3D QSAR study of hypolipidemic asarones by comparative molecular surface analysis

Article abstract of DOI:10.1016/j.bmc.2005.10.014

Three-dimensional quantitative structure-activity relationship (3D QSAR) modeled for α-asarone derivatives using the comparative molecular surface analysis (CoMSA) allowed us to reveal a correlation between the activity of these compounds and the electros

Full text of DOI:10.1016/j.bmc.2005.10.014

Bioorganic & Medicinal Chemistry 14 (2006) 1630–1643
3D QSAR study of hypolipidemic asarones by
comparative molecular surface analysis
Tomasz Magdziarz,a Bozena Łozowicka,b Rafał Gieleciak,a Andrzej Ba˛k,a
a,
b,c
*
´
Jarosław Polanski and Zdzisław Chilmonczyk
aDepartment of Organic Chemistry, Institute of Chemistry, University of Silesia, PL-40-006 Katowice, Poland
bDepartment of Organic Chemistry, Institute of Chemistry, University of Bialystok, J. Pilsudskiego 11/4, PL-15-443 Bialystok, Poland
cNational Institute of Public Health, Chelmska 30/34, 00-725 Warszawa, Poland
Received 3 August 2005; revised 30 September 2005; accepted 6 October 2005
Available online 3 November 2005
Abstract—Three-dimensional quantitative structure–activity relationship (3D QSAR) modeled for a-asarone derivatives using the
comparative molecular surface analysis (CoMSA) allowed us to reveal a correlation between the activity of these compounds
and the electrostatic potential at the molecular surface. The grid formalism (s-CoMSA) allowed us to indicate a pharmacophore
that is of key importance for compound activity. The CoMSA formalism coupled with the iterative variable elimination method
gives a highly predictive model.
ꢀ 2005 Elsevier Ltd. All rights reserved.
1. Introduction
Increasing attention is also being focused on other lipo-
protein fractions, such as high-density lipoprotein cho-
lesterol (HDL-C) and triglycerides (TG), as additional
potential targets of therapy. Elevated serum TG com-
bined with low HDL-C, a condition often associated
with smaller, dense LDL particles, is frequently referred
to as atherogenic dyslipidemia or the Ôlipid triad.Õ9
Atherosclerosis and subsequent cardiovascular diseases
still remain one of the major death risks in the industri-
alized and developing countries. The initiation of ath-
erosclerosis is most likely caused by inflammatory
responses, hyperlipidemia, and blood clotting factors.
Many factors were found to be associated with the
development of atherosclerosis and cardiovascular dis-
eases.1–5 The link between elevated cholesterol and cor-
onary heart diseases (CHD) has been clearly established,
and clinical trials have found that a 1% reduction in ser-
um total cholesterol reduced CHD risk by 2%.6
At present, the therapy of lipid metabolism disorders is
based on the use of drugs having hypolipidemic activity
like fibric acid derivatives, anion exchange resins which
sequester the bile acids, the inhibitors of 3-hydroxymeth-
yl-coenzyme A (3-HMG Co-A) reductase, en enzyme in-
volved in de novo sterol synthesis, probucol, lifibrol,
and many others.10,11 a-Asarone (1) is an active compo-
nent of Acorus calamus Linn, Acorus gramineus Soland,
and Guatteria gaumeri.12 a-Asarone and its analogues
are known to be endowed with hypolipidemic activity13,14
and have been the subject of many pharmacological,15–17
toxicological,18,19 synthetic,20 and QSAR studies.21,22
Basically, cells (except for hepatic and ileum cells) do
not synthesize cholesterol de novo but obtain it from
blood and the cholesterol that accumulates in athero-
sclerotic lesions originates, primarily, in plasma lipopro-
teins.7 The lowering of abnormally elevated levels of
atherogenic lipoproteins (chylomicrons, very low densi-
ty lipoproteins—VLDL-C, and low density lipopro-
teins—LDL-C) is now accepted as the first line of
approach to the treatment of hyperlipidemic patients.8
Different a-asarone structural features were found to
influence the hypolipidemic activity. Chamorro et al.
examined analogues possessing
a
dimethoxylated
unconjugated propenyl side chain (mice) and Cruz
et al.23 analogues with saturated side chain (rats). Anal-
ysis of a length of a hydrophobic chain has shown that
the most active analogues had the shortest side chain.
Keywords: a-Asarone analogues; Hypolipidemic activity; QSAR study.
*
Corresponding author. Tel.: +48 32 3591197; fax: +48 32 2599978;
e-mail addresses: bozena@uwb.edu.pl; polanski@us.edu.pl;
chilmon@il.waw.pl
0968-0896/$ - see front matter ꢀ 2005 Elsevier Ltd. All rights reserved.
doi:10.1016/j.bmc.2005.10.014

T. Magdziarz et al. / Bioorg. Med. Chem. 14 (2006) 1630–1643
1631
cholesterol concentration (HDL-C) [mmol/L]. The ath-
erogenic index for the analyzed series ranges from 0.10
(high activity) to 3.28 (low activity) (Table 2).
2.2. Hypolipidemic activity
Wistar rats weighing 300–320 g male were bought from
´ ´
Nofer Institute of Occupational Medicine, Łodz, Po-
land. The animals were equally divided into groups of
five animals each and maintained at temperature of
22 2 ꢁC, 45–80% relative humidity, and every 12 h
periods changing of light and darkness. All rats were
fed a high cholesterol diet (Murigran enriched with cho-
lesterol 1%, sodium cholate 0.2%, and olive oil 5%) for 7
days (Murigran-Łomna near Warsaw). Rats fed with
laboratory chow for the same duration as above were
used as noncholesterol control group. Compounds,
diluted in oil, were administered through gastric intuba-
tion at 80 mg/kg once a day for the duration of the
experiment. Group receiving clofibrate (150 mg/kg)
served as positive control. Compounds, diluted in oil,
were adjusted so that the rats were administered a vol-
ume of 5 mL/kg of body weight. Rats in the control
group received a similar volume of vehicle. At the end
of seven-day period, each animal was fasted for 16 h
and anesthetized. Blood samples were collected through
ocular puncture and centrifuged at 3000 rpm. Total cho-
lesterol (TC), HDL-cholesterol (HDL-C), and triglycer-
ides (TG) were determined using Alpha Diagnostics kits
on Clinic System—700 Beckman. All the data were sta-
tistically analyzed by StudentÕs t test. All the results were
calculated against noncholesterol and cholesterol con-
trol groups.
Figure 1. Molecular formulae of compounds 7–13.
During our work on a-asarone analogues, we synthe-
sized several new compounds and examined their hypo-
lipidemic activity.21,24–26 We also examined
relationship between the hypolipidemic activity and
molecular features. It appeared that the asarone activity
could be described, at least in part, using the pseudo- or
mini-receptor models.21,27 However, we could not have
indicated any clear molecular rule controlling the activ-
ity of these compounds.24–26
a
Some data concerning the compounds 1–40 were prelim-
inarly reported27 however, no experimental details are
available for compounds 7–13. Thus, in the present pub-
lication we describe the synthesis and biological evalua-
tion of compounds 7–13 (Fig. 1, Scheme 1).
We concentrate on the understanding of the structural
basis of the compoundsÕ pharmacological activity. Since
our previous models did not completely explain the
structure–activity relationship, we investigated 3D
QSAR by comparative molecular surface analysis, a
novel method described in our previous publica-
tions.28–36 We have found that this method indicates a
clear molecular basis for the activity.
2.3. CoMFA analysis and molecular alignment
All modeling work was performed using the Sybyl 6.2
software package run on Silicon Graphics O2 worksta-
tion. The initial geometry was optimized using the stan-
dard Tripos force field (POWELL method) with
0.005 kcal/mol energy gradient convergence criterion
and a distant-dependent dielectric constant. Charges
were calculated using the Gasteiger–Marsilli method
implemented in Sybyl. We used the FIT option of the
Sybyl to align the compounds analyzed. Parent a-asa-
rone, that is, a common fragment for all molecules,
was chosen as a template. Alignment was carried out
2. Methods
2.1. Data sets for the analysis
The chemical structures of the a-asarone derivatives are
shown in Table 1. Hypolipidemic activity reported was
measured (see Section 2.2, Table 1). As the activity data
in all our QSAR calculation, we used the atherogenic in-
dex ITG/HDL. It is calculated by the ratio of the triglycer-
ide concentration (TG) [mmol/L] to the HDL
Scheme 1. General procedure synthesis of compounds 7–13.

1632
T. Magdziarz et al. / Bioorg. Med. Chem. 14 (2006) 1630–1643
Table 1. Structures of a-asarone analogues and hypolipidemic activitya
Entry
Structure
TC
HDL-C
LDL-C
TG
Nonchole sterol diet 7–13
Cholesterol diet (CD) 7–13
. . .
. . .
ꢀ61 0.16**
+152 0.05**
100 0.06c
ꢀ91 0.07*
ꢀ39 0.02**
100 0.06b
100 0.33d
100 0.29e
CLO + CD
ꢀ23 0.14**
+24 0.06**
+7 0.14
ꢀ27 0.12**
1 + CD
+2.3 0.24
+57 0.06**
ꢀ43 0.17**
+75 0.30**
6 + CD
7 + CD
ꢀ16 0.28
+56 0.06**
ꢀ54 0.14**
+8 0.16
+7 1.16
+71.0 0.16*
ꢀ7.4 1.13
+54 0.31
8 + CD
9 + CD
10 + CD
11 + CD
12
ꢀ10 0.84**
+18 0.52**
+6 0.58**
+35 0.07**
+68 0.09*
+65 0.11*
+36 0.09
+10 0.06**
+26 0.09
ꢀ15.6 0.88
+12 0.64
ꢀ5 0.50**
ꢀ2 1.62*
+20 0.28*
+12 0.54
ꢀ16 0.19**
+12 0.09**
+26 0.18
ꢀ23 0.18
ꢀ15 0.10
+16 0.33
0
1.59
+13 0.25*
13
+13 0.38
a Expressed as a percentage of the cholesterol diet group (mean SD), n = 6.
b Cholesterol diet group 3.07 mmol/L.
c Cholesterol diet group 0.31 mmol/L.
d Cholesterol diet group 2.44 mmol/L.
e Cholesterol diet group 0.69 mmol/L; for compounds 7–13.
* Significantly different from the result for the cholesterol diet group control at p < 0.05.
** Significantly different from the result for the cholesterol diet group control at p < 0.01.

T. Magdziarz et al. / Bioorg. Med. Chem. 14 (2006) 1630–1643
Table 2. a-Asarone analogues and atherogenic index—ITG/HDL data
1633
Compound
Molecular formulae
ITG/HDL
TG (mmol/L)
HDL-C (mmol/L)
1
2.10
2.48
1.18
1.11
2
3
1.27
1.56
1.34
2.35
1.31
2.00
1.38
1.48
1.71
1.26
1.74
1.41
1.58
1.31
1.83
1.53
1.06
0.58
0.77
0.87
0.53
0.59
1.01
0.98
0.78
1.17
0.53
0.42
0.52
0.51
0.42
4
5
6
7
8
9
10
11
12
0.34
(continued on next page)

1634
T. Magdziarz et al. / Bioorg. Med. Chem. 14 (2006) 1630–1643
Table 2 (continued)
Compound
Molecular formulae
ITG/HDL
TG (mmol/L)
HDL-C (mmol/L)
0.39
13
14
15
16
17
18
19
20
21
22
23
24
2.05
0.8
0.36
0.45
0.16
0.27
0.13
0.14
0.47
0.15
0.10
0.45
0.22
0.21
0.24
0.09
0.19
0.07
0.08
0.2
0.59
0.53
0.57
0.71
0.56
0.57
0.43
0.66
0.73
0.64
0.1
0.07
0.29
0.14
0.64
(continued on next page)

T. Magdziarz et al. / Bioorg. Med. Chem. 14 (2006) 1630–1643
1635
Table 2 (continued)
Compound
Molecular formulae
ITG/HDL
TG (mmol/L)
HDL-C (mmol/L)
25
26
27
28
29
30
31
32
33
34
35
36
0.13
0.09
0.69
0.47
0.43
0.90
0.33
0.44
0.34
0.29
0.47
0.32
0.5
0.85
0.77
1.90
3.15
1.52
1.79
3.28
0.45
0.56
1.98
2.58
0.4
0.33
1.71
1.04
0.67
0.61
0.95
0.21
0.18
0.99
1.01
0.4
(continued on next page)

1636
T. Magdziarz et al. / Bioorg. Med. Chem. 14 (2006) 1630–1643
Table 2 (continued)
Compound
Molecular formulae
ITG/HDL
TG (mmol/L)
HDL-C (mmol/L)
0.5
37
38
39
40
1.92
0.96
1.06
0.77
0.43
1.96
1.28
0.80
0.54
0.6
0.54
by superimposing the atoms of benzene ring. We used
a
PLS procedures were programmed within the MATLAB
environment (MATLAB).37
single conformation for each molecule, which
provided an alignment illustrated in Supplementary
Materials Figure 1. The steric (Lennard-Jones) and
electrostatic fields around the set of compounds were
sampled with the probe atoms: sp3 carbon (charge +1
and 0) and hydrogen (charge +1), on the rectangular
grid that encompasses all aligned molecules (with mar-
gin of 3.0–4.0 A). For each molecule the energies of
˚
1769 grid points were calculated with 2 A spacing in a
lattice of 16 · 10 · 12. We kept a convention to truncate
the steric and electrostatic values at the level of
30.0 kcal/mol.
A PLS model was constructed for the centered data and
its complexity was estimated on the basis of the leave-
one-out cross-validation procedure (CV). In the leave-
one-out CV, one repeats the calibration m times, each
time treating the ith left-out object as the prediction ob-
ject. The dependent variable for each left-out object is
calculated on the basis of the model with one, two,
three, etc., factors.
˚
The root mean square error of CV for the model with j
factors is defined as
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
2.4. CoMSA analyses
P
2
ðobsd ꢀ predÞ
RMSECV ¼
;
For the calculation of shape descriptors we applied both
grid (s-CoMSA) and neural formalisms described in our
previous publications.28–36 Thus, each 3D molecular
representation is placed in its own virtual cubic grid
and molecular surface is calculated, respectively. The
electrostatic potential is calculated for the points ran-
domly sampled on the molecular surface and a mean
value of the electrostatic potential corresponding to
the respective points found in each grid cell is used to de-
scribe this cell. Grid cells are unfolded into vectors and
vectors describing all molecules of the series are aligned
into a matrix. Grid cells that are empty for all molecules
in the series analyzed are eliminated and the resulted
matrix was used for further calculations using the PLS
method. Alternatively, a CoMSA version with Kohonen
self-organizing neural network (SOM-CoMSA) was
used for comparison.
m
where obsd denotes the assayed value; pred is the pre-
dicted value of dependent variable, which ranges from
1 to m. Model with k factors, for which RMSECV
reaches a minimum, is considered as an optimal one.
We used the performance metrics that are accepted and
widely used in CoMFA analyses, that is, cross-validated
q2cv, s, RMS, and SDEP.
2.6. Iterative variable elimination
In our previous publications, we have shown that unin-
formative variable elimination (UVE)38 as well as its
modifications, that is, modified UVE (m-UVE) and iter-
ative variable elimination (IVE), can be used in 3D and
4D QSAR schemes.30,36,39 This enables the identifica-
tions of the molecular areas important for the interac-
tions with biological receptors or enzymes. In the
current calculation, we used iterative variable elimina-
tion (IVE-PLS)30 that is a modification of the UVE
2.5. PLS analysis
The obtained vectors were processed by the PLS analy-
sis with a leave-one-out cross-validation procedure. The

T. Magdziarz et al. / Bioorg. Med. Chem. 14 (2006) 1630–1643
1637
algorithm based on the analysis of the regression coeffi-
cients calculated by the PLS method. PLS allows pre-
senting the relation between the Y answer and X
predictors in a form of
15.6%, and 16%, respectively (as compared to clofibrate
with +24%, ꢀ23%, +7%, and ꢀ27% respective levels).
Moreover, compounds 11 and 12 elevated HDL-C while
diminishing TG levels.
Y ¼ Xb þ e;
where b is a vector of the regression coefficients and e the
vector of the errors.
The description of the molecular shape by spatial sectors
was originally proposed by Purcel and Testa, and fur-
ther improved by Motoc.41 In this method, a molecule
is separated into partitions of the spatial regions either
filled or unfilled by atoms or groups of atoms of certain
volumes. Using a similar idea but with improved formal-
ism, we have developed the CoMSA method and proved
that it can be a powerful tool for 3D QSAR modeling.
Hasegawa reported some further CoMSA modifica-
tions.42–44
Thus, the UVE algorithm analyzes the reliability of the
mean(b)/s(b) ratio (where s(b) means standard
deviation of b). Then, only the variables of the ÔrelativeÕ
high mean(b)/s(b) ratio are included into the final PLS
model. Instead of a single-step UVE procedure we used
here an iterative algorithm based on the abs(mean(b)/
s(b)) criterion to find the variables to be eliminated. This
procedure includes:
The statistical results of several CoMSA analyses are
summarized in Table 3. For the comparison, we per-
formed also standard CoMFA calculations. Cross-vali-
dated q2cv values obtained using these methods range
from 0.58 (CoMFA) to q2cv ¼ 0:60 (SOM-CoMSA). This
proves a correlation between descriptors and the hypo-
lipidemic activity of the asarone series.
1. Standard PLS analysis applied to analyze the matri-
ces yielded from the s-CoMSA procedure with the
leave-one-out cross-validation to estimate the perfor-
mance of the PLS model ðq2cvÞ,
2. The elimination of the matrix column of the lowest
abs(mean(b)/s(b)) value,
3. Standard PLS analysis of the new matrix without the
column eliminated in step 2,
4. Iterative repetition of steps 1–3 to maximize the LOO
q2cv parameter.
CoMFA calculations provide the best results for the H+
atomic probe, which points that electrostatic interac-
tions limit compound activity (Table 3, model 1). All ste-
ric fields provided unpredictive models (the best model
can be obtained for CH3(0) probe; q2cv ¼ 0:35).
All procedures were programmed within the MATLAB
environment (MATLAB).
All further results are reported for the s-CoMSA meth-
od. Standard method provides a relatively low q2cv value
of 0.53 (Table 3 model 2a). However, the exclusion of
three compounds (13, 29, and 32) increases model qual-
ity to q2cv ¼ 0:69 (s-CoMSA model—Table 3 model 2b).
As the experience in 3D QSAR modeling indicates that
q2cv alone cannot be a sufficient indicator of the model
quality, we performed further model validation. The
asarone series are split into two subseries. The com-
pounds were sampled into these subseries by performing
the Kennard–Stone calculations.45 Then a respective
model calculated for first subseries was used for the
activity predictions in the second, test group and the
SDEP error is calculated. The results and samplings
are reported in Table 3. This indicates reasonable pre-
dictivity. Figure 2a–d illustrates the relationships be-
tween the cross-validated (predicted) and observed
values of atherogenic index for the various models
reported in Table 3.
3. Chemistry
The compounds employed in this study 7–13 (Fig. 1)
were synthesized according to Scheme 1. The com-
pounds were prepared by a modified method of Blike
and Burckhalter.40 Thus, an appropriately substituted
amine was reacted with three equivalents of paraformal-
dehyde in absolute ethanol with the addition of equiva-
lent of concentrated HCl to form an intermediate
imminium salt, which was then reacted with the substi-
tuted trimethoxyphenone to afford the desired product
(Scheme 1). This reaction works very well for small
amines. Although the compounds were obtained in good
yields, the amount of product sharply decreases as the
size of the amine becomes bigger. Usually, the reaction
time ranges from 2 to 4 h; however; it needs to be in-
creased with the increasing amine bulkiness. The best re-
sults were obtained using ethanol/HCl solution and the
products formed can be efficiently purified by recrystal-
lization from ethanol/acetone (9:1).
Figure 3 (see also Supplementary Materials Figure 2)
illustrates the surface areas of the key importance for
the compound activity as indicated by s-CoMSA model
2a. We used that model because it includes all com-
pounds. The respective color-coding allows us to identi-
fy the influence of the respective points sampled on the
molecular surface by the combination of the electrostat-
ic potential value and a value of the b weight in the PLS
model. The variables contributing to the activity on a
level close to 0 (near 90% of the points sampled) are
omitted. Such an illustration suggests the key pharma-
cophore for the asarones investigated. Thus, an area
near the central aromatic ring substituted with an alkoxyl
4. Results and discussion
The hypolipidemic activity of compounds 7–13 was test-
ed on male rats fed with cholesterol diet against clofi-
brate as a reference drug. Compounds 7–13 exhibited
diversified hypolipidemic activity in rats. The most ac-
tive compound 8 elevated the HDL-C level by 35%
and diminished TC, LDL-C, and TG levels by 10%,

1638
T. Magdziarz et al. / Bioorg. Med. Chem. 14 (2006) 1630–1643
Table 3. The results of CoMFA and CoMSA 3D QSAR modeling
a
Compound
Model
q2cv
r2j
s
RMS
SDEP
f
f
1
CoMFAb
s-CoMSA
0.58
0.53
0.69
0.92
0.94
0.55
0.60
0.45
0.75
0.71
0.76
0.90
0.98
0.97
0.92
0.90
0.85
0.86
0.57
0.62
0.45
0.28
0.20
0.54
0.51
0.71
0.48
—
—
f
2a
2b
3a
3b
4
0.42
0.24
0.13
0.13
0.19
0.21
0.34
0.33
—
—
f
s-CoMSAc
f
s-CoMSA-IVE-PLSg
s-CoMSA-IVE-PLSc,h
s-CoMSA
—
f
—
0.80d
0.64e
0.49
5
SOM-CoMSA
s-CoMSAi
s-CoMSA-IVE-PLSi
6a
6b
0.30
a LOO cross-validated values; all compounds are included in the model.
b Calculation was performed with H(+1) as a probe atom.
c After the exclusion of three compounds: 13, 29, and 32.
d Test set (K–S subset selection): 1:5, 7, 11, 15, 23, 28, 29, 32, 36.
e Test set (K–S subset selection): 14:17, 19, 22, 24, 25, 29, 32, 35, 37, 38.
f Not tested.
g After IVE-PLS starting from model 2a.
h After IVE-PLS starting from model 2b.
i Test set 33:40.
j Fitted statistics.
a
b
Observed activity
Observed activity
c
d
Observed activity
Observed activity
Figure 2. The relationship between the cross-validated (predicted) and observed values of atherogenic index for the various models reported in Table
3; (a) model 2a, (b) model 3a, (c) model 2b and (d) model 3b.

T. Magdziarz et al. / Bioorg. Med. Chem. 14 (2006) 1630–1643
1639
Figure 3. Molecular surface areas of the highest electrostatic contribution to the compound activity according to model 2b, sorted by the decreasing
activity of some illustrative molecules (see also Supplementary Materials Figure 2). Blue indicates positive, and yellow and red negative influence on
the activity. Details in text.
functionality provides a negative contribution, i.e.,
increases the activity (blue-colored sections), while a
negatively charged carbonyl oxygen in the side chain
generally decides a lower activity, clearly decreasing
the activity as illustrated by the yellow and red molecu-
lar surface areas. This rule can be proved by the exam-
ination of the structures given in Table 2. It is worth
noticing that if a hydroxyl function replaces a carbonyl
oxygen, for example, compound 11, we do not observe
any activity decrease similar to that effected by a carbon-
yl-group, for example, compound 10. Thus, our analysis
reveals the molecular basis for the hypolipidemic activi-
ty of asarones. In Figure 4, we show the results obtained
after application of additional filters. This allowed us to
obtain highly predictive 3D QSARs (model 3a
q2cv ¼ 0:92, s = 0.28, RMS = 0.13 and 3b qc2v ¼ 0:94,
s = 0.20, RMS = 0.13, respectively); however it does not
give so clear molecular illustration. Thus, in Figure 4
we give examples of such plots for compound, of high,
medium, and low activity. In Figure 4a, the sectors con-
tributing to the activity are divided into two subsets.
Green sector increases the activity; orange—decreases.
The plots are now more specific for individual com-
pounds, for example, the contribution of the alliloxyl side
chain of compound 22 is pronounced more clearly than
the areas closer to the aromatic unit. However, the key
pharmacophore near the central aromatic unit and the
disadvantageous influence of carbonyl oxygen are pre-
served. Moreover, in Figure 4b we coded by colors the
different influences of the electro-negative and -positive
surface sectors. This reveals the disadvantageous contri-
bution of the electronegative carbonyl oxygen (dark blue)
and electropositive metoxyl (red). The plots shown in
Figure 4c allow for the estimation of the relative contri-
bution of the sectors indicated. It is worth noticing that
the contour interaction plots displayed in Figures 3 and
4 are completely different from that which resulted from
CoMFA modeling. Unlike in CoMFA, each plot is char-
acteristic for individual compound. This enables a clear
differentiation of the low and high activity attributes. In
particular, this provides also a novel insight into the
molecular basis for the asarone activity.
Since overfitting is an important problem in QSAR
modeling, we further validated the quality of the CoM-
SA models by the so-called model randomization. We
investigated this by generating 100 random permuta-
tions of the activity column. We used model 6a as an

1640
T. Magdziarz et al. / Bioorg. Med. Chem. 14 (2006) 1630–1643
a
b
c
Figure 4. Molecular surface sectors indicated by s-CoMSA-IVE-PLS (model 3b from Table 2) for compounds of the highest, medium, and lowest
activity; (a) color codes indicate a sign of activity) change: orange—decreases the activity, green—increases the activity, (b) combination of the
electrostatic potential sign and the sign of the b weight in the model: +/+ (red, decreases the activity), ꢀ/+ (cyan—increases the activity), +/ꢀ
(magenta, increases the activity), ꢀ/ꢀ (dark blue, decreases the activity), and (c) color codes indicate a relative value of activity change: warm color
(red)—decreases the activity, cool colors (blue)—increases the activity, respectively.
initial form to perform IVE in such random models.
Thus, each model with random activity column was pro-
cessed by the IVE procedure and q2cv was measured. Fig-
ure 5 illustrates the histogram of calculated q2cv. Red
asterisk points to the qc2v value of the unmodified model,
that is, model 6b. Clearly, the majority of models have a
relatively low q2cv, significantly lower than the value cal-
culated for a model with actual compound activity. This
validates the statistical significance of the model 6b
where the activity is properly arranged according to
the actual activity reported in Table 2.
5. Conclusions
Compounds 7–13 exhibited diversified hypolipidemic
activity in rats. The most active compound 8 elevated
HDL-C level by 35% and diminished TC, LDL-C, and
TG levels by 10%, 15.6%, and 16%, respectively (as com-
pared to clofibrate with +24%, ꢀ23%, +7%, and ꢀ27%
respective levels). Although some pseudo- or mini-recep-
tor models were reported for asarones previously, this did
not reveal any clear molecular basis for controlling the
activity of these compounds. The present analysis
employing the CoMSA allowed us to reveal a correlation
between the activity of these compounds and the electro-
Figure 5. The histogram of q2cv computed for 100 randomized CoMSA
models (random activity arrangement). An asterisk indicates a qc2v
value for the proper activity arrangement. Details in text.
static potential at the molecular surface. The grid formal-
ism (s-CoMSA) allowed us also to indicate
6. Experimental
a
pharmacophore that is of key importance for the com-
poundsÕ activity. The CoMSA formalism coupled with
the IVE (CoMSA–IVE) allowed us also to obtain highly
predictive models.
6.1. Chemical methods and reagents
Melting points were determined with a Ko¨ffler appara-
tus of the Bo¨etius type and are uncorrected. H and
1

T. Magdziarz et al. / Bioorg. Med. Chem. 14 (2006) 1630–1643
1641
13C NMR spectra were recorded on a Bruker AC 200F
spectrometer using CDCl3 solution with TMS as inter-
nal standard (chemical shifts in d ppm). IR spectra were
recorded on a Nicolet Magna 550 FTIR Spectrometer in
chloroform solutions. The UV spectra were collected on
a Hewlett-Packard UV–vis Diode Array Spectropho-
tometer 8452A. Mass spectra were obtained at 70 eV
with an AMD-604 spectrometer. The reaction products
were isolated by column chromatography or flash chro-
matography performed on a silica gel 70–230 mesh
ASTM (Merck, Darmstad, Germany). Thin-layer chro-
matograms were developed on aluminum TLC sheets
precoated with silica gel F254 (Merck, Darmstadt, Ger-
many). The spots were visualized with 50% sulfuric acid
after heating. All the solvents were dried and freshly dis-
tilled prior to use. Starting materials and reagents were
purchased from Aldrich Chemical Co. (Steinheim,
Germany).
2H); 13C NMR (CDCl3, 50 MHz) d: 194 (C), 152 (C);
142 (C); 130 (C); 105 (CH); 52 (CH2); 60 (CH3); 56
(CH3); 45 (CH2); 33 (CH2); 26 (CH2); 24 (CH2); 22
(CH2); MS m/z 112 (M+, 100).
6.6. 3-Piperidin-1-yl-1-(3,4,5-trimethoxyphenyl)propan-1-
one (10)
Mp 203–204 ꢁC; IR (CHCl3) m 2700–2400; 1697, 1200,
1131 cmꢀ1 1H NMR (CDCl3, 200 MHz) d: 11.4 (s,
,
1H); 7.2 (s, 2H); 3.9 (s, 6H); 3.8 (s, 3H); 3.7 (m, 2H)
3.6 (m, 4H); 2.6 (m, 2H); 2.1 (m, 2H); 2.1 (m, 2H); 1.7
(t, 2H); 1.5 (m, 2H); 13C NMR (CDCl3, 50 MHz) d:
194 (C); 152 (C); 142 (C); 130 (C); 105 (CH); 53
(CH2); 60 (CH3); 56 (CH3); 32 (CH2); 22 (CH2); 21
(CH2); MS m/z 112 (M+, 100).
6.7. Synthesis of 3-morpholin-4-yl-1-(3,4,5-trimethoxy-
phenyl)propan-1-ol (11) 3-piperidin-1-yl-1-(3,4,5-trimeth-
oxyphenyl)propan-1-ol (12)
6.2. General procedure for 3-amin-4-yl-1-(3,4,5-trimeth-
oxyphenyl)propan-1-one (7–10)
3.5 g (0.010 mol) of hydrochloride 3-morpholin-4-yl-
1-(3,4,5-trimethoxyphenyl)propan-1-one or 3-piperidin-1-
yl-1-(3,4,5-trimethoxyphenyl)propan-1-one was dissolved
in 2 ml water. After adding (20 mL) 6 M NaOH, the
mixture was extracted three times with ether (3·
30 mL). The ether extract was dried over MgSO4 and
evaporated in vacuo. The crude product was dissolved
in 30 mL methanol. The reaction mixture was dropped
to a cold mixture of 440 mg (0.010 mol) NaBH4 dissolved
in 20 mL methanol and 20 mL water. The reaction
mixture was stirred for 2 h at room temperature and
15 min at 45 ꢁC. The organic solvent was removed and
15 ml of 6 N NaOH was added. The aqueous solution
was extracted with chloroform (3· 30 mL). The organic
extract were combined and washed with water and dried
over MgSO4. The mixture was filtered off and concentrat-
ed under a reduced pressure. After evaporation of solvent,
the residue was purified by silica gel column chromatogra-
phy (elution with hexane/ether; 1:1; v/v) and crystallized
from an ethanol/acetone 9:1 (v/v) .W = 63.5%.
The mixture of 0.046 mol of hydrochloride cycloamine,
2.0 g (0.066 mol) of paraformaldehyde, 2 g 3,4,5-tri-
methoxyacetophenone (0.040 mol), 2.4 mL concd HCl,
and 20 mL absolute ethanol in a 100 mL three-necked
flask was refluxed for 2 h. Then 1.2 g (0.040 mol) para-
formaldehyde was added. The reaction mixture was re-
fluxed for 5 h. To a hot mixture 50 ml acetone was
added and refluxed for 15 min. After cooling down to
room temperature, the reaction mixture was evaporated
under reduced pressure to dryness. The product was
crystallized from ethanol/acetone 9:1 (v/v).
6.3. 3-Morpholin-4-yl-1-(3,4,5-trimethoxyphenyl)propan-
1-one (7)
Mp 199–201 ꢁC; IR (CHCl3, cmꢀ1) m: 2700–2400; 1697;
1
1200; 1118; 1131; H NMR (CDCl3, 200 MHz) d: 11.3
(s, 1H); 7.2 (s, 2H); 4.2 (m, 2H); 4.1 (m, 2H); 3.9 (s,
9H); 3.7 (m, 2H); 3.4 (m, 4H); 2.9 (m, 2H); 13C NMR
(CDCl3, 50 MHz) d: 194 (C); 153 (C); 143 (C); 130
(C); 105 (CH); 63 (CH2); 60 (CH3); 56 (CH3); 52
(CH2); 32 (CH2); MS m/z 100 (M+, 100).
6.8. 3-Piperidin-1-yl-1-(3,4,5-trimethoxyphenyl)propan-1-
ol
6.4. 3-(4-Methylpiperidin-1-yl)-1-(3,4,5-trimethoxyphe-
nyl)propan-1-one (8)
Mp 65–68 ꢁC; IR (CHCl3, cmꢀ1) m: 3181, 1227, 1132,
1
1112; H NMR (CDCl3, 200 Hz) d: 6.6 (s, 2H); 4.9 (t,
1H); 3.9 (s, 6H); 3.8 (s, 6H); 2.7 (m, 4H); 2,4 (m, 2H);
1.9 (m, 2H); 1.6 (m, 4H); 1.5 (m, 2H); 13C NMR
(CDCl3, 50 Hz) d: 153 (C); 139 (C); 136 (C); 106 (CH);
80 (CH); 54 (CH2); 52 (CH2); 33 (CH2); 22 (CH2); 60
(CH3); 57 (CH3); MS m/z 98 (M+, 100).
Mp 153–155 ꢁC; IR (CHCl3, cmꢀ1) m: 2700–2400; 1684,
1200, 1133; 1H NMR (CDCl3, 200 MHz) d: 12.2 (s, 1H);
7.3 (s, 2H); 3.9 (s, 6H); 3.8 (s, 3H); 3.7 (t, 2H); 3.5 (m,
4H); 2.7 (m, 2H); 1.8 (m, 5H); 1.1 (d, 2H); 13C NMR
(CDCl3, 50 MHz) d: 194 (C); 152 (C); 142 (C); 130
(C); 102 (CH); 66 (CH2); 60 (CH3); 56 (CH3); 52
(CH2); 32 (CH2); MS m/z 112 (M+, 100).
6.9. 3-Morpholin-4-yl-1-(3,4,5-trimethoxyphenyl)propan-
1-ol
6.5. 3-Azepan-1-yl-1-(3,4,5-trimethoxyphenyl)propan-1-
one (9)
Mp 86–88 ꢁC; IR (CHCl3) m: 3190, 1227, 1132, 1119,
1127 cmꢀ1 1H NMR (CDCl3, 200 MHz) d: 6.6 (s,
;
2H); 4.9 (t, 1H); 3.9 (s, 9H); 3.7 (t, 4H); 2.7 (m, 4H);
2.5 (m, 4H); 1.9 (m, 2H); 13C NMR (CDCl3, 50 MHz)
d: 154 (C); 140 (C); 138 (C); 102 (CH); 75 (CH); 68
(CH2); 61 (CH3); 58 (CH2); 57 (CH3); 53 (CH2);
33(CH2); MS m/z 100 (M+, 100).
Mp 171–173 ꢁC; IR (CHCl3, cmꢀ1) m: 2700–2400; 1680,
1
1202, 1131, 1131; H NMR (CDCl3, 200 MHz) d: 12.2
(s, 1H); 7.2 (s, 2H); 3.9 (s, 6H); 3.8 (s, 3H); 3.7 (m,
2H); 3.6 (m, 4H); 2.9 (m, 2H); 1.8 (m, 4H); 1.6 (m,

1642
T. Magdziarz et al. / Bioorg. Med. Chem. 14 (2006) 1630–1643
13. Chamorro, G.; Garduno, L.; Sanchez, A.; Labarrios, F.;
Salazar, M.; Martinez, E.; Diaz, F.; Tamariz, J. Drug Dev.
Res. 1998, 43, 105.
6.10. Synthesis of 4-[(2E)-3-(3,4,5-trimethoxyphe-
nyl)prop-2-enyl]morpholine (13)
14. Labarrios, F.; Garduno, L.; Vidal, M.; Garcia, R.;
Salazar, M.; Martinez, E.; Diaz, F.; Chamorro, G.;
Tamariz, J. J. Pharm. Pharmacol. 1999, 51, 1.
15. Dandiya, P. C.; Sharma, J. D. Ind. J. Med. Res. 1962, 50,
46.
A solution of crude alcohol, 2.7 g (0.009 mol) 3-mor-
pholin-4-yl-1-(3,4,5-trimethoxyphenyl)propan-1-ol
50 ml of dry toluene was treated with 3 g 0.02 mol of an-
hyd CuSO4. The reaction mixture was heated for 2 h at
reflux. After the evaporation of the solvent, the residue
was crystallized from ethanol/acetone 9:1 (v/v).
in
16. Dandiya, P. C.; Menon, M. K. Br. J. Pharmacol. 1963, 20,
436.
17. Belova, L.; Alibekov, S.; Baginskaya, A.; Sokolov, S.;
Pokrovskaya, G.; Stikhin, V.; Trumpe, T.; Gorodnyuk, T.
Farmak. Toksikol. 1985, 48, 17.
18. Morales, R.; Madrigal, B.; Mercader, M.; Cassani, M.;
Gonzalez, G.; Chamorro, G.; Salazar, M. Mutat. Res.
1992, 279, 269.
19. Salazar, M.; Salazar, S.; Ulloa, V.; Mendoza, T.; Pages,
N.; Chamorro, G. J. Toxicol. Clin. Exp. 1992, 12, 149.
20. Diaz, F.; Contreras, L.; Flores, R.; Tamariz, J.; Labarrios,
F.; Chamorro, G.; Munoz, H. Org. Prep. Proc. Int. 1992,
24, 78.
Mp 85–86 ꢁC; IR (CHCl3, cmꢀ1) m: 3181, 1227, 1132,
1
1112; H NMR (CDCl3, 200 Hz) d: 6.86 (s, 2H); 6.1
(d, 1H, J = 15.8 Hz); 5.9 (dt, 1H, J = 6.6 Hz,
J = 15.75 Hz); 3,85 (s, 6H); 3,8 (s, 6H); 3,38 (m, 4H);
2,2 (dt, 2H, J = 6.65 Hz, J = 7,04 Hz); 2.1 (m, 2H); 13C
NMR (CDCl3, 50 MHz) d: 154 (C); 138 (C); 133
(CH); 128 (CH); 120 (C), 104 (CH); 69 (CH2); 66
(CH2); 60 (CH3); 56 (CH3); 53 (CH2).
21. Filipek, S.; Łozowicka, B. Acta Pol. Pharm. 2000, 57, 106.
22. Cruz, M.; Salazar, M.; Garciafigueroa, Y.; Hernandez, D.;
Diaz, F.; Chamorro, G.; Tamariz, J. Drug Dev. Res. 2003,
60, 186.
23. Cruz, A.; Garduno, L.; Salazar, M.; Martinez, E.;
Jimenez-Vazquez, H.; Diaz, F.; Chamorro, G.; Tamariz,
J. Arzneim.-Forsch./Drug Res. 2001, 51, 535.
24. Poplawski, J.; Lozowicka, B.; Dubis, A.; Lachowska, B.;
Witkowski, S.; Siluk, D.; Petrusewicz, J.; Kaliszan, R.;
Cybulski, J.; Chilmonczyk, Z.; Strzalkowska, M. J. Med.
Chem. 2000, 43, 3671.
Acknowledgments
The authors thank Professor Johann Gasteiger of the
University of Erlangen-Nurnberg, Germany, both for
¨
facilitating access to the CORINA, PETRA, SUR-
FACE, and KMAP programs. The financial support
of the KBN Warsaw under Grants No. KBN P05F
01617, 4P05F019 19, KBN 4T09A 088 25, and PBZ
040 P04/08 is gratefully acknowledged. R.G. thanks
the Foundation for Polish Science for an individual
grant.
25. Lozowicka, B.; Filipek, S.; Dubis, A. Eur. J. Med. Chem.
(submitted).
26. Lozowicka, B.; Poplawski, J.; Dubis, A.; Chilmonczyk, Z.;
Cybulski, J.; Kita, K.; Kobes, S.; Filipek, S. Drug Dev.
Res. (submitted).
27. Chilmonczyk, Z.; Siluk, D.; Kaliszan, R.; Lozowicka, B.;
Poplawski, J.; Filipek, S. Pure Appl. Chem. 2001, 73, 1445.
28. Polanski, J.; Walczak, B. Comp. Chem. 2000, 24, 615.
29. Polanski, J.; Gieleciak, R.; Bak, A. J. Chem. Inf. Comput.
Sci. 2002, 42, 184.
Supplementary data
Supplementary data associated with this article can be
found, in the online version, at doi:10.1016/
j.bmc.2005.10.014.
30. Polanski, J.; Gieleciak, R. J. Chem. Inf. Comput. Sci. 2003,
43, 656.
31. Polanski, J.; Gieleciak, R.; Wyszomirski, M. J. Chem. Inf.
Comput. Sci. 2003, 43, 1754.
References and notes
32. Polanski, J.; Gieleciak, R.; Wyszomirski, M. Dyes Pig-
ments 2004, 62, 63.
1. Anderson, K. M.; Wilson, P. W. F.; Odell, P. M.; Kannel,
W. B. Circulation 1991, 83, 356.
33. Polanski, J.; Gieleciak, R. Mol. Diversity 2003, 7, 45.
34. Polanski, J.; Gieleciak, R.; Bak, A. Comb. Chem. High
Throughput Screen. 2004, 7, 793.
35. Polanski, J.; Bak, A.; Gieleciak, R.; Magdziarz, T.
Molecules 2004, 9, 1148.
2. Van Gaal, L. F.; Hang, A.; Steijaert, Mm.; De Leeuw, Ih.
Int. J. Obes. 1995, 19, S21.
3. McCully, K. S. Nat. Med. 1996, 2, 386.
4. Brown, Bg.; Zhao, X.; Sacco, De.; Albers, J. J. Circulation
1993, 87, 1781.
36. Polanski, J.; Gieleciak, R.; Magdziarz, T.; Bak, A. J.
Chem. Inf. Comput. Sci. 2004, 44, 1423.
5. Badimon, J. J.; Fuster, V.; Badimon, L. Circulation 1992,
86, II-86–III-94.
37. MATLAB 5.0 program. Available from: The Mathworks
Inc., Natick, MA, USA, http://www.mathworks.com.
38. Centner, V.; Massart, D. L.; de Noord, O. E.; de Jong, S.;
Vandeginste, B. M. V.; Sterna, C. Anal. Chim. Acta 1996,
330, 1.
6. Executive Summary of the Third Report of the National
Cholesterol Education Program (NCEP) Expert Panel on
Detection, Evaluation, and Treatment of High Blood
Cholesterol in Adults (Adult Treatment Panel III) JAMA
2001, 285, 2486.
39. Polanski,J.;Bak,A.J.Chem.Inf.Comput. Sci. 2003, 43,208.
40. Blicke, F.; Bruckhalter, J. J. Am. Chem. Soc. 1942, 64,
451.
41. Testa, B.; Purcell, W. P. Eur. J. Med. Chem. 1978, 13, 509;
Motoc, I. Molecular Shape Descriptors. In Steric Effects
in Drug Design; Charton, M., Motoc, I., Eds.; Akademie:
Berlin, 1983; pp 93–105.
7. Chilmonczyk, Z.; Siluk, D.; Kaliszan, R. Exp. Opin. Ther.
Pat. 2001, 11, 1301.
8. Levine, G.; Keaney, J.; Vita, J. N. Engl. J. Med. 1995, 332,
512.
9. Szapary, P. O.; Rader, D. J. Am. Heart. J. 2004, 148, 211.
10. Eghdamian, B.; Ghose, K. Drugs Today 1988, 35, 79.
11. Farmer, J. A.; Gotto, A., Jr. Adv. Pharmacol. 1966, 35, 79.
42. Hasegawa, K.; Matsuoka, S.; Arakawa, M.; Funatsu, K.
Comput. Chem. 2002, 26, 583.
´ ´
12. Enriquez, R.; Chavez, M.; Jauregui, F. Phytochemistry
1980, 19, 2024.

T. Magdziarz et al. / Bioorg. Med. Chem. 14 (2006) 1630–1643
1643
43. Hasegawa, K.; Morikami, K.; Shiratori, Y.; Ohtsuka, T.;
Aoki, Y.; Shimma, N. Chemometr. Intell. Lab. Syst. 2003,
69, 51.
44. Hasegawa, K.; Matsuoka, S.; Arakawa, M.; Funatsu, K.
Comput. Biol. Chem. 2003, 27, 381.
45. Kennard, R. W.; Stone, L. A. Technometrics 1969, 11, 137.