-
VU Research Portal
In silico Medicinal Chemistry
Kooistra, A.J.
2015
document versionPublisher's PDF, also known as Version of
record
Link to publication in VU Research Portal
citation for published version (APA)Kooistra, A. J. (2015). In
silico Medicinal Chemistry: Investigating GPCRs: key regulators of
signal transductionand cell function.
General rightsCopyright and moral rights for the publications
made accessible in the public portal are retained by the authors
and/or other copyright ownersand it is a condition of accessing
publications that users recognise and abide by the legal
requirements associated with these rights.
• Users may download and print one copy of any publication from
the public portal for the purpose of private study or research. •
You may not further distribute the material or use it for any
profit-making activity or commercial gain • You may freely
distribute the URL identifying the publication in the public portal
?
Take down policyIf you believe that this document breaches
copyright please contact us providing details, and we will remove
access to the work immediatelyand investigate your claim.
E-mail address:[email protected]
Download date: 22. Jun. 2021
https://research.vu.nl/en/publications/355d81ed-8d02-4443-b34f-36ef824d1812
-
125
Structure-based prediction of GPCR-ligand function
Chapter 6
-
126
Chapter 6 - Structure-based prediction of GPCR-ligand
function
6 6
-
127
Structure-based prediction of GPCR-ligand function - Chapter
6
6 6
G protein-coupled receptors (GPCRs) are versatile membrane-bound
switch nodes of cell function and constitute arguably one of the
most important classes of proteins that is encoded in the human
genome.373, 379 Next to stimulating (agonism), blocking
(an-tagonism) or reducing constitutive activity (inverse agonism),
GPCRs ligands are able to stim-ulate cell-signaling pathways that
are completely independent of G-proteins.363, 380, 384 This new
concept of GPCR biased-signaling is an intriguing phenomenon in
which different ligands can promote or block multiple
conformational states of the same receptor that are each linked to
a different functional outcome.78, 380, 384 The first human GPCR
crystal structures were only solved in the past seven years93, 176
and have further increased our understanding of the molecular
details of how GPCRs work.73, 117, 176 These spectacular advances
in GPCR structure determination have resulted in 109 crystal
structures410 at this moment, covering 20 class A93, 176, 2 class
B81, 82, 1 class C107 and 1 class F84 GPCRs. Despite these advances
less than 2% of all human GPCRs have been crystallized, and most of
these structures represent inactive-state conformations, with the
exception of the adenosine A2A receptor, the neurotensin receptor
type 1411, and the β-adrenoceptors 1 (β1R) and 2 (β2R).176 The 31
different β1R/β2R structures (while finalizing this chapter a 32nd,
covalently bound, β-adrener-gic structure was released394) cover
multiple receptor activation states in combination with 19 ligands
(Table 6.1) with different functional effects (Table 6.1).102, 104,
105, 187, 193, 206, 302, 303, 386-388, 390-393 This unique GPCR
structure-function data set offers a unique opportunity to assess
the possibil-ities and limitations of the structure-based
computational prediction of GPCR ligand function in virtual
screening (VS) studies. GPCR crystal structures157, 158, 161, 163,
164, 173 and homology models115, 132 have been successfully used to
discover new GPCR ligands. While in most VS studies new antagonists
were found by docking into inverse agonist/antagonist bound GPCR
crystal structures157, 158, 161, 163, 164, 173 or GPCR models based
on the inactive bovine rhodopsin (bRho) structure91 and refined
using known antagonists115, 132, several reports show that also
agonists can be found by structure-based VS in inactive GPCR
models.122, 141, 144, 146, 152, 156 In some of these studies,141,
146, 152 the initial bRho-based inactive state homology model was
refined using true agonists, but in other cases122, 144 the
inactive model was even refined by antagonists. The other way
around, agonist-biased models have also been suc-cessfully applied
to discover new antagonists.156 Previous docking-based VS
studies149 against β2R crystal structures nevertheless indicated
that the functional activity of in silico hits is the same as the
functional activity of the co-crystallized ligand bound to the
structure that is used for the docking simulations.149 While
docking against the first carazolol (antagonist/inverse agonist)
bound β2R crystal structure enabled the discovery of new
β-adrenergic antagonists/inverse ago-nists164, 178, docking studies
against the BI167107 (full agonist) bound active-state crystal
structure facilitated the identification of partial/full
agonists.395 Moreover, even before the determination of
partial/full agonist bound β2R crystal structures104, 303, 388
several studies showed that the in-verse-agonist-bound β2R crystal
structure could be used for the selective identification of
ago-nists by introducing (small) conformational changes in the β2R
structure and using customized scoring approaches to select docking
poses.162, 276, 404, 412 In independent modeling studies162, 276,
404, 412 the position and/or orientation of the side chains of
S2045.43 and S2075.46 (through rotation of
Co-authored by: Rob leurs, Iwan J.P. de Esch, and Chris de
Graaf
Submitted for publication
-
128
Chapter 6 - Structure-based prediction of GPCR-ligand
function
6 6
TM5 and/or adjustment of the rotameric states of S2045.43 and
S2075.46) was customized based on ligand SAR and receptor mutation
studies,248, 263, 389, 408, 409, 413 thereby enabling the
preferen-tial scoring of agonists over antagonists. The selective
identification of partial/full agonists was significantly increased
by employing an interaction fingerprint (IFP)45 scoring method that
de-termines ligand-binding-mode similarity instead of classical
(energy-based) scoring functions.162 These studies illustrate how
(small) differences in protein conformation can influence
struc-ture-based virtual screening results, as has also been
demonstrated by recent comparative VS studies against GPCR crystal
structures and homology models.157, 159, 166, 173 Systematic
consider-ation of alternative protein conformations/flexibility in
docking simulations can, for example, be taken into account either
by incorporating flexibility for (selected) binding-site residues
during the docking simulations414, 415, by creating binding-site
ensembles143, by using multiple crystal structures416, or refining
protein models with different ligands.115, 154, 417
GPCR crystal structure-based identification of new ligands with
a different functional effect than the co-crystallized ligand has
been defined as one of the challenges in the current era of GPCR
structural and chemical biology.117 The growing number of
β-adrenoceptor crystal structures meanwhile covers multiple
activation states and multiple co-crystallized ligands with
different efficacy classes.379, 383, 384 This unique GPCR
structure-function data set allowed us for the first time to
systematically and separately investigate the effect of both the
reference ligands and the receptor conformations in combination
with classical (energy-based)149, 418 and IFP-scoring45, 419
approaches. In this study we have investigated whether (and to what
extent) i) the crystal structures retain their preference for
ligands with the same functional effect as the co-crystal-lized
ligand, ii) the different binding site conformations of the crystal
structures have an impact on the outcome of VS studies, iii)
specific IFPs can change or amplify the preference of a crystal
structure for ligands with a specific functional effect and iv) the
IFP derived from the predicted docking pose of a small agonist
(norepinephrine) can be used for the selective retrieval of
par-tial/full agonists (p/fAGO) over antagonist/inverse agonists
(ANT/iAGO) and decoys.
6.1 Computational Methods
Crystal structures retrieval and preparation
All β-adrenergic crystal structures were downloaded from the PDB
(last accession date, 4th of April 2014).61 Subsequently, all
structures were superposed using MOE420 based on the resi-dues
surrounding the ligands. Based on the initial superposed structures
all (47) residues with-in 4.5Å of any ligand were determined as
being the pocket residues: V90/M822.53, V94/862.57, G98/902.61,
A99/912.62, L101/H932.64, V102/I942.65, G105/K972.68, W107/99,
C114/1063.25, W117/1093.28, T118/1103.29, D121/1133.32,
V122/1143.33, L123/1153.34, V125/1173.36, T126/1183.37,
V172/T1644.56, S173/1654.57, C198/19045.49, C199/19145.50,
D200/19245.51, F201/19345.52, V202/F19445.53, T203/19545.54,
N204/1965.35, Y207/1995.38, A208/2005.39, I209/2015.40,
S211/2035.42, S212/2045.43, I214/V2065.45, S215/2075.46,
F216/2085.47, W303/2866.48, F306/2896.51, F307/2906.52,
N310/2936.55, N313/H2966.58, V314/2976.59, D322/K3057.32,
F325/Y3087.35, V326/I3097.36, A327/L3107.37, N329/3127.39,
W330/3137.40, G332/3157.42, and Y333/3157.43.The first and second
residue number refer to the Uniprot residue number of β1R and β2R,
respec-tively, while the Ballesteros-Weinstein83 and extracellular
loop 2 (ECL2)160 residue numbers are reported as superscript. An
in-house protocol similar to the KLIFS protocol370 was performed to
align all β-adrenergic crystal structures based on the selected 47
pocket residues, protonate the structures and separately extract
the ligand, pocket, protein and (if present) waters, ions, or-
-
129
Structure-based prediction of GPCR-ligand function - Chapter
6
6 6
ganometallics and cofactors in MOL2 format. All structures were
subsequently inspected using MOE and when necessary the
protein-hydroxyl groups were adjusted for optimal H-bonding between
the co-crystallized ligand and the receptor.
Consideration of alternative ligand ring conformations in
3ZPQ
The ring conformation of the fragment-like arylpiperazine ligand
19 in chain B of the β1R crystal structure (PDB-code 3ZPQ) did not
allow for the secondary amine in the piperazine moiety to interact
with both N329/3127.39 and D121/1133.32 due to its twist
conformation. We therefore re-docked the co-crystallized ligand
allowing for all possible ring conformations and selected the
highest-ranking pose, which is in the more stable chair
conformation421 (named 3ZPQ_dock, see Figure S6.1).
Pocket and ligand analyses
Pocket volumes were calculated using POVME422 (version 1.1.0)
with a grid spacing of 0.5Å and a padding of 1.09Å. The initial
inclusion volume was generated based on three spheres with a radius
of 7.0Å of which the centers were based on the superposed
co-crystallized ligands (blue volume in Figure S6.7). RMSD values
between the pockets from the crystal structures were calculat-ed
using MOE. Distance calculations within the quadruplet (formed by
G98/902.61, D121/1133.32, N329/3127.39 and S211/2035.42) and the
subsequent Z-score analyses were performed using an in-house
script. The residues that form the quadruplet were selected based
on their location and function in the binding pocket: G2.61 is
opposite of S5.42 (an important interaction point for polar groups,
in particular for agonists), N7.39 is opposite of D3.32 (both are
key interaction points for the ligand amine group), the N7.39/D3.32
pair is perpendicular to the G2.61/S5.42 pair (Figure S6.3), and
all four residues are roughly positioned at the same height in the
TM helices with respect to the membrane. The 2D similarity of the
co-crystalized ligands was assessed using Extended Connectivity
fingerprint (ECFP-4) in the Pipeline Pilot Suite385 and EDprints30
(in-house imple-mentation).
Compound preparation and docking
26 ligands (13 f/pAGO and 13 ANT/iAGO) and 980 decoy compounds
were obtained from the supporting information accompanying the
publication from de Graaf et al..162 In total 24 addition-al
β-adrenergic ligands (12 f/pAGO and 12 ANT/iAGO) were selected
based on the classification of Baker.333, 423 Molecular (2D)
structures of all ligands are shown in Figure S6.4. All compounds
were prepared starting from the SMILES format by first protonating
all compounds at physiologi-cal pH using Chemaxon's Calculator
(version 5.1.4)364 and were subsequently converted to MOL2 format
using CORINA (version 3.49).365, 424 All molecular docking
simulations were carried out in fivefold using PLANTS345 (version
1.2) using the (default) ChemPLP scoring function, search speed 2,
flipping of free ring corners enabled, and generating 99 poses for
every compound with a clustering RMSD of 1.0. The docking
experiments were performed in fivefold in order to reduce the
effect of the stochastic nature of PLANTS345, as the use of a
stochastic algorithm (as used in multiple docking algorithms, e.g.
PLANTS345, GOLD425, and QXP426) can result in different docking
poses for the same compound when performing multiple docking
runs.The binding site was based on the center of carazolol in β2R
(PDB-code 2RH1) with a binding-site radius of 15Å, thereby covering
all pocket residues (except for G105/K972.68, which is pointing
away from the binding site). In total 1030 compounds were docked in
48 β-adrenergic structures yielding almost 24.5 million docking
poses in total.
-
130
Chapter 6 - Structure-based prediction of GPCR-ligand
function
6 6
Interaction Fingerprint calculation and scoring
From the individual pockets and ligands, that were extracted
from the crystal structures, in-teractions fingerprints were
calculated through application of the FingerPrintLib (version 2.2,
downloaded from http://bioinfo-pharma.u-strasbg.fr, using OpenEye’s
OEChem Toolkit).45, 427 All docking poses were post-processed with
IFP by combining the docking poses with the consistent pocket
residue selection as indicated earlier. The IFPs from all
docking-poses were scored using all 40 reference IFPs resulting in
~979 million IFP-scores representing their interaction profile
similarity (as calculated by the Tanimoto similarity
coefficient).
Virtual screening assessment
Based on the PLANTS and IFP scores of all docking poses the
highest scoring docking poses according to each of the scoring
methods were selected. The resulting hit lists were used to rank
both the f/pAGO and the ANT/iAGO ligands as well as the decoys and
determine the enrichment factor (EFx, where x is the false-positive
rate) for each reference IFP (or ChemPLP score) and docking
structure combination. The EFs were calculated by determining the
true-positive rate (TP-rate, i.e. the percentage of f/pAGO and
ANT/iAGO respectively) at a 1% false-positive rate347 (FP-rate,
i.e. 1% of the decoys, see Figure 6.3 and 6.5a/b) resulting in the
EF1% of 1968 scoring combinations (40 reference IFP scores and 1
PLANTS score combined with 48 structures) for both the ranking of
f/pAGO and ANT/iAGO ligands (see Figure 5).
6.2 Subtle ligand-type-dependent structural changes in the
receptor binding site
There currently are 31 β-adrenergic structures; 15 human β2R and
16 turkey β1R structures (Table 6.1).102-105, 187, 193, 206, 302,
303, 386-388, 390-393 While finalizing this chapter a 32nd
β-adrenergic structure was released394 with a covalently-bound
agonist that forms similar H-bond interactions with TM5 as the
epinephrine (PDB-code 4LDO)103 and has the same covalent linker as
FAUC50 (PDB-code 3PDS).388 Multiple crystallization methods have
been employed to obtain stable crystals for structure determination
of these receptor-ligand complexes,93 amongst them T4 lysozyme
insertion at the intracellular loop (ICL) 3 or at the N-terminus,
introduction of thermostabilizing mutations, introduction of
covalently binding ligands, and the addition of antibody
fragments.428 All crystal structures were obtained in combination
with one of the 19 unique ligands (apart from the ligand-free
structure,393 see Table 6.1 and 6.2); 9 (partial/full) inverse
agonists, 7 (partial/full) agonists, 1 covalently-bound full
agonist, and 2 unvalidated fragments (but suggested to be
antagonists,392 vide infra). Throughout this manuscript we
abbreviate full and partial agonists as f/pAGO, antagonists as ANT
and (full/partial) inverse agonists as iAGO (see legend Table 6.1).
It should be noted that we did not focus on the differences between
β1R and β2R in this manuscript, as the pocket-lining residues are
almost identical (only residue 7.35 differs, a phenylalanine in β1R
but tyrosine in β2R) we assume that the ligand-binding modes in β1R
and β2R are transferable.429 Also, in the set of ligands that we
analyzed there is only one highly selective compound (ICI 118,551,
which has a ~540 fold higher affinity for β2R than for β1R).423
Moreover, it is suggested that selectivity for one of the
receptors333, 423, 430 is based on ligand kinetics431 (i.e. binding
to and dissociation from the receptor) and receptor activation429
rather than differences in ligand binding modes.431
-
131
Structure-based prediction of GPCR-ligand function - Chapter
6
6 6
Table 6.1. Overview of all β-adrenoceptor crystal structures,
the co-crystallized ligands and their function.
PDB Ligand Funct-iona,p Biase PDB Ligand Funct-iona,p Bias
e
3NYA193 OOH
NH2+
Alprenolol
ANT
●●●●●
p/fAGOANT/iAGOANT (Not
validated)Fab−complexLigand−freep/fAGOANT/iAGOANT (Not
validated)Fab−complexLigand−free
Yesj,k
Nof,i3ZPR392
N
N+H2N
Arylpiperazine 20
ANTb●●●●●
p/fAGOANT/iAGOANT (Not
validated)Fab−complexLigand−freep/fAGOANT/iAGOANT (Not
validated)Fab−complexLigand−free
-
4AMI105
N
OOH
NH2+
HN
Bucindolol
ANTo
●●●●●
p/fAGOANT/iAGOANT (Not
validated)Fab−complexLigand−freep/fAGOANT/iAGOANT (Not
validated)Fab−complexLigand−free
Yesg,m3P0G303,q
3SN6104,q
4LDE103,q
O
OHNH
O
NH2+ OH
BI-167107
fAGO
●●●●●
p/fAGOANT/iAGOANT (Not
validated)Fab−complexLigand−freep/fAGOANT/iAGOANT (Not
validated)Fab−complexLigand−free
Yesn
2YCW390
2RH1102
4GBR391
2R4R386,d
2R4S386,d
3KJ6206,d
NH
ONH2+ OH
Carazolol
iAGOo●●●●●
p/fAGOANT/iAGOANT (Not
validated)Fab−complexLigand−freep/fAGOANT/iAGOANT (Not
validated)Fab−complexLigand−free
Noh 2Y02302
OHNH2+
O OHNH
O
Carmoterol
fAGO
●●●●●
p/fAGOANT/iAGOANT (Not
validated)Fab−complexLigand−freep/fAGOANT/iAGOANT (Not
validated)Fab−complexLigand−free
Yesm,n
4AMJ105OH
O
NH
NH2+
O
O
Carvedilol
iAGO
●●●●●
p/fAGOANT/iAGOANT (Not
validated)Fab−complexLigand−freep/fAGOANT/iAGOANT (Not
validated)Fab−complexLigand−free
Yesg,j,m
Nof2Y00302
2Y01302
Dobutamine
pAGO
●●●●●
p/fAGOANT/iAGOANT (Not
validated)Fab−complexLigand−freep/fAGOANT/iAGOANT (Not
validated)Fab−complexLigand−free
Yesf
Nol
2VT4387
2YCX390
2YCY390
4BVN101
NH
ONH2+
N
OH
Cyanopindolol
iAGOo●●●●●
p/fAGOANT/iAGOANT (Not
validated)Fab−complexLigand−freep/fAGOANT/iAGOANT (Not
validated)Fab−complexLigand−free
- 4LDO103,q OH
OH
OHNH2+
Epinephrine
fAGO
●●●●●
p/fAGOANT/iAGOANT (Not
validated)Fab−complexLigand−freep/fAGOANT/iAGOANT (Not
validated)Fab−complexLigand−free
Nof,g,l
3NY8193 NH2+
OOH
ICI 118,551
iAGO●●●●●
p/fAGOANT/iAGOANT (Not
validated)Fab−complexLigand−freep/fAGOANT/iAGOANT (Not
validated)Fab−complexLigand−free
Nof,h 3PDS388,c
FAUC50
fAGO
●●●●●
p/fAGOANT/iAGOANT (Not
validated)Fab−complexLigand−freep/fAGOANT/iAGOANT (Not
validated)Fab−complexLigand−free
-
2YCZ390
N
INH
ONH2+ OH
Iodocyanopindolol
iAGO●●●●●
p/fAGOANT/iAGOANT (Not
validated)Fab−complexLigand−freep/fAGOANT/iAGOANT (Not
validated)Fab−complexLigand−free
- 4LDL103,qOH
OH
OHNH2+
HO
Hydroxybenzyl-isoproterenol
fAGO
●●●●●
p/fAGOANT/iAGOANT (Not
validated)Fab−complexLigand−freep/fAGOANT/iAGOANT (Not
validated)Fab−complexLigand−free
Yesn
3D4S187
O
N NS
NOOH
NH2+
Timolol
iAGO●●●●●
p/fAGOANT/iAGOANT (Not
validated)Fab−complexLigand−freep/fAGOANT/iAGOANT (Not
validated)Fab−complexLigand−free
Nof 2Y03302 OH
OH
OHNH2+
Isoproterenol
fAGO
●●●●●
p/fAGOANT/iAGOANT (Not
validated)Fab−complexLigand−freep/fAGOANT/iAGOANT (Not
validated)Fab−complexLigand−free
Nof,g,-h,i,j,k,l
OHOH
NH2+
HO
NH2+ OH
OO OH
S O
NH
-
132
Chapter 6 - Structure-based prediction of GPCR-ligand
function
6 6
PDB Ligand Funct-iona,p Biase PDB Ligand Funct-iona,p Bias
e
3NY9193
OO
O
OOH
NH2+
VS hit (Kolb)
iAGO●●●●●
p/fAGOANT/iAGOANT (Not
validated)Fab−complexLigand−freep/fAGOANT/iAGOANT (Not
validated)Fab−complexLigand−free
- 2Y04302OH
OH
OHNH2+
Salbutamol
pAGO
●●●●●
p/fAGOANT/iAGOANT (Not
validated)Fab−complexLigand−freep/fAGOANT/iAGOANT (Not
validated)Fab−complexLigand−free
Nol
3ZPQ392 N+H2N
NH
Arylpiperazine 19
ANTb●●●●●
p/fAGOANT/iAGOANT (Not
validated)Fab−complexLigand−freep/fAGOANT/iAGOANT (Not
validated)Fab−complexLigand−free
- - OH
OH
OH
+H3N
Norepinephrine
fAGO
●●●●●
p/fAGOANT/iAGOANT (Not
validated)Fab−complexLigand−freep/fAGOANT/iAGOANT (Not
validated)Fab−complexLigand−free
Nog
4GPO393 Ligand-free
a) Function abbreviations: ANT – antagonist, AGO – agonist, i –
inverse, p – partial, f – full. b) Not validated.392 c)
covalent-ly-bound ligand. d) Ligand and extracellular part of the
receptor were not resolved due to weak electron density. e) This
column indicates whether a ligand has a signaling preference for
β-arrestins over G-proteins or not according to f) Casella et
al.430, g) Drake et al.432, h) Kahsai et al.433, i) Kaya et al.434,
j) Kim et al.435, k) Liu et al.205, l) Rajagopal et al.436, m)
Warne et al.105 and n) Weiss et al.395. o) Although these ligands
are considered antagonists/inverse agonists, they can also have a
partial agonist effect depending on the activation state of the
receptor.333, 437 p) Color coding: Full/partial agonist (red),
antagonist/inverse agonist (blue), or unvalidated antagonist
(cyan). Ligands with a circle do not show a signaling preference or
it is not known for these compounds, ligands with triangular icons
have been shown to have a β-arrestin signaling bias. q) The
receptor is in its active state.
Superposition of all individual monomers of the crystal
structures of β1R and β2R based on their pocket residues (see
Materials and Methods) shows the conserved heptahelical fold, but
also indicates some structural variation, in particular in TM1 (a
60 degree kink)387, 390, TM5 and TM6 (which differ between active
and inactive states104, 303 and the straight and bent
inactive-state TM6390). While the structural differences in the
intracellular regions have been described in sev-eral reviews73,
176, our systematic analysis will focus on the structural
variations in the ligand-bind-ing site within the β-adrenergic GPCR
family.Figure 6.2a/b shows that the pocket residues in all (β1R and
β2R) crystal structures have largely conserved conformations. The
outliers are the FAB-complexes (PDB-codes: 2R4R, 2R4S386 and
3KJ6206) in which the ECL domain, the top of the helices, and the
ligand (carazolol) were not solved due to the distorted density in
this region.206, 386 When comparing the pocket residues by means of
the mutual RMSD values (Table S6.2, Figure 6.2c), they are
generally low (average all-atom RMSD 0.9Å). By determining the
average RMSD of the binding site residues of each individual
structure to all f/pAGO and all ANT/iAGO structures it becomes
apparent that all f/pAGO structures and all inactive-state ANT/iAGO
structures cluster together (Figure 6.2c). This means that the
binding site conformation within the inactive-state f/pAGO
structures, but also within the ANT/iAGO structures is conserved.
Moreover, the active-state structures form a sep-arate cluster
compared to other f/pAGO structures, which can be ascribed to the
movement of TM5 and TM6104, 303 (Table 6.1, Table S6.2, Figure
S6.2).438 Outliers are the FAB-complexes (due to the distorted
electron density), but also the ligand-free structure (4GPO393),
which have a high RMSD value compared to the other structures.
Interestingly, this apo structure does not cluster with any of the
f/pAGO structures or with ANT/iAGO structures.To further quantify
the differences between the pockets we measured the distances
between the Cα atoms of G2.61, D3.32, N7.39, and S5.42 (Figure
6.1c, Figure S6.3, and Table S6.3). While the mea-surements show
contraction of the pocket upon agonist binding302 (mainly through
movement of TM5), the highest contraction of the pocket was found
for the ligand-free β1R structure.393
-
133
Structure-based prediction of GPCR-ligand function - Chapter
6
6 6
RMSD
a
(f/pAGO)
RMSD
a
(ANT/iAGO
)
Distanceb
Volumec
Receptor
2RH1_A 0.8 0.6 0.2 377 β22VT4_A 0.8 0.6 0.2 492 β12VT4_B 0.9 0.7
0.1 442 β12VT4_C 0.9 0.7 0.1 442 β12VT4_D 0.8 0.6 0.2 493 β12YCW_A
0.7 0.5 -0.2 411 β12YCW_B 0.8 0.6 -0.2 409 β12YCX_A 1.0 0.7 0.2 459
β12YCX_B 1.0 0.7 0.1 464 β12YCY_A 0.8 0.6 0.0 437 β12YCY_B 0.8 0.6
-0.1 441 β12YCZ_A 0.9 0.6 0.2 469 β12YCZ_B 0.9 0.6 0.2 472 β13D4S_A
0.8 0.6 0.2 278 β23NY8_A 0.9 0.7 0.8 387 β23NY9_A 0.8 0.6 0.5 299
β23NYA_A 0.9 0.7 0.5 248 β24AMI_A 0.7 0.6 -0.2 411 β14AMI_B 0.6 0.6
-0.2 396 β14AMJ_A 0.7 0.5 0.5 488 β14AMJ_B 0.7 0.5 0.3 424 β14BVN_A
0.7 0.5 0.5 302 β14GBR_A 0.9 0.7 0.8 360 β23ZPQ_A 0.7 0.5 0.0 445
β13ZPQ_B 0.7 0.5 0.0 391 β13ZPR_A 0.7 0.5 0.2 460 β13ZPR_B 0.7 0.5
0.1 414 β12Y00_A 0.6 0.6 -0.6 368 β12Y00_B 0.6 0.6 -0.7 325
β12Y01_A 0.6 0.6 -0.4 435 β12Y01_B 0.6 0.6 -0.6 319 β12Y02_A 0.6
0.7 -0.4 430 β12Y02_B 0.6 0.7 -0.5 389 β12Y03_A 0.7 0.7 -0.1 462
β12Y03_B 0.6 0.6 -0.4 348 β12Y04_A 0.6 0.6 0.0 493 β12Y04_B 0.6 0.5
-0.5 406 β13P0G_A 0.9 1.1 -0.6 270 β23PDS_A 0.9 0.7 -0.6 293
β23SN6_Rd 1.0 1.1 -0.7 388 β24LDE_A 0.9 1.2 -0.8 284 β24LDL_A 0.9
1.2 -1.0 273 β24LDO_A 1.0 1.2 -1.0 236 β24GPO_A 1.0 0.9 -1.5 282
β14GPO_B 1.0 0.9 -1.5 283 β12R4R_A 2.4 2.2 2.6 - β22R4S_A 2.4 2.2
2.7 - β23KJ6_A 2.2 2.0 1.4 - β2
H2.64
S5.42
S5.43
S5.46
D45.51
F45.52
Y5.38
D3.32
K/D7.32
Y/F7.35
H2.64
W3.28
F6.51
N6.55
F45.52
A5.39
V3.33
V3.36 W6.48 Y7.43
I7.36
b
a c
Figure 6.1 A visual (a-b) and quantitative (c) analysis of the
pocket residues from all X-ray structures. Pocket residues (carbon
atoms are colored blue, salmon, white and gray for ANT/iAGO,
f/pAGO, FAB-complexed and ligand-free structures respectively) of
all chains of all crystal structures (listed in C by their PDB-code
followed by their chain identifier) are shown superimposed with the
cartoon representation of β2R (2RH1102) and bucindolol (green
carbon atoms, 4AMI105), as seen from the side (a) and from the top
(b). An overview (c) of a) Average all-atom RMSD using the pocket
residues (Å) compared to all 16 f/pAGO and all 27 ANT/iAGO
structures (full overview in Table S6.2), b) the average Z-score of
the distances between G2.61, D3.32, N7.39 and S5.42 (full overview
in Table S6.3, Figure S6.3) and c) the pocket volume (Å3, Figure
S6.4). d) The side chains of the ECL2 residues are not all fully
resolved, thereby influencing the pocket volume (Figure S6.7) as
well as RMSD values. The red and blue background coloring mark
values associated with f/pAGO and ANT/iAGO properties,
respectively. Icons: f/pAGO (red), ANT/iAGO (blue), or unvalidated
ANT(cyan) with no or unknown signaling preference (circle), or
β-arrestin biased ligands (triangle) (see Table 6.1).
Note, however, that these differences are small and that mainly
the distance between D3.32 and N7.39 to S5.42 varies between the
different structures (on average 0.6Å and 1.0Å, respectively).
Analysis of volumes of the individual binding pockets (Figure 6.1c,
Figure S6.7) indicates that β1R generally has a bigger pocket
volume than β2R (on average 415 Å3 compared to 308 Å3). This can
mainly be ascribed to the residue differences at positions 7.32
(D/K in β1R/2 respectively) and 7.35 (F/Y in β1R/2 respectively)
resulting in a more confined ligand-binding space in β2R
(Figure
-
134
Chapter 6 - Structure-based prediction of GPCR-ligand
function
6 6
6.1, Figure S6.7). Although the RMSD values between all pocket
residues are low (Figure 6.1c, Table S6.7), the effect of all
subtle changes combined is reflected by the large differences in
the pocket volumes. For the β1R and β2R the pocket volumes range
from 282 to 493 Å3 and from 236 to 388 Å3, respectively (only
comparing structures with fully resolved pocket residues). In line
with the results of the distance measurements and the RMSD
calculations, the ligand-free structure has the smallest pocket
volume of all β1R structures. The active-state adrenaline-bound
structure (4LDO103) has the smallest pockets volume of all β2R
structures, adapted to this small endogenous agonist.
6.2.1 Identification of ligand-type specific molecular
interaction profiles
The binding modes and interaction profiles of the 19 different
co-crystallized ligands (Table 6.1 and 6.2) were subsequently
investigated using the aforementioned IFP method.45 This technique
encodes 7 different types of interactions between the ligand and
each of the binding pocket residues. The following interactions
with the ligand are encoded: hydrophobic contact, aromatic
face-to-edge and face-to-face, H-bond acceptor-donor and
donor-acceptor, negative-positive and positive-negative ionic
interaction (as shown in Figures 6.2 and 6.5).45 The IFPs for all
ligands in all monomers were calculated for the available
β-adrenergic crystal structures. The 31 crystal structures contain
in total 48 monomers (43 ligand-bound). Of the 14 crystallized
ligand-pro-tein complexes containing more than one monomer, only 3
structures had identical interaction patterns within different
monomers of the same crystal structure (see Figure 6.2). Moreover,
ligands present in multiple co-crystal structures (e.g.
cyanopindolol, carazolol, BI-167107) did also not yield identical
IFPs. This emphasizes once more that crystal structures are
snapshots in time439 of entities that are flexible by nature. The
overview of all generated reference IFPs from the ligand-bound
crystal structures (Figure 6.2) shows that there are several
conserved interactions within the pocket. Apart from dobu-tamine
and arylpiperazines 19 and 20, all co-crystallized β1R and β2R
ligands contain an etha-nolamine moiety (which is abundant in
β-adrenergic ligands368) that can form a tight H-bond interaction
network with the sidechains of D3.32 and N7.39 (Figures 6.2,
6.5a/b).367, 389 In chain B of the arylpiperazine 19 structure the
ligand does not form H-bonds with N7.39 nor with D3.32. We
therefore redocked the co-crystallized ligand into its structure
considering alternative ring conformations, resulting in an
additional IFP (annotated as 3ZPQ_dock, see Figure 6.2, Figure
S6.1, and Materials and Methods).Most β-adrenergic ligands contain
an H-bond donor with which they are able to interact with
S211/2035.42 (74%, Figure 6.2, Table S6.8).429 More specific for
full agonists (carmoterol, iso-proterenol, BI-167107 and FAUC50) is
a catechol or a catechol-mimicking moiety with two (or more) H-bond
donors that allows for an extra H-bond with another TM5 serine,
namely S215/2075.46.440 Partial agonists also have two H-bond
donors that can interact with TM5, but their interactions seem to
be less optimal (e.g. the shortest observed distance between the
H-bond donor from the catechol in dobutamine to the S5.46 acceptor
is 3.8Å).429 All ligands also contain at least one aromatic ring,
which allows for aromatic stacking with F307/2906.52, F306/2896.51,
and F201/19345.52 in 95%, 56% and 54% of all structures
respectively. We furthermore observed aromatic interactions with
W117/1093.28 for many agonists (60%), ionic interactions with the
key ionic anchor D121/1133.32 for all ligands and double H-bond
interactions with N329/3127.39 as de-scribed above (see Table
S6.8).
-
135
Structure-based prediction of GPCR-ligand function - Chapter
6
6 6
Liga
ndAl
pren
olol
3NYA
_Aβ2
89%
25%
Buci
ndol
ol4A
MI_A
β147
%31
%Bu
cind
olol
4AMI
_Bβ1
47%
38%
Cara
zolo
l2R
H1_A
β284
%6%
Cara
zolo
l2Y
CW_A
β189
%13
%Ca
razo
lol
2YCW
_Bβ1
89%
0%Ca
razo
lol
4GBR
_Aβ2
63%
0%Ca
rved
ilol
4AMJ
_Aβ1
42%
13%
Carv
edilo
l4A
MJ_B
β126
%6%
Cyan
opin
dolo
l2V
T4_A
-Dβ1
89%
44%
Cyan
opin
dolo
l2Y
CX_A
β174
%0%
Cyan
opin
dolo
l2Y
CX_B
β174
%13
%Cy
anop
indo
lol
2YCY
_Aβ1
95%
25%
Cyan
opin
dolo
l2Y
CY_B
β184
%13
%Cy
anop
indo
lol
4BVN
_Aβ1
89%
50%
ICI51
18,5
513N
Y8_A
β279
%25
%Io
docy
anop
indo
lol2Y
CZ_A
-Bβ1
95%
31%
Tim
olol
3D4S
_Aβ2
84%
31%
VS5h
it5(K
olb)
3NY9
_Aβ2
74%
6%Ar
ylpi
pera
zine
519
3ZPQ
_Aβ1
11%
0%Ar
ylpi
pera
zine
519
3ZPQ
_Bβ1
21%
0%Ar
ylpi
pera
zine
519
3ZPQ
_doc
kβ1
58%
6%Ar
ylpi
pera
zine
520
3ZPR
_Aβ1
68%
0%Ar
ylpi
pera
zine
520
3ZPR
_Bβ1
42%
0%BI
G167
107
3P0G
_Aβ2
16%
63%
BIG1
6710
73S
N6_R
β258
%56
%BI
G167
107
4LDE
_Aβ2
47%
69%
Carm
oter
ol2Y
02_A
β116
%75
%Ca
rmot
erol
2Y02
_Bβ1
16%
81%
Dobu
tam
ine
2Y00
_Aβ1
0%38
%Do
buta
min
e2Y
00_B
β111
%38
%Do
buta
min
e2Y
01_A
β10%
56%
Dobu
tam
ine
2Y01
_Bβ1
5%50
%Ep
inep
hrin
e4L
DO_A
β20%
50%
FAU
C50
3PDS
_Aβ2
0%44
%Hy
drox
yben
zylIS
O4L
DL_A
β211
%63
%Is
opro
tere
nol
2Y03
_A-B
β126
%75
%Sa
lbut
amol
2Y04
_Aβ1
42%
88%
Salb
utam
ol2Y
04_B
β163
%81
%63
%N
orep
inep
hrin
ea2Y
03_A
β10%
63%
Hydr
opho
bic5
cont
act
Arom
atic
5face
GtoG
face
Arom
atic
5edg
eGto
Gface
Hbon
d5(r
esid
ue5d
onor
)Hb
ond5
(res
idue
5acc
epto
r)Io
nic5
(res
idue
5pos
itive
)Io
nic5
(res
idue
5neg
ativ
e)
W10
73.
285.
466.
48
GL/
HV/
IW
99W
6.51
45.5
145
.50
6.52
45.5
245
.54
5.38
5.39
5.42
5.43
7.40
7.43
6.55
7.35
555555
555Re
sidu
e5C
hain
555555
555555
555555
5552.61
2.64
2.65
7.36
7.39
T
3.32
3.33
3.36
3.37
3.29
D
TM3
ECL2
TM5
FT
YA
SS
DV
VT
CS
Receptor
TM2
ECL1
ANT/iAGO5IFP5≥50.6
f/pAGO5IFP5≥50.6
TM7
V/I
NW
YF/
Y
TM6
WF
FN
Figu
re 6
.2 A
n ov
ervi
ew o
f the
uni
que
inte
ract
ion
finge
rprin
ts o
f all
co-c
ryst
alliz
ed li
gand
s. Th
e co
lors
indi
cate
the
pres
ence
of a
n in
tera
ctio
n (a
s see
n fr
om th
e re
sidue
) ac
cord
ing
to th
e co
lors
des
crib
ed a
t the
bot
tom
of t
he fi
gure
. Ide
ntic
al IF
Ps fo
r mul
tiple
mon
omer
s w
ithin
a P
DB-
entr
y ar
e gr
oupe
d (e
.g. 2
VT4_
chai
nA-D
). Th
e la
st
two
colu
mns
des
crib
e th
e am
ount
of t
imes
(as a
per
cent
age
of th
e to
tal c
ompa
rison
s) a
n IF
P co
mpa
rison
resu
lts in
a sc
ore ≥
0.6
whe
n co
mpa
red
with
the
ANT/
iAG
O
refe
renc
e IF
Ps (a
blu
e ba
ckgr
ound
indi
cate
s a h
igh
perc
enta
ge) a
nd th
e f/
pAG
O re
fere
nce
IFPs
(a re
d ba
ckgr
ound
indi
cate
s a h
igh
perc
enta
ge).
a) Th
e IF
P of
the
high
est
scor
ing
dock
ing
pose
of n
orep
inep
hrin
e in
2Y0
3-ch
ainA
(see
Fig
ure
6.6)
. Nam
es a
nd ic
ons:
f/pA
GO
(red
), AN
T/iA
GO
(blu
e), o
r unv
alid
ated
AN
T392
(cya
n) w
ith n
o or
un
know
n sig
nalin
g pr
efer
ence
(circ
le),
or β
-arr
estin
bia
sed
ligan
ds (t
riang
le) (
see
Tabl
e 6.
1).
-
136
Chapter 6 - Structure-based prediction of GPCR-ligand
function
6 6
All interaction profiles (IFPs) of co-crystallized ligands (8
f/pAGO and 11 ANT/iAGO) in 33 β1R and 15 β2R binding sites were
compared and their Tanimoto similarity scores were determined
(Table S6.4). Figure 6.2c reports for each individual IFP the
percentage of similar IFPs (Tanimoto similarity score ≥ 0.645)
derived from (other) ANT/iAGO and f/pAGO co-crystal structures
com-plexes. This analysis shows that the IFPs of ANT/iAGO ligands
are more similar to each other than to f/pAGO IFPs (75% versus 19%
similar pairs respectively). Vice versa, the pairwise similarity
between f/pAGO IFPs is higher (62%) than similarity to ANT/iAGO
(62% versus 21% respectively, Figure 6.2). Moreover, the
arylpiperazine ligands cluster together with the ANT/iAGO IFPs (on
average 40% similar pairs compared to 1% similarity with f/pAGO
IFPs) suggesting that these compounds are antagonists, as
previously proposed.392 These analyses show that the functional
effect of a ligand is encoded in the ligand-protein interaction
patterns, and suggest that these IFPs can be used to discriminate
ANT/iAGO from f/pAGO ligands in structure-based virtual screening
studies.
Figure 6.3 Overall enrichment at 1% false positive rate
(FP-rate) for the retrieval of 25 partial/full agonists and 25
inverse agonists/antagonists over a set of 980 decoy molecules
using IFP scoring (a). Full ROC curves visualizing the retrieval
rate (TP-rate) of f/pAGO (red) and ANT/iAGO (blue) in the best
ANT/iAGO structure (b) and best f/pAGO structure (c, legend shown
in b). The structures are indicated by their PDB code followed by
an underscore and the chain identifier (except when there was only
one chain or all chains of the structure had similar performance).
The 2D structures represent the co-crystallized ligand for selected
structures. The ANT/iAGO-axis is scaled from 0 to 60 and the f/pAGO
axis from 0 to 100. *1) 3ZPQ_A, 3ZPQ_dock, 2Y00_B. *2) 4AMJ_B,
3ZPQ_B. Icons: fpAGO (red), ANT/iAGO (blue), or unvalidated ANT392
(cyan) with no or unknown signaling preference (circle), or
β-arrestin biased ligands (triangle) (see Table 6.1).
6.2.2 Selective retrieval of f/pAGO over ANT/iAGO in
structure-based virtual screen-ing
In order to quantify to what extent the differences between the
individual structures and the reference IFPs have an influence on
the outcome and selectivity of a VS, we docked a set of 1030
-
137
Structure-based prediction of GPCR-ligand function - Chapter
6
6 6
compounds comprising 25 f/pAGO, 25 ANT/iAGO and 980
physicochemically similar decoys162 in all β1R and β2R structures.
The docking studies were performed using PLANTS345 using the
ChemPLP scoring function and the resulting binding modes were
post-processed using IFP45 and ranked according to their PLANTS and
IFP score. In 85% of all f/pAGO and ANT/iAGO structures IFP scoring
yielded equal or higher enrichment factors (EF1%) than
ChemPLP-scoring (Figure 6.4, Figure S6.6). Using IFP to rank the
docking poses of ligands of the same functional activity class
gives a virtual screening enrichment that is on average two times
as high as the enrichment obtained by ChemPLP-scoring. Moreover, in
90% of the structures IFP scoring yielded an equal or higher
selectivity (higher retrieval of f/pAGO than ANT/iAGO over decoys
for f/pAGO structures and vice versa) than when using ChemPLP
scoring (Figure 6.4, Figure S6.6). The resulting EF1% values of
f/pAGO and ANT/iAGO based on IFP similarity scoring (using the
co-crystallized ligand reference IFP) are shown in Figure 6.4a. For
36 of the 43 protein-ligand complexes, the EF1% of ligands with the
same functional effect as the co-crystallized ligand was higher
than the enrichment of ligands with a different functional effect.
Virtual screening runs against five (putative) ANT/iAGO bound
receptor struc-tures, namely the arylpiperazine-19-bound β1R
complexes (3ZPQ), the bucindolol-β2R complexes (4AMI), and one of
the carvedilol-β2R complexes (4AMJ), resulted in a higher
enrichment for p/fAGO ligands than for ANT/iAGO ligands. For the
arylpiperazine-20-bound β1R structures392 (PDB-code 3ZPR) no
enrichment was obtained for any of the known ligands (EF1% = 0).
For 11 unique crystal structures (of the 14 structures with
multiple monomers) a significant dif-ference in the EF1% is
observed (ΔEF1% > 5) between the individual monomers,
demonstrating the impact of subtle differences in the reference IFP
and/or binding site structure in different monomers of the same
crystal structure. For example, the EF1% for f/pAGO over decoys for
chain A of structure 2Y02 is 84 compared to 52 for chain B, while
the reference IFPs of both chains is identical. This can
potentially be ascribed to the larger pocket volume of chain A (490
versus 389 Å3 for chain B, which is mainly due to the rotation of
F3257.35), therefore the pocket of chain A is better able to
accommodate the larger ligands. In contrast, the small pocket of
the ligand-free β1R structure393 (4GPO) is not able to effectively
accommodate ligands as D1213.32, F3076.52, and N3297.39 are moved
inwards (Figure S6.6), thereby preventing the prediction of correct
binding modes using rigid-protein docking simulations. Similarly,
the small pocket of the active-state, adrenaline-bound β2R
structure (4LDO103) is not able to accommodate the relatively
bigger TM5/6 binding moieties of most ANT/iAGO and is therefore
more selective for f/pAGO.Figure 6.3 shows that overall the
ANT/iAGO ligand-protein complexes are more selective (av-erage EF1%
of 27 and 4 for ANT/iAGO and f/pAGO respectively) than the f/pAGO
complexes (average EF1% of 13 and 46 for ANT/iAGO and f/pAGO
respectively). However, virtual screening against f/pAGO bound
structures in many cases gives a high enrichment for both ANT/iAGO
and (an even higher enrichment for) p/fAGO ligands. For example,
virtual screening against chain A of the salbutamol-bound β1R
structure (PDB-code 2Y04), results in a remarkably high enrich-ment
factor for both f/pAGO and ANT/iAGO (EF1% of 72 and 32
respectively). To evaluate the individual contribution of the
reference IFPs and the protein structure we re-scored all docked
compounds in all structures with all reference IFPs (i.e. not only
with the IFP that was derived from the corresponding
receptor-ligand complex). This furthermore allowed us to rescore
the docked compounds with IFP in the structures without a ligand
(PDB-codes: 2R4R, 2R4S, 3KJ6, and 4GPO). The resulting
scoring-matrices containing the EF1% of f/pAGO and ANT/iAGO are
shown in Figure 6.4a and 6.4b respectively.
-
138
Chapter 6 - Structure-based prediction of GPCR-ligand
function
6 6
03
58
1013
1518
2023
2528
3033
3538
4043
4548
5053
5558
6063
6568
7073
7578
8083
8588
9093
9598
100
EF 1
% (f
/pA
GO
)
PLANTS
2RH1_A
2VT4_A-D
2YCW_A
2YCW_B
2YCX_A
2YCX_B
2YCY_A
2YCY_B
2YCZ_A-B
3D4S_A
3NY8_A
3NY9_A
3NYA_A
4AMI_A
4AMI_B
4AMJ_A
4AMJ_B
4BVN_A
4GBR_A
3ZPQ_A
3ZPQ_B
3ZPQ_dock
3ZPR_A
3ZPR_B
2Y00_A
2Y00_B
2Y01_A
2Y01_B
2Y02_A
2Y02_B
2Y03_A-B
2Y04_A
2Y04_B
3P0G_A
3PDS_A
3SN6_R
4LDE_A
4LDL_A
4LDO_A
Norepiphrine
2RH1
_Aβ2
80
80
00
88
08
40
012
2016
00
120
816
360
016
2012
1232
1656
4044
2028
4044
4452
482V
T4_A
β18
00
40
04
40
00
120
08
200
00
012
1628
00
128
812
2424
4452
4824
80
1216
5256
2VT4
_Bβ1
80
04
00
00
00
00
40
2424
88
40
164
240
016
2828
2424
3636
4036
2836
440
4432
442 V
T4_C
β112
00
40
00
00
04
00
424
2412
120
08
428
00
1216
1624
3632
4444
3632
248
4836
5244
2VT4
_Dβ1
80
00
04
88
00
012
00
1628
04
80
812
160
00
00
812
1648
4852
124
44
1656
602Y
CW_A
β120
00
00
44
80
84
120
424
124
420
04
1220
00
1216
1216
3216
6464
6036
1624
4436
3272
2YCW
_Bβ1
160
80
00
412
08
412
012
208
88
200
824
320
012
1620
1232
3272
5660
2416
816
4044
482Y
CX_A
β120
012
00
08
120
160
44
012
160
020
04
836
00
88
128
160
4444
368
164
1616
4048
2YCX
_Bβ1
280
80
00
88
08
48
04
812
44
160
412
440
012
1616
2020
844
4036
1612
828
2832
442 Y
CY_A
β116
00
00
00
00
00
00
012
2012
120
04
1216
00
424
164
2020
5244
4040
164
2828
4444
2YCY
_Bβ1
80
00
00
00
00
00
00
2816
00
00
84
40
08
812
1220
2044
4036
3616
828
3640
402Y
CZ_A
β18
00
00
00
00
00
00
016
160
80
08
412
00
80
412
128
4036
4020
40
1632
2044
2YCZ
_Bβ1
120
00
00
00
00
00
00
48
00
00
04
160
012
128
48
848
3636
2416
432
2020
443D
4S_A
β216
00
00
00
00
00
00
412
44
48
04
820
00
812
1216
4036
4040
4032
2412
4844
4036
3NY8
_Aβ2
820
00
124
00
00
00
00
2432
00
80
2020
320
012
128
436
1236
2432
168
848
5244
403 N
Y9_A
β212
08
04
00
80
80
00
420
160
08
08
820
00
812
88
3212
4036
3220
424
4848
3244
3NYA
_Aβ2
360
00
00
00
04
00
00
2816
44
80
44
240
012
1616
2420
2440
4036
248
1644
4440
404A
MI_A
β116
812
44
48
160
208
80
824
284
816
08
1220
00
1616
1620
3636
3644
3620
248
3240
2836
4AMI
_Bβ1
244
160
08
420
012
48
08
2828
04
80
1216
324
012
1216
2452
3656
5248
3228
832
2848
484A
MJ_A
β112
00
00
00
40
00
44
020
168
48
08
2436
00
2016
2028
4028
5632
4428
168
2428
5240
4AMJ
_Bβ1
120
00
00
44
00
40
00
44
128
40
1624
160
016
2416
1636
2464
4452
2820
420
3256
644 B
VN_A
β18
00
00
04
40
04
00
028
124
48
016
1216
00
1612
1216
3620
4840
4020
288
3628
5244
4GBR
_Aβ2
80
00
00
00
80
00
00
2028
44
40
88
160
04
44
04
1240
1624
1220
012
3224
403Z
PQ_A
β18
00
00
08
80
00
120
40
164
48
016
828
00
1620
1620
3220
5244
4024
288
2424
4840
3ZPQ
_Bβ1
124
00
00
88
04
08
04
124
04
40
168
160
016
1616
1644
3244
4836
1632
824
3256
443Z
PR_A
β112
00
04
08
40
00
120
816
244
40
08
1616
00
816
1616
412
5232
3612
160
020
4440
3ZPR
_Bβ1
120
80
00
128
08
84
04
84
40
00
128
80
020
1616
1212
840
4032
1620
08
2440
362 Y
00_A
β132
820
44
812
288
128
48
1212
2016
1612
012
824
00
1616
1616
7252
6464
6440
4412
6836
6460
2Y00
_Bβ1
288
248
44
820
412
412
04
1620
84
120
812
200
016
1616
2468
5652
5248
3648
2072
3260
562Y
01_A
β112
824
120
128
288
44
120
820
2012
416
48
1616
00
1216
1224
4852
5256
6024
3212
6424
4448
2Y01
_Bβ1
1212
164
04
012
44
812
00
2016
1212
124
1212
204
016
1212
2472
4436
5232
2840
2468
2040
362Y
02_A
β120
1216
44
48
2812
128
08
2012
1612
824
48
420
00
1620
1628
8456
6860
4840
4444
8060
6472
2Y02
_Bβ1
1620
1612
412
424
2412
124
412
2020
1624
164
48
284
012
2012
2080
5264
6868
4044
5284
5248
722 Y
03_A
β112
420
00
04
164
88
44
812
812
1216
08
412
00
1212
2020
6864
6468
6444
4016
8456
5268
2Y03
_Bβ1
3212
2412
48
412
124
200
012
168
1616
164
84
40
024
1620
2476
5256
4844
3640
2872
4068
562Y
04_A
β112
412
00
820
1612
120
40
416
1620
1220
04
1216
00
1620
1624
8052
6472
7236
4420
4440
8080
2Y04
_Bβ1
124
124
08
816
816
80
00
3216
124
204
816
440
016
1612
1636
3260
5644
2828
3240
3248
443P
0G_A
β244
012
00
48
164
1612
80
128
84
828
00
08
00
016
120
4440
4440
4036
3244
6440
4448
3PDS
_Aβ2
280
44
08
416
120
48
00
124
128
120
44
40
012
84
1248
3640
4840
2832
3660
4056
523S
N6_R
β248
1224
48
48
4012
248
124
832
284
2824
04
020
00
1220
1612
6032
6036
3644
4464
6432
6460
4 LDE
_Aβ2
164
40
04
416
40
124
40
2416
08
160
812
40
012
128
1676
4836
3636
4036
4872
3652
524L
DL_A
β216
40
80
84
120
012
40
012
324
08
08
48
00
08
88
6032
5240
4820
3648
5232
4452
4LDO
_Aβ2
284
40
04
48
40
84
00
128
88
40
44
40
012
208
2036
3636
5232
2428
2880
2448
404G
PO_A
β116
00
00
40
04
40
00
00
04
40
00
416
00
2012
2424
2832
820
820
368
324
1628
4GPO
_Bβ1
200
80
00
00
00
40
00
00
40
00
44
120
024
2028
2420
2024
2416
424
828
020
162 R
4R_A
β212
2032
4428
2416
3232
3228
128
1216
164
432
020
2044
164
40
48
2020
2020
2012
436
1212
2824
2R4S
_Aβ2
1212
3636
1624
2436
2436
208
04
2020
2020
168
2416
4432
80
120
1216
1616
164
128
4412
1620
123K
J6_A
β28
1620
2024
160
2012
2016
168
1216
164
48
416
2420
2416
44
84
48
1212
124
80
48
1220
IFP
Structure
Figure 6.4a Enrichment factors at a 1% false positive rate using
all reference IFPs (columns) on all β-adren-ergic monomers for the
retrieval of for f/pAGO over physicochemically similar decoys. The
white to red and white to blue gradients as background color mark a
low to high enrichment for f/pAGO (A) and ANT/iAGO respectively.
Icons: fpAGO (red), ANT/iAGO (blue), or unvalidated ANT392 (cyan)
with no or unknown signaling preference (circle), or β-arrestin
biased ligands (triangle) (see Table 6.1).
-
139
Structure-based prediction of GPCR-ligand function - Chapter
6
6 6
PLANTS
2RH1_A
2VT4_A-D
2YCW_A
2YCW_B
2YCX_A
2YCX_B
2YCY_A
2YCY_B
2YCZ_A-B
3D4S_A
3NY8_A
3NY9_A
3NYA_A
4AMI_A
4AMI_B
4AMJ_A
4AMJ_B
4BVN_A
4GBR_A
3ZPQ_A
3ZPQ_B
3ZPQ_dock
3ZPR_A
3ZPR_B
2Y00_A
2Y00_B
2Y01_A
2Y01_B
2Y02_A
2Y02_B
2Y03_A-B
2Y04_A
2Y04_B
3P0G_A
3PDS_A
3SN6_R
4LDE_A
4LDL_A
4LDO_A
Norepiphrine
2RH1
_Aβ2
812
4840
836
3652
4448
3624
1244
128
48
520
164
204
84
84
420
824
2824
88
4424
816
02V
T4_A
β112
1248
368
3220
4436
4836
3628
3212
208
1640
04
020
40
48
44
2016
2036
288
1232
204
2812
2VT4
_Bβ1
124
4824
836
3248
4848
4040
1644
2028
88
440
120
160
44
44
412
1224
3636
128
2832
88
82V
T4_C
β112
2044
3216
2824
4440
4436
3232
3612
168
844
08
012
00
88
412
168
1636
404
420
3212
208
2VT4
_Dβ1
816
3628
1628
2432
4436
4028
1632
128
812
444
40
168
04
44
424
1220
3628
1612
2820
824
202Y
CW_A
β116
2428
3620
4024
4440
3628
3220
3616
1212
1652
412
012
04
412
44
2012
2432
2416
1224
2012
012
2YCW
_Bβ1
1216
4432
2040
2832
4444
4040
2040
1212
88
444
120
120
44
412
416
824
2424
1212
2820
1212
82 Y
CX_A
β116
1244
2420
2832
4444
4444
448
248
88
440
48
016
04
44
44
204
2424
244
416
248
2016
2YCX
_Bβ1
1212
4012
1616
2032
4440
3636
1220
412
84
364
40
160
04
48
816
816
2420
164
2416
1212
02Y
CY_A
β18
1252
4024
2828
4852
5636
4016
4016
44
452
44
012
00
412
44
128
432
2016
832
2812
128
2YCY
_Bβ1
436
4036
3224
3240
4056
2820
040
1212
88
440
80
200
44
44
416
816
3624
1612
2020
168
122Y
CZ_A
β18
1640
2812
2836
4036
3636
408
408
128
844
44
028
00
44
44
84
1632
368
412
1612
1212
2YCZ
_Bβ1
84
3620
824
2840
4436
2816
1240
816
44
360
40
120
04
40
412
420
3632
84
1624
820
163D
4S_A
β220
1260
2416
2832
4448
5236
3232
488
124
448
48
04
00
48
44
2012
424
244
844
284
164
3NY8
_Aβ2
836
3632
3232
2036
3236
3224
1232
812
48
324
84
200
08
44
412
412
424
44
3228
88
203N
Y9_A
β216
1644
2020
3632
3228
4428
2020
284
84
444
012
416
04
44
48
88
2024
204
440
244
168
3NYA
_Aβ2
2012
3612
816
2040
2836
3232
2028
08
88
320
80
40
04
44
44
420
2020
40
2832
128
204A
MI_A
β124
2432
2824
2816
3236
3224
2016
3212
1212
428
48
416
416
412
84
84
1224
164
828
128
812
4AMI
_Bβ1
1632
2428
2820
2428
2832
3232
1620
1216
84
284
124
2012
124
84
416
88
2020
48
2420
84
124A
MJ_A
β124
1648
3220
2420
3636
4432
2420
368
2012
840
44
012
00
88
84
44
2020
408
1220
164
208
4AMJ
_Bβ1
2824
4428
1632
2036
4444
3636
1644
88
84
524
204
120
84
48
48
1236
4844
1212
1612
1212
84B
VN_A
β112
432
2412
1628
3624
3236
2420
248
168
436
08
48
00
44
44
124
3232
324
1236
324
824
4GBR
_Aβ2
816
3616
1236
2028
2036
3212
832
816
84
2016
120
120
04
44
08
016
2024
80
128
80
43Z
PQ_A
β112
2028
3228
4020
3236
2840
1616
244
1212
424
84
48
04
412
1216
1216
2028
3612
012
248
84
3ZPQ
_Bβ1
1628
3632
2432
2036
3236
4824
2020
1616
88
204
84
40
04
124
1220
824
1220
88
2020
816
243Z
PR_A
β18
1632
3220
4020
2824
3228
2420
2812
124
432
88
44
00
48
44
80
820
244
84
44
44
3ZPR
_Bβ1
812
3628
1236
1620
2832
4024
812
412
84
164
124
80
04
44
48
012
1616
40
04
88
02Y
00_A
β112
2816
2824
1612
2420
1616
84
1612
1612
1212
08
412
04
812
48
168
824
2012
420
208
128
2Y00
_Bβ1
1612
2424
1212
424
2820
168
1220
120
124
204
40
00
44
44
420
1212
2812
812
2416
84
42Y
01_A
β18
1616
2412
1212
2424
2820
1216
248
412
824
44
04
416
48
44
128
1216
2012
424
168
84
2Y01
_Bβ1
1212
168
44
04
812
84
128
1620
48
40
00
40
44
84
412
88
2012
84
88
44
42Y
02_A
β18
3216
2820
164
1224
124
812
812
1212
420
164
04
04
48
48
2412
1620
1612
1228
328
48
2Y02
_Bβ1
1224
2828
1624
820
2420
2016
820
88
1212
124
00
84
84
44
412
412
2020
44
2020
44
82Y
03_A
β112
2024
2428
168
2428
2020
88
164
412
1224
80
04
44
44
44
1616
2024
248
812
2812
48
2Y03
_Bβ1
1228
2832
1616
816
2020
2012
1216
44
88
244
48
80
04
84
424
128
1216
84
2020
820
122Y
04_A
β18
1224
288
3220
2836
2424
3212
2020
164
428
84
04
44
48
44
128
2032
368
824
208
1216
2Y04
_Bβ1
1216
2424
1224
1624
3220
2420
1220
88
124
284
00
88
124
48
44
1216
3232
124
2012
80
43P
0G_A
β212
48
04
00
40
124
44
120
04
412
00
124
00
04
44
124
44
44
416
168
84
3PDS
_Aβ2
80
120
00
1212
1212
1616
812
84
44
160
00
00
04
40
412
816
1624
1212
2820
88
123S
N6_R
β212
820
40
124
1224
2012
1616
48
012
1620
08
08
44
44
44
44
416
88
412
84
48
4LDE
_Aβ2
124
2420
88
1228
2412
2428
1212
44
88
240
80
80
00
44
416
40
2016
84
2820
412
44L
DL_A
β212
824
208
124
2420
2012
84
88
120
1228
04
88
48
08
04
164
44
84
416
124
04
4LDO
_Aβ2
120
164
08
812
812
812
128
88
48
40
44
80
04
44
44
84
124
84
1616
412
04G
PO_A
β10
00
00
04
40
00
00
04
04
40
00
00
00
00
00
00
00
04
04
40
00
4GPO
_Bβ1
40
40
00
44
44
00
04
00
00
00
00
00
04
04
40
04
44
04
00
00
02R
4R_A
β28
2024
2824
2012
2420
2420
128
816
164
412
00
016
816
00
00
1616
168
124
44
416
04
2R4S
_Aβ2
1212
2832
2824
1228
3228
2812
1616
2424
44
200
124
812
164
84
432
3228
84
44
284
324
03K
J6_A
β212
424
2420
44
2412
2424
208
2416
164
416
08
208
04
88
1212
88
1624
244
1212
1212
44
03
58
1013
1518
2023
2528
3033
3538
4043
4548
5053
5558
6063
6568
7073
7578
8083
8588
9093
9598
100
EF 1
% (A
NT/
iAG
O)
IFP
Structure
Figure 6.4b Enrichment factors at a 1% false positive rate using
all reference IFPs (columns) on all β-adren-ergic monomers for the
retrieval of for ANT/iAGO over physicochemically similar decoys.
The white to red and white to blue gradients as background color
mark a low to high enrichment for f/pAGO (A) and ANT/iAGO
respectively. Icons: fpAGO (red), ANT/iAGO (blue), or unvalidated
ANT392 (cyan) with no or unknown signaling preference (circle), or
β-arrestin biased ligands (triangle) (see Table 6.1).
-
140
Chapter 6 - Structure-based prediction of GPCR-ligand
function
6 6
In Figure 6.4a we see that also ANT/iAGO structures can be used
to efficiently retrieve f/pAGO ligands when using the right
reference IFP. This is surprising, as this contrasts the results of
the previous VS studies in which the rotamers of the TM5 serines
and/or whole TM5 had to be rotated for the efficient retrieval of
f/pAGO ligands.162, 276, 404, 412 Interestingly, the isoproterenol
IFP (PDB-code 2Y03) allows efficient retrieval of f/pAGO ligands in
all ANT/iAGO structures with an average EF1% of 47. Ranking docking
poses in f/pAGO structures with an ANT/iAGO reference IFP on the
other hand does not yield high enrichments of ANT/iAGO ligands
(with an average ANT/iAGO EF1% of 13). The highest average EF1%
(21) was obtained by the cyanopindolol IFP (2YCY_B) in the f/pAGO
structures. An unanticipated finding was that when pairing specific
f/pAGO IFPs with ANT/iAGO structures also high EF1% values were
obtained for the selective retrieval of ANT/iAGO. When using the
IFPs of salbutamol (pAGO, PDB-code 2Y04) an average ANT/iAGO EF1%
of 27 was obtained (Figure 6.4a/b). In short, these statistics
emphasize that not only the choice of structure but also the choice
of reference IFP has a high impact on the resulting enrichment
factors and, more inter-estingly, on the ligand selectivity of the
results.
6.2.3 Selective structure-based VS for ligands with the desired
functional effect
Based on the highest EF1%s (Figure 6.3) we selected one f/pAGO
and one ANT/iAGO ligand-pro-tein complex for a more detailed
investigation. From all ANT/iAGO-bound complexes the
cy-anopindolol-bound (iAGO) β1R structure 2VT4 (both chains A and
B) obtained the highest ANT/iAGO EF1% value (48) and a f/pAGO EF1%
value of 0. As both chains gave similar virtual screening
enrichments (Figure 6.4b), we randomly selected chain A of 2VT4
(Figure 6.5b) for further in-vestigation. From all f/pAGO complexes
chain A of the carmoterol (fAGO) bound structure β1R structure 2Y02
(Figure 6.5a) obtained the highest f/pAGO EF1% value (84) and an
ANT/iAGO EF1% value of 24. In figure 6.3b and 6.3c the individual
ROC-plots are displayed for the retrieval of ANT/iAGO and f/pAGO
over decoys by IFP-similarity scoring of docking poses generated in
the 2VT4 and 2Y02 complexes using the reference IFPs of the
corresponding co-crystallized ligands. Figures 6.3b/c show the high
early and overall enrichment for ligands with the same functional
effect of the co-crystallized ligand and much lower early and
overall enrichment for ligands with a different functional
effect.By rescoring all the docking poses of the compounds that
were docked in all 48 structures with the reference IFPs from the
two selected structures the impact of the reference IFP selection
was investigated (Figure 6.5d/e). For the selected f/pAGO IFP
(2Y02_A), 93% of all ligand-pro-tein complexes (excluding the 2Y02
structures) yielded an equal or higher f/pAGO EF1% than when using
their own IFP, in 64% of all cases a lower ANT/iAGO EF1% and for
almost all ligand-pro-tein complexes a higher f/pAGO EF1% than
ANT/iAGO EF1% (Figure 6.5d, 89% of all cases). When using the
selected ANT/iAGO IFP (2VT4_A) instead of the IFP of the
co-crystallized ligand to rescore all docked compounds in the
ligand-protein complexes a higher ANT/iAGO EF1% (80% of all cases),
a lower f/pAGO EF1% (51% of all cases) and a higher ANT/iAGO EF1%
than f/pAGO EF1% (82% of all cases) was obtained (Figure 6.5e). In
two cases the use of the selected ANT/iAGO IFP resulted in an even
higher ANT/iAGO EF1% than with its own structure, namely in
combination with the structure of cyanopindolol-bound β1R (EF1% 52,
2YCY_A) and timolol-bound β1R (EF1% 60, 3D4S). This is an increase
of 4 and 24 in ANT/iAGO EF1% for the 2YCY_A and 3D4S structures
respectively compared to the enrichment obtained with their native
IFP. This was not observed for the selected f/pAGO IFP as the
combination with its own structure (2Y02_A) resulted in
-
141
Structure-based prediction of GPCR-ligand function - Chapter
6
6 6
the highest f/pAGO EF1% (84). The results for one of the
Fab-complexes (2R4S) were somewhat surprising as the f/pAGO EF1% in
combination with the selected ANT/iAGO IFP were higher than with
the selected f/pAGO IFP (36 and 16 respectively). Combining this
with the fact that this structure is missing 23 of the binding
pocket residues (vide supra), it was even more surprising that
these enrichment factors can be obtained when docking in a
structure with only half of the binding pocket. This does, however,
match the success of previously published VS studies using homology
models without ECL2160 and low-resolution homology models.122
Overall it is clear that the selection of the reference IFP is
essential and can influence the enrichment as well as selec-tivity
of the outcome of a VS study regardless of which structure was
used.In order to separately quantify the impact of the structure on
the enrichment values we studied the effect of rescoring all
docking poses in the selected two structures with all 39 reference
IFPs (Figure 6.5f/g). The results of rescoring the docking poses in
the 2 selected structures with all reference IFPs show similar
trends as rescoring with the reference IFP of the selected
complexes (Figure 5d/e). For the selected f/pAGO structure (2Y02_A,
Figure 6.5f) we see a shift in the results towards the f/pAGO and
the opposite for when using the selected structure (2VT4_A, Figure
6.5g). The observed effect is, however, less pronounced. When we
compare the obtained enrichment factors they are on average 38%
lower for the selected structures than when using their IFP to
rescore the docking poses. Moreover, with the right IFP the
selectivity of a crystal structure the can be shifted towards
ligands with another functional effect than the co-crystallized
ligand (Figure 6.5d/e), the oppo-site (shifting the selectivity by
using the right structure) is less frequent and less prominently
observed. The selection of the crystal structure does, however,
clearly influence the extent to which the IFP to able to shift this
selectivity and also the final enrichment values. Combined these
results indicate that the correct IFP is more crucial for obtaining
selectivity as well as high enrichments than the structure, but
they also indicate that obtaining the best results are obtained by
finding the right combination of IFP and structure.
-
142
Chapter 6 - Structure-based prediction of GPCR-ligand
function
6 6
0 10 20 30 40 50 60
020
4060
8010
0
EF1% ANT/iAGO
EF1%
f/pA
GO
●●
●●●●
●
●●●
●
●
●
●●
●
●
● ●
●
●
●●
●
●●●
p/fAGOANT/iAGOANT (Not validated)
0 10 20 30 40 50 60
020
4060
8010
0
EF1% ANT/iAGO
EF1%
f/pA
GO
●
●●●●●
●●
●
●●●● ●●
●
● ●● ●●
●
●●
●●
● ●●
●
●●
●
●●●●●
p/fAGOANT/iAGOANT (Not validated)Fab−complexLigand−free
0 10 20 30 40 50 60
020
4060
8010
0
EF1% ANT/iAGO
EF1%
f/pA
GO
●
●●
●
●
●●
●●● ●
●●
●●
●●
●
●
●
●
●
●
●●
●
●
●
●
● ●●
●
●●●●●
p/fAGOANT/iAGOANT (Not validated)Fab−complexLigand−free
0 10 20 30 40 50 60
020
4060
8010
0
EF1% ANT/iAGO
EF1%
f/pA
GO
● ●●
● ●● ●
● ●●
●
● ● ●●
●●
●
●●
●
●●
●
●
●●●
p/fAGOANT/iAGOANT (Not validated)
e d
c
N6.55%
D3.32%
N7.39%
S5.46%
S5.42%S5.43%
N6.55%
D3.32%
N7.39%
S5.46%
S5.42%S5.43%
Agonist: 2Y02 (chain A) Antagonist:
2VT4 (chain A) a
2VT4 IFP in all structures 2Y02_A
IFP in all structures
All IFPs in 2VT4_A structure All
IFPs in 2Y02_A structure
4GPO_A
4GPO_A
4LDE
2YCZ
Ligand StructureCarmoterol 2Y02_chainAProcaterol 2Y02_chainA
Cyanopindolol 2VT4_chainAAlprenolol 2VT4_chainA
N7.39D3.32 S5.42 S5.43 S5.46 N6.55
b
g f
Figure 6.5 Analysis of VS runs using the IFPs (d-e) and
structures (f-g) of the most f/p AGO selective (2Y02_A) and
ANT/iAGO selective (2VT4_A) protein-ligand complexes (see Figure
3A) show that both the reference IFP and protein structure
determine enrichment and functional selectivity of the VS study.
The
-
143
Structure-based prediction of GPCR-ligand function - Chapter
6
6 6
binding modes of the selected f/pAGO (a, 2Y02_A) and selected
ANT/iAGO (b, 2VT4_A) protein-ligand complex, a docked ligand with
the same efficacy class (green carbon atoms) and their
corresponding IFPs (c) for the displayed residues. EF1% results for
agonists and antagonist versus the decoys when using the reference
IFP of the selected f/pAGO (d) and selected ANT/iAGO (e) for
scoring docked compounds in all 48 protein structures. EF1% results
for f/pAGO and ANT/iAGO versus the decoys when using the 38 unique
reference IFPs of all ligand-protein complexes for rescoring the
docked compounds in the 2Y02_A (f) and 2VT4_A (g) structure. Icons:
fpAGO (red), ANT/iAGO (blue), or unvalidated ANT392 (cyan) with no
or un-known signaling preference (circle), or β-arrestin biased
ligands (triangle) (see Table 6.1).
6.2.4 Selective structure-based virtual screening for
partial/full agonists with a com-putationally predicted
norepinephrine interaction fingerprint
From the f/pAGO and ANT/iAGO retrieval rates in all structures
(Figure 6.4a/b), we can see that in particular the IFP of
epinephrine (PDB-codes 4LDO) is able to selectively retrieve f/pAGO
ligands in all structures (with an average f/pAGO and ANT/iAGO EF1%
of 43.6 and 9.4 respectively). This small endogenous agonist makes
all key interactions with D1133.32, N3127.39 and S2035.42 in β2R
(Figure 6.2). Based on this observation we hypothesized that the
IFP of an even smaller agonist, that could also make all these
interactions, should also be highly effective for the retrieval of
f/pAGO over ANT/iAGO in all structures. As the crystal structure of
norepinephrine bound to β1R or β2R has not yet been resolved we
docked norepinephrine and used the docking pose (Figure 6.6b) with
the highest IFP-score (compared to the IFP of isoproterenol) to
obtain the reference IFP for norepinephrine in β1R (Figure
6.2).Indeed, when using this computationally predicted IFP to
rescore all docked compounds in all structures we were able to
retrieve f/pAGO ligands with a high efficiency (the average f/pAGO
and ANT/iAGO EF1% were 46 and 8 respectively, Figure 6.4a and 6.6).
On average the obtained f/pAGO EF1% was almost twice as high
compared to the 2Y02_A IFP (the IFP from the highest-scoring f/pAGO
complex). It should be noted that this is mostly due to higher
f/pAGO EF1% in the ANT/iAGO structures. At the same time a 20%
lower ANT/iAGO EF1% was obtained compared to the 2Y02_A IFP.
Moreover, the ANT/iAGO structure that obtained the highest f/pAGO
EF1% with this IFP was a cyanopindolol-bound β1R structure (2YCW_A,
Figure 6.6a), which coincides with the slight contraction of the
pocket and the low f/pAGO RMSD as previously de-termined (Figure
6.1c). Interestingly, also in combination with the selected
ANT/iAGO structure (cyanopindolol-bound 2VT4_A) the norepinephrine
IFP was able to obtain a high and selective retrieval rate for
f/pAGO (EF1% of 56 and 12 for f/pAGO and ANT/iAGO respectively,
Figure 6.6c). The norepinephrine IFP combined with the 2Y02_A
structure resulted in a slightly reduced f/pAGO EF1%, but at the
same time reduced the ANT/iAGO EF1% thereby making the results much
more selective (Figure 6.6d).By using the IFP from a small compound
that makes all key interactions these key interactions are highly
emphasized, yielding more selective results. Whenever a docking
pose of a compound misses one or more of the key interactions
present in the reference IFP this immediately results in a
relatively low IFP score.
-
144
Chapter 6 - Structure-based prediction of GPCR-ligand
function
6 6
0 10 20 30 40 50 60
020
4060
8010
0
EF1% ANT/iAGO
EF1%
f/pA
GO
●
●
●●
●
●
● ●● ●
●● ●
●●
● ●●●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●●●●
p/fAGOANT/iAGOANT (Not validated)Fab−complexLigand−free
0.1
0.5
1.0
5.0
10.0
50.0
100.0
020
4060
80100
2Y02_A
f/pAGOANT/iAGORandom
0.1
0.5
1.0
5.0
10.0
50.0
100.0
020
4060
80100
2VT4_A
f/pAGOANT/iAGORandom
b!
c!
1.0
1.2
1.4
1.6
1.8
2.0
020
4060
8010
0
FP−rate (%)
TP−r
ate
(%)
1.0
1.2
1.4
1.6
1.8
2.0
020
4060
8010
0
FP−rate (%)
TP−r
ate
(%)
a!
d!
1.0
1.2
1.4
1.6
1.8
2.0
020
4060
8010
0
FP−rate (%)
TP−r
ate
(%)
1.0
1.2
1.4
1.6
1.8
2.0
020
4060
8010
0
FP−rate (%)
TP−r
ate
(%)
N6.55%
D3.32%
N7.39%
S5.46%
S5.42%
S5.43%
2Y03_A
4LDE 4LDL
4LDO 4GPO_A
3KJ6_A
Figure 6.6 EF1% results for agonists and antagonists (a) when
using the reference IFP of docked norepi-nephrine (Figure 6.2) in
2Y03_A (b). The individual ROC-plots for the retrieval of agonists
(red curve) and antagonists (blue curve) in the selected ANT/iAGO
structure 2VT4_A (c) and f/pAGO structure 2Y02_A (d). Icons: fpAGO
(red), ANT/iAGO (blue), or unvalidated ANT392 (cyan) with no or
unknown signaling preference (circle), or β-arrestin biased ligands
(triangle) (see Table 6.1).
6.3 Biased agonists
G-protein-coupled receptors can signal via G-proteins as well as
β-arrestins.441 So-called biased ligands can induce a preference of
the receptor for one pathway over the other.407 An in-depth
literature search learned that for 7 of the 19 co-crystallized
ligands there is a record of biased signaling (Table 6.1). Amongst
them are bucindolol and carvedilol, two β-blockers that promote
signaling via the β-arrestin pathway.105 One should note that the
signaling bias of compounds is not always consistently observed and
reported (e.g. for cyanopindolol205, 430, 434, 435, carvedilol105,
430, 432, 435, dobutamine430, 436) as the effect seems to be
dependent on assay type and condi-tions436, receptor-activation
state437, and receptor subtype430 (β1 versus β2). For another 7 of
the 19 co-crystallized compounds we consistently found one or more
publications reporting that ligand signaling is not biased (Table
6.1). It should be noted that the 7 biased ligands are all, with
the exception of alprenolol, relatively large compounds (Figure
6.1, Table 6.1) that stretch from the major pocket (between TM3, 4,
5, and 6) into the minor pocket389 (between TM1, 2, 3, and 7).To
identify potential key interactions for β-arrestin biased ligand,
we grouped all IFPs (Figure 6.2) of both the reported 7 biased and
the reported 7 unbiased ligands. Subsequently, we calculated the
difference in abundance for each interaction with each pocket
residue (Table S6.8). This highlighted four interactions that had a
high abundance (≥50%) and were unique for the biased ligands. Three
of them are hydrophobic contacts, namely with L101/H932.64,
D200/19245.51, and V326/I3097.36 and the fourth is an aromatic
stacking interaction (face-to-face or face-to-edge) with
W117/1093.28. Interestingly, this last interaction was present for
all biased ligands, except for alprenolol. This could indicate that
the signaling bias for these ligands is (partially) due to the
aromatic stacking with this tryptophan residue. Moreover,
isoproterenol is a full agonist with no signaling bias (it is even
used as reference ligand442), but when a hydroxybenzyl moiety is
at-tached the ligand (hydroxybenzylisoproterenol, 4LDL) makes an
aromatic interaction with W3.28
-
145
Structure-based prediction of GPCR-ligand function - Chapter
6
6 6
and induces a β-arrestin signaling bias.395 This is also in line
with mutation studies of the homol-ogous W3.28 in the muscarinic M2
receptor (also a member of the aminergic-GPCR family72) that
changed the signaling preference of the receptor when stimulated by
specific ligands.443 Also the contacts with L/H2.64, D45.51, a