more likely to be chosen to mate and to pro-
pagate their genetic material to o€spring through the
crossover operation, in which pieces of genetic material
are taken from each parent and recombined to create
the child. After many mating steps, the average ®tness
of the individuals in the population increases as good
combinations of genes are discovered and spread
through the population. Genetic algorithms are espe-
cially good at searching problem spaces with a large
number of dimensions, as they conduct a very ecient
directed sampling of the large space of possibilities.
Friedman's MARS algorithm is a statistical technique
for modeling data. It provided an error measure, called
the lack of ®t (LOF) score, that automatically penalized
models with too many features. It automatically selects
which features are to be used in the models. It is better
at discovering combinations of features that take
advantage of correlations between multiple features. It
incorporates Friedman's LOF error measure, which
estimates the most appropriate number of features,
resists over®tting, and allows user control over the
smoothness of ®t. It also inspired the use of splines as a
powerful tool for non-linear modeling. The GFA uses a
genetic algorithm to perform a search over the space of
possible QSAR models using the LOF score to estimate
the ®tness of each model. Such evolution of a popula-
tion of randomly constructed models leads to the dis-
covery of highly predictive QSARs. The GFA algorithm
approach has a number of important advantages over
other techniques: it builds multiple models rather than a
single model; it can use a larger variety of equation term
types in construction of its models (for example, splines,
step functions, or high-order polynomials); it provides,
through study of the evolving models, additional infor-
mation not available from standard regression analysis
(such as the preferred model length and useful partitions
of the data set).
1
structures of the compounds were con®rmed by H and
13C NMR measured with a JEOL JNM-EX400 spec-
trometer at 400 MHz, tetramethyl silane (TMS) being
used as an internal standard for 1H NMR and elemental
analytical data. Tolazoline (10) and naphazoline (15)
were obtained from Janssen Chimica (Beerse, Belgium)
and Aldrich Chemical Company, Inc. (Milwaukee, WI),
respectively; clonidine (5) and synephrine (16) from
Sigma Chemical Company (St. Louis, MO).
Insect culture
The study was conducted on H. armigera. The larvae
were raised on an arti®cial diet at a constant tempera-
ture of 25ꢁC and 14:10 (light:dark) photoperiod as
reported previously.38 Pupae were sexed and males and
females were allowed to emerge separately.
Intracellular cAMP levels
Compounds found to inhibit pheromone production in
vivo by at least 50% were subsequently analyzed using
the assay for cAMP production by intersegmental
membranes.39 Ovipositor tips, consisting of the 8th and
9th abdominal segments with the attached interseg-
mental membrane, were removed during the 10±12th h
photophase from 2- or 3-day old females. The interseg-
mental membranes were isolated by dissection and
washed twice in Pipes bu€ered physiological medium
(21 mM KCl, 12 mM NaCl, 3 mM CaCl2, 18 mM
MgCl2, 85 mM glucose, and 43 mM trehalose in 5 mM
Pipes bu€er brought to pH 6.6 using 0.1 N KOH). The
intersegmental membranes were incubated in 10 mL
medium containing 1 mg/mL isobutylmethylxanthine
(Sigma, USA) in the presence of synthetic Hez-PBAN (1
pmol/intersegment) and the test compounds for 10 min.
The reaction was stopped by transferring the interseg-
mental membranes into a 1:5 mixture of 20% perchloric
acid and Hepes bu€er (50 mM) on ice. After incubation
on ice for 20 minutes the intersegmental membranes
were homogenized, titrated to pH=7 with 1.5 M KOH
and centrifuged at 2000Âg for 10 min. The supernatant
was removed and used for radioimmunoassay (RIA).
The RIA for cAMP was performed as described pre-
viously.38
Molecular modeling
Once a reasonable RSM has been de®ned, a series of
structures can be evaluated against the model. When a
receptor model has been generated and the models
have been aligned, a QSAR can be built using data
from the receptor±structure interactions. The results of
the minimization procedure were used as descriptors
either to re®ne the model or to predict activity. Three-
dimensional energetics descriptors were calculated from
QSAR calculations
All experiments were conducted with Molecular Simu-
lations Incorporated's Cerius2 3.5 QSAR environment

2628
A. Hirashima et al. / Bioorg. Med. Chem. 7 (1999) 2621±2628
RSM±ligand interaction. These three-dimensional
descriptors were used in QSAR analysis. A RSM repre-
sents the global volume that can accommodate one or
more molecules and can be seen as the shape of an
active site built from the ligands that ®t into it in their
``active'' conformation. The descriptors used in this
study account for phenomena that occur at the contact
surface between the ligands and the protein active site.
This concept di€ers from that used in classical 3-D-
QSAR in which ®elds of molecules are represented using
grids in MFA. The technique resembles MFA but,
instead of a rectangular grid, the points considered are
taken from the receptor surface. Therefore, they are
probably more chemically relevant than a rectangular
grid because they exist on a surface that is shaped like a
molecule, and even better, a surface constructed from a
subset of active molecules. The MFA formalism calcu-
lates probe interaction energies on a rectangular grid
around a bundle of active molecules. Receptor calcu-
lates molecule±receptor model interaction energies on a
receptor surface, and, like MFA, these energies can
serve as input for the calculation of a QSAR relation-
ship. Just as each energy associated with an MFA grid
point can be used, so can each point on the surface of a
receptor model. Hopefully, the receptor surface is better
able to sample the environment of the molecule than a
rectangular grid, leading to better results.
14. Fabrias, G.; Barrot, M.; Camps, F. Insect Biochem. Molec.
Biol. 1995, 25, 655.
15. Zhu, J.; Millar, J.; Loefstedt, C. Archs. Insect Biochem.
Physiol. 1995, 30, 41.
16. Fang, N.; Teal, P. E. A.; Tumlinson, H. Archs. Insect Bio-
chem. Physiol. 1996, 32, 249.
17. Foster, S.; Roelofs, W. L. Archs. Insect Biochem. Physiol.
1996, 32, 135.
18. Rafaeli, A.; Gileadi, C. Invertebrate Neuroscience 1997, 3,
223.
19. Jurenka, R. A. Arch. Insect Biochem. Physiol. 1996, 33,
245.
20. Soroker, V.; Rafaeli, A. Insect Biochem. 1989, 67, 1.
21. Rafaeli, A.; Soroker, V.; Kamensky, B.; Raina, A. K. J.
Insect Physiol. 1990, 36, 641.
22. Arima, R.; Takahara, K.; Kadoshima, T.; Numazaki, F.;
Ando, T.; Uchiyama, M.; Nagasawa, H.; Kitamura, A.;
Suzuki, A. Appl. Entomol. Zool. 1991, 26, 137.
23. Jurenka, R. A.; Jacquin, E.; Roelofs, W. L. Proc. Natl.
Acad. Sci. USA 1991, 88, 8621.
24. Fonagy, A.; Matsumoto, S.; Schoofs, L.; Loof, A. De;
Mitsui, T. Biosci. Biotech. Biochem. 1992, 56, 1692.
25. Matsumoto, S.; Ozawa, R.; Nagamine, T.; Kim, G.-H.;
Uchiumi, K.; Shono, T.; Mitsui, T. Biosci. Biotech. Biochem.
1995, 59, 560.
26. Rafaeli, A.; Gileadi, C. Insect Biochem. Molec. Biol. 1995,
25, 827.
27. Rafaeli, A.; Gileadi, C. Insect Biochem. Molec. Biol. 1996,
26, 797.
28. Rafaeli, A.; Gileadi, C.; Fan, Y.; Meixun, C. J. Insect
Physiol. 1997, 43, 261.
29. Rohrbaugh, R. H.; Jurs, P. C. Analytica Chimica Acta
1987, 199, 99.
Acknowledgements
30. Ghose, A. K.; Pritchett, A.; Crippen, G. M. J. Comp.
Chem. 1988, 9, 80.
31. Burke, B. J.; Hop®nger, A. J. J. Med. Chem. 1990, 33, 274.
32. Burke, B. J.; Hop®nger, A. J. In Molecular Similarity;
Johnson, M. A.; Maggiora, G. M., Eds.; John Wiley and Sons:
New York, 1990; pp 11±73.
This work was supported in part by a Grant-in-Aid for
Scienti®c Research from the Ministry of Education,
Science, and Culture of Japan.
33. Hop®nger, A. J.; Burke, B. J. In Concepts and Applications
of Molecular Similarity; Johnson, M. A.; Maggiora, G. M.,
Eds.; John Wiley and Sons: New York, 1990; p 173.
34. Hirashima, A.; Shinkai, K.; Kuwano, E.; Taniguchi, E.;
Eto, M. Biosci. Biotech. Biochem. 1998, 62, 1179.
35. Hirashima, A.; Pan, C.; Katafuchi, Y.; Taniguchi, E.; Eto,
M. J. Pestic. Sci. 1996, 21, 419.
36. Hirashima, A.; Tarui, H.; Eto, M. Biosci. Biochem. Bio-
tech. 1994, 58, 1206.
37. Nathanson, J. A.; Kaugars, G. J. Med. Chem. 1989, 32,
441.
38. Rafaeli, A.; Soroker, V. Mol. Cell. Endocrinol. 1989, 65,
43.
39. Soroker, V.; Rafaeli, A. Insect Biochem. Molec. Biol. 1995,
25, 1.
40. Holland, J. In Adaptation in Arti®cial and Natural Systems;
University of Michigan Press, 1975.
41. Friedman, J. In Multivariate Adaptive Regression Splines;
Technical Report 102, Laboratory for Computational Statis-
tics, Department of Statistics, Stanford University: Stanford,
CA, 1988 (revised 1990).
References
1. Evans, P. D. Comp. Mol. Neurobiol. 1993, 287.
2. Evans, P. D. J. Physiol. 1981, 318, 99.
3. Roeder, T.; Nathanson, J. A. Neurochem. Res. 1993, 18, 921.
4. Roeder, T.; Gewecke, M. Biochem. Pharm. 1990, 39, 1793.
5. Roeder, T. Life Science 1991, 50, 21.
6. Roeder, T. Br. J. Pharmac. 1995, 114, 210.
7. Roeder, T. Eur. J. Pharmac. 1990, 191, 221.
8. Hirashima, A.; Shinkai, K.; Pan, C.; Kuwano, E.; Tani-
guchi, E.; Eto, M. Pestic. Sci. 1999, 55, 119.
9. Hirashima, A.; Pan, C.; Shinkai, K.; Tomita, J.; Kuwano,
E.; Taniguchi, E.; Eto, M. Bioorg. Med. Chem. 1998, 6, 903.
10. Pan, C.; Hirashima, A.; Kuwano, E.; Eto, M. J. Molec.
Model. 1997, 3, 455.
11. Hirashima, A.; Pan, C.; Kuwano, E.; Taniguchi, E.; Eto,
M. Bioorg. Med. Chem. 1999, 7, 1437.
12. Raina, A. K. Ann. Rev. Entomol. 1993, 38, 320.
13. Ma, P. W. K.; Roelofs, W. Insect Biochem. Molec. Biol.
1995, 25, 467.