Synthesis, pharmacological and biophysical characterization, and membrane-interaction QSAR analysis of cationic amphiphilic model compounds

Article abstract of DOI:10.1021/jm980694a

Cationic amphiphilic drugs have a propensity to interact with biological interphases. This study was designed to gain more insight into the molecular properties of catamphiphilic drugs which govern this type of interaction. A series of phenylpropylamine model compounds were synthesized in which modifications were incorporated at the aromatic part of the molecule. The replacement of ⁴⁵Ca²⁺ from phosphatidylserine monolayers served to monitor drug binding to the phospholipid. The influence on the phase- transition temperature of liposomes of dipalmitoylphosphatidic acid was measured to assess the perturbing action of the drugs on the structural organization of phospholipid assemblies. The antiarrhythmic activity of the compounds was determined in Langendorff preparations of guinea pig hearts to assess the membrane-stabilizing action. Quantitative structure-activity relationship (QSAR) models for these endpoints were developed using both intra- and intermolecular QSAR descriptors. Intermolecular membrane- interaction descriptors were derived from molecular dynamics simulations of the compounds in a model phospholipid monolayer. QSAR models were derived for all endpoints using partial least-squares regression (PLS) and a genetic algorithm tool, the genetic function approximation (GFA). Membrane- interaction descriptors appear to be of a particular importance in explaining the influence of the compounds on the phase-transition temperature of DPPA liposomes, while the other endpoints can be adequately modeled by intramolecular descriptors. The calcium-displacing activity at phosphatidylserine monolayers is governed by the electrostatic properties of the compounds. Measures of lipophilicity and molecular size are of particular importance for antiarrhythmic activity. Possible improvements to both the molecular modeling and the applied computational protocol of membrane-solute systems are identified and discussed.

Full text of DOI:10.1021/jm980694a

3874
J . Med. Chem. 1999, 42, 3874-3888
Syn th esis, P h a r m a cologica l a n d Biop h ysica l Ch a r a cter iza tion , a n d
Mem br a n e-In ter a ction QSAR An a lysis of Ca tion ic Am p h ip h ilic Mod el
Com p ou n d s
Christian D. P. Klein,§ Martin Klingmu¨ller,‡ Christiane Schellinski,† Silke Landmann,† Stefanie Hauschild,†
Dieter Heber,‡ Klaus Mohr,† and A. J . Hopfinger*,§
Laboratory of Molecular Modeling and Design, M/ C 781, College of Pharmacy, University of Illinois at Chicago,
833 South Wood Street, Chicago, Illinois 60612-7231, Department of Pharmaceutical Chemistry, Pharmaceutical Institute,
University of Kiel, Gutenbergstrasse 76, D-24118 Kiel, Germany, and Department of Pharmacology and Toxicology,
Pharmaceutical Institute, University of Bonn, An der Immenburg 4, D-53121 Bonn, Germany
Received December 10, 1998
Cationic amphiphilic drugs have a propensity to interact with biological interphases. This study
was designed to gain more insight into the molecular properties of catamphiphilic drugs which
govern this type of interaction. A series of phenylpropylamine model compounds were
synthesized in which modifications were incorporated at the aromatic part of the molecule.
The replacement of 45Ca2+ from phosphatidylserine monolayers served to monitor drug binding
to the phospholipid. The influence on the phase-transition temperature of liposomes of
dipalmitoylphosphatidic acid was measured to assess the perturbing action of the drugs on
the structural organization of phospholipid assemblies. The antiarrhythmic activity of the
compounds was determined in Langendorff preparations of guinea pig hearts to assess the
membrane-stabilizing action. Quantitative structure-activity relationship (QSAR) models for
these endpoints were developed using both intra- and intermolecular QSAR descriptors.
Intermolecular membrane-interaction descriptors were derived from molecular dynamics
simulations of the compounds in a model phospholipid monolayer. QSAR models were derived
for all endpoints using partial least-squares regression (PLS) and a genetic algorithm tool, the
genetic function approximation (GFA). Membrane-interaction descriptors appear to be of a
particular importance in explaining the influence of the compounds on the phase-transition
temperature of DPPA liposomes, while the other endpoints can be adequately modeled by
intramolecular descriptors. The calcium-displacing activity at phosphatidylserine monolayers
is governed by the electrostatic properties of the compounds. Measures of lipophilicity and
molecular size are of particular importance for antiarrhythmic activity. Possible improvements
to both the molecular modeling and the applied computational protocol of membrane-solute
systems are identified and discussed.
In tr od u ction
group, and the aromatic ring system is directed toward
the hydrophobic interior of the phospholipid layer. The
cationic amphiphilic nature may have impact on drug
pharmacokinetics and pharmacodynamics.2
Drugs from a wide variety of pharmacological groups
are cationic amphiphilic in nature, such as antiarrhyth-
mics, local anesthetics, antimalarials, â-blockers, and
tricyclic antidepressants.1 The drugs often contain, in
close proximity, a lipophilic aromatic ring system and
a side chain with a nitrogen protonized at physiological
pH. The drugs are prone to interact with membrane
phospholipids: the cationic nitrogen is attracted to the
negatively charged phosphate of the phospholipid head-
Many efforts have been made to understand the
molecular principles of drug-phospholipid interaction
on the molecular level. However, a drawback of applying
selected therapeutics as test compounds is their het-
erogeneity in molecular structure and physicochemical
properties. Only very few structure-activity studies for
explicit membrane-interaction endpoints have been
reported to date.3 A possible reason for this situation
may be the lack of appropriate training datasets. The
goal of this study was to synthesize, test, and character-
ize a set of cationic amphiphilic model compounds (Table
1) and analyze their membrane-interaction behavior in
terms of quantitative structure-activity relationship
(QSAR) models. In this series of phenylpropylamine
derivatives the structure of the aromatic part is the
variable parameter while the side chain is kept con-
stant. With a propyl linker between the aromatic ring
and the nitrogen, its alkaline character is relatively
independent of the aromatic variations.
* Corresponding author. Phone: (312) 996-4816. Fax: (312) 413-
3479. E-mail: hopfingr@uic.edu.
§ University of Illinois at Chicago.
‡ University of Kiel.
† University of Bonn.
Abbreviations: CPSA, charged partial surface area; DMPC,
dimyristoylphosphatidylcholine; DPPA, dipalmitoylphosphatidic acid;
DPPC, dipalmitoylphosphatidylcholine; DPPG, dipalmitoylphosphati-
dylglycerol; DSC, differential scanning calorimetry; ESR, electron spin
resonance; GFA, genetic function approximation; HPLC, high-perfor-
mance liquid chromatography; LOF, lack of fit; MDR, multiple drug
resistance; MDS, molecular dynamics simulation; NMR, nuclear
magnetic resonance; PLS, partial least-squares (regression); QSAR,
quantitative structure-activity relationship; QSPR, quantitative struc-
ture-property relationship; TES, N-tris(hydroxymethyl)methyl-2-ami-
noethanesulfonic acid; Tt, phase-transition temperature.
10.1021/jm980694a CCC: $18.00 © 1999 American Chemical Society
Published on Web 08/26/1999

Membrane-Interaction QSAR Analysis
J ournal of Medicinal Chemistry, 1999, Vol. 42, No. 19 3875
Ta ble 1. Synthesis of the Compounds
compd
X
Y
H
Z
method
yield (%)
empirical formulaa
11H17Cl2N
mp (°C)
155
1
Cl
H
A
B
A
A
B
A
B
90
95
86
74
93
74
29
C
2
3
Br
HO
H
H
H
H
C11H17BrClN
C11H18ClNO
139
142-144
4
C10H18Cl2N2O8
160b
5
phenyl
H
H
A
B
B
B
B
B
B
B
B
B
B
B
B
90
82
55
86
41
11d
50
85
83
42
92
68
87
C17H22ClN
191
6
7
8
H
Me
Et
n-Pr
i-Pr
t-Bu
MeO
4-Cl-phenyl
4-MeO-phenyl
4-EtO-phenyl
H
H
H
H
H
H
H
H
H
H
H
H
H
H
H
H
H
H
H
H
C11H18ClN
C12H20ClN
144
174
178
198
196
183
158
197
231
239
167
C
13H22ClN
C14H24ClN
14H24ClN
C15H26ClN
12H20ClNO
C17H21Cl2N
18H24ClNO
9
10
11
12
13
14
15
16
C
C
C
C19H26ClNO
C15H20ClN
17
18
19
20
MeO
H
phenyl
Cl
H
H
H
B
C
C
B
82
90
89
86
C
12H19Cl2NO
C17H22ClN
23H26ClN
189
165
167-168
167-169
phenyl
phenyl
C
C17H22ClN
21
22
NO2
H
H
B
B
28
88
C11H17ClN2O2
C12H20NJ
154
175c
23
24
25
4-HO-phenyl
HO
H
Cl
H
H
B
B
B
95
85
90
C17H22ClNO
C11H17Cl2NO
C19H26NJ O
220-221
189
165c
26
NH2
H
H
B
80
C11H20Cl2N2
250 dec
a
Analyses for C, H, and N are within (0.4% of the theoretical value. All compounds were isolated as hydrochlorides except 4, 22, and
25. bAnion ) diperchlorate. c Anion ) iodide. With regard to 4-n-propylacetophenone (see Experimental Section).
d
Meth od s
to assess the membrane-stabilizing potency of the test
compounds in an excitable tissue.1
Gen er a l Con sid er a tion s. The test compounds were
studied in three model systems. The inhibition of 45Ca2+
adsorption to phosphatidylserine monolayers indicates
the interaction of the compounds with the phospholipid
surface charge and probably reflects drug binding
affinity to the phospholipid.1,4,5 The drug-induced de-
pression of the transition temperature of dipalmi-
toylphosphatidic acid (DPPA) liposomes, measured by
differential scanning calorimetry, indicates structural
perturbation of the phospholipid bilayer. DPPA was
chosen because catamphiphilic drugs induce, in this
phospholipid, in addition to the control transition, a
second, clearly separate signal, the temperature of
which is independent of the amount of drug added.6 The
extent to which the transition temperature is reduced
has been suggested to reflect the depth of drug penetra-
tion into the hydrophobic core of the phospholipid
bilayer.5 The drug-induced elevation of the threshold of
alternating current to elicit arrhythmia in isolated,
spontaneously beating guinea pig hearts was measured
The trial descriptors of a QSAR analysis are normally
computed from the chemical structures of the molecules
composing the training set from which the QSAR is to
be constructed. However, there is no reason that the
estimation of QSAR descriptors need be restricted to the
intramolecular properties of a molecule. For example,
if the geometry of the receptor to which the training set
molecules bind is available, then intermolecular ligand-
receptor properties can be computed and used as trial
QSAR descriptors along with the intramolecular de-
scriptors. The information inherent to the ligand-recep-
tor descriptors should allow the construction of a better
QSAR model than if this information is neglected or not
available. In the study reported in this paper, a mem-
brane model of a phospholipid monolayer has been as-
sumed to be a general receptor for the set of cationic
amphiphilic compounds studied as antiarrhythmic drug
candidates.
The impetus to explore intermolecular membrane-
interaction descriptors as trial QSAR descriptors is

3876 J ournal of Medicinal Chemistry, 1999, Vol. 42, No. 19
Klein et al.
Ta ble 2. Physicochemical Properties, Antiarrhythmic Activity,
and Biophysical Properties for the Dataset
are chiral, and DSC measurements were performed with
the racemate. For that reason, and in order to allow for
an external validation of the QSAR models, these two
analogues were excluded from the training data set.
Compound 18 has a significantly different influence on
the phase behavior of DPPA liposomes than the other
analogues. Compound 18 raises the phase-transition
temperature. Therefore, it was judged reasonable to
exclude this compound from the training data set
constructed for this investigation.
d
e
f
log AC50
Tt
log IC50
b
compd
log Pa
pKa
10.59
10.62
x
(µM)
(K)
(µM)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
2.21
2.37
0.35
0.57
2.99
1.50
2.00
2.45
2.89
2.75
2.96
1.48
3.45
2.95
3.30
2.62
1.92
2.76
3.36
2.59
1.50
0.44
2.10
0.97
2.17
0.12
4.31
x
3.29
vl
302
x
314
x
285
308
303
300
295
299
297
x
x
296
x
297
x
308
294
309
307
307
349
294
x
3.85
x
2.72
3.60
5.40
3.00
x
x
x
x
5.40
x
x
5.40
x
4.35
x
4.85
x
10.68, 7.15
9.88
4.80
3.55
4.02
3.94
4.85
4.22
4.52
3.62
5.70
4.96
4.80
4.52
4.44
4.37
x
10.96
10.84
10.63
10.54
10.49
10.03
10.91
9.96
9.85
9.82
10.27
10.76
10.00
9.85
The given dataset appears to be particularly suitable
for performing membrane-interaction QSAR analysis, as
we term this modeling method, because there is minimal
uncertainty about the orientation of these compounds
within the phospholipid membrane. The cationic amino
group is presumably anchored near the phospholipid
headgroup region, or the membrane-water interphase,
while the aromatic hydrocarbon part is located within
the core region of the membrane. Experimental evidence
for this arrangement was found by Kuroda and Fuji-
wara,12 who studied the interactions of cationic am-
phiphilic drugs, such as lidocaine, with phosphatidyl-
choline liposomes using H1-NOE NMR spectroscopy.
9.93
10.74
x
9.32
x
4.13
vl
vl
4.43
3.82
vl
x
4.15
3.40
4.40
x
x
x
The phospholipid dipalmitoylphosphatidylcholine
(DPPC) was used for all MDSs in this study. This choice
of phospholipid might, at first glance, surprise, since the
DSC measurements were made for DPPA liposomes.
However, several considerations led to the belief that
DPPC membrane models are, nevertheless, reasonable
and lead to valid results. First, most of the published
membrane MDSs have been done with either DPPC or,
as in our group, dimyristoylphosphatidylcholine (DMPC).
Thus far, no MDS study of phosphatidic acid (PA)
membranes has come to our attention. The lack of use
of PA in membrane modeling may be due to the
difficulties in assigning proper ionization states and
counterions to the anionic PA headgroups. Second,
Hanpft and Mohr,6 when examining the influence of
several cationic amphiphilic compounds on the phase-
transition behavior of different phospholipids, found
that the transition temperatures of the drug-containing
phospholipid domains were practically independent of
the phospholipid headgroup in DPPA, DPPC, and di-
palmitoylphosphatidylglycerol (DPPG). It was concluded
that the intercalation of cationic amphiphilic substances
quenches the specific influence of the phospholipid’s
headgroup on the phase-transition behavior. To check
the validity of this approach, measurements were made
for selected test compounds in DPPC liposomes. The
measured drug-induced transition temperatures favor-
ably corresponded with the predicted values.
x
10.87, 7.34
3.00
x
a
b
Determined by RP-HPLC. Measured by aqueous titration of
hydrochlorides. c Calculated, MOPAC 6.0.26
Concentration to
d
elevate the threshold of alternating current to induce arrhythmia
in isolated guinea pig hearts by 50%. e Drug-induced phase-
transition temperature of DPPA liposomes/drug mixture. f Drug
concentration at which Ca2+ binding to phosphatidylserine mono-
layers is reduced to 50% of control value. x, not measured; vl, very
low activity, could not be measured accurately. Underlinded values
were included in the QSAR/QSPR analyses of the corresponding
property endpoint.
because in an earlier study using intramolecular de-
scriptors,7 it was not possible to build reliable QSAR
and QSPR models for the drug effects on 45Ca2+ binding
and the phase-transition temperature. This failure can
be attributed to the fact that membranes do not offer
spatially well-defined binding sites, and a classical,
intramolecular QSAR approach may be inappropriate.
Thus, we developed an intermolecular modeling ap-
proach, based upon molecular dynamics simulations
(MDS), of the compounds in a phospholipid environ-
ment, the results of which will be reported here.
Numerous MDS studies of phospholipid membranes
have been reported during the past few years, and in
some cases, the behavior of solutes in a model mem-
brane was studied.8,9 For a recent comprehensive review
of membrane MDS studies, see Tieleman et al.10 To our
knowledge, only qualitative or, at best, semiquantitative
information on solute-membrane interactions, and the
resulting changes in membrane properties, has been
derived from MDSs of membranes. One of the main
reasons for this situation is the computational overhead
of performing membrane MDSs. The methodology used
in this approach incorporates a simplified membrane
system, to keep the computational cost at a reasonable
level. Classical intramolecular solute descriptors and
grid cell occupancy descriptors11 were also included in
the QSAR/QSPR model-building process.
Ch em istr y. Several synthetic routes for the prepara-
tion of the N,N-dimethyl-3-phenylpropylamine (6) and
substituted derivatives have been reported,13-17 but
these procedures often involve too laborious steps. In
this series of model compounds 1-26 the synthesis of
ketone Mannich bases and subsequent hydrogenation
of their oxo function16,18 gave unsatisfactory results.
Therefore, more convenient methods were developed
using the Wittig and Vilsmeier reaction as key steps
followed by catalytic hydrogenation of the unsaturated
side chain.
The chemical structures and the synthetic schemes
for the compounds are given in Table 1; biophysical
properties are listed in Table 2. Compounds 16 and 17
The first strategy developed for the preparation of our
model compounds was based on their structural simi-

Membrane-Interaction QSAR Analysis
J ournal of Medicinal Chemistry, 1999, Vol. 42, No. 19 3877
Sch em e 1a
a
(a) N,N-Dimethylaminoethylphosphonium bromide/butyllithium; (b) Pd-C/H2; (c) HCl.
Sch em e 2a
a
(a) DMF/POCl3; (b) HClO4; (c) Pd-C/H2; (d) HCl. R ) H or Ar.
Sch em e 3a
a
(a) PhMgBr; (b) DMF/POCl3; (c) HClO4; (d) Pd-C/H2; (e) HCl.
larity with cinnamylamine derivatives (Scheme 1) pre-
pared by the method of Marxer and Leutert.19 It
prompted us to treat appropriate para-substituted ben-
zaldehydes with N,N-dimethylaminoethylphosphonium
bromide and butyllithium as a strong base to give the
corresponding allylamines. Hydrogenation of the side
chain in the presence of palladium-on-carbon followed
by reaction with hydrochloric acid produced the amine
hydrochlorides 1-5. As expected, the phenolic amine 3
was directly formed by hydrogenation of the p-benzyl-
oxycinnamylamine derivative involving cleavage of the
benzyloxyphenyl ether moiety.
Since this procedure generally proved to be rather
complicated and, furthermore, specially substituted
carbaldehydes were not as accessible as the correspond-
ing acetophenones, another route was performed ac-
cording to the method depicted in Scheme 2. Vilsmeier
reaction of the appropriate acetophenones was ac-
complished using phosphorus oxychloride in N,N-di-
methylformamide to give 3-chloro-2-propeniminium salts
isolated as perchlorates20,21 except for the iminium salt
to be required for the preparation of 9 (see Experimental
Section). In this procedure, the carbonyl function is
invariably substituted by the chlorine atom via the
corresponding enol group. The 3-chloro-2-propenimini-
um salts were catalytically hydrogenated using pal-
ladium-on-carbon, and the resulting amines converted
to the hydrochlorides 1 and 4-17 in the usual manner.
All double bonds were reduced, accompanied by hydro-
genolytic cleavage of the chlorine atom. Starting from
benzyl phenyl ketone, this procedure could also be
successfully applied for the preparation of the 2,3-
diphenylpropylamine (20). The method proved not to be
suitable for the preparation of the nitramine 21 as well
as the phenolic amines 3, 23, and 24 which required
subsequent heating of 12, 14, and 17 in hydrobromic
acid. The resulting hydrobromides were converted to the
corresponding hydrochlorides. Furthermore, it was found
that the nitro group could be conveniently introduced
in the para-position of the aromatic ring by simple
nitration of 6 to yield the nitrobenzene derivative 21,
which was catalytically hydrogenated using palladium-
on-carbon to give para-substituted aniline 26. Finally,
some trialkylpropylammonium salts were prepared by
methylation using methyl iodide and subsequent anion
exchange to give the hydrochlorides 22 and 25.
N,N-Dimethylamino-3,3-diphenylpropylamine hydro-
chlorides 1822 and 19 were prepared by treatment of
the appropriate acetophenone with phenylmagnesium
bromide, subsequent Vilsmeier formylation of the 1,1-
diphenylethanol, and catalytic hydrogenation of the re-
sulting 3,3-diphenyl-2-propeniminium salt (Scheme 3).
Partition coefficients were determined by HPLC on
an RP18 column and different methanol/water mobile
phases. The capacity factors were extrapolated to a
100% aqueous medium. The system was calibrated
using different aromatic hydrocarbons of known lipo-
philicity. pKa values were determined by aqueous titra-
tion of the hydrochlorides.
P h a r m a cology. 1. 45Ca 2+ Bin d in g to Mon ola yer s
of P h osp h a tid ylser in e. A detailed description of the
experimental procedure was given before.1 In short, a
Teflon planchette of 4.5-cm diameter was filled with 5
mL of buffer (0.01 mM CaCl2 supplemented with trace
amounts of 45CaCl2, 5 mM NaCl, 2 mM TES, 2 mM
histidine; pH 7.5). Radioactivity was detected by a
Geiger-Mu¨ller counting tube (Frieseke & Hoepfner,
Erlangen, Germany) positioned above the Teflon dish;
3 nmol of phosphatidylserine (from bovine brain; purity
98-99%, Sigma Chemical) dissolved in 1 µL of chloro-

3878 J ournal of Medicinal Chemistry, 1999, Vol. 42, No. 19
Klein et al.
form was carefully applied to the surface of the buffer
to form a phospholipid monolayer. Adsorption of 45Ca2+
to the phospholipid was indicated by a 2-fold elevation
of the counting rate under control conditions. Test
compounds were dissolved in the buffer at the appropri-
ate concentrations.
2. P h a se-Tr a n sition Tem p er a tu r e of DP P A Lip -
osom es. As decribed previously in detail,6,1,5 lipo-
somes of dipalmitoyl-sn-glycero-3-phosphate monoso-
dium salt (purity 99%, Sigma Chemical) were prepared
in a buffer of 14 mM TES, 14 mM histidine, pH 6,
containing appropriate amounts of test compound; 10
µL of the suspension was placed into an aluminum
capsule (Perkin-Elmer, Ueberlingen, Germany) which
was placed in a differential scanning calorimetry equip-
ment (DSC-2C/Intracooler II, Perkin-Elmer). The sample
was heated from 12 to 72 °C at a rate of 5 °C/min. The
temperature of the upward deflection of the transition
signal was used to indicate the phase-transition tem-
perature which occurred under control conditions at
about 65 °C.
3. E xcit a bilit y of Isola t ed Bea t in g Gu in ea P ig
Hea r ts. The procedure to induce arrhythmia by alter-
nating current in guinea pig heart Langendorff prepa-
rations was described previously.1 The hearts were per-
fused through the coronary vasculature with Tyrode’s
solution modified according to von Muralt (137 mM
NaCl, 2.7 mM KCl, 1.8 mM CaCl2, 11.9 mM NaHCO3,
1.1 mM MgCl2, 0.2 mM NaH2PO4, 5.5 mM glucose;
oxygenated by 95% O2/5% CO2; pH 7.3; 35 °C). Contrac-
tions were recorded via a fine hook fixed to the apex of
the heart. To determine the threshold current for
arrhythmia, a 50-Hz alternating current of increasing
intensity was applied through two fine flexible brushes
(diameter 5 mm) soaked with Tyrode’s solution and
attached in opposite positions to the ventricular myo-
cardium. Test compounds were dissolved in the perfu-
sion medium and applied over a period of 30 min before
measurement of the threshold current for arrhythmia.
Concentrations were increased in cumulative fashion.
F igu r e 1. MDS sampling protocol for the membrane-solute
interaction modeling.
field and the molecular dynamics modeling program
MOLSIM30 were employed for all MDSs. The computa-
tional protocol for the MDS is outlined in Figure 1. A
time step of 0.001 ps was used in all MDSs. Trajectory
data, i.e., atomic positions and system energies, were
recorded every 100 steps. Two-dimensional (the ”plane”
of the monolayer) periodic boundary conditions were
applied. Temperature scaling was accomplished by
coupling the system to an external constant-tempera-
ture bath.31
The model monolayer was first heated to 20 K and
then to 50 K and from that point in increments of 50 K
to a final temperature of 350 K. At each temperature
increment, 4 ps of MDS was carried out to allow for
structural relaxation and distribution of kinetic energy
throughout the system. When 350 K was reached, 50
ps of MDS was performed in order to equilibrate the
system.
One DPPC molecule was removed from the equili-
brated monolayer model. The test solute molecule was
then inserted into the resulting ”hole”, its propylamino
side chains being oriented toward the “aqueous inter-
face” side of the monolayer and the protonated amino
groups located near the phospholipid headgroup region.
However, three different monolayer-solute alignments
were examined for compounds 1, 6, 11, and 18, respec-
tively. These compounds were chosen as a subset for
extensive alignment analysis because they cover a wide
range of influence on Tt. The four alignments examined
are depicted schematically in Figure 2.
To remove unfavorable high-energy van der Waals
interactions between solute and phospholipid molecules,
the energy of the system was minimized by a series of
steepest descent and conjugate gradient minimization
steps. The energy convergence criterion was a gradient
of less than 0.5 kcal/(Å‚mol). Convergence was generally
achieved within less than 200 steps.
Mem br a n e-In ter a ction QSAR An a lysis. 1. Bu ild -
in g th e Molecu les a n d th e Sim u la tion System . All
compounds were built using the building tool and library
fragments of Chemlab-II.23 The dimethylaminopropyl
side chains were assigned all-trans conformations. The
side chain amino groups are protonated under physi-
ological conditions (see pKa values, Table 2). Accord-
ingly, the compounds were built as monocations. DPPC
was initially built in HyperChem,24 using available
crystal structure data for DMPC.25 Calculations at the
semiempirical MO level were carried out on the solute
molecules and DPPC to assign partial atomic charges
and to further optimize molecular geometry. The AM1
Hamiltonian, implemented in the MOPAC 6.026 pro-
gram, was employed.
An assembly of 4*4*1 DPPC molecules, resembling a
monolayer, was used as a starting structure for an
equilibration MDS. The cell parameters for an indi-
vidual phospholipid molecule were 8*8*32 Å, γ ) 97.4.
These parameters result in an average surface area per
phospholipid of 64 Å2, which is close to the reported
value of about 62 Å2 for the fully hydrated fluid lamellar
(LR) phase of DPPC.27 As in the previous studies
performed by our group,9,28,29 an extended MM2 force
The monolayer-solute systems were reheated using
the same heating schedule as described previously for
the monolayers. Equilibration periods were 1 ps/heating

Membrane-Interaction QSAR Analysis
J ournal of Medicinal Chemistry, 1999, Vol. 42, No. 19 3879
Ta ble 4. Intramolecular Solute Descriptors
descriptor
source
log P (logarithm of the partition coefficient)
sum of atomic polarizabilities
superdelocalizability
HOMO
LUMO
a
b
b
c
c
c
c
b
dipole moment
heat of formation
J urs-Stanton CPSA (charged partial
surface area) descriptors
radius of gyration
van der Waals surface area
molecular volume
density
no. of rotatable bonds
no. of hydrogen bond acceptors
no. of hydrogen bond donors
molecular weight
b
b
b
b
b
d
d
c
F igu r e 2. Solute alignments explored in the phospholipid
matrix. Alignment 1 was used as the starting point for the
generation of the membrane-solute interaction descriptors.
calculated octanol/water partition coefficient
molar refractivity
desolvation free energy for water
desolvation free energy for octanol
b
b
b
b
Ta ble 3. Descriptors Derived from the Membrane-Interaction
MDSs
Einter(AVG)
interaction energy between solute and
membrane averaged over the MD
trajectory (sum of H-bond,
a
b
Experimental value.10 Computed.30 c Calculated, MOPAC
6.22 Determined manually.
d
electrostatic, and vdW contributions)
interaction energy between solute and
membrane at total system minimum
potential energy (sum of H-bond,
electrostatic, and vdW contributions)
Einter(AVG) for H-bond interactions
Einter(MIN) for H-bond interactions
Einter(AVG) for electrostatic interactions
Einter(MIN) for electrostatic interactions
Einter(AVG) for van der Waals interactions
Einter(MIN) for van der Waals interactions
change in membrane density upon uptake
of the solute
Einter(MIN)
employs partial least-squares (PLS) regression,35 com-
bined with a method based on genetic algorithms, the
genetic function approximation (GFA),36 which is used
for model building. Friedman’s lack-of-fit (LOF) measure
is used as the ranking/score function during model
evolution.37 The LOF contains a penalty term for
independent variable model size (the smoothing factor)
and is the default ranking score of the Wolf program.
All descriptors were mean-scaled and centered. An
initial random population of 300 models was used. The
20 best-scoring models of the final GFA populations
were cross-validated, using the leave-one-out (n-fold)
cross-validation technique. GFA crossover parameters
were varied for each endpoint until a GFA model
population of optimal statistic quality, defined in terms
of the average cross-validated r2 of the 20 best models,
was obtained.
Attempts were made to build QSPR models with
linear and with linear and quadratic basis set functions
(terms). Purely linear models were statistically less
significant than combined quadratic/linear models. In
general, however, the same descriptors were of major
significance in both the linear and quadratic/linear
model populations. Therefore, model populations were
initialized with the same creation probability for both
linear and quadratic basis functions.
Einter(AVG
Einter(MIN
Einter(AVG
Einter(MIN
Einter(AVG
Einter(MIN
dp
•
HBD)
HBD)
COU)
COU)
VDW)
VDW)
•
•
•
•
•
dS
D
d
change in membrane entropy upon uptake
of a solute molecule
diffusion coefficient of a solute in the
membrane
average depth of a solute in the membrane
increment. Production trajectories were recorded over
20-ps MDS runs at 350 K. Membrane-solute interaction
descriptors were calculated using the last 15 ps of these
runs. For reference purposes, an undisturbed monolayer
model, without any phospholipids removed or solutes
inserted, was modeled in the same way as the mem-
brane-solute systems.
2. Calcu lation of Mem br an e-In ter action Descr ip-
tor s. The membrane-interaction descriptors determined
from the production runs are defined in Table 3. Solute-
membrane interaction energies were extracted from the
intermolecular interaction energy trajectory files. Other
membrane-interaction descriptors were calculated as
described in the Computational Details section.
3. Ca lcu la tion of In tr a m olecu la r Descr ip tor s.
Intramolecular solute descriptors were calculated manu-
ally, or by means of the molecular modeling package
Cerius2, release 3.0.32 AM1-minimized solute structures
were used for the intramolecular descriptor calculations.
The compounds were aligned using the side chain
nitrogen, the phenyl-C1, and the phenyl-C4 as align-
ment atoms. All of the intramolecular descriptors
considered in QSPR model building are listed in Table
4. The charged partial surface area (CPSA) descriptors
of Stanton and J urs33 are not individually reported.
4. Con str u ction of QSAR/QSP R Mod els. QSPR
models were built using the program Wolf 6.2.34 Wolf
The two chiral compounds 19 and 20 were not
included in any of the training datasets, because the
pharmacological and biophysical testing had been done
with the racemates. However, predictions were made
for the stereoisomers of both compounds. Compound 23
elevates the phase-transition temperature of DPPA.
This behavior was not observed for any other compound
in the dataset, and it was thus judged reasonable to
exclude this compound from the training dataset for
phase-transition temperature.
Resu lts a n d Discu ssion
P h a r m a cology. The test compounds inhibit, in a
concentration-dependent manner, the binding of 45Ca2+
to phosphatidylserine monolayers as has been previ-
ously illustrated for cationic amphiphilic therapeutics.1

3880 J ournal of Medicinal Chemistry, 1999, Vol. 42, No. 19
Klein et al.
F igu r e 3. Stereographic representations of the lowest potential energy states from the MDS of compounds 6 and 15 inter-
acting with the model DPPC monolayer. Compound 6, the unsubstituted ”parent” compound of the dataset, is preferably located
within the headgroup region, while compound 15 inserts into the DPPC monolayer and aligns itself parallel to the hydrocarbon
chains.
The measure of potency is the concentration of the test
compound at which 45Ca2+ binding is reduced to 50% of
the control value, i.e., 45Ca2+ adsorption to the phos-
pholipid measured in the absence of test compound (IC50
values, cf. Table 2).
The phase transition of DPPA liposomes is affected
by the test compounds as has been illustrated previously
for cationic amphiphilic therapeutics1,6 and for the test
compounds 5, 6, and 18.5 The temperature of the drug-
induced signal is specific for a given test compound
(Table 2).
The threshold current for arrhythmia increases in
the presence of the test compounds in a concentration-
dependent fashion as has been previously illustrated for
cationic amphiphilic therapeutics.1 The drug concentra-
tion at which the threshold current is elevated by 50%
over the control value in the absence of test compounds
is used as a measure of the membrane-stabilizing poten-
cy (AC50 values, cf. Table 2).1 The AC50 to suppress myo-
cardial excitability may be taken to reflect the potency
of a drug to block voltage-dependent sodium channels.
Mem br a n e-Solu te MDSs. A comparison of the
MDS trajectories for the different solute arrange-
ments in the phospholipid matrix (cf. Figure 2) indicated
the initial alignment 1 to be generally more stable than
the other alignments. This finding can be inferred from
a visual inspection of the MDS atomic position trajec-
tories by means of the MOLSIM visualization tool,
MDMOVIE, as well as from an inspection of the mem-
brane-solute interaction energies and the total system
potential energies, which are, for most of the com-
pounds, the lowest (most stable) for alignment 1. The
MDS trajectories for alignment 3 generally show move-
ment of the solute toward the phospholipid headgroup
region, until a final state, similar to alignment 1, is
reached. Both alignments 2 and 4 do not lead to stable
positions of the solute in the membrane.
Visual inspection of the MDS trajectories obtained
with alignment 1 revealed that the majority of the
solutes did not show any pronounced overall movement
away from their initial positions. Some compounds,
however, did behave differently. Compounds 2, 11, and
18 adopt positions which can be described as ”parallel”
to the monolayer plane. In this alignment, the com-
pounds interact mainly with the headgroup region of
the model monolayer and are virtually inserted into it.
A similar, but less pronounced, tendency to migrate
from the initial alignment position toward the head-
group region can be seen with compounds 6, 13, 16, and
17. In any case, the major movement starts during the
heating process and is finished before the start of the
production run. This observation is consistent with the
calculated diffusion coefficients for the production runs,
which range from 10-10 to 10-9 cm2/s over the data set.
The highest diffusion coefficients are observed for the
small compounds, such as 10 and 14. Large compounds,
such as the stereoisomers of compound 16, experience
smaller degrees of mobility, which is reflected by lower
computed diffusion coefficients. Graphical representa-
tions of the minimum potential energy states sampled
over the course of the MDS for compounds 6 and 15,
respectively, are shown in Figure 3.

Membrane-Interaction QSAR Analysis
J ournal of Medicinal Chemistry, 1999, Vol. 42, No. 19 3881
F igu r e 4. Predicted vs observed (a) -log AC50, (b) -log IC50, (c) Tt, and (d) Tt for the QSAR models given by eqs 1-4 in the text.
The average depth of a solute in the membrane,
defined as the ”depth” descriptor, ranges from 12.5 to
18 Å. These values may seem large, given a monolayer
thickness of about 20 Å. However, the distance to the
single, closest phospholipid headgroup is always much
smaller and amounts to around 6 Å for most of the
solutes. Still, it does not seem appropriate to only look
at the distance to the closest phospholipid, since the
headgroups sometimes undergo a considerable degree
of vertical movement. If, instead, the average distance
to the five closest headgroups is taken as a measure for
solute depth, the idea of a phospholipid headgroup
region is retained. Some of the five closest phospholipid
phosphorus atoms have relatively large, lateral (parallel
to the monolayer plane) distances from the solute, which
increases the average distance values. Both the aver-
age and minimum membrane-solute interaction ener-
gies span a relatively wide range of values. The most
stabilizing interactions arise for compound 1, with
Einter(AVG) ) -58.9 kcal/mol, followed by compound 13
and the stereoisomers of compound 16. Differences in
membrane-solute interaction energies as well as other
membrane-dependent descriptors are, in most cases,
small for pairs of stereoisomers. For example, Einter(AVG)
is -52.0 kcal/mol for R-16 and -56.0 kcal/mol for S-16.
The entropy of the model membranes shows practically
no change after uptake of the solute, and membrane
density varies only slightly. This is reasonable for low
concentrations (1/15) of solutes in the monolayer.
QSAR for An tia r r h yth m ic Activity. The model
population resulting from the GFA run consists of two-,
three-, and four-term models, practically all of which
are linear. The most often used descriptor is clearly log
P, which is found in about 60% of all QSAR models.
Surprisingly, it is not incorporated in the best-scoring
model:
-log AC50 ) 2.572 + 7.37*(acp*4* - 2*10*0) +
0.000378*APol + 0.07647*F-H2O -
0.072175*WPSA-3 (1)
LOF ) 0.085, r2 ) 0.97, xv-r2 ) 0.94,
SE ) 0.155, n ) 19
where LOF, Friedman’s lack-of-fit score; xv-r2, cross-
validated r2; SE, standard error; (acp*4* - 2*10*0), grid
cell occupancy descriptor determined by MDS using 4D-
QSAR analysis;7 WPSA-3, surface weighted positive
surface area;33 APol, sum of atomic polarizabilities;38
F-H2O, aqueous desolvation free energy;39 GFA options,
80 000 crossover operations, smoothing factor ) 2.
The fit and predictive quality of this model is equal
to the hybrid grid cell/log P model reported in ref 7, but
it contains only four variables as compared to the five-
variable model reported previously. One of the descrip-
tors (WPSA-3) of eq 1 is modestly correlated to log P (r2
) 0.73 for the training set of 19 compounds) and might
be regarded as a substitute for this measure. Except for
WPSA-3 and the grid cell occupancy descriptor (r2
)
0.69), no substantial cross-correlations among the de-
scriptors of the model can be found. Plots of predicted
versus observed -log AC50 values are shownn in Figure
4, part a. An inspection of this plot reveals that there
are no major outliers or ranges in -log AC50 where the
QSAR fit is particularly poor.
Equation 1 is consistent with the local anesthetics
receptor concept, in that it incorporates a 4D-QSAR grid
cell occupancy descriptor11 as a measure of molecular

3882 J ournal of Medicinal Chemistry, 1999, Vol. 42, No. 19
Klein et al.
geometry. Furthermore, it has long been known1,40 that
lipophilicity is the major factor in determining the
antiarrhythmic activity of cationic amphiphilic drugs.
This suggests that the spatial requirements of the
receptor are comparably unspecific and lipophilic inter-
actions play a major role in the inactivation process.
Compound 20 is chiral, and since the antiarrhythmic
activity has been measured for the racemate, it was not
included in the training dataset. Predictions for both
stereoisomers were made using eq 1. The predicted
values are 4.5 and 4.4, respectively. Both of these values
are very close to the measured activity of 4.1, and this
emphasizes the predictive quality of the QSAR model.
Tt ) 298.18 - 0.023*(Einter(MIN•VDW) + 18.68)2 -
0.105*(PPSA•3 - 43.198)2 + 9.070*(ROG -5.199)2
(3)
LOF ) 21.55, r2 ) 0.89, xv-r2 ) 0.82,
SE ) 2.95, n ) 15
Tt ) 300.34 - 0.023*(Einter(MIN) + 32.14)2 +
97060.0*(FPSA•3 - 0.068)2 (4)
LOF ) 20.92, r2 ) 0.81, xv-r2 ) 0.68,
SE ) 3.90, n ) 15
QSP R for In h ibition of Ca lciu m Ion Bin d in g to
P S Mon ola yer s. The inhibition of calcium ion binding
to PS monolayers had been measured for only 12
compounds. QSPR models which are based on such a
small dataset will have to be evaluated with particular
carefulness. The best QSPR model incorporates three
descriptors:
where LOF, Friedman’s lack-of-fit score; xv-r2; cross-
validated r2; SE, standard error; ROG, radius of gyra-
tion; Einter(MIN), PPSA•3, and FPSA•3; see text.
Both QSAR models, eqs 3 and 4, include one descrip-
tor derived from the membrane MDS. These intermo-
lecular descriptors are the total and the van der Waals
interaction energies between solute and phospholipids
at the minimum potential energy state of the simulation
-log IC50 ) -5.24 + 0.019*TPSA +
system (Einter(MIN) and Einter(MIN•VDW)). Also, surface
0.111*(Dipole-Z - 1.34)2 + 12.11*RPCG (2)
charge measures are included in both models. Equation
3, in addition, incorporates another intramolecular
term, the radius of gyration of the compounds. Obvi-
ously, the intermolecular descriptors, which were de-
rived from the membrane MDS, are of significant value
for explaining the influence of the compounds on the
DPPA phase-transition temperature.
LOF ) 0.104, r2 ) 0.98, xv-r2 ) 0.79,
SE ) 0.155, n ) 12
where LOF, Friedman’s lack-of-fit score; xv-r2, cross-
validated r2; SE, standard error; TPSA, total polar
surface area; Dipole-Z, dipole moment in the Z direction,
where Z is one of Cartesian axes; RPCG, relative
positive charge (charge of most positive atom/total
positive charge); GFA options, 80 000 crossovers, smooth-
ing factor ) 1.5.
Although models 3 and 4 incorporate quadratic terms,
no real parabolic relationship exists between the de-
scriptors and the considered endpoint. The presence of
quadratic terms allows a better fit to the training data
set, but no minimum or maximum can actually be seen
with the descriptor values in the training set. Plots of
predicted versus observed Tt are given in Figure 4, parts
c and d. Equation 4, part d of Figure 4, shows more
scatter between predicted and observed Tt values,
particularly in the middle of the Tt range, than eq 3,
part c of Figure 4.
The solute-membrane interaction energy descriptors,
Einter(MIN) and Einter(MIN•VDW), are positively correlated
with membrane-transition temperature. This means
that a more endergonic interaction of a solute with the
membrane will lead to a more pronounced depression
of the phase-transition temperature. This observa-
tion can be interpreted in the following fashion: Dif-
The main conclusion that can be made from an
inspection of this QSPR model is that the replacement
of calcium ions is an electrostatically driven process. All
descriptors are positively correlated with potency. Cal-
cium ions and cationic amphiphilic compounds bind to
the negatively charged headgroup of phosphatidylserine,
which is exposed to the aqueous phase in the assay
system. The insertion of the analogues into the phos-
pholipid phase appears to be of minor importance for
their activity, since no lipophilicity measure or mem-
brane-interaction descriptor is of major significance in
the final QSPR model population. The most often used
descriptors are TPSA, RPCG, Dipole-Z (see eq 2), the
grid cell occupancy descriptors, and Dipole-X (dipole
moment in the X direction). Plots of predicted versus
observed -log IC50 values are given in Figure 4, part b.
The lower and upper ranges in -log IC50 appear to
contain the major lack of fits between the QSPR model
and the observed data.
QSP R for th e Dep r ession of Tt in DP P A Lip o-
som es. The GFA model-building process (80 000 cross-
overs, smoothing factor ) 1.5) for the Tt endpoint
resulted in a population of two- and three-descriptor
models. The most often used descriptors were one of the
CPSA descriptors (FPSA-3, a fractional positive surface
charge measure), the membrane-solute interaction en-
ergy at the total potential energy minimum state of the
simulation system, and the radius of gyration of the
compounds. The best two- and three-term models are
given as eqs 3 and 4 below:
ferences in interaction energies, from Einter(MIN
•
,
VDW)
Einter(MIN COU), or Einter(MIN), largely originate from dif-
•
ferences in the location and alignment of the solute in
the membrane. For instance, a solute which is oriented
parallel to the alkyl chain, with its hydrophobic moiety
inserted between them, will experience a more ender-
gonic interaction with its phospholipid neighbors than
a solute that is located between the headgroups on the
monolayer surface and has a larger part of its molecular
surface exposed to the ”aqueous” outside. Therefore,
interaction energies, in particular Einter(MIN•VDW), can
be regarded as measures of the extent to which a solute
interacts with the phospholipid hydrocarbon chains.
Compounds that experience a stronger interdigitation
between the chains, such as compound 5, introduce a
more pronounced destabilizing effect into this region of
the membrane. Thus, these compounds lower the phase-

Membrane-Interaction QSAR Analysis
J ournal of Medicinal Chemistry, 1999, Vol. 42, No. 19 3883
Ta ble 5. Predicted and Observed Transition Temperatures (K)
for Solutes That Were Not Included in the Training Set
transition temperature to a larger extent than solutes
which do not interdigitate as significantly, such as
compound 21.
compd
Tt (pred) model 3
Tt (pred) model 4
Tt (obsd)
R-19
S-19
R-20
S-20
23
310
306
320
325
314
292
292
323
322
299
294a
294a
309a
309a
349
The two CPSA descriptors, PPSA-3 and FPSA-3, that
appear to be of major significance in the Tt QSPR models
are both measures of positively charged solute molecular
surface area. PPSA-3 is defined as the sum of the
product of the [solvent-accessible surface area × partial
charge] for all positively charged atoms of the solute.
FPSA-3 is obtained from PPSA-3 by dividing by the total
molecular solvent-accessible surface area.33 Both of
these descriptors are positively correlated with transi-
tion temperature: i.e., molecules with more negatively
charged surfaces induce larger declines in Tt than
molecules with positively charged surfaces. One possible
interpretation for the importance of these descriptors
in the QSPR models is that these descriptors reflect the
influence of the solutes on the electrostatic environment
of the membrane-water interphase and the electro-
static stability of the headgroup region in particular.
DPPA is the phospholipid used in the experimental
studies and possesses anionic headgroups, while DPPC
is an effectively neutral molecule. The CPSA descriptors
may, in this respect, act as corrections which account
for electrostatic differences between these two phospho-
lipids. Another possible explanation can be inferred from
the fact that the phospholipid hydrocarbon chains have
slightly positively charged surfaces, resulting in a
repulsive interaction with solutes also having positive
surface charges. The charge repulsions would also lead
to less of an insertion of such a solute into the hydro-
carbon chain region of the membrane.
a
Measured for the racemate.
behavior might be a real phenomenon, arising from
interactions of the solute methoxyl group with the
phospholipid headgroups. However, a random effect,
introduced by insufficient sampling of thermodynamic
states, seems more likely to be the case. Future devel-
opment on the membrane-interaction QSAR/QSPR meth-
odology will have to take the possibility of insufficient
sampling into consideration, and techniques to overcome
this limitation will have to be explored. One obvious
improvement is to extend the sampling period, but due
to the computational overhead of MDSs on large sys-
tems, this seems a rather costly approach. A more
economic approach might be to run multiple, short
simulations on systems with slightly different initial
states, i.e., modified molecular positions and initial
velocities.
How well do the membrane-interaction QSPR models
predict the influence of compounds 19, 20, and 23, which
were not included in the training set, on the phase-
transition temperature? The predictions from models 3
and 4 are given in Table 5. These are the best two QSPR
models from the two distinct families of QSPR models
generated in the model optimization process.
The radius of gyration is calculated from the atomic
coordinates of a molecule relative to its center of mass
and can, therefore, for small molecules, be regarded as
an indirect measure of molecular branching and size.
This molecular property is often used in biopolymer
modeling, particularly to describe the folding state of
proteins and other biomolecules.39 For the solutes
considered in this study, a large radius of gyration is
associated with a large depression in Tt. Compounds in
the training set with large radii of gyration are larger
and more elongated than other analogues. Therefore, if
an anchoring of the solute amine group to the phos-
pholipid headgroup region is assumed, compounds with
large radii of gyration reach further into the hydro-
carbon chain region than other analogues. This inter-
pretation is similar to the one given for the mem-
brane-interaction energy descriptors, Einter(MIN) and
Predicted values for the stereoisomers of 19 and 20
do not differ substantially, using either QSPR model.
Comparing predicted and observed values, it is found
that both models give the correct relative ranking of
Tt: i.e., the Tt for compound 20 is correctly predicted to
be higher than that of compound 19. The predictions
from model 4 are closer to the observed values than the
predictions from model 3, especially for compound 19,
where the prediction is in excellent agreement with the
observed value. The rise in transition temperature,
observed for compound 23 in DPPA liposomes, is not
predicted.
It is remarkable that no elevation in Tt was observed
when compound 23 was added to a DPPC liposome
preparation. Figure 5 shows the transition temperatures
of both DPPC and DPPA liposomes under the influence
of selected analogues. The transition temperatures for
compounds 6, 3, and 5 are essentially independent of
phospholipid headgroup type. This is not the case with
compound 23, which decreases the Tt of DPPC liposomes
by about 4 K to a value of 310.9 K. This observation is
much closer to the predicted values of 314 and 298 K,
respectively, than the value of 348.8 K, which was
observed with DPPA. This supports the validity of our
model. Apparently, the interaction of compound 23 with
DPPA is governed by molecular events which cannot be
adequately modeled by a DPPC simulation system. It
would be quite interesting to study the behavior of
selected compounds in a DPPA simulation system.
Explanations for the unusual behavior of compound
23 in DPPA can, at best, only be hypothesized. The
reason for the increase in Tt might be because this
Einter(MIN•VDW). However, the membrane-interaction
descriptors are only weakly correlated to the radius of
gyration.
The radius of gyration might also provide corrections
and/or enhancements to data generated in the MDSs.
Such corrections and enhancements may be needed
because by performing only one MDS per compound,
metastable conformations and alignments of the solutes
in the membrane cannot always be identified. Thus,
descriptors derived from the simulations are necessarily
biased. For example, the aromatic moiety of compound
14 does not penetrate as far into the monolayer as is
the case with compound 5. Consequently, the mem-
brane-solute van der Waals interaction energy for
compound 14 is much lower than for compound 5. This

3884 J ournal of Medicinal Chemistry, 1999, Vol. 42, No. 19
Klein et al.
The statistical fit and significance, particularly in terms
of xv-r2, of this model are clearly inferior to the two
membrane-interaction models given as eqs 3 and 4.
Thus, the necessity and significance of intermolecular
membrane-interaction descriptors in modeling Tt are
established.
Con clu sion s
The main conclusions from this work can be sum-
marized as follows:
Significant QSAR models are developed for -log AC50,
-log IC50, and Tt from the pool of trial descriptors
generated from intramolecular properties and features
a n d from intermolecular membrane-interaction proper-
ties generated from the MDS studies.
The sets of descriptors in each of the three QSAR
models to respectively describe AC50, IC50, and Tt are
quite different from one another.
F igu r e 5. Influence of selected compounds on the phase-
transition temperatures of DPPA and DPPC liposomes.
Ta ble 6. Correlation Coefficients of Average and Minimum
The different descriptor sets for each of the AC50, IC50,
and Tt QSAR models suggest that different mechanisms,
and/or sites of action, are involved in each of these three
properties. The QSAR models also provide a physico-
chemical basis as to why AC50, IC50, and Tt measures
are not highly correlated among one another.
None of the QSAR descriptors of the AC50 model are
derived from explicit membrane interaction MDS mod-
eling. This form of the QSAR suggests, in turn, that the
lipid regions of the membrane do not play a major role
in the expression of antiarrhythmic activity. Conversely,
the QSAR descriptors for AC50 identify specific and/or
local physicochemical features of the cationic am-
phiphilic drugs that are indicative of sp ecific ligand-
receptor interactions.
The QSAR model for calcium ion replacement does
not contain membrane-interaction descriptors charac-
teristic of a compound embedded within the membrane
bilayer. However, all of the QSAR descriptors are
electrostatic/charge properties in nature and are con-
sistent with electrostatic interactions at the surface of
a membrane probably involving the headgroups of the
phospholipid.
The depression of the membrane phase-transition
temperature, Tt, is expressed by a QSAR model which
contains MDS membrane-interaction descriptors which,
in turn, suggests Tt behavior involves the incorporation
of compounds into the hydrocarbon chain region of the
membrane bilayer.
The question arises whether the same amount and
quality of structure-property information could have
been extracted from the ”right” set of purely intramo-
lecular solute descriptors. That is, could the membrane-
solute MDS modeling have been avoided? We think that
the answer is ”No”. Some of the membrane-interaction
QSAR descriptors are collinear within their own class,
but no significant correlation to any intramolecular
solute descriptor could be found. This strongly suggests
that the membrane-solute descriptors determined by
MDS provide significant additional structure-property
information that cannot be obtained by any intramo-
lecular measures.
Interaction Energies for All Compounds
total
0.85
electrostatic
0.94
van der Waals
0.75
H-bond
0.99
compound adopts a position parallel to the membrane-
water interface within the headgroup region, and ”cross-
links” two or more phospholipids, by means of hydrogen
bonds to the phenolic oxygen and, simultaneously, by
ionic interactions to the charged amine group. The
MDSs support this idea by adopting states in which a
solute is aligned parallel to the interface for all phenolic
compounds (3, 23, 24) and the nitro compound (21).
However, no elevation in Tt is observed for 24, 3, and
21. Moreover, compound 24, in fact, is one of the most
pronounced Tt-depressing solutes. A larger specific
distance between the amino and the hydroxyl groups
of a solute might be necessary to easily permit this
”cross-linking” effect. Thus, it would be interesting to
study derivatives of compound 23, e.g., compounds with
more than one phenolic hydroxyl group, or the corre-
sponding aromatic amine.
None of the average membrane-solute interaction
energies are found as descriptors in the best membrane-
interaction QSPR models. Instead, interaction energies
at minimum system potential energy are preferred in
the best QSPRs. This preference may again be due to
incomplete sampling efficiency. A true Boltzmann aver-
aged potential energy distribution over a MDS trajectory
is not achieved. Interaction energies at minimum sys-
tem potential energy may provide the most realistic
representation of the most highly populated states of
the membrane-solute system. However, all interaction
energies at the minimum system potential energy are
meaningfully correlated with their average energy
counterparts (cf. Table 6).
Finally, a variety of models without any membrane-
interaction descriptors emerged from the GFA runs, the
best of which (in terms of cross-validated r2) is given as
eq 5:
Tt ) 295.53 - 0.2276*(J urs-PPSA•3 - 37.543)2 +
15.203*(ROG - 4.674)2 (5)
LOF ) 25.652, r2 ) 0.77, xv-r2 ) 0.50,
SE ) 4.655, n ) 15
Membrane-interaction descriptors do not appear to be
of particular significance in explaining the antiarrhyth-
mic or the calcium-displacing activity of the compounds.
This suggests that both endpoints are not primarily

Membrane-Interaction QSAR Analysis
J ournal of Medicinal Chemistry, 1999, Vol. 42, No. 19 3885
NMR (CDCl3) δ 2.17-2.28 (2H, m), 2.72 (2H, t, J ) 7.32 Hz),
2.79 (6H, d, J ) 4.75 Hz), 2.93-3.00 (2 H, m), 7.08 (2 H, d, J
) 8.30 Hz), 7.43 (2 H, d, J ) 8.29 Hz).
determined by the influence of the compounds on
thermodynamic membrane properties. Spatial features,
reflected by the grid cell occupancy descriptor in the
AC50 QSAR model, are important for the expression of
antiarrhythmic activity and are supported by whole-
molecule property descriptors such as surface charge
measures. This underscores the importance of penetra-
tion and adsorption behavior at some point in the
pharmacokinetic-pharmacodynamic chain of the assay
system. The MDS sampling protocol employed in this
study did not succeed in collecting sufficient structure-
property information for certain types of solute-
membrane interaction properties. Certainly, the short
time scale prevented a reasonable estimation of diffu-
sion coefficients. However, the diffusion behavior is
likely to be explained largely by intramolecular descrip-
tors such as log P and molecular weight, and the
determination of diffusion coefficients by molecular
dynamics simulations might be unnecessary.
N,N-Dim et h yl-3-(4-h yd r oxyp h en yl)p r op yla m in e h y-
d r och lor id e (3):42 IR (KBr) 3160, 2980, 2680, 1610, 1010, 810
1
cm-1; H NMR (CDCl3) δ 1.95-2.03 (2H, m), 2.62 (2H, t, J )
7.5 Hz), 2.86 (6H, s), 3.07-3.11 (2 H, m), 6.73 (2 H, d, J )
8.48 Hz), 7.05 (2 H, d, J ) 8.40 Hz).
N,N-Dim eth yl-3-(3-p yr id yl)p r op yla m in e d ih yd r op er -
ch lor a te (4): IR (KBr) 2940, 1560, 680 cm-1; 1H NMR (CDCl3)
δ 2.12-2.20 (2H, m), 2.92 (6H, s), 2.99 (2H, t, J ) 7.87 Hz),
3.23-3.27 (2 H, m), 8.06 (1 H, dd, J 1 ) J 2 ) 5.83 Hz), 8.59 (1
H, d, J ) 8.13 Hz), 8.73 (1 H, d, J ) 5.69 Hz), 8.80 (1 H, s).
Meth od B (Sch em e 2): Gen er a l P r oced u r e for th e
Vilsm eier Rea ction of Keton es follow ed by Ca ta lytic
Hyd r ogen a tion . To anhydrous N,N-dimethylformamide (30.0
mL) was added dropwise phosphorus oxychloride (0.22 mol)
at 15-25 °C, and stirring was continued for 1 h at ambient
temperature. After addition of ketone (0.02 mol) to the
resulting mixture, the temperature was slowly raised to 40
°C. A solution of ketone (0.08 mol) in anhydrous N,N-
dimethylformamide (10 mL) was added while maintaining the
temperature at 70 °C. After 5 h at room temperature methanol
(500 mL) was added and then perchloric acid (25 mL, 70%) to
form the perchlorate recrystallized from methanol. The 3-chlo-
ropropeniminium salts were catalytically hydrogenated using
palladium-on-carbon and the resulting amines converted to the
hydrochlorides 1 and 4-17 according to method A.
Exp er im en ta l Section
Ch em istr y. Melting points were determined on the Re-
ichert microhotstage and are uncorrected. The 400-MHz proton
NMR spectra were recorded on a Bruker AM400 spectrometer,
and shifts are expressed in δ (ppm) with TMS as internal
standard. IR spectra were taken with a Perkin-Elmer FTIR
1600 spectrometer in KBr and are expressed in cm-1. Elemen-
tal analyses were performed from the Microlaboratory of Ilse
Beetz, Kronach, Germany. Analytical results were within 0.4%
of the theoretical values. Silicagel thin-layer chromatography
was performed on precoated plates Kieselgel 60F254 (E. Merck,
AG, Darmstadt, Germany). Extraction solvents were dried over
magnesium sulfate. Starting materials were commercially
available (Merck, Aldrich, Fluka), of the best grade, and used
without further purification.
Meth od A (Sch em e 1): Gen er a l P r oced u r e for th e
Wittig Rea ction of Su bstitu ted Ben za ld eh yd es follow ed
by Ca ta lytic Hyd r ogen a tion . To a stirred solution of N,N-
dimethylaminoethyltriphenylphosphonium bromide (40 mmol)
in dry tetrahydrofuran (150 mL) was added an equimolar
solution of 1.4 M n-butyllithium in n-hexane under N2 at 0
°C. After 30 min a solution of starting benzaldehyde (40 mmol)
was added slowly, and then the reaction mixture was stirred
overnight at 60 °C. The solution was acidified with 3 N
hydrochloric acid (60 mL) and tetrahydrofuran evaporated.
After extraction with toluene (twice with 100 mL) the water
phase was made alkaline with 3 N sodium hydroxide and
extracted with ether (three times with 50 mL). The N,N-
dimethylcinnamylamine was isolated and converted to the
hydrochloride in the usual manner, followed by recrystalliza-
tion from ethanol/ether.
To a stirred solution or suspension of the propene derivative
(30 mmol) in methanol (250 mL) was added 10% palladium-
on-carbon (2.0 g). The reaction mixture was held under a
positive pressure of hydrogen (300-400 kPa) on a Parr shaker
until the theoretical uptake of hydrogen had occurred. The
catalyst was removed by filtration, and the filtrate was
concentrated to dryness in order to give the crude product.
After addition of water (100 mL) the resulting mixture was
made alkaline using 3 N sodium hydroxide and subsequently
extracted three times with ether (50 mL). The combined
organic layer was washed with water and dried, and the
hydrochloride was precipitated by addition of ethereal hydro-
chloric acid, recrystallized twice from ethanol/ether.
3-(4-Biph en ylyl)-N,N-dim eth ylpr opylam in e h ydr och lo-
1
r id e (5): IR (KBr) 2940, 2440, 760 cm-1; H NMR (CDCl3) δ
2.22-2.30 (2H, m), 2.78-2.81 (8 H, m), 2.97-3.02 (2 H, m),
7.25-7.59 (9 H, m).
N,N-Dim eth yl-3-ph en ylpr opylam in e h ydr och lor ide (6):
41
IR (KBr) 2920, 2580, 1480, 700 cm-1 1H NMR (CDCl3) δ
;
2.18-2.26 (2H, m), 2.75 (2 H, t, J ) 2.70 Hz), 2.77 (6 H, d, J
) 4.90 Hz), 2.94-2.99 (2 H, m), 7.18-7.33 (5 H, m).
N,N-Dim eth yl-3-(4-m eth ylp h en yl)p r op yla m in e h yd r o-
ch lor id e (7):43 IR (KBr) 2950, 2490, 1150, 810 cm-1; 1H NMR
(CDCl3) δ 2.15-2.23 (2H, m), 2.32 (3 H, s), 2.70 (2H, t, J )
7.23 Hz), 2.77 (6H, d, J ) 4.87 Hz), 2.92-2.98 (2 H, m), 7.07
(2 H, d, J ) 8.04 Hz), 7.11 (2 H, d, J ) 7.35 Hz).
N,N-Dim et h yl-3-(4-et h ylp h en yl)p r op yla m in e h yd r o-
ch lor id e (8): IR (KBr) 2950, 2520, 1460, 760 cm-1; 1H NMR
(CDCl3) δ 1.23 (3H, t, J ) 7.62), 1.71-2.24 (2H, m), 2.62 (2 H,
q, J ) 7.62 Hz), 2.71 (2 H, t, J ) 7.24 Hz), 2.77 (6 H, d, J )
4.83 Hz), 2.93-2.98 (2 H, m), 7.09 (2 H, d, J ) 8.05 Hz), 7.14
(2H, d, J ) 7.89 Hz).
N,N-Dim eth yl-3-(4-n -p r op ylp h en yl)p r op yla m in e Hy-
d r och lor id e (9). Zinc powder (2.5 g) was added to 5 N
hydrochloric acid (250 mL) at 5 °C under stirring followed by
2 mL of the Vilsmeier reaction mixture. Alternating, this
procedure was continued avoiding too hazardous conditions.
After 3 h stirring under cooling the mixture was slowly
neutralized with sodium hydroxide and extracted three times
with ether (50 mL). The combined organic layer was washed
with water and dried, and the hydrochloride was precipitated
by addition of ethereal hydrochloric acid, recrystallized twice
from acetone/ether: IR (KBr) 2930, 2510, 1450, 1150, 810 cm-1
;
1H NMR (CDCl3) δ 0.94 (3 H, t, J ) 7.34 Hz), 1.62 (2 H, t, J )
7.47 Hz), 2.16-2.24 (2H, m), 2.56 (2 H, t, J ) 7.67 Hz), 2.71
(3 H, t, J ) 7.25 Hz), 2.77 (6 H, d, J ) 4.81 Hz), 2.93-2.98 (2
H, m), 7.08 (2 H, d, J ) 8.19 Hz), 7.12 (2 H, d, J ) 8.15 Hz).
N,N-Dim eth yl-3-(4-isop r op ylp h en yl)p r op yla m in e h y-
d r och lor id e (10): IR (KBr) 2950, 2510, 1460, 1160, 810 cm-1
;
1H NMR (CDCl3) δ 1.24 (6 H, d, J ) 6.95 Hz), 2.16-2.24 (2H,
m), 2.72 (2 H, t, J ) 7.26 Hz), 2.77 (6 H, d, J ) 4.84 Hz), 2.87
(1 H, dq, J ) 6.92 Hz), 2.94-2.99 (2 H, m), 7.10 (2 H, d, J )
8.04 Hz), 7.16 (2 H, d, J ) 8.04 Hz).
3-(4-Ch lor op h en yl)-N,N-d im eth ylp r op yla m in e h yd r o-
ch lor id e (1):41 IR (KBr) 2920, 2490, 1470, 1030, 820 cm-1
;
1H NMR (CDCl3) δ 2.17-2.24 (2H, m), 2.73 (2H, t, J ) 7.37
Hz), 2.79 (6H, d, J ) 4.91 Hz), 2.93-2.98 (2 H, m), 7.14 (2 H,
d, J ) 8.38 Hz), 7.27 (2 H, d, J ) 7.96 Hz).
3-(4-ter t-Bu tylp h en yl)-N,N-d im eth ylp r op yla m in e h y-
1
d r och lor id e (11): IR (KBr) 2950, 2580, 1480, 840 cm-1; H
NMR (CDCl3) δ 1.31 (9 H, s), 2.18-2.24 (2H, m), 2.71 (2 H, t,
J ) 7.28 Hz), 2.78 (6 H, d, J ) 4.92 Hz), 2.95-3.00 (2 H, m),
7.11 (2 H, d, J ) 8.25 Hz), 7.33 (2 H, d, J ) 8.31 Hz).
3-(4-Br om op h en yl)-N,N-d im eth ylp r op yla m in e h yd r o-
ch lor id e (2): IR (KBr) 3000, 2510, 1500, 1075, 820 cm-1; 1H

3886 J ournal of Medicinal Chemistry, 1999, Vol. 42, No. 19
Klein et al.
N,N-Dim et h yl-3-(4-m et h oxyp h en yl)p r op yla m in e h y-
d r och lor id e (12):44 IR (KBr) 2950, 2840, 2530, 1460, 1170,
1120, 820 cm-1; 1H NMR (CDCl3) δ 2.14-2.22 (2H, m), 2.69 (2
H, t, J ) 7.25 Hz), 2.77 (6 H, d, J ) 4.92 Hz), 2.92-2.97 (2 H,
m), 3.79 (3H, s), 6.85 (2 H, d, J ) 8.64 Hz), 7.10 (2 H, d, J )
8.61 Hz).
2.82 (6 H, d, J ) 3.35 Hz), 2.90 (2 H, t, J ) 7.48 Hz), 2.98-
3.05 (2 H, m), 7.41 (2 H, d, J ) 9.45 Hz), 8.81 (2 H, d, J ) 8.53
Hz).
3-P h en yl-N,N,N-t r im et h ylp r op yla m m on iu m Iod id e
(22).48 A mixture of 6 (6.75 g, 50 mmol), methyl iodide (43.2
g, 0.3 mol), and potassium carbonate (12 g) in water (80 mL)
was refluxed for 7 h. Excess of methyl iodide was evaporated
and the residue extracted three times with 50 mL of dichlo-
romethane. The combined organic layer was dried and evapo-
rated. The residue was recrystallized from ethanol/ether to give
22 (13.4 g, 88%) as colorless crystals: mp 175 °C (173 °C);28
IR (KBr) 3000, 1470, 720 cm-1; 1H NMR (CD3OD) δ 2.09-2.17
(2 H, m), 2.72 (2 H, t, J ) 7.57 Hz), 3.13 (9 H, s), 3.35-3.40 (2
H, m), 7.19-7.32 (5 H, m).
3-(4-Ch lor obip h en ylyl)-N,N-d im eth ylp r op yla m in e h y-
1
d r och lor id e (13): IR (KBr) 2950, 2470, 1110, 820 cm-1; H
NMR (CDCl3) δ 2.22 (2H, m), 2.79 (8 H, m), 2.97-3.03 (2H,
m), 7.26-7.53 (8 H, m).
N ,N -Dim e t h yl-3-(4’-m e t h o xy -4-b ip h e n yly l)p r op yl-
a m in e h yd r och lor id e (14): IR (KBr) 2940, 2830, 2460, 1490,
1
1240, 1020, 810 cm-1; H NMR (CDCl3) δ 2.23-2.29 (2H, m),
2.76-2.79 (8 H, m), 2.96-3.00 (3 H, m), 3.85 (3H, s), 6.98 (2
H, d, J ) 8.76 Hz), 7.23 (2 H, d, J ) 8.07 Hz), 7.49 (2 H, d, J
) 5.63 Hz), 7.52 (2 H, d, J ) 6.46 Hz).
N ,N -D im e t h y l-3-(4′-h y d r o x y -4-b ip h e n y ly l)p r o p y l-
a m in e Hyd r och lor id e (23). A mixture of the methyl ether
14 (4 g, 13 mmol) in concentrated hydrobromic acid (40 mL)
was refluxed for 2 h. After cooling, to the mixture was added
slowly water (100 mL), and the resulting mixture was alkalized
with concentrated ammonia to give pH 9. The resulting
solution was extracted three times with 75 mL of ether. The
combined organic layer was washed twice with water and
dried, and 120 mL of the solvent was evaporated. After
addition of ethereal hydrochloric acid the crude product was
recrystallized from ethanol/ether to give 23 as colorless
N ,N -D i m e t h y l-3-(4′-e t h o x y -4-b i p h e n y ly l)p r o p y l-
a m in e h yd r och lor id e (15): IR (KBr) 2960, 2530, 1480, 1250,
1160, 1050, 810 cm-1; 1H NMR (CDCl3) δ 1.44 (3H, t, J ) 6.98
Hz), 2.21-2.29 (2H, m), 2.76-2.78 (8 H, m), 2.96-3.01 (2 H,
m), 4.08 (2H, q, J ) 6.98 Hz), 6.97 (2 H, d, J ) 8.75 Hz), 7.23
(2 H, d, J ) 8.10 Hz), 7.50 (2 H, d, J ) 8.0 Hz), 7.51 (2 H, d,
J ) 8.0 Hz).
N,N-Dim eth yl-3-(â-n a p h th yl)p r op yla m in e h yd r och lo-
r id e (16):45 IR (KBr) 2940, 2440, 1470, 820 cm-1 1H NMR
;
crystals: IR (KBr) 3140, 2940, 2680, 1490, 1210, 800 cm-1
;
1H NMR (CD3OD) δ 2.02-2.10 (2H, m), 2.74 (2 H, t, J ) 7.53
Hz), 2.88 (6H, s), 3.12-3.31 (2 H, m), 6.84 (2 H, d, J ) 8.62
Hz), 7.27 (2 H, d, J ) 8.13 Hz), 7.42 (2 H, d, J ) 8.63 Hz), 7.49
(2 H, d, J ) 8.19 Hz).
(CDCl3) δ 2.28-2.35 (2H, m), 2.76 (6 H, d, J ) 4.68 Hz), 2.92
(2 H, t, J ) 7.2 Hz), 2.96-3.01 (2H, m), 7.27-7.81 (7H, m).
3-(3-Ch lor o-4-m et h oxyp h en yl)-N,N-d im et h ylp r op yl-
a m in e h yd r och lor id e (17): IR (KBr) 2950, 2840, 2600, 1500,
1
1060, 1020, 810 cm-1; H NMR (CDCl3) δ 2.14-2.22 (2H, m),
3-(3-Ch lor o-4-h yd r oxyp h en yl)-N,N-d im et h ylp r op yl-
a m in e Hyd r och lor id e (24): obtained following a procedure
similar to that for compound 23; IR (KBr) 3100, 2950, 1470,
2.68 (2 H, t, J ) 7.33 Hz), 2.77 (6 H, d, J ) 4.9 Hz), 2.95-2.98
(2H, m), 3.89 (3H, s), 6.87 (1 H, d, J ) 8.39 Hz), 7.08 (1 H, dd,
J 1 ) 2.12 Hz, J 2 ) 8.2 Hz), 7.18 (1 H, d, J ) 2.13 Hz).
1280, 1160, 1050, 810 cm-1 1H NMR (CD3OD) δ 1.95-2.03
;
(2H, m), 2.61 (2 H, t, J ) 7.57 Hz), 2.86 (6 H, s), 3.08-3.12
(2H, m), 6.85 (1 H, d, J ) 8.25 Hz), 7.01 (1 H, dd, J 1 ) 2.03
Hz, J 2 ) 8.2 Hz), 7.2 (1 H, d, J ) 1.97 Hz).
Meth od C (Sch em e 3): Gen er a l P r oced u r e for th e
P r ep a r a tion of 3,3-Dip h en yl-Su bstitu ted P r op yla m in es
18 a n d 19. 1,1-Diphenyl-substituted ethanol derivatives were
prepared by Grignard reactions according to standard proce-
dures. Vilsmeier formylation and catalytical hydrogenation
were carried out according to method B.
3-(4′-Meth oxy-4-biph en ylyl)-N,N,N-tr im eth ylpr opylam -
m on iu m Ch lor id e (25). Compound 25 was obtained following
a procedure similar to that for compound 22. The resulting
iodide (2.4 g, 6.0 mmol) was dissolved in methanol (40 mL),
and the solution was saturated with gaseous hydrochloric acid
and heated at 100 °C for 1 h in an open round-bottom flask.
The residue was recrystallized from ethanol/ether to give 25
(1.8 g, 94%) as colorless crystals: mp 137 °C; IR (KBr) 2940,
2830, 1490, 1030, 800 cm-1; 1H NMR (CD3OD) δ 2.12-2.20 (2
H, m), 2.75 (2 H, t, J ) 7.45 Hz), 3.13 (9 H, s), 3.36-3.39 (2 H,
m), 3.82 (3H, s), 6.99 (2 H, d, J ) 8.77 Hz), 7.31 (2 H, d, J )
8.05 Hz), 7.51 (2 H, d, J ) 8.40 Hz), 7.52 (2 H, d, J ) 8.00 Hz).
N,N-Dim et h yl-3,3-d ip h en ylp r op yla m in e h yd r och lo-
r id e (18):46 IR (KBr) 3010, 2340, 1460, 700 cm-1 1H NMR
;
(CDCl3) δ 2.61-2.67 (2H, m), 2.76 (6 H, d, J ) 5.15 Hz), 2.89-
2.94 (2 H, m), 4.00 (1 H, t, J ) 2.1 Hz), 7.19-7.33 (10 H, m).
3-(4-Biph en ylyl)-N,N-dim eth ylpr opylam in e h ydr och lo-
r id e (19): IR (KBr) 2910, 2510, 1460, 1150, 750, 680 cm-1; 1H
NMR (CDCl3) δ 2.65-2.71 (2H, m), 2.78 (6 H, d, J ) 4.82 Hz),
2.93-2.97 (2H, m), 4.05 (2 H, t, J ) 7.87 Hz), 7.21-7.56 (14
H, m).
3-N,N-Dim et h yl-3-(4-a m in op h en yl)p r op yla m in e H y-
d r och lor id e (26).42 To a stirred solution of 21 (2.4 g, 10 mmol)
in methanol (100 mL) was added 10% palladium-on-carbon (1.0
g). The reaction mixture was held under a positive pressure
N,N-Dim et h yl-2,3-d ip h en ylp r op yla m in e H yd r och lo-
r id e (20).47 20 was prepared from desoxybenzoin (19.6 g, 0.1
mol) according to method B and the reaction mixture heated
for 24 h at 60-70 °C. For the precipitation of the perchlorate,
2-propanol (400 mL) and perchloric acid (25 mL, 70%) were
added to the mixture. Catalytic hydrogenation of the resulting
iminium salt yielded 20 recrystallized twice from acetone/ether
(8.5 g, 86%) as colorless crystals: mp 167-169 °C; IR (KBr)
2950, 2520, 1460, 690 cm-1; 1H NMR (CDCl3) δ 2.52 (3 H, d, J
) 4.88 Hz), 2.62 (3 H, d, J ) 4.9 Hz), 2.95 (1 H, q, J ) 8.1 Hz),
3.09 (1 H, q, J ) 7.1 Hz), 3.28-3.35 (1H, m), 3.44 (1 H, dt),
3.58-3.65 (1H, m), 7.04-7.33 (10 H, m).
N,N-Dim eth yl-3-(4-n itr op h en yl)p r op yla m in e Hyd r o-
ch lor id e (21).42 To a solution of 6 (17 g, 0.1 mol) in concen-
trated sulfuric acid (40 mL) was added dropwise concentrated
nitric acid (6.6 g) at -10 °C. After stirring for 15 h, the mixture
was added to crushed ice (100 g) and subsequently alkalized
slowly using sodium hydroxide solution (50%). The mixture
was extracted three times with 50 mL of ether. The combined
organic layer was washed with water and dried, and the
hydrochloride was precipitated by addition of ethereal hydro-
chloric acid, recrystallized twice from ethanol/ether to give 21
(7.2 g, 28%) as yellow crystals: mp 154 °C; IR (KBr) 2980,
1610, 1490, 760 cm-1; 1H NMR (CDCl3) δ 2.27-2.31 (2 H, m),
of hydrogen (300-400 kPa) on
a Parr shaker until the
theoretical uptake of hydrogen had occurred. The catalyst was
removed by filtration, and the filtrate was concentrated to
dryness. To the residue was added water (100 mL); the
resulting mixture was made alkaline using 3 N sodium
hydroxide and subsequently extracted three times with ether
(50 mL). The combined organic layer was washed with water
and dried, and the hydrochloride was precipitated by addition
of ethereal hydrochloric acid, recrystallized twice from ethanol/
ether to give 22 (2.4 g, 80%) as colorless crystals: mp 250 °C
dec; IR (KBr) 2775, 2520, 1550, 1450, 1150, 810 cm-1; 1H NMR
(CD3OD) δ 2.03-2.11 (2 H, m), 2.77 (2 H, t, J ) 7.91 Hz), 2.89
(6 H, s), 3.15-3.19 (2 H, m), 7.36 (2 H, d, J ) 8.42 Hz), 7.46
(2 H, d, J ) 8.36 Hz).
QSAR An a lysis: Com p u ta tion a l Deta ils. 1. d p (ch a n ge
in m em br a n e d en sity u p on u p ta k e of th e solu te). The
Connolly volume for the lowest potential energy state of the
reference monolayer model (without any solute) was deter-
mined. A probe with a radius of 1.4 Å was employed. The
procedure was repeated for the lowest potential energy state
of each monolayer-solute system and the corresponding

Membrane-Interaction QSAR Analysis
J ournal of Medicinal Chemistry, 1999, Vol. 42, No. 19 3887
change in density was calculated as
However, these computational experiments revealed that
either the method of calculation could not be meaningfully
applied to the highly disordered membrane system (bulk
modulus) or the variances in the measures were not large
enough to promise any information in QSPR model construc-
tion. (The AM1 charges of the training set molecules and the
DPPC molecules are available from the principal author,
C.D.P.K.)
M(mem+sol)
Mmem
Vmem
dp )
-
(1)
V(mem+sol)
where Mmem is the mass of the model monolayer without the
solute, M(mem+sol) is the mass of the monolayer-solute system,
and V(mem+sol) and Vmem are the respective volumes.
2. D (solu te d iffu sion coefficien t). The mean-square
displacement method was used to calculate diffusion coef-
ficients. For each trajectory frame in the MDS trajectory, the
average movement ∆di of a solute molecule, relative to the
previous frame, was determined. The mean square distance
is defined as
Ack n ow led gm en t. This work was supported in part
(UIC staff) by the Procter & Gamble Co. Resources of
the Laboratory of Molecular Modeling and Design were
used to perform the modeling part of this study. The
stimulating discussions with Prof. U. Holzgrabe at the
University of Bonn are gratefully acknowledged. We
greatly appreciate computer support from Dr. J . Triebs
of 8810M55 Programming GmbH. C.K. acknowledges
financial support from the Konrad Adenauer-Stiftung.
n
∆d2i
∑
i)1
øj2 )
(2)
t
Refer en ces
(1) Girke, S.; Mohr, K.; Schrape, S. Comparison Between the
Activities of Cationic Amphiphilic Drugs to Affect Phospholipid-
Membranes and to Depress Cardiac Function. Biochem. Phar-
macol. 1989, 38, 2487-2496.
where t is the total simulation time and n denotes the number
of MDS frames. Using the Einstein49 diffusion equation, the
diffusion coefficient D can be calculated as
(2) Lu¨llmann, H.; Mohr, K. Drug Phospholipid Interactions. In
Metabolism of Xenobiotics; Gorrod, J . W., Oelschla¨ger, H.,
Caldwell, J ., Eds.; Taylor & Francis: London, 1988.
(3) Seydel, J . K.; et al. QSAR and Multivariate Data Analysis of
Amphiphilic Benzylamines and their Interaction with Various
Phospholipids Determined By Different Methods. QSAR 1989,
8, 266-278.
øj2
2t
D )
(3)
3.
d (a ver a ge d ep th of a solu te in th e m em br a n e).
For each trajectory frame, the distances beween the aromatic
carbon bearing the aminopropyl side chain of the solute and
five adjacent phospholipid phosphorus atoms were determined.
The trajectory average distance between the aromatic carbon
and a single phosphorus atom was computed as
(4) Lu¨llmann, H.; Plo¨sch, H.; Ziegler, A. Ca2+ Replacement by
Cationic Amphiphilic Drugs from Lipid Monolayers. Biochem.
Pharmacol. 1980, 29, 2969-2974.
(5) Borchardt, K.; Heber, D.; Klingmu¨ller, M.; Mohr, K.; Mu¨ller, B.
The Ability of Cationic Amphiphilic Compounds to Depress the
Transition Temperature of Dipalmitoylphosphatidic Acid Lipo-
somes Depends on the Spatial Arrangement of the Lipophilic
Moiety. Biochem. Pharmacol. 1991, 42 (Suppl.), S61-S65.
(6) Hanpft, R.; Mohr, K. Influence of Cationic Amphiphilic Drugs
on the Phase-Transition Temperature of Phospholipids with
Different Polar Headgroups. Biochim. Biophys. Acta 1985, 814,
156-162.
n
d(single)
)
pidi
(4)
∑
i)1
where pi is the thermodynamic probability of the ith state (see
below). The average of all five carbon-phosphorus trajectory
average distances was used as the average depth descriptor,
d . The thermodynamic probability of each state in the MDS
trajectory can be calculated as
(7) Klein, C.; Hopfinger, A. J . Pharmacological activity and mem-
brane interactions of antiarrhythmics: 4D-QSAR/QSPR analy-
sis. Pharmacol. Res. 1998, 15, 303-311.
(8) Bassolino, D.; Alper, T. R. Stouch Drug Membrane Interactions
Studied by Molecular Dynamics Simulation: Size Dependence
of Diffusion. Drug Des. Discovery 1996, 13, 135-141.
(9) J in, B. Q.; Hopfinger, A. J . Characterization of Lipid Membrane
Dynamics by Simulation: 3. Probing Molecular Transport Across
the Phospholipid Bilayer. Pharmacol. Res. 1996, 12, 1786-1794.
(10) Tieleman, D. P.; Marrink, S. J .; Berendsen, H. J . C. A computer
perspective of membranes: molecular dynamics studies of lipid
bilayer systems. Biochim. Biophys. Acta 1997, 1331, 235-270.
(11) Hopfinger, A. J .; Wang, S.; Tokarski, J . S.; J in, B.; Albuquerque,
M.; Madhav, P.; Duraiswami, C. Construction of 3D-QSAR
models using the 4D-QSAR analysis formalism. J . Am. Chem.
Soc. 1997, 119, 10509-10524.
e(-E /RT)
i
pi )
(5)
Q
where Ei denotes the potential energy of the ith state and Q
is the partition function, approximated from the MDS trajec-
tory total potential energies:
n
e(-E /RT)
(6)
i
Q )
(12) Kuroda, Y.; Fujiwara, Y. Locations and Dynamical Perturbations
for Lipids of Cationic Forms of Procaine, Tetracaine, and
Dibucaine in Small Unilamellar Phosphatidylcholine Vesicles
As Studied by Nuclear Overhauser Effects in 1H Nuclear
Magnetic Resonance Spectroscopy. Biochim. Biophys. Acta 1987,
903, 395-410.
(13) Mannich, C.; Heilner, G. Syntheses of â-Ketobases from Ac-
etophenone, Formaldehyde, and Amine Salts. Ber. Dtsch. Chem.
Ges. 1922, 55, 356-365.
(14) Mannich, C.; Chang, F. T. About the Synthesis of Organic Bases
with Threefold Binding. Chem. Ber. 1933, 66, 418-420.
(15) Yamada, S.; Kikugawa, Y.; Ikegami, S. Reduction of Acid Amides
to Amines with Sodium Borohydride. Chem. Pharm. Bull. 1965,
13, 394-402.
(16) Ison, R. R.; Hassan, M. M. A. Clemmensen Reduction of Mannich
Bases: Potential Ganglionic and Nicotinic Agents. J . Pharm.
Sci. 1973, 62, 1388-1390.
(17) Stu¨tz, A.; Georgopoulos, A.; Granitzer, W.; Petranyi, G.; Berney,
D. Synthesis and Structure-Activity Relationships of Naftifine-
Related Allylamine Antimycotics. J . Med. Chem. 1986, 29, 112-
125.
(18) Kindler, K.; Hedemann, B.; Scha¨rfe, E. Studies on the Mecha-
nism of Chemical Reactions. X. Phenyl- and Cyclohexylalkyl-
amines by Hydrogenation. J ustus Liebigs Ann. Chem. 1948, 560,
215-221.
∑
i)1
4. d S (ch a n ge in en tr op y of th e m em br a n e system
u p on u p ta k e of a solu te). The thermodynamic probability
of each state of the MDS trajectory was calculated for the
reference monolayer system (without solute) and the solute-
containing system. The change in entropy upon uptake of a
solute was computed as
n
R
dS ) -
[
(pi(mem+sol) ln pi(mem+sol) - pi(mem) ln pi(mem))]
∑
m
i)1
(7)
where m denotes the number of phospholipids in the system,
and pi(*) are the thermodynamic probabilities of the respective
systems for the ith state.
Additional membrane-interaction descriptors, namely the
change in bulk modulus, the cohesive energy density, the
change in free volume, and the free energy of transfer from
water to membrane, were calculated for several compounds.

3888 J ournal of Medicinal Chemistry, 1999, Vol. 42, No. 19
Klein et al.
(19) Marxer, A.; Leutert, T. The Aminoalkylation of Carbonyl Com-
pounds with Aid of the Wittig-Reaction. Helv. Chim. Acta 1978,
61, 1708-1716.
(20) Zemlicka, J .; Arnold, Z. Synthetic Reactions of Dimethylforma-
mide. XI. Polyformylation of Ketones of Other Types and the
Problem of the Reaction Course. Collect. Czech. Chem. Commun.
1961, 26, 2852-2864.
(21) Liebscher, J .; Hartmann, H. 3-Chloro-2-propeniminium Salts
(Vinylogous Amide Chloride) as Versatile Synthons in Organic
Chemistry. Synthesis 1979, 10, 241-264.
(22) Fischer, G. W. 1H NMR-Spectroscopic Studies on 3-Chlor-2-
propeniminium Salts. J . Prakt. Chem. 1989, 311, 111-114.
(23) Pearlstein, R. A. CHEMLAB-II Users Guide; CHEMLAB Inc.,
1780 Wilson Dr, Lake Forest, IL 60045; 1988.
(24) HyperChem, Release 4.5 for MS Windows; Hypercube Inc.,
Waterloo, Ontario, Canada; 1995.
(25) Hauser, H.; Pascher, I.; Pearson, R. H.; Sundell, S. Preferred
conformation and molecular packing of phosphatidylethanola-
mine and phosphatidylcholine. Biochim. Biophys. Acta 1981,
650, 21-51.
(26) Stewart, J . J . P. Mopac 6.0 Release Notes; Frank J . Seiler
Research Laboratory, United States Air Force Academy; 1990.
(27) Nagle, J . F. Area/lipid of bilayers from NMR. Biophys. J . 1993,
64, 1476-1481.
(28) J in, B. Q.; Hopfinger, A. J . Characterization of Lipid Membrane
Dynamics by Simulation: 1. Torsion Angle Motions of the Linear
Chains. Biopolymers 1997, 41, 37-50.
(29) J in, B. Q.; Hopfinger, A. J . Characterization of Lipid Membrane
Dynamics by Simulation: 2. Quantitative Representation of
Lipid Fluidity. Comput. Polym. Sci. 1996, 6, 95-101.
(30) Doherty, D. C. Molsim Version 3.0 User Guide; The Chem21
Group, Inc., Lake Forest, IL 60045; 1994.
(31) Berendsen, H. J . C.; Postma, J . P. M.; van Gunsteren, W. F.;
DiNola, A.; Haak, J . R. Molecular dynamics with coupling to an
external bath. J . Chem. Phys. 1984, 81, 3684-3590.
(32) Cerius2, V.3.0 User Guide; Molecular Simulations Inc., San
Diego, CA; 1997.
(33) Stanton, D. T.; J urs, P. C. Development and Use of Charged
Partial Surface Area Structural Descriptors in Computer-
Assisted Quantitative Structure-Property Relationship Studies.
Anal. Chem. 1990, 62, 2323-2329.
(34) Rogers, D. Wolf 6.2 GFA Program; Molecular Simulations Inc.:
1994.
(35) Glen, W. G.; Dunn, W. J ., III; Scott, D. R. Principal Components
Analysis and Partial Least Squares Regression. Tetrahedron
Comput. Methodol. 1989, 2, 349-376.
(36) Rogers, D.; Hopfinger, A. J . Application of Genetic Function
Approximation to Quantitative Structure-Activity Relationships
and Quantitative Structure-Property Relationships. J . Chem.
Inf. Comput. Sci. 1994, 34, 854.
(37) Friedman, J . Multivariate Adaptive Regression Splines; Techni-
cal Report No. 102; Laboratory for Computational Statistics,
Department of Statistics, Stanford University; 1988.
(38) Gasteiger, J .; Marsili, M. Iterative partial equalization of orbital
electronegativity - a rapid access to atomic charges. Tetrahedron
1998, 36, 3219-3228.
(39) Hopfinger, A. J . Conformational Properties of Macromolecules;
Academic Press: New York, London, 1973.
(40) Rauls, D. O.; Baker, J . K. Relationship of nonspecific antiar-
rhythmic and negative inotropic activity with physicochemical
parameters of propranolol analogues. J . Med. Chem. 1979, 22,
81-86.
(41) McMahon, R. E. Demethylation Studies I. The Effect of Chemical
Structure and Lipid Solubility. J . Med. Pharm. Chem. 1961, 4,
67-74.
(42) van Braun, J .; Deutsch, H. Syntheses in the Fatty Aromatic
Series (VII. Phenolic Bases). Ber. Dtsch. Chem. Ges. 1912, 45,
2504-2522.
(43) Brown, B. G.; Hey, P. Choline Phenyl Ethers as Inhibitors of
Amine Oxidase. Br. J . Pharmacol. 1956, 11, 58-64.
(44) Kindler, K.; Burghard, F. Reactions of Aromatic Carbaldehydes
to â-Arylethylamines. Arch. Pharm. (Berlin) 1927, 265, 396-
397.
(45) Chandross, E. A.; Thomas, H. T. Intramolecular Exciplex
Formation in Naphthylalkylamines. Chem. Phys. Lett. 1971, 9,
393-397.
(46) Ruddy, A. W. Antispamodics II Substituted ?-(â-Aminoethyl)-
phenylacetonitriles and 3-Phenylpropylamines. J . Am. Chem.
Soc. 1951, 73, 4096-4098.
(47) Mattocks, A. M.; Hutchison, O. S. Sympathomimetics: 2,3-
Diphenylpropylamines. J . Am. Chem. Soc. 1948, 70, 3516-3517.
(48) Fry, J . L.; Silverman, S. B.; Orfanopoulos, M. Reduction of
Ketones to Hydrocarbons with Triethylsilane: m-Nitroethyl-
benzene. Org. Synth. 1981, 60, 108-112.
(49) Einstein, A. Investigations on the theory of the brownian move-
ment; Methuen & Co. Ltd.: London, 1926.
J M980694A