Paradoxically, Most Flexible Ligand Binds Most Entropy-Favored: Intriguing Impact of Ligand Flexibility and Solvation on Drug-Kinase Binding

Article abstract of DOI:10.1021/acs.jmedchem.8b00105

Biophysical parameters can accelerate drug development; e.g., rigid ligands may reduce entropic penalty and improve binding affinity. We studied systematically the impact of ligand rigidification on thermodynamics using a series of fasudil derivatives inhibiting protein kinase A by crystallography, isothermal titration calorimetry, nuclear magnetic resonance, and molecular dynamics simulations. The ligands varied in their internal degrees of freedom but conserve the number of heteroatoms. Counterintuitively, the most flexible ligand displays the entropically most favored binding. As experiment shows, this cannot be explained by higher residual flexibility of ligand, protein, or formed complex nor by a deviating or increased release of water molecules upon complex formation. NMR and crystal structures show no differences in flexibility and water release, although strong ligand-induced adaptations are observed. Instead, the flexible ligand entraps more efficiently water molecules in solution prior to protein binding, and by release of these waters, the favored entropic binding is observed.

Full text of DOI:10.1021/acs.jmedchem.8b00105

Subscriber access provided by University of South Dakota
Article
Paradoxically, Most Flexible Ligand Binds Most Entropy-Favored: Intriguing
Impact of Ligand Flexibility and Solvation on Drug-Kinase Binding
Barbara Wienen-Schmidt, Hendrik R. A. Jonker, Tobias Wulsdorf, Hans-
Dieter Gerber, Krishna Saxena, Denis Kudlinzki, Sridhar Sreeramulu, Giacomo
Parigi, Claudio Luchinat, Andreas Heine, Harald Schwalbe, and Gerhard Klebe
J. Med. Chem., Just Accepted Manuscript • DOI: 10.1021/acs.jmedchem.8b00105 • Publication Date (Web): 16 Jun 2018
Downloaded from http://pubs.acs.org on June 17, 2018
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 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 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.
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 37
Journal of Medicinal Chemistry
1
2
3
4
5
6
7
8
9
Paradoxically, Most Flexible Ligand Binds Most Entropy-
Favored: Intriguing Impact of Ligand Flexibility and Solvation
on Drug-Kinase Binding
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
Barbara WienenꢀSchmidt‡, Hendrik R.A. Jonker†, Tobias Wulsdorf‡, HansꢀDieter Gerber‡, Krishna
Saxena†, Denis Kudlinzki†, Sridhar Sreeramulu†, Giacomo Parigi#, Claudio Luchinat#, Andreas Heine‡,
Harald Schwalbe†*, Gerhard Klebe‡*
‡Institut für Pharmazeutische Chemie, PhilippsꢀUniversität Marburg, Marbacher Weg 6, 35032 Marburg, Germany
†Institut für Organische Chemie und Chemische Biologie, Johann Wolfgang GoetheꢀUniversität Frankfurt, MaxꢀvonꢀLaueꢀStraße 7 ,
N160ꢀ3.14 , 60438 Frankfurt am Main, Germany
#Magnetic Resonance Center (CERM/CIRMMP) and Department of Chemistry, University of Florence, Via Luigi Sacconi 6, 50019, Sesto
Fiorentino, Italy
ABSTRACT: Biophysical parameters can accelerate drug development, e.g. rigid ligands may reduce
entropic penalty and improve binding affinity. We studied systematically the impact of ligand
rigidification on thermodynamics using a series of fasudil derivatives inhibiting protein kinase A by
crystallography, isothermal titration calorimetry, nuclear magnetic resonance and molecular dynamics
simulations. The ligands varied in their internal degrees of freedom but conserve the number of
heteroatoms. Counterintuitively, the most flexible ligand displays the entropically most favored binding.
As experiment shows, this cannot be explained by higher residual flexibility of ligand, protein or formed
complex nor by a deviating or increased release of water molecules upon complex formation. NMR and
crystal structures show no differences in flexibility and water release although strong ligandꢀinduced
adaptations are observed. Instead, the flexible ligand entraps more efficiently water molecules in
solution prior to protein binding and by releasing these waters, the favored entropic binding is observed.
1
ACS Paragon Plus Environment

Journal of Medicinal Chemistry
Page 2 of 37
KEYWORDS: Thermodynamics, Protein Kinase A (PKA), Fasudil, Xꢀray Crystallography, NMR, 15NꢀT2ꢀ
1
relaxation time, (1H15N)ꢀTROSY, Degrees of Freedom, Molecular Dynamics Simulation, GIST
2
3
4
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
2
ACS Paragon Plus Environment

Page 3 of 37
Journal of Medicinal Chemistry
INTRODUCTION
1
2
3
4
5
6
7
8
9
Finding selective, effective and clinically successful drugs is a long and expensive enterprise. Hence, it
would be beneficial to accelerate this process by improving criteria to predict successful drug candidates
and to identify inappropriate ones early on. Therefore, the analysis of parameters beyond affinity is
required. In this context, thermodynamic analysis of proteinꢀligand interactions is used with increasing
popularity. Accordingly, the design of drugs with certain thermodynamic properties has been described
as a promising approach in many articles and reviews.1ꢀ7 Moreover, design guidelines to accomplish
purposefully tailored profiles have been intensely discussed, frequently advising the development of
rigid, correctly preꢀorganized ligands in order to influence the entropic penalty upon binding of the
ligand to its target.7ꢀ12
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, there remains a lack of systematic studies that explore correlations between ligand structure,
target protein characteristics and thermodynamic signature. From our perspective, it is important to
distinguish between the different characteristics of proteins. Particular factors that have to be taken into
account are: (1) size, (2) flexibility of the target, and (3) impact of water molecules. In this study, we
investigated the wide and clinically relevant family of protein kinases. Protein kinases are highly
flexible proteins as indicated by the presence of flexible loops in and next to the active site and helix
movements upon ligand binding.13ꢀ15 All protein kinases used ATP as phosphorylation substrate. This
reagent transfers its terminal γꢀphosphate group to an intermediately bound protein substrate, which is
phosphorylated at a Ser, Thr, or Tyr residue. Kinases have been rarely investigated by means of
comprehensive thermodynamic characterization. Only ten unique kinases are listed in databases that are
specialized for thermodynamic annotations, such as Scorpio16 and BindingDB17 (CDK2, ERK1/2,
JNK2, Pim1, AuroraꢀA, Thymidine Kinase, Nucleosid Diphosphate Kinase, vSrc, cSrc). Indeed for
none of these targets has a systematic investigation of ligand series been conducted. Considering that
according to Manning the human genome might encode for up to 518 protein kinases18 and with respect
to their importance as a drug target class, there is a clear need for thermodynamic data to characterize
3
ACS Paragon Plus Environment

Journal of Medicinal Chemistry
Page 4 of 37
protein kinases. The aim of this study is to start filling this gap, in order to broaden our understanding of
kinase dynamics and thermodynamics of ligand binding.
1
2
3
4
5
6
7
8
Here, the use of wellꢀstudied model proteins allows for the capture of extensive information from a wide
range of experimental techniques. Using the cAMPꢀdependent protein kinase (PKA) as a model system,
elaborate information on proteinꢀligand interaction was obtained using Xꢀray crystallography,
isothermal titration calorimetry (ITC) and nuclear magnetic resonance spectroscopy (NMR). The
selected ligands are derived from the approved drug fasudil,19 which has been developed as a rhoꢀkinase
inhibitor20 but, nonetheless, displays nanomolar affinity toward PKA and blocks the ATPꢀbinding site.21
This site also accommodates the second substrate, the protein to be phosphorylated. In our studies, it is
occupied by a fraction of protein, the inhibitory peptide PKI, comprising in our case 20 amino acids. It
bears instead of the catalytically important serine an alanine residue, hence it cannot be phosphorylated
in contrast to a physiological protein substrate. All of the studied fasudilꢀlike ligands have equal atom
count and comprise similar types of heteroatoms but they vary in their internal degrees of freedom
(Figure 1). In this way, the effect of ligand flexibility on the thermodynamic protein binding profile can
be investigated. Thorough analysis of the underlying structural factors contributing to this
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
1
thermodynamic profile has been made using Xꢀray crystallography and H15N bestꢀTROSY (transverse
relaxation optimized spectroscopy) NMR spectra. Furthermore, proteinꢀligand complex dynamics were
analyzed using 15NꢀT2ꢀrelaxation measurements. Additional insights about changes in the protein
1
hydration pattern were obtained from water H NMRD profiles by measuring the water proton
relaxation rates. Here, we present a unique study where the static data from high resolution crystal
structures are faced and combined with the information on the dynamics from detailed NMR
measurements and complemented molecular dynamics (MD) simulations.
For the thermodynamic data, ITC was used to measure KD and ꢁH at chemical equilibrium.
Subsequently, Gibbs free energy of binding ꢁG and the entropic contribution –TꢁS were calculated
using the following expression (eq. 1):
4
ACS Paragon Plus Environment

Page 5 of 37
Journal of Medicinal Chemistry
ꢁG = ΔH – TΔS = RT ln KD (1)
1
2
3
4
5
6
7
8
9
Our study displays striking results that could hardly have been predicted, as counterintuitively the most
flexible ligand binds entropically most favorably to the protein. Consequently, the hypothesis stating
that a significant loss of ligand’s degrees of freedom is, in all cases, entropically unfavorable to binding,
needs to be questioned and analyzed in the context of the entire binding event. We conclude that a
generalization of simple design guidelines is not constructive and will not lead to satisfactory results.
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
Figure 1. Chemical structure of the five ligands used in this study. From the left to the right the internal degrees
of freedom increase and will, in any case, require an entropic price to be paid upon binding, as an increasing
number of torsional degrees of freedom will be lost. In purple: Sꢀmethylꢀpiperazine substituted fasudilꢀderivative
(Ligand 01); in blue: Rꢀmethylꢀpiperazine substituted fasudilꢀderivative (Ligand 02); in green: Fasudil; in red:
openꢀchain fasudilꢀderivative (Ligand 04); in orange: longꢀchain fasudilꢀderivative (Ligand 05). According to
pKa calculations all ligands will be protonated at their sidechain amino group under the applied buffer conditions
of pH 7.2 (not indicated in the formulas).
RESULTS
Strongest induced fit is triggered by the most flexible ligand. In order to determine the binding
modes of all five ligands discussed, coꢀcrystal structures were obtained. For all structures, resolutions
between 1.4 and 1.6 Å could be achieved. In all five cases, difference electron densities are well defined
and indicate every heteroatom of the bound ligands (supplementary information, Table S1) with 100%
occupancy in the binding site. All structures have been deposited in Protein Data Bank (PDB). The
5
ACS Paragon Plus Environment

Journal of Medicinal Chemistry
Page 6 of 37
respective PDBꢀcodes are ligand 01: 5LCU, ligand 02: 5LCT, fasudil: 5LCP, ligand 04: 5LCR, ligand
05: 5LCQ.
1
2
3
4
5
6
7
8
Importantly, the crystallographic data confirmed that all five ligands display a congruent hinge binding
position of their respective isoquinoline moiety (Figure 2B). Moreover, their adjacent sulfonamides
occupy a common orientation in all structures, where one oxygen points toward the Glycineꢀrich loop
(Glyꢀloop) and the second toward the hinge region. Hence, proteinꢀligand interactions are highly similar
for the isoquinolineꢀ5ꢀsulfonamide portions of all five ligands. In all structures, a hydrogen bond is
accepted by the isoquinolineꢀnitrogen of the ligands and the backbone nitrogen of Val123 of the protein.
Besides these interactions, the compounds’ sulfonylꢀgroups do not directly interact with the protein.
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
In contrast to the very similar ligand core binding, significant changes were noted amongst the five
different proteinꢀligandꢀcomplexes considering the substituents attached to sulfonamide group.
In particular, the αG helix, the APE motif and the position of the proteinꢀkinaseꢀinhibitorꢀpeptide (PKI)
are visibly shifted (Figure 2A). Yet, the most prominent difference was revealed in the active site. Here,
the Glyꢀloop adopts three distinct positions ranging from a wideꢀopen to a closed conformation when
compared to the apoꢀprotein (Figure 2B).
The first, most open position of the Glyꢀloop exhibits the structure with the openꢀchained ligand 04
(red). Here, the loop is pushed out of the ligandꢀbinding site as the result of steric hindrance as the
binding of ligand 04 requires more space in the area of the Glyꢀloop than any of the other ligands.
The second, halfꢀopen conformation of the Glyꢀloop is found in the structures of Sꢀmethylꢀpiperazine
substituted ligand 01 (purple), the Rꢀmethylꢀpiperazine substituted ligand 02 (blue) and fasudil (green).
Interestingly, all three ligands share a common position of the Glyꢀloop even though the interaction
pattern of the homopiperazine moiety and 2ꢀmethylꢀpiperazine moieties of the ligands with the protein
differ significantly as will be described later.
6
ACS Paragon Plus Environment

Page 7 of 37
Journal of Medicinal Chemistry
Finally, the third, closed state is depicted in the structure of the longꢀchain ligand 05 (orange). A strong
induced fit resulting in a pullingꢀdown of the Glyꢀloop can be observed. Responsible for this
rearrangement are two hydrogen bonds formed between the backbone oxygen of Thr51 and the
sulfonamideꢀnitrogen as well as the secondary amine in the long chain of the ligand as shown in Figure
2C. Only in ligand 05 is a secondary amide present in the sulfonamide position and it can hence act as a
hydrogenꢀbond donor. All other ligands possess a tertiary amide in the sulfonamide, which does not
have the ability to act as a hydrogenꢀbond donor. Consequently, the formation of the key interaction to
Thr51 is impossible for all other compounds. In addition, steric hindrance would prevent the closed
position of the Glyꢀloop for all ligands other than ligand 05.
1
2
3
4
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
In the case of ligand 05, the closed conformation of the Glyꢀloop is facilitated by additional interactions
of the loop involving the following amino acids of the protein: Asp184, Ser53 and Lys72. In total, five
new interactions are formed (Figure 2C, black dotted lines).
From further analysis of the ligand binding mode, differences can be discerned with respect to the
deviating substituents (Figure 3B and S1). The interactions established by the terminal amino groups
differ, however, the observed Hꢀbonding patterns suggest that in all ligands, this nitrogen binds as
protonated ammonium group. The highest number of polar interactions to the protein is recognized by
the homopiperazine portion of fasudil (green) and the terminal aminoethyl moiety of the openꢀchain
ligand 04 (red). Interestingly, both ligands form comparable interaction patterns. In either case, the
terminal amino group forms hydrogen bonds with the backbone carbonyl oxygen of Glu170 as well as
the terminal carboxamide or carboxylate group of the side chains of Asn171 and Asp184. In contrast to
fasudil (green) and 04 (red), the interaction pattern of the terminal aminoethyl nitrogen of the longꢀ
chain ligand 05 (orange) differs. The only common interaction occurs with the backbone carbonyl
oxygen of Glu170. Furthermore, 05 interacts with the side chain of Glu127.
7
ACS Paragon Plus Environment

Journal of Medicinal Chemistry
Page 8 of 37
It is also notable that the Sꢀmethylꢀpiperazine substituted 01 (purple) and the Rꢀenantiomer 02 (blue)
1
2
3
each establish one hydrogenꢀbond to Asp184 of the DFGꢀloop via their terminal NH group.
4
5
6
7
8
A difference in the position of water molecules of the residual hydration pattern can be observed for the
different complexes. As a matter of fact, the protein flexibility takes impact on the observability of the
adjacent water positions. By optimizing the diffraction quality of the studied crystals, the resolution of
the collected datasets was improved. Nonetheless, many putative water molecule positions remained
unresolved due to an ambiguous density distribution next to the region showing enhanced residual
mobility in the crystal structure. A distinct analysis and quantification of the water molecule pattern and
hence a comparison across the active sites of all complexes is limited due to the flexible nature of the
protein, which affects the diffraction pattern defining the electron density in this region.
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
Figure 2. Superimposition of the coꢀcrystal structures of all five ligands. A: Overall view of the protein.
Ligand 01 is displayed in purple; 02 in blue; fasudil in green; 04 in red and 05 in orange. B: Blow up of
the active site. The longꢀchain ligand 05 induces the strongest conformational change dragging the Glyꢀ
loop toward the ligand. C: Interactions between 05 and the protein are displayed as orange dotted lines.
The key interaction responsible for the downward movement of the Glyꢀloop is a hydrogen bond to
8
ACS Paragon Plus Environment

Page 9 of 37
Journal of Medicinal Chemistry
Thr51. This transition is facilitated by the new interactions formed by the Glyꢀloop to the remaining part
of the protein. The latter contacts are displayed as black dotted lines.
1
2
3
4
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
Figure 3: Polar interactions of all five ligands to the corresponding protein structures. Ligand 04 is
displayed in red; fasudil in green; ligand 02 in blue; ligand 01 in purple and ligand 05 in orange. Polar
interactions are represented as dotted lines and colored in conformity with the corresponding ligand. A:
Interaction of the isoquinoline with the hinge region of the protein. B: The five ligands form different
polar interactions via their substituents attached to the sulfonamide group. In all cases the observed
interaction pattern suggests that the terminal sidechain amino group binds in protonated state (an
itemized depiction of the binding modes is shown in Figure S1).
The most flexible ligand binds entropically most favored to the protein. The thermodynamic
signature of ligand binding to the protein was determined by ITC at pH 7.2. All profiles were assessed
for putative buffer dependence to record possibly overlaid changes in protonation state upon complex
formation. The selected buffers differ in their ionization enthalpy by approximately 30 kJ/mol.22 A slope
between ꢀ0.07 and ꢀ0.15 could be observed for fasudil, 04 and 05 and +0.15 for 01 and 02 across the
9
ACS Paragon Plus Environment

Journal of Medicinal Chemistry
Page 10 of 37
three considered buffers. We believe, this scatter is insignificant within experimental accuracy.
Accordingly, changes in protonation states upon complex formation can be excluded.
1
2
3
4
5
6
7
8
We assume, all ligands adopt protonated state in solution prior to protein binding and carry this state
over to the proteinꢀbound situation. This is in accordance with the observed interaction patterns in the
crystal structures. Fasudil, 04 and 05 have calculated pKa values between 8.0 and 10.1, definitely
suggesting positively charged state in solution. The calculated pKa value for the piperazine nitrogen of
ligands 01 and 02 is 7.3, which is slightly below the applied buffer pH (calculated using
http://www.chemicalize.org/, accessed Sep, 9, 2015 and Sep, 23, 2015). Hence, there might be a small
partial proton uptake to reveal fully protonated state, if compared to the slopes of the other compounds
(s. above). Even though we believe these differences are insignificant, we used for the evaluation of the
thermodynamic signatures the profiles determined in phosphate buffer. This buffer exhibits very low
ionization enthalpy,22 thus per se, contributions from protonation effects will be minor under these
conditions.
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 reveal a closer match with the crystallographic data we applied two distinct scenarios during the ITC
titrations. In the first set of titrations, the ligand was directly titrated to the protein in the sample cell. In
the second set, the 20ꢀresidue peptide inhibitor PKI (s. above) was incubated with the protein and then
the ligand added.
Accordingly, measurements were performed in the presence (triangle symbols) and absence (circular
symbols) of the substrateꢀmimicking inhibitor peptide PKI (Figure 4).
Upon comparison of the relative differences of the two sets of thermodynamic profiles, it is apparent
that on absolute scale an offset between both sets of profiles is given. This leads to a shift toward more
favorable entropy and less beneficial enthalpy in the presence of PKI (triangles). Thus, the relative
differences between the ligands correlate in both cases.
10
ACS Paragon Plus Environment

Page 11 of 37
Journal of Medicinal Chemistry
All five ligands bind to PKA with virtually the same affinity. Moreover, it is striking that the binding
signatures of fasudil (green), the two methylꢀpiperazine substituted 01 (purple) and 02 (blue) as well as
the openꢀchain ligand 04 (red) are similar and scatter maximally in ꢁꢁH=4.2 kJ/mol and –TꢁꢁS=7.8
kJ/mol. In contrast, the longꢀchain ligand 05 (orange) displays a unique thermodynamic profile
significantly reduced in its enthalpic and simultaneously enhanced in its entropic contribution to
binding. In consequence, 05 is the most entropically and least enthalpically favored binder. In case of
the presence of PKI, the effect concerning ligand 05 is similar, however, somewhat less pronounced.
1
2
3
4
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
Figure 4. Thermodynamic profiles for all five ligands in absence (circles) and presence of PKI
(triangles). The entropy contribution is displayed on the yꢀaxis and enthalpy on the xꢀaxis. The dotted
line represents isoꢀaffinity for the mean ꢁG values of 32 kJ/mol. Titrations with the different ligands are
colored according to the colorꢀcode presented in Figure 1. The triangular symbol for ligand 04 (red) is
virtually hidden under the same profile as found for fasudil (green).
11
ACS Paragon Plus Environment

Journal of Medicinal Chemistry
Page 12 of 37
1
2
3
4
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
Figure 5. Amide chemical shift perturbations for a selection of residues and their corresponding
1
location in the Xꢀray crystal structure. Sections from an overlay of H15NꢀbestꢀTROSY spectra are
shown for PKA with and without (grey) ligands in the presence and absence of the PKI. Ligand 01
(purple) and 02 (blue) show nearly identical CSPs.
Correlation of amide chemical shift perturbations in the presence and absence of PKI. To further
analyze the structural influence of the peptidic inhibitor PKI, which is present in all of the studied
1
crystal structures, H15N bestꢀTROSY spectra of PKA were measured and the amide chemical shift
perturbation (CSP) of 88 amino acids were determined for each ligand in the presence and absence of
PKI (Figure S7AꢀC). Figure 5 shows the CSP for a selection of residues in the absence and presence of
PKI for the five different ligands and the corresponding amide location in the Xꢀray structure. These
data indicate similar binding properties of the ligands either in the absence or presence of the PKI
peptide and thus demonstrate the relevance of the crystal structures only determined in presence of PKI.
The largest spatial difference can be observed for residues Gly55 (not assignable for the longꢀchain
ligand 05), Ala188 and Arg256 (Figure S7D). Gly55 is located in the Glyꢀloop, which is involved in the
12
ACS Paragon Plus Environment

Page 13 of 37
Journal of Medicinal Chemistry
strongest inducedꢀfit adaptations observed in the five crystal structures. Ala188 resides two amino acids
Cꢀterminal from the DFGꢀmotif and is in close proximity to the PKIꢀpeptide’s Cꢀterminus. Arg256 is
remote from the active site at the bottom of the large subunit located in close proximity to the Nꢀ
terminus of the PKI peptide. The NMR data fully support and agree with the structural information from
crystallographic analysis. The spatial alignment of the crystal structures in Figure 2 was performed
using the backbone atoms of the residues with the smallest CSPs, thereby ensuring an unbiased
alignment.
1
2
3
4
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
The most flexible ligand does not form a flexible proteinꢀligand complex. In order to analyze the
residual flexibility of the resulting proteinꢀligand complexes 15NꢀT2ꢀrelaxation NMR measurements
were performed. These data (Figure S7D) indicate higher flexibility for the Nꢀterminal 13 residues. The
15
average of the remainder NꢀT2 relaxation values (75 of the 88 backbone amide signals that could be
tracked reliably) are presented in Figure 6. The 15NꢀT2ꢀrelaxation time can be used as a measure for
backbone dynamics (a more rigid protein backbone will show a smaller 15NꢀT2 relaxation time). On
average, the dynamics of all five complexes are very similar. Interestingly, ligand 05 with the largest
amount of internal degrees of freedom does not form a significantly more flexible proteinꢀligand
complex compared to the other more constrained members of the series.
13
ACS Paragon Plus Environment

Journal of Medicinal Chemistry
Page 14 of 37
Figure 6. The average amide 15NꢀT2ꢀrelaxation time for the PKA protein (75 out of 88 residues,
excluding the flexible Nꢀterminal residues) in presence of the PKI peptide, with and without ligands.
The overall differences are small. Error bars present the averaged standard deviation.
1
2
3
4
5
6
7
8
9
Protein hydration with and without fasudil or ligand 05. Protein hydration can be conveniently
monitored through NMR relaxometry, which provides the field dependence of the water proton
longitudinal relaxation rates in protein solutions, in turn reflecting the dynamics of the proteins and the
strength of the interactions between proteins and water molecules.23ꢀ26 NMR relaxometry is carried out
with a dedicated instrument that provides the measurement of the nuclear longitudinal relaxation rates
from 0.01 to tens of MHz ꢀ indicated as nuclear magnetic relaxation dispersion (NMRD) profiles. The
NMRD profiles report on the water proton spectral density functions, which depend on the
superposition of the different motional processes, in turn associated with characteristic motional
correlation times occurring on timescales from nanoseconds to microseconds.
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
Relaxometry measurements were performed at 298 K for PKA with and without the inhibitor peptide
PKI and/or with and without fasudil (Figure 7). The data show that in the presence of PKI and/or
fasudil the relaxation rates are slightly larger than for the free PKA samples. Furthermore, while the
relaxation rates are basically constant at low fields in the presence of PKI and/or fasudil, the profiles for
the free PKA samples show a slight decrease between 0.1 and 1 MHz. Consequently, the relaxometry
profiles measured in the presence of PKI/fasudil can be nicely reproduced from the sum of two
Lorentzian dispersions with different correlation times, whereas the profiles of free PKA require three
Lorentzian dispersions. All profiles were thus fitted using Eq. 3,23, 24, 27 and the best fit values are found
in Table S5. Water molecules interacting with the protein or with the protein complexes as well as
exchangeable protein protons contribute to the observed relaxation rates depending on the correlation
time modulating the dipoleꢀdipole interactions. The effective correlation time for the different protons is
the fastest between the reorientation time and the exchange time. The parameter a reports on the
contribution to water relaxation from protons relaxing with correlation times smaller than few ns,
14
ACS Paragon Plus Environment

Page 15 of 37
Journal of Medicinal Chemistry
whereas b1, b2 and b3 report on the contributions from protons relaxing with correlation time τslow, τR
(the protein reorientation time) and τfast, respectively, with τfast shorter than τR but larger than few ns.
This correlation time can either be due to fast local mobility at the protein site or to a water exchange
rate faster than protein rotation. The contribution from protons relaxing with such a correlation time
much shorter than τR is typically between 40% and 80% of the total.24
1
2
3
4
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
This fit shows that the contribution from protons relaxing with a correlation time τR = 28 ns or with a
correlation time τfast = 5 ns slightly increases on passing from free PKA to PKA bound to PKI and/or
fasudil, whereas the small amount of protons relaxing with a correlation time τslow =240 ns decreases.
The best fit value of τR (28 ns) is in good agreement with the value calculated with HYDRONMR42d
using the PDB structure 4WIH28, for which a correlation time of 26 ns is calculated. The presence of a
decreased contribution from water molecules with correlation time
upon PKI and fasudil binding
slow
may suggests that in free PKA some aggregation is present, and that the aggregated protein
decreases/disappears in the presence of the ligands. Overall, the data indicate that the hydration of the
protein does not decrease, but rather slightly increases upon PKI and/or fasudil binding, although the
difference is close to the error of the best fit values.
Figure 7. Water proton relaxometry profile of PKA with and without PKI and/or fasudil at 298 K. The
lines are the best fit profiles calculated from the parameters reported in Table S5.
15
ACS Paragon Plus Environment

Journal of Medicinal Chemistry
Page 16 of 37
In addition, we investigated whether relaxometry could detect differences between the binding of
fasudil and ligand 05 as a result of a larger number of water molecules removed from the active site
upon ligand 05 binding than upon fasudil binding. This could possibly explain the entropic advantage
found for 05 binding to PKA. We therefore collected relaxometry profiles of PKA in presence and
absence of PKI with either ligand 05 or fasudil (Figure S8). These measurements show, however, that
no differences can be detected between PKA+fasudil and PKA+ligand 05. This clearly indicates that
putative differences in the inventory of water displacement do not explain the entropic advantage of
ligand 05 over fasudil.
1
2
3
4
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
Entropic desolvation contributions investigated by molecular dynamics simulation. In order to
obtain insights which additional molecular contributors to entropy are important in the current case,
molecular dynamics (MD) simulations were carried out. We focused on fasudil and ligand 05 as
experimental relaxometry data are available for these ligands indicating analogous hydration patterns for
both proteinꢀligand complexes. Simulations were conducted for these two ligands, first, in their PKAꢀ
bound state in presence of PKI and, second, in aqueous solution prior to protein binding in absence of
the enzyme (details about the setup of the simulations are given in the Supporting Information, chapter
6). The MD simulations were subsequently used for an inꢀdepth analysis of the ligand desolvation
thermodynamics in aqueous solution. Finally, the simulation of the ligands in aqueous solution were
compared to and validated with experimental NMR ROE distance information (taken from rotating
frame NOEs) in order to obtain an idea about the predominantly populated conformers of the ligand in
buffered aqueous solution.
Ligand conformational entropy paradigm confirmed by simulation. At first, MD simulations of
fasudil and 05 were conducted in their proteinꢀbound states (each 100 ns), complemented by MDs in
aqueous solution describing the ligands in their unbound solution state prior to protein binding (each
100 ns). These calculations were performed to estimate on the conformational entropy contributions of
the ligand involving the transfer from aqueous solution to the proteinꢀbound state. The conformational
16
ACS Paragon Plus Environment

Page 17 of 37
Journal of Medicinal Chemistry
entropy difference for each of the two ligands between bound and unbound state was computed using
CENCALC.29 As anticipated by their chemical composition, the loss in conformational binding entropy
contribution (ꢀTꢁSconf) upon PKA binding is larger for ligand 05 than for fasudil (05: 9.3±0.3 kJ/mol,
fasudil: 4.8±1.3 kJ/mol, see Table S7 in SI, all values calculated at 300 K), indicating a higher loss of
flexibility upon binding of 05 over fasudil. As described above, the NMR spectroscopic measurements
did not reveal major structural differences of PKA experienced upon binding of the two ligands
including putative differences in the inventory of water displacement. Therefore, any contributions from
the protein will not matter in a relative comparison of the two complexes. As a consequence, in our
subsequent evaluations, we solely focused on the internal degrees of freedom of fasudil and 05.
1
2
3
4
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
Calculation of ligand desolvation entropy. The trajectories obtained from 100 ns unconstrained MD
simulations of unbound fasudil and 05 in aqueous solution were clustered with cpptraj (V15, see
section 6.3.1 in SI). The conformational ensemble found for fasudil can be well described by two,
equally populated clusters (50.1 vs 49.9%) of closely related conformers. In contrast, conformational
variance of 05 was best described using three clusters (67.8, 18.4 and 13.9%). Most interesting, the
crystallographically observed conformation of 05 in the bound state is found in the least populated
cluster.
The best representative conformers for each cluster were then used as starting geometries for followꢀup
MD simulations, however, now applying positional restraints, so that the conformations of the cluster
representatives were preserved throughout the simulations. This fixation of internal degrees of freedom
is necessary to subsequently perform an evaluation of the solvation pattern around the ligands using the
GIST method (Grid Inhomogeneous Solvation Theory) as a postꢀprocessing procedure.30, 31 The latter
approach calculates solvation thermodynamic quantities, such as entropy, and maps them on a threeꢀ
dimensional grid for graphical evaluation.
17
ACS Paragon Plus Environment

Journal of Medicinal Chemistry
Page 18 of 37
Visual inspection of the GIST entropy maps revealed that the secondary amino group is well solvated
for both ligands, fasudil and 05 (Figure 8). The desolvation of this group would be entropically
beneficial, since solvent hydrogenꢀbonding interactions result in both, translational and orientational
restrictions of the entrapped water molecules. On the contrary, the sulfonamide nitrogen atom is less
well solvated in fasudil (Figure 8(A)) compared to 05 (Figure 8(B)) for all clusters. This is probably
enhanced by the pseudoꢀmacrocyclic backꢀfolded shape of the predominant conformers of 05 (Figure
8(B1+B2)), which assist in binding water molecules more tightly to the sulfonamide group.
Consequently, the predominantly populated conformations of ligand 05 lead to an increase of the
probability to find a water molecule in the vicinity of the solvated ligand and therefore increases the
entropy difference between ligandꢀentrapped water molecules and those in bulk solvent phase.
1
2
3
4
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
To quantify these differences, we determined the solvation enthalpy ꢁH0 and entropy ꢁS0
They are
.
obtained as weighted sum of the GIST calculations using N cluster representatives as obtained from the
conformational clustering of the MD trajectory (eq. 2):
ꢄ
ꢅ
ꢂ
∑
ꢆ ꢈ
ꢂꢆ ꢈ
ꢃ ꢇꢅ ∙ ꢀꢁ ꢇꢅ
< ꢀꢁ >=
A0 either enthalpy or entropy (2)
Here, p(q) is the probability to find the solvated ligand in conformation qi (i.e. the occupancy of a
conformational cluster, cf. Figure 8) and the ꢁ indicates that all quantities A0, are calculated with respect
to bulk solvent state. The solvation quantities in the vicinity of the solvated ligand can be accessed
quantitatively by spatially summing ꢁA0 over a volume that captures the first layer of solvation of the
ligand (background of the method is found in section 6.1.4 of the SI).
Due to the enhanced hydrogenꢀbonding properties of the sulfonamide group in ligand 05 in the
predominantly populated pseudoꢀmacrocyclic backꢀfolded conformers, increased solvation entropy
contributions are suggested in comparison to fasudil (Table 1). A difference in ꢀTꢁStot at 300 K
between 05 and fasudil is computed to 15.2 ± 0.7 kJ/mol. With respect to the solvation free energy ꢁG
no significant difference between both ligands can be assigned within the error bars of the calculations.
18
ACS Paragon Plus Environment

Page 19 of 37
Journal of Medicinal Chemistry
Due to error propagation of the large individual energy contributions, the uncertainty in the ꢁG
difference becomes rather sizable. Since ligand 05 and fasudil do not differ in their experimentally
determined binding affinities, our calculation is supported by the fact that, overall, the price for the
desolvation free energy should be indistinguishable for both ligands.
1
2
3
4
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
Figure 8. A1+A2: Cluster centroids for the two conformational clusters of fasudil, which are populated
in 50.1% and 49.9% of the MD ensemble, respectively. B1+B2+B3: Cluster centroids of 05, populated
in 67.8%, 18.4% and 13.9% of the MD ensemble. The red isoꢀsurface displays solvent entropy at ꢀ2.5
kJ/mol/Å3.
Combined NMR experiments and MD simulations. To validate the findings suggested by our MD
simulations in aqueous solution, we consulted NMR experiments. The water molecules trapped by
ligand 05 in buffer solution cannot be characterized by NMR directly, due to the fast exchange rate of
19
ACS Paragon Plus Environment

Journal of Medicinal Chemistry
Page 20 of 37
ligandꢀtrapped water molecules. It is assumed that the solvated ligand perturbs the otherwise
homogeneous solvent and therefore gives rise to enhanced solvent structuring. Following this rationale,
the match of the in silico ensemble with an experimentally observed conformational ensemble would
strongly indicate that the proposed water entrapment takes place in solution and leads to the aboveꢀ
computed favorable solvation entropy. For this, we analyzed our unconstrained MD simulation of 05 in
water and compared it to NMRꢀobserved ROE distances. Since a particular distribution of these
distances can be realized by more than one conformational ensemble, we performed additional MD
simulations with 05, however, now applying NMRꢀderived distance potentials of varying strength.
Important enough, no significant differences have been recorded for our unconstrained, unbiased
simulations and the simulations applying NMRꢀbased distance restraints (details on these simulations
can be found in section 6.52. of the SI, including the 36 MD simulations with deviating settings of the
distance potentials). We therefore believe that the conformational distributions found by the simulations
are well representative for the conformational properties of fasudil and 05 in aqueous solution.
1
2
3
4
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
Table 1. Different solvation thermodynamic properties and conformational entropy difference for the
ligand during binding, as obtained from GIST and CCꢀMLA calculations at 300K.
ꢁA [kJ/mol]
fasudil
ligand 05
ꢁ b
ꢀTꢁSconf
4.8±1.38 a
9.3±0.3
4.5±1.4
ꢀTꢁSorient
ꢀTꢁStrans
ꢀTꢁStot
ꢁEww
41.5±0.1
53.5±0.3
95.1±0.6
360.1±5.9
ꢀ328.1±5.7
48.0±0.2
6.5±0.3
62.2±0.2
8.7±0.4
110. 2±0.4
367.6±4.1
ꢀ343.7±3.6
15.1±0.7
7. 5±7.2
ꢀ14.8±6.7
ꢁEsw
20
ACS Paragon Plus Environment

Page 21 of 37
Journal of Medicinal Chemistry
ꢁEtot
31.1±8.3
23.9±5.5
ꢀ7.2±9.9
7.9±9.9
1
2
3
4
ꢁG
126.3±8.3
134.2±5.5
5
6
7
8
9
a The uncertainty reported for each value is the ±standard deviation from the mean.
b This column reports the differences between ligand 05 and fasudil.
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 trajectory that agreed best with experimental data (91% within experimental boundaries) was used
for a principle component analysis (PC) of the conformational subspace. An overlap of 0.51 was found
with the in silico ensemble (section 6.5.3 in SI), which is quite promising considering the high
dimensionality of the configuration space compared to the PC subspace. Since subspace overlap can be
obscured by different effects, the significance of the result was further assessed by its Zꢀscore.32 The Zꢀ
score is constructed by comparing the actual overlap to the overlap with a background model consisting
of 500 random trajectories. Positive values indicate high significance, zero or negative values indicate
insignificant overlap. The calculated Zꢀscore in the present case is 5.84, and therefore indicates that
similarity between the experimental ROEꢀderived ensemble and the in silico generated main cluster of
05 is far above noise and therefore can be assumed significant for our attempted analysis of the
conformation and solvation properties.
DISCUSSION
For a series of five congeneric and nearly equipotent ligands, studied in this contribution, we thoroughly
analyzed the interactions with PKA. Highꢀresolution crystallographic data revealed a comparable hinge
binding mode of all ligands. As special feature, a strong inducedꢀfit mechanism for ligand 05 (orange) is
observed. NMR chemical shift perturbation data reveal high similarity between the structural properties
of the different complexes in crystalline and solution state. NMR relaxation time measurements
unraveled the structural influence of inhibitor peptide PKI on the protein and allowed analysis of the
backbone dynamics, which however, agree across all studied complexes. Thermodynamic signatures
21
ACS Paragon Plus Environment

Journal of Medicinal Chemistry
Page 22 of 37
were measured either in the absence or in the presence of the substrate mimicking peptide PKI. Even
though all ligands bind with nearly the same affinity, they distinctly factorize in enthalpy and entropy.
This also uncovers that the most flexible ligand 05 with the largest amount of internal degrees of
freedom is, on first sight paradoxically and quite counterintuitive, the entropically most favored binder
of the series, independent of the presence or absence of PKI. Usually, the binding of more rigid and
conformationally preorganized ligands reveals an entropic advantage to binding.33, 34 In the present case,
this would be expected for the ligands fasudil, 01 and 02 possessing the conformationally restricted
cyclic substituents.
1
2
3
4
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
It is remarkable that the relative differences in the thermodynamic profiles between the five ligands
remain similar with and without the presence of the substrateꢀlike PKI. However, the overall shift of all
profiles on absolute scale toward more entropyꢀfavored binding appears reasonable upon binding of the
substrate peptide as the protein becomes preꢀstabilized and structurally organized either in a way to
better recognize the coꢀsubstrate ATP or, in our case, the inhibitors of the fasudilꢀtype. The
thermodynamic profiles of ligand binding are all shifted toward an entropically more favorable but
enthalpically less beneficial signature in the presence of PKI. This profile is in accordance with a better
preorganization of one of the binding partner, here of the recipient protein.
Among the individual ligands, the entropically more favored binding of the conformationally less
restricted ligand 05 to PKA could result from (i) the formation of a proteinꢀligand complex with
strongly enhanced residual flexibility, (ii) from increased displacement of activeꢀsite water molecules
upon ligand binding to the uncomplexed protein, or (iii) from a difference in the solvation properties
among the five ligands in aqueous solution prior to binding. These contributions must even
overcompensate the entropic losses to be paid by restricting the conformational flexibility of ligand 05
to the bound state. Calculations, which evaluate the entropic losses for the spatial restriction of the more
rigid fasudil in comparison to the more flexible ligand 05 upon the transfer to the bound state, confirm
this assumption.
22
ACS Paragon Plus Environment

Page 23 of 37
Journal of Medicinal Chemistry
The first aspect, listed above, suggests the formation of a complex of enhanced residual flexibility. We
suppose that this is unlikely, as ligand 05 shows overall the lowest Bꢀfactors in the crystal structure. It
reflects the lowest residual mobility in bound state in the series (Table S2), which correlates with a
strong inducedꢀfit adaptation, pulling the Glyꢀloop towards the bound ligand. Our hypothesis is further
supported by the NMR relaxation data, which do not indicate enhanced mobility for the complex with
this ligand compared to the others. Clearly, these findings do not speak for an entropic advantage of
ligand 05 upon binding.
1
2
3
4
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
A second putative explanation for the favorable entropy contribution of 05 would originate from the
displacement of a larger amount of previously wellꢀordered activeꢀsite water molecules upon proteinꢀ
ligand complex formation. Unfortunately, the amount of water molecules in the active site cannot be
reliably evaluated for the entire series by crystallography due to a lessꢀwell defined electron density in
this area in consequence of enhanced loop flexibility. However, NMR relaxometry measurements show
that there is no significant difference in the hydration pattern between the complexes formed with 05
and fasudil. Thus, we believe that also differences in the amount of displaced water molecules cannot
explain the entropic advantage of the more flexible ligand 05.
A third alternative explanation, additionally influencing the thermodynamic signature of ligand 05
compared to the other members of the series, can arise from differences of the ligands in aqueous
solution prior to PKA binding. Remarkably, this effect has nothing to do with the actual binding to the
protein. Once the ligands are released from the bulk water phase and accommodate in the protein, they
have to shed their solvation shells. If these shells show structural differences in the local hydration
pattern, deviating thermodynamic signatures will result. Indeed MD simulations strongly indicate that
the entropically beneficial signal of ligand 05 is due to a unique hydration pattern triggered by a backꢀ
folded conformation that favorably entraps water molecules. The frequent occurrence of this
conformation (68 % of the time) in aqueous solution was confirmed within the boundaries of NMR
spectroscopic data (ROE distances).
23
ACS Paragon Plus Environment

Journal of Medicinal Chemistry
Page 24 of 37
Overall our calculations suggest that the entropic losses, required to immobilize the more flexible ligand
05 compared to fasudil in the binding pocket of PKA, are overcompensated by the entropic advantage
resulting from the water release once ligand 05 is desolvated. Nevertheless, we are careful comparing
such values quantitatively on a common scale including our microcalorimetry data, since the
methodologies of the applied methods to estimate such values differ strongly.
1
2
3
4
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
Our experimental data demonstrate that the entropically more favored binding of 05 to the protein
mainly results from an entropic benefit of the ligand while shedding its hydration shell upon leaving the
bulk water phase. Our hypothesis that a ligand with a larger amount of internal degrees of freedom does
not necessarily lead to entropically lessꢀfavored binding was also observed in other cases, however,
without providing a conclusive explanation.35 Therefore, the assumption that the number of the ligand’s
degrees of freedom to be lost upon complex formation has always a dominating impact on the entropic
binding component needs to be questioned or at least requires a more detailed analysis of the binding
event as a whole. It demonstrates that simple design guidelines cannot be generalized unreflectively and
deserve a much more careful consideration. In particular, the frequently neglected factor of ligand
solvation has to be more thoroughly considered.
CONCLUSION
Commonly applied design guidelines need to be systematically investigated and assessed for different
protein and ligand classes. Rules that hold for small and rigid proteins do not necessarily apply for
larger or flexible proteins. From our systematic study on a protein kinase we conclude that the
thermodynamic profile is not dominated by the loss of degrees of freedom due to fewer rotatable bonds
on the ligand. The hypothesis that ligand rigidification necessarily results in an overall significantly
more entropic binding profile could not be confirmed. Strikingly, we even recorded the opposite trend in
that the most flexible ligand was the most favored entropic binder. Therefore, the loss of degrees of
freedom of the ligand, which, as a matter of fact, has to be paid, does not necessarily dominate the
24
ACS Paragon Plus Environment

Page 25 of 37
Journal of Medicinal Chemistry
thermodynamic profile. Even in the present case of binding to a flexible protein like a kinase, the effect
of residual flexibility of the formed proteinꢀligand complex appears to be minor as all complexes under
consideration show very similar NMR relaxation data. In fact, the predominant effect seems to result
from structural differences in the hydration pattern of the ligands in the bulk water phase prior to any
protein binding. They appear to determine the resulting entropically more favored binding signal.
1
2
3
4
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
In summary, our results indicate that the complexity of structureꢀactivity relationships increases with
protein flexibility and desolvation differences of either the active site or the ligand molecules. Most
strikingly, these contributions cannot be treated as orthogonal design parameters in rational optimization
of potential drug candidates. Our data signalize that global understanding and reliable predictions of
thermodynamic profiles need a comprehensive view on the binding event and require many more
systematic studies of congeneric ligand series.
EXPERIMENTAL SECTION
Protein Expression and Purification for ITC and Crystallization. The catalytic subunit of cAMPꢀ
dependent protein kinase from Chinese hamster ovary cells (98% sequence identity with human
isoform) was expressed with a Hisꢀtag in a modified pET16bꢀVector with an introduced TEVꢀcleavage
site between the protein Nꢀterminus and Hisꢀtag. This plasmid was transformed into E. coli strain Bl21
(DE3)/pLysS (Novagen).28
Cell disruption was performed using a highꢀpressure homogenizer for multiple cycles. After
centrifugation (1h at 30.000g) cell lysate supernatant was purified in a first step using a NiꢀNTA column
that binds the Hisꢀtag of the protein and was eluted by an imidazole gradient. The Hisꢀtag was then
cleaved off by TEVꢀprotease. Afterwards, an inverse NiꢀNTA column was employed collecting PKA in
the flowꢀthrough. Finally, ion exchange chromatography was performed using a MonoS column
separating threeꢀfold phosphorylated PKA from the fourꢀfold phosphorylated form using a HEPES
buffer with a sodium chloride gradient.28
25
ACS Paragon Plus Environment

Journal of Medicinal Chemistry
Page 26 of 37
Protein Expression and Purification for NMR. The catalytic subunit of cAMPꢀdependent protein
1
2
3
kinase was expressed, isotope labeled (N15) and purified as previously described.36
4
5
6
7
8
Crystallization. Coꢀcrystallization was performed using the hanging drop method at 4 °C. The
crystallization drops were composed as follows: 10 mg/mL PKA (240 µM), 30 mM MBT (MES/Bisꢀ
Tris buffer pH 6.9), 1 mM DTT, 0.1 mM EDTA, 75 mM LiCl, 0.03 mM Mega 8, 0.07 mM PKI (Sigma:
P7739), 1.2 mM ligand dissolved in DMSO from a 50ꢀ100 mM stock. The wellꢀcontained mixture of
methanol in water with varying methanol concentrations (v/v) for the different ligands (fasudil: 18%
methanol; ligand 05: 18% methanol; ligand 04: 14% methanol; ligand 01: 16% methanol; ligand 02:
19% methanol). In the crystallization setup streakꢀseeding was performed using a horse hair in order to
initialize crystal growth. For crystal mounting, crystals were cryoꢀprotected in 5 mM MBT (MES/Bisꢀ
Tris buffer pH 6.9), 1 mM DTT, 0.1 mM LiCl, 1.2 mM ligand dissolved in DMSO from a 50ꢀ100 mM
stock, 16 % (v/v) methanol, 30% (v/v) MPD and flash frozen in liquid nitrogen.
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
Crystal Structure Determination. All structures were collected at the storage ring Bessy II Helmholtzꢀ
Zentrum Berlin, Germany at Beamline 14.1 on a Pilatus 6M pixel detector with resolutions between
1.42 – 1.62 Å. The datasets were processed using XDS37 and molecular replacement was performed
using CCP4 Phaser38 and PDBꢀstructure of PKA from bos taurus 1Q8W as a model.21 This was
followed by simulated annealing, multiple refinement cycles of maximum likelihood energy
minimization and Bꢀfactor refinement using Phenix.39 Coot40 was used to fit aminoꢀacid side chains into
σꢀweighted 2Fo – Fc and Fo – Fc electron density maps. If appropriate electron density was observed,
multiple side chain conformations were built into the model and maintained during the refinement if the
minor populated side chain displayed at least 20 % occupancy. Hydrogen atoms were included using a
riding model. Ramachandran plots for structure validation were generated using PROCHECK.41 Data
collection, unit cell parameters and refinement statistics are given in the supplementary information
(Table S2). Analysis of temperature factors was performed with Moleman.42 Protein and PKI Bꢀfactors
were anisotropically refined, water Bꢀfactors were isotropically refined for all structures. Rfree was
26
ACS Paragon Plus Environment

Page 27 of 37
Journal of Medicinal Chemistry
calculated using 5% of all reflections which were randomly chosen and not used for the refinement. The
required ligand restraint files were created using the Grade webserver (accessed Dec, 12, 2014; Feb, 27,
2015; Apr, 22, 2015; Dec, 16, 2015; Mar, 16, 2016).43, 44 For figure preparation Pymol was used.45
1
2
3
4
5
6
7
8
9
Isothermal Titration Calorimetry. The buffer used for the ITC experiments contained: 30 mM sodium
phosphate buffer pH 7.2, 10 mM MgCl2, 100 mM NaCl, 3% (v/v) DMSO. All measurements were
repeated 3ꢀ5 times. Further buffers were used in order to check for protonation linkage. In these buffers,
30 mM sodium phosphate buffer was replaced by 30 mM HEPES and 30 mM triethanolamine (TEA),
respectively (both at pH 7.2). Buffer dependency was tested for all measurements in the absence of PKI.
For the measurements expressed, purified and dialyzed PKA was used in the ITCꢀmeasuring cell. A 15ꢀ
20 fold higher concentrated ligand solution, diluted in dialysis buffer, was then stepwise injected to the
protein solution during the measurement. All measurements were performed at 25 °C. ITC data were
analyzed using NITPIC and Sedphat.46, 47 Raw data and exact values and standard deviations for ꢁG,
ꢁH and –TꢁS can be found in the supplementary information (Table S3, S4).
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
Compound purity was analyzed using quantitative nuclear magnetic resonance spectroscopy (qNMR)
and in case of deviation from 1:1 binding stoichiometry, ligand concentration was corrected
accordingly. The purity of the used ligands was at or above 95%.
NMR Measurements. The NMR samples contained about 0.1ꢀ0.2 mM PKA protein, 20 mM sodium
phosphate buffer (pH 6.5), 80 mM NaCl and 2 mM TCEP (tris(2ꢀcarboxyethyl)phosphine) and 10%
D2O (for spectrometer lock). The NMR experiments were conducted at a temperature of 298 K on
Bruker Avance 600, 800 and 950 MHz spectrometers equipped with cryogenic tripleꢀresonance probes.
For the PKA+PKI complex, titration experiments were performed in which the unlabeled PKI peptide
was added until no significant changes were observed anymore in the 1H15NꢀHSQC spectra after the last
addition. The samples of PKA with PKI contained a slight excess of the PKI peptide and the samples
with the ligands contained a 5 times excess of the ligand to assure the full complex formation. NMR
27
ACS Paragon Plus Environment

Journal of Medicinal Chemistry
Page 28 of 37
spectra were acquired and processed using Topspin version 3.2 (Bruker Biospin) and analyzed using
Sparky 3.114 (T. D. Goddard and D. G. Kneller, University of California, San Francisco). Resonances
for the PKA protein have been assigned before.36, 48 The spectra of PKA alone, PKA with PKI, PKA
with ligands and PKA with PKI and ligands have been assigned by overlaying the 1H15NꢀHSQC spectra.
Only the backbone amide signals that could be tracked reliably were further analyzed. Heteronuclear
1
2
3
4
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
15
15NꢀT2 relaxation experiments were performed on the uniformly N labeled PKA protein in complex
with PKI with and without ligands. The relaxation rates were determined from a series of spectra with
delays of 0, 16.96, 33.92, 50.88 and 67.84 ms. The amide chemical shift perturbations (CSPs) were
acquired from 1H15NꢀbestꢀTROSY49 spectra.
Relaxometry. Water 1H NMRD profiles were obtained by measuring the water proton relaxation rates,
R1, as a function of the applied magnetic field. The NMRD profiles were measured with a
SPINMASTER2000 fast field cycling relaxometer (Stelar, Mede (PV), Italy) operating in the 0.01–40
MHz proton Larmor frequency range, at 298 K. The measurements are affected by an error of about
±1%, when fitted to a monoexponential decay/recovery of the magnetization in the field cycling
experiment. Measurements were performed either in the presence or in the absence of PKI, with and
without fasudil and with and without ligand 05. Protein concentrations were 0.5 mM and 0.21 mM in
the absence of PKI, and 0.5 mM when PKI was added. Ligand concentration was 4ꢀ6 times larger than
the protein concentration.
The profiles were fit using Eq. (3), where protons are assumed in fast exchange and the modelꢀfree
model is applied23, 24, 27
:
R1=(0.3+a)+ b1J(τslow)+b2J(τR)+b3J(τfast
)
(3)
with J(τ)=0.2τ/[1+(ωτ)2]+0.8τ/[1+(2ωτ)2] and where a, b1, b2, b3, τslow, τR, τfast are fitting parameters. In
order to reduce the covariance among the many fit parameters, all profiles were fit simultaneously by
imposing common values for τslow, τR and τfast in the absence of PKI, and τR and τslow values 5.4% larger
28
ACS Paragon Plus Environment

Page 29 of 37
Journal of Medicinal Chemistry
in the presence of PKI (due to the estimated slower reorientation time of the complex upon PKI
binding).
1
2
3
4
5
6
7
8
Molecular Dynamics Calculations. Details about the setting of our molecular dynamics simulations
and the applied GIST calculations to estimate the effects of ligand hydration are documented in detail in
the Supporting Information, chapter 6.
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
Ligands. Ligands 01, 04 and 05 were purchased from Uorsy (Ukraine). The R and S isomers of the 2ꢀ
methylpiperazine inhibitors 01 and 02 were synthesized starting from 5ꢀ(chlorosulfonyl) isoquinolineꢀ
hydrochloride, prior to this freshly prepared from isoquinolineꢀ5ꢀsulfonic acid via a known literature
procedure,50, 51 which was reacted with the respective, commercially available, enantiomerically pure R
or Sꢀconfigured monoꢀNꢀBocꢀprotected 3ꢀmethylpiperazine thus rendering the corresponding inhibitor
precursors, respectively. Finally, NꢀBoc deprotection with 4 M HCl in dioxane gave rise to the
corresponding inhibitors 01 and 02 as their hydrochloride salts. We performed qNMR measurements to
check the purity of our ligands which exceeded 95% purity.
Accession Codes. Atomic coordinates and experimental details for all new crystal structures will be
released in the PDB upon publication: PKAꢀligand 01: 5LCU, PKAꢀligand 02: 5LCT, PKAꢀfasudil:
5LCP, PKAꢀligand 04: 5LCR, PKAꢀligand 05: 5LCQ.
ASSOCIATED CONTENT
Supporting Information.
The following is supplied as Supporting Information:
Crystallographic tables; pictures of the mFoꢀDFcꢀdensities of all five ligands; NMR spectra of the
15
ligands; qNMR data. ITCꢀtitration curves; amide CSPs and NꢀT2 relaxation per amino acid, synthesis
protocols, analytical data, MD simulations and GIST calculations to derive thermodynamic quantities,
MD simulations with NMR distance potentials, comparison of MD and NMR ensembles in Cartesian
and distance space.
29
ACS Paragon Plus Environment