Predicted ligands for the human urotensin-II G protein-coupled receptor with some experimental validation

Article abstract of DOI:10.1002/cmdc.201402087

Human Urotensin-II (U-II) is the most potent mammalian vasoconstrictor known.1 Thus, a U-II antagonist would be of therapeutic value in a number of cardiovascular disorders.2 Here, we describe our work on the prediction of the structure of the human U-II receptor (hUT₂R) using GEnSeMBLE (GPCR Ensemble of Structures in Membrane BiLayer Environment) complete sampling Monte Carlo method. With the validation of our predicted structures, we designed a series of new potential antagonists predicted to bind more strongly than known ligands. Next, we carried out R-group screening to suggest a new ligand predicted to bind with 7 kcal mol^-1 better energy than 1-{2-[4-(2-bromobenzyl)-4-hydroxypiperidin-1-yl]ethyl}-3-(thieno[3,2-b] pyridin-7-yl)urea, the designed antagonist predicted to have the highest affinity for the receptor. Some of these predictions were tested experimentally, validating the computational results. Using the pharmacophore generated from the predicted structure for hUT₂R bound to ACT-058362, we carried out virtual screening based on this binding site. The most potent hit compounds identified contained 2-(phenoxymethyl)-1,3,4-thiadiazole core, with the best derivative exhibiting an IC₅₀ value of 0.581 μM against hUT ₂R when tested in vitro. Our efforts identified a new scaffold as a potential new lead structure for the development of novel hUT₂R antagonists, and the computational methods used could find more general applicability to other GPCRs.

Full text of DOI:10.1002/cmdc.201402087

CHEMMEDCHEM
FULL PAPERS
DOI: 10.1002/cmdc.201402087
Predicted Ligands for the Human Urotensin-II G Protein-
Coupled Receptor with Some Experimental Validation
Soo-Kyung Kim,[a] William A. Goddard, III,*[a] Kyu Yang Yi,[b] Byung Ho Lee,[b] Chae Jo Lim,[b]
and Bartosz Trzaskowski[c]
Human Urotensin-II (U-II) is the most potent mammalian vaso-
constrictor known.[1] Thus, a U-II antagonist would be of thera-
peutic value in a number of cardiovascular disorders.[2] Here,
we describe our work on the prediction of the structure of the
human U-II receptor (hUT2R) using GEnSeMBLE (GPCR Ensem-
ble of Structures in Membrane BiLayer Environment) complete
sampling Monte Carlo method. With the validation of our pre-
dicted structures, we designed a series of new potential antag-
onists predicted to bind more strongly than known ligands.
Next, we carried out R-group screening to suggest a new
ligand predicted to bind with 7 kcalmolÀ1 better energy than
1-{2-[4-(2-bromobenzyl)-4-hydroxypiperidin-1-yl]ethyl}-3-
ed to have the highest affinity for the receptor. Some of these
predictions were tested experimentally, validating the compu-
tational results. Using the pharmacophore generated from the
predicted structure for hUT2R bound to ACT-058362, we carried
out virtual screening based on this binding site. The most
potent hit compounds identified contained 2-(phenoxymeth-
yl)-1,3,4-thiadiazole core, with the best derivative exhibiting an
IC50 value of 0.581 mm against hUT2R when tested in vitro. Our
efforts identified a new scaffold as a potential new lead struc-
ture for the development of novel hUT2R antagonists, and the
computational methods used could find more general applica-
bility to other GPCRs.
(thieno[3,2-b]pyridin-7-yl)urea, the designed antagonist predict-
Introduction
Urotensin-II (U-II) isolated from goby fish urophysis is a C-ter-
minal conserved cyclic peptide (ETPDCFWKYCA) with a disulfide
constraint.[3] The human isoform of U-II, an undecapeptide (AG-
TADCFWKYCV), is the most potent mammalian vasoconstrictor
known—10 to 100 times more potent than endothelin-1.[1,3]
The identification of this peptide as the endogenous ligand for
the orphan G protein-coupled receptor (GPCR), originally desig-
nated GPR14,[1] which is mainly expressed in the cardiovascular
system, has stimulated interest in developing novel nonpepti-
dic human U-II receptor (hUT2R) agonists and antagonists that
might be of therapeutic value in cardiovascular disorders char-
acterized by increased vasoconstriction, myocardial dysfunc-
tion, and even atherosclerosis.[2]
peptide revealed that the W7K8Y9 triad is required for receptor
recognition and activation.[4] Correlating this pharmacophore
study, all reported compounds of hUT2R antagonists contain
a basic amino group and at least two aromatic moieties.[5] The
recently described P5U superagonist, with higher affinity than
U-II at hUT2R as well as higher activity in the rat thoracic aorta
assay, where a cysteine residue in position 5 is replaced by
penicillamine (b,b-dimethylcystein), adopts
a well-defined
type II’ beta-hairpin structure as determined by nuclear mag-
netic resonance (NMR) studies.[2] Structure–activity relationship
(SAR) studies indicate that the replacement of the tryptophan
residue with the corresponding d-isomer switches the activity
from agonist to partial agonist. Similarly, partial agonists are
obtained by replacement of the lysine residue with ornithine,
which lacks one methylene unit. The simultaneous presence of
a d-tryptophan residue in position 7 and an ornithine residue
in position 8 leads to the potent antagonist urantide.[6] The
pharmacophoric distances of key residues, W7, K8, Y9, for the
hUT2R interaction, provided the first successful application of
virtual screening (VS) based on a pharmacophore generated
from SAR and NMR studies on the physiological peptidic
ligand of its receptor.[4]
The sequence WKY for hUT2R appears to be very important
for biological activities, whereas the disulfide bridge of hUT2R
is of minor importance. Alanine scans of the native undeca-
[a] Dr. S.-K. Kim, Prof. W. A. Goddard, III
Materials & Process Simulation Center, California Institute of Technology
Beckman Institute (BI);Code (139-74)
1200 E. California Blvd., Pasadena, CA 91125 (USA)
E-mail: wag@wag.caltech.edu
[b] Dr. K. Y. Yi, Dr. B. H. Lee, C. J. Lim
In order to understand the species selectivity of human and
rat UT2R, we previously studied a hUT2R-selective antagonist,
ACT-058362 (hUT2R Kd: ~4 nm vs rUT2R Kd: ~1500 nm),[7] and
Division of Drug Discovery Research
Korea Research Institute of Chemical Technology
Yuseong-gu, Daejeon 305-343 (Republic of Korea)
[c] Prof. B. Trzaskowski
a
non-selective arylsulphonamide-based antagonist,[8] SB-
Chemistry Department, University of Warsaw
706375 (hUT2R Kd: ~9 nm vs rUT2R Kd: ~21 nm).[2] We investi-
gated the binding site of ACT-058362 and SB-706375 on both
human and rat UT2R to explain the dramatic (~400-fold) lower
´
Krakowskie Przedmiescie 26/28 Warszawa, 00-927 (Poland)
Supporting information for this article is available on the WWW under
http://dx.doi.org/10.1002/cmdc.201402087.
ꢀ 2014 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim
ChemMedChem 2014, 9, 1732 – 1743 1732

CHEMMEDCHEM
FULL PAPERS
www.chemmedchem.org
affinity of ACT-058362 for rUT2R and the similar (~10 nm) affini-
ty of SB-706375 for both UT2Rs. We predicted that ACT-058362
binds in the transmembrane (TM) 3456 region while SB-706375
binds in the TM 1237 region. These predicted sites explained
the known differences in binding ACT-058362 to the rat and
human receptor, while explaining the similar binding of SB-
706375 compound.
human chemokine CXCR4 (hCXCR4),[20] human H1 histamine re-
ceptor (hH1HR),[21] rat M3 muscarinic receptor (rM3MR),[22]
human M2 muscarinic receptor (hM2MR),[23] and bovine rhodop-
sin (bRho).[24] We selected these nine templates based on high
sequence identity to hUT2R from our GEnSeMBLE analysis (TM
sequence identities in parentheses): hNOPR (36.0%), mOPRD
(34.5%), mOPRM (34.4%), hOPRK (32.8%), hD3DR (31.3%),
hAA2AR (27.9%), hb2AR (27.3%), hCXCR4 (27.1%) and bRho
(21.8%). We also selected the templates with lower than 30%
sequence identity for the diversity of templates and the valida-
tion of our methods. For families in which more than one
GPCR is known, we chose the structure with the highest se-
quence identity.
In order to provide a structural model for understanding the
experimental results and for predicting new ligands for hUT2R,
we report now the results of applying the “GPCR ensemble of
structures in membrane bilayer environment” (GEnSeMBLE)
complete sampling Monte Carlo method, which samples ~1.7
trillion combinations of TM helix rotations (h) and tilts (q, f)
for each of 12 cases of 9 templates from experiment (X-ray) or
previous calculations to select an ensemble of 10 low-energy
packings to be used to then predict the binding site and
energy for both known and new ligands.
To obtain the optimum packings of the hUT2R helices in
these templates, we first considered rotating each of the 12
transmembrane domains (TMD) by increments of 158 over the
range of Æ608, a total of (9)7 =4.7 million configurations using
the BiHelix method. This was carried out for the following 12
cases:
With the aim of providing a basis for rational design of new
hUT2R antagonists, we first used the DarwinDock Monte Carlo
complete sampling method to predict the binding site to
hUT2R of four known hUT2R selective antagonists: ACT-058362,
SB-706375, GSK-1440115, and GSK-1562590, which will here-
after be referred to as ACT, SB, GSK1, and GSK2, respectively.
*
Nine cases using homology templates: 1. hNOPR, 2.
mOPRD, 3. mOPRM, 4. hOPRK, 5. hD3DR, 6. hAA2AR, 7.
hb2AR, 8. hCXCR4, 9. bRho.
*
One case using the area hydrophobic center (HPC) method
1) ACT-058362: a 4-ureido-quinoline palosuran; 1-[2-(4-benzyl-
4-hydroxy-piperidin-1-yl)ethyl]-3-(2-methyl-quinolin-4-
yl)urea sulfate salt.[8]
to position the helices with respect to a common mem-
brane midplane: 10. hb2AR GEnSeMBLE.
*
One case using the rawmid HPC method to position the
2) SB-706375: a arylsulphonamide; 2-bromo-4,5-dimethoxy-N-
[3-(R)-1-methyl-pyrrolidin-3-yloxy]-4-trifluoromethylphenyl)-
benzene sulfonamide hydrochloride.[9]
helices with respect to a common membrane midplane: 11.
hb2AR GEnSeMBLE.
*
One case that we had previously predicted using an early
3) GSK-1440115: (4’-[(1R)-1-[[(6,7-dichloro-3-oxo-2,3-dihydro-
4H-1,4-benzoxazin-4-yl)acetyl](methyl)amino]-2-(4-morpholi-
nyl)ethyl]-4-biphenylcarboxylic acid trifluoroacetate.[10]
version of GEnSeMBLE, which had been subjected to 1 ns
of MD after inserting in the membrane and water box: 12.
hUT2R MembStruk.[25]
4) GSK-1562590:
4’-[(1R)-1-[[(6,7-dichloro-3-oxo-2,3-dihydro-
4H-1,4-benzoxazin-4-yl)acetyl](methyl)amino]-2-(1-pyrrolidi-
For each of these 12 cases, we selected the lowest energy
2000 combinations out of the 4.7 million possibilities to build
into 7-helix bundles (CombiHelix) for which we optimized the
side chains followed by some minimization. Then we com-
pared the interhelical interaction energies for all 12*2000=
24000 cases using an energy scoring in which each amino acid
side chain is neutralized (NiH). We finally selected the best 20
to consider optimizing the tilts of the helices.
nyl)ethyl]-3-biphenylcarboxamide hydrochloride.[10]
In each case, we considered the lowest 10 structures from
the ensemble of hUT2R structures. Using this binding site, we
proceeded to predict modifications of GSK1 expected to im-
prove binding, and we used virtual screening with the result-
ing pharmacophore to suggest a new scaffold expected to im-
prove binding.
From these 20 configuration, we selected six based on con-
siderations of energy and diversity, and considered 1.7 trillion
combinations of the rotation angles (h) and tilt angles (q, f) to
select the best 2000 to pack into 7-helix bundles. Using an
energy scoring in which each amino acid side chain is neutral-
ized (NiH), we selected from these 12000 cases. All ensembles
of the top 20 are from the mOPRD template (Table 1), suggest-
ing a close correspondence of UT2R with mOPRD, which may
be responsible for the observation that a side effect of pain re-
lieving drugs that target opioid receptors is to activate hUT2R.
Figure 1 shows the final best structure for hUT2R, which de-
rives from mOPRD only by h variations of À158 for TM 5, q de-
viations of À108 for TM 4, and f changes of À158 for TM 3–5.
However, many of the other low-energy structures have larger
deviations that may play a role in activation.
Results
GEnSeMBLE predictions of apo-protein structures for hUT2R
Details of the GEnSeMBLE predictions are given in Appendix A
in the Supporting Information. For hUT2R, we considered the
following nine X-ray structure templates: human nociceptin/or-
phanin FQ receptor (hNOPR),[11] mouse delta opioid receptor
(mOPRD),[12] mouse mu opioid receptor (mOPRM),[13] kappa
opioid receptor (hOPRK),[14] human D3 dopamine receptor
(hD3DR),[15] turkey b1 adrenergic receptor (tb1AR),[16] human ad-
enosine A2A receptor (hAA2AR),[17] human b2 adrenergic receptor
(hb2AR),[18] human sphingosine 1-phosphate (hS1P1R),[19]
ꢀ 2014 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim
ChemMedChem 2014, 9, 1732 – 1743 1733

CHEMMEDCHEM
FULL PAPERS
www.chemmedchem.org
Table 1. Top 20 predicted structures of the human urotensin-II receptor (hUT2R) selected as the lowest energy configures from the 12000 final structures
from the SuperBiHelix analysis of the 6 BiHelix cases (shaded in gray in Table S2 and S3 in the Supporting Information).[a]
Apo-#
V
f
h
BiHelix RMSD
H1 H2
H3 H4
H5 H6 H7
H1
H2
H3
H4
H5
H6
H7
H1
H2
H3
H4
H5
H6 H7
Rank
[ꢁ]
1
2
3
4
5
6
7
8
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
À10
0
0
0
0
0
À10
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
10
10
0
0
0
0
0
0
10
10
10
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
À15 À15 À15
0
0
0
0
15
15
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
À15
À15
À30
15
0
0
0
0
0
0
15
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
Top1
Top1
Top1
Top9
Top9
Top9
Top1
Top9
Top1
Top9
Top9
Top1
Top9
Top9
Top1
Top1
Top9
Top1
Top9
Top9
1.07
1.16
0.70
1.13
1.41
1.21
2.09
1.40
1.48
1.42
1.49
0.87
1.73
1.05
0.76
0.73
1.19
1.72
1.18
1.09
À15 À15
15
À15
À15
0
0
0
0
À15
À15 À30 À15
0
À15
0
0
À15
15
À15 À15
À10 10
À15 À15
15 À15 À15 À30 À15
À30 À30 À15
15
0
0
À15
À30
0
0
0
0
0
10
10
10
10
0
0
0
0
0
0
0
0
0
0
0
À15 À15 À15
15
15
15
15
0
0
15
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
15
9
0
À15
0
À15 À30
À15
0
10
11
12
13
14
15
16
17
18
19
20
0
À15 À15
À15 À30
30 À30
À15 À30
0
0
15
15
15
0
À10 10
À15 À15
15 À15 À15
15 À15
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
À15 À15 À15
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
À15 À30
15 À30
0
0
0
À30 À15
À15 15
À30 À15
0
15 À30
10
15
30
15
15
À15 À15
10 10
0
0
À15 À15 À30
À15 À15
À15 À15 À30
0
À15
0
À30
15
0
10
0
0
0
0
[a] This SuperBiHelix analysis considered: Æ108 sampling of q tilt angles, Æ308 sampling of both f and h angles by 158 increments, leading to a total of
(3ꢂ5ꢂ5)7 ~13 trillion combinations. The structures were ordered by the average rank from four energy scorings: charge interhelical (CiH), charge total
(CTot), neutral interhelical (NiH), and neutral total (NTot). The top 20 structures identified all came from the mOPRD template.
Predicted binding sites for the predicted hUT2R structures
The low-lying packing structures have important interhelical
couplings. Classical interhelical H-bonding networks were
found in TM 1,2–7 (N1.50, D2.50 and N7.49) between D2.50
and N7.49 except for apo-1 and between N1.50 and D2.50 in
close proximity of 5 ꢁ among the top 20. However, none of
the top 20 structures contain a salt-bridge between R3.50 and
D(E)6.30 in the D/ERY region that is observed in many other
class A GPCRs. This is expected because residue 6.30 is K273 in-
stead of D or E.
Predicting the binding site of ACT to hUT2R
We used the DarwinDock complete sampling Monte Carlo
method to predict ~50000 poses for each of the seven confor-
mations of ACT to finally select the best binding site (for de-
tails, see Experimental Section). We did this for each of the top
10 low-energy 7-helix bundles predicted from SuperBiHelix (all
based on mOPRD). We found that the best structure binds to
apo-10 of the ensemble based on the best binding energy
from all ensemble docking conformations.
Instead, we find that R1554.42 makes a salt bridge with
E1473.49 coupling TM 3 and TM 4 in the D/ERY region, as do
apo-5, 8, 10, 11, and 14 out of the top 20 structures. There are
H-bonding networks among S852.38, Y892.42 and E1473.49, cou-
pling TM 2 with TM 3 in the D/ERY region between S2.38 and
E3.49 (except for apo-7, 11, 13, 17, 19) and between Y2.42 and
E3.49 (except for apo-11, 12, 15, 17, and 19 out of the top 20
structures). TM 2–4 H-bonding networks—either between
N922.45 and R1664.42 for apo-1, 2, 7, 13, 18 or between S852.38
and R1664.42 for apo-3-6, 8-11, 1, 17, 19—in the top 20 struc-
tures are expected to make these configurations inactive.
Some structures like apo-5, 8, and 10 have an extra TM 3–5
H-bonding network between R1483.50 and Y2315.58. Some struc-
tures have additional TM 3–5 interactions between H1353.37
and S2195.46 at the middle of TMD for apo-2, 14, 20, as well as
between S1453.47 and Y2315.58 only for apo-5 in the proximity
of the D(E)RY region. Apo-5-11, 14, 17, 19 and 20 structures
have TM 2–6 among Y872.40 and R2556.32. Finally, all structures
except for apo-13 have TM 1–7 interaction between D471.38
As shown in Figure 2A, critical components of the binding
are: 1) a salt bridge from the protonated nitrogen to D1303.32
;
2) an H-bond of the carbonyl oxygen atom with T3057.42; 3) the
phenyl group in a hydrophobic pocket formed by I1072.60 and
L1263.28; 4) the indole ring in a hydrophobic pocket formed by
L2125.39 and, L2155.42, F2165.43, I2205.47 and W2756.48
.
To validate this predicted binding pose for ACT, we matched
five ACT analogues to the low-energy predicted structures of
hUT2R from DarwinDock and minimized. Two of the binding
modes of ACT63 and ACT97 (the number from the conforma-
tional search) have a salt bridge at D1303.32, which is the major
anchoring point in hUT2R. The binding mode of ACT63, which
has a cis amide bond in the ureido group, shows additional in-
teractions at L2155.42 with the terminal protonated amine in
the indoline ring, while the binding mode of ACT179, which
has a trans conformation, and ACT211 with opposite orienta-
tion, has no additional H-bonding.
and R2947.32
.
We examined five ACT analogues (ACT-1 to ACT-5) to assess
how well our predicted structure explains the experimental
SAR (Table 2). Matching the five ACT analogues into the three
ꢀ 2014 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim
ChemMedChem 2014, 9, 1732 – 1743 1734

CHEMMEDCHEM
FULL PAPERS
www.chemmedchem.org
We selected six extended conformations for the docking. The
best pose of SB bound to the hUT2R–mOPRD leads to the
structure in Figure 2B with the following features: 1) a salt
bridge between the protonated amine and D1103.32, the major
anchoring point for SB to hUT2R; 2) an H-bonding at Y3067.43
with the O atoms of the sulfonamide group; 3) a hydrophobic
interaction at the terminal methoxy phenyl group with hydro-
phobic residues (F1112.64 and Y2987.36); 4) a pi–pi interaction at
the center phenyl group with aromatic residues F2746.51 and
F1273.29
.
Docking of GSK1 and GSK2 to hUT2R
We predicted the binding site and energy of GSK1 (IC50 =
0.048 mm) to the top 10 low-lying packing protein conforma-
tions from SuperBiHelix, predicting the structure shown in Fig-
ure 3A. The best pose is from the conformation number 40
out of seven different ligand conformations and WT (wild-type)
10 out of top 10. The predicted structures have the following
features: 1) a salt bridge at D1303.32 to the protonated amine;
2) an H-bonding between the carbonyl O atom and Y1112.64
;
3) an aromatic–aromatic interaction at F1273.29 (À5.63 kcal
molÀ1), Y2987.35 (À4.04 kcalmolÀ1), F2746.51 (À2.86 kcalmolÀ1),
and F1313.33 (À2.72 kcalmolÀ1) with the biphenyl group; 4) a
hydrophobic interaction at the hydrophobic residues, F1313.33
,
M1343.36, L2125.39, L2155.42, and W2756.48 with the benzoxazoli-
none ring.
The binding of GSK1 (UCavE=À42.09 kcalmolÀ1) is 2.72 kcal
molÀ1 less favorable than the binding of ACT (UCavE=
À44.81 kcalmolÀ1), which is consistent with the observed two-
times better binding of ACT. We also predicted the binding
site and energy of GSK2 (IC50 =0.004 mm) to 10 low-lying pack-
ing protein conformations from SuperBiHelix, predicting the
structure shown in Figure 3B and displaying the same interac-
tions as GSK1.
Figure 1. Modelled H-bonding networks of human urotensin-II receptor
(hUT2R) based on the mouse delta opioid receptor (hUT2R-mOPRD)[12] in
upper (A), middle (B), and lower (C) transmembrane (TM) regions. The salt
bridge or H-bonding is shown in a red circle. We used Ballesteros–Weinstein
numbering consisting of the TM helix number followed by residue number
relative to the highly conserved residue in the helix, numbered as 50.
Predicting new ligands using multiple R-group optimization
best binding poses, ACT63, ACT97, and ACT211, we found that
the binding mode of ACT97 has the best correlation (r2 =0.83)
compared with the binding modes of ACT63 (r2 =0.18) and
ACT211 (r2 =0.33). Thus, only this binding mode of ACT97
agrees with the experimental observations.
Starting from the predicted best binding site of ACT-1 bound
to the apo-10 conformation of hUT2R from mOPRD, we
searched multiple R groups, considering the optimization for
three positions as shown in Table 3. Since a CH3 group in the
R1 position is better than H, we considered the functional
groups of CH3, CH2CH3, CH=CH2, and CH2OH in the R1 posi-
tion. We found that single substitution at R1 with any of the
four substituents increased the binding affinity. Since a Br
group in the R2 group is slightly better than H in the R1 posi-
tion, we considered OCF3, CF3, N=C=S and I in the R2 position
as plausible analogues of Br. We found that for single substitu-
tion at the R2 position, the CF3 and OCF3 groups are best. To
optimize the R3 position, we considered CH3, Cl, I, and phenyl
groups to providing maximum diversity. We found that for
single substitution at the R3 position, I, Cl, and CH3 are better
than H (ACT-1).
From our docking study, we find that the protonated amine
in the corresponding part of K8 interacts with D1303.32 while
the indoline ring of W7 interacts with L2155.42 and W2756.48. For
the Y9 position, the terminal phenyl group forms a pi–pi inter-
action with Y1112.64
.
Docking of SB to hUT2R
We predicted the binding site and energy of the SB antagonist
to the top 10 low-energy predicted structures of hUT2R–
mOPRD from SuperBiHelix. The observed IC50 value for SB
against the human reception is 0.073 mm, three times higher
than the experimental IC50 value of ACT (IC50 =0.024 mm). The
conformational search of SB generated 17 conformations out
of 360 conformations with the diversity of 2.5 and 1.5 ꢁ RMSD.
To suggest new ligands with improved binding, we then
considered nine di-substituted derivatives, and finally 22 candi-
dates in Table 3 with up to triple excitations. Based on this R-
ꢀ 2014 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim
ChemMedChem 2014, 9, 1732 – 1743 1735

CHEMMEDCHEM
FULL PAPERS
www.chemmedchem.org
Figure 2. The docking result of human urotensin-II receptor (hUT2R)-selective ligands at the hUT2R-mOPRD model derived using the mouse delta opioid re-
ceptor (mOPRD) as a template.[12] The binding site and 2D ligand interaction diagram, determined using the Maestro program[46] for A) ACT-058362 and B) SB-
706375. We used Ballesteros–Weinstein numbering consisting of the TM helix number followed by residue number relative to the highly conserved residue in
the helix, numbered as 50. H-bonding is indicated by the dotted lines.
Table 2. The experimental IC50 and predicted binding energy of ACT-
058362 analogues.[a]
(Figure 4). Comparing the binding of ACT-1 and the other ana-
logues, we predict that these compounds might bind up to
~7 kcalmolÀ1 better energetically.
Predicting new ligands using virtual screening
To identify novel hUT2R-selective lead compounds, we per-
formed virtual screening using the predicted pharmacophore
of ACT to three sets of hUT2R structures, hUT2R_fRho, hUT2R_
CXCR4 and hUT2R_mOPRD. Experimental tests were carried out
for the hUT2R_fRho results. Thus, to compare with experimen-
tal results, we will discuss just the results for hUT2R_fRho.
We used LigandScout 3.0[26] to for search new leads for
hUT2R ligands. Based on the pharmacophore score, we select-
ed 5036 hit molecules from the ZINC database that includes
19.6 million compounds, each with 100 conformations. For
each of these 5036 compounds, we used our predicted protein
structure with our scoring methods to select the best 200
poses from which we selected the best 10 molecules using
two alternative scoring methods. These are listed in Table S1
and Figure S1 of the Supporting Information. Of these top 10
from each scoring method, we selected the six predicted to be
best for experimental study, leading to the compounds shown
Compd
R1 R2 R3
X
IC50 [mm]
UCavE
BE
[kcalmolÀ1
]
[kcalmolÀ1
]
ACT-058362
ACT-1
ACT-2
ACT-3
ACT-4
H
H
H
H
H
H
Br
H
H
0.015Æ0.008
À37.08
À35.42
À36.37
À39.15
À37.45
À23.62
À51.94
À63.28
À61.41
À65.80
À62.48
À64.58
H
H
Br 0.021Æ0.007
Me
H
H
H
H
H
H
H
0.026Æ0.003
0.030Æ0.010
0.069Æ0.048
3.607Æ0.966
H
H
Me
H
ACT-5
[a] Unified cavity energy (UCavE) and binding energy (BE) were calculat-
ed; BE=complexEÀproteinEÀligandE). IC50 data shown are the meanÆstandard
deviation of three separate experiments.
group screening for ACT-1, we suggest that CH=CH2, CH2OH,
and CH3 in the R1 position and I, Cl, and CH3 in the R3 position
are the best candidates for improved hUT2R interactions
ꢀ 2014 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim
ChemMedChem 2014, 9, 1732 – 1743 1736

CHEMMEDCHEM
FULL PAPERS
www.chemmedchem.org
mOPRD template has the best
intermolecular H-bonding net-
works compared with the other
two templates, as indicated by
the better energies in the Super-
BiHelix analysis.
When we examined the H-
bonding networks for all 20 en-
sembles of the hUT2R-mOPRD
template, we found the classical
H-bonding networks in TM 2–4
(N2.45 and W4.50) known for
many members of the rhodopsin
family A GPCRs (the top 20 also
shared this interaction except for
apo-7, 13, 17, and 18). We also
found hUT2R-specific interac-
tions: TM 2–4 networks in the
D(E)RY region among S852.48
R1664.42 and E1473.49
The H-
,
,
.
bonding network between S2.38
and E3.49 was shown for all
except for apo-7, 11, 13, 17, and
19, and the interaction between
Y2.42 and E3.49 was shown for
all except for apo-11, 12, 15, 17,
and 19. In addition, there are
TM 2–4 H-bonding networks
either between N922.45 and
R1664.42 for apo-1, 2, 7, 13, 18 or
between S852.38 and R1664.42 for
apo-3-6, 8-11, 1, 17, 19 in the
top 20, and TM 3–5 H-bonding
Figure 3. The binding site and 2D ligand interaction diagram, determined using the Maestro program, for A) GSK-
1440115 (GSK1) and B) GSK-1562590 (GSK2) at the hUT2R-mOPRD model derived using the mouse delta opioid re-
ceptor (mOPRD) as a template.[12]
in Table 4. The best measured affinity for hUT2R leads to 89.7%
binding inhibition at 10 mm concentration. The top two ranked
compounds by calculated UCavE value have similar structures
and exhibit the most potent inhibition of hUT2R; compounds
#1 and #2 (ZINC02322315) in Table 4 were determined to have
IC50 values of 1.40 mm and 0.581 mm, respectively.
networks between R1483.50 and Y2315.58 for apo-2, 14, 20, as
well as between S1453.47 and Y2315.58 only for apo-5 in the
proximity of the D(E)RY region. However, none of the top 20
contain a salt bridge between R3.50 and D(E)6.30 in the D/ERY
region, which is observed in many other class A GPCRs. This is
expected because residue 6.30 is K273 instead of D or E. In-
stead, we find that R1554.42 makes a salt bridge with E1473.49
,
coupling TM 3 and TM 4 for apo-5, 8, 10, 11, and 14 structures.
Thus, diverse ensemble structures were stabilized by various
H-bonding networks, and conserved interactions between
TM 3 and TM 6 in the D/ERY motif in class A GPCRs were re-
placed by TM 2–4 networks.
Discussions
To predict the structures of antagonists bound to hUT2R that
might be useful in developing improved ligands, we applied
the GEnSeMBLE method that samples ~13 trillion potential
packings (considering rotations and tilting of all seven TMD) to
each of nine GPCR templates to finally select the best 10 that
might be important in binding antagonists or activation by ag-
onists.
To determine the binding mechanism of hUT2R antagonists,
we docked various hUT2R-selective antagonists, ACT, SB, GSK1,
and GSK2. From the docking of ACT, SB, GSK1, and GSK2 with
the apo-10 low-energy predicted structures from SuperBiHelix
on hUT2R-mOPRD, we find major interactions at D1303.32 with
the protonated nitrogen through a salt-bridge interaction, indi-
cating that such interaction is necessary for hUT2R recognition.
The presence of both aromatic and positively charged group
leads to compounds with similar interactions. However, addi-
tional H-bonding depends on the ligand type. The main bind-
ing difference observed in three structures is an additional H-
bond. For ACT, H-bonding forms between the carbonyl oxygen
The best 20 7-helix bundles all derive from the mOPRD ho-
mology template, which has the second highest sequence
identity in TM (34.5%) among the nine templates evaluated.
The H-bonding networks are highly dependent on the tem-
plate chosen. The hUT2R-mOPRD structure displays an extra H-
bonding that is not found in the other two templates: TM 2–
6–7 among Y872.40, R2556.32 and T3197.58, and TM 1–7 between
D471.38 and R2947.32. Thus, the final best structure from the
ꢀ 2014 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim
ChemMedChem 2014, 9, 1732 – 1743 1737

CHEMMEDCHEM
FULL PAPERS
www.chemmedchem.org
Table 3. The predicted binding energies from R-group screening of ACT-
1. Compounds are ordered by unified cavity energy, thus we suggest #32
and 25 as the best candidates.[a]
#
R1
R2
R3
UCavE
[kcalmolÀ1
SnapBE
[kcalmolÀ1
]
]
32
17
27
25
16
22
31
3
29
26
30
15
11
4
CH2OH
Me
CH=CH2
CH=CH2
Me
Br
Br
Br
Br
Br
Br
Br
Br
CF3
Br
Br
Br
Br
Br
Br
CF3
CF3
Br
I
I
I
À42.47
À41.77
À39.81
À39.51
À39.50
À39.47
À39.16
À38.24
À38.22
À38.10
À37.96
À37.80
À37.77
À37.33
À37.19
À36.82
À36.49
À36.49
À35.57
À35.36
À35.23
À35.05
À61.30
À60.17
À61.09
À64.97
À56.87
À61.06
À57.15
À62.87
À55.12
À63.12
À57.28
À56.75
À54.67
À55.64
À54.98
À55.72
À53.56
À57.08
À57.45
À53.75
À55.89
À53.05
Me
Cl
I
Cl
H
H
Cl
Me
Me
I
H
H
H
H
Cl
Me
H
H
H
Et
CH2OH
CH=CH2
CH2OH
CH=CH2
CH2OH
Me
H
Figure 4. Prediction of improved substituents from R-group screening (top)
based on the predicted binding site of ACT-1 at human urotensin-II receptor
(hUT2R) (bottom). N-term: N-terminal and EC2: the second extracellular loop.
CH2OH
Me
Et
Me
Et
Et
CH2OH
Et
1
19
14
21
20
28
2
Br
OCF3
Br
hit has an IC50 value of 0.581 mm (Table 4), representing a new
lead for the development of hUT2R antagonists. Since most
known hUT2R antagonists have poor pharmacology profiles,
new leads will provide the opportunity to develop improved
compounds as selective renal vasodilator in renal ischemia.
13
Me
OCF3
[a] Unified cavity energy (UCavE) and binding energy (BE) were calculat-
ed; BE=complexEÀproteinEÀligandE).
atom and T3057.42. However, an additional H-bond is shown at
the sulfonamide group of SB with Y3067.43/Y1112.64 and the car-
bonyl atom of GSK with Y2987.36/Y1112.64. The binding site of
Conclusions
This study shows the importance of considering the ensemble
of the best 10 to 20 models of the 7-helix bundle for a GPCR,
as exemplified using hUT2R, a validated target for the treat-
ment of renal ischemia. Thus, the best antagonist binding site
was found in the 10th structure of the apo-protein hierarchy.
We consider that these GEnSeMBLE and DarwinDock tech-
niques, which estimate the energies for massive numbers of
conformations and structures but rapidly decrease them to an
ensemble of 10–100 for detailed scoring, will be useful for
many other GPCRs, most of which have little experimental
data available that is useful for ligand design. Our predicted
3D structures for hUT2R are expected to provide increased un-
derstanding of the binding mechanism of hUT2R, which should
be valuable in developing improved selective antagonists.
three classes of hUT2R antagonists is overlapped at D1303.32
.
Except for SB, which mainly interacts at TM 1–2–7, the main in-
teraction of ACT and the GSK compounds forms at TM 3–5.
The benzopyran ring of both GSK1 & 2 binds at the extra sites
close to upper TM 7 and the second extracellular loop (EC2).
Thus, our docking study predicted a different binding mode
for three classes of hUT2R antagonists, displaying a different
spatial disposition of the pharmacophoric elements.
For ligand optimization, we used R-group screening
(Table 3), with the predicted structure for ACT-1 in our predict-
ed hUT2R structure to select the compounds shown in
Figure 4, which we suggest might exhibit improved bind over
ACT-1 to hUT2R by ~7 kcalmolÀ1. According to the information
from the docking experiments, we could extend the hydropho-
bic interactions at three leucine residues (45, 198, and 212)
from TM 5 and loops.
Experimental Section
Synthesis of ACT derivatives
For lead generation, based on the binding site for ACT-1, we
used our best ligand–protein structure as the basis for virtual
screening. This led to the identification of a new scaffold: 2-
(phenoxymethyl)-1,3,4-thiadiazole. The best virtual screening
The general synthetic route employed for the preparation of ACT
derivatives (ACT-1–5) is outlined in Scheme 1. Thieno[3,2-b]pyridin-
yl carboxylic acids (1) were reacted with diphenyl phosphoryl azide
ꢀ 2014 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim
ChemMedChem 2014, 9, 1732 – 1743 1738

CHEMMEDCHEM
FULL PAPERS
www.chemmedchem.org
tography was carried out using
silica gel (230–400 mesh). All sol-
vents and reagents were commer-
cially available and used without
purification.
Table 4. The experimentally determined inhibitory activities (inhibition at 10 mm and IC50) of virtual screening
hits against human urotensin-II receptor (hUT2R). Hits were identified by virtual screening against the crystal
structure of frog rhodopsin (hUT2R-fRho) in complex with ACT-058362.[a]
#
Compd
Inhib. @10 mm
IC50 [mm]
UCavE
[kcalmolÀ1
BE
[%]
]
[kcalmolÀ1
]
Thieno[3,2-b]pyridine-7-carbonyl
azide (2): A solution of thieno[3,2-
b]pyridine-7-carboxylic
acid
(568 mg, 3.18 mmol) in DMF
(10 mL) was treated with DPPA
(1.02 mL, 4.77 mmol) and Et3N
(0.89 mL, 6.36 mmol) and stirred at
RT for 5 h. The mixture was diluted
with water (50 mL), and extracted
with EtOAc (2ꢂ50 mL). The organ-
ic layer was washed with brine (2ꢂ
25 mL), dried over anhyd Na2SO4,
filtered and concentrated in vacuo.
Purification of the crude residue
by column chromatography on
silica gel (EtOAc/n-hexane, 1:3)
gave 2 as a white solid (195 mg,
30%): 1H NMR (300 MHz, CDCl3):
d=7.94 (d, J=5.4 Hz, 1H), 7.83 (d,
J=4.7 Hz, 1H), 7.66 (d, J=5.4 Hz,
1H), 7.39 ppm (d, J=4.7 Hz, 1H).
1
89.7
87.7
17.2
1.40Æ0.19
À44.20
À50.88
À43.79
À47.53
À58.69
À59.62
2
3
0.59Æ0.07
–
4
5
0.0
0.0
–
–
À40.49
À38.33
À47.73
À52.07
1-[2-(4-Benzyl-4-hydroxypiperi-
din-1-yl)ethyl]-3-(thieno[3,2-
b]pyridin-7-yl)urea (ACT-3): A so-
lution of 2 (50 mg, 0.24 mmol) in
toluene (3 mL) was heated at
110 8C for 1 h. After cooling to RT,
a solution of 1-(2-aminoethyl)-4-
benzylpiperidin-4-ol
(84 mg,
6
7
0.0
–
–
À38.22
À23.05
À46.84
À54.64
0.36 mmol) and Et3N (0.1 mL,
0.72 mmol) in CH2Cl2 (3 mL) was
added to the reaction mixture, and
stirring was continued for 1 h. The
mixture was diluted with water
(20 mL), and extracted with CH2Cl2
(2ꢂ20 mL). The organic layer was
washed with brine (2ꢂ20 mL),
dried over anhyd Na2SO4, filtered
and concentrated in vacuo. Purifi-
cation of the crude residue by
column chromatography on silica
gel (MeOH/CH2Cl2, 20%) gave ACT-
3 as a yellow solid (20 mg, 20%):
1H NMR (300 MHz, CD3OD): d=
8.43 (d, J=3.3 Hz, 1H), 8.01 (d, J=
3.4 Hz, 1H), 7.93 (d, J=3.3 Hz, 1H),
7.46 (d, J=3.4 Hz, 1H), 7.20–7.29
(m, 5H), 3.61–3.66 (m, 2H), 3.47–
3.51 (m, 2H), 3.17–3.28 (m, 4H),
2.82 (s, 2H), 1.90–1.95 (m, 2H),
20.9
8
13.4
–
À36.68
À52.75
[a] Unified cavity energy (UCavE) and binding energy (BE) were calculated; BE=complexEÀproteinEÀligandE). IC50 data
shown are the meanÆstandard deviation of three separate experiments.
1.75–1.81 ppm (m, 2H); 13C NMR (125 MHz, DMSO): d=36.1, 36.8,
49.4 (2C), 55.3, 57.6, 60.2, 68.3, 106.6, 121.1, 125.8, 126.2, 128.0
(2C), 130.5, 131.0 (2C), 138.3, 142.6, 148.6, 154.7, 157.6; HRMS (EI):
calcud for C22H26N4O2S [M]+: 410.1776, found: 410.1796.
(DPPA) using Et3N as the base to give corresponding acyl azides 2.
The target ureas were obtained by reaction of thieno[3,2-b]pyridin-
yl isocyanates 3, which were generated in situ by Curtius rear-
rangement of azides 2, with amine 4.
1H NMR spectra were recorded on a Varian Gemini 200 or Bruker
DRX-300 spectrometer. High-resolution mass spectra (HRMS) were
recorded on a JEOL JMS-700 mass spectrometer. Column chroma-
1-{2-[4-(2-Bromobenzyl)-4-hydroxypiperidin-1-yl]ethyl}-3-(thieno-
[3,2-b]pyridin-7-yl)urea (ACT-1): Reaction of 1-(2-aminoethyl)-4-(2-
bromobenzyl)piperidin-4-ol (114 mg, 0.37 mmol, 1.5 equiv) as de-
ꢀ 2014 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim
ChemMedChem 2014, 9, 1732 – 1743 1739

CHEMMEDCHEM
FULL PAPERS
www.chemmedchem.org
hUT2R binding assay
Filtration-based time-resolved fluo-
rescence (TRF) receptor binding
assays with europium-labeled U-II
(Eu-UII) were performed in 96-well
AcroWell plates with GH polypro
(GHP) membrane (PALL Life Scien-
ces, Ann Arbor, USA) by incubating
10 mg per well HEK293-hUT mem-
branes with 2 nm of Eu-UII in
a total assay volume of 100 mL. U-II
was labeled with europium (Eu) at
the N1 position by the PerkinElmer
labeling service (Waltham, USA),
and hUT2R membranes were pre-
pared from human embryonic
kidney (HEK293) cells, stably ex-
pressing hUT2R. The assay buffer
contained 25 mm 4-(2-hydroxy-
ethyl)piperazine-1-ethanesulfonic
acid (HEPES), 5 mm MgCl2, 1 mm
CaCl2, 0.5% bovine serum albumin
(BSA) pH 7.4. Nonspecific Eu-la-
Scheme 1. Reagents and conditions: a) diphenyl phosphoryl azide (DPPA), Et3N, DMF, RT, 5 h; b) toluene, 1108C,
1 h; c) Et3N, CH2Cl2, RT, 1 h.
beled hUT2R binding was determined experimentally by the pres-
ence of 1 mm unlabeled U-II. After incubation at room temperature
for 90 min, the incubation mixtures were filtered and washed in
the automatic vacuum filtration system for filter plates. The europi-
um was dissociated from the bound ligand by the addition of
150 mL of DELFIA enhancement solution (PerkinElmer Oy) and incu-
bated for 15 min with shaking. Dissociated europium created
highly fluorescent complexes, which were measured in a multilabel
counter with a TRF option (Envision, PerkinElmer). The counter set-
ting was as follows: excitation at 340 nm, 200 ms delay, and emis-
sion collection for 400 ms at 615 nm. The extent of antagonism was
expressed as % displacement. The IC50 value was characterized in
an eight-dose response study to generate the compound concen-
tration required to yield 50% displacement.[27]
scribed for ACT-3 gave ACT-1 as a yellow solid (31 mg, 26%):
1H NMR (500 MHz, CD3OD): d=8.41 (d, J=4.5 Hz, 1H), 8.02 (d, J=
4.5 Hz, 1H), 7.90 (d, J=5.6 Hz, 1H), 7.54 (d, J=7.5 Hz, 1H), 7.45 (d,
J=5.6 Hz, 1H), 7.38~7.39 (m, 1H), 7.23 (t, J=7.0 Hz, 1H), 7.09 (t,
J=7.0 Hz, 1H), 3.87–3.50 (m, 2H), 3.00 (s, 2H), 2.71–2.74 (m, 2H),
2.51–2.55 (m, 2H), 2.36–2.43 (m, 2H), 1.78–1.87 (m, 2H), 1.57–
1.61 ppm (m, 2H).
1-[2-(4-Benzyl-4-hydroxypiperidin-1-yl)ethyl]-3-(5-methylthieno-
[3,2-b]pyridin-7-yl)urea (ACT-2): Reaction of 5-methylthieno[3,2-
b]pyridine-7-carbonyl azide (77 mg, 0.35 mmol) as described for
1
ACT-3 gave ACT-2 as a yellow solid (12 mg, 8%): H NMR (500 MHz,
CD3OD): d=7.91 (s, 1H), 7.90 (d, J=5.6 Hz, 1H), 7.41 (d, J=5.6 Hz,
1H), 7.30 (m, 2H), 7.25 (m, 3H), 3.60 (t, J=6.0 Hz, 2H), 3.36 (m,
2H), 3.12 (t, J=6.0 Hz, 2H), 3.09 (t, J=11.8 Hz, 2H), 2.85 (s, 2H),
2.62 (s, 3H), 2.04 (m, 2H), 1.91 ppm (m, 2H).
1-(2-(4-Benzyl-4-hydroxypiperidin-1-yl)ethyl)-3-(2-methylthieno-
[3,2-b]pyridin-7-yl)urea (ACT-4): Reaction of 2-methylthieno[3,2-
b]pyridine-7-carbonyl azide (122 mg, 0.56 mmol) as described for
ACT-3 gave ACT-4 as a yellow solid (129 mg, 54%): 1H NMR
(500 MHz, CD3OD): d=8.31 (d, J=5.7 Hz, 1H), 7.92 (d, J=5.7 Hz,
1H), 7.27 (m, 2H), 7.21 (m, 3H), 7.12 (s, 1H), 3.47 (t, J=6.3 Hz, 2H),
2.94 (m, 2H), 2.78 (s, 2H), 2.76 (t, J=6.3 Hz, 2H), 2.69 (m, 2H), 2.62
(s, 3H), 1.77 (m, 2H), 1.61 ppm (m, 2H); 13C NMR (125 MHz, DMSO):
d=16.6, 35.2, 36.3, 49.2 (2C), 55.3, 57.2, 68.0, 106.6, 120.8, 123.7,
126.3, 128.1 (2C), 131.0 (2C), 138.1, 142.0, 144.0, 148.2, 155.0, 158.1;
HRMS (EI): calcd for C23H28N4O2S [M]+: 424.1933, found: 424.1928.
Computational methods
GEnSeMBLE predictions of apo-protein structures for hUT2R: We
have been developing methods for predicting the 3D structures of
GPCRs from sequence information without use of homology since
1998,[28] with significant improvements in 2002 and 2004.[29,30] The
MembStruk method[30] was validated by a series of applications to
human D2 dopamine,[31] human b2 adrenergic,[32] human M1 acetyl-
choline,[33] human CCR1 cytokine,[34] mouse MrgC11,[35] and human
DP prostaglandin receptor.[36] However, the more recent availability
of X-ray structures has allowed us to improve on these techniques
significantly, leading to the GEnSeMBLE method.[37]
1-(2-(4-Benzyl-4-hydroxypiperidin-1-yl)ethyl)-3-(3-bromothieno-
[3,2-b]pyridin-7-yl)urea (ACT-5): Reaction of 3-bromothieno[3,2-
b]pyridine-7-carbonyl azide (35 mg, 0.12 mmol) as described for
The GEnSeMBLE method samples a complete set (trillions) of possi-
ble packings (rotations and tilts of the helices), all of which are
scored using a fast approximate (BiHelix) analysis from which we
extract smaller sets (1000s), which are scored more accurately to
select the ensemble of ~20 low lying structures expected to play
a role in binding ligands and activating the GPCR. The GEnSeMBLE
prediction methodology has been described elsewhere,[37] but it is
summarize here in brief:
ACT-3 gave ACT-5 as
a
yellow solid (12 mg, 20%): 1H NMR
(500 MHz, CD3OD): d=8.48 (d, J=5.5 Hz, 1H), 8.05 (d, J=5.5 Hz,
1H), 8.00 (s, 1H), 7.29–7.22 (m, 2H), 7.21–7.17 (m, 3H), 3.48 (t, J=
6.0 Hz, 2H), 3.02 (d, J=9.7 Hz, 2H), 2.83 (t, J=6.0 Hz, 2H), 2.77 (s,
2H), 2.80–2.71 (m, 2H), 1.81–1.74 (m, 2H), 1.61 ppm (d, J=13.5 Hz,
2H).
1) PredicTM: We use PredicTM to perform a multiple sequence
alignment of 1726 GPCR sequences using the MAFFT[38] program
and to predict the TM domains for the target GPCR. These TM do-
ꢀ 2014 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim
ChemMedChem 2014, 9, 1732 – 1743 1740

CHEMMEDCHEM
FULL PAPERS
www.chemmedchem.org
mains were extended by capping rules, and the secondary struc-
ture prediction using PORTER[39] and SSPro,[40] as shown in Fig-
ure S2 of the Supporting Information.
GPCRs, was constructed by homology. Then keeping the 7-TM
region fixed, we minimized the loops (up to 1000 steps or down to
0.5 RMS force) followed by quench annealing between 50 K and
600 K each for 0.1 ps for 10 cycles. The lowest energy structures,
which were at eight cycles for hUT2R_hCXCR4 and at seven cycles
for hUT2R_mOPRD, were selected for further docking. Both the an-
nealed structures had 68 H-bonds (increased from 30 and 45 H-
bonds at the beginning of hUT2R hCXCR4 and hUT2R mOPRD, re-
spectively).
2) Helix shape: We used two methods to predict the shape of the
TM domains: a) OptHelix uses molecular dynamics to predict the
structure for each TM region, which is usually helical except for
kinks that may be caused by proline residues; b) homologize heli-
ces is used rather than OptHelix, particularly when closely related
X-ray structures are available. For hUT2R, we considered the follow-
ing nine templates, the X-ray structures for human nociceptin/or-
phanin FQ receptor (hNOPR),[11] mouse delta opioid receptor
(mOPRD),[12] mouse mu opioid receptor (mOPRM),[13] human kappa
opioid receptor (hOPRK),[14] human D3 dopamine receptor
(hD3DR),[15] human adenosine A2A receptor (hAA2AR),[17] human b2
adrenergic receptor (hb2AR),[18] human chemokine CXCR4
(hCXCR4),[20] and bovine rhodopsin (bRho).[24] The sequence identi-
ties of TM regions for these systems are 36.0% for hNOPR, 34.5%
for mOPRD, 34.4% for mOPRM, 32.8% for hOPRK, 31.3% for
hD3DR, 27.9% for hA2AAR, 27.3% for hb2AR, 27.1% for hCXCR4, and
21.8% for bRho. All nine templates were selected based on high
sequence identity and diversity of GPCR family.
DarwinDock: Our earlier studies used the HierDock[32a] and
MSCDock,[44] which have now been replaced by DarwinDock. For
each ligand conformation, DarwinDock iteratively generates
~50000 poses into the putative binding regions of the bulky-resi-
due-alanized protein. This is followed by the energy scoring of
family heads to select the top 10% ordered by total energy. The
top 100 conformations are chosen for further optimization. For
each of these, we dealanize the protein side chains (using
SCREAM) to find the optimum side chains for each of the best 100
poses. Then, we neutralize the protein and ligand by transferring
protons appropriately within salt bridges and protonating or de-
protonating exterior ligands, followed by further full geometry
minimization. We consider that use of these neutral residue charg-
es improves the accuracy for comparing different docked struc-
tures. The result is that small changes in geometries of charged li-
gands far from the binding site can lead to large differential bind-
ing energies of 10–30 kcalmolÀ1. We find that neutralizing these
exposed residues removes the sensitivity to details of the distances
of charged residues (and counter ions) remote from the active site.
This neutralization leads to differential binding energies that are
dominated by the local cavity interactions and leads to much
smaller solvation energies.
3) BiHelix optimization of helix rotations: Since the sequence
identity in TM region (22–36%) was high, we sampled Æ608 rota-
tion about each TM axis (h angle) by 158 increments, leading to
a total of (9)7 ~4.7 million packings. To make it practical to evaluate
the energies for all 4.7 million packings, we use the BiHelix method
in which we first approximate the interactions within the 7-helix
bundle by partitioning the interactions into 12 sets of pairwise
helix interactions, which are added together (mean field approxi-
mation). For each of the 972 (9ꢂ9ꢂ12) pairwise interactions, we
use SCREAM[41] to optimize the side chains. The BiHelix mean field
energies for all packings are used to select the best 2000. Then, we
build the full helix bundle for each of these 2000, optimize the
side chains for each using SCREAM, and neutralize the charged res-
idues for more accurate energy scoring. The Dreiding D3 force
field (D3FF)[42] was used throughout wherever energies were evalu-
ated.
This same procedure was followed for each of seven ligand confor-
mations generated as follows. Starting from the minimized struc-
ture of the ACT-058362 (ACT), SB-706375 (SB), GSK-1440115 (GSK1),
and GSK-1562590 (GSK2), we performed the conformational search
of mixed torsional/low-mode sampling (1000 steps, 100 steps per
rotatable bond, 5 kcalmolÀ1 of energy window, 0.5 ꢁ of maximum
atom deviation cutoff) using the Maestro software.[45] The low-
energy conformations were reminimized by D3FF and clustered by
2.5 and 1.5 ꢁ of RMSD in two steps. For docking, the lowest 7–10
ligand conformations (within 10 kcalmolÀ1 of the best energy)
were selected out of 227 for ACT, 360 for SB, and 158 for GSK2.
The final docked structure with the best binding energy from all
ligand conformations was selected.
4) SuperBiHelix simulations optimization of helix tilts and rota-
tions: Starting with the X-ray tilt angles the BiHelix step always
identifies the experimental rotation angles correctly in the BiHelix
ensemble,[37] but even for closely related GPCRs, optimization of
only the rotation angles does not lead to the X-ray structure for
other GPCRs. To do this, we must optimize the tilt angles: q (tilting
away from the z-axis) and f (the azimuthal angle of the tilting
from the xy-plane) angle[43] simultaneous with the rotation angles
(h). Since the best rotation angles from BiHelix match the template
mOPRD, we considered it sufficient to sample Æ108 for q tilt angle
with simultaneous sampling by Æ158 and Æ308 for both f and h
angles, leading to a total of (3ꢂ5ꢂ5)7 ~10 trillion combinations.
For cases in which the optimum angle is at the boundary of these
variations, we do a second round starting from the new best con-
formation until it is consistent.
DarwinDock has been validated for a number of X-ray co-crystals
including three crystal structures of ligand/GPCR complexes: hb2AR
(0.4 ꢁ RMSD),[18] hA2AAR (0.8 ꢁ RMSD),[17] and turkey b1-adrenergic
receptor (0.1 ꢁ RMSD).[16] This shows that we can accurately identi-
fy ligand binding sites in proteins, which can then be used to opti-
mize the ligands with desirable properties.
Virtual screening: We used LigandScout 3.0[26] to identify new
leads for hUT2R ligands, based on the 3D pharmacophore form de-
rived from our predictions on ACT1. For the pharmacophore, we
selected: 1) one pi–pi stacking at F1313.33; 2) three hydrophobic in-
teractions of the methyl group at V184 and L2125.39, the phenyl
ring at I54 and L126, and the phenyl ring from fused ring with
F2165.43, I2205.37, and P271; 3) one H-bond donor at D1303.32; and
4) one positive N at D1303.32 with 15 exclusion volume.
When we pack the helices into a 7-helix bundle, we choose the hy-
drophobic center (HPC) for each TM from the PredicTM analysis to
place at z=0, for the common mid-membrane plane. SuperBiHelix
can also be used to optimize translation along the TM axis simulta-
neously with optimizing rotations.
5) Loop generation: The loops and the N/C terminus including
helix 8 for a new structure are modeled through homology with
the each template structure. The disulfide bridge between
C1233.25and C199 for hUT2R, which is conserved among class A
ꢀ 2014 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim
ChemMedChem 2014, 9, 1732 – 1743 1741

CHEMMEDCHEM
FULL PAPERS
www.chemmedchem.org
[10] D. J. Behm, N. V. Aiyar, A. R. Olzinski, J. J. McAtee, M. A. Hilfiker, J. W.
Dodson, S. E. Dowdell, G. Z. Wang, K. B. Goodman, C. A. Sehon, M. R.
Harpel, R. N. Willette, M. J. Neeb, C. A. Leach, S. A. Douglas, Br. J. Phar-
macol. 2010, 161, 207–228.
[11] A. A. Thompson, W. Liu, E. Chun, V. Katritch, H. Wu, E. Vardy, X. P. Huang,
C. Trapella, R. Guerrini, G. Calo, B. L. Roth, V. Cherezov, R. C. Stevens,
Nature 2012, 485, 395–399.
[12] S. Granier, A. Manglik, A. C. Kruse, T. S. Kobilka, F. S. Thian, W. I. Weis,
B. K. Kobilka, Nature 2012, 485, 400–404.
[13] A. Manglik, A. C. Kruse, T. S. Kobilka, F. S. Thian, J. M. Mathiesen, R. K. Su-
nahara, L. Pardo, W. I. Weis, B. K. Kobilka, S. Granier, Nature 2012, 485,
321–326.
[14] H. Wu, D. Wacker, M. Mileni, V. Katritch, G. Han, E. Vardy, W. Liu, A. A.
Thompson, X.-P. Huang, F. I. Carroll, S. W. Mascarella, R. B. Westkaemper,
P. D. Mosier, B. L. Roth, V. Cherezov, R. C. Stevens, Nature 2012, 485,
327–332.
[15] E. Y. Chien, W. Liu, Q. Zhao, V. Katritch, G. W. Han, M. A. Hanson, L. Shi,
A. H. Newman, J. A. Javitch, V. Cherezov, R. C. Stevens, Science 2010,
330, 1091–1095.
[16] T. Warne, M. J. Serrano-Vega, J. G. Baker, R. Moukhametzianov, P. C. Ed-
wards, R. Henderson, A. G. W. Leslie, C. G. Tate, G. F. X. Schertler, Nature
2008, 454, 486–491.
[17] V. P. Jaakola, M. T. Griffith, M. A. Hanson, V. Cherezov, E. Y. Chien, J. R.
Lane, A. P. Ijzerman, R. C. Stevens, Science 2008, 322, 1211–1217.
[18] V. Cherezov, D. M. Rosenbaum, M. A. Hanson, S. G. F. Rasmussen, F. S.
Thian, T. S. Kobilka, H.-J. Choi, P. Kuhn, W. I. Weis, B. K. Kobilka, R. C. Ste-
vens, Science 2007, 318, 1258–1265.
Supporting Information
The following are available in the Supporting Information: details
of GEnSeMBLE prediction of the 3D structure of hUT2R (Appendix A
in the Supporting Information); details of predicting the binding
site of ACT-058362 and SB-706375 to hUT2R (Appendix B in the
Supporting Information); hits from virtual screening result of ACT-
058362 with hUT2R from the rhodopsin template (Figure S1); TM
region and the secondary structure prediction of hUT2R (Figure S2);
the hydropathy prediction from PredicTM for hUT2R (Figure S3);
the final TM regions for hUT2R from the PredicTM multiple align-
ment method with the X-ray structures for nine templates (Fig-
ure S4); the predicted structure of hUT2R with three templates (Fig-
ure S5); the H-bonding networks of hUT2R with three templates
(Figure S6); the binding site of ACT-058362 at hUT2R with three
templates (Figure S7); the binding site of SB-706375 at hUT2R with
the three templates (Figure S8), The post virtual screening result of
hUT2R from the template of frog rhodopsin (hUT2R-fRho) bound
with ACT-058362 (Table S1), Sequence identities of the hUT2R with
all predicted structures as well as 14 currently available X-ray struc-
tures of GPCRs (Table S2), Top 20 predicted structures for the
hUT2R selected from the total 24000 of the BiHelix analysis for the
top 9 templates (Table S3), top 10 predicted structures of the
hUT2R based on three templates: hb2AR, hCXCR4, mOPRD
(Table S4); top 20 predicted structures of the hUT2R from the Su-
perBiHelix analysis of hb2AR and hCXCR4 (Table S5); the x, y, z, h, q,
f, and RMSD (ꢁ) comparison of hUT2R predicted with the hUT2R-
fRho and hCXCR4 templates compared with that from mOPRD
(Table S6).
[19] M. A. Hanson, C. B. Roth, E. Jo, M. T. Griffith, F. L. Scott, G. Reinhart, H.
Desale, B. Clemons, S. M. Cahalan, S. C. Schuerer, M. G. Sanna, G. W.
Han, P. Kuhn, H. Rosen, R. C. Stevens, Science 2012, 335, 851–855.
[20] B. Wu, E. Y. Chien, C. D. Mol, G. Fenalti, W. Liu, V. Katritch, R. Abagyan, A.
Brooun, P. Wells, F. C. Bi, D. J. Hamel, P. Kuhn, T. M. Handel, V. Cherezov,
R. C. Stevens, Science 2010, 330, 1066–1071.
[21] T. Shimamura, M. Shiroishi, S. Weyand, H. Tsujimoto, G. Winter, V. Ka-
tritch, R. Abagyan, V. Cherezov, W. Liu, G. W. Han, T. Kobayashi, R. C. Ste-
vens, S. Iwata, Nature 2011, 475, 65–70.
Acknowledgements
Funding for this project was provided by the Korea Research In-
stitute of Chemical Technology (South Korea).
[22] A. C. Kruse, J. Hu, A. C. Pan, D. H. Arlow, D. M. Rosenbaum, E. Rose-
mond, H. F. Green, T. Liu, P. S. Chae, R. O. Dror, D. E. Shaw, W. I. Weis, J.
Wess, B. K. Kobilka, Nature 2012, 482, 552–556.
[23] K. Haga, A. C. Kruse, H. Asada, T. Yurugi-Kobayashi, M. Shiroishi, C.
Zhang, W. I. Weis, T. Okada, B. K. Kobilka, T. Haga, T. Kobayashi, Nature
2012, 482, 547–551.
Keywords: computational chemistry · docking · GPCRs · R-
group screening · urotensin II · virtual screening
[24] a) K. Palczewski, T. Kumasaka, T. Hori, C. A. Behnke, H. Motoshima, B. A.
Fox, I. Le Trong, D. C. Teller, R. E. Stenkamp, M. Yamamoto, M. Miyano,
Science 2000, 289, 739–745; b) D. C. Teller, T. Okada, C. A. Behnke, K.
Palczewski, R. E. Stenkamp, Biochemistry 2001, 40, 7761–7772; c) T.
Okada, Y. Fujiyoshi, M. Silow, J. Navarro, E. M. Landau, Y. Shichida, Proc.
Natl. Acad. Sci. USA 2002, 99, 5982–5987; d) J. Li, P. C. Edwards, M. Burg-
hammer, C. Villa, G. F. X. Schertler, J. Mol. Biol. 2004, 343, 1409–1438;
e) T. Okada, M. Sugihara, A. N. Bondar, M. Elstner, P. Entel, V. Buss, J. Mol.
Biol. 2004, 342, 571–583.
[25] S.-K. Kim, Y. Li, R. Abrol, J. Heo, W. A. Goddard III, J. Chem. Inf. Model.
2011, 51, 420–433.
[26] a) G. Wolber, A. A. Dornhofer, T. Langer, J. Comput.-Aided Mol. Des. 2007,
20, 773–788; b) C. R. Underwood, P. Garibay, L. B. Knudsen, S. Hastrup,
G. H. Peters, R. Rudolph, S. Reedtz-Runge, J. Biol. Chem. 2010, 285, 723–
730.
[27] K. S. Oh, S. Lee, B. H. Lee, Assay Drug Dev. Technol. 2011, 9, 514–521.
[28] W. B. Floriano, N. Vaidehi, W. A. Goddard III, M. S. Singer, G. M. Shepherd,
Proc. Natl. Acad. Sci. USA 2000, 97, 10712–10716.
[29] R. J. Trabanino, S. E. Hall, N. Vaidehi, W. B. Floriano, V. Kam, W. A. God-
dard III, Biophys. J. 2004, 86, 1904–1921.
[30] N. Vaidehi, W. B. Floriano, R. Trabanino, S. E. Hall, P. Freddolino, E. J.
Choi, W. A. Goddard III, Proc. Natl. Acad. Sci. USA 2002, 99, 12622–
12627.
[1] R. S. Ames, H. M. Sarau, J. K. Chambers, R. N. Willette, R. V. Aiyar, A. M.
Romanic, C. S. Louden, J. J. Foley, C. F. Sauermelch, R. W. Coatney, Z. Ao,
J. Disa, S. D. Holmes, J. M. Stadel, J. D. Martin, W.-S. Liu, G. I. Glover, S.
Wilson, D. E. McNutty, C. E. Ellis, N. A. Eishourbagy, U. Shabon, J. J. Trill,
D. V. P. Hay, E. H. Ohlstein, D. J. Bergsma, S. A. Douglas, Nature 1999,
401, 282–286.
[2] A. Carotenuto, P. Grieco, P. Campiglia, E. Novellino, P. Rovero, J. Med.
Chem. 2004, 47, 1652–1661.
[3] S. A. Douglas, Curr. Opin. Pharmacol. 2003, 3, 159–167.
[4] S. Flohr, M. Kurtz, E. Kostenis, A. Brkovich, A. Fournier, T. Klabunde, J.
Med. Chem. 2002, 45, 1799–1805.
[5] A. Carotenuto, P. Grieco, P. Rovero, E. Novellino, Curr. Med. Chem. 2006,
13, 267–275.
[6] P. Grieco, A. Carotenuto, P. Campiglia, L. Marinelli, T. Lama, R. Patacchini,
P. Santicioli, C. A. Maggi, P. Rovero, E. Novellino, J. Med. Chem. 2005, 48,
7290–7297.
[7] S. K. Kim, Y. Li, C. Park, R. Abrol, W. A. Goddard III, ChemMedChem 2010,
5, 1594–1608.
[8] M. Clozel, C. Binkert, M. Birker-Robaczewska, C. Boukhadra, S.-S. Ding,
W. Fischli, P. Hess, B. Mathys, K. Morrison, C. Mꢃller, C. Mꢃller, O. Nayler,
C. Qiu, M. Rey, M. W. Scherz, J. Velker, T. Weller, J.-F. Xi, P. Ziltener, J.
Pharmacol. Exp. Ther. 2004, 311, 204–212.
[31] Y. Kalani, N. Vaidehi, S. E. Hall, W. B. Floriano, R. J. Trabanino, P. L. Fred-
dolino, V. Kam, W. A. Goddard III, Proc. Natl. Acad. Sci. USA 2004, 101,
3815–3820.
[32] a) W. B. Floriano, N. Vaidehi, G. Zamanakos, W. A. Goddard III, J. Med.
Chem. 2004, 47, 56–71; b) P. Spijker, N. Vaidehi, P. L. Freddolino, P. A.
[9] S. A. Douglas, D. J. Behm, N. V. Aiyar, D. Naselsky, J. Disa, D. P. Brooks,
E. H. Ohlstein, J. G. Gleason, H. M. Sarau, J. J. Foley, P. T. Buckley, D. B.
Schmidt, W. E. Wixted, K. Widdowson, G. Riley, J. Jin, T. F. Gallagher, S. J.
Schmidt, L. Ridgers, L. T. Chirstmann, R. M. Keenan, S. D. Knight, D.
Dhanak, Br. J. Pharmacol. 2005, 145, 620–635.
ꢀ 2014 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim
ChemMedChem 2014, 9, 1732 – 1743 1742

CHEMMEDCHEM
FULL PAPERS
www.chemmedchem.org
Hilbers, W. A. Goddard III, Proc. Natl. Acad. Sci. USA 2006, 103, 4882–
4887.
[33] J. Y. Peng, N. Vaidehi, S. E. Hall, W. A. Goddard III, ChemMedChem 2006,
1, 878–890.
[40] G. Pollastri, D. Przybylski, P. Rost, B. Baldi, Proteins 2002, 47, 228–235.
[41] V. W. T. Kam, W. A. Goddard III, J. Chem. Theory Comput. 2008, 4, 2160–
2169.
[42] S. L. Mayo, B. D. Olafson, W. A. Goddard III, J. Phys. Chem. 1990, 94,
8897–8909.
[43] R. Abrol, A. R. Griffith, J. K. Bray, W. A. Goddard III, Methods Mol. Biol.
2012, 914, 237–254.
[44] A. E. Cho, J. A. Wendel, N. Vaidehi, P. M. Kekenes-Huskey, W. B. Floriano,
P. K. Maiti, W. A. Goddard III, J. Comput. Chem. 2005, 26, 48–71.
[45] K. S. Watts, P. Dalal, R. B. Murphy, W. Sherman, R. A. Friesner, J. C. Shelley,
J. Chem. Inf. Model. 2010, 50, 534–546.
[34] N. Vaidehi, S. Schlyer, R. J. Trabanino, W. B. Floriano, R. Abrol, S. Sharma,
M. Kochanny, S. Koovakat, L. Dunning, M. Liang, J. M. Fox, F. L. de Men-
donca, J. E. Pease, W. A. Goddard III, J. Biol. Chem. 2006, 281, 27613–
27620.
[35] a) J. Heo, S.-K. Han, N. Vaidehi, J. Wendel, P. Kekenes-Huskey, W. A. God-
dard III, ChemBioChem 2007, 8, 1527–1539; b) J. Heo, N. Vaidehi, J.
Wendel, W. A. Goddard III, J. Mol. Graphics Modell. 2007, 26, 800–812.
[36] Y. Li, F. Zhu, N. Vaidehi, W. A. Goddard III, F. Sheinerman, S. Reiling, I.
Morize, L. Mu, K. Harris, A. Ardati, A. Laoui, J. Am. Chem. Soc. 2007, 129,
10720–10731.
[46] V. M. Unger, P. A. Hargrave, J. M. Baldwin, G. F. Schertler, Nature 1997,
389, 203–206.
[37] R. Abrol, J. K. Bray, W. A. Goddard III, Proteins 2012, 80, 505–518.
[38] K. Katoh, K. Kuma, H. Toh, T. Miyata, Nucleic Acid Res. 2005, 33, 511–
518.
Received: March 26, 2014
[39] G. Pollastri, A. McLysaght, Bioinformatics 2005, 21, 1719–1720.
Published online on July 2, 2014
ꢀ 2014 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim
ChemMedChem 2014, 9, 1732 – 1743 1743