A virtual screen for diverse ligands: Discovery of selective G protein-coupled receptor antagonists

Article abstract of DOI:10.1021/ja077620l

Virtual screening has become a major focus of bioactive small molecule lead identification, and reports of agonists and antagonists discovered via virtual methods are becoming more frequent. G protein-coupled receptors (GPCRs) are the one class of protein targets for which success with this approach has been limited. This is likely due to the paucity of detailed experimental information describing GPCR structure and the intrinsic function-associated structural flexibility of GPCRs which present major challenges in the application of receptor-based virtual screening. Here we describe an in silico methodology that diminishes the effects of structural uncertainty, allowing for more inclusive representation of a potential docking interaction with exogenous ligands. Using this approach, we screened one million compounds from a virtual database, and a diverse subgroup of 100 compounds was selected, leading to experimental identification of five structurally diverse antagonists of the thyrotropin-releasing hormone receptors (TRH-R1 and TRH-R2). The chirality of the most potent chemotype was demonstrated to be important in its binding affinity to TRH receptors; the most potent stereoisomer was noted to have a 13-fold selectivity for TRH-R1 over TRH-R2. A comprehensive mutational analysis of key amino acid residues that form the putative binding pocket of TRH receptors further verified the binding modality of these small molecule antagonists. The described virtual screening approach may prove applicable in the search for novel small molecule agonists and antagonists of other GPCRs.

Full text of DOI:10.1021/ja077620l

Published on Web 03/22/2008
A Virtual Screen for Diverse Ligands: Discovery of Selective
G Protein-Coupled Receptor Antagonists
Stanislav Engel,† Amanda P. Skoumbourdis,‡ John Childress,† Susanne Neumann,†
Jeffrey R. Deschamps,§ Craig J. Thomas,‡ Anny-Odile Colson,† Stefano Costanzi,†
and Marvin C. Gershengorn*,†
The Clinical Endocrinology Branch and Laboratory of Biological Modeling, National Institute of
Diabetes and DigestiVe and Kidney Diseases, National Institutes of Health, 50 South DriVe,
Bethesda, Maryland 20892, NIH Chemical Genomics Center, National Human Genome Research
Institute, National Institutes of Health, 9800 Medical Center DriVe, RockVille, Maryland 20850,
and Laboratory for the Structure of Matter, NaVal Research Laboratory,
Washington, D.C. 20375
Received October 3, 2007; E-mail: marving@intra.niddk.nih.gov
Abstract: Virtual screening has become a major focus of bioactive small molecule lead identification, and
reports of agonists and antagonists discovered via virtual methods are becoming more frequent. G protein-
coupled receptors (GPCRs) are the one class of protein targets for which success with this approach has
been limited. This is likely due to the paucity of detailed experimental information describing GPCR structure
and the intrinsic function-associated structural flexibility of GPCRs which present major challenges in the
application of receptor-based virtual screening. Here we describe an in silico methodology that diminishes
the effects of structural uncertainty, allowing for more inclusive representation of a potential docking
interaction with exogenous ligands. Using this approach, we screened one million compounds from a virtual
database, and a diverse subgroup of 100 compounds was selected, leading to experimental identification
of five structurally diverse antagonists of the thyrotropin-releasing hormone receptors (TRH-R1 and TRH-
R2). The chirality of the most potent chemotype was demonstrated to be important in its binding affinity to
TRH receptors; the most potent stereoisomer was noted to have a 13-fold selectivity for TRH-R1 over
TRH-R2. A comprehensive mutational analysis of key amino acid residues that form the putative binding
pocket of TRH receptors further verified the binding modality of these small molecule antagonists. The
described virtual screening approach may prove applicable in the search for novel small molecule agonists
and antagonists of other GPCRs.
activity.4 However, screening efforts on a large scale are often
prohibitively expensive and technically inaccessible to all but
Introduction
A recent survey of known drug targets confirmed G protein-
coupled receptors (GPCRs or seven transmembrane-spanning
receptors) as the most widely targeted class of cellular proteins
by pharmacological agents.1 The cell surface accessibility of
the GPCR extracellular domain and seven transmembrane
helices, as well as the control of wide-ranging physiological
processes that are engendered by GPCR activation or inhibition,
provides strong rationale for the “drugability” of this cellular
target. While the completion of the human genome and other
milestones in cellular and molecular biology has advanced our
understanding of cellular targets, so too have advances in the
drug discovery process. Traditional efforts rooted in the screen-
ing of small molecules against specific cellular targets have been
joined by novel technologies, including structure-based methods2
and virtual or in silico screening approaches.3 High-throughput
screening is, perhaps, the most powerful method for identifying
small molecule ligands that possess intrinsic biochemical
the industrial sector and singular screening efforts within the
NIH Molecular Libraries Initiative.5 Structure or fragment-based
drug design is more accessible; however, targets such as GPCRs,
for which crystallographic or NMR-based structural information
is difficult to generate, are not readily studied by such methods.
Thus, it is difficult to discover small molecule ligands for these
compelling pharmacological targets.
The recent success of in silico methods in identifying
compound leads for targets such as kinases6 and transcription
factors7 has highlighted the mainstream use of virtual screening.
(2) Congreve, M.; Murray, C. W.; Blundell, T. L. Drug DiscoVery Today
2005, 10, 895–907.
(3) Muegge, I.; Oloff, S. Drug DiscoVery Today: Technol. 2006, 3, 405–
411.
(4) Inglese, J.; Johnson, R. L.; Simeonov, A.; Xia, M.; Zheng, W.; Austin,
C. P.; Auld, D. S. Nat. Chem. Biol. 2007, 3, 466–479.
(5) Austin, C. P.; Brady, L. S.; Insel, T. R.; Collins, F. S. Science 2004,
306, 1138.
† National Institute of Diabetes and Digestive and Kidney Diseases.
‡ National Human Genome Research Institute.
§ Naval Research Laboratory.
(6) Toledo-Sherman, L.; Deretey, E.; Slon-Usakiewicz, J. J.; Ng, W.; Dai,
J.-R.; Foster, J. E.; Redden, P. R.; Uger, M. D.; Liao, L. C.; Pasternak,
A.; Reid, N. J. Med. Chem. 2005, 48, 3221–3230.
(7) Derksen, S.; Rau, O.; Schneider, P.; Schubert-Zsilavecz, M.; Schneider,
G. ChemMedChem 2006, 1, 1346–1350.
(1) Imming, P.; Sinning, C.; Meyer, A. Nat. ReV. Drug DiscoVery 2007,
5, 821–834.
9
10.1021/ja077620l This article not subject to U.S. Copyright. Published 2008 by the American Chemical Society J. AM. CHEM. SOC. 2008, 130, 5115–5123 5115

A R T I C L E S
Engel et al.
Figure 1. Compound structures of thyrotropin-releasing hormone (1) and the compounds identified by the virtual screen.
Proteins with well-documented crystallographic or NMR-based
structures constitute the most successfully pursued targets of
virtual screening. In the absence of protein structural informa-
tion, an alternate virtual screening format utilizes detailed
knowledge of known ligands and is commonly referred to as
ligand-based virtual screening.3 There is little ability to establish
a virtual screen in cases where protein structure is unavailable
and ligand classes have yet to be discovered. Few reports of in
silico generated ligands of GPCRs have been reported,8–11 most
likely due to the paucity of information of the structure of the
receptors. For several years, the only experimentally delineated
GPCR structure was that of rhodopsin12 (a class A GPCR).
Recently, after this work was completed, crystal structures for
the human ꢀ2-adrenergic receptor (ꢀ2-AR) have also been
published.45–47 These findings disclosed a substantial structural
similarity between rhodopsin and the ꢀ2-AR, amply supporting
the widespread practice of modeling GPCRs by homology.
Although homology modeling has gained sophistication and
credibility in recent years and new techniques for GPCR
structural modeling continue to evolve,13 these advances have
yet to generate widespread success in GPCR virtual screens.
At least three factors contribute to the limited applicability of
in silico GPCR methods. The first factor is the inherent
conformational flexibility of GPCRs. The idea of induced fit
for both receptor and ligand may represent a general feature of
GPCR activation, and the ligand binding pocket of a GPCR
may have intrinsic plasticity, resulting in the ability to adopt a
number of alternative conformations. Consequently, there is a
high level of uncertainty surrounding an exact designation of
a receptor-based pharmacophore. Moreover, the use of static
representations of receptor/ligand complexes as templates for
formulation of a pharmacophore hypothesis may erroneously
restrict the available binding potential. Shoichet and co-workers
have introduced technical advances in “soft docking” to aid
virtual methods in overcoming these issues.14 The second factor
involves the availability of multiple sites of binding interaction
within the transmembrane binding cavity of GPCRs. An
approach that includes the entire binding potential of the putative
binding pocket may allow for the discovery of ligands that bind
to the receptor via interactions not exploited by the cognate
ligand. The third factor, which does not concern GPCRs
exclusively, is the inherent inaccuracy of the scoring functions.
While a number of docking procedures proved capable of
reproducing the geometry of receptor–ligand complexes, the
associated scoring functions are not yet a viable solution for
predicting affinity of the ligands.48,49 This is especially true when
the structures of the receptors are derived by homology
modeling.
Thyrotropin-releasing hormone (TRH), Figure 1, plays an
important role in the regulation of thyrotropin (thyroid-stimulat-
ing hormone, TSH) and prolactin synthesis and secretion in the
pituitary gland and as a neurotransmitter/neuromodulator within
the central and peripheral nervous systems.15 TRH action is
mediated by the two TRH receptor isotypes, TRH-R1 and TRH-
R2. On the basis of the marked differences in tissue distribution
and basal signaling of TRH-R1 and TRH-R2, it is likely that
the two receptors mediate differing effects of TRH.16 A potent,
efficacious small molecule ligand with selectivity between TRH-
R1 and TRH-R2 would be of use to delineate the unique physi-
ological roles of each receptor. In addition, nonpeptide TRH
receptor ligands, which are metabolically stable and able to cross
the blood-brain barrier, might be valuable for the treatment of
several neuropathological conditions including neurodegenera-
tion. Thus, novel TRH receptor agonists and/or antagonists
would be of general use to the biomedical community.
In this study, we applied a virtual screening approach to
identify novel small molecule ligands of TRH-R1 and TRH-
R2, with a particular interest in identifying isotype-selective
ligands.
(8) Marriott, D. P.; Dougall, I. G.; Meghani, P.; Liu, Y.-J.; Flower, D. R.
J. Med. Chem. 1999, 42, 3210–3216.
Structural alignment of the binding cavities of TRH-R1 and
TRH-R2 has revealed that the composition and arrangement of
residues within the transmembrane domains (TMDs) that are
important for TRH binding are nearly identical.17 This is
(9) Freddolino, P. L.; Kalani, M. Y. S.; Vaidehi, N.; Floriano, W. B.;
Hall, S. E.; Trabanino, R. J.; Kam, V. W. T.; Goddard, W A. Proc.
Natl. Acad. Sci. U.S.A. 2004, 101, 2736–2741.
(10) Evers, A.; Klebe, G. J. Med. Chem. 2004, 47, 5381–5392.
(11) Evers, A.; Hessler, G.; Matter, H.; Klabunde, T. J. Med. Chem. 2005,
48, 5448–5465.
(14) Ferrari, A. M.; Wei, B. Q.; Costantino, L.; Shoichet, B. K. J. Med.
Chem. 2004, 47, 5076–5084.
(12) Palczewski, K.; Kumasaka, T.; Hori, T.; Behnke, C. A.; Motoshima,
H.; Fox, B. A.; Le, T. I.; Teller, D. C.; Okada, T.; Stenkamp, R. E.;
Yamamoto, M.; Miyano, M. Science 2000, 289, 739–745.
(13) Zhang, Y.; DeVries, M. E.; Skolnick, J. PLoS Comp. Biol. 2006, 2,
88–99.
(15) Metcalf, G.; Jackson, I. M. D. Ann. N.Y. Acad. Sci. 1989, 533.
(16) Sun, Y.; Lu, X.; Gersengorn, M. C. J. Mol. Endocrinol. 2003, 30, 87.
(17) Engel, S.; Gershengorn, M. C. Pharmacol. Ther. 2007, 113, 410–
419.
9
5116 J. AM. CHEM. SOC. VOL. 130, NO. 15, 2008

A Virtual Screen for Diverse Ligands
A R T I C L E S
consistent with the observed binding affinity similarities for TRH
and its numerous analogues at these two receptors.16 Thus, a
virtual screen utilizing a ligand binding cavity with binding
potentials distinct from those utilized by TRH may be beneficial
for the discovery/design of novel, diverse and selective modula-
tors of TRH receptor activities.
Our screening was based on a homology model of TRH-R1
previously built based upon a 3D projection map of bovine
rhodopsin as structural template.18 Numerous studies exploring
both chemically modified TRH and mutationally altered recep-
tors have allowed detailed assessment and refinement of the
predicted binding cavity.19–26
Overall, the screening comprised four sequential stages: two
actual virtual screenings, each of which was followed by a
diverse subset selection (Supplementary Figure 1). Each stage
was designed to reduce the input virtual library of compounds
to 10% of its size. To mitigate the limits of currently used
procedures, our virtual screening scheme was designed to (a)
consider receptor flexibility; (b) take into account all the
available binding potential of the ligand binding cavity; and
(c) minimize the use of scoring functions by integrating them
with selections based on pharmacophore matching and diversity.
The virtual screening led to experimental identification of five
structurally diverse antagonists of TRH receptors. In particular,
a stereoisomer of N-(1-(2,3-dihydrobenzo[b][1,4]dioxin-2-yl-
)ethyl)-2,2-diphenylacetamide was identified as a novel, potent,
and selective small molecule antagonist of TRH-R1.
Figure 2. Interaction potential (receptor-based 3D pharmacophore) of the
ligand binding cavity in TRH-R1. Blue, red, and yellow spheres represent
a potential field of interaction with H-donor, H-acceptor, and hydrophobic
molecular probes, respectively. TRH (in orange) is superimposed with the
receptor-based pharmacophore, demonstrating that only a part of the
available interaction potential is utilized for TRH binding. Flexibility was
granted to the residues lining the binding pocket in the calculation of the
interaction potentials.
cavity, the clustering points of high potential interaction energy
for each molecular probe (less than -0.1 kcal/mol for the “dry”
probe, less than -7 kcal/mol for the NH2 probe, and less than -12
kcal/mol for the carbonyl oxygen probe) were represented as
spherical features of corresponding radius using Unity and combined
into a single query (Figure 2). This spherical representation of the
interaction potential mimics the entire conformational space of the
key residues, thus addressing the issue of protein flexibility in
the ligand docking process and partially compensating for model
imprecision. The partial-match constraints protocol (Unity) was used
to define four distinct groups, reflecting the spatial localization of
the spherical features.
One million commercially available compounds from the ZINC
database of diverse drug-like compounds29 were screened using
the Flex Search protocol (Unity) with the conformational space of
the ligands sampled in an attempt to fulfill all the partial-match
constraints. The partial constraints in the query were empirically
set to produce a hit rate of 10% (∼100 000 compounds). Only
ligands concurrently matching from 1 to 3 spherical features in
each partial query were considered as “hits”. A 10% diverse subset
(∼10 000 compounds) of the Unity pharmacophoric search hit-list
was subsequently evaluated by flexible docking to the putative TRH
binding pocket in TRH-R1 using the program FlexE (Tripos).30
The selection of this subset was carried out based on the position
of the compounds in 6D chemical space, defined by the six
optimized BCUT descriptors (elecneg_S_burden_001.000_K_H;
elecneg_S_invtopd_000.100_K_L;gastchrg_S_invtopd_000.010_K_H;
gastchrg_S_invtopd_010.000_K_L; tabpolar_S_burden_001.000_K_L;
tabpolar_S_invdist_000.250_K_H) using a methodology imple-
mented in the Diverse Solution software31 available within Sybyl
7.1.32 During docking with FlexE, flexibility of the ligands was
allowed and the binding pocket was represented by an assembly
of the five alternative conformations in order to mimic flexibility
of the protein target (see Generation of Alternative Conformations
of the Binding Cavity). The FlexX scoring function was used to
evaluate the compound-receptor interaction in the docking pro-
cedure. One thousand compounds with docking scores higher than
26 kJ/mol were selected (the interacting energy for TRH was 28
kJ/mol), and a 10% diverse subset (100 compounds) was generated
using MACCS structural keys fingerprints (Tanimoto coefficient
Methods
Modeling: Virtual Screening. A model of the TRH/TRH-R1
complex based on the topological homology to the 3D projection
map of rhodopsin was used as the structural template for this study.
This model was obtained through a series of optimizations using
various computer simulation techniques to rationalize the effects
of complementary modifications in the structure of TRH and
TRH-R1.18,20
Here, a receptor-based pharmacophore describing the interaction
potential of the TRH-R1 binding pocket was generated using the
combination of the programs Grid (Molecular Discovery)27 and
Unity (Sybyl 7.1).28 A box of 15 × 15 × 15 Å3 was defined around
the position of TRH in the complex. After TRH extraction, the
potential to form hydrophobic, proton donor or proton acceptor
interactions within the box was explored using “dry”, neutral flat
NH2 and sp2 carbonyl oxygen molecular probes, respectively (Grid).
The effective dielectric constant was set to 8.0. Flexibility of the
side chains was allowed, thereby extending the area of potential
interactions. A methyl (CH3) molecular probe was used to define
the van der Waals boundaries of the binding cavity. Within this
(18) Colson, A. O.; Perlman, J. H.; Smolyar, A.; Gershengorn, M. C.;
Osman, R. Biophys. J. 1998, 74, 1087–1100.
(19) Rutledge, L. D.; Perlman, J. H.; Gershengorn, M. C.; Marshall, G. R.;
Moeller, K. D. J. Med. Chem. 1996, 39, 1571–1575.
(20) Perlman, J. H.; Laakkonen, L. J.; Guarnieri, F.; Osman, R.; Gersh-
engorn, M. C. Biochemistry 1996, 35, 7643–7650.
(21) Laakkonen, L. J.; Guarnieri, F.; Perlman, J. H.; Gershengorn, M. C.;
Osman, R. Biochemistry 1996, 35, 7651–7663.
(22) Perlman, J. H.; Colson, A-O.; Wang, W.; Bence, K.; Osman, R.;
Gershengorn, M. C. J. Biol. Chem. 1997, 272, 11937–11942.
(23) Colson, A-O.; Perlman, J. H.; Jinsi-Parimoo, A.; Nussenzveig, D. R.;
Osman, R.; Gershengorn, M. C. Mol. Pharmacol. 1998, 54, 968–978.
(24) Jain, R.; Singh, J.; Perlman, J. H.; Gershengorn, M. C. Bioorg. Med.
Chem. 2002, 10, 189–194.
(25) Kaur, N.; Lu, X.; Gershengorn, M. C.; Jain, R. J. Med. Chem. 2005,
48, 6162–6165.
(29) Irwin, J. J.; Shoichet, B. K. J. Chem. Inf. Model. 2005, 45, 177–182.
(30) FlexE; Tripos Inc.: St. Louis, MO 63144, 2006.
(31) Pearlman, R. S.; Smith, K. M. Perspect. Drug DiscoVery Des. 1998,
9, 339–353.
(26) Kaur, N.; Monga, V.; Lu, X.; Gershengorn, M. C.; Jain, R. Bioorg.
Med. Chem. 2007, 15, 433–443.
(27) Grid 1.2.2; Molecular Discovery Ltd.: Oxford, U.K., 1998.
(28) Unity; Tripos Inc.: St. Louis, MO 63144, 2006.
(32) Sybil 7.1; Tripos Inc.: St. Louis, MO 63144, 2006.
9
J. AM. CHEM. SOC. VOL. 130, NO. 15, 2008 5117

A R T I C L E S
Engel et al.
similarity metric) implemented in the Diverse Subset feature of
MOE33 and submitted to experimental testing.
sites NecoI and HindIII. Mutations were introduced using the
QuikChange II XL site-directed mutagenesis kit (Stratagene). All
constructs were verified by sequencing (MWG Biotech).
Cell Culture. HEK293 (human embryonic kidney) cells were
grown in DMEM containing 10% fetal bovine serum, 100 units/
mL penicillin, and 10 µg/mL streptomycin (Life Technologies Inc.)
at 37 °C in a humidified 5% CO2 incubator.
Ligand Binding Assays. The cells were transfected with 0.8
µg/mL of plasmid DNA encoding for TRH-R1, TRH-R2, or mutant
receptors for 48 h using FuGENE6 reagent (Roche, Basel,
Switzerland). Competition binding assays were performed in a
monolayer of intact cells at 4 °C for 4 h with 1–5 nM [3H]MeTRH
and various concentrations of test compounds as described.40 The
cells were preincubated with compound for 15 min before addition
of radioligand. Equilibrium dissociation constants (Kd) for
[3H]MeTRH were determined in saturation binding experiments
as described41 or in competition experiments using unlabeled
MeTRH. Equilibrium binding constants for tested compounds were
derived using the formula: Ki ) (IC50)/(1 + ([L]/Kd)), where IC50
is the concentration of unlabeled ligand that half-competes with
specifically bound [3H]MeTRH, and Kd is the equilibrium dissocia-
tion constant for [3H]MeTRH.
Generation of Alternative Conformations of the Binding
Cavity. A conformational search of the residues lining the receptor
cavity as defined by a methyl (CH3) molecular probe (Grid) was
conducted with the Monte Carlo Multiple Minima (MCMM)
protocol (MacroModel 9.1, Schrödinger).34 The search consisted
of 1000 steps of the Monte Carlo torsional sampling intervened by
500 iterations of PRCG minimization (0.05 kJ/mol) using OPLS2005
force field in vacuum. All of the structures with an energy content
higher than that of the most stable structure by more than 100 kJ/
mol were rejected, as they could potentially represent geometrically
distorted conformations. Of the remaining structures, the most stable
one and the four most diverse, in terms of atomic rmsd, were
selected for docking studies. The five selected structures showed
rmsd values of about 0.4 Å from each other. This is indicative of
significant diversity, considering that only the residues lining the
binding pocket were involved in the conformational search (the
rest of the protein was frozen), and the rmsd values were calculated
with respect to all the atoms of the receptor. The selection of five
diverse structures was judged sufficient since the FlexE docking
program combines them to generate a higher number of possible
alternative structures.
Luciferase Assay. HEK293 cells were transfected with 0.8 µg/
mL of plasmid DNA encoding for TRH-R1 or TRH-R2 and 0.8
µg/mL of pAP(Activator Protein)-1Luc vector (PathDetect In Vivo
Signal Transduction Pathway trans- and cis-Reporting System;
Stratagene, La Jolla, CA). After 6 h of transfection, the cells were
stimulated with or without 3 nM TRH in the presence or absence
of 30 µM of 2 and incubated for an additional 18 h. The
luminescence was measured as previously described.42
Neighbor Search. A search of the original library for related
structures was performed using MACCS structural keys fingerprints
and Tanimoto coefficient similarity metric implemented in the
similarity feature of MOE.33
Scoring of the Individual Stereoisomers of Compound 2. The
LiaisonScore method (Liaison 4.0, Schrödinger)35 was used to
predict free binding energy of the individual stereoisomers of 2
(2a-d) within the model of TRH-R1. This method uses the same
conformational sampling approach as the Linear Response Method
(LRM)36,37 but is based on an empirical binding score (Gli-
deScore)38 with a discrete representation of the protein. The main
advantage of the LiaisonScore over the typical docking scoring
function is that it treats both ligand and protein as flexible entities,
leading to a greater precision in the evaluation of protein–ligand
interactions. Molecular dynamics was used as a sampling method
with a temperature of 300 K. The truncated Newton algorithm was
used for energy minimization. The calculations were performed with
the OPLS2005 force field in a surface-generalized Born (SGB)
continuum model for solvation.39 Due to the absence of a set of
known active compounds (training set), the calculated free binding
energies have relative but not absolute meaning.
Measurement of Intracellular Calcium Mobilization Using
Fluorometric Imaging Plate Reader (FLIPRTETRA). HEK293
cells stably expressing TRH-R1, TRH-R2, or GPR40 were seeded
in black-walled, clear-bottomed 96-well plates (Corning, NY) at a
density of 6 × 104 cells/well in DMEM with 10% fetal bovine
serum and incubated for 24 h at 37 °C and 6% CO2. The following
day, the culture media was replaced with 100 µL of Hank’s
Balanced Salt Solution with 20 mM HEPES, and the cells were
loaded with 100 µL of calcium 3 fluorescent dye (Molecular
Devices, Sunnyvale, CA) for 1 h at room temperature before
addition of compounds. Transient changes in [Ca2+] induced by
ligands were measured using the FLIPRTETRA system (Molecular
Devices, Sunnyvale, CA). Changes in fluorescence were detected
at the emission wavelength of 515–575 nm. Agonistic and
antagonistic properties of ligands were measured in a concentration
range from 10 nM to 50 µM. The agonistic response was determined
immediately upon compound addition, followed by 15 min con-
tinued incubation before addition of EC50 concentration of TRH
(1 nM, TRH receptors) or linoleic acid (10 µM, GPR40) to measure
antagonistic effects. Responses were measured as peak fluorescent
intensity minus basal fluorescent intensity at each compound
concentration.
Experimental Section
Materials. Dulbecco’s modified Eagle’s Medium (DMEM) and
fetal bovine serum were purchased from Biosource (Rockville, MD).
TRH (pyroGlu-His-ProNH2), linoleic acid (sodium salt), and
luciferin were purchased from Sigma (St. Louis, MO). [3H]MeTRH
was purchased from NEN Life Science Products (Boston, MA).
Compounds for screening were obtained from ChemBridge (San
Diego, CA) and Asinex (Winston-Salem, NC).
Construction of Vectors and Site-Directed Mutagenesis of
TRH Receptor. cDNAs for mouse TRH-R1 and TRH-R2 were
inserted into the pcDNA3.1(-)/hygromycin vector using restriction
Data Analysis. All data were analyzed by linear or nonlinear
regression using the Prism software version 3 (GraphPad, Inc., San
Diego, CA).
Results and Discussion
To achieve a robust and reliable virtual screen for ligands of
a GPCR, there must exist a detailed model of the core putative
binding pocket. From our longstanding efforts to dissect the
binding of TRH (1) (Figure 1) to TRH-R1,18–26 an experimen-
tally supported homology model was available and was used
in the virtual screen. As discussed, a four-stage virtual screening
(33) Molecular Operating EnVironment; Chemical Computing Group, Inc.:
Montreal Quebec, Canada, 2005.
(34) MacroModel 9.1; Schrodinger, LLC.: New York, 2003.
(35) Liason 4.0; Schrodinger, LLC.: New York, 2005.
(36) Aqvist, J.; Medina, C.; Samuelsson, J. E. Protein Eng. 1994, 7, 385–
391.
(37) Jones-Hertzog, D. K.; Jorgensen, W. L. J. Med. Chem. 1997, 40, 1539–
1549.
(38) 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.
(39) Zhou, R.; Friesner, R. A.; Ghosh, A.; Rizzo, R. C.; Jorgensen, W. L.;
Levy, R. M. J. Phys. Chem. B 2001, 105, 10388–10397.
(40) Lu, X.; Huang, W.; Worthington, S.; Drabik, P.; Osman, R.; Gersh-
engorn, M. C. Mol. Pharmacol. 2004, 66, 1192–1200.
(41) Perlman, J. H.; Nussenzveig, D. R.; Osman, R.; Gershengorn, M. C.
J. Biol. Chem. 1992, 267, 24413–24417.
(42) Gershengorn, M. C.; Osman, R. Physcol. ReV. 1996, 76, 175–191.
9
5118 J. AM. CHEM. SOC. VOL. 130, NO. 15, 2008

A Virtual Screen for Diverse Ligands
A R T I C L E S
Figure 3. Pharmacological analysis of commercially available 2 that is an equal mixture 2a-d. (A) Radioligand competition binding was performed at 4
°C in cells expressing mouse TRH-R1 (9) or TRH-R2 ((). The cells were preincubated for 15 min with increasing concentrations of 2 followed by 4 h
incubation in the presence of 1 nM [3H]MeTRH. The binding was determined as described in Methods. (B) Inhibition of TRH-stimulated signaling of
TRH-R1 and TRH-R2 and of basal signaling of TRH-R2 by 2. Luciferase activity in HEK293 cells expressing TRH-R1 or TRH-R2 was measured after
stimulation with or without (basal) 3 nM TRH in the absence or presence of 30 µM of 2 as described in Methods. Rlu ) relative light units. (C) Demonstration
of 2 inhibiting agonist-stimulated formation of inositol phosphates. HEK293 cells expressing TRH-R1 were preincubated for 15 min with increasing
concentrations of 2 followed by 90 min incubation at 37 °C in the presence of EC50 concentration of TRH (2 nM) (9) or (R)-desaza-TRH (2.5 µM) (O).
(R)-Desaza-TRH is a superagonist at TRH receptors (see Engel, S.; Gershengorn, M. C. J. Biol. Chem. 2006, 281, 13103). Formation of inositol phosphates
was measured as described (see Engel, S.; Gershengorn, M. C. J. Biol. Chem. 2006, 281, 13103.). All of the curves represent the nonlinear regression
analyses of the data using one-site competition function. All data are presented as the mean ( sem from three independent experiments performed in
duplicates.
procedure (Supplementary Figure 1) was designed in order to
consider the flexibility of ligands and receptor, to use the entire
binding potential presented by the ligand binding cavity, and
to minimize the usage of scoring functions. Molecular probes
(Grid) were used to map the interaction fields associated with
all amino acids lining the pocket, while granting flexibility to
the side chains to account for induced fit and intrinsic uncertainty
in the homology model. On the basis of this potential, a receptor-
based 3D pharmacophore was generated (Figure 2). The
superimposition of this pharmacophore with TRH reveals that
the available interaction potential is much larger than that
utilized by TRH, the cognate ligand. In the first stage of the
screening, one million compounds from the ZINC database of
commercially available, diverse compounds were screened
virtually for flexible fitting to the receptor-based 3D pharma-
cophore, resulting in a hit-list of 100 000 compounds. In the
second stage, a diverse subset of 10 000 compounds from this
hit-list was selected to be processed in the third stage, which
consisted of molecular docking at a model of the TRH-R1
binding pocket represented by an assembly of alternative
conformations to mimic the structural flexibility within the
binding site. To minimize the impact of scoring functions, we
introduced a fourth and final stage, in which a diverse subgroup
of 100 compounds among the 1000 top ranking docked
compounds was selected for experimental testing.
These compounds and several related compounds (see
Neighbor Search in Methods) were purchased and analyzed for
their ability to stimulate TRH receptor-mediated increases in
intracellular Ca2+, that is, act as agonists, or to inhibit TRH-
stimulated increases in intracellular Ca2+, act as antagonists,
and inhibit binding of the radiolabled TRH analogue [3H](Nτ-
methyl)His-TRH using TRH-R1- and TRH-R2-expressing cells.
Among the compounds tested, only antagonists were observed.
We found five unique scaffolds including a novel substituted
N-((tetrahydrofuran-2-yl)methyl)-1-naphthamide 3 and a sub-
stituted 1-phenethylindoline 4 (Figure 1) (see Supplementary
Figure 2 for related IC50 values). As a preliminary gauge of
selectivity, we determined the level of activity of these
compounds against GPR40 (a GPCR that like TRH-R1 and
TRH-R2 couples to Gq43) and found no activity. The assortment
of chemotypes that were found to be novel antagonists of TRH-
R1 and TRH-R2 strongly suggests that using an expanded
binding potential and granting flexibility to the residues within
the binding cavity promotes the discovery of more diverse small
molecule classes. The most compelling lead structures were
baseduponthecorescaffoldofsubstituted2,3-dihydrobenzo[b][1,4]dioxin-
2-yl (represented by compounds 2), and several small molecules
with this core structure (11–14, as representative compounds)
were among the most highly ranked compounds from the virtual
screen.
The compound 2 (N-(1-(2,3-dihydrobenzo[b][1,4]dioxin-2-
yl)ethyl)-2,2-diphenylacetamide) contains two stereocenters at
carbons C1 and C2 (Figure 5).The pharmacological analysis
was performed utilizing a commercial preparation which
included all four stereoisomers (Figure 1). This mixture was
found to compete with [3H]MeTRH for binding to TRH-R1 and
TRH-R2 (Figure 3A) with affinities of 0.68 ( 0.20 and 7.3 (
1.4 µM, respectively (an 11-fold binding selectivity for TRH-
R1 over TRH-R2), to inhibit TRH-mediated upregulation of
expression of luciferase reporter gene and basal (ligand-
independent) signaling activity of TRH-R2 (Figure 3B). The
mixture of 2 was shown also to inhibit TRH and (R)-desaza-
TRH-stimulated formation of inositol phosphate second mes-
sengers (Figure 3C). These data confirmed that the screen had
indeed revealed a novel chemotype with potent and selective
TRH receptor antagonism.
The presence of two chiral centers was of interest. As a
preliminary study, we estimated the relative free energy of
binding of each stereoisomer using molecular docking and
LiaisonScore empirical scoring function. The computational
analysis predicted that the C2-S, C1-R isomer 2a would bind
with the highest affinity to TRH-R1 relative to the other isomers
(Figure 4A). We synthesized and tested all four individual
stereoisomers of 2. The synthesis (shown in Figure 4D)
originated with chirally pure 2,3-dihydrobenzo[b][1,4]dioxine-
2-carboxylic acids 7a and 7b (the R isomer is commercially
available and the S isomer was obtained via a classical chiral
resolution). With chiral starting materials, it was assured that
one stereocenter was set. From this starting point, a Weinreb
amide was formed (8a or 8b) and displaced with methyl
Grignard reagent to give acetates 9a and 9b. Reductive
amination catalyzed by Ti(OiPr)4 produced the primary amines
10a-d. This step provided equal mixtures of diastereomers that
(43) Tayasam, G. V.; Tulasi, V. K.; Davis, J. A.; Bansal, V. S. Exp. Opin.
Ther. Targets 2007, 11, 661–671.
9
J. AM. CHEM. SOC. VOL. 130, NO. 15, 2008 5119

A R T I C L E S
Engel et al.
Figure 4. (A) Equilibrium binding constants and calculated LiaisonScores of 2a-d complexes with TRH-R1. HEK293 cells were transfected for 48 h with
0.8 µg/mL of plasmid DNA encoding for TRH-R1, and equilibrium binding constants were determined as described in Methods. Data are presented as the
mean ( sem from three independent experiments performed in duplicate. LiaisonScores of the stereoisomers docked to TRH-R1 were calculated as described
in Methods. (B) Linear relationship between calculated free energies of interaction (LiaisonScore) and experimental binding affinities of 2a-d. (C) Radioligand
competition binding was performed at 4 °C in cells expressing mouse TRH-R1. The cells were preincubated for 15 min with increasing concentrations of
2a followed by 4 h incubation in the presence of 0.3, 1.0, 3.0 or 10 nM [3H]MeTRH. IC50 values of inhibition of radioligand binding by 2a are directly
proportional to the concentration of [3H]MeTRH, obeying the Cheng-Prusoff relationship: IC50 ) Ki × ([A*]/KdisA* + 1), the characteristic of competitive
ligand–receptor interactions. [A*] is concentration of radioligand and KdisA* is the equilibrium dissociation constant of the TRH-R1/[3H]MeTRH complex.
Data are presented as the mean ( sem from three independent experiments performed in duplicates. Boundaries of the 95% confidence interval for the linear
regression analysis are shown. (D) Synthetic scheme for the production of 2a-d. Reagents and conditions: (a) CDI (1.1 equiv), CH2Cl2, 1 h;
N,O-dimethylhydroxylamine hydrochloride; (b) MeMgCl (1.5 equiv), THF, 0 °C, 30 min; (c) Ti(OiPr)4 (2 equiv), NH3/EtOH (5 equiv), 6 h, then NaBH4 (1.5
equiv), 3 h; (d) diphenyl acetic acid (1.3 equiv), HATU (2 equiv), DMF, then DiPEA (3 equiv), 5 min, 6a-d (1 equiv) in DMF dropwise, 45 min. CDI )
carbonyl diimidazole; HATU ) O-(7-azabenzotriazole-1-yl)-N,N,N′,N′-tetramethyluronium hexafluorophosphate.
Figure 5. Models of TRH/TRH-R1 and 2a/TRH-R1 complexes. (A) The optimized model of the TRH/TRH-R1 complex used as a starting point to
design the virtual screening protocol. (B) The optimized model of the 2a/TRH-R1 complex as generated by the virtual screening (see Methods). Note
the difference in the position of Gln105 in TM-3 (in green) between the two complexes. Gln105 is not important for TRH binding but, because of
its high flexibility, may contribute to binding of 2a. In the virtual screening, the interaction field associated with Gln105 contributed to the selection
of 2a as a “hit” by superimposition with one of its dioxin oxygen atoms. In the ligands, the spheres in cyan, red, and blue correspond to carbon,
oxygen, and nitrogen atoms, respectively.
were noted by LCMS analysis. Aliquots of this material were
purified via LC methods, and the purified materials were
crystallized. A purified, as yet undetermined crystal formed from
the fast running eluent of the products derived from chirally
enriched 7a (S isomer) and crystallographic analysis revealed
the structure as primarily the R,R isomer 10d (the stereochemical
designation of the C2 position changes from S to R upon
reductive amination). From this structure, chiral HPLC analysis
and optical rotation measurements were able to define the
remaining stereoisomers. Coupling of 10a-d with diphenyl
acetic acid provided the final products 2a-d as diastereomeric
mixtures separable by LC methods. Aliquots of the purified,
stereochemically defined 10a-d were also submitted to this final
procedure to allow precise definition of each final product.
The pharmacological analysis of the individual stereoisomers
supported the accuracy of the model since ranking of the
predicted free energies of binding was consistent with the
relative affinities demonstrated by 2a-d (Figure 4A,B). Isomer
2a (the lead compound identified in the virtual screen) was found
to have a Ki of 0.29 µM and a nearly 13-fold selectivity for
TRH-R1 over TRH-R2. Compound 2a represents the most
potent inhibitor of TRH-R1 reported to date. The models of
TRH-R1 complexed with TRH (1) or 2a indicate overlap
between the binding sites of the two ligands (Figure 5A,B). To
evaluate this prediction, we determined the nature of 2a
antagonism. A linear relationship was observed between IC50
values for inhibition of [3H]MeTRH binding by 2a and
9
5120 J. AM. CHEM. SOC. VOL. 130, NO. 15, 2008

A Virtual Screen for Diverse Ligands
A R T I C L E S
Table 1. Equilibrium Binding Constants of [3H]MeTRH and 2a to TRH-R1, TRH-R2, and Mutant Receptorsa
a HEK293 cells were transfected for 48 h with 0.8 µg/mL of plasmid DNA encoding for TRH-R1, TRH-R2, or a mutant receptor and equilibrium
binding constants were determined as described in Methods. Data are presented as the mean ( sem from three independent experiments performed in
duplicates. NA ) not applicable.
concentration of the radioligand (Figure 4C), which is consistent
with competitive character of interaction of these ligands with
TRH-R1.
thereby optimizing existing interactions with key residues and/
or allowing new interactions to form. According to the model
of TRH-R1, Ile309 flanks Trp279 in TM-6, restricting its side
chain to pointing outside of the ligand binding pocket (Figure
5A,B). Thus preventing its interaction with the bound ligand.
Consistent with the model, substitution of Trp279 by smaller
Phe or Met in TRH-R1 has no effect on 2a binding (Table 1).
However, in the Ile309Ala mutant receptor, substitution of
Trp279 by Phe or Met results in 40- and 90-fold reduced
affinities for 2a, respectively (Table 1). This finding is consistent
with the idea that reduced bulkiness in the vicinity of Trp279
in the Ile309Ala mutant allows its access to the ligand binding
cavity and interaction with the bound 2a, thereby contributing
to the increased affinity of 2a in this mutant compared to TRH-
R1. In TRH-R2, by contrast, the nature of the residue at position
297 (corresponding to position 309 in TRH-R1) has no
significant effect on 2a binding (Table 1). These results support
the model of the 2a/TRH-R1 complex, but they do not support
a role of the Ile/Val variation in 2a isotype selectivity; they do
further indicate that different interactions are involved in the
formation of 2a complexes with TRH-R1 and TRH-R2. The
described relationships between the affinity of 2a and the nature
of the residue at position 309 in TRH-R1 may be valuable for
future design of new ligands with improved potency and/or
selectivity.
To test the veracity of our docking model further, binding
analyses of 2a to a number of TRH-R1 and TRH-R2 mutants
were performed (Table 1 and Supplementary Figure 3). Ac-
cording to the model, Arg306, a residue in TM-7 that is known
to interact with Pro-NH2 of TRH,20 appears to form a hydrogen
bond with the amide carbonyl of 2a (Figure 5A,B). Substitution
of Arg306 by Lys reduced affinity for 2a by 200-fold, indicating
its important role in 2a binding. In contrast, substitution of the
corresponding residue in TRH-R2 (Arg294Lys) shows no effect
on 2a binding, revealing an important difference in binding
modes of 2a to TRH-R1 and TRH-R2. According to the model
of the TRH-R1/2a complex, Asp195 in TM-5 acts in conjunction
with Arg306 to position 2a within the binding pocket by
formation of a H–bond with the amide hydrogen of 2a (Figure
5A,B). However, substitutions of Asp195 abolish high affinity
binding of MeTRH (not shown), thus preventing performance
of competition binding experiments to evaluate 2a affinity at
this mutant TRH receptor.
Ile309 in TM-7 of TRH-R1 (Val297 in TRH-R2) sits at the
bottom of the TRH binding cavity and represents the only
difference in composition of the ligand binding pockets in the
models of TRH-R1 and TRH-R2. Ile309 directly contacts one
of the phenyl rings of 2a (Figure 5B), raising the possibility
that the Val/Ile variation may contribute to TRH-R1 isotype
selectivity of 2a through formation of additional hydrophobic
contacts. Mutagenesis studies revealed that space-generating
substitutions at this position in TRH-R1 increase affinity for
2a. The potency of 2a was as much as 1100-fold higher for
alanine at position 309 as compared to leucine (Table 1). It is
possible that extra space in the proximity of a phenyl ring of
2a causes it to shift deeper into the transmembrane domain,
In the model of the TRH-R1/TRH complex, Gln105 (Gln102
in TRH-R2), located at the entrance of the ligand binding pocket,
is situated in an outward pointed direction, therefore having no
direct interaction with bound TRH (Figure 5A). In support of
the model, the substitution of Gln105 by Ala does not affect
TRH binding in TRH-R1 and TRH-R2 (Table 1). Conforma-
tional analysis of the residues in TRH-R1 reveals a high degree
of flexibility associated with Gln105, allowing it access to the
ligand binding pocket (Figure 5B). By approaching the ligand
9
J. AM. CHEM. SOC. VOL. 130, NO. 15, 2008 5121

A R T I C L E S
Engel et al.
Figure 6. Effect of prolonged exposure to 2 or midazolam on the surface expression and signaling of TRH-R1. HEK293 cells stably expressing a low level
of TRH-R1 (clone R1–17, ∼70 000 receptors/cell) (see Engel, S.; Gershengorn, M. C. J. Biol. Chem. 2006, 281, 13103) were incubated without (control)
or with 20 µM of 2 or midazolam in DMEM (2% of heat-deactivated FBS) for 16 h at 37 °C. (A) After washing with HBSS (10% Hepes, pH 7.4), the cells
were incubated for 1 min with ice-cold acid solution (0.2 M acetic acid, 0.5 M NaCl, pH 2.5) to remove the bound compounds, followed by wash (×3) with
HBSS. The cells were incubated in the presence of 10 nM [3H]MeTRH for 7 h at 4 °C, and specific binding was determined as described. (B) The cells were
washed five times (10 min of incubation each) with HBSS at room temperature, and TRH-stimulated release of the intracellular Ca2+ was measured in the
presence of 1 nM TRH (EC50 concentration) using FLIPR, as described in Methods. All data are presented as the mean ( sem from three independent
experiments performed in triplicate.
binding cavity, the Gln105 field of interaction with the H-donor/
acceptor probe (Grid, see Methods) contributed to the selection
of 2a in the virtual screening by superimposition with one of
its benzodioxin oxygens (not shown). A similar analysis in TRH-
R2 revealed that Gln102 is conformationally restricted in the
position outside of the binding pocket, making its interaction
with the bound 2a highly unlikely. In agreement with the model,
the substitution of Gln105 in TRH-R1 by alanine reduced the
binding affinity of 2a by more than 50-fold (Table 1), whereas
the corresponding substitution in TRH-R2 (Gln102Ala) showed
no effect on 2a binding. In fact, 2a exhibits reversed selectivity
for the Gln to Ala variants of TRH receptors, indicating that
interaction of Gln105 with the bound ligand may contribute to
2a TRH-R1 selectivity. Alternatively, the highly flexible Gln105
located at the entrance of the ligand binding cavity may facilitate
2a transition into the TMD of TRH-R1 by formation of transient
interactions with the ligand. Identification of Gln105 as a
potential interaction site for 2a exemplifies the need to use the
entire available binding potential in formulation of a receptor-
based pharmacophore, rather than the binding determinants of
the cognate ligand only. Overall, the mutagenesis studies support
the verity of the model of 2a/TRH-R1 complex. Furthermore,
they demonstrate that 2a and TRH bind to TRH receptors by
interacting with different transmembrane domain residues, and
that different interactions are involved in stabilization of 2a in
TRH-R1 and TRH-R2, despite the similarities of their sites for
interaction with TRH.
Finally, to begin to gauge the utility of this novel TRH
receptor antagonist, we studied its effects on TRH receptors
upon prolonged cellular exposure. Benzodiazepines (and their
analogues) are the only class of nonpeptide compounds known
to antagonize TRH signaling.40 However, there are several
disadvantages to their use as in vivo probes of inhibition of
TRH signaling. For example, prolonged exposure of TRH
receptor-expressing cells to benzodiazepines results in an
increased number of surface receptors and potentiation of the
TRH response (Figure 6 and ref 44), thereby compromising the
antagonistic effect of these compounds. In contrast to benzo-
diazepines, prolonged incubation of TRH-R-expressing cells
with 2 does not significantly affect surface expression of TRH
receptors or efficacy of TRH signaling (Figure 6). This feature
of 2 represents a significant advantage over benzodiazepines
for application in intact animals.
Conclusion
Virtual screening has been used in many efforts to identify
novel small molecule ligands for diverse cellular targets
including kinases, phosphatases, and transcription factors. Novel
compounds originating from in silico methods have entered
clinical evaluation in disease states such as cancer and diabetes.
However, virtual methods targeting GPCRs have trailed behind,
owing in large part to structural uncertainty of the receptors
and a lack of novel in silico methods. The virtual screening
methodology described in this study allows for a more inclusive
representation of the structural potential available for interaction
with artificial ligands, thereby increasing the likelihood of
identification of ligands with diverse chemical scaffolds and
potentially different binding modalities. By employing both an
expanded interaction potential that differs from that utilized by
the natural ligand and conformational flexibility within the
binding cavity, and by minimizing the usage of scoring
functions, chemically diverse antagonists of TRH receptors have
been identified. Of note, these ligands are capable of distin-
guishing between TRH receptor isotypes that share nearly
identical structural determinants for TRH binding and represent
the first known small molecule ligands with a significant degree
of TRH receptor isotype selectivity. Also, the chemical diversity
of the new ligands and corresponding variation in their binding
affinities can contribute to development of structure–activity
(45) Rasmussen, S. G.; Choi, H. J.; Rosenbaum, D. M.; Kobilka, T. S.;
Thian, F. S.; Edwards, P. C.; Burghammer, M.; Ratnala, V. R.;
Sanishvili, R.; Fischetti, R. F.; Schertler, G. F.; Weis, W. I.; Kobilka,
B. K. Nature 2007, 450, 383–387.
(46) Rosenbaum, D. M.; Cherezov, V.; Hanson, M. A.; Rasmussen, S. G.;
Thian, F. S.; Kobilka, T. S.; Choi, H. J.; Yao, X. J.; Weis, W. I.;
Stevens, R. C.; Kobilka, B. K. Science 2007, 318, 1266–1273.
(47) Cherezov, V.; Rosenbaum, D. M.; Hanson, M. A.; Rasmussen, S. G.;
Thian, F. S.; Kobilka, T. S.; Choi, H. J.; Kuhn, P.; Weis, W. I.;
Kobilka, B. K.; Stevens, R. C. Science 2007, 318, 1258–1265.
(48) Ferrara, P.; Gohlke, H.; Price, D. J.; Klebe, G.; Brooks, C. L., III
J. Med. Chem. 2004, 47, 3032–3047.
(49) 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.
(44) Heinflink, M.; Nussenzveig, D. R.; Grimberg, H.; Lupu-Meiri, M.;
Oron, Y.; Gershengorn, M. C. Mol. Endocrinol. 1995, 9, 1455–1460.
9
5122 J. AM. CHEM. SOC. VOL. 130, NO. 15, 2008

A Virtual Screen for Diverse Ligands
A R T I C L E S
relationships for application in drug design using ligand-based
approaches. Finally, the selectivity and cellular effects exhibited
by 2a suggest that it may be useful as a probe to distinguish
the physiological roles of TRH-R1 and TRH-R2 in animal
models.
tive of the National Institutes of Health Roadmap for Medical
Research, the Intramural Research Program of the National
Human Genome Research Institute, and the National Institute
of Drug Abuse (Contract Y1-DA6002-02), National Institutes
of Health.
Acknowledgment. We thank Dr. Bruce Raaka and Dr.
Oksana Gavrilova for advice and help in preparation of this
manuscript. We thank Ms. Allison Peck for critical reading of
this manuscript. We thank Mr. John Lloyd for HRMS data.
A.O.C. gratefully acknowledges Dr. David Stanton for his
helpful suggestions and discussions regarding the virtual screen-
ing methodology. This research was supported by the Intramural
Research Program of the National Institute of Diabetes and
Digestive and Kidney Diseases, the Molecular Libraries Initia-
Supporting Information Available: Elaboration of virtual
screening procedure, structures of alternate ligands discovered,
chemical synthetic methods and characterization of products and
intermediates, crystallographic analysis of 10d and determination
of absolute configuration of 2a-d. This material is available
free of charge via the Internet at http://pubs.acs.org.
JA077620L
9
J. AM. CHEM. SOC. VOL. 130, NO. 15, 2008 5123