Structure-based drug design of chromone antagonists of the adenosine A 2A receptor

Article abstract of DOI:10.1039/c3md00338h

The structure-guided optimisation of a hit series of chromone derivatives, previously identified using virtual screening of homology models of the adenosine A_2A receptor, has led to the discovery of potent, selective and ligand efficient antagonists. Lipophilic hotspots and calculated water networks were modelled within the receptor binding site to facilitate rational ligand design. This journal is the Partner Organisations 2014.

Full text of DOI:10.1039/c3md00338h

MedChemComm
View Article Online
View Journal
CONCISE ARTICLE
Structure-based drug design of chromone
antagonists of the adenosine A2A receptor†
Cite this: DOI: 10.1039/c3md00338h
*
Stephen P. Andrews,‡ Jonathan S. Mason, Edward Hurrell and Miles Congreve
Received 4th November 2013
Accepted 30th December 2013
The structure-guided optimisation of a hit series of chromone derivatives, previously identified using virtual
screening of homology models of the adenosine A2A receptor, has led to the discovery of potent, selective
and ligand efficient antagonists. Lipophilic hotspots and calculated water networks were modelled within
the receptor binding site to facilitate rational ligand design.
DOI: 10.1039/c3md00338h
www.rsc.org/medchemcomm
High resolution crystal structures have been published for
Initial analysis of 1 with the A2A receptor homology models
twenty G protein-coupled receptors (GPCRs), including 17 resulted in two putative binding modes (A and B; Fig. 1), from
members of the rhodopsin family, a frizzled receptor and two Glide13 docking studies. In order to improve the activity of 1 by
members of the secretin family.1–4 These advances in structural rational design it was rst necessary to validate binding mode A
biology have given enormous insight into the binding sites of which, aer further modelling and careful analysis, appeared to
this superfamily of receptors, facilitating structure-based drug be the most plausible. To this end, small sets of analogues were
design and providing templates for the construction of high designed iteratively and tested in a competitive radioligand
condence homology models.5–7
binding assay14 with hA2A receptor and [3H]-ZM241385.
The adenosine A2A receptor is a member of the rhodopsin
Initially, the negatively charged carboxylate functionality was
family of GPCRs (class A). Many ligands are known for this removed from 1 as it had been modelled to sit unfavourably in a
receptor, including a number of antagonists which have been lipophilic region lined by Leu167ECL2, Ile662.64 and Met2707.35
investigated clinically for the treatment of central nervous (superscript numbers refer to Ballesteros–Weinstein15 numbering).
system disorders, particularly Parkinson's disease.8 For As predicted, replacement with a positively charged group such as
example, preladenant reached phase 3 clinical trials but was an aliphatic amine (2) did not invoke an increase in affinity,
discontinued owing to lack of efficacy versus placebo and, in whereas conversion to a neutral group such as an ester (3) gave rise
2013, istradefylline was launched in Japan for the treatment of to a 20-fold improvement in affinity (Table 1).
Parkinson's disease. Currently, there are no other marketed A2A
Modications with simple alkyl groups led to improvements
receptor antagonists and the exploration of further chemotypes in ligand efficiency16 (4; LE ¼ 0.42), and larger chains were also
is warranted.
well-tolerated (5; pKi ¼ 7.6). A simple acetate group at position
Compound 1 was previously reported as a 2.0 mM hit R1 was found to maintain a good level of affinity (6) and the
following the virtual screening of homology models of the A2A
receptor which were built from the crystal structure of the b1
adrenergic receptor.9 Herein, we describe the efficient optimi-
sation of a series of chromone ligands with structure-based
approaches driven by molecular modelling and biophysical
techniques.
Compound 1 was identied prior to solving the X-ray crystal
structure of the A2A receptor; however, there are now multiple
3D structures available for this receptor, which has been solved
in both its active and inactive forms,10 facilitated by either a
fusion protein11 or thermostabilisation technology.12
Heptares Therapeutics Ltd., Biopark, Broadwater Road, Welwyn Garden City, AL7 3AX,
UK. E-mail: steve.andrews@heptares.com
† Electronic supplementary information (ESI) available. See DOI:
10.1039/c3md00338h
Fig. 1 Putative binding modes of compound 1. In mode A (green
carbon) the thiazole N hydrogen bonds to Asn253 and in mode B (purple
carbon) the chromone carbonyl group hydrogen bonds to Asn253.
‡ ©Heptares Therapeutics 2013. The HEPTARES name is a trade mark of Heptares
Therapeutics Ltd.
This journal is © The Royal Society of Chemistry 2014
Med. Chem. Commun.

View Article Online
MedChemComm
Concise Article
Table 1 Initial hit exploration
Table 2 Investigation of substituents at vectors R2 and R3
R1
pKi
LE
R2
R3
pKi
LE
1
2
3
4
5
6
7
–CH2CO2H
–(CH2)2NMe2
–CH2CO2Et
–methylcyclopropyl
–(CH2)3C^CH
–C(]O)Me
–H
5.7
5.8
7.0
7.1
7.6
7.5
6.5
0.34
0.33
0.38 10
0.42 11
0.43 12
0.46
8
9
H
Et
n-Pr
n-pentyl
H
H
H
H
H
5.7
7.1
7.6
7.7
5.1
0.44
0.49
0.50
0.46
0.37
Me
0.47
areas outside of the hotspot and shows only a small increase in
potency when compared to 10. Compounds 9 and 10 were
therefore progressed in preference to 11 as they showed the best
balance of LE and lipophilic ligand efficiency18 (LLE ¼ 3.6, 3.7
and 2.8, respectively).
unsubstituted hydroxyl derivative (7) showed the highest LE
(0.47). This phenol group was kept constant in the following
round of design where the effects of substituents at other
positions on the chromone scaffold were rationally explored.
During a druggability analysis,6 a lipophilic hotspot was
identied with a GRID aromatic CH (C1]) probe,17 in a pocket
lined by Ile662.64 and Tyr2717.36 (Fig. 2). This led to the design of
compounds with lipophilic groups at the vector dened by R2
(C-6 of the chromone core), with the minimum chain length of
two atoms required to reach the hotspot (Table 2). When
compared to the R2-unsubstituted derivative (8), compounds 9,
10 and 11 all showed very good affinities for the receptor.
The size of the lipophilic hotspot indicated that a propyl
group would most efficiently ll this pocket and this was indeed
found to be the case, with 10 showing the highest LE. A later
analysis using WaterFLAP and WaterMap in conjunction with
A2A receptor crystal structures showed that moderately high
energy or ‘unhappy’ waters6 are in this region and are displaced
by 10 (see ESI†). The larger n-pentyl derivative (11) explores
The proposed binding mode (A) was further conrmed by
testing compound 12 with a methyl substituent at position R3.
Compound 12 was predicted to impose a steric clash on
Asn2536.55 in binding mode A (Fig. 2; the carbon at position C-2
of the chromone is almost at the GRID C3 surface), and was
found to be four times less active than 8. By contrast, in binding
mode B, the methyl group of 12 would face a large pocket and a
minimal change of activity would be expected (see Fig. 1).
Selected analogues from this series were screened for
stability in rat liver microsomes (RLM) and selectivity vs. the
adenosine A1 receptor. They were found to have half-lives in
RLM of 14–18 minutes (see ESI†) and, in several cases, much
higher affinity was observed for the A2A receptor than the A1
receptor. For example, compound 6 was 13-fold selective for the
A2A receptor and compounds 4, 9 and 11 were all >100-fold
selective for the A2A receptor (see ESI†). In order to understand
and rationalise these observations, homology models of the A1
receptor were constructed from crystal structures of the A2A
receptor.
Compound 4 contains a bulky and lipophilic methyl-
cyclopropyl group at vector R1 which, in the A2A receptor binding
site, sits in a lipophilic sub-pocket with a relatively large cavity.
This pocket was found to be smaller and more polar in the
corresponding A1 site, thus disfavouring the binding of 4 (Fig. 3).
Compound 11 contains a lipophilic group at the vector R2,
and this group also sits in a sub-pocket which is smaller and
more polar in the A1 binding site. In A2A, the polar hydroxyl of
Ser672.65 was found to point away from the sub-pocket (Fig. 2),
whereas in A1, the corresponding residue (Asn702.65) was
modelled with its polar side chain facing into the binding site,
H-bonding to the backbone of Ala662.61, or to Gln91.32 and
Tyr2717.36
.
To further corroborate these ndings, compound 8 was
found to have a two-fold higher affinity for the adenosine A1
receptor, as it does not have substituents at either vector R1 or
R2 to access these ‘selectivity pockets’.
Fig. 2 Compound 10 in binding mode A; GRID C3 surface (1.0 kcal
molꢀ1; grey) and C1] lipophilic hotpspots (ꢀ2.8 kcal molꢀ1; yellow) for
A2A receptor; and WaterMap waters calculated for the pseudo-apo
protein from the ligand complex PDB:3UZC (coloured spheres).
Med. Chem. Commun.
This journal is © The Royal Society of Chemistry 2014

View Article Online
Concise Article
MedChemComm
interacted with Ile66ECL2 and Tyr2717.36, which further supports
binding mode A, in which the chromone C-6 substituent is
projected towards this lipophilic sub-pocket of the binding site.9
The nal area of the molecules to be investigated was the
4-methylthiazole group. The N atom of this heterocycle provides
the core H-bond found for A2A receptor ligands to Asn2536.55
.
Furthermore, the methyl group at the 4 position of this
heterocycle makes a crucial contribution to the binding affinity
of this series of molecules, with a severe (33-fold) loss of activity
upon its removal (16 / 17; Table 3). This is a clear example of a
‘magic methyl’ effect, and is of similar magnitude to a recent
example published with opioid receptor antagonists,20 but here
there is a clear structural understanding of its origin. The
methyl group displaces an ‘unhappy’ water (red sphere in
Fig. 2), identied from a WaterMap calculation of the water
network and energetics for the pseudo-apo X-ray structure from
PDB: 3UZC.6,14,21,22 This water is in a lipophilic hotspot cavity in
the binding site, giving a strong benecial effect; furthermore,
without the thiazole methyl group, this water would be trapped
in an even more unfavourable position (owing to the remaining
apolar CH of the ligand), that would bind less deeply or create a
‘dewetted’ vacuum region.23 This phenomenon was not
observed with ZM241385 which contains a furan ring that sits
more deeply than the thiazole of 14 and is unable to accom-
modate a methyl substituent (Fig. 6).
Fig. 3 Compound 4 modelled in the 1.8 A˚ resolution A2A receptor
crystal structure (PDB: 4EIY), with its GRID C3 surface shown in solid
grey. In this model, His264ECL3 H-bonds to Glu169ECL2 which leaves its
aromatic ring facing the binding site and creates a lipophilic cavity. In
the corresponding A1 receptor model (also based on PDB: 4EIY; GRID
C3 surface shown in lilac mesh), the longer Lys265ECL3 (shown in lilac
stick) moves into the cavity to interact with Glu172ECL2 (Glu169ECL2 in
A2A), thus changing the size and polarity of the pocket. The cyclopropyl
group of 4 is seen to pierce the A1 pocket (see ESI†).
The power of water network energetic analyses to ratio-
nalise SAR and drive ligand design is further illustrated when
changing the position of the sulphur atom within the thiazole
ring. A ten-fold reduction in activity is found between
compounds 14 and 15, yet no changes in favourable interac-
tions or steric clashes are observed with the receptor. Both
ligands are still able to hydrogen bond to Asn2536.55 via their N
atoms. Scoring of the two isomers using the docking program
Glide in both SP and XP modes showed insignicant differ-
ences (approx. 0.1 kcal better or worse, respectively) and no
differences were observed in a rescoring with Hyde24 that
Compound 13 was selected as a representative of the series
and tested for affinity at all of the adenosine receptor sub-types:
A
2A, A1, A2B and A3. It was found to have >100-fold higher affinity
for the A2A receptor than the A1 receptor, and no measurable
binding affinity for the A2B or A3 receptors (Fig. 4).
The combination of the SAR observed in Table 1 and 2 led to
compound 14 which was found to have a good affinity for the A2A
receptor and maintained a high LE (Fig. 5; LE ¼ 0.48, pKi ¼ 8.5).
Subsequent Biophysical Mapping™ (BPM – a technique which
combines site-directed mutagenesis with surface plasmon
resonance screening in order to determine the individual
contributions of binding site residues to the binding of
ligands)19 of 14 suggested that the propyl group of this ligand
Table 3 Investigation of the methylthiazole substituent
R4
R5
pKi
LE
Fig. 4 Compound 13.
15
16
Me
7.5
0.43
H
H
7.3
5.8
0.43
0.36
17
Fig. 5 Compound 14.
This journal is © The Royal Society of Chemistry 2014
Med. Chem. Commun.

View Article Online
MedChemComm
Concise Article
In summary, the structure-guided optimization of a chro-
mone hit series has led to the discovery of potent antagonists of
the A2A receptor with high LE and selectivity. Important aspects
of the series' SAR can be explained by the effect of displacing
waters from lipophilic hotspots (‘unhappy’ waters) or by per-
turbing the calculated water network within the binding site.
This study is an example of how high quality GPCR binding
mode information is starting to signicantly impact on the
discovery of new agents for this important class of receptors. In
particular, the ability to consider the position and energy of
lipophilic hotspots and of calculated water molecules offers
great promise for rational drug design.
Notes and references
1 N. Bertheleme, P. S. Chae, S. Singh, D. Mossakowska,
M. M. Hann, K. J. Smith, J. A. Hubbard, S. J. Dowell and
B. Byrne, Biochim. Biophys. Acta, Gen. Subj., 2013, 1828, 2583.
2 C. Wang, H. Wu, V. Katritch, G. W. Han, X. P. Huang, W. Liu,
F. Y. Siu, B. L. Roth, V. Cherezov and R. C. Stevens, Nature,
2013, 497, 338.
Fig. 6 Compound 14 (green stick), modelled in the 1.8 A˚ resolution
structure of the A2A receptor (PDB: 4EIY). ZM241385 (cyan stick) is
shown to bind very deeply with the furan C5 atom touching the
surface of the binding site (GRID C3 surface shown in solid grey).
3 K. Hollenstein, J. Kean, A. Bortolato, R. K. Cheng, A. S. Dore,
A. Jazayeri, R. M. Cooke, M. Weir and F. H. Marshall, Nature,
2013, 499, 438.
explicitly considers hydrophobic interactions via atom logP
contributions.
4 F. Y. Siu, M. He, G. C. de, G. W. Han, D. Yang, Z. Zhang,
C. Zhou, Q. Xu, D. Wacker, J. S. Joseph, W. Liu, J. Lau,
V. Cherezov, V. Katritch, M. W. Wang and R. C. Stevens,
Nature, 2013, 499, 444.
5 B. Kobilka, Angew. Chem., Int. Ed., 2013, 52, 6380.
6 J. S. Mason, A. Bortolato, M. Congreve and F. H. Marshall,
Trends Pharmacol. Sci., 2012, 33, 249.
7 M. Congreve, C. J. Langmead, J. S. Mason and F. H. Marshall,
J. Med. Chem., 2011, 54, 4283.
8 J. F. Chen, H. K. Eltzschig and B. B. Fredholm, Nat. Rev. Drug
Discovery, 2013, 12, 265.
The signicant difference in potency observed with this
small change of an indirectly-interacting ligand atom can be
rationalized by analysing the computed energies of the waters
in the bottom of the pocket (affected in 15 by the S atom; Fig. 7)
and the le of the pocket (affected in 14 by the S atom). Using
the sum of the energies of the GRID probes for water and C1]
in WaterFLAP, the two waters at the bottom of the pocket were
found to increase in energy by 9 kcal (14 / 15), whereas the ve
waters to the le were only 4.5 kcal lower. The signicant
increase in the total energy of the waters affected by 14 / 15
could thus rationalise the potency drop.
9 C. J. Langmead, S. P. Andrews, M. Congreve, J. C. Errey,
E. Hurrell, F. H. Marshall, J. S. Mason, C. M. Richardson,
N. Robertson, A. Zhukov and M. Weir, J. Med. Chem., 2012,
55, 1904.
10 S. P. Andrews and B. Tehan, Med. Chem. Commun., 2013, 4,
52.
11 E. Chun, A. A. Thompson, W. Liu, C. B. Roth, M. T. Griffith,
V. Katritch, J. Kunken, F. Xu, V. Cherezov, M. A. Hanson and
R. C. Stevens, Structure, 2012, 20, 967.
12 N. Robertson, A. Jazayeri, J. Errey, A. Baig, E. Hurrell,
A. Zhukov, C. J. Langmead, M. Weir and F. H. Marshall,
Neuropharmacology, 2011, 60, 36.
13 T. A. Halgren, R. B. Murphy, R. A. Friesner, H. S. Beard,
L. L. Frye, W. T. Pollard and J. L. Banks, J. Med. Chem.,
2004, 47, 1750.
14 M. Congreve, S. P. Andrews, A. S. Dore, K. Hollenstein,
E. Hurrell, C. J. Langmead, J. S. Mason, I. W. Ng, B. Tehan,
A. Zhukov, M. Weir and F. H. Marshall, J. Med. Chem.,
2012, 55, 1898.
15 J. A. Ballesteros, H. Weinstein, and C. S. Stuart, in Methods in
Neurosciences, Academic Press, New York, 1995, vol. 25, pp.
366–428.
Fig. 7 Waters in the proximity of the thiazole ring from the computed
water networks for 14 (blue spheres; ligand shown) and 15 (blue
crosses; ligand not shown) from a WaterFLAP analysis, used to
compare water network perturbation.
Med. Chem. Commun.
This journal is © The Royal Society of Chemistry 2014

View Article Online
Concise Article
MedChemComm
16 A. L. Hopkins, C. R. Groom and A. Alex, Drug Discovery Today, 21 A. Bortolato, B. G. Tehan, M. S. Bodnarchuk, J. W. Essex and
2004, 9, 430.
J. S. Mason, J. Chem. Inf. Model., 2013, 53, 1700.
22 C. Higgs, T. Beuming and W. Sherman, ACS Med. Chem. Lett.,
2010, 1, 160.
23 J. S. Mason, A. Bortolato, D. R. Weiss, F. Deorian, B. Tehan
and F. H. Marshall, In Silico Pharmacology, 2013, 1, 1.
17 P. J. Goodford, J. Med. Chem., 1985, 28, 849.
18 P. D. Leeson and B. Springthorpe, Nat. Rev. Drug Discovery,
2007, 6, 881.
19 A. Zhukov, S. P. Andrews, J. C. Errey, N. Robertson, B. Tehan,
J. S. Mason, F. H. Marshall, M. Weir and M. Congreve, J. Med. 24 N. Schneider, S. Hindle, G. Lange, R. Klein, J. Albrecht,
Chem., 2011, 54, 4312.
20 G. Lunn, B. J. Banks, R. Crook, N. Feeder, A. Pettman and
Y. Sabnis, Bioorg. Med. Chem. Lett., 2011, 21, 4608.
H. Briem, K. Beyer, H. Claussen, M. Gastreich,
C. Lemmen and M. Rarey, J. Comput.-Aided Mol. Des.,
2012, 26, 701.
This journal is © The Royal Society of Chemistry 2014
Med. Chem. Commun.