Enhancement of hydrophobic interactions and hydrogen bond strength by cooperativity: Synthesis, modeling, and molecular dynamics simulations of a congeneric series of thrombin inhibitors

Article abstract of DOI:10.1021/jm9016416

Accurately predicting the binding affinity of ligands to their receptors by computational methods is one of the major challenges in structure-based drug design. One of the potentially significant errors in these predictions is the common assumption that the ligand binding affinity contributions of noncovalent interactions are additive. Herein we present data obtained from two separate series of thrombin inhibitors containing hydrophobic side chains of increasing size that bind in the S3 pocket and with, or without, an adjacent amine that engages in a hydrogen bond with Gly 216. The first series of inhibitors has a m-chlorobenzyl moiety binding in the S1 pocket, and the second has a benzamidine moiety. When the adjacent hydrogen bond is present, the enhanced binding affinity per ?² of hydrophobic contact surface in the S3 pocket improves by 75% and 59%, respectively, over the inhibitors lacking this hydrogen bond. This improvement of the binding affinity per ?² demonstrates cooperativity between the hydrophobic interaction and the hydrogen bond.

Full text of DOI:10.1021/jm9016416

2126 J. Med. Chem. 2010, 53, 2126–2135
DOI: 10.1021/jm9016416
Enhancement of Hydrophobic Interactions and Hydrogen Bond Strength by Cooperativity: Synthesis,
Modeling, and Molecular Dynamics Simulations of a Congeneric Series of Thrombin Inhibitors
Laveena Muley,† Bernhard Baum,‡ Michael Smolinski,† Marek Freindorf,§ Andreas Heine,† Gerhard Klebe,*,‡ and
David G. Hangauer*,†
†Department of Chemistry, University at Buffalo, The State University of New York, Buffalo, New York 14260,
‡Department of Pharmaceutical Chemistry, Philipps University Marburg, Marbacher Weg 6, 35032 Marburg, Germany, and
§The Center for Computational Research, University at Buffalo, The State University of New York, Buffalo, New York 14260
Received November 5, 2009
Accurately predicting the binding affinity of ligands to their receptors by computational methods is one
of the major challenges in structure-based drug design. One of the potentially significant errors in these
predictions is the common assumption that the ligand binding affinity contributions of noncovalent
interactions are additive. Herein we present data obtained from two separate series of thrombin
inhibitors containing hydrophobic side chains of increasing size that bind in the S3 pocket and with, or
without, an adjacent amine that engages in a hydrogen bond with Gly 216. The first series of inhibitors
has a m-chlorobenzyl moiety binding in the S1 pocket, and the second has a benzamidine moiety. When
2
˚
the adjacent hydrogen bond is present, the enhanced binding affinity per A of hydrophobic contact
surface in the S3 pocket improves by 75% and 59%, respectively, over the inhibitors lacking this
2
hydrogen bond. This improvement of the binding affinity per A demonstrates cooperativity between
˚
the hydrophobic interaction and the hydrogen bond.
Introduction
ability of the ligand to access the binding site, the desolvation
free energyof the ligand, and the protein binding cavity as well
as entropy and enthalpy changes in the ligand, protein host,
and water.3-5 Also, scoring functions typically make the
assumption that the binding affinity contributed by each
of the individual interactions to the total binding affinity
is additive and does not depend upon the surrounding inter-
actions.6 This assumption does not account for one of the
confounding properties of noncovalent interactions, namely,
that of cooperativity.7 Cooperativity, as described by Williams,8
is a phenomenon whereby multiple ligand-host binding inter-
actions can give a ligand with a binding affinity greater than
(positive cooperativity) or less than (negative cooperativity) the
sum of the individual interactions. For example, a recent paper
by Lafleur et al described positive cooperativity between van der
Waals interactions and a hydrogen bond for a series of tyrosine
kinase inhibitors.9
Molecular recognition in an aqueous biological system
occurs through a collection of reversible and individually
weak, binding interactions between the interacting species.
As such, it is one of the most important processes in drug
discovery. There have been substantial advances in the deter-
mination of the structures of proteins and their complexes by
X-ray crystallography and NMRa spectroscopy.1 However
the ability to quantitatively, or even semiquantitatively, pre-
dict the binding affinity of ligands to these structures using
computational methods is still a major challenge. In a recent
analysis by Warren et al,2 10 docking programs and 37scoring
functionswere evaluated and itwas concluded thatthe scoring
functions are in need of significant improvements for predict-
ing binding affinity. This task is particularly challenging
because the observed binding affinity not only is due to the
collective weak noncovalent interactions but also includes the
The current study focuses on our initial efforts toward
experimentally probing the cooperativity of noncovalent
interactions in a systematically varied series of ligands.
Thrombin was selected as the model host system because it
has a well-defined active site with numerous X-ray structures
of ligand complexes available. The basic questions this study
addresses are the following: (i) Can a hydrogen bond and a
hydrophobicinteraction reinforce each other through positive
cooperativity? (ii) If yes, then what is the magnitude of the
positive cooperativity and can it depend upon the number and
type of pre-existing interactions?
*To whom correspondence should be addressed. For G.K. (for
kinetic and the collaborative interpretation of the data): phone, þ49
6421 282 1313; fax, þ49 6421 282 8994; e-mail, klebe@mailer.
uni-marburg.de. For D.G.H. (for the design and synthesis of the
compounds, modeling studies, MD simulations, and collaborative
interpretation of the data): phone, þ1-716-645-4183; fax, þ1-716-645-
6963; e-mail, hangauer@buffalo.edu.
a Abbreviations: NMR, nuclear magnetic resonance; Boc, di-tert-
butyloxycarbonyl; Fmoc, 9-fluorenylmethyloxycarbonyl; DCM, di-
chloromethane; DIEA, diisopropylethylamine; DMF, N,N-dimethyl-
formamide; DMSO, dimethyl sulfoxide; EDCI, 1-(3-dimethylamino-
propyl)-3-ethylcarbodiimide hydrochloride; HOBT, hydroxybenzotria-
zole; HCl, hydrochloric acid; HPLC, high pressure liquid chromato-
graphy; HRMS, high resolution mass spectrometry; ITC, isothermal
titration calorimetry; LCMS, high pressure liquid chromatography/
mass spectrometry; PAB-Z, N-benzyloxy-4-aminomethylbenzamidine;
Z, benzyloxycarbonyl; MD, molecular dynamics.
Two separate series of thrombin inhibitors were de-
signed. The structural features of the two series and their
binding mode in the active site of thrombin are illustrated
in Figure 1.
r
pubs.acs.org/jmc
Published on Web 02/11/2010
2010 American Chemical Society

Article
As shown in Figure 1, the first series has a m-chlorobenzyl
Journal of Medicinal Chemistry, 2010, Vol. 53, No. 5 2127
bond with the carbonyl oxygen of the Gly 216 residue. In both
inhibitor series a proline moiety was selected to bind under
Tyr60A and Trp60D in the S2 pocket, as does the natural
substrate.14 If there is no positive cooperativity, then the
moiety binding in the S1 pocket. This side chain was selected
based upon earlier thrombin inhibitor studies.10-12 In the
second series the m-chlorobenzyl moiety is replaced by the
more potent benzamidine moiety which forms a salt bridge
and two hydrogen bonds with Asp 189 at the bottom of the S1
pocket. While these P1 side chains were held constant within
each series, the size of the hydrophobic side chain (R) that
binds in the well-defined S3 pocket13 was systematically
increased, from a hydrogen atom to a benzyl group, in the
presence or absence of the X = NH2 that forms a hydrogen
2
˚
increased binding affinity per added A of P3 hydrophobic
side chain contact surface would be the same in the absence or
presence of the X = NH2 hydrogen bonding side chain.
Furthermore, if the magnitude of the cooperativity depends
upon the number and type of pre-existing interactions, then
the increased hydrophobic binding affinity in the presence of
the X = NH2 hydrogen bonding side chain could be different
for the two series because series II has a salt bridge with two
hydrogen bonds in the S1 pocket whereas Series I does not. As
described herein the presence of the X = NH2 hydrogen bond
2
˚
significantly increased the binding affinity per added A of
hydrophobic contact surface and to a somewhat different
extent in the two series. Consequently, not accounting for
cooperativity of this type in scoring functions could add
substantially to the error in predicting binding affinity, since
hydrophobic binding is a major component of most ligand-
2
˚
proteincomplexes, and the error is accumulatedover many A
of hydrophobic contact surface.
Chemistry
Synthesis of Series I. The synthesis of series IA starts
with an amide coupling between Boc-proline and m-chloro-
benzylamine using EDCI/HOBT as the coupling reagents
(Scheme 1). The amide 1 was recrystallized from MeOH/
water before quantitative Boc-deprotection using metha-
nolic HCl. The amine 2 was then coupled to the different
carboxylic acids (Scheme 1) to give products 3a-k or with
N-protected D-amino acids (Scheme 2) to give the products
(4a-k). Each product was purified by reverse-phase
HPLC. This was followed by the Boc/Fmoc deprotection
in case of the amino acids to obtain the final products
5a-k.
Synthesis of Series II. The synthesis of the series II
compounds begins with the synthesis of N-benzyloxycar-
bonyl-4-aminomethylbenzamidine 10 as shown in Scheme 3.
The synthesis is guided by the process reported by Lila et al.15
The synthesis begins by conversion of 4-cyanobenzyl bro-
mide to the bis-Boc protected benzylamine 6. The bromide
is displaced with di-tert-butyliminodicarboxylate using
sodium hydride as a base in anhydrous THF. The cyano
compound 6 was converted to the amidoxime 7 by reaction
with hydroxylamine. Complete consumption of the starting
material was realized when 5 equiv each of DIEA and
hydroxylamine hydrochloride were used in refluxing DCM.
Figure 1. Structural features and binding mode of the designed
inhibitors.
Scheme 1. General Scheme for Synthesis of Series IAa
a (a) EDCI, HOBT, DIEA, 18 h, 72%; (b) gaseous HCl, ether, 100%; (c) EDCI, HOBT, DIEA, DCM, 12-18 h.

2128 Journal of Medicinal Chemistry, 2010, Vol. 53, No. 5
Scheme 2. General Scheme for Synthesis of Series IBa
Muley et al.
a (a) EDCI, HOBT, DIEA, DCM, 12-18 h; (b) gaseous HCl, methanol; (c) 20% piperidine in DCM, 2 h.
Scheme 3. Synthesis of N-Benzyloxycarbonyl-4-aminomethylbenzamidine Hydrochloride 10 (PAB-Z HCl)a
3
a Z = benzyloxycarbonyl.
The recrystallized amidoxime 7 was converted into the
free benzamidine by acetylating the hydroxyl group with
acetic anhydride and subjecting the resulting acetate to
hydrogenolysis over Pd/C in acetic acid. Protection of
the benzamidine 8 with a benzyloxycarbonyl group pro-
vided 9. Removal of the Boc groups was achieved with
1 M HCl in methanol to give the hydrochloride salt 10
crystallization or silica gel chromatography in order to be
pure enough to carry forward in the synthesis. The next step
was hydrogenation of the benzyl ester and was conducted on
a Parr hydrogenator over 10% Pd/C in methanol. The
reaction had to be monitored and promptly removed upon
completion (2-6 h) to avoid formation of the methyl ester. In
most cases the reaction had yields >95% with very good
purity. In cases where the methyl ester formed, it was
removed by base-acid back-extraction. If needed, the com-
pound was purified using reverse phase HPLC. The next step
(PAB-Z HCl).
3
An attempt to streamline the synthesis was made by
coupling PAB-Z to Boc-L-proline. However, the reaction
produced low yields of the coupled product (30-40%) and
the need for reverse phase HPLC could not be avoided.
Preparation of the dipeptide portion prior to the attach-
ment of the benzamidine was a longer route but proved to
be more efficient. A general scheme for the synthesis of this
series is shown in Schemes 4 and 5. First, the alkyl-
carboxylic acids and Boc-D-amino acids are coupled to
L-proline benzyl ester for series IIA and IIB, respectively.
The ester is then saponified to the carboxylic acid followed
in the synthesis is the coupling with PAB-Z HCl. This was
3
performed using EDCI/HOBT with DIEA as the base. DMF
was used in place of DCM to increase solubility. The desired
products were purified by reverse-phase HPLC. Deprotec-
tion of the Z-amidines 13a-k was achieved by catalytic
hydrogenation at atmospheric pressure over 10% Pd/C in
methanol. The reaction was monitored by LC/MS and
typically took between 18 and 36 h to reach completion.
The products were converted to the hydrochloride salts
with HCl in ether and submitted for biological testing. In
the case of 17a-h the intermediate was subjected to Boc-
deprotection using methanolic HCl. The mixture was then
allowed to stir for 2 h. The solvent was removed to give the
desired products as the dihydrochloride salts for biological
testing. Further details can be found in the Supporting
Information.
by coupling with PAB-Z HCl. Last, the protecting groups
3
are removed and the product is converted to the hydro-
chloride salt.
The coupling between L-proline benzyl ester and each of
the alkylcarboxylic acids, or Boc-D-amino acids, produced
some products that were >95% pure after acid-base work-
up and carried on to the next step, but others required either

Article
Journal of Medicinal Chemistry, 2010, Vol. 53, No. 5 2129
Scheme 4. General Scheme for Synthesis of Series IIAa
a (a) EDCI, HOBT, DIEA, DCM, 12-18 h; (b) H2/Pd, methanol, 18 h; (c) PAB-Z HCl, EDCI, HOBT, DIEA, DMF, 12-18 h; (d) H2/Pd, methanol,
3
18 h, HCl-methanol, 1 h.
Scheme 5. General Scheme for Synthesis of Series IIBa
Table 1. Series I Ki Values and Corresponding ΔG Values for Binding,
and Hydrophobic Contact Surface Area As Obtained from Molecular
Modeling
hydrophobic
contact surface
2
area (A )
˚
compd Ki (μM) (sd ΔG (kJ/mol) (sd
(sd
3a
3b
3c
3d
3e
3f
1.46 ꢀ 103 409
-16.2
-19.1
-19.9
-21.7
-18.9
-22.6
-23.6
-25.5
-23.0
-25.8
-23.0
-29.1
-25.6
-29.5
-31.9
-34.5
-35.8
-36.5
-36.3
-39.0
-39.9
-38.5
1.68
0.69
0.01
0.68
0.26
0.72
0.26
0.73
0.41
0.57
0.65
0.59
0.20
0.38
0.47
0.38
0.86
0.39
0.36
0.79
0.88
0.96
292
311
334
342
342
365
368
380
400
391
384
299
312
334
342
338
368
378
370
384
409
387
3.76
2.59
2.02
1.36
1.15
1.62
2.20
0.81
1.66
2.27
0.65
0.47
0.74
1.33
1.24
1.88
1.58
1.94
0.85
1.82
0.76
0.40
444
322
133
190
155
484
86.0
139
109
46.3
22.1
6.34
11.0
7.45
34.3
4.22
11.8
1.60
0.66
0.28
0.19
0.10
0.06
0.11
0.04
0.15
3g
3h
3i
72.4
34.5
92.9
29.7
94.3
8.08
33.0
6.81
2.54
0.92
0.54
0.41
0.43
0.15
0.10
0.18
a (a) EDCI, HOBT, DIEA, DCM, 12-18 h; (b) H2/Pd, methanol,
18 h; (c) PAB-Z HCl, EDCI, HOBT, DIEA, DMF, 12-18 h; (d) H2/Pd,
3
methanol, 18 h; (e) gaseous HCl, methanol.
3j
3k
5a
5b
5c
5d
5e
5f
Results and Discussion
Binding Affinity and Binding Mode Evaluation. Ki values
for compounds 3a-k, 5a-k, 14a-k, and 18a-k binding to
thrombin were measured using a standard kinetic photo-
metric assay as described in the Experimental Section. The
hydrophobic contact surface areas between the ligands and
the active site were calculated using SYBYL 7.3 as described
in the Experimental Section. Molecular modeling studies
were carried out starting with the crystal structure 1TA211 as
the starting reference structure for series I. 1TA2 is the
crystal structure of thrombin in a complex with an inhibitor
similar to 5k. The only difference between the two ligands is
that the inhibitor in the 1TA2 structure has an additional
chlorine atom on the phenyl ring that binds in the S1 pocket.
For series II, 1K2213 was used as the starting reference
structure. 1K22 is the crystal structure of thrombin in
complex with the drug melagatran. The structure of this
drug is similar to that of 18g. The only difference between the
two is that melagatran has a four-membered ring in the
central S2 pocket instead of the five-membered ring in 18g
and acetic acid is attached to the terminal amine.
5g
5h
5i
5j
5k
the SYBYL force field. Once the optimized geometry was
obtained, the hydrophobic contact surface area was calcu-
lated. The detailed procedure used for the calculation of the
hydrophobic contact surface area is described in the Experi-
mental Section.
The experimentally obtained Ki values, the corresponding
calculated Gibbs energy of binding, and the calculated
hydrophobic contact surface areas for series I and II are
listed in Tables 1 and 2, respectively.
The Tables 1 and 2 data for the inhibitor series I and II are
plotted in Figure 2a and Figure 2b, respectively. Four
important observations and data interpretations can be
made from these plots:
(1) Both series produce a linear correlation between the
improvement in the ΔG of binding and the increase in
hydrophobic contact surface in the S3 pocket, in the presence
and absence of the additional hydrogen bond (X = H or
NH2, respectively). This suggests that any cooperativity that
The modeling studies were carried out by replacing the
additional chlorine atom in 1TA2 with a hydrogen atom for
series I or replacing the four-membered ring in the central S2
pocket with a five-membered proline ring and deleting the
terminal acetic acid in 1K22 for series II. For both series the
P3 side chain was modified to correspond to the designed
inhibitors (3a-k, 5a-k, 14a-k, and 18a-h). The resulting
modeled complexes were minimized to convergence using

2130 Journal of Medicinal Chemistry, 2010, Vol. 53, No. 5
Muley et al.
Table 2. Series II Ki Values and Corresponding ΔG Values, and
Hydrophobic Contact Surface Area As Obtained from Molecular
Modeling
lacking the added hydrogen bond (X = H). This demon-
strates that the presence of the adjacent hydrogen bond
would require a scoring function to award almost twice as
much binding affinity for each A of hydrophobic contact
surface in the S3 pocket.
2
˚
hydrophobic
contact surface
2
area (A )
˚
compd Ki (μM) (sd ΔG (kJ/mol) (sd
(sd
(3) The slope of the line for series IIB, with an added
2
˚
14a
14b
14c
14d
14e
14f
14g
14h
14i
56.1
39.0
5.70
3.81
0.94
1.28
1.64
3.61
1.18
1.17
0.75
1.42
0.18
0.07
0.15
0.02
0.011
0.023
0.004
8.28
13.8
1.19
0.57
0.33
0.27
0.28
0.53
0.30
0.41
0.19
0.24
0.03
0.03
0.02
0.02
0.002
0.006
0.001
-24.3
-25.2
-29.9
-30.9
-34.4
-33.6
-33.0
-31.1
-33.8
-33.9
-35.0
-33.4
-38.5
-41.0
-39.0
-43.5
-45.5
-43.7
-47.9
0.35
0.85
0.47
0.35
0.75
0.51
0.40
0.33
0.86
0.53
0.77
0.40
0.42
0.99
0.31
1.17
0.46
0.71
1.34
287
303
323
354
364
365
367
374
387
393
380
304
322
333
328
365
355
377
365
0.21
2.01
0.69
3.88
2.76
2.85
2.40
2.43
2.74
2.34
0.60
1.56
0.98
0.04
3.26
3.17
2.82
0.37
0.30
hydrogen bond (X = NH2), is -0.154 (kJ/mol)/A of hydro-
phobic contact surface area added, which is 59% more than
2
˚
-0.097 (kJ/mol)/A for series IIA lacking the added hydro-
gen bond (X = H). The series IIB vs series IIA data confirm
the trend obtained with series IB vs series IA but also
suggests that the presence of the benzamidine salt bridge in
series II, with a bifurcated hydrogen bonding to Asp 189 in
the S1 pocket, may somewhat dampen the cooperativity
between the P3 hydrophobic side chain and the adjacent
hydrogen bond when X = NH2.
14j
14k
18a
18b
18c
18d
18e
18f
18g
18h
(4) The presence of additional strong interactions else-
where in the complex may be expected to help position the
hydrophobic side chain in the S3 pocket so that the binding
affinity will not be enhanced as much with the installation of
another hydrogen bond immediately adjacent to it. This
concept is supported by comparing the binding affinity
2
˚
enhancement -0.097 (kJ/mol)/A of added hydrophobic
contact surface in the S3 pocket in the presence of the
benzamidine hydrogen bonds/salt bridge in the S1 pocket
2
˚
(series IIA, X = H) to the -0.073 (kJ/mol)/A for the series
IA compounds (X = H) lacking any hydrogen bonding/salt
bridge in the S1 pocket. These data suggest that the presence
of the benzamidine hydrogen bonds/salt bridge in the S1
pocket has already accomplished some of the P3 side chain
positioning such that a 33% increase in the strength of the
hydrophobic interaction in the S3 pocket occurs. This in-
dicates that cooperativity between hydrogen bonds/salt
bridges and hydrophobic interactions can also be long-
range, albeit with a dampened magnitude. The cooperativity
between the adjacent hydrogen bond (X = NH2) and the
hydrophobic binding in the S3 pocket was 75% for series I
and is reduced to 33% when the hydrogen bonds/salt bridge
are installed in the S1 pocket (benzamidine) instead of
immediately adjacent to the P3 hydrophobic side chain.
When the hydrogen bond is already present adjacent to the
P3 side chains, and we add the S1 pocket benzamidine
hydrogen bonds/salt bridge (i.e., series IB, X = NH2 to
series IIB, X = NH2), the long-range cooperativity is further
2
˚
reduced to 21% (-0.128 to -0.154 (kJ/mol)/A , res-
pectively) from 33%. This suggests that the magnitude
of cooperativity is dependent not only upon the distance
between the interacting side chains but also upon how much
ligand positioning has already been accomplished by
pre-existing hydrogen bonds. These interpretations are
also supported by the molecular dynamics results discussed
later.
Figure 2. (a) Plot of binding Gibbs energy vs hydrophobic contact
surface area for series I (A) and series I (B). (b) Plot of binding
Gibbs energy vs hydrophobic contact surface area for series
II (A) and series II (B). The error in the two slopes was obtained
by fitting the data to a linear regression model using WinNonLin
Professional. The Student’s t test was also performed on the two
slopes to confirm that the values were outside experimental error
(P < 0.01).
A linear correlation between the Gibbs energy of binding
and hydrophobic contact surface area was previously
reported for buried hydrophobic residues at the interface in
a protein-protein interaction.16 The protein interface stabi-
lization free energy (slope) was reported to improve
by -0.0627 ( 0.005 kJ/mol upon burial of each additional
is present between the hydrophobic R side chain and
the hydrogen bonding X side chain does not change in
magnitude as the R side chain increases in size to fill the S3
pocket.
2
˚
A of hydrophobic surface, comparable to the -0.073 (
2
0.006 (kJ/mol)/A obtained for series I in the current study
˚
(2) The slope of the line for the series IB, with an added
2
when the adjacent hydrogen bond was absent (X = H). The
values obtained in the current study with ligand-thrombin
interactions and in the protein-protein interaction study16
are in the range of the values reported in the literature
˚
hydrogen bond (X = NH2), is -0.128 (kJ/mol)/A of hydro-
phobic contact surface area added, which is 75% more than
2
˚
that of -0.073 (kJ/mol)/A for the series IA compounds

Article
Journal of Medicinal Chemistry, 2010, Vol. 53, No. 5 2131
2
˚
(0.08-0.20 (kJ/mol)/A ) for hydrophobic interactions in a
ligand-receptor complex.17 The current study suggests that
this 2.5-fold range may be at least partially due to the
presence or absence of cooperative interactions when the
ΔG values for the hydrophobic interactions were measured.
To put into perspective the potential underestimate of
binding affinity for not including positive cooperativity, one
can consider that the larger P3 side chain ligands in the
2
˚
current study have about 400 A of hydrophobic contact
2
surface compared to about 300 A of hydrophobic contact
˚
surface when there is no P3 side chain. If a scoring func-
tion were to calculate the difference in the binding affinity
2
˚
based upon -0.073 (kJ/mol)/A instead of the correct
2
˚
-0.128 (kJ/mol)/A (when X = NH2) for series I, then the
calculated Ki would be underestimated by about an order of
magnitude.
The data can also be analyzed using double functional
group replacement cycles as illustrated in Figure 3. These
cycles for the indicated series of inhibitors show that the
X = NH2 group and benzyl P3 side chain have a positive
cooperativity of 5.97 kJ/mol for series I (Figure 3a). Similar
analysis for series II (Figure 3b) again indicates positive
cooperativity (4.37 kJ/mol) but somewhat smaller than for
series I.
In order to validate our modeling results, the thrombin
complex crystal structures with compounds 5c (small side
chain) and 5k (large side chain) from series I and with
compounds 14c, 18b (small side chain) and 14k, 18h (large
side chain) from series II were solved.15 The bound inhibitor
geometries obtained from these crystal structures were com-
pared with the geometries obtained from the modeling.
Figure 4 shows the modeled inhibitor geometries super-
imposed upon the corresponding crystal structure geo-
metries. As shown, the modeled geometries are essentially
identical to that obtained from the crystal structures, which
confirms that the averaged binding modes obtained from
crystal analysis and modeling agree well. This indicates that
the modeled complexes and the corresponding calculated
hydrophobic contact surfaces are valid.
Crystal structures for the compounds from series I which
lack the additional hydrogen bond (X = H) could not be
obtained because of the weaker binding affinity of the
compounds in this series. However, crystal structures
were successfully determined when X = H for series II
(compounds 14c and 14k). On the basis of the close corre-
spondence that was obtained with series II for the crystal-
lographic and modeled P3 side chains when X = NH2, and
the modeled and crystallographic P3 side chains when X =
H,18 it was assumed that the modeled geometry obtained for
series I with X = H is also relevant.
Figure 3. (a) Double functional group replacement cycle demon-
strating positive cooperativity for series I. (b) Double functional
group replacement cycle demonstrating positive cooperativity for
series II.
With this first experimental demonstration of how signifi-
cantly hydrophobic binding strength can vary because of
neighboring interactions, the question now becomes “What
is the cause of this?” As indicated earlier a possible explana-
the hydrogen bond between the terminal NH2 and the Gly
216 residue. However, in order to be consistent with the
linear correlation shown in Figure 2, the improvement in the
strength of this hydrogen bond would need to be linearly
dependent upon the gradual increase in hydrophobic contact
surface in the S3 pocket as the side chain size is increased. In
order to gain further insight into these two possibilities,
molecular dynamics (MD) simulations were performed on
select inhibitors in each series. The P3 isopropyl side chain
derivatives were selected as examples with a small hydro-
phobic side chain because crystal structure analyses showed
that the S3 pocket accommodates bound water molecules, in
addition to the hydrophobic side chain, with side chains
2
˚
tion for the increase in the binding affinity per A of hydro-
phobic contact surface in the presence of the additional
hydrogen bond is the likely restriction of P3 side chain
movement inside the hydrophobic S3 pocket. This restriction
could cause the side chain to spend more time in proximity to
the hydrophobic S3 pocket surface and thereby potentially
enhance the associated van der Waals interactions. A second
possibility is that as the P3 side chain is increased in size, the
reduction in ligand residual motion (due to a complementary
steric filling of the S3 pocket) may result in a strengthening of

2132 Journal of Medicinal Chemistry, 2010, Vol. 53, No. 5
Muley et al.
Figure 4. (a) Overlay of crystal vs modeled inhibitors for series I.
(b) Overlay of crystal vs modeled inhibitors for series II.
smaller than the isopropyl group.18 These co-occupancy
water molecules add a complication to the relative interpre-
tation of the MD simulations, since they are not present with
the larger side chains that are used in the comparison.
Consequently, we selected compounds 3d and 5d (series I,
m-chlorobenzyl) and 14d and 18c (series II, benzamidine), all
of which contain a P3 isopropyl side chain, as an example of a
small alkyl group for the MD simulations (Figure 5a). The
benzyl group was selected as the large P3 side chain so that
compounds 3k and 5k (series I, m-chlorobenzyl) and 14k and
18h (series II, benzamidine) were utilized for the large side
chain MD simulations (Figure 5b). The detailed procedure
of the MD simulations is described in the Experimental
Section. An analysis of the trajectory over 1 ns revealed
interesting differences in the two series as illustrated in
Figure 6. Table 3 summarizes the corresponding average
˚
distances in A between the P3 hydrophobic side chains and
CD1 of the adjacent Leu 99 residue at the back of the S3
pocket, as well as the average hydrogen bond distance
between the terminal amino group and Gly 216.
Figure 5. Schematic drawing of the binding mode of the inhibitors
with isopropyl side chain (a) and benzyl side chain (b) in S3 pocket.
The distances recorded along the molecular dynamics trajectory are
shown as dotted lines.
In Figure 6a the distances between the centroid of the
isopropyl side chain of compounds 3d, 5d (X = H, NH2) and
14d, 18c (X = H, NH2) of series I and II, respectively, and
CD1 of the adjacent Leu 99 residue at the edge of the S3
pocket are shown. Figure 6b shows the corresponding dis-
tance for the corresponding benzyl derivatives 3k, 5k and
14k, 18h. Along the trajectories for these eight compounds it
is observed that the distance from the P3 side chain to the S3
earlier. Hence, the installation of an additional hydrogen
bond in series II does not reduce the residual motion of the
inhibitors as much as in series I.
The hydrogen bond distance between the terminal NH2
and the Gly 216 carbonyl oxygen atom was also moni-
tored over 1 ns for compounds 5d, 5k and 18c, 18h, and the
results are illustrated in Figure 6c. This MD simulation
suggests that the larger side chains in the S3 pocket reduce
the time-average length of the hydrogen bond and thus the
strength.
˚
pocket wall only fluctuates by an average of (0.38 A across
both series (Table 3), which indicates that these side chains
remain inside the hydrophobic S3 pocket even in the absence
of the additional hydrogen bond. Also, the isopropyl and
benzyl side chains (small and large) spend more time closer to
the Leu 99 residue located in the S3 pocket in the presence of
the additional hydrogen bond than in its absence for both
series I and II. This suggests that the hydrogen bond
strengthens the hydrophobic interaction of the attached side
chain. However, the average distance between the hydro-
phobic side chain and the Leu 99 residue in the absence of the
additional hydrogen bond for series II is smaller than for
series I. This is probably due to the presence of the benza-
midine moiety which forms a hydrogen bonding salt bridge
with Asp 189 in the S1 pocket, thereby already reducing the
residual entropy of the inhibitor somewhat as indicated
Overall the MD simulation data suggest the following:
(1) The additional hydrogen bond to Gly 216 (X = NH2)
holds the P3 side chains closer to the wall of the hydrophobic
pocket for series I and II, most likely resulting in the
2
˚
additional binding free energy per A of hydrophobic con-
tact surface area. For series I (m-Cl) the time-averaged mean
hydrophobic side chain distance to the S3 pocket wall
˚
decreased by 1.57 and 0.49 A for the small (isopropyl) and
large (benzyl) side chains, respectively, when the Gly 216
hydrogen bond is installed (X = NH2). For series II
(benzamidine) the corresponding mean side chain distance
˚
decreases are reduced to 0.71 and 0.39 A, respectively.

Article
Journal of Medicinal Chemistry, 2010, Vol. 53, No. 5 2133
Figure 6. Distances recorded along the molecular dynamics trajectory: (a) fluctuations of representative distances over 1 ns from the centroid
of the inhibitor side chain to CD1 of Leu 99; (b) fluctuations of representative distances over 1 ns from the terminal atom of the benzyl side chain
to CD1 of Leu99; (c) fluctuations of representative distances over 1 ns from terminal amine of the inhibitor to the carbonyl oxygen of Gly 216.
˚
Table 3. Average Distances in A between the Hydrophobic Side Chain
and CD1 of the Adjacent Leu99 Residue at the Back of the S-3 Pocket
and the Average Hydrogen Bond Distance between the Terminal Amine
and Gly 216 Recorded over 1 ns
(X = NH2), then the small (isopropyl) and large (benzyl) side
chains are positioned closer to the S3 pocket wall by 0.46 and
˚
0.13 A for the benzamidine compounds 18c and 18h com-
pared to the corresponding m-Cl compounds 5d and 5k,
respectively. However this closer positioning is less dramatic
than in the absence of the hydrogen bond to Gly 216. This
again suggests that the presence of the benzamidine hydro-
gen bonding in the S1 pocket has already accomplished some
of the close P3 side chain positioning that the X = NH2
hydrogen bond to Gly 216 can provide so that adding the
hydrogen bond to Gly 216 is less beneficial to binding in the
benzamidine series compared to the m-Cl series.
time averaged distance
between inhibitor side chain centroid/
average
H-bond
˚
terminal atom CT to Leu99 (A)
˚
compd
(sd distance (A) (sd
3d
3k
7.02
4.31
5.45
3.82
5.70
4.08
4.99
3.69
0.45
0.49
5d
5k
0.44
0.27
0.43
0.37
0.31
0.26
3.73
3.24
0.52
0.33
14d
14k
18c
18h
(3) The hydrogen bond appears to get shorter and stronger
as the size of the hydrophobic side chain increases. For the m-
Cl series and the benzamidine series the hydrogen bond to
4.17
3.06
0.27
0.21
˚
(2) The presence of the benzamidine S1 pocket hydrogen
bonding salt bridge has already reduced the distance of the
side P3 side chain to the S3 pocket wall. The small (isopropyl)
and large (benzyl) side chains are closer to the S3 pocket wall
Gly 216 shortens by 0.49 and 1.11 A when the side chain is
increased from isopropyl to benzyl (from 5d to 5k and from
18c to 18h, respectively).
(4) When one compares the standard deviations of the
distances along the MD trajectory for the key interac-
tions (Table 3), suggestions regarding the residual motion of
these interactions emerge. Upon installation of the hydrogen
bond to Gly 216 in the m-Cl series I P3 isopropyl side
˚
by 1.32 and 0.23 A for the benzamidine compounds 14d and
14k compared to the corresponding m-Cl compounds 3d and
3k, respectively, all of which lack the hydrogen bond to Gly
216 (X = H). When the hydrogen bond to Gly 216 is present

2134 Journal of Medicinal Chemistry, 2010, Vol. 53, No. 5
Muley et al.
inhibitor (3d vs 5d), the standard deviation of the hydro-
phobic contact distance does not change significantly
error in binding affinity predictions for not including co-
operativity between a typical hydrophobic side chain of
2
100 A contact surface and an adjacent hydrogen bond can
˚
˚
((0.45 to 0.44 A) even though the average distance to the
S3 pocket wall is reduced from 7.02 to 5.45 A. On the other
˚
be about an order of magnitude.
hand when this hydrogen bond is installed in the benzami-
dine series II P3 isopropyl side chain inhibitor (14d to 18c),
Experimental Section
˚
the analogous standard deviations are reduced by (0.12 A
Molecular Modeling. All modeling experiments were per-
formed using SYBYL software (Tripos, St. Louis, MO), ver-
sions 6.9 or 7.0, on an HP or SGI workstation. Each modeling
experiment used thrombin coordinates downloaded from the
Brookhaven Web site (www.rcsb.org) as described earlier. It
was observed that the thrombin crystal structure is essentially
unchanged with the various inhibitor complexes evaluated in
this study.15 Hydrogens were added, and charges were com-
puted using the Gasteiger-Marsilli protocol. The ligand was
extracted from the binding site, and the inhibitors discussed
herein were manually constructed and docked using analogous
thrombin inhibitor crystal structures as a guide. The complexes
were minimized to convergence while holding the protein fixed,
using the Tripos force field and the default parameters with the
exception of using a dielectric constant of 80.0.
˚
((0.43 to 0.31 A) while the distance to the S3 pocket wall is
reduced from 5.70 to 4.99 A. This comparison indicates that
˚
as the P3 isopropyl side chain is moved closer to the S3
pocket wall the residual motion of the side chain is decreased.
When the analogous comparisons are made with the larger
P3 benzyl side chain in the m-Cl series I (3k vs 5k), the
standard deviation of the hydrophobic contact distance now
˚
˚
decreases by a more significant (0.22 A ((0.49 to 0.27 A)
while the distance to the S3 pocket wall is reduced from 4.31
˚
to 3.82 A. In the benzamidine series II with the P3 benzyl side
chain (14k vs 18h) the standard deviation of the hydrophobic
˚
˚
contact distance decreases by ( 0.11 A ((0.37 to 0.26 A)
while the distance to the S3 pocket wall is reduced from 4.08
˚
to 3.69 A. This comparison analogously indicates that as the
Calculation of Hydrophobic Contact Surface Area. Once the
energy minimized binding mode is obtained using the above
procedure, the MOLCAD surface is generated for the protein.
This surface is first mapped according to lipophilic potential.
Only the hydrophobic surface is used for the analysis. In order to
do so, the midpoint of the lipophilic potential range for the
protein was calculated within SYBYL and the potential from
the midpoint to the maximally lipopholic surface was defined as
the protein lipophilic surface. The next step is to take the
distance between the protein and the ligand into account. In
order to do so, the protein MOLCAD surface is mapped again
except this time according to the distance between the ligand and
the protein. A new MOLCAD protein surface is then created
that satisfies both criteria: (1) It has a lipophilic potential from
the midpoint of the lipophilic scale to the maximum value, and
(2) it has a distance between the ligand (including hydrogens)
and the protein of 2.6 A or less so that a water molecule cannot
be accommodated between the two surfaces (diameter of water,
˚
2.8 A). The hydrophobic protein surface area that meets both
criteria is then calculated. The above procedure is repeated three
times in order to obtain an average measurement of the contact
surface area and a standard deviation.
Molecular Dynamics Simulations. The protein-ligand com-
plexes were placed in a sphere of TIP3 water molecules of radius
˚
50 A, about 50 000 atoms. For the ligands the Amber GAFF
force field was used. The atomic charges of the ligands were
calculated by fitting (program RESP) to the HF/6-31G* elec-
trostatic potential (Gaussian 98) based on the crystallographi-
cally observed conformation. The MD was started by heating
the solvent to 300 K over a period of 20 ps and cooling to 100 K
over a period of 5 ps, keeping the solute fixed. Then the system
was gradually brought to 300 K over 25 ps. The simulation was
carried on for 1050 ps under constant temperature and pressure
(NPT), applying periodic boundary conditions (Amber, version
8.0). Energy data were stored every 10 time steps, solute
coordinates every 0.5 ps, and solvent coordinates every 5 ps.
All results presented refer to the 1 ns trajectories, excluding the
first 1 ns required for temperature adjustment and equilibration
(an equilibrated state with respect to the total potential energy in
the system was reached after 100 ps). Analyses were carried out
primarily using modules from the Amber program suite, while
VMD19 and DS Viewer Pro (http://accelrys.com/products) were
used for visualization purposes.
P3 benzyl side chain is moved closer to the S3 pocket wall, the
residual motion of the side chain is decreased. Finally, the
effect of the size of the P3 side chain on the standard
deviation of the hydrogen bond to Gly 216 can be analyzed
from the data in Table 3. In the m-Cl series I increasing the P3
side chain from an isopropyl group to a benzyl group (5d to
5k) reduces the standard deviation of the hydrogen bond to
˚
˚
Gly 216 by (0.19 A ((0.52 to 0.33 A) while the hydrogen
bonding distance decreases from 3.73 to 3.24 A. In the
˚
benzamidine series II increasing the P3 side chain from an
isopropyl group to a benzyl group (18c to 18h) reduces the
standard deviation of the hydrogen bond to Gly 216 by
˚
˚
(0.06 A ((0.27 to 0.21 A) while the hydrogen bonding
distance decreases from 4.17 to 3.06 A. This comparison
˚
˚
indicates that increasing the size of the P3 side chain both
shortens the length of the hydrogen bond to Gly 216 and
reduces the residual motion along the hydrogen bond.
Conclusions
This study demonstrates that the presence of a hydrogen
bond adjacent to an inhibitor’s hydrophobic side chain can
mutually increase the strength of these interactions compared
to what would be expected based upon simple additivity. This
increase in strength is suggested to be due to the hydrogen
bond holding the hydrophobic side chains closer and more
firmly against the hydrophobic pocket wall and conversely the
side chain increasing the strength of the hydrogen bond due to
stabilizing its geometry. Thus, a hydrophobic interaction and
the hydrogen bond can mutually reinforce each other. The
data also demonstrate that cooperativity between hydro-
phobic interactions and hydrogen bonds can be longer range,
albeit with reduced amplitude, extending from the thrombin
S1 pocket to the S3 pocket. Finally, the magnitude of the
cooperativity between the adjacent hydrogen bond and the
hydrophobic interaction in the S3 pocket was shown to be
reduced in the presence of additional hydrogen bonds/salt
bridge in the S1 pocket. To the best of our knowledge, this is
thefirst in-depthexperimental demonstration of cooperativity
between hydrophobic interactions and hydrogen bonds in
protein-ligand complexes. The results obtained suggest that
scoring function predictions of ligand binding affinities could
be improved if cooperativity were included. The potential
Biological Assay. Kinetic inhibition of human thrombin
(from Beriplast, CSL Behring, Marburg, Germany) was deter-
mined photometrically at 405 nm using the chromogenic
substrate Pefachrom tPa (LoxoGmbH, Dossenheim, Germany)
€
according to the protocols described by Sturzebecher et. al.
20

Article
Journal of Medicinal Chemistry, 2010, Vol. 53, No. 5 2135
under the following conditions: 50 mM Tris/HCl, pH 7.4,
154 mM NaCl, 5% DMSO, 0.1% PEG 8000 at 25 °C using
different concentrations of substrate and inhibitor. Ki values
(n g 3) were determined as described by Dixon.21
(9) Lafleur, K.; Huang, D.; Zhou, T.; Caflisch, A.; Nevado, C.
Structure-based optimization of potent and selective inhibitors of
the tyrosine kinase erythropoietin producinghuman hepatocellular
carcinoma receptor B4 (EphB4). J. Med. Chem. 2009, 52, 6433–
6446.
(10) Lumma, W. C., Jr.; Witherup, K. M.; Tucker, T. J.; Brady, S. F.;
Sisko, J. T.; Naylor-Olsen, A. M.; Lewis, S. D.; Lucas, B. J.; Vacca,
J. P. Design of novel, potent, noncovalent inhibitors of thrombin
with nonbasic P-1 substructures: rapid structure-activity studies
by solid-phase synthesis. J. Med. Chem. 1998, 41, 1011–1013.
(11) Tucker, T. J.; Brady, S. F.; Lumma, W. C.; Lewis, S. D.; Gardell,
S. J.; Naylor-Olsen, A. M.; Yan, Y.; Sisko, J. T.; Stauffer, K. J.;
Lucas, B. J.; Lynch, J. J.; Cook, J. J.; Stranieri, M. T.; Holahan,
M. A.; Lyle, E. A.; Baskin, E. P.; Chen, I. W.; Dancheck, K. B.;
Krueger, J. A.; Cooper, C. M.; Vacca, J. P. Design and synthesis of
a series of potent and orally bioavailable noncovalent thrombin
inhibitors that utilize nonbasic groups in the P1 position. J. Med.
Chem. 1998, 41, 3210–3219.
Acknowledgment. We kindly acknowledge CSL Behring,
Marburg, Germany, for supplying us with generous amounts
of human thrombin from production of Beriplast. L.M.,
M.S., M.F., and D.G.H. contributed the design and synthesis
of the compounds, the modeling studies, and the MD simula-
tions, as well as collaborative interpretation of the data.
B.B., A.H., and G.K. contributed to the kinetic and the
collaborative interpretation of the data.
Supporting Information Available: Experimental details for
the synthesis of all the compounds. This material is available free
of charge via the Internet at http://pubs.acs.org.
(12) Baum, B.; Mohamed, M.; Zayed, M.; Gerlach, C.; Heine, A.;
Hangauer, D.; Klebe, G. More than a simple lipophilic contact: a
detailed thermodynamic analysis of nonbasic residues in the s1
pocket of thrombin. J. Mol. Biol. 2009, 390, 56–69.
(13) Dullweber, F.; Stubbs, M. T.; Musil, D.; Sturzebecher, J.; Klebe,
G. Factorising ligand affinity: a combined thermodynamic and
crystallographic study of trypsin and thrombin inhibition. J. Mol.
Biol. 2001, 313, 593–614.
References
(1) Whitesides, G. M.; Krishnamurthy, V. M. Designing ligands to
bind proteins. Q. Rev. Biophys. 2005, 38, 385–395.
(2) Warren, G. L.; Andrews, C. W.; Capelli, A. M.; Clarke, B.;
LaLonde, J.; Lambert, M. H.; Lindvall, M.; Nevins, N.; Semus,
S. F.; Senger, S.; Tedesco, G.; Wall, I. D.; Woolven, J. M.; Peishoff,
C. E.; Head, M. S. A critical assessment of docking programs and
scoring functions. J. Med. Chem. 2006, 49, 5912–5931.
(3) Hubbard, R. E. 3D structure and the drug-discovery process. Mol.
BioSyst. 2005, 1, 391–406.
(4) Talhout, R.; Villa, A.; Mark, A. E.; Engberts, J. B. Understanding
binding affinity: a combined isothermal titration calorimetry/
molecular dynamics study of the binding of a series of hydropho-
bically modified benzamidinium chloride inhibitors to trypsin.
J. Am. Chem. Soc. 2003, 125, 10570–10579.
(14) Stubbs, M. T.; Oschkinat, H.; Mayr, I.; Huber, R.; Angliker, H.;
Stone, S. R.; Bode, W. The interaction of thrombinwith fibrinogen.
A structural basis for its specificity. Eur. J. Biochem. 1992, 206,
187–195.
ꢀ
(15) Lila, C.; Gloanec, P.; Cadet, L.; Herve, Y.; Fournier, J.; Leborgne,
F.; Verbeuren, T. J.; De Nanteuil, G. Large-scale preparation of
protected 4-aminomethylbenzamidine. Application to the syn-
thesis of the thrombin inhibitor, melagatran. Synth. Commun.
1998, 28, 4419–4429.
(16) Vallone, B.; Miele, A. E.; Vecchini, P.; Chiancone, E.; Brunori, M.
Free energy of burying hydrophobic residues in the interface
between protein subunits. Proc. Natl. Acad. Sci. U.S.A. 1998, 95,
6103–6107.
(17) Bohm, H.-J., Schneider, G., Eds. Protein-Ligand Interactions:
From Molecular Recognition to Drug Design; Wiley-VCH: Wein-
heim, Germany, 2003; pp 3-20.
(18) Baum, B.; Muley, L.; Smolinski, M.; Heine, A.; Hangauer, D. G.;
Klebe, G. Unpublished results.
(19) Humphrey, W.; Dalke, A.; Schulten, K. VMD: visual molecular
dynamics. J. Mol. Graphics 1996, 14, 33–38.
(20) Sturzebecher, J.; Sturzebecher, U.; Vieweg, H.; Wagner, G.;
Hauptmann, J.; Markwardt, F. Synthetic inhibitors of bovine
factor Xa and thrombin comparison of their anticoagulant effi-
ciency. Thromb. Res. 1989, 54, 245–252.
(5) Nguyen, B.; Neidle, S.; Wilson, W. D. A role for water molecules in
DNA-ligand minor groove recognition. Acc. Chem. Res. 2009, 42,
11–21.
(6) Dill, K. A. Additivity principles in biochemistry. J. Biol. Chem.
1997, 272, 701–704.
(7) Williams, D. H.; Searle, M. S.; Mackay, J. P.; Gerhard, U.;
Maplestone, R. A. Toward an estimation of binding constants in
aqueous solution: studies of associations of vancomycin group
antibiotics. Proc. Natl. Acad. Sci. U.S.A. 1993, 90, 1172–1178.
(8) Williams, D. H.; Stephens, E.; O’Brien, D. P.; Zhou, M. Under-
standing noncovalent interactions: ligand binding energy
and catalytic efficiency from ligand-induced reductions in motion
within receptors and enzymes. Angew. Chem., Int. Ed. 2004, 43,
6596–6616.
(21) Dixon, M. The graphical determination of Km and Ki. Biochem.
J. 1972, 129 (1), 197–202.