Molecular mechanisms of biased and probe-dependent signaling at CXC-motif chemokine receptor CXCR3 induced by negative allosteric modulators

Article abstract of DOI:10.1124/mol.117.110296

Our recent explorations of allosteric modulators with improved properties resulted in the identification of two biased negative allosteric modulators, BD103 (N-1-{[3-(4-ethoxyphenyl)-4-oxo-3,4-dihydropyrido[2,3-d]pyrimi-din2yl]ethyl}-4-(4-fluorobutoxy)-N-

Full text of DOI:10.1124/mol.117.110296

Molecular Pharmacology Fast Forward. Published on January 17, 2018 as DOI: 10.1124/mol.117.110296
This article has not been copyedited and formatted. The final version may differ from this version.
MOL #110296
Title:
Molecular mechanisms of biased and probe-dependent signaling at CXCR3 induced
by negative allosteric modulators
Authors:
Regine Brox, Lampros Milanos, Noureldin Saleh, Paul Baumeister, Armin Buschauer, Dagmar
Hofmann, Markus R. Heinrich, Timothy Clark, Nuska Tschammer
Department of Chemistry and Pharmacy, Medicinal Chemistry, Emil Fischer Center, Friedrich
Alexander University, Schuhstrasse 19, 91052 Erlangen, Germany (R.B., D.H., M.R.H., N.T.)
Computer Chemistry Center, Friedrich Alexander University, Nägelsbachstrasse 25, 91052 Erlangen,
Germany (L.M., N.S., T.C.)
Institute of Pharmacy, University of Regensburg, Universitätsstraße 31, 93040 Regensburg, Germany
(P.B., A.B.)
Current Address: NanoTemper Technologies GmbH, Floessergasse 4, 81369 München, Germany (N.T.).
1

Molecular Pharmacology Fast Forward. Published on January 17, 2018 as DOI: 10.1124/mol.117.110296
This article has not been copyedited and formatted. The final version may differ from this version.
MOL #110296
Running Title Page
Running Title: Biased and probe dependent signaling at CXCR3
Corresponding author: Nuska Tschammer, PhD.
Email: nuska.tschammer@nanotempertech.com; Address: Floessergasse 4, 81369 München, Germany;
Telephone: +49 (0) 89 45 22 895 802
Pages number: 34
Figures number: 8
Tables number: 0
References: 59
Abstract words count: 246
Introduction words count: 734
Discussion words count: 1475
Abbreviations:
AMs: Allosteric modulators
BRET: Bioluminescence resonance energy transfer
cAMP: Cyclic adenosine monophosphate
cDNA: complementary deoxyribonucleic acid
CXCL9-11: CXC-motif chemokine 9-11
CXCR3: CXC-motif chemokine receptor 3
ELISA: Enzyme-linked immunosorbent assay
GPCR: G protein-coupled receptor
INF-γ: Interferon-γ
MD: Molecular dynamics
OPLS: Optimized Potentials for Liquid Simulations
PCR: Polymerase chain reaction
pKb: Equilibrium dissociation constant
Rluc: Renilla luciferase
Th1: T helper cells
TM: Transmembrane
2

Molecular Pharmacology Fast Forward. Published on January 17, 2018 as DOI: 10.1124/mol.117.110296
This article has not been copyedited and formatted. The final version may differ from this version.
MOL #110296
α: cooperativity factor
3

Molecular Pharmacology Fast Forward. Published on January 17, 2018 as DOI: 10.1124/mol.117.110296
This article has not been copyedited and formatted. The final version may differ from this version.
MOL #110296
Abstract
Our recent explorations of allosteric modulators (AMs) with improved properties resulted in the
identification of two biased negative AMs, N-1-{[3-(4-Ethoxyphenyl)-4-oxo-3,4-dihydropyrido[2,3-
d]pyrimi-din2yl]ethyl}-4-(4-fluorobutoxy)-N-[(1-methylpiperidin-4-yl)me-thyl}]butanamide (BD103)
and {5-[(N-{1-[3-(4-ethoxyphenyl)-4-oxo-3,4-dihydropyrido[2,3-d]pyrimidin-2-yl]ethyl-2-[4-fluoro-3-
(trifluoromethyl)phenyl]acetamido)methyl]-2-fluorophenyl}boronic acid (BD064), that exhibited probe-
dependent inhibition of the chemokine receptor CXCR3 signaling. With the intention to elucidate the
structural mechanisms underlying their selectivity and probe-dependency, we used site-directed
mutagenesis combined with homology modeling and docking to identify amino acids of CXCR3 that
contribute to modulator binding, signaling and transmission of cooperativity. With the use of allosteric
radioligand RAMX3 we identified F1313.32 and Y3087.43 that contribute specifically to the binding pocket
of BD064, whereas D1864.60 solely participate in the stabilization of binding conformation of BD103.
The influence of mutations on the ability of negative allosteric modulators to inhibit the chemokine-
mediated activation (CXCL11 and CXCL10) was assessed with BRET based cAMP and β-arrestins
recruitment assay. Obtained data revealed complex molecular mechanisms governing biased and probe-
dependent signaling at CXCR3. In particular, F1313.32, S3047.39 and Y3087.43 emerged as key residues
for the compounds to modulate the chemokine response. Notably, D1864.60, W2686.48 and S3047.39 turned
out to play a role in signal pathway selectivity of CXCL10 as mutations of these residues led to a G
protein active but β-arrestin inactive conformation. These diverse effects of mutations suggest the
existence of ligand- and pathway-specific receptor conformations and give new insights in the
sophisticated signaling machinery between allosteric ligands, chemokines and their receptors, which can
provide a powerful platform for the development of new allosteric drugs with improved pharmacological
properties.
4

Molecular Pharmacology Fast Forward. Published on January 17, 2018 as DOI: 10.1124/mol.117.110296
This article has not been copyedited and formatted. The final version may differ from this version.
MOL #110296
Introduction
Biased allosteric modulation of G protein-coupled receptors represents an attractive approach to drug
design (Khoury et al., 2014; Leach et al., 2007c). While allosteric ligands act at sites that are
topographically distinct from the orthosteric binding site, they display unique properties that are reflected
in greater receptor subtype selectivity, permissivity, saturability of effect, probe-dependence and biased
signaling (Christopoulos, 2002; Kenakin, 2010, 2012). The latter has attracted special attention in the
past decade as some ligands have emerged that stabilize diverse active receptor states, which resulted in
a selective biological response (Kenakin, 2007). We recently discovered biased allosteric agonists of the
chemokine receptor CXCR3 that either lead to β-arrestin 2 recruitment and internalization or solely
activate G proteins (Milanos et al., 2016a). Another interesting phenomenon with substantial implications
for drug discovery is related to the permissive nature of allosteric agents and is known as probe-
dependence. In this case, the allosteric effects can vary depending on the co-bound orthosteric ligand
(“probe”) and are manifested by a change in cooperativity (Kenakin, 2005; Leach et al., 2007b). For
example, the well-characterized allosteric anti-HIV modulator of the chemokine CCR5 receptor,
aplaviroc, can selectively block the binding of [125I]-MIP1α (CCL3) while the binding of [125I]-RANTES
is unaffected (Watson et al., 2005).
The chemokine system itself is an excellent example for studying functional selectivity. The promiscuity
of chemokines with multiple shared ligands and receptors is often wrongly referred to as redundancy.
However, several studies have shown that the binding of different chemokines to a single chemokine
receptor can differ in the biological response resulting in functional selectivity (Corbisier et al., 2015;
Leach et al., 2007a; Rajagopal et al., 2013; Tian et al., 2004). This is also the case for the chemokine
receptor CXCR3, which is activated by gamma-inducible chemokines CXCL11, CXCL10 and CXCL9
(Groom and Luster, 2011). Dysfunctions of the CXCR3 signaling network are linked to a myriad of
inflammatory diseases, such as rheumatoid arthritis (Lacotte et al., 2009), multiple sclerosis (Sorensen
5

Molecular Pharmacology Fast Forward. Published on January 17, 2018 as DOI: 10.1124/mol.117.110296
This article has not been copyedited and formatted. The final version may differ from this version.
MOL #110296
et al., 1999), cancer (Kawada et al., 2007; Ma et al., 2009), atherosclerosis (Mach et al., 1999) and
allograft rejection (Cox et al., 2001). Therefore, intensive efforts have been devoted to developing
allosteric modulators as therapeutics with greater potential for treating such diseases through enhanced
selectivity (Bernat et al., 2014; Scholten et al., 2012; Wijtmans et al., 2008). Despite the strong focus in
small-molecule ligands for CXCR3, the lack of structural data on chemokine binding and activation
makes it challenging to understand the precise mechanisms of the interactions of allosteric modulators
with the receptors. Nonetheless, some crystal structures of chemokine receptors including CXCR4 in
complex with small-molecule antagonists IT1t and CVX15 (Wu et al., 2010) or the viral chemokine
vMIP-II (Qin et al., 2015), afforded some structural insights in the interaction between ligands. Three
GPCR binding pockets have been identified; 1T1t was detected to interact with residues in the minor
binding pocket, CVX15 with residues in the major pocket and vMIP-II extending to both subpockets
(Nicholls et al., 2008; Salchow et al., 2010). Site-directed mutagenesis has also been used to dissect the
binding mode of novel allosteric modulators of CXCR3. Scholten et al. showed that NBI-74330 is mainly
anchored into the minor pocket, whereas VUF11211 extends from the minor pocket in the major pocket
(Scholten et al., 2014). In addition, Nedjai et al. described the binding site of the CXCR3 antagonist
VUF10085 in the minor pocket with Asp1122.63, F1313.32 and Y3087.43 forming direct contact points, but
distinct from the broad-spectrum antagonist TAK-779, which is assumed to bind outside the minor pocket
(Nedjai et al., 2015).
Our recent explorations of allosteric modulators with improved properties resulted in the identification
of two biased negative allosteric modulators that exhibited probe-dependent inhibition of CXCR3
signaling (Bernat et al., 2015). Such probe-dependent allostery may serve to fine tune the chemokine
response in the seemingly redundant area of multiple chemokine agonists for receptors. In the current
study, we aimed to elucidate the binding mode of our novel allosteric modulators and to identify amino-
acid residues that steer the bias and probe-dependent mechanisms using site-directed mutagenesis and
6

Molecular Pharmacology Fast Forward. Published on January 17, 2018 as DOI: 10.1124/mol.117.110296
This article has not been copyedited and formatted. The final version may differ from this version.
MOL #110296
computational studies. By doing so, we provide novel insights into the complex interactions between
allosteric ligands, chemokines and their receptor, which open a powerful platform for further
development and rational design of allosteric modulators with improved bias and probe-dependency.
Moreover, we report for the first-time amino-acid residues that are specifically involved in CXCL10-
mediated β-arrestin recruitment but not in G protein activation.
7

Molecular Pharmacology Fast Forward. Published on January 17, 2018 as DOI: 10.1124/mol.117.110296
This article has not been copyedited and formatted. The final version may differ from this version.
MOL #110296
Materials and Methods
CXCR3 Ligands: The synthesis of cRAMX3 was previously described (Bernat et al., 2014). BD064
[compound 14 in the study by Bernat et al. (2014) (Bernat et al., 2014)] and BD103 [compound 1b in the
study by Bernat et al. (2015) (Bernat et al., 2015)] were synthesized according to the previously described
procedures (Bernat et al., 2014; Bernat et al., 2015). Briefly, for the synthesis of the boronic acid
derivative BD064 the primary amine (2-(1-aminoethyl)-3-(4-ethoxyphenyl)pyrido[2,3-d]pyrimidin-
4(3H)-one) was reductively alkylated using the protocol for reductive amination developed by Molander
et al. (Molander and Cooper, 2008). The required trifluoroborate-substituted aromatic aldehyde was
prepared from commercially available boronic acid ((2-flouro-5-formylphenyl)boronic acid) by
treatment with potassium hydrogen difluoride in methanol. The resulting secondary amine was
subsequently acylated by activated 2-[4-fluoro-(3-trifluoromethyl)phenyl]acetic acid. The
Trifluoroborate moiety was converted to boronic acid by the treatment with trimethylsilyl chloride in
aqueous acetonitrile
The synthesis of BD103 was similar. Also the primary amine (2-(1-aminoethyl)-3-(4-
ethoxyphenyl)pyrido[2,3-d]pyrimidin-4(3H)-one) was reductively alkylated using corresponding
substituted benzaldehyde and triacetroxyborohydride. The resulting secondary amine was subsequently
acylated by 4-(4-fluorobutoxy)butanoic acid, which was synthesized according to the previously
published procedure (Bernat et al., 2014).
Modeling and docking of BD064 and BD103: The human CXCR3 homology model used in our
docking studies was that reported previously (Milanos et al., 2016b). Ligand structures were optimized
using OPLS force field (Harder et al., 2016) in the LigPrep tools and the relevant protonation states at
physiological pH were determined using Epik. The ligands were docked using a 25-Å box centered on
cRAMX3 and default settings in GLIDE standard precision (SP) (Halgren et al., 2004).
8

Molecular Pharmacology Fast Forward. Published on January 17, 2018 as DOI: 10.1124/mol.117.110296
This article has not been copyedited and formatted. The final version may differ from this version.
MOL #110296
Cell Culture and Transfection: Human embryonic kidney (HEK) 293 T cells were grown at 37°C and
5% CO2 in DMEM/F-12 (Gibco, Darmstadt, Germany) supplemented with 10% fetal bovine serum
(FBS), 2 mM L-glutamine, 1% penicillin-streptomycin. Cells were transiently transfected with 5 µg of
pcDNA 3.1 containing either FLAG-CXCR3 wild type or mutants using TransIT-293 transfection
reagent (Mirus Bio, Madison, WI, USA)). Cells were harvested after 48 h for membrane preparations.
For the cAMP biosensor assay, HEK293T cells were transiently transfected with 1 µg/ml of wild type or
mutant CXCR3 and 2 µg/ml of DNA encoding for the cAMP biosensor CAMYEL (cAMP sensor using
YFP-Epac-RLuc, American Type Celture Collection, Manassas, VA, USA) (Jiang et al., 2007). For β-
arrestin recruitment experiments, HEK293T cells were transfected with cDNA coding for β-arrestin1- or
2-Rluc and CXCR3-mVenus (either wild type or mutant) and linear polyethyleneimine (linear PEI, 25
kDA, Polysciences Inc., Warrington, PA, USA) in a 1:3 ratio. Cells were plated in 96-well F-Bottom
white tissue culture plates (Greiner-Bio-One GmbH, Frickenhausen, Germany) at a density of 3 × 105.
DNA Constructs and Site-Directed Mutagenesis: The wild type CXCR3 was a kind gift of Prof. Dr.
R. Leurs (VU Amsterdam, Netherlands). cDNA encoding for the BRET based cyclic AMP (cAMP) was
purchased from American Type Culture Collection (Manassas, VA, USA) (pcDNA3.1-(L)-His-
CAMYEL #ATCC-MBA-277). The pcDNA3.1-β-arrestin1/2-Renilla luciferase (Rluc) constructs were
kind gifts from Dr. M. Bouvier (University of Montréal, QC, Canada). The FLAG-CXCR3-mVenus
construct was generated by substituting the stop codon of CXCR3 and fusing them to yellow fluorescent
protein (mVenus). We generated the point mutations by polymerase chain reaction using oligonucleotide
primers with desired amino acid change and either pc DNA 3.1 CXCR3 with a FLAG tag encoded at the
N terminus or pc DNA 3.1 FLAG-CXCR3 with a C-terminal mVenus as a template. Amplification
conditions were 10 s at 98 °C, 15 s at 68 °C and 30 s at 72 °C for 30 cycles. The PCR amplified fragment
was digested with BmtI and XhoI. All mutations were verified by restriction endonuclease mapping and
DNA sequence analysis (LGC Genomics, Berlin, Germany).
9

Molecular Pharmacology Fast Forward. Published on January 17, 2018 as DOI: 10.1124/mol.117.110296
This article has not been copyedited and formatted. The final version may differ from this version.
MOL #110296
Membrane Preparation: At 100% confluency cells were washed with PBS twice, treated with Tris
EDTA buffer (10 mM Tris, 0.5 mM EDTA, 5 mM KCl, 140 mM NaCl, pH 7.4), and harvested using a
cell scraper. Cells were pelleted at 1100×g for 8 min at 4 °C, resuspended in Tris-EDTA-MgCl2 buffer
(50mM Tris, 5 mM EDTA, 1.5 mM CaCl2, 5 mM MgCl2, 5 mM KCl, 120 mM NaCl, pH 7.4) and
followed by lysis with an Ultra-Turrax. After centrifugation at 50,000×g, at 4 °C for 18 min the
membranes were resuspended in the binding buffer (50 mM Tris, 1 mM EDTA, 5 mM MgCl2 and
subsequently homogenized with a glass-Teflon homogenizer (20 strokes). The homogenized membranes
were shock-frozen in liquid nitrogen and stored at -80°C. The protein concentration was determined with
the Lowry method with bovine serum albumin used as a standard (Lowry et al., 1951).
Enzyme-Linked Immunosorbent Expression Assay (ELISA): The day after transfection, cells were
detached by trypsin, resuspended into fresh culture medium and plated in poly(D-lysine)-coated 48-well
assay plates. After additional 24h the cells were fixed with 4% formaldehyde solution (Roti®-Histofix
4%, Carl Roth GmbH, Karlsruhe, Germany). After three washes with washing buffer (150 mM NaCl, 25
mM Tris, pH 7.5) and being blocked with 3% skim milk in washing buffer pH 8, cells were incubated
with monoclonal mouse ANTI-FLAG® M2 antibody (Sigma Aldrich, Steinheim, Germany) in the
blocking buffer. After 1 h the cells were washed three times and incubated with Anti-Mouse IgG (whole
molecule)-Peroxidase antibody produced in rabbit (Sigma Aldrich). Subsequently, cells were incubated
with substrate buffer containing 2 mM o-phenylenediamine, 35 mM citric acid, 66 mM Na2HPO4 and
0.015% H2O2 at pH 5. The coloring reaction was stopped with 1 M H2SO4 and the absorption at 490 nm
was determined using a CLARIOstar® microplate reader (BMG Labtech GmbH, Ortenberg, Germany).
Allosteric Radioligand Displacement Assay: Receptor binding studies were performed on membrane
preparations of HEK293T cells expressing the corresponding receptor. The enantiopure tritium labeled
RAMX3 (specific activity: 85.8 Ci/mmol) at the concentration of 1 nM was used for the assays. To
determine unspecific binding 5M of cRAMX3 was used. The assays were carried out in 96 well plates
10

Molecular Pharmacology Fast Forward. Published on January 17, 2018 as DOI: 10.1124/mol.117.110296
This article has not been copyedited and formatted. The final version may differ from this version.
MOL #110296
at a protein concentration of 30 µg/mL in a total volume of 200 µL. The incubation buffer contained 20
mM HEPES, 10 mM MgCl2, 100 mM NaCl and 0.1% BSA at pH 7.4. After incubating for 1 h at 37°C,
the binding was stopped by filtration through Whatman GF/B filters using a 96-channel cell harvester
(Brandel, Unterföhring, Germany). The filters were rinsed five times with ice cold Tris-NaCl buffer.
After drying for 3h at 60 °C, filters were sealed with melt-on scintillator sheets Melti-Lex G/HS
(Whatman/Perkin-Elmer, Freiburg, Germany) and the trapped radioactivity was measured in a microplate
scintillation counter (Micro Beta Trilux scintillator, Whatman/Perkin-Elmer). Three to five experiments
per compound were performed with each concentration in triplicate.
cAMP Biosensor Assay: The experimental procedure for this assay has been adapted as previously
described (Barak et al., 2008). HEK293T cells transiently transfected with the CAMYEL sensor and the
corresponding receptor were split into 96-well half area white tissue culture plates (Greiner-Bio-One
GmbH, Frickenhausen, Germany) at a density of 2 × 104 cells per well. The next day, cells were incubated
with 30 µl Hank`s balanced salt solution (HBSS) for 1 h before treatment. Next, 10 µl substrate
coelenterazine-h (Promega GmbH, Mannheim, Germany) was added to a final concentration of 5 µM.
To measure the effects of ligands on cAMP levels, chemokine alone or EC80 of chemokine and various
concentration of negative allosteric modulator in HBSS buffer containing 50 µM forskolin (Sigma
Aldrich) and 0.2% BSA were added 5 min after coelenterazine-h. After additional 15 min of incubation
the YFP emission (535-30 nm), as well as the Rluc emission (475-30 nm) were measured using a
CLARIOstar® microplate reader. The BRET signal (BRET ratio) was determined by calculating the ratio
between the YFP and the Rluc emission.
β-Arrestin BRET Assay: Cell plating and transfection was performed as mentioned earlier. Forty-eight
hours post-transfection, the medium was exchanged with 80 µl HBSS and incubated 30 min before
treatment. The coelenterazine-h was added at a final concentration of 5 µM. After 5 min. of incubation,
chemokine alone or EC80 of chemokine and various concentration of negative allosteric modulators were
11

Molecular Pharmacology Fast Forward. Published on January 17, 2018 as DOI: 10.1124/mol.117.110296
This article has not been copyedited and formatted. The final version may differ from this version.
MOL #110296
added and incubated for an additional 10 min. The plate was measured on a CLARIOstar® microplate
reader and BRET ratios were calculated.
Data Analysis: The data was analyzed by nonlinear regressions using the algorithms in PRISM 6.0
(GraphPad Software, La Jolla, CA, USA). The equations and models used are described in detail below.
Obtained parameters were compared by one-way ANOVA followed by Dunnett`s Multiple Comparison
Test to estimate the significance in observed differences between wild type CXCR3 and its mutants. The
data from allosteric radioligand displacement assays were analyzed by two different approaches. The
data obtained in the homologous competition assays were analyzed by nonlinear regression:
(
)
bottom + top-bottom
Y=
(1)
ꢁlogEC -Xꢂ∙Hillslope)
50
ꢀ1+10
ꢃ
Because we often observed incomplete displacement of the radioligand by our negative allosteric
modulators, we applied the ternary complex model of allosterism to analyze these data. The data from
allosteric radioligand binding studies were fitted to an equation (Christopoulos and Kenakin, 2002) using
PRISM 6.0:
[
]
B
K
A
ꢄ1+
ꢅ
K
B
Kapp
=
(2)
(3)
α[B]
K
B
ꢄ1+
ꢅ
(
)
Y
1+K
0
A
Y=
[ ]
ꢁ c +K
ꢂ
app
where Kapp describes the occupancy of the orthosteric site, KA is the KD value of RAMX3 for the
investigated receptor, [c] is the concentration of the radioligand, [B] is the concentration of allosteric
modulator, KB is the equilibrium dissociation constant of modulator binding, and α is the ternary complex
constant, which denotes cooperativity factor (Christopoulos and Kenakin, 2002). Values of α > 1 denote
positive cooperativity, whereas α < 1 denotes negative cooperativity. Values of α approaching zero are
indistinguishable from competitive antagonism. When α approaches zero, the Kb value approaches the
Ki value (Christopoulos and Kenakin, 2002). An α value equal to 1 denotes an allosteric interaction that
result in an unaltered ligand affinity.
12

Molecular Pharmacology Fast Forward. Published on January 17, 2018 as DOI: 10.1124/mol.117.110296
This article has not been copyedited and formatted. The final version may differ from this version.
MOL #110296
To obtain the EC50 and Emax (maximal response) values for CXCL11 and CXCL10 in wild type CXCR3
and its mutants from cAMP BRET and β-arrestin BRET Assays, dose-response curves were fitted by
nonlinear regression and normalized to the CXCR3 WT response:
(
)
bottom + top-bottom
Y=
(4)
ꢁlogEC -Xꢂ∙Hillslope)
50
ꢀ1+10
ꢃ
Subsequently, the index of agonism log(max/EC50) was calculated to quantify the effects of receptor
mutation on CXCL11- and CXCL10-mediated signaling. The parameter is derived from agonists
concentration-response curves and comprises the maximal response to the agonist (max) and the EC50.
Finally, comparisons of log(max/EC50) values between mutants yielded Δlog(max/EC50), which allows
the system independent scaling of agonism within a given functional system. When this is done, the
agonism of the agonist at mutant CXCR3 is compared with the agonist at WT CXCR3 and system effects
are cancelled (Kenakin, 2017).
Δlog(max/EC50) = log(max/EC50)WT - log(max/EC50)mutant
(5)
To account error propagation, standard errors of the mean for the normalized Δlog(max/EC50) values
were calculated employing equation 6.
푆. 퐸. 푀.ꢆ,ꢇ = ꢈ(푆. 퐸. 푀.ꢆ )ꢇ + (푆. 퐸. 푀.ꢇ )ꢇ
(6)
To account for the change in activity between the two chemokines CXCL11 and CXCL10 the
ΔΔlog(max/EC50) values were calculated for each mutant. For WT CXCR3 the ΔΔlog(max/EC50) was
set to 0.
ΔΔlog(max/EC50) = Δlog(max/EC50)CXCL10 - Δlog(max/EC50)CXCL11
(7)
13

Molecular Pharmacology Fast Forward. Published on January 17, 2018 as DOI: 10.1124/mol.117.110296
This article has not been copyedited and formatted. The final version may differ from this version.
MOL #110296
Results
We previously described a series of novel negative allosteric modulators with biased functional
properties (Fig. 1). BD064 preferentially inhibits CXCL11-mediated β-arrestin 2 recruitment over G
protein activation, whereas BD103 preferably blocks CXCL11-mediated activation of G proteins rather
than β-arrestin 2 recruitment (Bernat et al., 2015).
Based on the homology model of CXCR3 and docking studies, our aim was to define the binding mode
of given biased negative allosteric modulators experimentally and to dissect the molecular mechanisms
responsible for biased signaling. We therefore combined site-directed mutagenesis with detailed
functional characterization of mutant receptors in allosteric radioligand displacement and BRET-based
cAMP and β-arrestin recruitment assays.
Computational studies on the binding mode of BD064 and BD103. We recently determined a binding
mode for the RAMX3, a closely related CXCR3 antagonist, to its CXCR3 receptor based on a homology
model after 1.8 µs of molecular dynamics simulation using metadynamics enhanced sampling (Milanos
et al., 2016b). Two closely related biased negative allosteric agonists BD064 and BD103 were then
docked based on this binding mode. The binding mode of BD064 is analogous to that of cRAMX3
(Milanos et al., 2016b). BD064 spans from the minor to the major pocket, mainly binding to the interface
between these two pockets and interacting with residues in TM3, 6 and 7 of CXCR3 (Fig. 2). Specifically,
F1313.32 can form π-π stacking interactions with both the aromatic bicyclic unit and the boron phenyl ring
of BD064. Additional π-π stacking interactions are observed with Y2716.51 and Y3087.43. K3007.35
hydrogen bonds to the nitrogen atom of the heterocycle. Further amino acids that were not tested by
mutagenesis appear to interact with BD064 in the proposed binding pocket including F1353.36, W1092.60
,
H2726.52 and D2786.58. The binding mode of BD103 differed from those of cRAMX3 and BD064. It sits
higher and further left towards TM 5and 6 in the binding cavity of CXCR3 (Fig. 3). The binding site is
14

Molecular Pharmacology Fast Forward. Published on January 17, 2018 as DOI: 10.1124/mol.117.110296
This article has not been copyedited and formatted. The final version may differ from this version.
MOL #110296
mainly defined by residues from TM 5, 6 and 7. The upward shift of the binding pocket of BD103 is
predominantly caused by the cation-π interaction of the piperidine moiety with Y205ECL2, which prevents
BD103 binding to the deep binding pocket of cRAMX3 and BD064. Two additional residues contribute
to this “hot spot”; R2165.39 and D1864.60 build an ionic lock to stabilize the binding conformation of
BD103. F1313.32 and Y3087.43, important anchor residues for BD064 and cRAMX3, do not interact with
BD103 due to the higher binding site and the resulting increased distance between the modulator and
these two residues. The binding pocket of BD103 partially overlapped with that of BD064 and shared
the residues Y2716.51 and W1092.60, which are predicted to form multiple π- π stacking interactions with
BD103. Additionally, the methoxyphenyl moiety of BD103 interacted with I2796.59, V2756.55, T2135.36
and V2175.40 via lipophilic interactions.
Influence of amino acid substitutions on the affinity and cooperativity of biased negative allosteric
modulators towards RAMX3. Changes in the receptor sequence induced by site-directed mutagenesis
can have significant impact on the receptor expression level. The expression of the CXCR3 receptor
mutants was analyzed by whole cell-based ELISA. HEK293T cells were transiently transfected with
cDNA coding for FLAG-CXCR3 wild type or mutant receptors. The cell surface expression level of
mutants was found to be between 61% and 109% of the wild type expression (Supplemental Information
Table S1). The mutations W2686.48F andY2716.51F resulted in a most significant reduction of the receptor
surface expression, reaching only about 60% of the wild type expression.
To examine the effects of amino-acid substitutions on the binding of the allosteric modulators, we next
performed an allosteric radioligand (RAMX3) (Bernat et al., 2012) displacement assay. The use of
allosteric radioligand enables the measurement of affinity of the allosteric ligands directly for their
allosteric binding pocket in a competitive manner. In this assay, the membrane preparations of HEK293T
cells that express the corresponding receptor were used. In general, the specific binding detected in this
assay correlated well with the ELISA findings (Supplemental Information Fig. S1, Table S1). Decreased
15

Molecular Pharmacology Fast Forward. Published on January 17, 2018 as DOI: 10.1124/mol.117.110296
This article has not been copyedited and formatted. The final version may differ from this version.
MOL #110296
level of bound RAMX3 at the W2686.48F mutant correlates with a reduced receptor expression as
determined by ELISA. The mutation K3007.35R with an expression level comparable to the wild type,
resulted in strongly reduced radioligand binding, suggesting that this residue is an important anchor for
the binding of RAMX3 to CXCR3. Remarkably, reduced surface expression level of the Y2716.51
F
mutant did not influence the extent of specific RAMX3 binding. Most likely, this mutation influences
the extent of the receptor cell surface expression, but not the overall expression of the protein. Membrane
preparations were used for the radioligand binding-studies that include not only the plasma membrane
but also other intracellular membranes that might contain immature receptors. The ELISA assay only
detects mature receptors expressed on the surface.
Next, the ability of cold RAMX3 (cRAMX3) and two biased negative allosteric modulators BD064 and
BD103 to displace RAMX3 from the wild type and mutant receptors was evaluated. We previously
showed that BD064 and BD103 were unable to suppress the binding of RAMX3 to CXCR3 completely,
so that the ternary complex model of allosterism was used to analyze the radioligand binding data (Bernat
et al., 2014; Bernat et al., 2015). This model was also used here to analyze the radioligand binding data.
It describes allosteric interactions only in terms of the equilibrium dissociation constant pKb and the
cooperativity factor α. Values of α > 1 denote positive and α < 1 negative cooperativity. Values of α
approaching zero are indistinguishable from competitive antagonism. α = 1 denotes an allosteric
interaction that results in unaltered ligand affinity (Christopoulos and Kenakin, 2002; Ehlert, 1988). The
pKb and α values from these experiments are listed in Supplemental Information Table S1 and depicted
on Fig. 4; none of the mutants had a significant effect on the affinity of cRAMX3. As expected, the
binding cooperativity between the allosteric radioligand RAMX3 and cRAMX3 was indistinguishable
from competitive antagonism with α equaling or approaching zero. As reported previously, BD064
displayed nearly the same affinity as cRAMX3 at the wild type but weaker negative cooperativity against
RAMX3 with α value of 0.38 ± 0.05 (Bernat et al., 2014; Bernat et al., 2015). Consistent with our
16

Molecular Pharmacology Fast Forward. Published on January 17, 2018 as DOI: 10.1124/mol.117.110296
This article has not been copyedited and formatted. The final version may differ from this version.
MOL #110296
hypothesis that cRAMX3 and BD064 can bind to CXCR in different orientations (Bernat et al., 2014;
Bernat et al., 2015), few mutations influenced the BD064 affinity. Specifically, alanine substitution of
the residue Y3087.43 led to a 15-fold loss in affinity of BD064. An even more drastic effect was observed
at the F1313.32A and S3047.39E mutant, where the low binding affinity did not allow pKb and α values to
be calculated. These results indicate that Y308, F131 and S304 contribute substantially to the binding of
BD064 to CXCR3, but not RAMX3. The negative cooperativity between BD064 and RAMX3 was
significantly reduced only at the W2686.48F mutant (α value of 0.38 vs. 0.72), thus suggesting that this
residue plays a role in the transmission of cooperativity between BD064 and the radioligand RAMX3
and does not contributes to the binding affinity. For BD103 the observed affinity for wild type CXCR3
was a somewhat lower than previously reported with a comparable incomplete displacement of the
radioligand (α value of 0.22) (Bernat et al., 2015). The S3047.39E mutant, which displayed a large
reduction in affinity for BD064, also caused a striking loss of affinity for BD103. A considerable effect
on the binding of BD103 was also observed for the mutation of D1864.60N. In contrast to BD064 the
mutation F1313.32A had no effect on affinity and cooperativity of BD103. The residues S3047.39 and
D1864.60 are thus essential for the binding of BD103 to CXCR3. The serine mutation S3017.36A, which
did not have any significant influence on binding affinity of BD103, resulted in a significant weakening
of the negative cooperativity between BD103 and RAMX3 (α value of 0.47 vs. 0.22). All other mutants
had no appreciable effect on the affinity and cooperativity of BD103 towards RAMX3.
In summary, the radioligand displacement experiments confirmed that the biased allosteric modulators
BD064 and BD103 bind to overlapping allosteric sites in CXCR3, and thus react to mutations differently.
Key residues like F1313.32, D1864.60 and Y3087.43 arose as the most important anchor points for the
investigated biased negative allosteric modulators. These interactions are also consistent with the
proposed binding mode from the docking studies.
17

Molecular Pharmacology Fast Forward. Published on January 17, 2018 as DOI: 10.1124/mol.117.110296
This article has not been copyedited and formatted. The final version may differ from this version.
MOL #110296
Effects of Mutations on Signaling of Chemokines CXCL11 and CXCL10. Before investigating the
interplay between the endogenous agonists and biased negative allosteric modulators, we investigated
the influence of mutations on chemokine signaling in G protein-dependent and independent pathways.
CXCR3 is a Gαi-coupled receptor, thus the G protein-dependent signaling can be monitored as a change
in the intracellular cAMP level (Denis et al., 2012). While in the G protein-independent signaling β-
arrestin 1 and β-arrestin 2 are involved (Oakley et al., 2000), which evoke isomer-specific functions (Ahn
et al., 2004; Kohout et al., 2001), we decided to monitor the influence of mutations on both pathways.
We chose BRET-based approaches for all three functional assays. BRET allows the real-time detection
of protein-protein interactions in living cells (Eidne et al., 2002). For the quantitatively and rapid
monitoring of intracellular levels of cAMP we used the Epac based BRET sensor CAMYEL (Jiang et
al., 2007). CXCR3 couples to Gαi, which leads to a negative regulation of the adenylate cyclase and a
decrease in cAMP levels (Denis et al., 2012) and requires a pre-stimulation of adenylate cyclase with
forskolin. To measure the recruitment of β-arrestin with BRET, CXCR3 was fused to the modified EYFP
(mVenus) and β-arrestin was fused to Rluc as described by Hamdan et al. (Hamdan et al., 2005). Agonist
concentration-response curves were fitted by nonlinear regression to yield Emax (maximal response) and
EC50 (Supplemental Information Table S2-S4) and subsequently analyzed applying the Δlog(max/EC50)
scale (Fig. 5). Subsequently the ΔΔlog(max/EC50) values were calculated for the investigated pathways
to get the change in CXCL10 activity vs. CXCL11 at each mutant (Fig. 6). This single index of agonism
allows a system-independent comparison of agonist potencies and signaling efficacies across different
receptor mutants in that the effects of the system processing of agonist response and differences in assay
sensitivity and receptor expression are cancelled (Kenakin, 2017).
All mutant receptors were readily activated by CXCL11 and CXCL10 as shown in a dose-dependent
inhibition of forskolin-stimulated cAMP production. However, some mutations induced the opposite
effect on the activation of G proteins by either decreasing or increasing the signaling depending on the
18

Molecular Pharmacology Fast Forward. Published on January 17, 2018 as DOI: 10.1124/mol.117.110296
This article has not been copyedited and formatted. The final version may differ from this version.
MOL #110296
chemokine used. W2683.32F and D2977.32N gave a five to eight-fold reduction in the potency of CXCL11
and CXCL10, although for CXCL11 the change in log(max/EC50) was not statistically significant (Fig.
5 and 6). However, the efficacy of CXCL11 increased significantly at these mutations, but not that of
CXCL10. The mutation D1864.60N resulted in a drastic increase of CXCL11 efficacy (+51%) with no
influence on potency. This mutation led, on the other hand, to the most dramatic drop in potency of
CXCL10 and therefore to a nearly 10-fold decrease in relative activity of CXCL10 over CXCL11 (Fig
6). In contrast, the mutation Y2716.51F produced a 2-fold selective decrease on the activity of CXCL11
over CXCL10 (Fig. 6).
The influence of mutations on the CXCL11- and CXCL10- mediated recruitment of β-arrestins differed
significantly from their influence on the G protein mediated signaling (Supplemental Information Table
S3 and S4, Fig. 5 and 6). Mutation of D2977.32 to asparagine led to reduced efficacy of both chemokines
(31% and 49%, respectively) in recruiting β-arrestin 1, without affecting β-arrestin 2 recruitment (Table
3 and 4). Moreover, this mutant had no influence on the potency of CXCL11, but induced a 16-fold
reduction in recruitment for β-arrestin 1 and 7-fold for β-arrestin 2 (Fig. 7A and B). Similarly, exchange
of S3017.36 for alanine had no effect on the CXCL11- and CXCL10-mediated β-arrestin 2 recruitment,
yet both the efficacy of CXCL11 and CXCL10 and the potency were decreased for β-arrestin 1
recruitment (Supplemental Information Table S3 and S4, Fig. 5). Both alanine mutations of F1313.32 and
Y3087.43 led to a pronounced decrease in CXCL11 mediated recruitment of β-arrestins (Fig. 5 and 6).
The substitution of K3007.35 by either methionine or arginine did not affect the recruitment of β-arrestin
mediated by either CXCL11 or CXCL10. The mutations D1864.60N, W2686.48F and S3047.39E almost
completely abolished the CXCL10-mediated β-arrestin 1 and 2 recruitment to CXCR3. It is important
to note, that this effect was exclusively observed for CXCL10 (Supplemental Information Table S3 and
S4, Fig 6). However, CXCL11 was still able to mediate the recruitment of β-arrestin to D1864.60N and
W2686.48F mutants, despite a significant loss of efficacy. The mutation S3047.39E had no effect at all on
19

Molecular Pharmacology Fast Forward. Published on January 17, 2018 as DOI: 10.1124/mol.117.110296
This article has not been copyedited and formatted. The final version may differ from this version.
MOL #110296
CXCL11- mediated β-arrestin-recruitment. Although the mutation S3047.39E almost completely
abolished the CXCL10-mediated recruitment of β-arrestins to CXCR3 wild type, this mutation had no
influence on CXCL10- mediated G protein signaling (Fig. 5). Collectively, D1864.60, W2686.48 and
S3047.39 play a role in signal pathway selectivity as mutations of these residues led to a G protein active
but β-arrestin inactive conformation.
In general, the mutations had a pronounced effect on the potency of CXCL11- and CXCL10- mediated
G protein activation, in contrast to their effect on CXCL11- and CXCL10- mediated β-arrestin
recruitment, where the efficacy was strongly influenced. These observations are in accord with the
accepted notion that different receptor conformations initiate either G protein activation or recruitment
of β-arrestin (Liu et al., 2012; Rahmeh et al., 2012). Moreover, these conformations can also differ
depending on the bound chemokine, which consequently results in a probe-dependent behavior (Fig. 6).
Effects of Mutations on the Ability of Negative Allosteric Modulators to Inhibit Chemokine -
mediated cAMP Signaling. The binding of compounds to the allosteric site on a GPCR changes the
receptor conformation and can therefore affect the binding affinity and /or the signaling efficacy of the
orthosteric ligand (May et al., 2007). To estimate the influence of mutations on the ability of allosteric
modulators to modulate the potency and efficacy of CXCL11 and CXCL10, we first analyzed their
influence on the G protein-dependent signaling (Supplemental Information Table S5 and S6). The results
were analyzed using a simple allosteric ternary complex model as described before (Bernat et al., 2014).
The assumptions were that the allosteric modulators neither depress the maximal response nor suppress
the basal activity. These effects are not accounted for in an allosteric ternary complex model. Importantly,
20

Molecular Pharmacology Fast Forward. Published on January 17, 2018 as DOI: 10.1124/mol.117.110296
This article has not been copyedited and formatted. The final version may differ from this version.
MOL #110296
even if these assumptions are not entirely true for all the allosteric modulators, this analysis enables an
initial approximation and a semi-empirical estimate of cooperativity (Christopoulos and Kenakin, 2002).
All three negative allosteric modulators suppressed the chemokine-induced activation at CXCR3 with
the following rank order of functional affinities: cRAMX3 > BD064 > BD103, where slight probe
dependence was observed for BD064. Compound BD064 inhibited the response of CXCL10 more
strongly than that of CXCL11 (6.5-fold difference in pKb: 6.69 vs. 7.50, two-tail unpaired t-test, p < 0.05,
Supplemental Information Table S5 and S6), in accord with the results of an [35S]GTPγS incorporation
assay (Bernat et al., 2015). The modulators demonstrated behavior undistinguishable from competitive
antagonism with α equal or nearly zero. Neutralizing the two aspartic acid residues D1864.60 and D2977.32
by mutation to asparagine generally improved the functional affinity of cRAMX3 and BD064. A
significant improvement was observed for BD064 to modulate CXCL11-mediated G protein activation
at the D1864.60N mutant (7.7-fold, Fig. 7I) and for cRAMX3 to modulate CXCL10-mediated G protein
activation at D1864.60N and D2977.32N (12-fold and 8-fold, respectively, Fig. 8B). The mutations
D1864.60N and D2977.32N resulted in enhanced negative cooperativity between BD103 and CXCL11.
Furthermore, cRAMX3 was more potent at Y2716.51F to inhibit CXCL11- and CXCL10- mediated cAMP
response (5-fold and 7-fold, respectively). The substitution of K3007.35 for methionine resulted in a 6-
fold pKb improvement of cRAMX3 to modulate CXCL10-induced G protein signaling. However, this
mutation induced a 5.5fold loss of functional affinity and cooperativity between BD064 and CXCL11
(Fig. 7D). Moreover, this effect was probe-dependent because the mutation K3007.35M did not affect
BD064 ability to inhibit CXCL10-mediated response. Substitution with arginine at the same position was
tolerated for all given negative allosteric modulators as well as the mutation of S3017.36 to alanine. The
mutation of W2686.48 to phenylalanine caused a large reduction (8-fold) in cRAMX3 ability to inhibit
CXCL11-mediated activation (Supplemental Information Table S5), but had no effect on BD064 and
BD103. The same mutation resulted in a minor reduction (3-fold) of BD064 ability to inhibit CXCL10-
21

Molecular Pharmacology Fast Forward. Published on January 17, 2018 as DOI: 10.1124/mol.117.110296
This article has not been copyedited and formatted. The final version may differ from this version.
MOL #110296
mediated activation (Fig. 8E) and a complete loss of BD103 activity towards CXCL10 (Fig. 8H). The
most striking effects on the signaling properties of the negative allosteric modulators were observed at
F1313.32A, Y3087.43A and S3047.39E. Regarding CXCL11-mediated G protein activation, F1313.32A let to
a considerable lower pKb of cRAMX3 (27-fold) and BD064 (4.5-fold) and a weaker negative
cooperativity between BD064 and CXCL11 (23-fold, Fig. 7A-C). In contrast, this mutation had no
significant effect on the ability of both negative allosteric modulators to inhibit CXCL10-mediated
activation indicating a probe-dependent effect of this mutant (Fig. 8A and D). Likewise, strong probe-
dependence was detected by the exchange of Y3087.43 for alanine, which caused a dramatic reduction of
the functional affinity and cooperativity of cRAMX3 to inhibit CXCL11-mediated signaling (19.5-fold
and 57-fold, respectively, Fig. 7G), whereas only the functional affinity of cRAMX3 towards CXCL10
was decreased (26-fold) with a nearly unchanged cooperativity (α value of 0.07). In addition, this
mutation seems to play an even more important role for the functional activity of BD064 as the pKb and
α value could not be determined accurately because of the pronounced reduction in signaling. Although
this compound was still able to inhibit CXCL10-mediated G protein activation at the Y3087.43 mutant,
the cooperativity and functional affinity were substantially decreased (30-fold and 56-fold, respectively).
The last mutant S3047.39E displayed a significant drop in functional affinity and cooperativity for
cRAMX3 and BD064 to inhibit the chemokine elicited response (Fig. 7E). As reported earlier the
residues involved in the signaling of BD103 were difficult to detect due to poor affinity, but we noted
some slight weaker negative cooperativity for the Y2716.51F, K3007.35M, S3017.36A and Y3087.43
A
(Supplemental Information Table S5 and Fig. 7H) mutant in presence of CXCL11 as well as for the
D2977.32N, K3007.36M, S3047.39E and Y3087.43A (Supplemental Information Table S6 and Fig. 8I) mutant
in the presence of CXCL10. The residues contributing to the signaling transduction of BD103 appear to
differ in some parts compared to cRAMX3 and BD064, consistent with the results from the radioligand
binding studies. The cAMP BRET assay also showed that at some mutants (F1313.32A, D2977.32N,
22

Molecular Pharmacology Fast Forward. Published on January 17, 2018 as DOI: 10.1124/mol.117.110296
This article has not been copyedited and formatted. The final version may differ from this version.
MOL #110296
K3007.35R and Y2716.51F) the modulators elicited higher levels of cAMP as the basal level pre-stimulated
with forskolin (e.g. Fig. 8A). This indicates inverse agonistic activity of the modulators, which was
confirmed by experiments without the addition of endogenous ligand (Supplemental Information Fig.
S2). In general, the data suggest that the amino-acid residues involved in the binding of the modulators
(F1313.32A, Y3087.43A and S3047.39E), also play an important role in signaling transmission. However,
there is a need to differentiate the modifications, that influence the compound affinity, from those that
influence the cooperativity exhibited toward the endogenous ligand, as the two properties are not
correlated (Kenakin and Miller, 2010; Keov et al., 2011).
Effects of Mutations on the Ability of Modulators to Inhibit Chemokine - mediated β-Arrestin
Recruitment. To evaluate the importance of amino-acid substitutions for β-arrestin biased negative
allosteric modulation, we tested cRAMX3, BD064 and BD103 for their ability to inhibit the chemokine-
induced recruitment of β-arrestin 1 and 2 on the CXCR3 WT and its mutants using a BRET based assay
(Supplemental Information Table S7-S10, Fig. 7 and 8). Because only a few amino-acid substitutions
resulted in differentiation between the β-arrestin 1 and 2 pathways, the general discussion of the
observations applies for the both β-arrestin isomers. The exceptions will be pointed out explicitly. The
rank order of functional affinities at wild type CXCR3 was similar as for the G protein activation.
The only deviation within the two investigated pathways at wild type CXCR3 was noted for the
cooperativity between BD103 and CXCL10; the cooperativity was diminished 17-fold (β-arrestin 1) and
26-fold (β-arrestin 2), whereas the α value for G protein activation was approximately zero, and thus
undistinguishable from competitive antagonism (Supplemental Information Tables S8 and S10).
However, this was a probe-dependent effect because in both pathways BD103 showed a negative
cooperativity towards CXCL11. Substitution of D1864.60 for asparagine followed a similar trend as
observed in the cAMP assay, by increasing the functional activities of cRAMX3 and BD064 to inhibit
CXCL11-mediated β-arrestin 2 recruitment (6-fold and 9.5-fold, respectively, Supplemental Information
23

Molecular Pharmacology Fast Forward. Published on January 17, 2018 as DOI: 10.1124/mol.117.110296
This article has not been copyedited and formatted. The final version may differ from this version.
MOL #110296
Table S7, S9 and Fig. 7I). However, the increase was not statistically significant in the β-arrestin 1
pathway. The D2977.32N mutant showed no notable effect on cRAMX3 and BD064 to modulate
CXCL11-induced recruitment of β-arrestins. However, for CXCL10- induced recruitment of β-arrestin 1
this mutation led to opposite effects on the functional activity of cRAMX3 between the G protein and β-
arrestin 1 pathways with a 6-fold increase in the former and a 5-fold decrease in the latter (Fig. 8B). The
resulting difference in functional activity (30-fold) indicates that this residue is an intersection point
between G protein and β-arrestin 1 recruitment for cRAMX3. For BD103, the introduced asparagine
improved the negative cooperativity exhibited towards CXCL10 in the β-arrestin pathway to such an
extent that it reached an α value of approximately 0 undistinguishable from competitive antagonism (Fig.
8F), and the functional activity between BD103 and CXCL11 (8.5 and 5-fold, respectively β-arrestin 1
and β-arrestin 2). Further gains in the pKb value of BD103 were detected at Y2716.51F (6.5-fold) to
modulate CXCL11 mediated β-arrestin 1 recruitment (Table 7 and 9). In contrast, W2686.48F completely
abolished the response of BD103 towards CXCL10 for β-arrestin 1 recruitment (Fig. 8H). Moreover, this
effect was probe dependent as W2866.48F had no effect on CXCL11-induced β-arrestin 1 recruitment.
The other two modulators were mostly unaffected by theses residues except for the W2686.48F mutant,
which caused significantly reduced pKb values for cRAMX3 and BD064 to inhibit CXCL10-mediated
β-arrestin 1 recruitment (Fig. 8E). However, this effect could be also a reason for the dramatically
reduced efficacy of CXCL10 and the resulting small measurement window. Variable reductions in
functional affinity (5- to 19-fold) and cooperativity were noted for all three compounds to inhibit the
CXCLL- and CXCL10-elicited response at the F1313.32A mutant with BD103 having the most drastic
effect (Fig. 7A-C). Interestingly, the influence of the mutation was less pronounced in the presence of
CLXC10, especially in the β-arrestin 2 pathway, which indicates probe-dependence (Fig. 8A, D and G).
The substitution of S3047.39 for glutamic acid revealed a remarkably lower functional affinity of
cRAMX3 to inhibit CXCL11 elicited recruitment of β-arrestin 1 (4-fold) and an even more drastic impact
24

Molecular Pharmacology Fast Forward. Published on January 17, 2018 as DOI: 10.1124/mol.117.110296
This article has not been copyedited and formatted. The final version may differ from this version.
MOL #110296
for the β-arrestin 2 pathway (10-fold). In contrast BD064 suffered the greatest losses in cooperativity
leading to an almost complete abrogation of signaling (Fig. 7E). In addition, BD103 followed a similar
trend with significant reductions in functional affinity and cooperativity, whereas this mutation had no
influence on the modulator in the cAMP assay (Fig. 7F). The most profound results were seen at the
Y3087.43A mutant, where the inhibitory effect on CXCL11-mediated β-arrestin recruitment was totally
abolished irrespective of the allosteric modulator tested. Again, the most drastic influence, switching the
negative cooperativity to positive, was seen for BD103 (Fig. 7H). No α values between the modulators
and CXCL10 could be derived from the results due to the low functional affinity and the precipitous
dose-response curve (Fig. 8I). These findings are consistent with the results from the G protein activation
experiments suggesting that these two residues likely contribute to a region in the network of interactions
that govern the transmission of signaling for both pathways investigated. The pKb values estimated
obtained from the β-arrestin recruitment studies at the remaining mutants were not significantly different
from the wild type CXCR3.
25

Molecular Pharmacology Fast Forward. Published on January 17, 2018 as DOI: 10.1124/mol.117.110296
This article has not been copyedited and formatted. The final version may differ from this version.
MOL #110296
Discussion
Allosteric ligands can robustly affect domain coupling and the conformational rearrangement upon
chemokine activation of the chemokine receptors. Because the complex pharmacology of chemokines
and their receptors might be a reason for the high attrition rate in the development of drugs that target
chemokine receptors (Tschammer et al., 2014), it is essential to dissect molecular mechanisms that
govern the interactions between allosteric modulators, chemokines and their receptors. With the aim to
improve the understanding of the complex molecular mechanisms that govern the functions of CXCR3,
we recently designed and characterized novel biased negative allosteric modulators of CXCR3, which
elicit a bias in the inhibition of a given signaling pathway (G-protein-dependent signaling or recruitment
of β-arrestin) and additionally tend to inhibit a specific pathway depending on the endogenous agonist
used (Bernat et al., 2014; Bernat et al., 2015). In this work, we have rationalized molecular mechanisms
that direct the effects of the biased negative allosteric modulators BD064 and BD103, and the unselective
cRAMX3. Starting with a homology model (Milanos et al., 2016b) and docking of BD064 and BD103
we created a model that helped us to define the region suitable for the site-directed mutagenesis. By the
combination of site-directed mutagenesis and detailed functional assays, we identified and categorized
the contribution of various amino acids to the binding affinity and activity of investigated CXCR3
ligands. Obtained data were analyzed by the allosteric ternary complex model (ATCM), which is the
simplest mechanism to describe allosteric interactions (Christopoulos and Kenakin, 2002). Although the
general application of this model is to binding data, it can also be applied to functional data, but requires
the additional assumption that the allosteric modulator does not perturb the signaling capacy of the
receptor. When allosteric modulators change orthosteric ligand signaling efficacy in addition to, or
instead of, any observed effects on binding affinity, the ATCM is insufficient (Christopoulos, 2002). For
allosteric ligands that also act as agonists in their own right, an extended model is needed that consider
the isomerization of a GPCR between active and inactive states (Hall, 2000); or for ligands that change
26

Molecular Pharmacology Fast Forward. Published on January 17, 2018 as DOI: 10.1124/mol.117.110296
This article has not been copyedited and formatted. The final version may differ from this version.
MOL #110296
kinetics of orthosteric ligand-receptor interactions (Avlani et al., 2004). Although the ATCM shows some
limitations related to the correlation between parameters and the dependence of some of these
parameteres on tissue effects such as receptor number, receptor coupling efficiency and amplification
within functional assays, it can still be very useful in providing semi-quantitative estimates of allosteric
effects (Keov et al., 2011). To our knowledge, this is a first comprehensive study of the interactions
between the endogenous chemokines and biased negative allosteric inhibitors of CXCR3.
The mutation of aspartic acids D1864.60 and D2973.32 to asparagine abolished the probe-dependency
previously displayed by BD064, which preferentially inhibits the activation of CXCR3 by CXCL10
(Bernat et al., 2015). This indicates that the aspartic acid contributes to a conformation of CXCR3 that
allows BD064 to inhibit CXCL10 preferentially. The introduction of the asparagine abolished constraints
responsible for the probe-dependency, resulting in an undistinguishable inhibition between CXCL11 and
CXCL10. The mutations D1864.60N and D2973.32N also increased both the functional affinity of
cRAMX3 to modulate CXCL10-mediated G protein activation and the cooperativity between BD103
and CXCL11. In contrast, only D1864.60N had a similar enhancing effect on all three allosteric modulators
to inhibit CXCL11-induced β-arrestin 2 recruitment. The mutation of residue D2973.32 induced probe-
dependence for cRAMX3 as the functional activity of cRAMX3 to inhibit CXCL10-mediated
recruitment of β-arrestin 1 was dramatically decreased, whereas CXCL11-mediated recruitment of β-
arrestin 1 was unaffected. In contrast, the inhibition of CXCL10-induced G protein activation by
cRAMX3 was improved at this mutant. The resulting difference in functional affinity indicates that
D2973.32 functions as a molecular switch or gatekeeper, which constrains the recruitment of β-arrestin 1
and may serve as a key residue to trigger bias signaling between the two arrestin isoforms as well as G
protein signaling. The asparagine introduced at this position yielded an enhanced cooperativity between
BD103 and CXCL10 and higher functional affinity to activity to inhibit CXCL11-mediated β-arrestin
recruitment without affecting the inhibition of G-protein activation.
27

Molecular Pharmacology Fast Forward. Published on January 17, 2018 as DOI: 10.1124/mol.117.110296
This article has not been copyedited and formatted. The final version may differ from this version.
MOL #110296
A highly conserved residue W2686.48 is involved in the activation of the GPCRs (Schwartz et al., 2006).
It was found that mutation of W2486.48 into phenylalanine led to a G protein active, but β-arrestin inactive,
and thus biased CCR5 conformation (Steen et al., 2013). In CXCR3, the mutation of W2686.48 to
phenylalanine caused variable probe-dependent effects on the modulators irrespective of the pathway
investigated. Generally, the ability of the biased negative allosteric modulators BD064 and BD103 to
inhibit CXCL10-mediated signaling was impaired; BD103 showed the most dramatic loss of function.
Markedly, although the ability of cRAMX3 to inhibit CXCL11-mediated G protein activation was
impaired, its ability to inhibit CXCL10-mediated G protein activation was not affected. According to our
docking studies, the residue W2686.48 does not interact directly with any of the ligands; as reported for
other GPCRs (Schwartz et al., 2006), it steers between the active and inactive conformations of the
GPCR. Our data show that in CXCR3, W2686.48 is not a simple molecular on-off switch, but rather fine-
tunes the responses, depending on the nature of a ligand and thus importantly contributes to the probe-
dependency.
Another vital residue for the probe-dependency is F1313.32. The mutation F1313.32A impaired the ability
of cRAMX3 and BD064 to inhibit CXCL11-induced G protein activation and β-arrestin recruitment.
Noteworthy, in presence of CXCL10 the signaling of cRAMX3 and BD064 was mostly unchanged,
which clearly indicates probe-dependent contribution of F1313.32 to the overall signaling of CXCR3.
Also, the F1313.32A mutant did not impact BD103 to modulate G protein activation, however, it
completely abolished the response of BD103 in the β-arrestin recruitment pathway irrespective of the
chemokine used. This finding underscores the importance of F1313.32 as the residue responsible for the
biased inhibition of chemokine-induced signaling by BD103. The importance of this residue as the
anchor point was also described for the structurally related ligand VUF10085, where the mutation to
alanine significantly decreased the ability of VUF10085 to inhibit chemotactic response to CXCL11
(Nedjai et al., 2015). Similarly, the Y3087.43A mutant rendered VUF10085 impotent, which agrees with
28

Molecular Pharmacology Fast Forward. Published on January 17, 2018 as DOI: 10.1124/mol.117.110296
This article has not been copyedited and formatted. The final version may differ from this version.
MOL #110296
our current findings. In the cAMP assay, the mutation Y3087.43A diminished both the functional affinity
and the cooperativity of cRAMX3 towards CXCL11, whereas only the functional affinity of cRAMX3
was reduced upon activation with CXCL10 indicating a specific implication of Y3087.43 in the
transmission of cooperativity between cRAMX3 and CXCL11. An even more striking effect was
observed for BD064 where the Y3087.43A mutant induced a complete loss of allosteric modulation of
CXCL11-mediated G protein activation, although BD064 was still able to inhibit CXCL10-mediated
response but with significant reductions in functional affinity and cooperativity. The importance of this
residue was further confirmed by the finding that all three allosteric modulators elicited no discernible
β-arrestin recruitment at the Y3087.43A mutant. Likewise, substitution of S3047.39 with glutamic acid
substantially affected the ability of cRAMX3 and BD064 to modulate CXCL11- and CXCL10 induced
G protein activation and β-arrestin recruitment. It is noteworthy that this residue plays a key role in the
transmission of cooperativity between BD064 and either CXCL11 or CXCL10 as mutation to glutamic
acid results in an almost complete abrogation of signaling. The inhibitory properties of BD103 were only
influenced in the β-arrestin recruitment assay at S3047.39E. Overall, all three mutations (F1313.32A,
Y3087.43A and S3047.39E) showed significant impact on the binding of BD064 and BD103 and affected
the signaling of CXCL11 and CXCL10 significantly. Finally, they substantially influenced the functional
properties of the negative allosteric modulators in all pathways investigated.
Besides the discovery of amino acids that contribute to the phenomenon of bias and probe-dependent
signaling of the negative allosteric modulators, we further identified D1864.60, W2686.48 and S3047.39 as
key residues that are specifically involved in CXCL10-mediated β-arrestin recruitment but not G protein
activation, as mutating them led to a G protein active but β-arrestin inactive conformation. Note that both
CXCL11-mediated G protein activation and β-arrestin recruitment were mainly unaffected by the
investigated mutations, which confirms different receptor conformations stabilized by each chemokine.
29

Molecular Pharmacology Fast Forward. Published on January 17, 2018 as DOI: 10.1124/mol.117.110296
This article has not been copyedited and formatted. The final version may differ from this version.
MOL #110296
In summary, we have identified key residues in the CXCR3 receptor that are involved in the binding,
signaling and transmission of cooperativity between the three negative allosteric modulators and the
endogenous chemokine ligands CXCL11 and CXCL10. We confirmed that BD064 and BD103 adopt
some slightly different binding modes, which are in turn responsible for the probe-dependent inhibition
of G protein–dependent or –independent signaling. The differential effects of mutations on either
negative allosteric modulators or chemokines clearly indicate that CXCR3 adopts multiple active states
stabilized by the bound ligand(s). The transmission of the functional response is triggered by diverse
molecular switches responsible for the phenomenon of ligand-directed signaling. Our findings provide
further understanding of the structural basis of allosteric modulation in the chemokine receptors with
special emphasis on probe-dependence, and can help guide structure-based design and optimization of
allosteric ligands with improved physiological properties.
30