Binding of serotonin to the human serotonin transporter. Molecular modeling and experimental validation

Article abstract of DOI:10.1021/ja076403h

Molecular modeling and structure-activity relationship studies were performed to propose a model for binding of the neurotransmitter serotonin (5-HT) to the human serotonin transporter (hSERT). Homology models were constructed using the crystal structure of a bacterial homologue, the leucine transporter from Aquifex aeolicus, as the template and three slightly different sequence alignments. Induced fit docking of 5-HT into hSERT homology models resulted in two different binding modes. Both show a salt bridge between Asp98 and the charged primary amine of 5-HT, and both have the 5-HT C6 position of the indole ring pointing toward Ala173. The difference between the two orientations of 5-HT is an enantiofacial discrimination of the indole ring, resulting in the 5-hydroxyl group of 5-HT being vicinal to either Ser438/Thr439 or Ala169/Ile172/Ala173. To assess the binding experimentally, binding affinities for 5-HT and 17 analogues toward wild type and 13 single point mutants of hSERT were measured using an approach termed paired mutant-ligand analogue complementation (PaMLAC). The proposed ligand-protein interaction was systematically examined by disrupting it through site-directed mutagenesis and reestablishing another interaction via a ligand analogue matching the mutated residue, thereby minimizing the risk of identifying indirect effects. The interactions between Asp98 and the primary amine of 5-HT and the interaction between the C6-position of 5-HT and hSERT position 173 was confirmed using PaMLAC. The measured binding affinities of various mutants and 5-HT analogues allowed for a distinction between the two proposed binding modes of 5-HT and biochemically support the model for 5-HT binding in hSERT where the 5-hydroxyl group is in close proximity to Thr439.

Full text of DOI:10.1021/ja076403h

Published on Web 03/04/2008
Binding of Serotonin to the Human Serotonin Transporter.
Molecular Modeling and Experimental Validation
Leyla Celik,§ Steffen Sinning,‡ Kasper Severinsen,‡ Carsten G. Hansen,‡
Maria S. Møller,† Mikael Bols,† Ove Wiborg,*,‡ and Birgit Schiøtt*,§
The iNANO and inSPIN Centers, the Department of Chemistry, UniVersity of Aarhus, Aarhus,
Denmark, and the Laboratory of Molecular Neurobiology, Centre for Psychiatric Research,
Aarhus UniVersity Hospital, RisskoV, Denmark
Received August 25, 2007; E-mail: owiborg@post.tele.dk; birgit@chem.au.dk
Abstract: Molecular modeling and structure-activity relationship studies were performed to propose a
model for binding of the neurotransmitter serotonin (5-HT) to the human serotonin transporter (hSERT).
Homology models were constructed using the crystal structure of a bacterial homologue, the leucine
transporter from Aquifex aeolicus, as the template and three slightly different sequence alignments. Induced
fit docking of 5-HT into hSERT homology models resulted in two different binding modes. Both show a salt
bridge between Asp98 and the charged primary amine of 5-HT, and both have the 5-HT C6 position of the
indole ring pointing toward Ala173. The difference between the two orientations of 5-HT is an enantiofacial
discrimination of the indole ring, resulting in the 5-hydroxyl group of 5-HT being vicinal to either Ser438/
Thr439 or Ala169/Ile172/Ala173. To assess the binding experimentally, binding affinities for 5-HT and 17
analogues toward wild type and 13 single point mutants of hSERT were measured using an approach
termed paired mutant-ligand analogue complementation (PaMLAC). The proposed ligand-protein
interaction was systematically examined by disrupting it through site-directed mutagenesis and re-
establishing another interaction via a ligand analogue matching the mutated residue, thereby minimizing
the risk of identifying indirect effects. The interactions between Asp98 and the primary amine of 5-HT and
the interaction between the C6-position of 5-HT and hSERT position 173 was confirmed using PaMLAC.
The measured binding affinities of various mutants and 5-HT analogues allowed for a distinction between
the two proposed binding modes of 5-HT and biochemically support the model for 5-HT binding in hSERT
where the 5-hydroxyl group is in close proximity to Thr439.
1. Introduction
transmembrane (TM) R-helices with the N- and C-termini
located intracellularly.6 Experimental data10,11 have confirmed
the predicted topology. The first crystal structure of a bacterial
homologue of the NSS proteins, the Aquifex aeolicus leucine
transporter, LeuTAa, was published in 200512 and new structures
of LeuTAa with bound TCAs, have recently appeared.13,14 The
structures show a novel fold for a membrane bound transporter
with 12 TM helices where TM1-5 and TM6-10 are related
by a pseudo 2-fold axis having TM1 and TM6 unwound in the
middle to make up the binding site along with TM3 and TM8,
both of which are kinked around the binding site.12 The
amphiphilic binding site contains a substrate leucine residue
along with two sodium ions, one of which is directly coordinated
The serotonin transporter (SERT), responsible for reuptake
of serotonin (5-hydroxytryptamine, 5-HT) into the presynaptic
neuron, is the primary target for treatment of anxiety and
depression1-3 and belongs to the monoamine transporter sub-
family of neurotransmitter sodium symporters (NSSs).4 Depres-
sion can be treated by tricyclic antidepressants (TCAs), as
imipramine and amitriptyline, or by selective serotonin reuptake
inhibitors (SSRIs) such as citalopram, paroxetine, fluoxetine,
sertraline, and fluvoxamine.5,6 SERT is also a target for drugs
of abuse, e.g., 3,4-methylenedioxymethamphetamine (ecstasy)7
and cocaine.8
The 3D structure of SERT is so far unknown.9 However, from
hydropathy plots it has been predicted to consist of 12
(7) Trigo, J. M.; Renoir, T.; Lanfumey, L.; Hamon, M.; Lesch, K.; Robledo,
P.; Maldonado, R. Biol. Psych. 2007, 62, 669-679.
(8) Sora, I.; Hall, F. S.; Andrews, A. M.; Itokawa, M.; Li, X.; Wei, H.;
Wichems, C.; Lesch, K.; Murphy, D. L.; Uhl, G. R. Proc. Natl. Acad. Sci.
U.S.A. 2001, 98, 5300-5305.
§ iNANO and inSPIN Centers, Department of Chemistry.
‡ Centre for Psychiatric Research.
† Department of Chemistry.
(9) Rudnick, R. J. Membr. Biol. 2006, 213, 101-110.
(10) Androutsellis-Theotokis, A.; Rudnick, G. J. Neurosci. 2002, 22, 8370-
8378.
(1) Fuller, R. W.; Wong, D. T. Ann. N. Y. Acad. Sci. 1990, 600, 68-80.
(2) Hahn, M. K.; Blakely, R. D. Pharmacogen. J. 2002, 2, 217-235.
(3) Owens, M. J.; Nemeroff, C. B. Clin. Chem. 1994, 40, 288-295.
(4) Nelson, N. J. Neurochem. 1998, 71, 1785-1803.
(5) Oh, S. J.; Ha, H. J.; Chi, D. Y.; Lee, H. K. Curr. Med. Chem. 2001, 8,
999-1034.
(11) Chen, J.; Liu-Chen, S.; Rudnick, G. J. Biol. Chem. 1998, 273, 12675-
12681.
(12) Yamashita, A.; Singh, S. K.; Kawate, T.; Jin, Y.; Gouaux, E. Nature 2005,
437, 215-223.
(6) Ramamoorthy, S.; Bauman, A. L.; Moore, K. R.; Han, H.; Yang-Feng, T.;
Chang, A. S.; Ganapathy, V.; Blakely, R. D. Proc. Natl. Acad. Sci. U.S.A.
1993, 90, 2542-2546.
(13) Singh, S. K.; Yamashita, A.; Gouaux, E. Nature 2007, 448, 952-956.
(14) Zhou, Z.; Zhen, J.; Karpowich, N. K.; Goetz, R. M.; Law, C. J.; Reith, M.
E. A.; Wang, D.-N. Science 2007, 317, 1390-1393.
9
10.1021/ja076403h CCC: $40.75 © 2008 American Chemical Society
J. AM. CHEM. SOC. 2008, 130, 3853-3865
3853

A R T I C L E S
Celik et al.
to the substrate. It has been suggested that this sodium ion (Na1)
is cotransported, while the second ion (Na2) may play a
structural role.12 Toward the extracellular side of the transporter
the binding site is only closed by an aromatic lid (residues
Tyr108 and Phe252) and a salt bridge (residues Arg30 and
Asp404). The LeuTAa structures with bound TCAs have these
located outside the salt bridge. On the basis of the occluded
conformation of the LeuTAa structure, the facilitated transport
is suggested to consist of at least three different conformations
of the protein12 to follow the alternate access theory for
transporters.15 The high-resolution 3D structure of LeuTAa with
bound substrate leucine has paved an avenue toward the
construction of homology models of the neurotransmitter
transporters and further medicinal chemistry approaches.9
The observation by Chotia and Lesk16 that 3D structure is a
more conserved property during evolution than is the sequence
of proteins belonging to the same family, supports the use of
LeuTAa as a template for homology modeling of the human
NSSs despite low overall sequence identity.9,12 As pointed out
by Yamashita et al.12 an interesting difference between LeuTAa
and monoamine transporters is the substitution of a glycine
residue (Gly24) in the LeuTAa binding site to an acidic residue
in the monoamine transporters. They propose that the carboxyl-
ate in the substrate leucine in LeuTAa is replaced by a
carboxylate functionality in the NSSs at the position corre-
sponding to Gly24 in LeuTAa; for SERT this is Asp98, hereby
complementing the difference in substrate between LeuTAa (an
amino acid) and SERT (an amine). This is a classic example of
the deletion model as proposed for the evolution of selective
receptors.17 A further indication that the LeuTAa structure is a
good template for modeling of NSS proteins is found in an
elegant study by Dodd and Christie.18 On the basis of the LeuTAa
structure and sequence alignments of the creatine and γ-ami-
nobutyric acid (GABA) transporters, both of which also belong
to the NSS family, they were able to change the substrate
specificity of the creatine transporter by mutating a few residues
in the putative binding pocket of the creatine transporter to their
respective counterparts of the GABA transporter thereby switch-
ing it to a transporter selective for GABA.18 This suggests that
substrate specificity of the transporters is defined by the actual
amino acids lining the central binding cavity and that this central
cavity is common to the members of the NSS family.
In order to eliminate the possibility of observed trends in
measured binding affinities being due to allosteric effects, we
decided to explore the binding of 5-HT in human SERT
(hSERT) in a systematic and quantitative manner by using
molecular modeling to rationally design pairs of protein mutants
and substrate analogues. A structural change of a ligand that
affects its binding affinity can be reversed by a specific,
complementing, mutation of the transporter. We call this
experimental paradigm “paired mutant-ligand analogue com-
plementation” (PaMLAC) and find that it alleviates several of
the reservations associated with mutational analysis of ligand
binding by supplying a second layer of experimental evidence
and additionally firmly establish which part of the ligand
interacts with which residue in the transporter. The PaMLAC
paradigm relies heavily on the availability of a wide ensemble
of ligand analogues and protein mutants but also very much on
reliable structural molecular models to provide testable predic-
tions of protein-ligand interactions. The idea of coupling protein
mutational studies with binding affinities for substrate analogues
has previously been attempted by Strader et al. in the 1980s22,23
in a study of the â-adrenergic receptor. However, they did not
have a 3D model from which rational predictions could be made
and therefore did not explore the interactions systematically.
To the best of our knowledge, resolution of ligand-protein
interactions in hSERT to such a degree of detail has been very
scarce in the literature, a study from Barker et al. being the
most comprehensive so far.19
To predict the binding of substrate 5-HT to hSERT, we first
have to build a homology model of hSERT based on the
template LeuTAa structure and then predict the structure and
orientation of the substrate in the binding cavity by using
molecular docking. This approach possesses several computa-
tional challenges as are evident from the preliminary homology
models of NSS proteins that have appeared in the literature.24-30
In these studies a ligand was modeled in the binding site of the
selected transporters by a manual placement since the use of
automated docking procedures for this purpose so far has failed.
A comprehensive examination of a model describing the binding
orientation of 5-HT in the binding site of hSERT, including
biochemical validation, has not yet appeared in the literature.
Homology modeling of membrane proteins is, compared to
homology modeling of globular water soluble proteins, a new
research area, as the number of templates is rather limited.
However, it has recently been shown that homology modeling
methods are as applicable to membrane proteins as they are to
water soluble proteins,31 meaning that good quality homology
Site-directed mutagenesis has previously been employed in
numerous studies in an attempt to identify residues lining the
primary ligand binding site in SERT, as well as in the dopamine
transporter (DAT) and the norepinephrine transporter (NET).
Some have been successful, but after the publication of the
12
structure of LeuTAa a significant proportion of the residues
(22) Strader, C. D.; Sigal, I. S.; Candelore, M. R.; Rands, E.; Hill, W. S.; Dixon,
R. A. F. J. Biol. Chem. 1988, 263, 10267-10271.
suggested to interact directly with a ligand are more likely to
influence ligand binding in an indirect manner. However, with
the mutagenesis studies and the LeuTAa structure at hand it
seems very evident that Asp9819 and Ala16920 interact with
5-HT, and in addition Ile17220,21 is likely to interact with some
inhibitors.
(23) Strader, C. D.; Candelore, M. R.; Hill, W. S.; Sigal, I. S.; Dixon, R. A. F.
J. Biol. Chem. 1989, 264, 13572-13578.
(24) Zomot, E.; Bendahan, A.; Quick, M.; Zhao, Y.; Javitch, J.; Kanner, B.
Nature 2007, 449, 726-730.
(25) Beuming, T.; Shi, L.; Javitch, J. A.; Weinstein, H. Mol. Phamacol. 2006,
70, 1630-1642.
(26) Jørgensen, A. M.; Tagmose, L.; Jørgensen, A. M. M.; Topiol, S.; Sabio,
M.; Gundertofte, K.; Bøgesø, K. P.; Peters, G. H. ChemMedChem 2007,
2, 815-826.
(15) Jardetzky, O. Nature 1966, 211, 969-970.
(27) Paczkowski, F. A.; Sharpe, I. A.; Dutertre, S.; Lewis, R. J. J. Biol. Chem.
2007, 282, 17837-17844.
(16) Chotia, C.; Lesk, A. M. EMBO J. 1986, 5, 823-826.
(17) Topiol, S. Trends Biochem. Sci. 1987, 12, 419-421.
(18) Dodd, J. R.; Christie, D. L. J. Biol. Chem. 2007, 282, 15528-15533.
(19) Barker, E. L.; Moore, K. R.; Rakhshan, F.; Blakely, R. D. J. Neurosci.
1999, 19, 4705-4717.
(28) Ravna, A. W.; Sylte, I.; Kristiansen, K.; Dahl, S. G. Bioorg. Med. Chem.
2006, 14, 666-675.
(29) Sen, N.; Shi, L.; Beuming, T.; Weinstein, H.; Javitch, J. A. Neuropharm.
2005, 49, 780-790.
(20) Larsen, M. B.; Elfving, B.; Wiborg, O. J. Biol. Chem. 2004, 279, 44147-
44156.
(21) Henry, L. K.; DeFelice, L. J.; Blakely, R. D. Neuron 2006, 49, 791-796.
(30) Forrest, L. R.; Tavoulari, S.; Zhang, Y.-W.; Rudnick, G.; Honig, B. Proc.
Natl. Acad. Sci. U.S.A. 2007, 104, 12761-12766.
(31) Forrest, L. R.; Tang, C. L.; Honig, B. Biophys. J. 2006, 91, 508-517.
9
3854 J. AM. CHEM. SOC. VOL. 130, NO. 12, 2008

Binding of Serotonin to hSERT
A R T I C L E S
Chart 1. (A) 5-HT and 17 Analogues Used in the PaMLAC
models can be expected if the sequence identity in the TM
regions is approximately 30% or above and the template
structure is of high resolution (meaning below 2.0 Å in the TM
regions). If these criteria are fulfilled, the homology models
generated are expected to have a CR-RMSD (root-mean-square
deviation) of ∼2 Å from the native structure of the protein.31
Specifically, it was found that the TM regions of homology
models of membrane proteins are modeled with higher accuracy
than the domains of the protein not embedded in the membrane,
possibly due to the fact that the surrounding lipids in the
membrane restrict the flexibility of the TM regions.31 In our
case these two prerequisites are fulfilled as the structure of
LeuTAa was solved at a resolution of 1.65 Å, and the sequence
identities between LeuTAa and NNS proteins of about 30% in
the TM regions were found by Yamashita et al.12 The next
computational challenge is to predict the orientation of the
cognate substrate, 5-HT, in the binding cavity of the produced
homology models of hSERT. Over the past decade, molecular
docking has developed into a method of great use in medicinal
chemistry,32,33 and it works well in predicting the structure of
the bound ligand in the protein, especially if a consensus docking
approach is utilized.34 However, so far molecular docking is
not very good at predicting accurate binding affinities by use
of the scoring functions, and more advanced methods must be
applied.35 Recently, it has been shown that molecular docking
to homology models36 of a protein is a good choice in virtual
screening studies when no structure of the target is available.
Some care must be taken, however, especially related to the
orientation and flexibility of amino acid side chains lining the
binding cavity. The binding cavity in the homology models of
hSERT is expected to be rather tight for the larger substrate in
hSERT compared to the leucine substrate in LeuTAa. This is
partly overcome by the fact that some of the amino acids lining
the binding cavity of hSERT are smaller than those in LeuTAa,
thereby giving more space for 5-HT to fit. To further allow for
a flexible binding cavity in order to accommodate 5-HT, we
included protein flexibility by using the newly developed
induced fit docking (IFD) methodology.37 This protocol is
among the first methods for including some degree of protein
flexibility during a molecular docking simulation.
Studies. Numbering of Important Atoms in the Indole Ring Is
Indicated on 5-HT. (B) Selected Drugs Targeting hSERT; â-CIT )
2â-2-carboxymethoxy-3-(4-[125I]iodophenyl)tropane
In this study we present a systematic exploration of the
hSERT binding site and outline a preferred binding mode of
5-HT as obtained from IFD computations into carefully selected
homology models of hSERT. The obtained models for ligand-
protein interactions were challenged by the PaMLAC approach
using 18 tryptamine analogues, Chart 1A, against wild type (wt)
hSERT and 13 rationally chosen single point mutated transport-
ers.
2. Computational Methods
2.1. Sequence Alignment. The amino acid sequence for hSERT was
acquired from the UniProt Database,38 accession number P31645. The
sequence of LeuTAa was extracted from pdb-entry code 2A65,12
downloaded from the Protein Data Bank.39 The long intracellular
N-terminal of hSERT was not included in the homology modeling,
only residues 79-630 covering the 12 R-helixes with intervening loops
(32) Chen, H.; Lyne, P. D.; Giordanetto, F.; Lovell, T.; Li, J. J. Chem. Inf.
Model. 2006, 46, 401-415.
(33) Cummings, M. D.; DesJarlais, R. L.; Gibbs, A. C.; Mohan, V.; Jaeger, E.
P. J. Med. Chem. 2005, 48, 962-976.
(34) Yang, J.-M.; Chen, Y.-F.; Shen, T.-W.; Kristal, B. S.; Hsu, D. F. J. Chem.
Inf. Model. 2005, 45, 1134-1146.
(35) Warren, G. L.; Andrews, C. W.; Capelli, A.-M.; Clarke, B.; LaLonde, J.;
Lambert, M. H.; Lindvall, M.; Nevins, N.; Semus, S. F.; Senger, S.;
Tedesco, G.; Wall, I. D.; Woolven, J. M.; Peishoff, C. E.; Head, M. S. J.
Med. Chem. 2006, 49, 5912-5931.
(38) The UniProt Consortium. Nucl. Acids Res. 2007, 35, D193-D197.
(39) Berman, H. M.; Battistuz, T.; Bhat, T. N; Bluhm, W. F.; Bourne, P. E.;
Burkhardt, K., Feng, Z.; Gilliland, G. L.; Iype, L.; Jain, S.; Fagan, P.; Marvin
J.; Padilla, D.; Ravichandran, V.; Schneider, B.; Thanki, N.; Weissig, H.;
Westbrook, J. D.; Zardecki, C. Acta Crystallogr., Sect. D: Biol. Crystallogr.
2002, 58, 899-907.
(36) Kairys, V.; Fernandes, M. X.; Gilson, M. K. J. Chem. Inf. Model. 2006,
46, 365-379.
(37) Sherman, W.; Day, T.; Jacobson, M. P.; Friesner, R. A.; Farid, R. J. Med.
Chem. 2006, 49, 534-553.
9
J. AM. CHEM. SOC. VOL. 130, NO. 12, 2008 3855

A R T I C L E S
Celik et al.
Table 1. hSERT Homology Models Built Using Alignments A, B,
and C. Structural and Energetic Data for the 15 Models Are Listed
Ramachandran coordinates are optimized during the homology model-
ing procedure, it is obvious that the models with lowest DOPE score
and objective function within one sequence alignment will have the
best possible Ramachandran plots. In this way, the different alignments
can be evaluated by comparing the computed Ramachandran scores;
the better the alignment, the more favorable Ramachandran plots are
expected. The volume of the binding cavity was measured by the
Molegro Virtual Docker suite of programs, using the solvent accessible
surface method with a maximum of 25 cavities per model.45 Sodium
ions were included by manual placement in the same relative place as
they are located in the LeuTAa structure. A chloride ion was manually
inserted between residues Tyr121, Ser336, Asn368, and Ser372 in a
few models (please see Supporting Information). The four residues and
the Cl- ion were minimized in MacroModel version 9.146 with the
OPLS force field.
homology Ramachandran
DOPE
distance
GlideScore
volume of
model
plot (%)a
(kJ/mol)
D98(O
δ)−
Na1 (Å)b (kcal/mol)c binding site (Å3)
A-1
A-2
A-3
A-4
A-5
B-1
B-2
B-3
B-4
B-5
C-1
C-2
C-3
C-4
C-5
94.9
94.6
94.7
95.7
94.7
93.8
93.5
94.0
94.6
94.4
95.3
94.4
95.3
94.9
96.4
-78607
-78374
-78040
-78360
-78627
-77388
-76433
-78284
-77728
-77268
-79159
-78962
-79844
-79205
-79703
4.1
90.1
101.9
84.0
128.0
100.9
46.6
67.6
36.9
67.6
61.4
3.3
3.5
3.3
3.3
-6.6
3.2
4.2
62.4
114.2
115.2
68.0
3.3
3.3
3.0
2.3. Ligand Modeling. 5-HT was drawn in Maestro46 as charged
on the primary amino group and minimized with the OPLS 2005 force
field as implemented in MacroModel 9.146 for 10 000 steps of conjugate
gradient iterations or until convergence, according to default settings.
-6.7
143.9
a Percentage of residues with psi- and phi-values in the “favored” or
“most favored regions” of the Ramachandran plot. b Only distances <5.0
Å are included. c GlideScore computed for docking of 5-HT in a rigid protein
in Glide 4.0 using the SP scoring function.
2.4. Docking. Initial docking of 5-HT in the 15 homology models
of hSERT was performed with Glide47,48 version 4.0 using standard
procedures.46 This docking, with a rigid protein and a flexible ligand,
was generally performed with the default standard precision (SP) scoring
function in Glide;47 however, the Glide extra precision (XP) scoring
function49 was tested for two models to evaluate if the special extra
terms in this scoring function is of importance for this particular
protein-ligand system. The binding site was defined from two residues
(Asp98 and Ile172) which make up the two ends of the binding cavity
parallel to the membrane plane.
and the short C-terminal were included. Three alignments of the target
sequence of hSERT to the template, LeuTAa, are evaluated. The first
alignment, A, is based on an automatic alignment created in MOD-
ELLER (version 8.1),40 which uses the align2d methodology.41 Ad-
ditional manual adjustments were performed to eliminate unwanted gaps
in the produced alignment for hSERT and to incorporate knowledge
from available experimental data placing Asp9819 and Ala169,20 in the
binding site by iteratively evaluating the DOPE (Discrete Optimized
Protein Energy) scores42 after model building. Ile172 has been shown
to be important for binding of inhibitors to hSERT,20,21 and its sequence
proximity to Ala169 also places this residue in the binding cavity. Ile172
corresponds to Val104 in LeuTAa which is indeed placed in the
hydrophobic part of the binding cavity in LeuTAa.12 No further attempts
were done to refine the extra- and intracellular loops or the C-terminal,
since they are distant from the transmembrane region and the putative
ligand binding cavity (more than 20 Å), which is the focus of the present
study, and thus are probably not affecting the binding of the substrate.
The second alignment, B, was published with the LeuTAa structure12
and is used for generating the second set of homology models of
hSERT. The third alignment, C, stems from a study by Beuming et
al.25 where experimental data was included for all known proteins in
the NNS family to propose a detailed alignment.
2.2. Model Building. Five homology models, 1-5, were built for
each of the three alignments, A-C, using MODELLER in stand-alone
mode and default settings.40 The produced 15 homology models, A1-
5, B1-5, and C1-5, were evaluated by visual inspection (with respect
to reproducing the protein backbone in the 12 TM helices, especially
the kinks in TM1 and TM6) and by quantitative examinations through
the Objective Function, DOPE scores, the volume of the binding site,
and sterical features through Ramachandran plots (Table 1). The
MODELLER Objective Function describes how well the model fits
with all the input structural data,43 and the DOPE-score42 evaluates,
per residue basis, the quality of the model against the template. The
objective function and DOPE scores are energetic measurements and
should thus both be as low as possible.42,43 Stereochemical features of
the obtained models were examined from Ramachandran plots generated
in VMD 1.8.4 using the ramaplot plugin, version 1.0.44 As the
2.5. Induced Fit Docking. To fully explore the concerns about
flexible amino acid side chains when performing molecular docking
simulations into homology models, we decided to include protein
flexibility in the docking protocol using one homology model of hSERT
from each alignment. Models of hSERT that allowed docking of 5-HT
into a rigid protein were chosen for alignments A and C, whereas for
B, where no poses were generated during normal docking, the homology
model with the best energy scores and an interaction between Na1 and
Asp98 was selected for further evaluation, resulting in selection of
homology models A-4, B-3, and C-5 for further calculations. The IFD
protocol37,46 from Schro¨dinger Inc. that combines Glide 4.046 and Prime
1.546 was employed for docking of 5-HT into hSERT allowing for
protein flexibility.37,50 The IFD workflow consists of three steps. First,
the ligand is docked flexibly into a rigid protein with a soft Van der
Waals potential, thereby allowing for some steric clash. During this
step it is possible to make point mutations of highly flexible residues
to alanine, to artificially create more room and secure that at least a
few poses of the protein-ligand complex are generated; this was not
necessary for introducing 5-HT into the hSERT binding site. In the
second step during an IFD workflow the protein is optimized, using
Prime, within 5.0 Å of the ligand poses from the first step, and if any
residues were mutated in the first step, they are reintroduced. The final
step is a redocking of the ligand into the relaxed protein binding cavity,
where normal Van der Waals terms are used. Either the SP or XP
scoring function can be applied in the last step. The final scoring from
an IFD calculation is the computed GlideScore (SP or XP) reflecting
the interaction between the protein and the ligand. Another reported
(45) Thomsen, R.; Christensen, M. H. J. Med. Chem. 2006, 49, 3315-3321.
(46) Schro¨dinger, LLC, 2006.
(47) Friesner, R. A.; Banks, J. L.; Murphy, R. B.; Halgren, T. A.; Klicic, J. J.;
Mainz, D. T.; Repasky, M. P.; Knoll, E. H.; Shelley, M.; Perry, J. K.;
Shaw, D. E.; Francis, P.; Shenkin, P. S. J. Med. Chem. 2004, 47, 1739-
1749.
(40) Sali, A.; Blundell, T. L. J. Mol. Biol. 1993, 234, 779-815.
(41) Sanchez, R.; Sali, A. Proc. Natl. Acad. Sci. U.S.A. 1998, 95, 13597-13602.
(42) Eramian, D.; Shen, M.; Devos, D.; Melo, F.; Sali, A.; Marti-Renom, M.
A. Protein Sci. 2006, 15, 1653-1666.
(48) Halgren, T. A.; Murphy, R. B.; Friesner, R. A.; Beard, H. S.; Frye, L. L.;
Pollard, W. T.; Banks, J. L. J. Med. Chem. 2004, 47, 1750-1759.
(49) Friesner, R. A.; Murphy, R. B.; Repasky, M. P.; Frye, L. L.; Greenwood,
J. R.; Halgren, T. A.; Sanschagrin, P. C.; Mainz, D. T. J. Med. Chem.
2006, 49, 6177-6196.
(43) Eswar, N.; John, B.; Mirkovic, N.; Fiser, A.; Ilyin, V. A.; Pieper, U.; Stuart,
A. C.; Marti-Renom, M. A.; Madhusudhan, M. S.; Yerkovich, B.; Sali, A.
Nucl. Acids Res. 2003, 31, 3375-3380.
(44) Humphrey, W.; Dalke, A.; Schulten, K. J. Mol. Graph. 1996, 14, 33-38.
(50) Teague, S. J. Nat. ReV. Drug DiscoVery 2003, 2, 527-541.
9
3856 J. AM. CHEM. SOC. VOL. 130, NO. 12, 2008

Binding of Serotonin to hSERT
A R T I C L E S
Table 2. IFD Simulations of 5-HT in the Three Homology Models of hSERT
setup
homology model
sodium
binding site
scoring function
poses
cluster1
cluster2
cluster3
outlier
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
A-4
A-4
A-4
A-4
A-4
A-4
B-3
B-3
B-3
B-3
B-3
C-5
C-5
C-5
C-5
C-5
C-5
C-5
none
Na1
Na2
Na1, Na2
Na1
Na1, Na2
none
Na1
D98, I172
D98, I172
D98, I172
D98, I172
5-HTa
SP
SP
SP
SP
7
10
9
9
9
2
5
4
6
7
4
2
4
1
3
2
2
2
1
3
1
1
SP
D98, I172
D98, I172
D98, I172
D98, I172
D98, I172
D98, I172
D98, I172
D98, I172
D98, I172
D98, I172
5-HTa
SP/XPb
SP
7
1
1
3c
3c
6c
1c
1c
9
SP
SP
SP
Na2
Na1, Na2
Na1, Na2
none
Na1
Na2
Na1, Na2
Na1
Na1
SP/XPb
SP
7
7
7
6
6
10
5
76
2
2
3
1
1
SP
SP
SP
SP
10
10
8
7
10
6
1
5
1
5-HTd
D98, I172
SP
Na1, Na2
SP/XPb
total
1
16
125
15
a 5-HT as docked in the rigid homology model with the SP scoring function was used for definition of the binding site in IFD. b The XP-scoring function
is used in the redocking step of the IFD simulation. c All but one pose in model B-3 are placed outside the binding cavity. d 5-HT as docked in the rigid
homology model with the XP scoring function was used for definition of the binding site in IFD.
B12 and C.25 A figure of the three alignments against each other
score is the IFDScore, which is an empirical measure of the energetics
of the total protein-ligand complex. It is a sum of the reported G-score
and 5% of the Prime energy, which is a force field-calculated energy
of the protein after the refinement in step two of the IFD workflow.37
Different setups were created examining the effect of the three
homology models, the presence of the two sodium ions, the definition
of the binding site, and the scoring function utilized in step three of
IFD. In total, 18 different IFD calculations were carried out as listed
in Table 2.
and the LeuTAa template is included in the Supporting Informa-
tion. Our alignment was manually refined to remove gaps from
within the transmembrane segments and to place Asp98,19
Ala169,20 and Ile17220,21 in the putative binding site. This was
a procedure including several iterations of (i) refining the
alignment, (ii) building models, (iii) comparing DOPE scores42
between template and target for individual residues, and (iv)
visually inspecting the binding cavity. When comparing the
DOPE score for individual residues between the template and
the model, a high similarity generally implies a good model,
allowing us to further refine the alignment. In the final
alignment, A, the highest differences in DOPE scores were
located in loops outside the membrane layer. These have not
been further refined at this point since they are located more
than 20 Å from the binding pocket; therefore, their detailed
structure is not expected to influence ligand binding. Figures
of DOPE scores for the three alignments are included in the
Supporting Information. The three alignments show sequence
identities for hSERT against LeuTAa of 23%, 20%, and 19%,
respectively, for the full protein and about 28% sequence identity
in the TM regions. The differences between the three alignments
are located in R-helices TM4, TM5, TM9, and TM12 all of
which are located distant from the binding cavity. If conservative
mutations of amino acids are considered, sequence similarities
of 55-56% are observed between LeuTAa and hSERT. Within
5 Å of the binding site the identity is even higher (19 out of 25
residues ) 76%), while more differences are found in the non-
membrane embedded loop regions. Since the LeuTAa structure
was solved to 1.65 Å12 and computed sequence identity numbers
between template and target are around 30%, one can expect
the resulting homology models to be of a good quality with
around 2 Å deviations between CR-atoms of the models and
the true native structure of hSERT.31
3. Experimental Methods
3.1. Ligand Synthesis. Compounds 1C (Gramine), 2A (Tryptamine),
keto-2C, 5-F-R-MeT, 5-HT, 5-MeT, 5-MeOT, 5,7-diHT, 6-FT, 6-MeOT,
and 7-MeT were purchased from Sigma-Aldrich, 1A from Maybridge,
and 6-HT from National Institute of Mental Health’s Chemical
Synthesis and Drug Supply Program; all were used as received.
Compounds 1B, 2C, 3A, 3B, and 3C were synthesized by us (see the
Supporting Information).
3.2. Site-Directed Mutagenesis. Site-directed mutagenesis was
performed using the method of mismatched complimentary primer pairs
in a polymerase reaction with Phusion (Finnzyme) using hSERT in
the pCDNA3 (Invitrogen) vector as template. Mutants were verified
and checked for unwanted secondary mutations by automated DNA
sequencing using Big Dye 3.1 (Applied Biosystems) chemistry. For
further details see the Supporting Information.
3.3. Cell Culture. HEK-293 MSR cells (Invitrogen) were cultured
as monolayer cultures in DMEM (BioWhitaker) supplemented with
10% FCS (Gibco Life Technologies), 100 U/mL penicillin, 100 µg/
mL streptomycin (BioWhitaker), and 6 µg/mL of Geneticin (Invitrogen)
at 95% humidity and 5% p(CO2) at 37 °C. Cells were detached from
the culture flasks by Versene (Invitrogen) and trypsin/EDTA (Bio-
Whitaker) treatment for subculturing or seeding into microtiter plates.
3.4. Uptake Assay. Measurement of 5-HT uptake was performed
as described by Larsen et al.20 HEK-293 MSR cells (Invitrogen) were
used instead of COS-1 cells. For further details see the Supporting
Information.
4. Results
The structures of the 15 homology models are highly similar
and all include the important feature found in the LeuTAa
structure;12 helices TM1 and TM6 are unwound in the middle
to allow for substrate binding in an occluded binding site
between helices TM1, TM3, TM6, and TM8 (Figure 1). Because
4.1. Modeling. 4.1.2. Sequence Alignment and Model
Building. A total of 15 homology models (Table 1) were built
in MODELLER40 from three different alignments: one devel-
oped by us (A) and two that have previously been published,
9
J. AM. CHEM. SOC. VOL. 130, NO. 12, 2008 3857

A R T I C L E S
Celik et al.
et al.12 Since B-3 has a lower DOPE score than B-1 and no
significant differences could be detected around the binding
cavities of the two models, B-3 was chosen for the IFD studies,
resulting in one model per alignment. The three alignments were
identical around the binding cavity, which indicates that the
differences in the measured volumes must originate from
differences in the side chain positioning, as during the building
of the homology models in MODELLER, no major backbone
manipulations away from the template structure is possible.
Since IFD is employed in the next stage of our theoretical
prediction of a binding geometry of 5-HT in hSERT, the exact
rotamer state of the side chain is not maintained during the
docking simulation, and it may thus be less important which
homology model was initially chosen for each of the three
alignments, the important feature being that the ligand can be
docked in the first step of the IFD workflow.
4.1.3. Induced Fit Docking. IFD37,46 of 5-HT into hSERT
was performed to compensate for the rather small binding site
in the homology models in an unbiased fashion that is likely to
be consistent with natural protein dynamics upon ligand binding.
To examine the favored binding mode of 5-HT, several aspects
were examined through 18 different setups as listed in Table 2,
thereby introducing a kind of consensus scoring scheme for the
study. The preference of the two sodium ions from the LeuTAa
structure12 was examined in four setups for each homology
model, setups 1-4, 7-10, and 12-16, respectively. The
influence of a predocked 5-HT was tested in setups 5, 16, and
17, and the effect of the Glide XP scoring function49 in the
redocking of ligands was explored in one setup for each
homology model (6, 11, and 18). Since protein flexibility is
included in IFD, up to 10 poses were written from each setup,
entailing a total of 125 poses of 5-HT binding in hSERT out
the theoretical maximum of 180 poses. Computed docking
scores (IFDScore and GlideScore) for all poses are tabulated
in the Supporting Information.
Figure 1. Structure of the hSERT homology model showing helices TM1,
TM3, TM6, and TM8 around the binding site together with the two sodium
ions from the LeuTAa structure and 5-HT. The binding site is closed toward
the extracellular lumen by an aromatic lid (residues Tyr176 and Phe335)
and a salt bridge (Arg104 and Glu493). Helices farther away from the
binding site are shaded.
of the very high similarity between the models neither the
MODELLER Objective Function40 (not shown) nor the DOPE
scores42 can be employed to directly distinguish their accuracy.
However, it is interesting to note that the five homology models
generated from alignment B generally exhibit the smallest
binding cavities, the poorest DOPE scores, and the lowest
number of residues in the allowed parts of the Ramachandran
plots. Furthermore, only two of the five homology models from
alignment B show the expected coordination of Asp98 to Na1,
compared to four out of five models from alignment A and C.
Ramachandran plots (data in Table 1) reveal that an overall
favorable stereochemistry is found in all models implying that
the alignments are in accordance with the secondary structure
elements. The residues occupying disallowed areas of the
conformational space are either glycine residues or they are
located in the extra- and intracellular loops, consistent with their
observed flexible nature. One residue does stand out here,
namely Asp98 with (æ,ψ) around (-150°, -150°) in all models.
However, Asp98 is located in the unwound part of TM1 in the
binding site and is thus not obliged to a particular conformation
to maintain any secondary structure.
4.1.2. Initial Docking and Selection of Homology Models
for IFD. The ligand binding site in the hSERT models is
generally too small for direct docking of 5-HT in the 15
homology models using Glide.46 This was also observed by
others25,26 who had to manually place the ligand in the binding
cavity by superimposition and subsequent minimization, some-
times imposing rather severe constraints. In contrast, we
succeeded in using molecular docking for placing 5-HT in the
binding cavity in two of the 15 homology models of hSERT,
A-4 and C-5, by using Glide 4.046 with default settings. These
are the two models with the largest cavities. However, rather
poor GlideScores were computed. Because of the possibility of
directly docking 5-HT into models A-4 and C-5, these two
models were selected for further IFD studies. Of the five
homology models from alignment B only two (B-1 and B-3)
allow Asp98 to interact with Na1 as suggested by Yamashita
The results reveal that the presence of none, one, or two
sodium ions did neither significantly influence the statistics of
the docking nor the structural features of the generated poses;
by comparison of the statistics from setup 1-4, 7-10, and 12-
15, respectively, for alignments A, B, and C, Table 2, it is clear
that the major differences in results from the setups with varied
sodium ion inclusion are found in the computed Prime energies,
which reveal the energy of the protein. The results (see
Supporting Information) indicate that the Na2 seems to be more
important than Na1 in stabilizing the protein and supports the
suggestion from Yamashita et al. that one of the sodium ions
may be structural, while the other is transported.12
It was possible to group 107 of the poses into three binding
clusters, each defined by the position and orientation of the 5-HT
indole ring (Figure 2). The statistics of the division of poses
into clusters as well as the total number of poses from each
setup are included in Table 2. Only one of the poses from model
B-3 has 5-HT located in the binding site; the other 13 poses all
have 5-HT positioned in the extracellular vestibule of the
transporter. They lie outside the aromatic lid, consisting of
Tyr176 and Phe335, in the space normally occupied by a salt
bridge, Arg104‚‚‚Glu493 (corresponding to Arg30‚‚‚Asp404 in
LeuTAa), which is therefore not present in these poses. The poses
with 5-HT located in the extracellular vestibule have the ligand
in random orientations, and at a first glance they do not seem
9
3858 J. AM. CHEM. SOC. VOL. 130, NO. 12, 2008

Binding of Serotonin to hSERT
A R T I C L E S
Figure 2. Orientation of 5-HT in the hSERT binding site. (A) Schematic representation of the binding site with possible hydrogen bonds shown in blue and
the aliphatic “bottom” of the binding site in green. Coordination of 5-HT in the three clusters identified from IFD are shown (B) cluster 1, setup 5, (C)
cluster 2, setup 5, and (D) cluster 3, setup 18. Carbon atoms of 5-HT are shown in pink while those in the protein are colored cyan. Putative hydrogen bonds
to 5-HT are included in the three IFD clusters.
to be relevant compared to the recently published structures of
LeuTAa with TCAs bound in the extracellular vestibule.13,14 It
should be noted here that the identified imipramine binding site
in LeuTAa has been suggested to be a secondary site, of no or
little importance related to the functioning of the NSSs.51
Imipramine has been shown to be a competitive inhibitor of
transport of 5-HT in hSERT, and it seems unlikely that the
primary binding site of the tricyclic amines should be so distant
from Ile172, which has been shown to be important in inhibitor
binding.20,21 Also, kinetic evidence for hSERT has suggested
that a secondary low-affinity binding site for imipramine exists
that is different from the inhibitory site.52,53 In line with these
arguments from the literature it is interesting to observe that
the computed docking scores of poses in the extracellular
vestibule after IFD were much less favorable than those placing
5-HT in the putative central binding pocket (see numbers in
the Supporting Information). In this study we focus on the
primary binding site; thus, the poses with 5-HT in the vestibule
were excluded from further analysis. All 14 poses from model
B-3 were consequently excluded, since the single pose in the
binding site has very poor docking scores and represents a
structural outlier compared to the three identified clusters. An
additional four poses from models A-4 and C-5 were similarly
found to represent structural outliers and were not included in
the analysis. All poses from 5-HT docking in A-4 and C-5 with
IFD were placed in the putative biding site and none in the
vestibule. Furthermore, it should be noted that the two poses
generated in the initial rigid protein docking belongs to the
clusters identified in IFD. IFD studies with setup 4, 6, 15, and
18 were subsequently repeated with inclusion of a chloride ion
in the recently proposed anion binding site.24,30 Results from
these simulations are equivalent to those described in Table 2,
and the chloride ion, thus, does not seem to be affecting the
binding of 5-HT. This is probably due to the distance (around
9 Å) between the 5-HT amino group and the chloride ion and
the presence of Na1 between the two groups. Results for these
simulations are tabulated in the Supporting Information.
4.1.4. Analysis of 5-HT Binding Clusters. Figure 2A shows
a schematic representation of the hSERT binding site. Parallel
to the membrane plane, Asp98 anchors all 5-HT poses in one
end while the other end (the 5-HT C6-position) is located near
Ala173 in a pocket that is generally hydrophobic, but includes
(51) Rudnick, G. ACS Chem. Biol. 2007, 2, 606-609.
(52) Wennogle, L. P.; Meyerson, L. R. Eur. J. Pharmacol. 1982, 86, 303-307.
(53) Plenge, P.; Mellerup, E. T. Eur. J. Pharmacol. 1985, 119, 1-8.
9
J. AM. CHEM. SOC. VOL. 130, NO. 12, 2008 3859

A R T I C L E S
Celik et al.
Table 3. Mean Ki (µM) and 95% Confidence Intervals (in brackets) for Inhibition of [3H]-5-HT Uptake from at Least Three Independent
Experiments for 18 Tryptamine Analogues against wt hSERT and 13 Single Point Mutants
substrate
analogue
wt hSERT
Ala173Cys
Ala173Ser
Ala173Thr
Ala173Asp
Ala173Met
Ala173Leu
1A
1B
1C
2A
2C
keto-2C
3A
3B
3C
123 [59-250]
22 [13.7-37]
112 [14.5-870]
13.2 [7.2-24]
9.0 [3.5-23]
1.05 [0.55-1.98]
1.13 [0.05-24]
5.5 [0.82-37]
0.81 [0.37-1.76]
137 [12.3-1500]
9.8 [1.28-75]
147 [5.6-3800]
15.7 [4.7-52]
9.0 [2.6-31]
2.0 [1.51-2.8]
1.25 [0.028-56]
9.9 [3.4-29]
4.2 [1.95-9.2]
0.74 [0.0182-30]
1.94 [0.136-28]
0.55 [0.25-1.23]
5.2 [3.2-8.3]
18.9 [10.4-34]
12.9 [8.1-21]
21 [18.2-24]
60 [4.8-740]
43 [3.3-570]
7.1 [2.5-20]
3.2 [1.92-5.2]
1.94 [0.95-3.9]
0.63 [0.025-15.7] 0.82 [0.33-2.0]
9.3 [3.02-28.4]
1.74 [1.11-2.7]
18.1 [9.6-34]
3.6 [1.52-8.3]
2.1 [1.02-4.4]
4.6 [0.32-67]
7.5 [2.8-20]
15.7 [10.0-25]
2.4 [1.05-5.3]
2.8 [1.19-6.5]
11.5 [7.0-18.6]
0.95 [0.5-1.81]
0.36 [0.07-2.0]
7.5 [2.1-27]
4.7 [3.3-6.7]
1.26 [0.40-4.0]
0.95 [0.031-29]
4.9 [0.168-140]
1.10 [0.166-7.4]
0.95 [0.46-1.95]
0.43 [0.035-5.3]
3.2 [0.31-33]
0.66 [0.049-8.8]
2.7 [1.93-3.7]
0.91 [0.74-1.11]
NDa
0.26 [0.0185-3.6] 0.27 [0.024-3.0]
0.101 [0.0134-0.7 6] 0.31 [0.032-3.0]
0.74 [0.54-1.02] 0.66 [0.35-1.25]
0.75 [0.21-2.7]
0.47 [0.22-1.03]
1.30 [0.38-4.5]
26 [16.0-42.5]
11.3 [2.6-50]
21 [7.7-57]
2.1 [1.78-2.4]
2.8 [1.75-4.4]
0.40 [0.163-0.96] 0.78 [0.60-1.02]
0.31 [0.135-0.72]
0.56 [0.40-0.80]
2.5 [2.2-2.9]
0.84 [0.31-2.3]
0.86 [0.61-1.22]
7.5 [3.8-14.6]
185 [99-350]
4.3 [4.0-4.7]
1.87 [1.28-2.8]
0.7 [0.35-1.4]
16.5 [6.6-41]
280 [86-890]
2.6 [2.3-3.0]
4.6 [2.3-9.2]
9.8 [3.3-29]
1.86 [0.8-4.4]
0.43 [0.26-0.71]
5-F-R-MeT 0.42 [0.23-0.77] 0.69 [0.46-1.03]
5-HT
0.92 [0.51-1.65] 1.54 [0.65-3.6]
5,7-diHT
5-MeT
5-MeOT
6-HT
6-MeOT
6-FT
42 [18.3-96]
18.4 [10.7-31]
56 [37-83]
42 [27-64]
2.6 [2.1-3.3]
0.21 [0.11-0.4]
46 [16.0-133]
10.5 [4.7-24]
22 [9.4-52]
17.0 [9.1-32]
4.4 [3.9-5.0]
0.42 [0.32-0.55]
81 [27-240]
11.2 [4.8-26]
13.1 [4.8-36]
16.3 [9.2-29]
7.9 [5.3-11.8]
1.08 [0.57-2.0]
4.5 [3.5-5.7]
2.2 [0.82-5.7]
9.8 [5.4-16.1]
9.1 [4.8-17.3]
1.87 [1.30-2.7]
0.67 [0.33-1.36]
7-MeT
0.37 [0.26-0.55] 0.40 [0.169-0.94] 0.38 [0.25-0.57]
0.37 [0.091-1.47] 0.30 [0.21-0.44]
0.34 [0.093-1.28] 0.32 [0.07-1.37]
substrate
analogue
Asp98Glu
Tyr175Phe
Tyr176Phe
Thr439Ala
Thr439Ser
Thr439Val
Ala169Ile
1A
1B
1C
2A
2C
keto-2C
3A
3B
3C
105 [35-320]
146 [16.7-1270]
13.4 [6.9-26]
74 [3.7-1500]
22 [11.7-40]
32 [8.2-127]
4.8 [2.5-8.9]
17.7 [13.9-22]
5.4 [2.1-13.8]
5.3 [4.1-6.9]
51 [15.4-167]
9.4 [5.0-17.8]
3.2 [1.87-5.4]
0.83 [0.60-1.13]
1.81 [0.61-5.4]
5.6 [5.0-6.2]
1.84 [1.00-3.4]
ND
55 [7.2-430]
8.7 [2.8-27]
6.2 [2.4-16.3]
2.0 [0.31-13.1]
1.74 [0.40-7.5]
3.4 [0.88-12.9]
105 [30-370]
26 [6.9-94]
9.1 [7.0-11.8]
7.6 [1.41-41]
2.1 [1.27-3.5]
28 [14.1-44]
13.3 [2.9-61]
1.19 [0.63-2.3]
0.37 [0.14-0.98]
1.42 [0.60-3.4]
2.8 [1.25-6.1]
0.65 [0.22-1.92]
0.31 [0.068-1.43] 0.37 [0.038-3.5]
0.44 [0.128-1.51] 0.98 [0.38-2.6]
5.6 [1.03-31]
0.84 [0.138-5.1]
1.47 [0.153-14.1] 3.7 [0.38-36]
7.1 [1.81-28]
0.53 [0.118-2.4]
1.41 [0.182-10.9]
2.2 [0.80-5.9]
4.2 [2.2-8.0]
22 [13.3-37]
4.1 [1.18-14.3]
0.68 [0.32-1.46]
0.39 [0.075-2.04] 2.6 [1.41-4.9]
ND ND
0.26 [0.0181-3.8] 4.8 [1.22-18.5]
0.088 [0.051-0.15 1] 0.10 [0.0081-1.1 3] 0.183 [0.035-0.97] 0.92 [0.69-1.22]
0.30 [0.026-3.4] ND
0.80 [0.153-4.2] 0.078 [0.0130-0.4 6] 0.40 [0.042-3.9]
5-F-R-MeT 0.107 [0.025-0.46] 0.21 [0.10-0.44]
5-HT
0.79 [0.48-1.3]
1.35 [0.78-2.3]
37 [15.9-88]
40 [25-64]
47 [35-62]
52 [22-124]
3.9 [1.80-8.6]
0.64 [0.25-1.65]
4.0 [0.119-133]
3.3 [2.5-4.4]
ND
0.43 [0.10-1.93]
9.7 [4.6-20]
10.1 [1.40-73]
26 [6.4-103]
17.8 [8.3-38]
0.76 [0.50-1.17]
3.7 [1.08-12.6]
179 [65-500]
0.62 [0.3-1.28]
22 [1.12-420]
21 [4.2-97]
1.15 [0.24-5.5]
68 [8.3-560]
6.9 [1.79-27]
28 [7.0-111]
24 [6.0-98]
1.63 [0.88-3.0]
0.158 [0.036-0.71] 0.30 [0.17-0.53]
3.8 [1.98-7.2]
520 [120-2300]
6.4 [2.9-14.5]
6.6 [2.2-19.7]
57 [31-105]
5,7-diHT
5-MeT
5-MeOT
6-HT
6-MeOT
6-FT
1.36 [0.191-9.7]
2.55 [0.57-11.5]
12.0 [7.1-20]
25 [12.4-49]
ND
12.3 [0.93-162]
1.23 [0.21-7.3]
0.079 [0.022-0.29]
1.36 [1.18-1.58]
0.40 [0.029-57]
1.82 [1.2-2.8]
0.126 [0.022-0.73] 0.141 [0.043-0.47] 0.28 [0.173-0.46] 0.099 [0.028-0.35]
7-MeT
ND
0.122 [0.048-0.31] 0.34 [0.161-0.71] 0.152 [0.058-0.40]
0.077 [0.0085-0.7 0] 0.175 [0.021-1.47] 1.80 [0.79-4.1]
a ND: Not determined.
some hydrophilic character. Residues Tyr176 and Phe335,
constituting the aromatic lid, are located on the extracellular
side of the bound ligand. The 5-HT‚‚‚Asp98 interaction is either
purely ionic or has an additional hydrogen bond character,
depending on the distances (tabulated in the Supporting
Information). The difference between the clusters is an ap-
proximately 180° rotation of the indole ring in the binding site
(Figure 2B-D). In cluster 1 the indole N-H points toward
Tyr176, while it faces Phe341 in clusters 2 and 3; in cluster 2
it points toward the edge of the Phe341 aromatic ring while it
is orthogonal to and points directly toward the center of the
aromatic ring in cluster 3. The only difference between clusters
2 and 3 is, thus, a parallel displacement of the indole ring
positioning it approximately 2-3 Å deeper into the binding
pocket in cluster 2. The retained interaction with Asp98 can be
rationalized by different conformations of the 5-HT ethylamine
side chain; in cluster 2 it is elongated while it is less extended
in cluster 3 (Figure 2C and 2D). Because of the high structural
similarity between the binding of 5-HT in clusters 2 and 3, they
are referred to as one single binding mode versus another
binding mode represented by cluster 1, leading to two possible
orientations of 5-HT in the hSERT binding site differing in the
indole ring orientation. Additional interactions to 5-HT in the
binding site includes a large number of hydrophobic interactions
as well as possible hydrogen bonds to the 5-hydroxyl group:
In cluster 1 a hydrogen bond to the Ala169 carbonyl oxygen
can be found while there is a possible hydrogen bond to Thr439
in clusters 2 (hydroxyl group) and 3 (carbonyl) and another
possible interaction to Ser438 (carbonyl) in cluster 3, as
indicated in Figure 2.
4.2. Biochemistry. 4.2.1. PaMLAC Study of 5-HT in
hSERT. In order to be able to confirm the predicted binding
of 5-HT in hSERT and to discriminate between the two
orientations found with IFD (cluster 1 and clusters 2/3,
respectively), we undertook a SAR study measuring the binding
affinities of a battery of substituted tryptamines, Chart 1A,
toward wt hSERT and 13 single point mutants of the protein.
The location and nature of the mutations were selected in a
way that would allow us to achieve experimental data that would
either challenge or support the two orientations of 5-HT as
predicted from the IFD simulations. All data are listed in Table
3. Below we present the studied protein-ligand interactions one
by one and analyze the effects on binding affinities of essential
single point mutations according to the PaMLAC paradigm,
starting with Asp98 and Ala173 to examine the identical features
of the two orientations, establishing the position of the ligand
axis in the binding site. The rotation around the axis, as revealed
in the three clusters, is then evaluated by mutations of Ala169
and Thr439.
9
3860 J. AM. CHEM. SOC. VOL. 130, NO. 12, 2008

Binding of Serotonin to hSERT
A R T I C L E S
4.2.2. Interpretation of Experimental Data. Assumptions.
Conclusions about 5-HT binding in hSERT by the use of
tryptamine analogues and transporter mutants in biochemical
assays rest upon the assumptions that (a) tryptamines bind to
the same site as 5-HT, (b) the mutations do not alter the overall
binding orientation but solely affect the binding affinity, (c)
mutant hSERTs are not subject to changes in the conformational
equilibrium of the transporter that affects binding of 5-HT and
tryptamine analogues differently, and (d) observed Ki values
primarily reflect initial binding affinity and not KM.
The assumption that the tryptamine analogues bind to the
same site as 5-HT appears reasonable since they are competitive
inhibitors of 5-HT transport.54 Additionally, the observation that
our models are able to predict an interaction between both ends
of the ligand with both ends of the putative binding site, which
can be supported experimentally (Vide infra) under the PaMLAC
paradigm, justifies that the tryptamines bind to hSERT in a very
similar fashion.
If mutations in hSERT or substitutions on the ligand were
severe it is possible that the ligand might assume another
orientation in the binding site. Such a situation could occur if,
e.g., an acidic residue (apart from Asp98) is mutated into the
binding cavity introducing a competition for interaction with
the positively charged nitrogen atom of the tryptamines. Results
from IFD suggest that changes in the ligand orientation would
initially manifest itself as a rotation of the indole ring, as these
are the only poses observed. However, if an aspartate or
glutamate residue is introduced in the hydrophobic bottom of
the cavity by mutation of either Ala169, Ile172, Ala173, or
Gly442, one can speculate that such a dramatic mutation could
cause a longitudinal swap of the ligand. We do not detect any
signs of such changes of ligand orientation in our experimental
data and therefore believe that variations in the substitution
pattern are neither sufficient to flip the indole ring nor to revert
the ligand axis.
4.2.3. Experimental Test of Salt Bridge between Asp98
and 5-HT. The acidic side chain of Asp98 is predicted to form
a salt bridge with the positively charged amine of 5-HT.12 Asp98
is indeed very sensitive to mutagenesis, underscoring the
importance of this residue for transporter function. Asp98Gly,
Asp98Ala, and Asp98Asn mutants were all found to be inactive
with regards to uptake of 5-HT and binding of â-CIT and [3H]-
S-citalopram (data not shown; for structures see Chart 1B). Only
the Asp98Glu mutant was active and is included in this study
to elaborate on the elegant experiments with N-methylated
tryptamines by Barker et al.,19 addressing the existence of this
putative salt bridge. We have extended their study by including
tryptamine analogues that are not N-methylated and thus mimic
the cognate substrate, 5-HT, more closely. Shortening the
alkylamine chain, analogues 1A (123 µM) and 1C (15.7 µM)
relative to 2A (2.4 µM) and 2C (2.8 µM), substantially reduces
affinity for the wt transporter (Table 3). As stated in the
PaMLAC approach, a measured loss of affinity when perturbing
the substrate analogue must be followed by regaining some of
the affinity by a single point mutant compensating for the
chemical changes in the ligand. Indeed, the loss of affinity for
the short tryptamines can be partially or fully rescued by
simultaneously extending the side chain of Asp98 by mutating
it to a glutamate, as is expected if the Asp98‚‚‚ligand salt bridge
is present. Furthermore, all tryptamines show improved binding
affinities toward the Asp98Glu mutant, but 1A and 1C gain
most. This can be interpreted as the alkylamine group in 1A
and 1C that assumes an extended conformation to gain an
interaction with Glu98, whereas the longer alkylamine chains
most likely are found in a more folded conformation for the
other analogues.
4.2.4. Exploration of Environment around the C6-Position
of 5-HT. From the IFD simulations it was predicted that the
5-HT C6-position may provide a basis for an interaction with
Ala173. We set up a series of PaMLAC experiments to test
this hypothesis, and if true, to characterize the electronic nature
of the protein near C6. Wild type hSERT allows some diversity
in substitutions on C6 of the tryptamines; 6-MeOT (2.6 µM)
and 6-FT (0.21 µM) are both well tolerated and have binding
properties similar to 5-HT (0.92 µM) and tryptamine (2.4 µM),
whereas 6-HT (42.0 µM) shows a drastically decreased affinity.
A small hydrophobic pocket in the protein near C6 of the ligand
thus seems likely. In accordance with the PaMLAC paradigm,
we want to test this hypothesis by reverting the nature of the
apparent hydrophobic interaction between C6 of the tryptamine
analogues and the Ala173 side chain to create a novel interac-
tion, thereby firmly establishing the existence of a direct ligand-
protein contact point. The fact that a methoxy substituent at C6
can be accommodated without loss of affinity toward wt hSERT
compared to tryptamine indicates that sufficient space exists
around the C6 position of the ligands to allow for a mutation
of the hydrophobic side chain of Ala173 to serine or threonine,
both hydrophilic. This is expected to be harmful for the affinity
of 6-MeOT. This pattern can be observed for Ala173Thr (9.8
µM) but (surprisingly) not for Ala173Ser (2.8 µM), see Table
3. This observation is difficult to directly explain; however, it
may be related to the observation that in the IFD models the
Ala173 side chain in wt hSERT is not pointing directly into
the binding cavity (Figure 2). It rather occupies a more tangential
orientation at the protein-ligand interface, and maybe the effect
of the methyl group in Ala173Thr becomes important for
orienting the hydroxyl group properly for interaction with the
ligand. It can further be predicted that Ala173Ser and Ala173Thr
may present a novel hydrogen-bonding partner to appropriate
substituents of the C6-atom of tryptamine, if the hydroxyl group
can indeed reach into the binding cavity. Hence, the poor affinity
of 6-HT for wt hSERT is expected to improve for Ala173Ser
and Ala173Thr if the orientation predicted from the IFD
simulations is correct. This pattern is, indeed, observed with a
20-fold increase in affinity of 6-HT for Ala173Ser (2.1 µM)
and Ala173Thr (2.2 µM) relative to wt hSERT (42 µM). This
complementing mutation effectively restores the affinity of 6-HT
to the same level as that of tryptamine for wt hSERT (2.4 µM),
demonstrating a full reversal from the native hydrophobic
interaction of the C6 hydrogen in tryptamine with Ala173 to a
hydrophilic interaction of 6-HT with single point mutated
proteins Ala173Ser or Ala173Thr. Having now established the
two longitudinal anchor points of the ligand, we moved to
experiments designed to probe the rotation of the ligand around
this axis. This can be accomplished by identification of
interaction partners with either the N-H bond of the indole
ring or with substituents at the C5- or C7-positions of the
tryptamine analogues.
(54) Adkins, E. M.; Barker, E. L.; Blakely, R. D. Mol. Pharmacol. 2001, 59,
514-523.
9
J. AM. CHEM. SOC. VOL. 130, NO. 12, 2008 3861

A R T I C L E S
Celik et al.
diHT is significantly reduced in affinity by the Ala169Ile
mutation relative to 5-HT, from 42 µM toward the wt to 520
µM against Ala169Ile. This could be interpreted as a repulsion
between Ala169Ile and the hydroxyl substituent on C7 of the
ligand and thus suggests that the Ala169 side chain and C7 of
the ligand are close in space. Curiously, hydrophobic substituents
on the C5 position (5-MeT, 5-MeOT) seem to have a positive
effect on affinity when paired with Ala169Ile that could be
interpreted as proximity of C5 of the ligand and Ala169. As
noted above, the absence of a clear pattern in measured binding
affinities of the Ala169Ile mutant may be due to secondary
effects, which is consistent with structures predicted from IFD
(Figure 2); the side chain of Ala169 is pointing away from the
binding cavity, thus making a direct interaction less likely.
Figure 3. Two views, rotated by approximately 90°, of the two binding
pockets around C5 (mauve) and C7 (ice-blue) of 5-HT displayed in cluster
2 (model A-4). Van der Waals surfaces of residues around the indole ring
are shown.
4.2.7. Ala173 and the 5-HT C5-Position. From the IFD
clusters it can be speculated that mutants with long hydrophobic
side chains at position 173 can potentially affect the affinity of
substituted tryptamines in a manner that would allow us to
deduce the correct location of the 5-HT C5 and C7 positions in
the binding site. Accordingly, both leucine and methionine
residues at position 173 have a negative impact on the affinity
of tryptamine analogues with a hydrophilic substituent on C5
(5-HT) whereas the affinity of 7-MeT is unaffected. The affinity
of 5,7-diHT is reduced 4-7 fold by the Ala173Met and
Ala173Leu mutations which is actually less than the 7-18 fold
reductions in affinity seen for 5-HT, indicating that the long
hydrophobic side chains at position 173 do not interact
negatively with a hydrophilic substituent on C7 but only with
hydrophilic substituents on C5. In accordance, long hydrophobic
residues on position 173 increase the affinity of tryptamine
analogues with hydrophobic substituents on C5 (5-MeT and
5-MeOT) whereas it leaves 7-MeT and 7-MeOT affinities
unaffected. All of these observations demonstrate that Ala173Met
and Ala173Leu can transform the hydrophilic pocket near C5
to a hydrophobic one but will leave the hydrophobic pocket
around C7 unaffected.
4.2.5. Environment around the C5- and C7-Positions of
5-HT. The SAR data listed in Table 3 reveals some flexibility
with respect to substituents at C5 of the tryptamine analogues
toward wt hSERT; hydrogen, fluoro, or hydroxyl groups in this
position result in very similar binding affinities with mean Ki
values between 0.42 µM and 2.4 µM. On the other hand,
hydrophobic substituents, such as methyl and methoxy groups,
result in poor binding and elevated mean Ki values of 18.4-56
µM. This indicates that a small group is important, and some
hydrophilic character is preferred (OH and F), however, not
strictly required (H). For the C7-position of the tryptamine
skeleton, a methyl substituent is favorable; 7-MeT binds equally
well as 5-HT toward wt hSERT and 6-fold better than
tryptamine (Table 3). The additional observation that placing a
hydroxyl group at C5 and another at C7 at the same time, 5,7-
diHT (42 µM), results in a poor binding compared to 5-HT (0.92
µM) strongly indicates that the binding pocket of the protein
has a preference for a hydrophobic substituent at the C7-position
of tryptamine.
The pockets surrounding the ligand apparently have different
electrostatic characteristics in the directions defined from the
C5 and C7 positions of 5-HT; the milieu of the pocket near C7
seems to be mostly hydrophobic, whereas the milieu around
C5 substituents seemingly is more hydrophilic (Figure 3). Note
that the two binding modes from the IFD simulations exactly
differ in their opposite orientations of C5 and C7 of 5-HT in
the binding site. Thus, mutations of residues lining the ligand
binding cavity along C5, C6, and C7 of the indole ring (Ala169,
Ala173, and Thr439) were constructed to further discriminate
the two binding modes employing the PaMLAC paradigm.
4.2.6. Ala169 and Substituted Tryptamines. Ala169 is
predicted to be located near 5-HT C5 in IFD cluster 1 and near
C6 and C7 in IFD cluster 2/3 but with the side chain pointing
away from the substrate and probably not interacting directly
with 5-HT. When Ala169 is mutated to the bulkier isoleucine,
the affinities of the three C6-substituted tryptamines are virtually
unaffected compared to wt hSERT. The affinity of Ala169Ile
for tryptamine (7.6 µM) decreases 3-fold compared to wt hSERT
(2.4 µM), indicative of an overall negative impact of this
mutation on tryptamine binding. The same relative decrease is
observed for 5-HT and 7-MeT in the Ala169Ile mutant
compared to the wt, indicating that neither the apparent
hydrophilicity of the pocket near C5 nor the hydrophobicity of
the pocket near C7 is affected by this mutation. However, 5,7-
4.2.8. Thr439 and the 5-HT C5-Position. Mutagenesis of
Thr439 to serine, alanine, and valine reveals a pattern in the
affinity for C5-substituted tryptamines consistent with the side
chain of Thr439 lining the hydrophilic pocket near C5. While
tryptamine affinity is unchanged by Thr439Ala or Thr439Val
mutations and slightly improved by the conservative mutation
to serine, the affinity of tryptamines with hydrophobic substit-
uents at the C5-position (5-MeT and 5-MeOT) benefit 14- to
22-fold by a Thr439Ala mutation. The Thr439Val mutants also
exhibit improvements in affinity for tryptamines with hydro-
phobic substituents on C5 but to a lesser extent than seen for
Thr439Ala. Furthermore, there is a trend toward reduced affinity
of tryptamines with hydrophilic C5-substituents (5-HT and 5,7-
diHT) for Thr439Val. This trend is even more pronounced for
Thr439Ala. There is thus a clear indication that Thr439
mutations can influence the hydrophobicity of the pocket near
C5 while leaving the pockets near C6 and C7 almost unaffected.
4.2.9. Mutation of Tyr175 and Tyr176. Mutation of the
conserved Tyr17525 to phenylalanine produces a mutant that
has a dramatically decreased Vmax (data not shown). An overall
moderate increase in affinity for the tryptamines against
Tyr175Phe is observed but otherwise no remarkable patterns
can be detected. The completely conserved Tyr176 can be
mutated to phenylalanine without significant changes in Km or
9
3862 J. AM. CHEM. SOC. VOL. 130, NO. 12, 2008

Binding of Serotonin to hSERT
A R T I C L E S
Vmax even though it forms a hydrogen bond to Asp98 in all
theoretical models. No changes or patterns in the affinity of
tryptamine analogues are seen for Tyr176 mutations. This may
indicate that Tyr175 and Tyr176 are mostly important in
substrate permeation and not particularly in the binding.
5. Discussion
5.1. Model of hSERT Structure and 5-HT Binding. It can
be assumed that the substrate binding site of hSERT has the
same approximate location as that of leucine in LeuTAa since it
would be highly improbable that a homologous transporter with
similar ionic coupling should utilize a different binding site for
the substrate.9,12 A substrate binding site deep within the
transporter, halfway through the membrane, is also consistent
with predictions from studies employing a fluorescent NET
substrate.55
The three employed sequence alignments of hSERT with
LeuTAa are highly similar, showing sequence identities of
approximately 28% in the TM region. For membrane proteins
this has been shown to produce homology models that will have
RMSD values in the TM region of around 2 Å for CR atoms
compared to the native structure31 when using a high-resolution
template structure. Most importantly, the alignment of the helices
making up the binding cavity, TM1, TM3, TM6, and TM8,
surrounding the binding site are exactly identical in the three
alignments, while differences are located in helices TM4, TM5,
TM9, and TM12. Despite the high degree of similarity, though,
alignment B resulted in homology models of hSERT, which
are not suitable for placing 5-HT in the binding cavity, even
when including protein flexibility during docking as in the IFD
protocol. This points to the effects distant structural elements
may have on the size and shape of the binding site and
emphasizes the importance of a correct sequence alignment for
the protein as a whole. Furthermore, it shows that choosing
appropriate homology models with a large cavity is indeed
important.
The unwound regions of TM1 and TM6 in the LeuTAa
structure are convincingly reproduced in the homology models
(Figure 1). Figure 4 displays the two coordination environments
for the sodium ions imported from the LeuTAa structure (Figures
4A and 4B). Although the coordination around the ions is not
perfect in the models, they are similar to the two sites in
LeuTAa.9,12 As predicted by Yamashita et al., Asp98 in hSERT
substitutes for the coordination of Na1 to the carboxylate group
from the substrate leucine in the LeuTAa structure.12 Residues
Tyr121, Ser336, Asn368, and Ser372 are furthermore positioned
so they can coordinate a chloride ion, forming a Cl- binding
site (Figure 4C) as has recently been shown.24,30 In Figure 4D
the substrate leucine from the template is depicted together with
hSERT Asp98 and bound 5-HT. It is obvious that the oxygen
atoms from the Asp98 carboxylate replaces those from the
leucine carboxylate group. The same is seen for the 5-HT and
leucine ammonium nitrogen atoms. In the three clusters identi-
fied from IFD the charged nitrogen of 5-HT is placed within 1
Å of the nitrogen atom from the superimposed leucine substrate
from LeuTAa. This can be interpreted as an example of the
deletion model as proposed for the evolution of selective
receptors.17
Figure 4. Coordination of (A) Na1, (B) Na2, and (C) Cl- in hSERT models.
(D) Asp98 and 5-HT bound in hSERT have similar functional groups as
leucine in LeuTAa. (E) Differences in the hSERT‚‚‚5-HT (cyan) and LeuTAa
(pink) binding sites. LeuTAa residue numbers are given in italics. Based on
cluster 3.
Only six out of 25 residues within 5 Å of 5-HT in hSERT
differ from those in LeuTAa (Figure 4E). A major difference is
the mutation of LeuTAa Gly24 to hSERT Asp98. This mutation
is generally found between amino acid and monoamine trans-
porters. Two other mutations are LeuTAa Asn21 and Ser256 to
hSERT Tyr95 and Gly338, respectively. These two residues
form the protein surface that interacts with the aliphatic side
chain of 5-HT. An extra aromatic residue was thereby introduced
in hSERT and to allow for possible π-π interactions between
Tyr95 and 5-HT. It could be interesting to mutate this residue
in hSERT and examine the need for an aromatic residue at the
position of Tyr95 as well as the possibility of a larger residue
instead of Gly338. The mutations of LeuTAa Val104 and Thr254
to hSERT Ile172 and Ser336 are probably by themselves the
least significant among the six differences; the methyl groups
introduced or removed in hSERT do not face the binding site
but the surrounding helices, thereby neither changing the size
nor the electrostatics properties of the binding site. More
important is the mutation of LeuTAa Ile359 to hSERT Gly442.
This residue is residing in the bottom of the cavity making up
(55) Schwartz, J. W.; Blakely, R. D.; DeFelice, L. J. J. Biol. Chem. 2003, 278,
9768-9777.
9
J. AM. CHEM. SOC. VOL. 130, NO. 12, 2008 3863

A R T I C L E S
Celik et al.
the hydrophobic floor together with Ala169, Ile172, and Ala173
in the hSERT models. When superimposing the LeuTAa and
hSERT structures, it is seen, Figure 4E, that the position of the
LeuTAa Ile359 side chain coincides with that of the six-
membered ring of the 5-HT indole ring. Altogether it can be
suggested that the difference in substrate selectivity between
LeuTAa and hSERT may be found in the LeuTAa to hSERT
Ile359 to Gly442, Asn21 to Tyr95, and Gly24 to Asp98
mutations. A consequence of the mutations of Val104 and Ile359
in LeuTAa to Ile172 and Gly442 in hSERT is that the
hydrophobic pocket is opened and the two residues residing
behind the hydrophobic contact of Val104 and Ile359 in LeuTAa
are exposed to the ligand in the hSERT binding cavity. These
residues are Ala173 and Ala169. On the basis of the results for
mutation of a creatine transporter into a GABA selective
transporter,18 one can envision that a few core mutations of
hSERT may give a transporter that is, for example, selective
for leucine.
The IFD simulations resulted in the identification of two
possible binding orientations of 5-HT in the binding cavity. Both
orientations show a salt bridge from Asp98 to the protonated
amine of 5-HT, and the ligand was oriented to involve a contact
between C6 of 5-HT and Ala173. The differences between the
two binding modes are related to the orientation of the indole
ring in the cavity; that is the interactions to C5, C7, and the
indole N-H. Neither of the preliminary homology modeling
studies24-26,28,30 were able to place the ligand in the hSERT
binding site by automated docking techniques nor could they
provide any details regarding the orientation of 5-HT in the
cavity. Thus, the binding mode identified in an MD simulation
of 5-HT in hSERT,56 where 5-HT was manually placed in a
large cavity originally created by S-citalopram and introducing
further restrained minimizations, differs from both of the binding
modes detected from our more objective docking study. To the
best of our knowledge, no other study in literature has considered
the orientation of 5-HT in the cavity. The role of having a
chloride ion present in the recently proposed Cl--binding site
was evaluated by a few additional IFD computations (listed in
the Supporting Information). Briefly, it can be concluded that
all homology models accommodate a chloride ion nicely.
Furthermore, neither the statistics of the poses generated in IFD
nor the computed docking scores change significantly depending
on the presence of the anion. Thus, it seems likely that the role
of the anion is related to transport and not directly involved in
substrate binding.
binding affinity for tryptamines in hSERT, unless a hydroxy or
methoxy substituent is present at the C5-position. Binding of a
tryptamine is thus a prerequisite for transport but not sufficient
to trigger it. Accordingly, 5-HT, 5,7-diHT, 6-HT, and 5-MeOT
were transported with efficiencies mirroring their relative
affinities whereas tryptamine, 5-MeT, 2C, 6-MeOT, and 6-FT
were transported poorer (or not at all) than expected from their
inhibitory potencies. The observation that 6-HT can be trans-
ported indicates some promiscuity in hSERT and could be a
result of structural flexibility of either substrate or protein,
however, not extending this to a methoxy group at the
C6-position, 6-MeOT. A study detailing the kinetic properties
of a fluorescent substrate for hNET, showing that binding and
unbinding of the substrate will occur 36 000 times for each
transport event,58 supports the notion that binding and triggering
of translocation are two separate events. The rate-limiting
translocation is not initiated by binding per se but can be
speculated to be so by the presence of a hydrophilic (oxygen)
substituent on C5 of the tryptamine. Since both IFD cluster 2/3
and the mutational studies suggest a possible interaction between
5-HT and Thr439 located in TM8 next to Ser438, which
interacts with Na2, one can speculate that simultaneous presence
of substrate and Na2 might induce the formation of the observed
kinks in TM8. It is known that TM8 undergoes conformational
changes as part of the transport process,59 and it can be
hypothesized that this serves as the communication link between
favorable binding events in the binding site and the formation
of an intracellular permeation pore.60,61
5.3. Salt Bridge between Asp98 and 5-HT. Within the
monoamine transporter subfamily of NSSs, Asp98 is fully
conserved. Other members of the NSS family, the amino acid
transporters, predominantly have a glycine at this position. The
inference that the acidic side chain forms a salt bridge with the
protonated primary amine of the cognate substrate of SERT,
DAT, and NET was originally reported by Kitayama et al.62
but later questioned by Wang et al.63 In hSERT the findings of
Barker et al.19 and the present study show that the loss of affinity
when shortening the alkylamine chain of tryptamine can be
partially rescued by lengthening the acidic side chain at position
98 of hSERT. The inability to fully rescue the affinity of 1A in
Asp98Glu can be rationalized by the fact that 1A has a reduced
flexibility due to the short alkyl chain, when adapting to the
interaction with an acidic residue at position 98.
5.4. Enantioface Discrimination of the Indole Moiety. The
5-HT indole ring constitutes a prochiral fragment and the two
binding modes identified from IFD can be regarded as the
recognition of the two enantiofaces. The main difference is the
interchanging positions of C5 and C7 as well as the direction
of the indole N-H bond. The enantioface discrimination of the
two 5-HT binding modes might thus be due to interactions from
the indole N-H to the binding site. It has been suggested that
5.2. Binding vs Transport of Tryptamines. Tryptamines
can either be substrates,57 i.e., they are transported, or inhibitors
of hSERT. Thus, the measured Ki may reflect affinities toward
other conformational states of the transport cycle. Apart from
the natural substrate 5-HT, the most potent substrate for wt
hSERT in this study is 5,7-diHT. It is transported with only
12% of the efficiency of 5-HT57 but also exhibits a 46-fold lower
Ki than 5-HT (see Table 3), easily accounting for the reduced
transport. Comparison of the binding affinities (by Ki values)
with the results for transport measured by Pratuangdejkul et
al.57 reveals a poor correlation between transport efficiency and
(58) Schwartz, J. W.; Novarino, G.; Piston, D. W.; DeFelice, L. J. J. Biol. Chem.
2005, 280, 19177-19184.
(59) Norregaard, L.; Loland, C. J.; Gether, U. J. Biol. Chem. 2003, 278, 30587-
30596.
(60) Loland, C. J.; Norregaard, L.; Litman, T.; Gether, U. Proc. Natl. Acad.
Sci. U.S.A. 2002, 99, 1683-1688.
(61) Guptaroy, B.; Zhang, M.; Binda, F.; Bowton, E.; Johnson, L.; Galli, A.;
Javitch, J.; Gnegy, M. Society for Neuroscience Annual Meeting, Atlanta,
GA, U.S.A., October 14-18, 2006 and personal communication.
(62) Kitayama, S.; Shimada, S.; Xu, H.; Markham, L.; Donovan, D. M.; Uhl,
G. R. Proc. Natl. Acad. Sci. U.S.A. 1992, 89, 7782-7785.
(63) Wang, W.; Sonders, M. S.; Ukairo, O. T.; Scott, H.; Kloetzel, M. K.; Surratt,
C. K. Mol. Pharmacol. 2003, 64, 430-439.
(56) Jørgensen, A. M.; Tagmose, L.; Jørgensen, A. M. M.; Bøgesø, K. P.; Peters,
G. H. ChemMedChem 2007, 2, 827-840.
(57) Pratuangdejkul, J.; Schneidier, B.; Jaudon, P.; Rosilio, V.; Baudoin, E.;
Loric, S.; Conti, M.; Launay, J. M.; Manivet, P. Curr. Med. Chem. 2005,
12, 2393-2410.
9
3864 J. AM. CHEM. SOC. VOL. 130, NO. 12, 2008

Binding of Serotonin to hSERT
A R T I C L E S
Phe341 in hSERT is involved in π-π stacking with 5-HT.12
Indeed, according to the models, Phe341 receives π-stabilization
by interacting with the indole N-H in an orthogonal “T-shape”-
manner in clusters 2 and 3 from IFD. This has been shown to
be as favorable as the parallel-displaced π-π stacking confor-
mation for benzene rings.64,65 In cluster 1 an orthogonal
T-shaped interaction of indole N-H is found with Tyr176
located at the opposite face of the binding cavity. Similar
hydrogen bonds to π-systems have previously been shown to
be important in proteins for stabilizing the local 3D structure.66
Mutations of these two aromatic residues to non-aromatic ones
may accordingly be able to shed further light on this interaction,
assuming that the transporter remains active.
In IFD clusters 2 and 3, which were shown above to be
consistent with the data from the PaMLAC studies, the
hydrophilic pocket around C5 is predicted to be formed at the
interface of TM3 and TM8, and more precisely by the side
chains of residues Ala173 and Thr439. As shown above, results
from mutations of these two residues and binding of tryptamine
analogues substituted at C5, C6, or C7 are consistent with the
predictions formed on the basis of IFD clusters 2 and 3 while
inconsistent with predictions from IFD cluster 1.
5-HT C5-position to a hydrophobic area around the 5-HT C7-
position (Figure 3), (d) long hydrophobic side chains on hSERT
residue 173 can extend into the pocket near the 5-HT C5-
position, and (e) hSERT Thr439 interacts with the hydroxyl
group of 5-HT. These observations are only consistent with a
5-HT binding mode in hSERT as found in clusters 2 and 3 in
the IFD calculations. We expect that this model will be of great
importance for understanding the function of hSERT and for
future drug design projects. Current research in our group is
aimed at describing the movements of the cytoplasmic parts of
TM1, TM6, and TM8 relative to one another, which would be
useful in taking the understanding of hSERT function one step
further. Studies to further elucidate the monoamine transporters
as drug targets are similarly underway.
Acknowledgment. We thank Bente Ladegaard for skilful
technical assistance and the iNANO Center, the Danish Center
for Scientific Computing, the Danish Natural Science Research
Council, the Danish National Research Foundation, and the
Carlsberg and Lundbeck Foundations for financial support.
Note Added after ASAP Publication: The version of this
paper published on March 4, 2008, contained errors involving
refs 14 and 24. The version published on March 6, 2008 has
the correct information.
6. Conclusion
In conclusion we have provided evidence from computations,
SAR, and mutagenesis studies under the PaMLAC paradigm
that (a) hSERT Asp98 interacts with the protonated alkylamine
of 5-HT, (b) hSERT Ala173 mutants interact with tryptamines
substituted at the C6 position (Figure 2D), (c) the electrostatic
properties of hSERT changes from a hydrophilic vicinity of the
Supporting Information Available: Details of the experi-
mental methods; figures of computed DOPE per residue scores
and the alignments against LeuTAa; a table with GlideScore,
IFDScore, and hydrogen bond distances for Asp98‚‚‚5-HT;
results from simulations with Cl- included in the models. This
material is available free of charge via the Internet at http://
pubs.acs.org. A pdb file with 5-HT bound in hSERT can be
obtained from the authors.
(64) Sinnokrot, M. O.; Valeev, E. F.; Sherrill, C. D. J. Am. Chem. Soc. 2002,
124, 10887-10893.
(65) Sinnokrot, M. O.; Sherrill, C. D. J. Phys. Chem. A 2004, 108, 10200-
10207.
(66) Steiner, T.; Koellner, G. J. Mol. Biol. 2001, 305, 535-557.
JA076403H
9
J. AM. CHEM. SOC. VOL. 130, NO. 12, 2008 3865