Quantitative insight into the design of compounds recognized by the L-type amino acid transporter 1 (LAT1)

Article abstract of DOI:10.1002/cmdc.201402281

L-Type amino acid transporter 1 (LAT1) is a transmembrane protein expressed abundantly at the blood-brain barrier (BBB), where it ensures the transport of hydrophobic acids from the blood to the brain. Due to its unique substrate specificity and high expression at the BBB, LAT1 is an intriguing target for carrier- mediated transport of drugs into the brain. In this study, a comparative molecular field analysis (CoMFA) model with considerable statistical quality (Q²=0.53, R²=0.75, Q² SE=0.77, R² SE=0.57) and good external predictivity (CCC=0.91) was generated. The model was used to guide the synthesis of eight new prodrugs whose affinity for LAT1 was tested by using an in situ rat brain perfusion technique. This resulted in the creation of a novel LAT1 prodrug with l-tryptophan as the promoiety; it also provided a better understanding of the molecular features of LAT1-targeted high-affinity prodrugs, as well as their promoiety and parent drug. The results obtained will be beneficial in the rational design of novel LAT1-binding prodrugs and other compounds that bind to LAT1.

Full text of DOI:10.1002/cmdc.201402281

CHEMMEDCHEM
FULL PAPERS
DOI: 10.1002/cmdc.201402281
Quantitative Insight into the Design of Compounds
Recognized by the l-Type Amino Acid Transporter 1 (LAT1)
Henna Ylikangas,[a] Kalle Malmioja,[a] Lauri Peura,[a] Mikko Gynther,[a]
Emmanuel O. Nwachukwu,[a] Jukka Leppꢀnen,[a] Krista Laine,[a] Jarkko Rautio,[a] Maija Lahtela-
Kakkonen,[a] Kristiina M. Huttunen,[a] and Antti Poso*[a, b]
l-Type amino acid transporter 1 (LAT1) is a transmembrane
protein expressed abundantly at the blood–brain barrier (BBB),
where it ensures the transport of hydrophobic acids from the
blood to the brain. Due to its unique substrate specificity and
high expression at the BBB, LAT1 is an intriguing target for car-
rier-mediated transport of drugs into the brain. In this study,
a comparative molecular field analysis (CoMFA) model with
considerable statistical quality (Q2 =0.53, R2 =0.75, Q2 SE=0.77,
R2 SE=0.57) and good external predictivity (CCC=0.91) was
generated. The model was used to guide the synthesis of
eight new prodrugs whose affinity for LAT1 was tested by
using an in situ rat brain perfusion technique. This resulted in
the creation of a novel LAT1 prodrug with l-tryptophan as the
promoiety; it also provided a better understanding of the mo-
lecular features of LAT1-targeted high-affinity prodrugs, as well
as their promoiety and parent drug. The results obtained will
be beneficial in the rational design of novel LAT1-binding pro-
drugs and other compounds that bind to LAT1.
Introduction
l-Type amino acid transporter 1 (LAT1) is a sodium-independ-
ent heterodimeric transmembrane protein. LAT1 is found in
several peripheral tissues, but is expressed at 100-fold higher
levels in the blood–brain barrier (BBB) than in peripheral tis-
sues (e.g., placenta, retina, and gut).[1,2] The natural substrates
of LAT1 are large neutral amino acids such as l-leucine, l-tryp-
tophan, and l-phenylalanine. These compounds have excellent
affinity for the LAT1 transporter protein and thus rapidly pass
through the BBB.[3] However, LAT1 also transports thyroid hor-
mones and amino acid derived drug molecules such as levodo-
pa,[4] gabapentin,[5] and baclofen.[6] Its distinctive substrate spe-
cificity and relatively high expression at the BBB make LAT1 an
interesting target for drug delivery to the central nervous
system (CNS).
previous studies have focused mostly on a few amino acids
and their derivatives in clinical use.[3,9,15] Little information is
available on the structural and chemical features that influence
the molecular size and flexibility of the prodrug or the pre-
ferred position of the parent drug with respect to the promoi-
ety essential for achieving LAT1-binding prodrugs with high
affinities.
To design better LAT1 prodrugs, it is of the utmost impor-
tance to gain a good understanding of the three-dimensional
(3D) structure–activity relationships of LAT1-binding com-
pounds. For this reason, we were interested in analyzing the
3D quantitative structure–activity relationships (3D QSAR) of
LAT1-binding compounds.
In this study, the first 3D QSAR study of LAT1-binding com-
pounds was carried out by using classical and topomer compa-
rative molecular field analysis (CoMFA).[16,17] The models are
based on biological data determined by an in situ rat brain
perfusion technique, which reveals the ability of the com-
pounds to inhibit the brain uptake of the LAT1-selective sub-
strate, [14C]l-leucine.[18] By using the information obtained from
the CoMFA models, it was possible to design and synthesize
eight new prodrugs. The model predicted the affinities of the
synthesized compounds with reasonable evidence that the
topomer CoMFA model can be used in the rational design of
new LAT1 prodrugs. Herein we discuss the quantitative insights
obtained from the CoMFA model and propose new concepts
for LAT1-binding compounds and prodrugs, which can be
useful for improved drug delivery to the brain. As far as we are
aware, these novel 3D QSAR models are the first to be pub-
lished for compounds that target LAT1.
Natural substrates of LAT1 can be used as potential promoi-
eties for LAT1-targeted prodrugs in order to improve drug de-
livery to the brain. In general, the substrates should have un-
substituted carboxylic acid and amine functionalities, and the
parent drug should be attached to the side chain of amino
acids in order to maintain efficient LAT1 binding.[7–9] These
kinds of prodrugs have been reported earlier.[8,10–14] However,
[a] H. Ylikangas, K. Malmioja, Dr. L. Peura, Dr. M. Gynther, E. O. Nwachukwu,
Dr. J. Leppꢀnen, Dr. K. Laine, Prof. J. Rautio, Dr. M. Lahtela-Kakkonen,
Dr. K. M. Huttunen, Prof. A. Poso
School of Pharmacy, University of Eastern Finland
P.O. Box 1627, 70211 Kuopio (Finland)
E-mail: antti.poso@uef.fi
[b] Prof. A. Poso
University Hospital Tꢁbingen, Department of Internal Medicine 1
Division of Translational Gastrointestinal Oncology
Otfried-Mꢁller-Strasse 10, 72076 Tꢁbingen (Germany)
Supporting information for this article is available on the WWW under
http://dx.doi.org/10.1002/cmdc.201402281.
ꢁ 2014 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim
ChemMedChem 2014, 9, 2699 – 2707 2699

CHEMMEDCHEM
FULL PAPERS
www.chemmedchem.org
The data series used in CoMFA models consist of 47 com-
pounds which were selected based on their relatively diverse
structures in order to obtain as much information on 3D QSAR
as possible. The training set (Table 1) consists of 39 and an ex-
ternal test set of eight compounds with varying affinities for
LAT1 (0–100% l-leucine uptake inhibition). Their ability to in-
hibit l-leucine uptake into the brain was used as a measure of
biological activity, as it describes the ability of the investigated
compounds to replace competing substrates, that is, their af-
finity for LAT1.[8] High affinity is a prerequisite for efficient
uptake into the brain; this
Results and Discussion
CoMFA models
In general, prodrugs can be divided into two main sections:
the parent drug and the promoiety, which are connected to
each other by a cleavable prodrug bond. Following this no-
menclature, these sections in the CoMFA coefficient maps
were analyzed based on the natural substrates of LAT1 and
previously reported LAT1 prodrugs (Table 1).
makes it a reliable method for
Table 1. Training set used in CoMFA models with percent l-leucine uptake inhibition determined by the in
situ rat brain perfusion technique.
determining biological activity,
although as such, it does not di-
rectly guarantee uptake into the
brain.
Compd
(Inh. [%])[a]
Structure
Compd
(Inh. [%])[a]
Structure
In addition to our previously
published
compounds[7,8,11,19]
1
2
and commercial compounds
(Table 1), the training set in-
cludes one novel prodrug (com-
pound 26), and an amino acid
derivative (39) which were de-
signed as a continuum to our
previously published prodrugs
in order to broaden the chemi-
cal space presented so far. The
detailed synthesis procedures of
these compounds are described
in the Supporting Information.
Six 3D QSAR models were
generated, and from those, one
model with two components
was selected and used in guid-
ing the synthesis of the new
LAT1 substrates. The model was
selected based on results ob-
tained from partial least squares
(PLS) and progressive scram-
bling stability analyses (topomer
CoMFA model E: Table 2). The
topomer CoMFA model revealed
that the steric component ex-
plains most of the variance in
biological activity of LAT1-bind-
ing compounds. The contribu-
tion of steric interactions was
stronger than the effect of elec-
trostatics for both topomers, R1:
amino acid terminal and R2:
side chain, prodrug bond, and
(99ꢀ4)
(94ꢀ2)
3
4
(93ꢀ2)
(93ꢀ3)
5
6
(92ꢀ1)
(88ꢀ3)
7
8
(86ꢀ2)
(86ꢀ4)
9
10
(81ꢀ1)
(83ꢀ2)
11
(79ꢀ2)
12[b]
(79ꢀ6)
13
(75ꢀ2)
14
(66ꢀ1)
15
(66ꢀ5)
16
(64ꢀ3)
17
(64ꢀ4)
18
(63ꢀ7)
19
(59ꢀ4)
20
(57ꢀ20)
21
22
(56ꢀ1)
(39ꢀ6)
parent
drug,
respectively
(Figure 1).
23
24
(38ꢀ5)
(36ꢀ7)
ꢁ 2014 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim
ChemMedChem 2014, 9, 2699 – 2707 2700

CHEMMEDCHEM
FULL PAPERS
www.chemmedchem.org
Model analysis
Table 1. (Continued)
Compd
Structure
Compd
Structure
Our CoMFA model indicates that
adding steric moieties that
occupy regions beyond the
amino acid terminal decreases
affinity for LAT1 (Figure 1: yellow
contour in R1). These findings
are in agreement with our previ-
ous studies.[20] The model also
revealed that increasing the pos-
itive charge near the amine
function (blue contours at R1)
can be beneficial for affinity. In
(Inh. [%])[a]
(Inh. [%])[a]
25
(35ꢀ11)
26[c]
(33ꢀ6)
27
(31ꢀ1)
28
(25ꢀ7)
29
(23ꢀ3)
30
(21ꢀ7)
addition, adding
a
negative
charge near the carboxylic acid
can increase LAT1 affinity (red
cubic-shaped contours at R1). So
far, all the tested compounds
that fulfill the electrostatic fea-
tures of the model but do not
contain an amino acid function-
ality (28, 30, and 32) have been
rather poor inhibitors (i.e.,
<25% of l-leucine uptake into
the brain; Table 1).
31
(8ꢀ5)
32
(5ꢀ10)
33[d]
(0ꢀ0)
34[d]
(0ꢀ0)
35[d]
(0ꢀ0)
36[d]
(0ꢀ0)
The model shows that the
promoiety for LAT1 prodrugs
should be relatively planar, be-
cause large, branched substitu-
ents can reach areas that can de-
crease affinity (yellow contour in
R2). Moreover, the addition of
steric substitutions near the 5-
and 6-positions of l-tryptophan
and its derivatives can be benefi-
37[d]
(0ꢀ0)
38[d]
(0ꢀ0)
39[d]
(0ꢀ0)
[a] [14C]l-leucine brain capillary surface area (PA) percent inhibition data are the averageꢀSD of n=3
experiments conducted at a compound concentration of 100 mm. [b] Measured at 70 mm compound con-
centration. [c] n=4. [d] n=2.
cial
for
efficient
binding
(Figure 1: green crescent-shaped
contour at R2). These regions correspond roughly to the 3-
and 4-positions of l-phenylalanine and indicate positions
where one can attach the parent drug in the promoiety. This is
in agreement with results of previous studies.[7,15]
The most efficient valproic acid prodrugs in the training set
(3 and 4; >90% l-leucine uptake inhibition) contain the
parent drug substituted at the 3-position of l-phenylalanine.[8]
These correspond better with steric features revealed by the
CoMFA model (green crescent-shaped contour at R2) than pro-
drugs 16 and 17 (64% l-leucine uptake inhibition). Prodrugs
16 and 17 contain parent drugs at the 4-position and are less
efficient in inhibiting l-leucine uptake than 3 and 4 (Table 1).
The CoMFA model shows that prodrugs with parent drug at
position 4 (e.g., 1, 16, and 17) do not occupy sterically disfa-
vored areas. However, their orientations do not reach the steri-
cally favored regions described by the topomer CoMFA model
either (green contours in R2). In addition to the position of the
parent drug, the CoMFA model indicates that relatively rigid
and large prodrugs bind more efficiently to LAT1 than more
Figure 1. Topomer CoMFA model presented with l-tryptophan prodrug 40
(99% l-leucine uptake inhibition). Green and red represent areas where
adding steric features and negative charge or hydrogen bond acceptors are
favored, respectively. Blue represents areas where more positive charge or
hydrogen bond donors are favored, whereas yellow contours designate ster-
ically disfavored regions. R1 and R2 indicate the common core; amino acid
function and variable topomer; side chain and parent drug.
ꢁ 2014 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim
ChemMedChem 2014, 9, 2699 – 2707 2701

CHEMMEDCHEM
FULL PAPERS
www.chemmedchem.org
fulfill sterically favored areas. In
addition, the following charac-
teristics can decrease affinity for
LAT1: 4) adding steric features
above the aromatic plane of the
amino acid side chains; 5) insert-
ing substituents beyond the
amino acid terminal which reach
sterically disfavored regions and
disturb the amino acid function-
ality of the promoiety.
Table 2. Progressive scrambling stability of the six CoMFA models generated for LAT1-binding compounds rep-
resented as meanꢀSD of 20 independent scrambling tests.
[g]
Model
n[a]
N[b]
Qs*2[c]
Q0*2[d]
SDEPs*[e]
SDEP0*[f]
dq2/dr2
CCC[h]
yy
1
2
3
0.415ꢀ0.013
0.460ꢀ0.02
0.421ꢀ0.03
0.488ꢀ0.01
0.541ꢀ0.03
0.494ꢀ0.03
0.872ꢀ0.01
0.847ꢀ0.02
0.892ꢀ0.02
0.836ꢀ0.01
0.803ꢀ0.03
0.853ꢀ0.03
0.237ꢀ0.08
0.408ꢀ0.12
0.648ꢀ0.18
A
34
0.813
2
3
4
0.523ꢀ0.02
0.534ꢀ0.02
0.465ꢀ0.03
0.616ꢀ0.02
0.628ꢀ0.02
0.547ꢀ0.03
0.747ꢀ0.01
0.753ꢀ0.01
0.820ꢀ0.02
0.668ꢀ0.01
0.961ꢀ0.08
0.751ꢀ0.02
0.819ꢀ0.13
0.961ꢀ0.08
1.135ꢀ0.18
B
C
D
E
32
33
36
33
35
0.696
0.569
1.00
2
3
4
0.450ꢀ0.01
0.450ꢀ0.02
0.349ꢀ0.02
0.529ꢀ0.01
0.529ꢀ0.02
0.411ꢀ0.03
0.833ꢀ0.01
0.845ꢀ0.01
0.935ꢀ0.02
0.781ꢀ0.01
0.792ꢀ0.01
0.896ꢀ0.02
0.868ꢀ0.12
1.139ꢀ0.15
1.108ꢀ0.23
Model validation
1
2
3
0.304ꢀ0.01
0.342ꢀ0.01
0.321ꢀ0.02
0.358ꢀ0.02
0.402ꢀ0.02
0.377ꢀ0.02
0.925ꢀ0.01
0.912ꢀ0.01
0.939ꢀ0.01
0.904ꢀ0.01
0.886ꢀ0.01
0.770ꢀ0.20
0.517ꢀ0.61
0.726ꢀ0.12
0.770ꢀ0.20
The CoMFA model showed ac-
ceptable internal predictivity
(Q2 >0.5) when this was deter-
mined with leave-one-out (LOO)
cross-validation. The experimen-
tal and predicted activities for
the compounds showed good
accuracy with the exception of
two compounds: valproic acid
prodrug 26 and aliphatic com-
pound 35. The predictions of
activity for them (72 and 43%
l-leucine uptake inhibition, re-
spectively) differ by >1.2 loga-
rithmic units from measured ac-
tivities, which translate into 39
and 43 percentage units differ-
1
2
3
0.382ꢀ0.01
0.436ꢀ0.01
0.386ꢀ0.02
0.450ꢀ0.01
0.513ꢀ0.01
0.454ꢀ0.03
0.868ꢀ0.01
0.843ꢀ0.01
0.892ꢀ0.02
0.783ꢀ0.10
0.803ꢀ0.01
0.846ꢀ0.02
0.783ꢀ0.10
0.922ꢀ0.11
1.11ꢀ0.12
0.907
0.823
2
3
4
0.377ꢀ0.02
0.380ꢀ0.03
0.313ꢀ0.04
0.443ꢀ0.02
0.447ꢀ0.03
0.368ꢀ0.04
0.901ꢀ0.01
0.912ꢀ0.02
0.976ꢀ0.02
0.870ꢀ0.02
0.880ꢀ0.03
0.902ꢀ0.03
0.642ꢀ0.11
0.672ꢀ0.19
0.765ꢀ0.23
F
[a] Number of ligands in the training set. [b] Number of PLS components. [c] Predictivity at the critical thresh-
old level of perturbation s (0.85); maximum value of Qs*2 s. [d] Q0*2 =Qs*2 s: adjusted Qs*2, corresponding to the
value expected for an unperturbed, non-redundant Q2. [e] Standard error of prediction at the critical threshold
level of perturbation s. [f] SDEP0*={[(2ꢁs)(nꢁNꢁ1)(SDEPs)2ꢁ(1ꢁs)(nꢁ1)SDy2]/(nꢁNꢁ1)}1/2; N=number of com-
pounds in the training set; SDy =1.1 (response standard deviation); adjusted SDEPs*, corresponding to the
value expected for an unperturbed, non-redundant model. [g] Sensitivity to perturbation. [h] Concordance cor-
relation coefficient.[20,21]
flexible prodrugs. For example, the ketoprofen prodrug with
a derivative of l-phenylalanine as a promoiety (compound 1)
inhibits >90% of l-leucine uptake, although it is substituted
at position 4. Prodrug 12, which has a more flexible promoiety
(l-lysine), inhibits l-leucine uptake by almost 20 percentage
units less than prodrug 1 (79% l-leucine uptake inhibition). It
is possible that the large aromatic portion of 1 allows the pro-
drug to adopt a conformation beneficial for LAT1 binding. Sim-
ilarly, dopamine prodrug 8 inhibits at 86%, whereas the more
flexible prodrugs 22 and 25 inhibit <40% of l-leucine uptake
into the brain. Based on the CoMFA model, it seems that the
flexible prodrugs protrude into sterically disfavored areas
(yellow contours at R2) which can decrease their affinity for
LAT1. This finding was further studied with the newly synthe-
sized compounds.
ence in l-leucine uptake inhibition (Supporting Information).
This might be due to the fact that the simultaneously meta-
and para-substituted prodrug 26 and non-aromatic compound
35 are located outside the chemical space presented by the
CoMFA model. Therefore, activity predictions for 26 and 35 are
less reliable due to the large extrapolation. Another modest
outlier, compound 39, is observed in the plot of predicted
versus measured activities (Figure 2). It is an amino acid deriva-
tive which is unable to inhibit the uptake of l-leucine. Exclud-
ing 39 from the training set decreases the statistical quality of
the CoMFA model which is suggested to result from the
strong electrostatic interactions of the compound (i.e., the
nitro substituent at position 3 withdrawing electrons from the
aromatic ring). However, compound 39 provides important in-
formation on electrostatics in the CoMFA model which is
missed if the compound is omitted from the training set. Nev-
ertheless, the predicted activity values of CoMFA model for the
training set are in agreement with the experimental data
within a statistically tolerable error range. Subsequently, the
CoMFA model was used to guide the synthesis of eight novel
prodrugs (40–47) and then used to predict their activities,
which were compared with [14C]l-leucine uptake inhibition de-
termined by using the in situ rat brain perfusion method.
These compounds (prediction set) were further used to deter-
mine the external predictivity of the CoMFA model.
In summary, the 3D QSAR model created for LAT1-binding
compounds indicates that the following chemical modifica-
tions to currently known substrates can increase their affinity
for LAT1: 1) increasing the negative charge over the aromatic
ring of l-phenylalanine and l-tryptophan; 2) adding substitu-
ents that increase the negative charge in proximity to the 5-
and 6-positions of l-tryptophan, and similarly to the 3- and 4-
positions of l-phenylalanine; 3) adding substituents to l-phe-
nylalanine and l-tryptophan at the same plane with their aro-
matic side chains at the 5- and 6-positions of l-tryptophan,
and similarly to the 3- and 4-positions of l-phenylalanine to
ꢁ 2014 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim
ChemMedChem 2014, 9, 2699 – 2707 2702

CHEMMEDCHEM
FULL PAPERS
www.chemmedchem.org
residues. Prodrugs 41, 42, 43, 45, and 46 were prepared by
starting from commercially available modified amino acids
(Scheme 2). Amino acid 63 was prepared by reacting 4-nitro-3-
fluorobenzylbromide with diethyl acetamidomalonate in the
presence of sodium hydride. The amino groups of all amino
acid derivatives were protected with Boc if the commercial
starting compounds were not protected already. The carboxylic
acid groups were converted into their methyl esters before
palladium-catalyzed reduction of nitro or cyano groups. The
parent drugs and amino acids 63, 66, 71, and 74 were then
coupled by EDC. The carboxylic acid of the amino acid deriva-
tive with a cyano group was left unprotected, as compound
59 was reacted with valproic acid chloride after reduction of
the cyano group. Finally, all the protecting groups were re-
moved through base-catalyzed hydrolysis for the methyl ester
and by acid-catalyzed hydrolysis for the Boc group, respective-
ly. The products were treated either with HCl(g) or HCl in diox-
ane to remove any remaining TFA. Valproic acid prodrug 47
(Table 3) was prepared by coupling 2-propylpentanoyl chloride
Figure 2. Predicted versus measured binding affinities of compounds in the
training set shown on a logarithmic scale. Those compounds with the high-
est percent [14C]l-leucine uptake inhibition are located at the upper right-
hand side of the coordinates. The arrow indicates compound 39.
Chemistry of the prediction set
The synthesized prodrugs in-
clude two benzoic acid prodrugs
(40 and 44), four valproic acid
prodrugs (41, 42, 45, and 47),
and two prodrugs of kynurenic
acid (43 and 46). Prodrugs 40
and 44 were synthesized by pro-
tecting amino and carboxylic
acid groups of l-5-hydroxytryp-
tophan and l-tyrosine with tert-
butyloxycarbonyl
(Boc)
and
benzyl (Bn) groups, respectively,
and coupling the obtained com-
pounds with benzoic acid using
1-ethyl-3-(3-dimethylaminopro-
pyl)carbodiimide
(EDC)
(Scheme 1). The Bn and Boc pro-
tecting groups were cleaved se-
quentially by hydrogenolysis on
palladium and trifluoroacetic
acid (TFA), respectively. Finally,
the products were treated with
hydrochloric acid to remove TFA
Scheme 2. Synthesis of 41–43 and 45 and 46. Reagents and conditions: a) 1. NaH, acetamidomalonate, DMF, RT,
4 h; 2. HCl, reflux 18 h; b) Boc2O, TEA, THF, RT, 18 h; c) dimethyl sulfate, K2CO3, acetone, RT, 18 h, or SOCl2 in
MeOH, reflux, 4 h, 69–99%; d) 10% Pd/C, MeOH, RT, 4 h, 95–99%; e) 2-propylpentanoyl chloride, NaOH, CH2Cl2, RT,
3 h, 72%; f) valproic or kynurenic acid, EDC, DMAP, DMF, RT, 18 h, 31–66%; g) 1. NaOH or LiOH(aq), MeOH, RT,
30 min; 2. TFA, CH2Cl2, RT, 30 min, 81–91% or HCl(g), CH3CN, RT, 30 min, 99%.
Scheme 1. Synthesis of 40 and 44. Reagents and conditions: a) 1. Boc2O, TEA, THF, RT, 18 h; 2. benzyl alcohol, EDC, DMAP, CH2Cl2, RT, 18 h, 58–70%; b) benzoic
acid, EDC, DMAP, CH2Cl2, RT, 18 h, 35–67%; c) Pd (10% on activated charcoal), MeOH, RT, 45 min; d) TFA, CH2Cl2, RT, 30 min; e) 4m HCl in dioxane, RT, 10 min,
48–72%.
ꢁ 2014 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim
ChemMedChem 2014, 9, 2699 – 2707 2703

CHEMMEDCHEM
FULL PAPERS
www.chemmedchem.org
can be used to prioritize compounds for synthesis
and further optimization of the LAT1 prodrugs.
Table 3. Prediction set synthesized based on the statistically best CoMFA model with the
values of percent l-leucine uptake inhibition determined by the in situ rat brain perfu-
sion technique.
Compd
(Inh. [%])[a]
Structure
Compd
(Inh. [%])[a]
Structure
QSAR of synthesized compounds
Topomer CoMFA suggested that the steric and
electrostatic properties of l-tryptophan have a pos-
itive effect on affinity for LAT1. Consequently, we
were interested in introducing l-tryptophan and
its analogues as promoieties for new LAT1 pro-
drugs. As far as we are aware, l-tryptophan has
not been previously used as a promoiety for LAT1.
The CoMFA model indicated that adding steric fea-
tures at position 5 of l-tryptophan could be bene-
ficial for affinity (Figure 1; green crescent-shaped
contour at R2). For this reason, benzoic acid was
substituted into position 5 of l-tryptophan (com-
pound 40), which is a representative prodrug with
the highest affinity for LAT1 in our data set (99%
l-leucine uptake inhibition). Relative to natural
l-tryptophan (6; 88% l-leucine uptake inhibition)
and 5-methoxy-d,l-tryptophan (12; 75% l-leucine
uptake inhibition), prodrug 40 expresses signifi-
cantly higher affinity for LAT1. This indicates that
l-tryptophan is a valuable promoiety for novel
LAT1-targeted prodrugs.
40
(99ꢀ1)
41
(92ꢀ2)
42
(87ꢀ4)
43
(85ꢀ4)
44
(73ꢀ6)
45
(65ꢀ3)
46
(25ꢀ3)
47
(3ꢀ0)[b]
[a] [14C]l-leucine brain capillary surface area (PA) percent inhibition data are the aver-
ageꢀSD of n=3 experiments conducted at a compound concentration of 100 mm.
[b] n=2.
with Boc-l-lysine and subsequently removing the Boc group
Table 4. Measured versus predicted percent l-leucine uptake inhibition
by the prediction set.
by acid-catalyzed hydrolysis. Detailed synthesis procedures are
described in the Supporting Information.
Compd
Inhibition [%][a]
log(Inhib. [%])[b]
Measured
Predicted
Measured
Predicted
External predictivity
40
41
42
43
44
45
46
47
99ꢀ1
92ꢀ2
87ꢀ4
85ꢀ4
73ꢀ6
65ꢀ3
25ꢀ3
3ꢀ0[b]
86
87
84
84
80
85
77
43
6.00
5.06
4.83
4.75
4.44
4.26
3.51
2.43
4.78
4.83
4.73
4.71
4.60
4.74
4.52
3.87
The CoMFA model predicted compounds 40–45 to be active
(>50% l-leucine uptake inhibition; >4.00 in logit transforma-
tion), although the measured and predicted l-leucine uptake
inhibition values of compounds 40, 44, and 45 differ from
each other by >10 percentage units (Table 4). Benzoic acid
prodrug 40 is the only l-tryptophan derivative substituted at
position 6 of the indole ring (Figure 1). This was suggested by
the CoMFA model, and it is a singularity in the chemical space
of the data set. The model also anticipated the meta- and
para-substituted prodrug 46 to be threefold more active (77%
l-leucine uptake inhibition) than determined by the in situ rat
brain perfusion (25% l-leucine uptake inhibition) (Table 4). Pro-
drug 46 can be considered to locate slightly outside the chem-
ical space of the data set, which complicates the correct pre-
diction of its activity. The model concluded that aliphatic com-
pound 47 would be inactive (<4.00 logit transformation),
which is in agreement with its affinity (3% l-leucine uptake in-
hibition). Altogether, the CoMFA model was shown to possess
statistically significant external predictivity (concordance corre-
lation coefficient, CCC=0.91; Table 2).[20,21] CCC was selected as
an indicator for this study, as it has been proposed to be the
most stable and restrictive in determining real predictivity of
QSAR models.[20] These results show that the CoMFA model
[a] [14C]l-leucine brain capillary surface area percent inhibition data are
the averageꢀSD of n=3 experiments conducted at a compound con-
centration of 100 mm. [b] Logit transformation of l-leucine uptake inhibi-
tion=Sc+log(I/(100ꢁI)), in which I=percent l-leucine uptake inhibition;
I>99!I=99, I<1!I=1; Sc (scale factor)=ꢁlog(c), for which c=stan-
dard molar concentration (10ꢁ4).[33]
Similar to 40, prodrug 44 was substituted at position 4 of
l-phenylalanine which was also observed to increase affinity
for LAT1 by the CoMFA model (green crescent-shaped contour
at R2). Both representative prodrugs inhibit l-leucine uptake
by >50%, but the results show that substitution at position 5
of l-tryptophan is preferred over position 4 of l-phenylalanine
(99 and 73% uptake inhibition, respectively).
In our previous studies we also demonstrated that the 3-po-
sition of l-phenylalanine is preferred over position 4.[11] This
was further studied with kynurenic acid prodrugs 43 and 46,
ꢁ 2014 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim
ChemMedChem 2014, 9, 2699 – 2707 2704

CHEMMEDCHEM
FULL PAPERS
www.chemmedchem.org
as the CoMFA model suggests that adding steric features at
positions 3 and 4 of l-phenylalanine can enhance affinity for
LAT1 (Figure 1). Similarly to previous studies, kynurenic acid at
position 3 (43) is considerably more efficient than the parent
drug at position 4 (46) (85 and 25% l-leucine uptake inhibi-
tion, respectively). Adding kynurenic acid successfully as the
parent drug (43 and 46) shows that in addition to flexible pro-
drugs, LAT1 can accommodate prodrugs with larger and more
rigid parent drugs than ketoprofen, which was the parent drug
used in our earlier work.[8]
aliphatic promoiety and aromatic parent drug (Tables 1 and 2).
This further shows that complete removal of aromaticity has
a significant effect on LAT1 affinity.
Conclusions
The topomer CoMFA model generated in this study provides
novel insight into the molecular features required for LAT1-
binding compounds with high affinity. It clarifies the molecular
size, flexibility, and preferred position of the parent drug of
LAT1-targeted prodrugs. The CoMFA model was used success-
fully in the design of new prodrugs, which had high affinity for
LAT1; it resulted in the discovery of the first representative
l-tryptophan prodrug. This indicates that the CoMFA model
can be used in the rational design of novel prodrugs and other
compounds that bind efficiently to LAT1.
Inserting steric features at position 3 was also studied by in-
troducing a methylene group between the amide and aromat-
ic ring of l-phenylalanine (compound 41). The modification
did not disrupt binding to LAT1, as prodrug 41 inhibits l-leu-
cine uptake inhibition to the same extent as high-affinity pro-
drugs 3 and 4 (>90% l-leucine uptake inhibition). This finding
indicates that the distance between the parent drug and pro-
moiety can be modified, and that prodrugs are relatively toler-
ant in the sense of geometry of the parent drugs. This is en-
couraging when one considers its implications for the future
design of larger LAT1 prodrugs.
Experimental Section
Computational design
Ligands: The training set contains substrates of LAT1 and their de-
rivatives, ketoprofen, valproic acid, and dopamine prodrugs with
activities spanning three log units. The prediction set contains
close structural analogues with training set compounds and in-
cludes new valproic acid and kynurenic acid prodrugs.
The steric and electrostatic properties of l-tryptophan are
similar to those of l-phenylalanine, which has been successful-
ly applied as a promoiety in several previously reported LAT1
prodrugs (1, 3, 4, 16, and 17).[8,11,19] Based on the CoMFA
model, electrostatic and steric modifications were made to the
aromatic ring of l-phenylalanine to increase their affinity for
LAT1. Following from this step, fluoro and hydroxy substituents
were placed at position 4 of l-phenylalanine (42 and 45). Pro-
drugs 42 and 45 are close analogues of the methoxy-substitut-
ed valproic acid prodrug 26 in the training set (Tables 1 and 3).
The substituents used in 42 and 45 are similar in size, and
both can activate the aromatic ring. Both prodrugs exhibit
high affinity for LAT1 (87 and 65% l-leucine uptake inhibition,
respectively), whereas prodrug 26 inhibits only 33% of l-leu-
cine uptake. The molecular size of the methoxy substituent in
26 is slightly larger than either fluoride or hydroxy and it can
rotate more freely. This could affect the conformation of the
prodrug, directing the parent drug into an disfavored region,
thereby decreasing affinity for LAT1. Nevertheless, compounds
26, 42, and 45 show that it is possible to employ para and
meta substitutions simultaneously and still have efficient LAT1
prodrugs. Successful simultaneous substitution of 42 and 45
also provides an interesting possibility to modify the chemical
stability of the prodrug bond by a suitable substituent at the
meta position which could affect the bioconversion rate of
LAT1 prodrugs. However, this needs to be further studied
before any definitive conclusions can be made.
Training set (n=39): O-[2-(3-benzoylphenyl)-1-oxopropyl]-l-tyro-
sine (1),[8] l-1-naphthylalanine (2), 3-(2-propylpentanamido)-l-phe-
nylalanine (3),[11] 3-((2-propylpentanoyl)oxy)-l-phenylalanine (4), 5-
fluoro-l-tryptophan (5), l-tryptophan (6), l-phenylalanine (7), 3-
((3,4-dihydroxyphenethyl)carbamoyl)-l-phenylalanine (8),[19] d,l-cy-
clopentanealanine (9), 3-cyano-l-phenylalanine (10), l(+)-a-phenyl-
glycine (11), N6-[2-(3-benzoylphenyl)-1-oxopropyl]-l-lysine (12),[6] 5-
methoxy-d,l-tryptophan (13),[7] levodopa (14), 3-methoxy-l-tyro-
sine (15), 4-((2-propylpentanoyl)oxy)-l-phenylalanine (16),[11] 4-(2-
propylpentanamido)-l-phenylalanine (17),[11] d,l-homophenylala-
nine (18), gabapentin (19), N-phenyl-l-glutamine (20), 5-hydroxy-l-
tryptophan (21), N6-[2-(3,4-dihydroxyphenyl)ethyl]-6-oxolysine
(22),[19] a-methyl-l-tyrosine (23), d,l-b-phenylalanine (24), 4-((3,4-di-
hydroxyphenethyl)amino)-4-oxo-l-asparagine (25),[19] 4-methoxy-3-
(2-propylpentanamido)-l-phenylalanine (26), l-alanyl-l-tyrosine
(27), 2-amino-5-methoxybenzoic acid (28), l-2-amino-3-phenylpro-
panol (29), saclofen (30), l-alanyl-l-tyrosyl-l-alanine (31), sulfanila-
mide (32), 2-[2-(3-benzoylphenyl)-1-oxopropoxy]ethyl ester-l-leu-
cine (33), 2-[2-(3-benzoylphenyl)-1-oxopropoxy]ethyl ester-l-phe-
nylalanine (34), diaminopimelic acid (35), N-[2-(3-benzoylphenyl)-1-
oxopropyl]-l-leucine (36), N-[2-(3-benzoylphenyl)-1-oxopropyl]-l-
phenylalanine (37), N2-benzoyL-l-lysine (38), 2-Amino-4-((3-nitro-
phenyl)amino)-4-oxobutanoic acid (39).
Prediction set (n=8): 5-benzoyloxy-l-tryptophan (40), 3-((2-pro-
pylpentanamido)methyl-l-phenylalanine (41), 4-fluoro-3-(2-propyl-
pentanamido)-d,l-phenylalanine (42), 3-(4-hydroxyquinoline-2-car-
boxamido-l-phenylalanine (43), 4-benzoyloxy-l-phenylalanine (44),
(4-hydroxy-(3-(2-propylpentanamido)-l-phenylalanine (45), 4-(4-hy-
droxyquinoline-2-carboxamido-l-phenylalanine (46), 2-propylpenta-
namido-l-lysine (47).
Lastly, compound 47 was designed to examine the flexibility
of LAT1 prodrugs in greater detail. In 47, exchanging the rigid
promoiety for a flexible l-lysine resulted in an almost complete
loss of inhibitory activity (3% l-leucine uptake inhibition).
However, poor affinity does not necessarily result from in-
creased flexibility alone, as replacing aromaticity with aliphatic
features has been shown in previous studies to decrease affini-
ty.[7] Non-aromatic prodrug 47 inhibits l-leucine uptake over
70 percentage units less efficiently than prodrug 12 with an
Virtual screening: Brutus[22,23] was used to identify novel com-
pounds with affinity for LAT1 by considering 3D molecular interac-
tion fields of a compound database. The databases from Asinex
Merged Libraries (n=398926), Bachem (n=4874), Acros (n=
ꢁ 2014 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim
ChemMedChem 2014, 9, 2699 – 2707 2705

CHEMMEDCHEM
FULL PAPERS
www.chemmedchem.org
16545), and Sigma–Aldrich (n=41390) were searched by using 3,
13, and 40 as queries. The query compounds and databases were
pre-processed with the LigPrep package of Schrçdinger Suite
2011.[24] Tautomers and stereoisomers for the database compounds
and query compounds were generated at pH 7.4ꢀ1 using Epik.[25]
Only S configurations were selected for the query compounds. Par-
tial charge calculation and minimization of the molecular structures
in 500 steps were performed using the OPLS_2005 force field.[26]
For Brutus screening, partial charges were assigned with modified
MMFF94 force field as implemented in MOE 2012.17.[27] Default set-
tings for minimum hydrogen bond (0.5), van der Waals (0.7), hydro-
phobic (0.2), and steric (0.5) similarity indices in Brutus were em-
ployed.[22,23] The top 5% of the hit compounds (n=172) were se-
lected for further analysis based on Brutus total similarity and hy-
drophobic, van der Waals, steric, and hydrogen bond similarities,
respectively. After visual inspection, a small selection (n=5) of hit
compounds was selected based on their novelty with respect to
known LAT1-binding compounds and synthetic feasibility in pro-
drug design. None of them had affinity for LAT1.
pounds based on their interaction potential. Conformational sam-
pling of the dataset was performed using LigPrep packages of
Schrçdinger suite 2011.[24] Tautomers and stereoisomers were gen-
erated at pH 7.4 using Ionizer and MMFFs charges. Conformations
were generated using ConfGen and Fast CF protocol,[30] which re-
sulted in a total of 7325 conformations. The lowest-energy confor-
mation of compound 40, which possesses high affinity for LAT1
and is relatively rigid, was selected as the template compound. De-
fault similarity filters were disabled to guarantee the whole data-
base to be aligned with the template. Alternative alignments for
each compound with the template were then ranked according to
total similarity, and the top solution for each compound was se-
lected for further analysis. The alignment of the compound series
was visually checked to identify compounds with poorly aligned
conformations. Alignment was followed by generating electrostatic
and steric interaction fields using Coulomb and Lennard–Jones po-
tentials, respectively.[16] The compounds were positioned in a regu-
lar 3D cubic grid with 2 ꢂ spacing. The interaction energies be-
tween the compound and probe atoms were calculated at each
grid point. Subsequently, the relationships of the intermolecular in-
teraction fields with biological activity ([14C]l-leucine uptake inhibi-
tion) were analyzed by multivariate statistical technique, PLS.[31]
CoMFA models: The structural features of a series of LAT1-binding
compounds (n=47) were correlated with experimentally deter-
mined LAT1 affinity using topomer and classical CoMFA methodol-
ogies.[16,28] Both methods focus on 3D molecular interaction fields
of molecules instead of 2D structures. Consequently, they can
mimic the way a protein binding site recognizes its substrates.
Contrary to classical CoMFA, topomer CoMFA does not require pre-
determined alignment of compounds. This is beneficial for com-
pound series with varying molecular sizes such as our dataset,
which contains small promoiety-sized compounds (<250 Da) and
larger prodrugs (270–420 Da) Topomer CoMFA is also able to eluci-
date the effect of small structural changes on a series of structural-
ly similar compounds more specifically than classical CoMFA. This
results in identifying a common core in the data series and divid-
ing the rest of the molecule into varying parts, i.e., topomers. For
this reason, the effect of different topomers and their effect on bio-
logical activity can be studied in greater detail. In classical CoMFA
the effect of small structural changes on biological activity can be
overlooked more easily because the method perceives the entire
molecule and its molecular interaction fields due to variations in
alignment instead.
Validation of CoMFA models: LOO cross-validation was used to
evaluate the optimal number of components for non-cross-validat-
ed analysis. Each compound of the training was omitted in turn,
and the derived model was used to predict the biological activity
of the omitted compound. The optimal number of components for
non-cross-validated analysis was determined based on the “5%
rule” (Table 2). If adding one component increased Q2 by 5%, the
additional component was considered justified. In general, Q2
values >0.5 were considered statistically significant.[33] Predicted
versus measured affinities were plotted to identify outliers. Com-
pounds 27 and 38 were omitted from the model which increased
the Q2 value and decreased the standard error of prediction to
>5%. The plot of predicted versus measured affinities in logarith-
mic scale (Figure 2) suggested compound 39 as being a potential
outlier, but its inclusion in the training set was needed in order to
produce a model with reasonable internal predictivity. The regres-
sion equation with thousands of coefficients is represented as
CoMFA contour maps, which show regions in space where adding
negative or positive charge or steric bulk either increases or de-
creases ligand activity (Figure 1). Information obtained from the
best model (Model E) was used in further lead optimization to sug-
gest structural modifications that could increase the compounds’
affinity for LAT1. Detailed information on the measured and pre-
dicted activities for training set can be found in the Supporting
Information.
With the topomer approach the compounds were fragmented into
two variable R groups between the single bond connecting Ca
and Cb. The majority (n=31) were automatically fragmented,
whereas four compounds had to be manually fragmented (11, 23,
24, 35) (Table 1). 3D representations of the compounds solely
based on their 2D topology were created using deterministic
rules.[29] Alignment was performed by defining a common core for
the compounds which was followed by a single fragment confor-
mation being generated for each compound. As a result, self-con-
sistent and absolute configurations and 3D conformations were
generated for the compound data set.[29] Four compounds (19, 28,
30, and 32; Table 1) could not be used in topomer CoMFA model-
ing due to the lack of a common core, that is, an amino acid func-
tionality. In total, six models were generated with slightly different
compounds included in the training set (model A: n=34; model B:
n=32, model C: n=33; model D: n=36; model E: n=35, model F:
n=35; Table 2). The best model was selected based on statistics
and external predictivity.
The sensitivity of the CoMFA models to chance correlations was
determined by using progressive scrambling[32] (Table 2) as imple-
mented in Sybyl-X 2.0.[33] Scramblings were performed in 20 indi-
vidual runs to obtain statistically valid results. The model consisting
of two components with dq2/dr2 slope closer to the optimal unity
yy
and smaller error margin than the model with three components
was considered valid and employed in guiding the synthesis of
new compounds whose activity (affinity) was predicted with the
CoMFA models.
The set of synthesized prodrugs (n=8) (Table 3) was used as a pre-
diction set for analyzing the external predictivity of the model. CCC
was determined for the prediction set based on measured affinity
versus predicted affinity.[20,21] Topomer CoMFA models E and D with
CCC>0.85 were considered significant.
In classical CoMFA, the complete series of compounds (n=52) was
aligned with Brutus.[22,23] Brutus considers compounds based on
their 3D molecular interaction fields instead of simple 2D topologi-
cal similarities, which allows it to align structurally diverse com-
ꢁ 2014 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim
ChemMedChem 2014, 9, 2699 – 2707 2706

CHEMMEDCHEM
FULL PAPERS
www.chemmedchem.org
[1] F. Verrey, Pfluegers Arch. 2003, 445, 529–533.
Synthesis
[2] R. J. Boado, J. Y. Li, M. Nagaya, C. Zhang, W. M. Pardridge, Proc. Natl.
Acad. Sci. USA 1999, 96, 12079–12084.
[3] Y. Kanai, Y. Fukasawa, S. H. Cha, H. Segawa, A. Chairoungdua, D. K. Kim,
H. Matsuo, J. Y. Kim, K. Miyamoto, E. Takeda, H. Endo, J Biol Chem. 2000,
275, 20787–20793.
All of the purchased commercial compounds were of analytical
grade or better. The synthesized prodrugs possess >95% purity,
which was determined by elemental analysis (C, H, N). Characteri-
1
zation of the prodrugs was carried out with H and 13C NMR spec-
[4] P. Gomes, P. Soares-da-Silva, Brain Res. 1999, 829, 143–150.
[5] K. C. Cundy, R. Branch, T. Chernov-Rogan, T. Dias, T. Estrada, K. Hold, K.
Koller, X. Liu, A. Mann, M. Panuwat, S. P. Raillard, S. Upadhyay, Q. Q. Wu,
J.-N. Xiang, H. Yan, N. Zerangue, C. X. Zhou, R. W. Barrett, M. A. Gallop, J.
Pharmacol. Exp. Ther. 2004, 311, 315–323.
[6] J. B. van Bree, K. L. Audus, R. T. Borchardt, Pharm. Res. 1988, 5, 369–371.
[7] H. Ylikangas, L. Peura, K. Malmioja, J. Leppꢀnen, K. Laine, A. Poso, M.
Lahtela-Kakkonen, J. Rautio, Eur. J. Pharm. Sci. 2013, 48, 523–531.
[8] M. Gynther, K. Laine, J. Ropponen, J. Leppꢀnen, A. Mannila, T. Nevalai-
nen, J. Savolainen, T. Jarvinen, J. Rautio, J. Med. Chem. 2008, 51, 932–
936.
[9] H. Uchino, Y. Kanai, D. K. Kim, M. F. Wempe, A. Chairoungdua, E. Mori-
moto, M. W. Anders, H. Endou, Mol. Pharmacol. 2002, 61, 729–737.
[10] M. Gynther, A. Jalkanen, M. Lehtonen, M. Forsberg, K. Laine, J. Roppo-
nen, J. Leppꢀnen, J. Knuuti, J. Rautio, Int. J. Pharm. 2010, 399, 121–128.
[11] L. Peura, K. Malmioja, K. Laine, J. Leppꢀnen, M. Gynther, A. Isotalo, J.
Rautio, Mol. Pharm. 2011, 8, 1857–1866.
troscopy and mass spectrometry. Reactions were monitored by
thin-layer chromatography using aluminum sheets coated with
silica gel 60 F245 (0.24 mm) with suitable visualization. Purifications
by flash chromatography were performed using silica. 1H and
13C NMR spectra were recorded on a Bruker Avance 500 spectrom-
eter (Bruker Biospin, Fallanden, Switzerland) operating at 500.13
and 125.75 MHz. The products were also characterized by MS with
a Finnigan LCQ quadrupole ion trap mass spectrometer (Finnigan
MAT, San Jose, CA, USA) equipped with an electrospray ionization
source. Purity was determined by elemental analysis (C, H, N) with
a ThermoQuest CE Instruments EA 1110-CHNS-O elemental ana-
lyzer (CE Instruments, Milan, Italy). Detailed synthetic procedures
and characterization data can be found in the Supporting Informa-
tion.
[12] I. Walker, D. Nicholls, J. W. Irwin, S. Freeman, Int. J. Pharm. 1994, 104,
157–167.
Animal testing
[13] F. P. Bonina, L. Arenare, F. Palagiano, A. Saija, F. Nava, D. Trombetta, P. de
Caprariis, J. Pharm. Sci. 1999, 88, 561–567.
[14] D. M. Killian, S. Hermeling, P. J. Chikhale, Drug Delivery 2007, 14, 25–31.
[15] Q. R. Smith, International Congress Series 1277, 2005, 63–74.
[16] R. D. Cramer, D. E. Patterson, J. D. Bunce, J. Am. Chem. Soc. 1988, 110,
5959–5967.
[17] R. D. Cramer, J. Med. Chem. 2003, 46, 374–388.
[18] Y. Takasato, S. I. Rapoport, Q. R. Smith, Am. J. Physiol. 1984, 247, H484–
H493.
[19] L. Peura, K. Malmioja, K. Huttunen, J. Leppꢀnen, M. Hꢀmꢀlꢀinen, M. M.
Forsberg, M. Gynther, J. Rautio, K. Laine, Pharm. Res. 2013, 30, 2523–
2537.
[20] N. Chirico, P. Gramatica, J. Chem. Inf. Model. 2011, 51, 2320–2335.
[21] N. Chirico, P. Gramatica, J. Chem. Inf. Model. 2012, 52, 2044–2058.
[22] A. J. Tervo, T. Rçnkkç, T. H. Nyronen, A. Poso, J. Med. Chem. 2005, 48,
4076–4086.
Determination of LAT1 affinity: The ability of the investigated
compounds to bind LAT1 was studied by the in situ rat brain per-
fusion technique, which has been described and validated in detail
previously.[8,11,18] The 100% permeability surface area (PA) product
of [14C]l-leucine, a known substrate of LAT1, was determined by
30 s perfusion of 0.2 mCimLꢁ1 [14C]l-leucine (0.64 mm). To determine
the interactions of the investigated compounds with LAT1 at the
BBB, compounds at 100 mm were co-perfused for 30 s with
0.2 mCimLꢁ1 [14C]l-leucine. The PA product of 0.2 mCimLꢁ1 [14C]-
l-leucine co-perfused with the investigated compounds was then
compared with the 100% PA product of 0.2 mCimLꢁ1 [14C]l-leucine.
The LAT1 binding of the investigated compounds is shown by the
decrease in PA product of [14C]l-leucine caused by competitive
binding to LAT1.
[23] T. Rçnkkç, A. J. Tervo, J. Parkkinen, A. Poso, J. Comput. Aid. Mol. Des.
2006, 20, 227–236.
[24] LigPrep ver. 2.5, Schrçdinger LLC, New York, NY (USA), 2012.
[25] Epik ver. 2.2, Schrçdinger LLC, New York, NY (USA), 2011.
[26] W. L. Jorgensen, J. Tirado-Rives, J Am Chem. Soc. 1988, 110, 1657–1666.
[27] Chemical Computing Group Inc., 1010 Sherbooke Street West, Suite
910, Montreal, QC, H3A 2R7 (Canada) 2009.
[28] R. D. Cramer, J. Comput. Aid. Mol. Des. 2012, 26, 805–819.
[29] R. J. Jilek, R. D. Cramer, J. Chem. Inf. Comput. Sci. 2004, 44, 1221–1227.
[30] I. J. Chen, N. Foloppe, J. Chem. Inf. Model. 2010, 50, 822–839.
[31] B. L. Bush, R. B. Nachbar, Jr., J. Comput. Aided Mol. Des. 1993, 7, 587–
619.
Acknowledgements
This work was supported by funding from the Academy of Fin-
land (grant no. 256837), the FinPharma Doctoral Program, the
Magnus Ehrnrooth Foundation, the Jenny and Antti Wihuri Foun-
dation, the Kuopio University Foundation, the Finnish Pharma-
ceutical Society, the Finnish Cultural Foundation, the Orion Re-
search Foundation, the Alfred Kordelin Foundation and the Emil
Aaltonen Foundation. The CSC-IT Center for Science Limited is
acknowledged for providing computing resources, as is Dr. Ewen
MacDonald (University of Eastern Finland) for reviewing the
language of this text.
[32] R. D. Clark, P. C. Fox, J. Comput. Aid. Mol. Des. 2004, 18, 563–576.
[33] Sybyl-X ver. 2.0, Tripos International, 1699 South Hanley Road, St. Louis,
MO, 63144 (USA).
Received: July 31, 2014
Keywords: 3D QSAR · CoMFA · drug design · LAT1 · prodrugs
Published online on September 9, 2014
ꢁ 2014 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim
ChemMedChem 2014, 9, 2699 – 2707 2707