3D-QSAR study of bis-azaaromatic quaternary ammonium analogs at the blood-brain barrier choline transporter

Article abstract of DOI:10.1016/j.bmc.2005.04.020

Previously, we have developed 3D-QSAR models of the blood-brain barrier (BBB) choline transporter, a transport system that may have utility as a vector for central nervous system drug delivery. In this study, we extended the model by evaluating five bis-azaaromatic quaternary ammonium compounds for their affinity for the choline binding site on the BBB-choline transporter. The compounds, and their affinities for the transporter, were then incorporated into our existing molecular model, in order to update our knowledge on the molecular recognition factors associated with interaction of ligands at the choline binding site. The current compounds are structurally related to previous substrates that we have evaluated, but offer additional three dimensional aspects compared to the series of compounds previously utilized to define the original models. The compounds showed good affinity for the BBB-choline transporter, exhibiting inhibition constants ranging from 10 to 68 μM, as determined by the in situ rat brain perfusion method. Comparative molecular field analysis (CoMFA) and comparative molecular similarity index analysis (CoMSIA) methods were used to build the new 3D QSAR models. When the new bis-azaaromatic quaternary ammonium compounds were included in the model, the best cross-validated CoMFA q² was found to be 0.536 and the non-cross-validated r² was 0.818. CoMSIA hydrophobic cross-validated q² was 0.506 and the non-cross-validated r² was 0.804. This new model was able to better predict BBB-choline transporter affinity of hemicholinium-3 (predicted 65 μM, actual 54 μM), when compared to an earlier model (predicted 316 μM).

Full text of DOI:10.1016/j.bmc.2005.04.020

Bioorganic & Medicinal Chemistry 13 (2005) 4253–4261
3
D-QSAR study of bis-azaaromatic quaternary ammonium
analogs at the blood–brain barrier choline transporter
a
Werner J. Geldenhuys, Paul R. Lockman, Tiffany H. Nguyen,
a
a
a,b
Cornelis J. Van der Schyf, Peter A. Crooks, Linda P. Dwoskin and David D. Allen
c
c
a,
*
a
Department of Pharmaceutical Sciences, Texas Tech University Health Sciences Center, School of Pharmacy,
1
Pharmaceutical Chemistry, North West University, Potchefstroom, 2520, South Africa
300 S. Coulter Dr., Amarillo, TX, 79106-1712, USA
b
c
Department of Pharmaceutical Sciences, College of Pharmacy, University of Kentucky, Lexington, KY 40536-0082, USA
Received 18 January 2005; revised 9 April 2005; accepted 11 April 2005
Available online 4 May 2005
Abstract—Previously, we have developed 3D-QSAR models of the blood–brain barrier (BBB) choline transporter, a transport sys-
tem that may have utility as a vector for central nervous system drug delivery. In this study, we extended the model by evaluating
five bis-azaaromatic quaternary ammonium compounds for their affinity for the choline binding site on the BBB-choline trans-
porter. The compounds, and their affinities for the transporter, were then incorporated into our existing molecular model, in order
to update our knowledge on the molecular recognition factors associated with interaction of ligands at the choline binding site. The
current compounds are structurally related to previous substrates that we have evaluated, but offer additional three dimensional
aspects compared to the series of compounds previously utilized to define the original models. The compounds showed good affinity
for the BBB-choline transporter, exhibiting inhibition constants ranging from 10 to 68 lM, as determined by the in situ rat brain
perfusion method. Comparative molecular field analysis (CoMFA) and comparative molecular similarity index analysis (CoMSIA)
methods were used to build the new 3D QSAR models. When the new bis-azaaromatic quaternary ammonium compounds were
2
2
included in the model, the best cross-validated CoMFA q was found to be 0.536 and the non-cross-validated r was 0.818. CoMSIA
2
2
hydrophobic cross-validated q was 0.506 and the non-cross-validated r was 0.804. This new model was able to better predict
BBB-choline transporter affinity of hemicholinium-3 (predicted 65 lM, actual 54 lM), when compared to an earlier model
(
predicted 316 lM).
Ó 2005 Elsevier Ltd. All rights reserved.
1
. Introduction
such as D,L-2-amino-7-bis[(2-chloroethyl)amino]-1,2,3,4-
tetrahydro-2-naphthoic acid (D,L-NAM), a treatment
3
,4
The blood–brain barrier (BBB) is a major obstacle for
the entry of drugs into the central nervous system
CNS). About 98% of potential CNS-active drugs are
excluded from the brain by the BBB. A viable alterna-
tive option to structural modification of a drug to facili-
tate brain entry is to utilize native nutrient transporters
located at the barrier interface to bind to, and transport
such drugs into the CNS. Thus, drugs can be designed
to have affinity for a BBB transporter, enabling rela-
tively polar, water-soluble CNS-active drugs to enter
for brain tumors that was modified to use the BBB
large neutral amino acid transport system. Another example
is the neuronal nicotinic receptor antagonist N-n-octyl-
nicotinium iodide (NONI, Fig. 1), which has been
shown to bind the BBB-choline transporter with an
(
1
affinity comparable (K ꢀ 49 lM) to that of choline
i
5
(K ꢀ 45 lM).
m
2
Furthermore, NONI was demonstrated to be trans-
ported by the BBB-choline transporter, which confirmed
that this transport system has utility as a brain delivery
vector. In the absence of an X-ray crystal structure for
1
the brain from the periphery. This method has been
successfully employed in early studies for compounds
6
the human BBB-choline transporter, we have previ-
ously developed a comparative molecular field analysis
Keywords: Blood–brain barrier; Choline; Brain drug delivery; Trans-
port; Molecular modeling.
(CoMFA) model, to yield further insight into the
requirements for ligand-transporter binding.
7,8
The
results of this study generated the model for binding
*
Corresponding author. Tel.: +1 806 356 4015x286; fax: +1 806 356
034; e-mail: david.allen@ttuhsc.edu
4
0
968-0896/$ - see front matter Ó 2005 Elsevier Ltd. All rights reserved.
doi:10.1016/j.bmc.2005.04.020

4254
W. J. Geldenhuys et al. / Bioorg. Med. Chem. 13 (2005) 4253–4261
Figure 1. Structure of NONI, a quaternary ammonium compound
that has affinity for the BBB-choline transporter. NONI has been
shown to be taken up into the brain by the BBB-choline transporter,
demonstrating the use of this transporter as a drug delivery vector.
Figure 3. Structures of bis-azaaromatic quaternary ammonium com-
Figure 2. Theoretical rendering of the active binding site of the BBB
choline transporter.
2,7,8
pounds used in this study.
2
,7,8
interaction illustrated in Figure 2 below.
Structure–
some of these compounds are similar to that of NONI,
they are of interest in expanding our understanding of
the nature of the pharmacophoric requirements of the
BBB-choline transporter, since the charged quaternary
ammonium moiety largely prevents passive uptake into
the brain due to the BBBÕs selective permeability.
activity relationships suggested the need for a quater-
nary ammonium group, a free hydroxyl group, and
an intramolecular distance between the quaternary
ammonium nitrogen and the hydroxyl group of
˚
ꢀ
3.3 A. Surprisingly, it was also shown that bis-quater-
nary ammonium compounds lacking a hydroxyl group
9
also inhibit choline uptake into brain synaptosomes.
In the present study, we supplemented and refined the
predictive utility of our earlier BBB-choline transporter
3D-QSAR model by adding the novel bis-azaaromatic
quaternary ammonium compounds to the equation.
We propose that information gleaned from this study
provides a better understanding of how the rational
incorporation of specific structural moieties into CNS-
active drug molecules may also allow them to act as sub-
strates for the BBB-choline transporter. Additionally,
the incorporation of these bis-azaaromatic quaternary
ammonium compounds into our model may provide
us with a better understanding of why these compounds
have affinity for the BBB-choline transporter, even
though they lack a free hydroxyl group, which was pre-
In this report, we discuss the exploitation of the BBB-
choline transporter as a drug delivery vector for bis-
compounds containing two charged quaternary ammo-
5
,7,8
nium moieties in their molecular structure.
The
BBB-choline transporter is responsible for transporting
choline, a charged cation, into the CNS, where it acts
as a precursor for the neurotransmitter acetylcholine
and also as an essential component of membrane phos-
1
0
pholipids such as phosphatidylcholine.
Novel N-n-alkylnicotinium analogs have been demon-
1
1,12
strated to modulate the dopaminergic system
by
9
inhibiting nicotine-evoked dopamine overflow from rat
striatal slices. In addition, these compounds are reported
to cross the BBB by acting as substrates for the BBB-
choline transporter. A series of bis-azaaromatic quater-
nary ammonium compounds (Fig. 3) have recently been
synthesized and evaluated for CNS activity.
compounds were shown to be ligands for neuronal nico-
tinic receptors, with suggested possible utility in Alz-
heimerÕs and ParkinsonÕs disease, and smoking cessation
therapies. Because the pharmacological profiles of
viously thought to be necessary for high affinity.
5
2. Methods
1
3–15
These
2.1. Chemistry
1
3
11
NONI was prepared as previously described. The bis-aza-
aromatic quaternary ammonium compounds, bIQHxI,
bIQDI, bIQDDB, bQHxI, and bPiDDB, were pre-
1
3

W. J. Geldenhuys et al. / Bioorg. Med. Chem. 13 (2005) 4253–4261
4255
1
3
pared by the procedures described by Ayers et al.
ated at concentrations of 250 lM, as described previ-
ously and briefly detailed below.
1
9
Briefly, the compounds were synthesized by reacting
an appropriate azaaromatic compound with a variety
of diiodo- or dibromoalkanes for 24 h at 65 °C. Synthe-
sis, structure determination using H and C NMR
spectroscopy and mass spectroscopy, and purity deter-
mination by elemental analysis were carried out as previ-
Rats were decapitated and brain samples obtained as de-
16
scribed previously. Briefly, the brain was removed
1
13
from the skull, the arachnoid membrane and meningeal
vessels were removed, and the perfused cerebral hemi-
sphere dissected on ice. Brain regions were placed in
scintillation vials and weighed. Two 50 ll aliquots of
the perfusion fluid were also transferred to the individ-
ual scintillation vials and weighed. Brain and perfusion
fluid samples were digested overnight at 50 °C in 1 ml
piperidine (1 M) solution. Fisher chemical scintillation
cocktail (10 ml; Beckman, Fullerton, CA) was added to
each vial and the tracer contents assessed by dual-label
liquid scintillation counting, and corrected for quench,
background, and efficiency.
1
3
ously described. The structures of all compounds were
consistent with the spectroscopic data.
2
.2. In situ brain perfusion technique
3
Brain uptake of [ H]-choline was assessed using a modi-
1
6–18
fied in situ rat brain perfusion technique. Briefly,
for this study, short perfusions (60 s) were used to deter-
3
mine [ H]-choline brain uptake. Once uptake values
were estimated, subsequent experiments evaluated the
3
inhibition of [ H]-choline brain uptake by 250 lM con-
3
Brain [ H]-choline uptake was determined by a single
centrations of each of the test compounds. This concen-
tration is our standard starting point for compound
screening, because it represents approximately five times
time point blood-to-brain transfer coefficient (K ) calcu-
in
1
6,20
lation as previously described,
relationship:
using the following
1
0
the km of choline for the transporter.
K
in ¼ ½Ctot ꢂ V
v
C
pf ꢃ=ðCpf =T Þ ð1Þ
Male Fischer-344 rats (220–250 g) were obtained from
Charles River Laboratories, Inc. (Wilmington, MA).
All studies were conducted in accordance with the
NIH Guide for the Care and Use of Laboratory Ani-
mals, and were approved by the Institutional Animal
Care and Use Committee of Texas Tech University
Health Sciences Center. Rats were anesthetized with so-
dium pentobarbital (50 mg/kg intraperitoneal) by a PE-
where Ctot = Cbr + Cvas is the sum amount of choline
contained in the perfusate in the blood–brain vessels
(Cvas) and the amount of choline that penetrated into
the brain (C ). Cerebral vascular volume (V ) and per-
br
v
fusion flow rate were estimated in separate experiments
2
as previously described. C is the tracer choline perfu-
pf
sion fluid concentration, and T is the net perfusion time
assumed to be under linear uptake conditions.
6
0 catheter containing heparinized saline (100 U/ml),
which was inserted into the left common carotid artery
after ipsilateral external carotid, occipital, and common
carotid artery ligation (the latter ligation was caudal to
the catheter implantation). The pterygopalatine artery
Apparent cerebrovascular permeability surface-area
products (PA) were determined using the following
relationship:
1
0
was patent during these experiments. Rat body tem-
perature was evaluated and maintained at 37 °C by a
feedback-controlled heating pad (YSI Indicating Con-
troller, Yellow Springs, Ohio). The left common carotid
artery catheter was connected to a syringe, which con-
tained physiologic buffered perfusion fluid (containing
PA ¼ F lnð1 ꢂ k =F Þ
ð2Þ
In this relationship, cerebral blood flow (F) was assessed
in
1
4
for the brain region of interest from uptake of [ C]-
1
6
diazepam, as described earlier.
Apparent inhibition constants (K ) for bis-azaaromatic
i
[
in mM]: NaCl 128, NaPO 2.4, NaHCO 29.0, KCl
3
3
4
.2, CaCl 1.5, MgCl 0.9, and D-glucose 9) with 1
2
quaternary ammonium compounds were determined as
19
previously described, from the equation:
3
14
lCi/ml [ H]-choline and 0.3 lCi/ml [ C]-sucrose (to
determine vascular volume). Perfusion fluid was filtered,
warmed to 37 °C, and gassed with 95% O and 5%
½ðPA ꢂ K Þ=ðPA ꢂ K Þꢃ ¼ 1 þ C =K
ð3Þ
2
o
D
i
D
i
i
CO . Just before perfusion, pH and osmolarity of this
solution were ꢁ7.35 and 290 mOsm, respectively. Perfu-
2
Competition kinetics were assumed, and from this rela-
tionship, PA represents the [ H]-choline PA in the ab-
3
sion fluid was infused into the left carotid artery at
10 ml/min for 60 s using a Harvard Apparatus infusion
pump (South Natick, MA). This perfusion rate main-
o
3
sence of competitor. PA represents the [ H]-choline
i
PA in the presence of the inhibitor and C represents
i
1
0,16
tains a carotid artery pressure of ꢀ120 mmHg.
After
0 s perfusion with [ H]-choline (in the absence and
the perfusate concentration of inhibitor. Apparent Ki
is defined as the inhibitor concentration that reduces sat-
3
6
3
urable brain [ H]-choline influx by 50% at tracer choline
presence of the unlabeled compound of interest), perfu-
sion fluid was switched to contain tracer-free perfusion
fluid for 15 s to wash out residual [ H]-choline loosely
concentration (C ꢄ K ), in the absence of competing
pf
m
3
compounds. We were able to demonstrate competition
at the BBB-choline transporter over a 0.25–12.5 lM
concentration range of hemicholinium-3, the defining
1
8
bound to the cerebrovasculature.
3
18
Inhibition of [ H]-choline brain uptake was ascertained
by addition of the appropriate unlabeled bis-azaaro-
matic quaternary ammonium analog into the perfusion
substrate for choline transport. This range has previ-
ously demonstrated that competition kinetics occur at
the BBB-choline transporter and provides the basis for
1
8
fluid. To determine K values, compounds were evalu-
i
evaluating K values for the inhibitors. Additionally,
i

4256
W. J. Geldenhuys et al. / Bioorg. Med. Chem. 13 (2005) 4253–4261
previous studies indicate that monocations such as
3.1. Radiochemicals
1
8
sodium have little effect on choline transport. Further-
more, while the concentrations of the compounds uti-
lized in the present study were used solely to evaluate
their effects on BBB choline uptake, it should be pointed
out that therapeutic concentrations of the same com-
pounds would likely be much lower, due to their effects
3
14
[ H]-Choline (79.2 Ci/mmol, >98% purity) and [ C]-
sucrose (4.75 mCi/mmol) were obtained from Dupont-
New England Nuclear (Boston, MA). Prior to each
3
perfusion, [ H]-choline was thoroughly dried before
being dissolved in perfusion buffer to ensure that volatile
1
5
3
tritium contaminants (such as [ H]–H O) were not present.
on brain dopamine systems.
2
3
. Molecular modeling
4. Results and discussion
Molecular modeling was accomplished as described pre-
8
An established approach for augmenting brain drug
delivery is to utilize native BBB saturable nutrient trans-
porters, such as has been demonstrated with the BBB-
viously. Briefly, molecules were built using SYBYL 6.91
(
Tripos Inc., St. Louis, MO) molecular modeling soft-
2
,5
ware on a Silicon Graphics Octane computer. All mole-
cules were energy minimized for CoMFA and CoMSIA
analysis by iteratively searching for lowest energy con-
formations by rotating bonds in 10° increments followed
by energy minimization. All compounds were aligned
manually using the FIT ATOMS function in SYBYL,
by superimposing the cationic N-atom, the a-carbon
adjacent to the nitrogen atom and the hydroxyl group.
CoMFA analyses were carried out using a standard SY-
choline transporter. In this report, we evaluated how
the inclusion of a new series of bis-azaaromatic quater-
nary ammonium compounds would influence our earlier
3D-QSAR model of the BBB-choline transporter. This
study was initiated to further understand how this trans-
port system could serve as a vector for CNS drug deliv-
ery. Additionally, these bis-azaaromatic quaternary
ammonium compounds may provide insight as to why
they show high affinity for the BBB-choline transporter,
even though they lack a free hydroxyl group in their
structure. The bis-azaaromatic quaternary ammonium
compounds were evaluated for their ability to block
˚
BYL/QSAR routine. CoMFA grid spacing was 2.0 A in
the X, Y, and Z directions, and grid regions were auto-
matically generated by the CoMFA routine to encom-
3
˚
pass all molecules with an extension of 4.0 A in each
[ H]-choline uptake into brain via the BBB-choline
transporter, utilizing the in situ brain perfusion
3
direction. An sp carbon for measuring steric interac-
1
0
tions and a point charge of +1.0 to explore electrostatics
were used as probes to generate the interaction energies
at each lattice point, with a default energy cutoff at
technique. Inhibition constants (logK ) estimated for
i
these compounds are given in Table 1.
3
0 kcal/mol. Partial least squares (PLS, cross-validated
CoMFA and CoMSIA techniques were employed to
gain insight into the structural requirements for these
compounds to bind to the transporter, as well as to de-
rive extended predictive models for future drug design.
Results of CoMFA and CoMSIA analyses are shown
in Tables 2 and 3, respectively. Two models are pre-
and non-cross-validated) regression analyses were per-
formed from the resulting field. With cross-validation,
one compound is taken out of the group, and the
remaining compounds are then used to generate a new
model. This model is then used to predict the biological
2
8
activity of the deleted compound. This value (q ) differs
sented in this study: (1) our current model (Model 1)
and (2) a new model with the bis-azaaromatic quater-
nary ammonium compounds included (Table 2). For
2
2
from the normal regression value (r ), in that the q va-
lue can be negative. Column filtering (2.0 kcal/mol) was
also investigated to see if such an approximation would
8
2
the first model, a cross-validated q of 0.406 was esti-
2
result in improved q values. In PLS, each compound is
mated, which improved when column filtering of
2.0 kcal/mol was applied, to give a final q of 0.408 (Ta-
2
described by different dimensions, denoted principal
components, and PLS attempts to then identify compo-
nents that are relevant in interactions between the
2
ble 2). A non-cross-validated r of 0.943 was determined
for this model. Steric and electrostatic contributions
were calculated to be 72% and 28%, respectively, sup-
porting our earlier findings suggesting that the steric
properties of the cationic compounds are of paramount
importance in transporter binding.
biological parameter (e.g., logK ) and molecular descrip-
i
tors (e.g., steric or hydrophobic fields). The number of
components was chosen such that the smallest standard
error of prediction was achieved, where these compo-
nents describe the degree of complexity of the model.
Usually, the minimum number is chosen, since adding
more components may lead to over-fitting of the data,
which results in low predictability. These relative contri-
butions are represented as a 3D coefficient map illustrat-
ing the favored effects [80%; steric (in green) and
electrostatic (in blue)] and the unfavorable effects
3
Table 1. Inhibition of brain [ H]-choline uptake, logK
i
(lM) ± stan-
dard error of the mean (SEM) values obtained from in situ brain
9
perfusion studies for bis-azaaromatic quaternary ammonium com-
pounds used in this study
Compound
i
logK (lM) ± SEM
[
20%; steric (in yellow) and electrostatic effects (in
red)]. The green map areas indicate where sterically
bulky groups may enhance binding affinity. Blue colored
regions indicate where electronegative groups will en-
hance binding affinity. LogK values were used as the
i
biological descriptor in the SYBYL spreadsheets.
bPiDDB
bQHxI
1.57 ± 0.48
1.83 ± 1.18
1.46 ± 0.70
0.98 ± 0.26
1.35 ± 0.18
bIQHxI
bIQDI
bIQDDB

W. J. Geldenhuys et al. / Bioorg. Med. Chem. 13 (2005) 4253–4261
4257
Table 2. Summary of CoMFA partial least square (PLS) analysis of the original model (Model 1) and the new model with the bis-azaaromatic
quaternary ammonium compounds (Model 2) included in the training sets
2
q
2
r
Model
S.E.P.
Components
S.E.E.
Steric
Electrostatic
a
a
1
2
0.408
0.536
0.875
0.881
2
2
0.943
0.818
0.290
0.130
0.717
0.637
0.283
0.363
a
Column filtering of 2 kcal/mol; S.E.P. = standard error of prediction; S.E.E. = standard error of estimate.
Table 3. Summary of CoMSIA (hydrophobic) PLS analysis of the
original model (Model 1) and the new model with the bis-azaaromatic
quaternary ammonium compounds (Model 2) included in the training
sets
2
q
2
Model
S.E.P.
Components
R
S.E.E.
1
2
0.433
0.506
0.864
0.819
1
3
0.883
0.804
0.416
0.531
S.E.P. = standard error of prediction; S.E.E. = standard error of
estimate.
Figure 6. Contour maps representing the new CoMFA analysis
Model 2). The bis-azaaromatic quaternary ammonium compound
(
bPiDDB (one of the new compounds) is shown docked into the
binding site.
aromatic quaternary ammonium compounds included,
and the original model (Model 1) for comparison. The
original CoMFA model was used both for comparison,
and to investigate its robustness to predict a 3D-QSAR
for new, chemically diverse compounds. Most 3D-
QSAR training (and test) sets contain compounds with
a similar substructure, with side chain replacement as
the major variant among the compounds. The structural
models investigated herein display a variety of structural
features in which even the substructures differ consider-
ably. These range from ring structures to alkyl
compounds.
Figure 4. Contour maps of the CoMFA analysis for the original
training set without inclusion of the bis-azaaromatic quaternary
ammonium compounds (Model 1). Green areas indicate regions where
bulky substituents can be accommodated in a sterically favorable way,
and yellow, in an unfavorable way. Electrostatic fields are shown
where blue areas would favorably accommodate cationic groups.
Cationic interactions that would be unfavorable are shown in the red
areas. The docking of choline into the binding site is shown in the
The green contours in Figures 4–6 represent high steric
tolerance regions (80% contribution) while yellow
contours represent regions of unfavorable steric interac-
tion (20% contribution). Electrostatic contours are
shown in blue and red. Blue contours represent regions
where positively charged groups will enhance activity
Figures 4–6 show the resulting CoMFA contour plots
representing the new model (Model 2) with the bis-aza-
Figure 5. Contour maps of the new CoMFA analysis (Model 2), which include the bis-azaaromatic quaternary ammonium compounds in the
training set. The steric and electrostatic fields are shown separated. Green areas indicate regions where bulky substituents can be accommodated in a
sterically favorable way, and yellow, in an unfavorable way. Electrostatic fields are shown where blue areas would favorably accommodate cationic
groups. Cationic interactions that would be unfavorable are shown in the red areas. The docking of choline into the binding site is shown in the above
model.

4258
W. J. Geldenhuys et al. / Bioorg. Med. Chem. 13 (2005) 4253–4261
Figure 7. Contour maps of the CoMSIA hydrophobic fields with (A) the original model (Model 1), and (B) the new model (Model 2) with the bis-
azaaromatic quaternary ammonium compounds included in the training set. Areas that favor hydrophobic interactions are indicated by yellow, and
areas that are unfavorable for hydrophobic interactions are indicated in white. Choline docked into the binding site is illustrated in the left panel (A)
and bPiDDB, a representative of the bis-azaaromatic quaternary ammonium compounds, is shown docked into the binding site in the right panel (B).
3
predicted and actual brain [ H]-choline uptake inhibi-
(
80% contribution), whereas red regions denote where
negatively charged groups will enhance activity (20%
contribution). Green areas surround the alkyl chain
segment of choline, as well as an area near the N-methyl
groups (Fig. 5). The latter areas indicate where bulky
groups would sterically complement the interaction
between a ligand and the choline transporter binding
site. The same green ÔbandÕ is also seen in Figure 5,
suggesting a commonality between these two models.
Figures 4 and 5 show a large (ÔfavorableÕ) blue area
surrounding the cationic nitrogen, as would be expected
for a cationic transporter, and indicates a region where
positively charged substituents would be predicted to
improve affinity. Figure 6 shows the new CoMFA model
tion (i.e., logK s, predicted logK > 4) for the new bis-
i
azaaromatic quaternary ammonium compounds.
i
The bis-azaaromatic quaternary ammonium compounds
differ substantially from the training set compounds
used in the previous study to create Model 1. First,
the compounds used for creating Model 2 contain two
positively charged centers, as opposed to one positively
charged center in the molecules used in the original
training set for Model 1. Second, the bis-quaternary
ammonium compounds are devoid of the hydroxyl
group present in the choline molecule, which is promi-
nent in a large proportion of the compounds in the ori-
ginal training set for Model 1. The hydroxyl group is
thought to contribute to the high affinity of molecules
for the choline binding site in the Model 1 approxima-
tion (Fig. 2).
(
Model 2) with bPiDDB docked into the binding site.
Figure 7 shows the results of CoMSIA hydrophobic
field analyses. Favorable hydrophobic fields are indi-
cated by yellow contours and unfavorable regions are
indicated by white contours. The original model (Model
8
Two possible explanations could account for these
observations. First, one of the cationic moieties may ini-
tially bind to the anionic site on the BBB-choline trans-
porter, anchoring the molecule to the transporter;
subsequently, the second cationic moiety may then bind
to a second anionic site on the transporter, which is dif-
ferent from the hydrogen bonding site in the original
model. Because of the high degree of flexibility and the
length of the tethered n-alkyl chain in the bis-azaaro-
matic compounds, it is likely that the anchored molecule
can adopt a large number of conformations. For exam-
ple, two extreme conformations of the bis-azaaromatic
compounds that may be involved in binding to the
transporter are: a fully extended conformation, where
the two cationic moieties are furthest apart and interact
with two separate anionic binding sites; or alternatively,
a folded or hairpin conformation where both cationic
moieties interact with one anionic binding site, or with
two closely juxtaposed binding sites on the transporter.
Of course, other possible conformations of these flexible
molecules may also be recognized by the transporter
binding site. The n-alkyl linker may serve a secondary
role in the binding process by interacting with hydro-
phobic residues adjacent to the anionic binding site(s)
lining the transporter cavity. Secondly, it is possible that
the bis-azaaromatic compounds interact with a dimeric
complex of the choline transporter by spanning the dis-
1
, Fig. 7A) shows primarily two areas surrounding the
cationic quaternary ammonium moiety of choline. The
white areas show where more hydrophilic substituents
in the structure of the compounds would likely increase
affinity for the choline transporter, while yellow areas
show where the addition of hydrophobic substituents
may increase affinity for the binding site. The new model
(
Model 2, Fig. 7B) with the bis-azaaromatic quaternary
ammonium compounds included in the modeling, shows
similar contour maps with the yellow regions surround-
ing the cationic quaternary ammonium group in
bPiDDB, which is also shown docked into the binding
site (Fig. 7B).
It is important to note that the bis-quaternary ammo-
nium compounds exhibit a high affinity for the choline
transporter, even though they lack a hydrogen bonding
moiety in their structure. When the new bis-quaternary
ammonium compounds were docked into the Model 1
binding site, we were unable to adequately predict the
affinities of these compounds. The bis-azaaromatic qua-
ternary ammonium compounds take up molecular space
not encountered in the training set used to create the ori-
ginal CoMFA model. This was evident when our origi-
nal model was used to predict the logK for these bis-
i
compounds. No correlation could be found between

W. J. Geldenhuys et al. / Bioorg. Med. Chem. 13 (2005) 4253–4261
4259
tance between two adjacent anionic sites on individual
transporter molecules (i.e, the bis-azaaromatic com-
pounds may bind in an intermolecular manner with
the transporter dimer). In this respect, it has been
proposed that transporters, such as the neuronal dop-
amine transporter, consist of several transmembrane
As seen in Figure 8, these cationic moieties were aligned
with choline in such a way as to orient the n-alkyl linker
toward a green, sterically favored area in the first CoM-
FA model (Model 1), and with the cationic quaternary
ammonium groups overlapping. The cationic moieties
appear to fit comfortably inside the CoMFA model in
this configuration. As expected, the predicted logKis
were not in agreement with experimental values
2
1
units. Also, G-protein-coupled receptors are known
2
2,23
to exist as functional dimeric units,
with closely
associated binding pockets between the different recep-
tor subunits.
(logK ꢁ 3), likely due to the absence of the hydroxyl
i
group, which appears to be important for high affinity
binding of most, but not all, BBB-choline transporter li-
2
Both models could explain why bis-quaternary ammo-
nium compounds have affinity for the BBB-choline
transporter, even though they do not possess a free hy-
droxyl group. Our proposed syntheses of rigid analogs
that conformationally restrict these molecules into
folded, extended or other unique structures will help
determine the active conformation of these flexible
bis-azaaromatic aromatic quaternary ammonium
compounds.
gands. This confirmed our speculation that the tethered
second cationic group is required for high affinity inter-
action or binding at the BBB-choline transporter (as is
seen with these bis-quaternary ammonium compounds),
when compared to choline. Thus, in the bis-compounds,
the incorporation of two cationic moieties into the
molecule may compensate for the lack of a hydroxyl
group thought to be necessary for hydrogen bonding
at the choline binding site (see Fig. 2).
The second cationic quaternary ammonium group teth-
ered to the first seems to compensate for the loss of the
hydrogen bond interaction proposed for ligands with a
Because our original model (Model 1) failed to ade-
quately predict the affinity of the bis-quaternary ammo-
nium compounds, we evaluated whether incorporation
of these compounds into the first model would improve
our 3D QSAR descriptors. The new modelÕs (Model 2,
9
free hydroxyl group. To address this concept in our
molecular modeling approach, we carried out the fol-
lowing: we docked only one of the cationic moieties of
the bis-azaaromatic quaternary ammonium structure
into the original CoMFA model. Accordingly, we trun-
cated the bis-azaaromatic quaternary ammonium com-
pounds, revealing three distinct substructures (Fig. 8).
2
Table 2) cross-validated q -value was lower than that
of the original model with an estimated q value of
2
0.374. With column filtering set at 2.0 kcal/mol, the
new model improved further to yield a final model with
a q of 0.536 and a non-cross-validated r of 0.881.
2
2
To test the second modelÕs predictive capability, and as
proof of concept, we searched for compounds that
incorporate two quaternary ammonium groupings
joined by an n-alkyl chain linker in the bis-azaaromatic
compounds, leading us to evaluate hemicholinium-3 and
hemicholinium-15 (Fig. 9).
Hemicholinium-3 is a bis-quaternary ammonium com-
pound similar to the bis-azaaromatic compounds, but
it also contains additional hydroxyl groups in its struc-
ture. Hemicholinium-3 is the defining substrate for cho-
2
line transport systems. Hemicholinium-15 is the
monomeric analog of hemicholinium-3, and was a com-
ponent of the training set in both CoMFA models de-
scribed above. The original CoMFA model predicted
hemicholinium-3 to have a K of 316 lM, which changed
i
to 63 lM, when the new CoMFA model was utilized.
2
The experimental K of hemicholinium-3 is 54 lM.
i
Hemicholinium-15 binds with much lower affinity than
hemicholinium-3, suggesting that the binding of two cat-
ionic quaternary ammonium groups results in higher
affinity and potentially greater transport than binding
of just one cationic group. This also indicated that, by
incorporating bis-compounds in our previous model, it
is possible to use this new model for future studies that
will attempt to predict the binding properties of bis-qua-
ternary ammonium compounds.
Figure 8. Contour maps of the original CoMFA analysis (Model 1),
without the bis-azaaromatic quaternary ammonium compounds in the
training set. The three truncated cationic moieties derived from the bis-
azaaromatic quaternary ammonium compounds are shown docked
into the original CoMFA model.
The results of this study suggest that determining the
maximum size of the ligand that can be accommodated
by the BBB-choline transporter will aid in the design of

4260
W. J. Geldenhuys et al. / Bioorg. Med. Chem. 13 (2005) 4253–4261
Figure 9. Structures of hemicholinium-3, a bis-quaternary ammonium compound, and its monomer, hemicholinium-15.
pro-drugs or co-drugs of molecules that poorly pene-
trate the BBB. Thus, tethering the poorly permeable
drug to a high affinity ligand may allow shuttling of
the drug into the CNS via the BBB-choline transporter.
Since the BBB-choline transporter has not yet been
crystallized, structural information about the protein
binding site is not available. Therefore, establishing
the relationship of the mapped CoMFA fields to the
topography of the protein site with respect to the amino
acid residues lining the binding pocket, is not possible.
However, the synthesis of additional novel (i.e., rigid
and/or bulky) ligands will continue to provide further
insight into size limitations for ligands at the binding site
about which the current model is unable to give clear in-
sight. In addition, utilization of other molecular model-
ing approaches, such as pharmacophore analysis, may
provide additional information. It is also important to
point out that, as discussed above, the mono-azaaro-
matic quaternary ammonium compound, NONI, enters
space encountered in these ÔbisÕ compounds. A new
model was constructed by including the bis-azaaromatic
quaternary ammonium compounds into our previous
data sets. The molecular models showed an excellent
2
predictability value, which cross-validated at q > 0.4.
The current model was able to predict the affinity of
hemicholinium-15, a ÔmonomericÕ compound, which
binds with much less affinity than hemicholinium-3, a
structurally related ÔdimericÕ or ÔbisÕ analog. From these
studies it is evident that there is a need for the synthesis
and evaluation of more rigid or conformationally re-
strained bis-quaternary ammonium analogs, to further
determine the topographical architecture of the BBB-
choline transporter binding site, and to elucidate the
structural requirements for binding and transport.
Acknowledgements
5
,15
the brain via the BBB-choline transporter.
Further-
This project was partly funded under the NIH grant
number U19DA017548.
more, supporting evidence demonstrates that the
bis-azaaromatic compound, bPiDDB, decreases self-
administration of nicotine in rats after peripheral
2
4
3
administration, and is a potent inhibitor of [ H]-cho-
line transport via the BBB-choline transporter. Thus, ta-
ken together, these results suggest that both NONI and
bPiDDB enter the brain via the BBB-choline trans-
porter, and that CNS active compounds can be designed
to utilize this transporter system as a portal of entry into
the brain.
References and notes
1
2
. Pardridge, W. M. Nat. Rev. Drug Discov. 2002, 1, 131.
. Lockman, P. R.; Allen, D. D. Drug Dev. Ind. Pharm. 2002,
2
. Takada, Y.; Greig, N. H.; Vistica, D. T.; Rapoport, S. I.;
Smith, Q. R. Cancer Chemother. Pharmacol. 1991, 29,
8
8, 749.
3
9.
4
. Takada, Y.; Vistica, D. T.; Greig, N. H.; Purdon, D.;
Rapoport, S. I.; Smith, Q. R. Cancer Res. 1992, 52, 2191.
. Allen, D. D.; Lockman, P. R.; Roder, K. E.; Dwoskin, L.
P.; Crooks, P. A. J. Pharmacol. Exp. Ther. 2003, 304,
5. Conclusion
5
We have investigated a series of nAChR antagonists in
earlier CoMFA and CoMSIA models of the BBB-cho-
line transporter. Our earlier models could not ade-
quately predict the BBB-choline transporter affinity of
the novel bis-azaaromatic quaternary ammonium com-
pounds. This prediction failure was likely due to the
models not being able to accommodate the n-alkyl link-
age separating the two cationic quaternary ammonium
moieties, as these data sets did not include the molecular
1
268.
6
7
8
. Okuda, T.; Haga, T. Neurochem. Res. 2003, 28, 483.
. Allen, D. D.; Lockman, P. R. Life Sci. 2003, 73, 1609.
. Geldenhuys, W. J.; Lockman, P. R.; McAfee, J. H.;
Fitzpatrick, K. T.; Van der Schyf, C. J.; Allen, D. D.
Bioorg. Med. Chem. Lett. 2004, 14, 3085.
. Batzold, F.; DeHaven, R.; Kuhart, M. J.; Birdsall, N.
Biochem. Pharmacol. 1980, 29, 2413.
10. Allen, D. D.; Smith, Q. R. J. Neurochem. 2001, 76, 1032.
9

W. J. Geldenhuys et al. / Bioorg. Med. Chem. 13 (2005) 4253–4261
4261
1
1
1. Crooks, P. A.; Ravard, A.; Wilkins, L. H.; Teng, L. H.;
Buxton, S. T.; Dwoskin, L. P. Drug Dev. Res. 1995, 36, 91.
2. Wilkins, L. H., Jr.; Haubner, A.; Ayers, J. T.; Crooks, P.
A.; Dwoskin, L. P. J. Pharmacol. Exp. Ther. 2002, 301,
16. Takasato, Y.; Rapoport, S. I.; Smith, Q. R. Am. J.
Physiol. 1984, 247, 484.
17. Smith, Q. R. Pharm. Biotechnol. 1996, 8, 285.
18. Allen, D. D.; Smith, Q. R. J. Neurochem. 2001, 76, 1.
19. Smith, Q. R.; Momma, S.; Aoyai, M.; Rapoport, S. I. J.
Neurochem. 1987, 49, 1651.
20. Smith, Q. R.; Nagura, H.; Takada, Y.; Duncan, M. W. J.
Neurochem. 1992, 58, 1330.
21. Ravna, A. W.; Sylte, I.; Dahl, S. G. J. Comput. Aided Mol.
Des. 2003, 17, 367.
22. Moro, S.; Spalluto, G.; Jacobson, K. A. Trends Pharma-
col. Sci. 2005, 26, 44.
23. Miligan, G. Mol. Pharmacol. 2004, 66, 1.
24. Neugebauer, N. M.; Zhang, Z.; Crooks, P. A.; Dwoskin,
L. P.; Bardo, M. T. Psychopharmacology, in press.
1088.
3. Ayers, J. T.; Dwoskin, L. P.; Deaciuc, A. G.; Grinevich, V.
P.; Zhu, J.; Crooks, P. A. Bioorg. Med. Chem. Lett. 2002,
1
1
1
1
4. Dwoskin, L. P.; Sumithran, S. P.; Zhu, J.; Deaciuc, A. G.;
2, 3067.
Ayers, J. T.; Crooks, P. A. Bioorg. Med. Chem. Lett. 2004,
1
4, 1863.
5. Crooks, P. A.; Ayers, J. T.; Xu, R.; Sumithran, S. P.;
Grinevich, V. P.; Wilkins, L. H.; Deaciuc, A. G.; Allen, D.
D.; Dwoskin, L. P. Bioorg. Med. Chem. Lett. 2004, 14,
1
869.