Thermodynamic characterization of new positive allosteric modulators binding to the glutamate receptor A2 ligand-binding domain: Combining experimental and computational methods unravels differences in driving forces

Article abstract of DOI:10.1021/ci500559b

Positive allosteric modulation of the ionotropic glutamate receptor GluA2 presents a potential treatment of cognitive disorders, for example, Alzheimer's disease. In the present study, we describe the synthesis, pharmacology, and thermodynamic studies of a series of monofluoro-substituted 3,4-dihydro-2H-1,2,4-benzothiadiazine 1,1-dioxides. Measurements of ligand binding by isothermal titration calorimetry (ITC) showed similar binding affinities for the modulator series at the GluA2 LBD but differences in the thermodynamic driving forces. Binding of 5c (7-F) and 6 (no-F) is enthalpy driven, and 5a (5-F) and 5b (6-F) are entropy driven. For 5d (8-F), both quantities were equal in size. Thermodynamic integration (TI) and one-step perturbation (OSP) were used to calculate the relative binding affinity of the modulators. The OSP calculations had a higher predictive power than those from TI, and combined with the shorter total simulation time, we found the OSP method to be more effective for this setup. Furthermore, from the molecular dynamics simulations, we extracted the enthalpies and entropies, and along with the ITC data, this suggested that the differences in binding free energies are largely explained by the direct ligand-surrounding enthalpies. Furthermore, we used the OSP setup to predict binding affinities for a series of polysubstituted fluorine compounds and monosubstituted methyl compounds and used these predictions to characterize the modulator binding pocket for this scaffold of positive allosteric modulators.

Full text of DOI:10.1021/ci500559b

Subscriber access provided by University of Washington | Libraries
Article
Thermodynamic Characterization of New Positive
Allosteric Modulators Binding to the Glutamate Receptor
A2 Ligand-Binding Domain: Combining Experimental and
Computational Methods Unravels Differences in Driving Forces
Ann-Beth Nørholm, Pierre Francotte, Eric Goffin, Iuliana Botez, Laurence Danober,
Pierre Lestage, Bernard Pirotte, Jette S. Kastrup, Lars Olsen, and Chris Oostenbrink
J. Chem. Inf. Model., Just Accepted Manuscript • DOI: 10.1021/ci500559b • Publication Date (Web): 24 Nov 2014
Downloaded from http://pubs.acs.org on November 29, 2014
Just Accepted
“Just Accepted” manuscripts have been peer-reviewed and accepted for publication. They are posted
online prior to technical editing, formatting for publication and author proofing. The American Chemical
Society provides “Just Accepted” as a free service to the research community to expedite the
dissemination of scientific material as soon as possible after acceptance. “Just Accepted” manuscripts
appear in full in PDF format accompanied by an HTML abstract. “Just Accepted” manuscripts have been
fully peer reviewed, but should not be considered the official version of record. They are accessible to all
readers and citable by the Digital Object Identifier (DOI®). “Just Accepted” is an optional service offered
to authors. Therefore, the “Just Accepted” Web site may not include all articles that will be published
in the journal. After a manuscript is technically edited and formatted, it will be removed from the “Just
Accepted” Web site and published as an ASAP article. Note that technical editing may introduce minor
changes to the manuscript text and/or graphics which could affect content, and all legal disclaimers
and ethical guidelines that apply to the journal pertain. ACS cannot be held responsible for errors
or consequences arising from the use of information contained in these “Just Accepted” manuscripts.
Journal of Chemical Information and Modeling is published by the American
Chemical Society. 1155 Sixteenth Street N.W., Washington, DC 20036
Published by American Chemical Society. Copyright © American Chemical Society.
However, no copyright claim is made to original U.S. Government works, or works
produced by employees of any Commonwealth realm Crown government in the
course of their duties.

Page 1 of 48
Journal of Chemical Information and Modeling
1
2
3
4
5
6
7
8
9
Thermodynamic Characterization of New Positive Allosteric Modulators Binding to the
Glutamate Receptor A2 Ligand-Binding Domain: Combining Experimental and
Computational Methods Unravels Differences in Driving Forces
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
Ann-Beth Nørholm1,2, Pierre Francotte3, Eric Goffin3, Iuliana Botez4, Laurence Danober5, Pierre
Lestage5, Bernard Pirotte3, Jette S. Kastrup1, Lars Olsen1, and Chris Oostenbrink2,*
Address:
1Department of Drug Design and Pharmacology, Faculty of Health and Medical Sciences,
University of Copenhagen, Universitetsparken 2, DK-2100 Copenhagen, Denmark.
2Institute of Molecular Modeling and Simulation, University of Natural Resources and Life
Sciences, Muthgasse 18, A-1190 Vienna, Austria.
3Department of Medicinal Chemistry, Center for Interdisciplinary Research on Medicines (CIRM),
University of Liege, Avenue de l’Hôpital, 1, B36, B-4000 Liège, Belgium.
4Pôle d’Expertise Chimie-Chimie Médicinale, Institut de Recherches Servier, 129 Chemin de
Ronde, F-78290 Croissy-sur-Seine, France
5Pôle d’Innovation Thérapeutique en Neuropsychiatrie, Institut de Recherches Servier, 129 Chemin
de Ronde, F-78290 Croissy-sur-Seine, France
* Corresponding author e-mail: chris.oostenbrink@boku.ac.at
1
ACS Paragon Plus Environment

Journal of Chemical Information and Modeling
Page 2 of 48
1
2
3
4
5
6
7
8
9
Keywords: Ionotropic glutamate receptor A2, positive allosteric modulators, isothermal titration
calorimetry, thermodynamic integration, one step perturbation, enthalpy-entropy compensation
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
Abbreviations
ADHD, attention deficit hyperactivity disorder; AMPA, α-amino-3-hydroxy-5-methylisoxazole-4-
propionic acid; BTD, 3,4-dihydro-2H-1,2,4-benzothiadiazine 1,1-dioxide; CNS, central nervous
system; GluA2, AMPA receptor subunit 2; ITC, isothermal titration calorimetry; LBD, ligand-
binding domain; MD, molecular dynamics; OSP, one step perturbation; PEG, poly(ethylene glycol);
RMS, root mean square; TI, thermodynamic integration; TMS, tetramethylsilane.
Abstract
Positive allosteric modulation of the ionotropic glutamate receptor GluA2 presents a potential
treatment of cognitive disorders, e.g. Alzheimer’s disease. In the present study, we describe the
synthesis, pharmacology, and thermodynamic studies of a series of monofluoro-substituted 3,4-
dihydro-2H-1,2,4-benzothiadiazine 1,1-dioxides. Measurements of ligand binding by isothermal
titration calorimetry (ITC) showed similar binding affinities for the modulator series at the GluA2
LBD, but differences in the thermodynamic driving forces. Binding of 5c (7-F) and 6 (no-F) is
enthalpy driven, 5a (5-F) and 5b (6-F) are entropy driven, and for 5d (8-F) both quantities were
equal in size. Thermodynamic integration (TI) and one step perturbation (OSP) were used to
calculate the relative binding affinity of the modulators. The OSP calculations had a higher
predictive power than those from TI, and combined with the shorter total simulation time, we found
the OSP method to be more effective for this setup. Furthermore, from the molecular dynamics
simulations we extracted the enthalpies and entropies and along with the ITC data, this suggested
2
ACS Paragon Plus Environment

Page 3 of 48
Journal of Chemical Information and Modeling
1
2
3
4
5
6
7
8
9
that the differences in binding free energies are largely explained by the direct ligand-surrounding
enthalpies. Furthermore, we used the OSP setup to predict binding affinities for a series of
polysubstituted fluorine compounds and monosubstituted methyl compounds, and used these
predictions to characterize the modulator binding pocket for this scaffold of positive allosteric
modulators.
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
3
ACS Paragon Plus Environment

Journal of Chemical Information and Modeling
Page 4 of 48
1
2
3
4
5
Introduction
6
7
8
9
The α-amino-3-hydroxy-5-methylisoxazole-4-propionic acid (AMPA) receptors mediate influx of
metal ions across post-synaptic membranes in the mammalian central nervous system (CNS) in
response to the neurotransmitter glutamate.1 AMPA receptors form homo- and heterotetrameric ion
channels; each subunit consists of an N-terminal domain (NTD), a ligand-binding domain (LBD), a
membrane-spanning region, and an intracellular C-terminal domain. In the full-length
homotetrameric glutamate receptor A2 (GluA2) the LBD subunits are arranged as a dimer of
dimers.2 The endogenous ligand (S)-glutamate binds in the agonist binding site, whereas positive
allosteric modulators have been shown to bind at the dimer interface (Figure 1A-C). Upon (S)-
glutamate binding, the LBD closes like a clamshell, which leads to a structural rearrangement and
opening of the ion channel.3 The receptor may undergo desensitization, triggered by a
rearrangement of the LBD dimer where (S)-glutamate is still bound but the ion channel is closed.1
Positive allosteric modulators potentiate the effect of (S)-glutamate by either stabilizing the (S)-
glutamate bound conformation and thus slowing deactivation or by stabilizing the subunit interface,
postponing the interface rearrangement and subsequent closure of the ion channel with (S)-
glutamate bound and hence slowing desensitization.4, 5 This fine-tuning of receptor signaling
represents a promising therapeutic strategy in the treatment of several neurological diseases such as
Alzheimer’s disease, schizophrenia, and attention deficit hyperactivity disorder (ADHD).
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
In recent years, significant improvements in the thermodynamic characterization of small-molecule
inhibitor complexes with receptors and enzymes have been made, both by experimental as well as
by computational means. Currently, the main experimental approach is isothermal titration
calorimetry (ITC),6 while computationally, the necessary simulation time has become accessible to
compute free-energy differences from statistical mechanical approaches.7 Whereas successful
calculations of binding free energies are now more common,8 calculation of the individual enthalpic
4
ACS Paragon Plus Environment

Page 5 of 48
Journal of Chemical Information and Modeling
1
2
3
4
5
6
7
8
9
and entropic contributions remain more challenging.9 Furthermore, difficulties in drug design arise
from enthalpy-entropy compensation,10, 11 a thermodynamic concept where changes in the entropic
contribution are counteracted by changes in the enthalpic contribution and vice versa. Moreover, it
was shown that binding affinity is correlated to neither enthalpy nor entropy alone.12 For using
computational methods in drug development, it is important to understand how the individual
interactions in the protein-ligand system contribute to the enthalpy-entropy compensation. It has
been shown that the energetic contributions from the solvent reorganization exactly cancel13-15 and
therefore, the calculation of the direct ligand-surrounding contributions rather than the contributions
of the interactions of the surroundings with itself can lead to additional insight. The contribution of
the interactions of the surroundings with itself are also included in the experimentally determined
data from ITC and thus, the combined use of experimental and computational methods allows for a
more thorough decomposition of the energetic contributions, where the ligand-surrounding energy
terms are analyzed. This provides a more detailed analysis of the energetic contributions that
contribute directly to ligand binding, which would not be possible using experimental methods
alone.
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
In the present study, we report the synthesis and pharmacological characterization of three new
positive allosteric modulators, where the substitution pattern of fluorine on the benzene nucleus of
3,4-dihydro-2H-1,2,4-benzothiadiazine 1,1-dioxide (BTD) is varied. These modulators are derived
from the previously described 4-cyclopropyl-7-fluoro-3,4-dihydro-2H-1,2,4-benzothiadiazine 1,1-
dioxide (5c)16 (Figure 2). We characterized the thermodynamics of binding to the dimeric GluA2
LBD of this modulator series using ITC. To complement the experimentally determined
thermodynamic details of binding, we used both thermodynamic integration (TI) and one step
perturbation (OSP) to set up a method to calculate the relative free energy of binding. TI is a highly
accurate method to estimate free-energy differences between compounds, but it also has a high
5
ACS Paragon Plus Environment

Journal of Chemical Information and Modeling
Page 6 of 48
1
2
3
4
5
6
7
8
9
computational expense.17 The OSP method takes advantage of using a single reference molecule
that is representative of several compounds of interest, and thus significantly improves the
efficiency.18 From these simulations we extracted the reduced enthalpies and entropies to perform a
detailed analysis of the direct enthalpy-entropy contributions to ligand binding. Finally, we used the
OSP data to make predictions of new compounds covered by the reference molecule, allowing for a
detailed QSAR analysis of the modulator binding pocket. To our best knowledge, the present work
represents the first free energy calculations of GluA2 LBD positive allosteric modulators.
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
Materials and Methods
Chemistry
All commercial chemicals (Sigma-Aldrich, Belgium; Appolo Scientific, United Kingdom and
Fluorochem, United Kingdom) and solvents were reagent grade and used without further
purification. Melting points were determined on a Stuart SMP3 apparatus in open capillary tubes
and are uncorrected. NMR spectra were recorded on a Bruker Avance 500 spectrometer (1H: 500
MHz; 13C: 125 MHz) using DMSO-d6 as solvent and tetramethylsilane (TMS) as internal standard;
chemical shifts are reported in δ values (ppm) relative to internal TMS. Elemental analyses (C, H,
N, S) were carried out on a Thermo Flash EA 1112 series elemental analyser and were within ±
0.4% of the theoretical values. This analytical method certified a purity of ≥95% for each tested
compound. All reactions were followed by TLC (silica gel 60F254 Merck) and visualization was
accomplished with UV light (254 or 366 nm). The synthesis of 5c starting from 2,5-
difluorobenzenesulfonamide (2c) has been previously described.16 Synthesis of the other
compounds was performed according to the scheme in Figure 3.
General procedure for the synthesis of fluoro-substituted 2-fluorobenzenesulfonamides (2):
Glacial acetic acid (30 mL) introduced in a round bottom flask (500 mL) was saturated for 30 min.
6
ACS Paragon Plus Environment

Page 7 of 48
Journal of Chemical Information and Modeling
1
2
3
4
5
6
7
8
9
with gaseous sulfur dioxide. An aqueous solution of cupric chloride (1.5 g in 10 mL) was added to
this solution giving a suspension of cuprous chloride (suspension A). In another flask, the
appropriate fluoro-substituted 2-fluoroaniline (1; 5 g) was dissolved in a mixture of glacial acetic
acid (30 mL) and 12N HCl (15 mL) and the resulting solution (solution B) was cooled on an ice/salt
bath (-5 °C). A solution of sodium nitrite (2.5 g) in water (10 mL) was added drop wise to solution
B and the resulting mixture was slowly added to suspension A and stirred on an ice bath for 15 min.
The reaction mixture was then poured on water (200 mL) and diethyl ether (200 mL). The organic
layer was separated, washed with water (100 mL) and concentrated to dryness under reduced
pressure. The residue of the fluoro-substituted 2-fluorobenzenesulfonyl chloride was dissolved in
dioxane (25 mL) and this solution was poured under stirring onto a cooled mixture of concentrated
ammonia (25 mL) and water (10 mL). After 30 min, the reaction mixture was concentrated under
reduced pressure and the resulting precipitate of the final compound was collected by filtration,
washed with water, and purified by crystallization in methanol-water (yield: 60-70%).
2,6-Difluorobenzenesulfonamide (2d) was used from a commercial source (Fluorochem, ref
017310).
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
General procedure for the synthesis of fluoro-substituted 2-cyclopropylaminobenzene-
sulfonamides (3): A solution of the appropriate fluoro-substituted 2-fluorobenzenesulfonamide (2)
(3 g) in dioxane (30 mL) and cyclopropylamine (3 mL) was heated in a closed vessel at 100-110 °C
for 24 h. The solvent and the excess of amine were removed by distillation under reduced pressure
and the residue was dissolved in methanol (20 mL). The methanolic solution was cooled and mixed
with water (60 mL) under stirring. The resulting precipitate was collected by filtration, washed with
water, dried and used in the next step without further purification (yield: 80-90%).
7
ACS Paragon Plus Environment

Journal of Chemical Information and Modeling
Page 8 of 48
1
2
3
4
5
6
7
8
9
General procedure for the synthesis of fluoro-substituted 4-cyclopropyl-4H-1,2,4-
benzothiadiazine 1,1-dioxides (4): In an open vessel, a mixture of the appropriate fluoro-
substituted 2-cyclopropylaminobenzenesulfonamide (3) (2 g) and triethyl orthoformate (20 mL) was
heated at 130-150 °C for 24-48 h. The resulting suspension was cooled on an ice bath and the
insoluble material was collected by filtration, washed with diethyl ether, dried and crystallized in
methanol (yield: 70-80%).
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
General procedure for the synthesis of fluoro-substituted 4-cyclopropyl-3,4-dihydro-2H-1,2,4-
benzothiadiazine 1,1-dioxide (5): A mixture of the appropriate fluoro-substituted 4-cyclopropyl-
4H-1,2,4-benzothiadiazine 1,1-dioxide (4) (1.9 g) and isopropanol (50 mL) was supplemented with
finely divided sodium borohydride (1 g) and then heated at 50-55 °C for 5-10 min. The solvent was
removed by distillation under reduced pressure and the residue was taken up with water (50 mL).
The resulting suspension was slightly acidified by the addition of 6N HCl. The title compound was
extracted three times with dichloromethane (3 x 30 mL). The combined organic layers were dried
over magnesium sulfate and the filtrate was concentrated under reduced pressure. The residue was
crystallized in methanol-water (yield: 80-85%).
Effect on AMPA-evoked membrane depolarization (in vitro fluorescence assay). This assay
was performed on rat primary brain cultures using fluorescent membrane potential dyes and an
imaging based plate reader (FDSS, Hamamatsu, JP) following our previously published
procedure.16, 19
Effect on AMPA-mediated release of noradrenaline on rat hippocampal slices. Potentiation of
noradrenaline release on rat hippocampal slices was measured according to our previously reported
8
ACS Paragon Plus Environment

Page 9 of 48
Journal of Chemical Information and Modeling
1
2
3
4
procedure.20, 21
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
Protein expression and purification
The dimeric GluA2-LBD double mutant L483Y-N754S was expressed and purified as previously
described.22 Briefly, Escherichia coli Origami B (DE3) cells were transformed with the GluA2-
LBD-L483Y-N754S pET-22b(+) plasmid. Cells were grown in LB medium to an OD600 of 0.9
where the cells were subsequently cooled on ice to 20°C. Protein expression was induced by
addition of 0.5 mM isopropyl β-D-thiogalactopyranoside. Cells were harvested and lysed after 18
hours, and the soluble protein was initially purified by Ni2+-affinity chromatography. The N-
terminal His-tag was cleaved by tryptic digest, and the cleaved protein was further purified by
anion-exchange chromatography and size-exclusion chromatography.
Isothermal titration calorimetry
An ITC200 MicroCalorimeter (GE Healthcare) with a cell volume of 200 µL was used for ITC at
25ºC. Titrations with 5b, 5d, and 6 were direct titrations, whereas the binding affinity of 5a was
determined in a displacement assay with 5c. The protein solution consisted of GluA2-LBD-L483Y-
N754S in 100 mM HEPES, 100 mM NaCl, 2 mM KCl, 5 mM L-glutamate, pH 7.0. UV absorption
was used to determine the protein concentration, which was: 47 µM (5b titration), 31 µM (6
titration), 29 µM (5d titration), and 20 µM (5a titration). The protein solution buffer was also used
to dissolve the compounds to the following concentrations used for ITC: 695 µM 5b, 490 µM 6,
and 543 µM 5d. For the displacement ITC assay of 5a with 5c, 285 µM 5c was used to displace 100
µM 5a. All experiments were set up with 20 injections with 3 min intervals; first injection was 0.4
µL, the remaining 2 µL. Data analysis was performed using the Origin 7.0 software (MicroCal)
9
ACS Paragon Plus Environment

Journal of Chemical Information and Modeling
Page 10 of 48
1
2
3
4
5
6
7
8
9
(single binding site model). For each binding curve the first data point was discarded and the heat of
dilution of injecting ligand into buffer was subtracted prior to fitting. All experiments were
performed three times and the reported Kd, ΔH, and –TΔS are mean values of three independent
titrations.
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
Simulation setup
The GROMOS11 suite of simulation programs was used for all simulations.23 The GROMOS++
suite of programs was used to set up and analyse the simulations.24
The coordinates for the protein were extracted from the crystal structure of GluA2-LBD-L483Y-
N754S in complex with BPAM97 at 1.95 Å (PDB code 3TDJ; the X-ray structure of GluA2-LBD-
L483Y-N754S in complex with BPAM344 was not available when the simulations were
initiated).22 The GROMOS 54a8 force field was applied for MD simulations.25 The endogenous
ligand L-glutamate was described according to the same force field in the zwitter-ionic form and
bearing a net charge of -1 e. Building blocks for the allosteric modulators were derived from similar
functional groups available in the force field. In particular, the sulphonamide group was previously
described in simulations of sildenafil26 and the charge distributions for the fluorine-substituents are
taken from a model of tri-fluoroethanol.27 The building blocks of all species are available in the
Supplementary Material. For the protein, all Glu, Asp, Lys, and Arg residues were described in their
charged states while the optimal protonation states of His (neutral) were chosen such that the
potential number of hydrogen bonds is maximised.
The protein-ligand complex was placed in a rectangular box with a size of 7.8 x 8.7 x 9.9 nm3 with
a minimum solute-solvent distance of 0.23 nm, with 13 Cl- and 5 Na+ ions added and containing
approximately 19000 simple point charge (SPC) water molecules. To obtain the free-energy
10
ACS Paragon Plus Environment

Page 11 of 48
Journal of Chemical Information and Modeling
1
2
3
4
5
6
7
8
9
differences of the free compounds in water, which are required to calculate the free energy of
binding, all real compounds and the reference compound (see below) were also simulated free in
solution. For this, a box with 727 SPC water molecules was used.
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
To relax unfavorable contacts between the protein-ligand complex and the solvent or ligand and
solvent, all systems were energy minimized before simulations, using a steepest descent energy
minimization with an energy threshold of 0.1 kJ mol-1. Initial velocities were subsequently assigned
randomly from a Maxwell-Boltzmann distribution corresponding to 60 K. In the thermalisation and
equilibration phase, the temperature was gradually increased from 60 K to 300 K using the weak
coupling scheme28 over six discrete simulation steps, while decreasing the force constant of an
initial positional restraint from 25000 kJ mol-1 nm-2 to zero. In a seventh step, the Nosé-Hoover
chains algorithm29 was used involving three chains and a reference temperature of 300 K.
Preliminary simulations showed rotations of the modulator up to +/- 90º relative to the crystal
structure and therefore, a distance restraint was applied between the centre of geometry of C4, C6,
and C10 of one modulator to the F2 on the other modulator with an optimal distance of 0.36 nm and
a force constant of 4000 kJ mol-1 nm-2 (for atom numbering, see the molecular building block in the
Supplementary Material).
All production simulations were performed with a constant volume, a constant temperature (300 K),
and a constant number of particles. Stability of the simulations was confirmed by calculating the
atom-positional root-mean-square deviations with respect to the initial starting structure and by
monitoring the occurrence of hydrogen bonds and secondary structure elements over the course of
the simulations. The presence of hydrogen bonds was determined using a geometric criterion
requiring a maximum hydrogen-acceptor distance of 0.25 nm and a minimum donor-hydrogen-
11
ACS Paragon Plus Environment

Journal of Chemical Information and Modeling
Page 12 of 48
1
2
3
4
5
6
7
acceptor angle of 135°.30 The occurrence of secondary structure elements was determined using the
Kabsch-Sander rules,31 as implemented in the GROMOS++ suite of analysis programs.
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
Free-energy calculations
The relative free energy of binding between two compounds A and B can be calculated using a
thermodynamic cycle,32 from which the following equations can be derived:
ΔΔGbind = ΔGbind (B) - ΔGbind (A) = ΔGAB (prot) – ΔGAB (wat)
(1)
where ΔGbind (B) and ΔGbind (A) represent the binding free energy of compound B and A,
respectively, and ΔGAB is the free energy of changing compound A into compound B as calculated
in the protein (prot) or free in solution (wat).
Thermodynamic integration
In thermodynamic integration (TI) ΔGAB is calculated by gradually modifying compound A into
compound B using a coupling parameter, λ, which connects the Hamiltonians describing the two
compounds. The Hamiltonian, H(λ), is parameterized such that at λ = 0 it represents compound A
(H(0) = HA) and at λ = 1 it represents compound B (H(1) = HB). Here, the GROMOS
parameterization of the Hamiltonian is used33 with a softcore potential34 with softness parameters
αLJ = 0.5 and αCRF = 0.5 nm2.
The derivative of the Hamiltonian with respect to λ is calculated at each λ point and the ensemble
average is numerically integrated providing the free energy difference between compounds A and
B.35 In the current work, performed at constant volume, we calculate the Helmholtz free-energy
12
ACS Paragon Plus Environment

Page 13 of 48
Journal of Chemical Information and Modeling
1
2
3
4
5
6
difference, ΔA. For the small alchemical changes described here, the volume-pressure work will be
7
8
negligible and the Helmholtz and Gibbs free energies will be directly comparable.
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
!!!
!!!
!!(!)
!!
∆ꢀ ≈ ∆ꢀ =
ꢀꢁ
(2)
In practice, the ensemble average, <ꢀH(λ)/ꢀλ>, is calculated from at least 11 independent
simulations performed at 11 equidistant λ-values. At every λ-value, a short equilibration was
performed for 100 ps, followed by production simulations of at least 1 ns. To obtain a smooth TI
curve up to three additional simulations were performed at intermediate λ-values. The free-energy
differences between the allosteric modulators were calculated as outlined in Figure 4A for the
compounds free in solution and while bound to the protein. In the protein simulations, both
modulators were modified simultaneously, with a distance restraint between them. The ensemble
averages were unbiased prior to the integration as in umbrella sampling36, 37 using
!!
!!
!
/!
!
!
!"#$
!
!
!!
!!
=
(3)
!"
!
/!
!
!
!"#$
!
!
where Ubias is the energy of the restraint, the subscript b indicates the ensemble averages obtained in
the presence of the bias while the subscript ub indicates the ensemble average in the unbiased case,
kB is the Boltzmann constant, and T is the absolute temperature.
One step perturbation
To estimate the free energy difference using OSP,38 a reference molecule, which can sample the
relevant phase space for all real compounds, was designed (Figure 4B). The challenge lies in
designing the reference compound to have the best possible resemblance to all real compounds.18
The reference compound contains four soft fluorine atoms for which the charge was set to 0 and the
13
ACS Paragon Plus Environment

Journal of Chemical Information and Modeling
Page 14 of 48
1
2
3
4
5
van der Waals interaction was softened34 to remove the singularity at the origin with a softness
6
7
8
parameter of αLJ = 1.51.39
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
To calculate the free energy difference between a real compound and the reference compound, the
Zwanzig perturbation formula was applied to a single simulation of the reference state.40 Thus, the
free energy difference between a real compound, A, and the reference compound, R, is calculated
via:
!
!
/! !
!
∆ꢀ!" = ꢀ! − ꢀ! = −ꢀ!ꢀln ꢀ! ! !!
(4)
!
where angular brackets indicate the ensemble average obtained using molecular dynamics (MD)
simulation of the reference compound, R. Again, in view of the small perturbations considered here,
we use the approximation ΔGRA ≈ ΔARA. In this particular case HR also contains the restraining
energy between the two reference molecules, such that the free energy of releasing the restraint is
included in ΔGRA. The free energy difference between the real compounds is subsequently
calculated according to ΔGAB = ΔGRB – ΔGRA. Production simulations of the reference state free in
solution and when bound to the protein were performed for 10 ns, writing the coordinates every 0.5
ps for later analysis.
Enthalpy-entropy compensation
Theoretically, the enthalpic and entropic contributions to the relative binding free energies are also
accessible from the simulations. The enthalpy difference can be obtained from the difference in the
average total energy for two compounds, incremented by the volume-pressure work:
∆ꢀ!" = ꢀ! ! − ꢀ! ! + ꢀ∆ꢀ
(5)
14
ACS Paragon Plus Environment

Page 15 of 48
Journal of Chemical Information and Modeling
1
2
3
4
5
6
7
8
9
where <HA>A and <HB>B indicate the ensemble average of the Hamiltonian of compound A and B,
respectively, as calculated over the ensemble for compounds A and B. The volume-pressure work
as appropriate in simulations in the NVT ensemble and due to the alchemical modifications
described in this work will be negligible. The entropy is subsequently calculated from the difference
between the enthalpy and free energy.41
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
However, equation (5) is known to converge poorly, as it is based on the total energy of the entire
system. Still, for a pairwise additive force field, the total enthalpic change can be divided into the
contributions due to the ligand and its surroundings as well as the contributions due to the
interactions within the surroundings (protein and solvent).
!"
∆ꢀ = ꢀ!!" ! − ꢀ!!"
(6)
!
!"
with Hls describing the ligand-surrounding energy terms of the Hamiltonian. It can be proven that
the surrounding-surrounding enthalpies are explicitly included in the corresponding entropy change,
largely explaining the often-observed enthalpy-entropy compensation.15 The reduced term ΔHls
does not correspond to an experimental observable, but can be considered as the direct enthalpic
driving force. Moreover, it can be readily computed from molecular simulations and may help to
dissect the origins of the experimentally determined thermodynamic quantities. Equation 6 was
applied directly to the end-state simulations of the TI calculations.
Results and Discussion
Chemistry
The synthesis of the monofluoro-substituted 4-cyclopropyl-3,4-dihydro-2H-1,2,4-benzothiadiazine
1,1-dioxides 5 is described in Figure 3. The appropriate o-fluoroaniline 1 bearing a second fluorine
15
ACS Paragon Plus Environment

Journal of Chemical Information and Modeling
Page 16 of 48
1
2
3
4
5
6
7
8
9
atom in another position of the benzene nucleus was converted into the corresponding o-
fluorobenzenesulfonamide 2 by means of the Meerwein variation of the Sandmeyer diazotization
reaction.42 Nucleophilic substitution of the fluorine atom of intermediates 2 by cyclopropylamine,
providing the corresponding 2-cyclopropylaminobenzenesulfonamides 3, was performed in a closed
vessel by heating the mixture at 130-150°C for 24-48 h. The use of o-fluorobenzenesulfonamides
instead of the corresponding o-chlorobenzenesulfonamides was preferred because of the greater
reactivity of fluoro-substituted aromatic compounds to nucleophilic substitution. Moreover, this
reaction was greatly facilitated by the fact that the fluorine atom was located at the ortho position of
the electron-withdrawing sulfonamide group. In the case of intermediate 2b, which bears a fluorine
atom in the para-position of the sulfonamide group, thus in competition with the ortho-position for
the reaction with the amine, the nucleophilic substitution was mainly observed in the ortho-position.
Ring closure of intermediates 3 by heating them in triethyl orthoformate provided the fluoro-
substituted 4-cyclopropyl-4H-1,2,4-benzothiadiazine 1,1-dioxides 4, which were converted with
sodium borohydride into the corresponding “saturated” final compounds 5. The characterization of
all compounds is given in the Supplementary Material.
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
Biological evaluation
The new fluoro-substituted BTD compounds 5a (5-F), 5b (6-F), and 5d (8-F) were evaluated as
AMPA receptor potentiators in an in vitro fluorescence assay and their activity was compared to
that of the previously reported BTD compound 5c (7-F) as well as the unsubstituted analogue 6.16
The experiments were performed on primary cultures of neurons from rat embryonic cortex,
measuring the effect of the modulators on AMPA-evoked membrane depolarization, as previously
described.16, 19 For each compound, the EC2x value was determined (concentration of modulator
giving a 2-fold increase of the fluorescence induced by 300 µM AMPA), as well as their maximum
16
ACS Paragon Plus Environment

Page 17 of 48
Journal of Chemical Information and Modeling
1
2
3
4
5
6
7
8
9
effect (Emax value normalized to unity for AMPA-evoked response taken as x1) (Table 1). It was
observed that the 8-fluoro-substituted compound 5d, with an EC2x value of 0.33 µM and a high
value of maximal potentiation of x15, exhibited a potentiator effect on AMPA receptors comparable
to that of the 7-fluoro-substituted reference compound 5c,16 while their 5- and 6-fluoro-substituted
analogues, although very active (EC2x ~ 2 µM), where clearly less efficient. This result indicated
that the best positions for a fluorine atom on the BTD ring system are the 7- and the 8-positions.
Such an observation was already made with their corresponding monochloro-substituted BTDs.
However, the fluoro-substituted analogues were systematically found to be more active on AMPA
receptors than the corresponding chloro-substituted compounds.43
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
It is known that positive allosteric modulators of the AMPA receptor are able to potentiate
noradrenaline (NA) release in rat hippocampal slices, an effect linked to their interaction with
44
presynaptic AMPA receptors.21,
At 300 µM, most of the compounds induced a strong
enhancement of the (S)-AMPA (10 µM) evoked [3H]-NA release (100% representing the effects
shown by (S)-AMPA alone). This result indicated that the new monofluoro-substituted compounds
5a, 5b, and 5d, as previously observed with compounds 5c and 6,16 were able to act on presynaptic
AMPA receptors, almost with the same efficacy.
Thermodynamic details of modulator binding
To measure modulator binding independently of dimer formation, we used a double mutant of the
GluA2 LBD, GluA2-LBD-L483Y-N754S. This mutant is a preformed dimer in solution, which
mimics the endogenous binding environment found in the dimer of dimers structure of the full-
length GluA2 receptor. We have previously used this mutant for ITC measurements, and the
mutations do not interfere with modulator binding.16, 22
17
ACS Paragon Plus Environment

Journal of Chemical Information and Modeling
Page 18 of 48
1
2
3
4
5
6
7
8
9
For compounds 5b, 5d, and 6 the thermodynamic details of binding were determined in a direct
titration with GluA2. For 5a we were not able to measure a saturation curve in a direct titration and
therefore, the thermodynamic properties were determined in an ITC displacement assay with the
previously reported 5c.16 The ITC experiments showed that for all four compounds, binding is an
exothermic process (Figure 5). The binding affinities (Kd) and thermodynamic quantities are given
in Table 2. The following rank order in binding affinity (Kd) was observed, with the most potent
compound listed first: 5c (7-F) > 5d (8-F) > 6 (no-F) > 5b (6-F) > 5a (5-F). For compound 6 the
complex formation is primarily enthalpy driven as also previously seen for 5c. On the contrary, the
complex formation for 5a and 5b is primarily entropy driven, whereas 5d complex formation is
equally enthalpy and entropy driven (Table 2).
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
To decipher the mode of action of the modulators, we compared the relative difference in EC2x for a
compound x (EC2x(x)/EC2x(5c)) with the relative difference in the Kd: (Kd(x)/Kd(5c)). A direct
comparison between EC2x and Kd is not possible, since the EC2x is the concentration of modulator
that leads to two-fold increase in receptor response to a given agonist compared to the agonist alone
and only indirectly contains information on binding affinity. The comparison of the relative
differences in EC2x to the ITC data enables a decomposition of the EC2x values into the binding
effect and the modulatory strength. Thus, it suggests if the relative difference between the effects of
two compounds is purely due to binding affinity or if it is due to a more efficient modulation: If the
EC2x(x)/EC2x(5c) equals Kd(x)/Kd(5c), it suggests that these compounds are different in their
binding affinity only and hence, the two compounds should have similar modulatory effects. If
EC2x(x)/EC2x(5c) does not equal Kd(x)/Kd(5c), then binding affinity alone does not explain the
differences and therefore, the two compounds likely have different modulatory strengths. From data
in Table 1 and Table 2, it is seen that for all compounds the EC2x(x)/EC2x(5c) is lower than
Kd(x)/Kd(5c). Thus, differences in the binding affinity between the modulators are most likely not
18
ACS Paragon Plus Environment

Page 19 of 48
Journal of Chemical Information and Modeling
1
2
3
4
5
6
7
8
9
the sole cause of the difference between these compounds. The largest difference is seen for 5a
relative to 5c where Kd(5a)/Kd(5c) amounts to 35, while EC2x(5a)/EC2x(5c) is only 8. The other
three compounds show smaller differences between the fractions. This suggests that 5a may have a
stronger modulatory effect than 5c, which is partially compensated by a weaker binding of 5a
relative to 5c. Therefore, it may be of interest to improve the binding affinity by introducing
modifications elsewhere in 5a.
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
Stability of the simulations
To set up a system for calculation of binding affinities, we used the GluA2-LBD-L483Y-N754S
protein in complex with either two copies of 5c for TI (Figure 4A) or two copies of the reference
molecule for OSP (Figure 4B).
To ensure that the MD simulations represent stable protein structures, we evaluated the root mean
square (RMS) deviation of the protein backbone from the initial starting structure. The RMS
deviation remained below 0.25 nm, which indicates stability of the protein. Furthermore, the
average occurrence of hydrogen bonds as well as the occurrence of secondary structural elements
remained stable throughout the simulations (data not shown). However, a closer inspection of
modulator binding in the protein showed that the modulators were turning ±90º relative to the
crystal structure of GluA2-LBD-L483Y-N754S with 5c (PDB code 4N07) in all MD simulations.
Therefore, we used two distance restraints between the two modulator molecules located at the
binding site, from the fluorine atom of one modulator to the centre of geometry on the benzene ring
of the other modulator and vice versa (see methods).
Free energy calculations
19
ACS Paragon Plus Environment

Journal of Chemical Information and Modeling
Page 20 of 48
1
2
3
4
5
6
7
8
9
The results of the free energy calculations using both TI and OSP are shown in Table 3 along with
the relative free energy differences calculated from ITC. As a consistency check, the free energy
differences from 5d to 5a, from 5a to 6, from 6 to 5b, and from 5b to 5d were calculated in water as
well, and the maximum free energy difference along the closed cycles was -0.19 kJ/mol. Comparing
the relative free energy calculated by TI with the free energy determined by ITC, we get an RMS
error of 5.1 kJ/mol, which is slightly higher than what is typically considered ‘chemical accuracy’
(~4 kJ/mol),45 but just below 5.6 kJ/mol corresponding to a factor 10 in Kd at room temperature.
This high RMS error of the TI calculations is primarily due to the poor prediction of 5a relative to
5c (-0.4 kJ/mol vs. 8.8 kJ/mol determined by ITC), but also the prediction for 6 is relatively poor
(2.5 kJ/mol vs. 6.0 kJ/mol), although still within chemical accuracy.
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
Next, we calculated the free energy of binding using the OSP method, which is a computationally
more efficient method than TI. When we compare the OSP free energy calculations with the relative
free energy differences from ITC we obtain an RMS error of 2.1 kJ/mol, which is even below
thermal fluctuations (kBT = 2.5 kJ/mol). Thus, for this series of compounds we find that the OSP
setup gives better predictions. Similar to the TI calculations the relative free energy of binding of 5a
was the most difficult to calculate (11.9 kJ/mol vs. 8.8 kJ/mol). However, the OSP method performs
slightly better than the TI method for the difference between 5c and 6, and conversely, slightly
worse for 5b.
Thus, the calculations of the relative binding free energy of 5a showed the poorest correlation for
both TI and OSP in comparison to the experimentally determined relative free energies from ITC.
For this setup we found the OSP method to have a somewhat higher predictive power compared to
the TI method. This is promising because for the four TI calculations in Figure 4A, we have
performed at least 4 x 11 x 1 ns = 44 ns of simulation, while the OSP calculations are based on a
single simulation of 10 ns of the reference state.
20
ACS Paragon Plus Environment

Page 21 of 48
Journal of Chemical Information and Modeling
1
2
3
4
5
6
7
Ligand orientations, conformations and interactions
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
The modulators bind in the dimeric interface of the GluA2 ligand binding domain. The only direct
hydrogen bond observed is between the N2 hydrogen and the carbonyl oxygen of Pro494 (Figure
1B & 1C.16, 22 For 5c and its analogue BPAM97, a weak hydrogen-fluorine bond has been observed
to Ser497, otherwise the binding of the modulators is stabilized primarily through non-bonded
interactions and shape complementarities. To check that the experimentally observed binding pose
was also seen in the simulations, we examined the final structures from the end-state simulations of
the TI and compared them to the experimentally observed structure of 5c (Figure 6). The structures
of 5b and 6 match best with the experimental structures for both copies of the modulators (Figure
6A and 6C). An alternative puckering of the ring system relative to the experimental structure was
observed for 5d in one of the modulators (Figure 6B). The puckering seen in the crystal structure
enables the only hydrogen bond from the N2 hydrogen of the modulator to the carbonyl oxygen of
Pro494.
The advantage of the OSP method lies in the usage of a reference structure, which may sample all
relevant conformations of multiple molecules of interest. For the OSP calculations, the number of
structures in the trajectories that contribute significantly to the free energy calculation was
determined for each real compound by counting the number of structures for which the energy
difference HA – HR in equation (4) was less than the final free energy difference increased by kBT
(HA – HR < ΔGAR + kBT). The number of contributing structures was found to be in the range of 0.3-
12% of the entire set of 20,000 structures, with the lowest amount of structures contributing to the
free energy estimates of compound 5a.
21
ACS Paragon Plus Environment

Journal of Chemical Information and Modeling
Page 22 of 48
1
2
3
4
5
6
7
8
9
We extracted the single structure, which contributed the most to the free energy calculation for each
of the real compounds from the OSP simulation. Figures 7A and 7C show the front view of the
single structures with the highest probability for each of the compounds in line representation
together with the X-ray crystal structure of 5c in complex with GluA2-LBD-L483Y-N754S. The
RMS deviations of the most contributing ligand structures were found to be: 2.4 Å (5a), 1.3 Å (5b),
0.95 Å (5c), 2.6 Å (5d), and 1.1 Å (6), with respect to the X-ray structure of 5c. For both copies of
the compounds some translation of the simulated structures relative to the crystal structure (in
green) is seen, the best overlap with the X-ray structure is seen for 5c and 6 (Figure 7A and 7C).
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
The side view of the same structures reveals a different puckering of the ring system for 5a and 5d
in both copies of the modulator (Figures 7B and 7D) relative to the puckering observed in the X-ray
structure of 5c. The observation of both conformations indicates that the reference structure
simulates a broad ensemble of conformations and the relevant structures for the individual
compounds may be selected. Furthermore, the cyclopropyl group is rotated 180º for 5a and 5d
(Figure 7B and 7D) and 90º for 5b (Figure 7B) relative to the conformation seen in the X-ray
crystal structure of 5c in GluA2-LBD-L483Y-N754S.
We also calculated the average puckering for the real compounds by reweighting the reference state
simulation (data not shown) and confirmed our observations of the single highest contributing
structures. Comparing the results from the TI and OSP, it is seen that 5a consistently goes to the
alternative puckering in the OSP, while it remains in the experimentally observed puckering in the
TI calculations. This may explain the differences observed between the free energy values using TI
and OSP (Table 3).
To see if the two different ring conformations represent an intrinsic effect of the compounds or if
they are induced by the protein, we also extracted the highest contributing structure for each of the
22
ACS Paragon Plus Environment

Page 23 of 48
Journal of Chemical Information and Modeling
1
2
3
4
5
6
7
8
9
real compounds from the simulation in water (Figures 7E and 7F). Here, the two different ring
conformations were also observed where 5c and 5d showed preference for the conformation seen in
the crystal structure, and 5a, 5b, and 6 were found with the other ring conformation. The
observation of both conformations in the water simulation suggests that there is not a large energy
difference associated with the transition in water and thus, the protein may easily influence the
preference for either of the conformations. 5c and 5a are the only compounds, which preserve the
intrinsic conformation seen in the water simulation when bound to the protein. For 5b, 5d, and 6 the
binding to the protein may result in a slightly shifted orientation, which induces another puckering
than seen in the water simulation (considering only the single most contributing structure). For 5a
the fluorine substituent may influence the puckering, by some steric hindrance caused by the
cyclopropyl group, thus favoring the alternative conformation to the X-ray structure both
intrinsically and in the protein, severely hampering the formation of the N2 hydrogen bond to the
carbonyl oxygen of Pro494, and possibly explaining the low number of contributing structures to its
free energy estimate.
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
These observations suggest that 5c shows the overall combined best modulatory effect and binding
affinity, since it is the only compound with preserved intrinsically favored conformation, which is
also the optimal conformation for hydrogen bonding when bound to the protein. Furthermore, 5d
also shows the alternative puckering of the ring in the protein and this ligand is shifted most to the
right from the X-ray structure. This observation may give a molecular explanation for the
previously suggested difference in modulation effects as compared to compounds 5a, 5b and 5c.
Enthalpy-entropy compensation
23
ACS Paragon Plus Environment

Journal of Chemical Information and Modeling
Page 24 of 48
1
2
3
4
5
6
7
8
9
To investigate the thermodynamic details of ligand binding to the protein, we extracted the ligand-
surrounding enthalpies, ΔHls, from the end-states of the TI simulations, according to equation (6)
(Table 4). TI simulations were preferred over OSP simulation data, since the endstate TI are
simulations of the real molecules. By comparing the computed ligand-surrounding enthalpy to the
values that were obtained experimentally, we computed the changes in enthalpy due to the
reorganization in the surroundings (protein and solvent; ΔHss). For ease of comparison, we have
also explicitly calculated the thermodynamic quantities relative to 5c (ΔΔH, ΔΔHls, and ΔΔHss).
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
Comparing 6 with 5c, we see that the ΔΔG(ITC) of 6.0 kJ/mol is almost entirely explained by the
ligand-surrounding enthalpy change (ΔΔHls(TI) = 6.7 kJ/mol). Since also -TΔΔS(ITC) is almost
zero, the surrounding-surrounding enthalpy is close to zero as well. Hence, for this compound there
is no significant reorganization of the surroundings and the difference in affinity is explained from
the direct interaction energy between the modulator and the protein. This is in line with the very
similar position of 6 at the binding site compared to 5c (Figure 7).
For 5d, the ΔΔG(ITC) is also very similar to the value of ΔΔHls(TI) in Table 4 (5.1 vs. 5.0 kJ/mol).
However, here the experiment indicates that the difference in binding of 5d relative to 5c is the
result of an unfavorable enthalpy (ΔΔH(ITC) = 13.5 kJ/mol), and favorable entropy (-TΔΔS(ITC) =
-8.4 kJ/mol). This effect entirely compensates the enthalpy-change due to reorganization of the
surroundings (ΔΔHss = 8.5 kJ/mol). So even if the experiment suggests entropic differences between
the compounds, a closer analysis reveals that the enthalpic differences in the direct interaction
(ΔΔHls(TI)) may explain the differences in affinity.
Finally, for 5a and 5b the entropically driven binding is compensated by an even larger value of
ΔΔHss. Also for these compounds, the largest portion of ΔΔG may be explained from ΔΔHls(TI),
and the favorable entropic contributions are entirely compensated by enthalpically unfavorable
24
ACS Paragon Plus Environment

Page 25 of 48
Journal of Chemical Information and Modeling
1
2
3
4
5
6
7
8
9
reorganization of the surroundings. Thus, the combination of experimental data with the calculated
ligand-surrounding enthalpies allows us to dissect the origin of the measured enthalpies and
entropies, and could give a better insight into the design of new compounds. It is arguably easier to
design compounds that have more favorable ΔHls, than to design compounds to have an effect on
the reorganization of the surroundings or on a largely compensated entropy.
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
Thus, our calculations of the energetic contributions to ligand binding indicate that enthalpic and
entropic differences for this modulator series have large (compensating) contributions from either
solvent reorganization or interactions between the protein and the solvent. This also shows that the
differences in the binding affinities are mostly explained from direct interactions between the
ligands and their surroundings.
Prediction of binding affinities
The strength of the OSP method lies in the need for only a single simulation of the reference
compound to cover multiple compounds of interest (Figure 4B). Therefore, to characterize the
modulator binding pocket with respect to the 4-cyclopropyl-3,4-dihydro-2H-1,2,4-benzothiadiazine
1,1-dioxide modulator scaffold, we designed a series of additional fluorine-derivatives and
calculated the free energy of these compounds relative to 6 for ease of comparison (Table 5).
Apart from 5c, the 6,7-di-F compound was predicted to have the most favorable free energy
(ΔΔGbind = -2.2 kJ/mol) and the fluorine substituted compound which was predicted to have the
least favorable energy was the 5,6,7,8-tetra-F compound with ΔΔGbind = 24.6 kJ/mol. Comparing
the predicted relative free energy for all the compounds with a fluorine at the 5-position the ΔΔG is
9–24 kJ/mol, whereas for the four remaining fluorine substituted compounds the ΔΔG is (-2)–3
25
ACS Paragon Plus Environment

Journal of Chemical Information and Modeling
Page 26 of 48
1
2
3
4
5
6
7
8
9
kJ/mol (i.e. without fluorine in the 5-position). Thus, an electronegative substituent or hydrogen-
bonding acceptor in the 5-position decreases the predicted binding affinity regardless of the number
and position of other electronegative substituents, in line with the unfavorable binding of 5a
(calculated and experimental).
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
Table 5, combined with the OSP data described in Table 3, allows us to investigate if the
contributions of the F substitutions are additive or not. The free energy of some compounds may be
approximated, such as e.g. for the 5,8-di-F compound (9.4 kJ/mol relative to 6), which is very close
to the sum of the relative free energies for 5a (7.0 kJ/mol) and 5d (1.4 kJ/mol). Further examples of
compounds for which additivity holds reasonably well are the 6,8-di-F and the 6,7-di-F compounds,
while additivity does not hold for the 5,7-di-F, 5,7,8-tri-F, 5,6,7-tri-F, or 5,6,7,8-tetra-F compounds.
This indicates that the predictions generated from the OSP simulation are more thorough than a
simple QSAR model, which can be derived from looking at only the monosubstituted compounds,
and take into account the mutual influences of the substitutions.
To investigate the impact of a small non-polar substituent on the binding affinity, we calculated the
binding free energies relative to 6 for a series of mono-substituted methyl compounds (Table 5).
From these four derivatives, it is seen that the 5-substituted compound binds with the least
favorable energy (ΔΔGbind = 63.3 kJ/mol). The predicted methyl derivative with the most favorable
energy is the 6-methyl compound with ΔΔGbind = 1.8 kJ/mol. Thus, for both mono-methyl and
mono-fluorine compounds the 5-substituted position is predicted to result in the weakest binding
compounds; however, the predictions indicate that it is more favorable to have an electronegative
atom compared to a bulkier methyl substituent in this position. Also, for 5a (the 5-F-compound) the
lowest number of structures contributing to the OSP calculations was observed, and this molecule
favors an alternative puckering. Therefore, an explanation may also be found in a steric clash of any
26
ACS Paragon Plus Environment

Page 27 of 48
Journal of Chemical Information and Modeling
1
2
3
4
5
6
type of substituent in this position, which may favor an alternative puckering of the ring system,
which in turn prevents the hydrogen bond between the modulator and the protein.
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
Comparing 5b (6-F, ΔΔG = 0.1 kJ/mol) with the compound with a methyl in the 6-position (ΔΔG =
1.8 kJ/mol) shows that the two compounds with a substituent in the 6-position bind with similar
relative free energies, thus, at this position there is no apparent preference for either a hydrophobic
or electronegative substituent. For both the 7- and 8-position it is more favorable to have an
electronegative substituent than a hydrophobic substituent, and it may also be concluded that it is
more favorable for a substituent to be located in the 7- rather than the 8-position.
In summary, for the 5-, 7-, and 8-position it is less favorable to fill up space with a hydrophobic
substituent than to introduce an electronegative substituent, with the largest effect seen for the 5-
position. For the 5-position, this may be due to interactions with the cyclopropyl group, leading to
an alternative puckering. For the 7- and 8-position, steric clashes with the protein environment
cause unfavorable interactions. In contrast, for the 6-position the difference between a methyl- and
an electronegative substituent is small. The 6-position is the position where a substituent would be
the closest to its similar substituent on the other modulator molecule when bound to the protein and
thus, a substituent in this position might strengthen interactions between the modulators.
Conclusion
This work is the first example of combining experimental and calculated free energies of positive
allosteric modulator binding to the GluA2 LBD, and furthermore, it gives a detailed decomposition
of the thermodynamic driving forces of modulator binding.
27
ACS Paragon Plus Environment

Journal of Chemical Information and Modeling
Page 28 of 48
1
2
3
4
5
6
7
8
9
The experimentally measured thermodynamic details of binding of this fluorine substituted series of
GluA2 positive allosteric modulators showed that despite similarities in the binding affinities,
complex formation was primarily enthalpy driven for 6 as also previously seen for 5c. The 5a and
5b complex formation was primarily entropy driven, and 5d complex formation was equally
enthalpy and entropy driven. Dissection of the thermodynamic contributions, however, indicated
that the main driving forces of binding stem from the ligand-surrounding enthalpies for all
compounds and that the experimentally determined enthalpic and entropic differences are largely
explained by reorganizations in the protein and the solvent. Our comparison of the ITC,
pharmacological, and simulation data suggested that since 5c is the only compound, which
preserves the intrinsically favored and optimal conformation for hydrogen bonding when bound to
the protein, it shows the overall best combined modulatory effect and binding affinity. Our
predictions generated using the OSP simulation provided an indication of favorable substitution
positions in this modulator scaffold. While a larger substituent in the 5-, 7- or 8-position would be
sterically unfavorable, a larger substituent in the 6-position might strengthen interactions between
the modulators.
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
Supporting Information
Experimental characterization of intermediates and synthesis products, free energy profiles for TI
and molecular building block for all simulated compounds. This information is available free of
charge via the internet at http://pubs.acs.org.
Acknowledgements
28
ACS Paragon Plus Environment

Page 29 of 48
Journal of Chemical Information and Modeling
1
2
3
4
5
6
7
8
9
The Danish Council for Independent Research (Mobility PhD grant) and GluTarget (A.B.N., L.O.,
J.S.K.) as well as the Vienna Science and Technology Fund (WWTF, grant number LS08-QM03,
C.O.) and the European Research Council (ERC grant number 260408, C.O.) are acknowledged for
financial support.
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
References
1. Traynelis, S. F.; Wollmuth, L. P.; McBain, C. J.; Menniti, F. S.; Vance, K. M.; Ogden, K. K.;
Hansen, K. B.; Yuan, H. J.; Myers, S. J.; Dingledine, R., Glutamate Receptor Ion Channels:
Structure, Regulation, and Function. Pharmacol. Rev. 2010, 62, 405-496.
2. Sobolevsky, A. I.; Rosconi, M. P.; Gouaux, E., X-Ray Structure, Symmetry and Mechanism of
an AMPA-Subtype Glutamate Receptor. Nature 2009, 462, 745-756.
3. Armstrong, N.; Gouaux, E., Mechanisms for Activation and Antagonism of an AMPA-Sensitive
Glutamate Receptor: Crystal Structures of the GluR2 Ligand Binding Core. Neuron 2000, 28, 165-
181.
4. Jin, R. S.; Clark, S.; Weeks, A. M.; Dudman, J. T.; Gouaux, E.; Partin, K. M., Mechanism of
Positive Allosteric Modulators Acting on AMPA Receptors. J. Neurosci. 2005, 25, 9027-9036.
5. Sun, Y.; Olson, R.; Horning, M.; Armstrong, N.; Mayer, M.; Gouaux, E., Mechanism of
Glutamate Receptor Desensitization. Nature 2002, 417, 245-253.
6. Leavit, S.; Freire, E., Direct Measurement of Protein Binding Energetics by Isothermal Titration
Calorimetry. Curr. Opin. Struct. Biol. 2001, 11, 560-566.
7. Shirts, M. R.; Mobley, D. L.; Chodera, J. D., Alchemical Free Energy Calculations: Ready for
Prime Time? Annu. Rep. Comp. Chem. 2007, 3, 41-59.
8. de Ruiter, A.; Oostenbrink, C., Free Energy Calculations of Protein-Ligand Interactions. Curr.
Opin. Chem. Biol. 2011, 15, 547-552.
29
ACS Paragon Plus Environment