Generation of ligand conformations in continuum solvent consistent with protein active site topology: Application to thrombin

Article abstract of DOI:10.1021/jm021028j

Using the crystal structure of an inhibitor complexed with the serine protease thrombin (PDB code 1UVT) and the functional group definitions contained within the Catalyst software, a representation of the enzyme's active site was produced (structure-based

Full text of DOI:10.1021/jm021028j

J . Med. Chem. 2003, 46, 1293-1305
1293
Articles
Gen er a tion of Liga n d Con for m a tion s in Con tin u u m Solven t Con sisten t w ith
P r otein Active Site Top ology: Ap p lica tion to Th r om bin
Paulette A. Greenidge,‡ Sandrine A. M. Me´rette,‡ Richard Beck, Guy Dodson, Christopher A. Goodwin,#
Michael F. Scully,# J ohn Spencer,‡ J o¨rg Weiser,§ and J ohn J . Deadman*,‡
Chemistry Department, Drug Discovery Division, and Biochemistry Department, Thrombosis Research Institute,
Emanuelle Kaye Building, Manresa Road, London, SW3 6LR. U.K., Anterio Consult & Research GmbH, Augustaanlage 26,
68165 Mannheim, Germany, and Protein Structure Division, National Institute for Medical Research, The Ridgeway,
Mill Hill, London, NW7 1AA. U.K.
Received August 20, 2002
Using the crystal structure of an inhibitor complexed with the serine protease thrombin (PDB
code 1UVT) and the functional group definitions contained within the Catalyst software, a
representation of the enzyme’s active site was produced (structure-based pharmacophore model).
A training set of 16 homologous non-peptide inhibitors whose conformations had been generated
in continuum solvent (MacroModel) and clustered into conformational families (XCluster) was
regressed against this pharmacophore so as to obtain a 3D-QSAR model. To test the robustness
of the resulting QSAR model, the synthesis of a series of non-peptide thrombin inhibitors based
on arylsuphonyl derivatives of an aminophenol ring linked to a pyridyl-based S1 binding group
was undertaken. These compounds served as a test set (20-24). The crystal structure for the
novel symmetrical disulfonyl compound 24, in complex with thrombin, has been solved. Its
calculated binding mode is in general agreement with the crystallographically observed one,
and the predicted Ki value is in close accord with the experimental value.
In tr od u ction
brinogen, the hirudin peptide, and sodium binding8 and
in the interactions of its mutants with these ligands and
reflects only small movements of the protein ternary
structure.9 For this reason, the structures with diverse
ligands should be comparable, and indeed it has even
been possible to identify conserved waters.10 Several
approaches11,12 have been described to rationalize this
extensive data and to allow extrapolation to the design
of new inhibitors. For example, the comparative molec-
ular similiarity indices approach (CoMSIA)11 was ap-
plied, but this method does not consider the structure
of the receptor. In contrast, the “linear response”12
method which does utilize the target was used to
develop several equations to predict thrombin inhibitor
affinities. While this method allows a range of factors
that are important to inhibitor-enzyme interactions to
be included, the Monte Carlo method is inherently
computationally expensive.
Bursi and Grootenhuis13 have previously correlated
theoretical and experimentally determined binding data
for a series of thrombin inhibitors. They found that
molecular mechanics minimization of inhibitors in the
receptor structure in which they had been cocrystallized
only provided a statistically significant correlation (R
) 0.74, 14 inhibitors) for those complexes with a
resolution of e 2.5 Å. Comparative molecular field
analysis (CoMFA)14 which only uses the information of
the inhibitors performed better (R ) 0.95, cross-
validated r2 (q2) ) 0.46) when used with high quality
ab initio charges and minimized conformations of crystal
structure inhibitors (crystal structure alignment). Thus,
in this case, the success of both the aforementioned
Thrombin (EC 3.4.21.5) has been implicated in the
aetiology of a number of disease states including throm-
botic disease,1 cancer, and endotoxic shock, and for this
reason inhibition of its proteolytic activity continues to
be the subject of inhibitor design after three decades.2
Second generation, nonpeptidic thrombin inhibitors are
now in phase II3 and phase III4 clinical development.
Pooling the combined research in the thrombin area
provides numerous libraries of inhibitors with diverse
structures and a range of affinities for thrombin span-
ning 9 orders of magnitude (millimolar to picomolar).
In addition, there are at least 124 structures of inhibitor
complexes with thrombin deposited in the Protein Data
Bank (Merops database, www.MEROPS.Sanger.Ac.Uk,
code S01.217). Thrombin, in contrast to the other
coagulation proteinases, shows only small changes in
catalytic constants for its substrates upon binding to
cofactors. For example, binding of thrombin to glyco-
protein 1b5 increases kcat/Km for cleavage of PAR-1 by
only 6-fold,6 while binding of FXa to FVa or FIXa to
FVIIIa cause an increase of 3 orders of magnitude.7 This
small scale of allosteric change for thrombin is also
reflected in its interactions with thrombomodulin, fi-
* To whom correspondence should be addressed. Tel: ++44+207-
351 8330 ext. 3622. Fax: ++44+207-351 8324. Email: jdeadman@trigen.
co.uk.
‡ Chemistry Department, Drug Discovery Division, Thrombosis
Research Institute.
#
Biochemistry Department, Thrombosis Research Institute.
§ Anterio Consult & Research GmbH.
National Institute for Medical Research.
10.1021/jm021028j CCC: $25.00 © 2003 American Chemical Society
Published on Web 03/18/2003

1294 J ournal of Medicinal Chemistry, 2003, Vol. 46, No. 8
Greenidge et al.
Ta ble 1. Structure, Experimental (exptl), Calculated (calcd) Ki Values, and Catalyst Conformational Energies of the 16 Training Set
Inhibitors Plus a Test Set Inhibitor Taken from the Literature (Compound 17)
methods apparently relies on access to crystal structure
data for each inhibitor-enzyme complex. However, this
dependence makes these approaches impracticable dur-
ing the early stages of a medicinal chemistry program;
hence, our investigation of the novel combination of a
structure-based pharmacophore15 (representation of
important inhibitor-enzyme interactions in which ex-
cluded volumes that are not penetrable by the inhibitors
are used to define the demarcation of the active site)
constructed from a single high-resolution crystal struc-
ture complex, the Catalyst16 pharmacophore mapping
method, and in continuum solvent conformations of
ligands generated in MacroModel.17 The work described
here considers a subset of that same set of thrombin
inhibitors.13 The structure-based pharmacophore (Fig-
ure 1) was built using the crystallographic coordinates
of compound 1 (Table 1) complexed with thrombin (PDB
code 1UVT)18 to predict the effects on the Ki values of
structural modification of a set of homologous 4-ami-
nopyridine (4-AP) thrombin inhibitors (Table 1).
Catalyst16 attempts to produce a 3D-QSAR model that
correlates estimated activities with measured activities.
A pharmacophore model consists of a collection of
features necessary for the biological activity of the
training set arranged in 3D-space. Common pharma-
cophore features include hydrogen bond donor, hydrogen
bond acceptor, and hydrophobe. The fewer features an
inhibitor maps to, and the poorer its fit to them, then
the lower its affinity will be predicted to be. The
simplistic scoring function does not explicitly contain
terms for ligand solvation nor protein and ligand
entropy; thus its use may be limited to congeneric ligand
series.19 The work of Vieth et al.20 supports the phar-
macophore approach21 and the use of conformational
models generated in solution. By means of molecular
dynamics simulations, atoms that play a key role in
ligands binding affinity were identified as those having
the lowest mobility and largest interaction energy with
the receptor. These so-called “anchor” atoms have a
similar spatial orientation both in the active site and
in solution. In an analysis of conformational energy
penalties of protein-bound ligands, Bostro¨m et al.22 also
concluded that it is more appropriate to use the aqueous
rather than the in vacuo conformational ensemble of the
unbound ligand when making a comparison with the
bioactive conformation. Several recent studies,23 includ-
ing our own,24 have appeared in which the conforma-
In Catalyst,16 a conformational model is an abstract
representation of the accessible conformational space of
a ligand. It is assumed that the biologically active
conformation of a ligand (or a close approximation
thereof) should be contained within this model. Given
a ligand training set (ligand conformational models and
binding affinity values) and pharmacophore model,

Ligand Conformations with Protein Active Site Topology
J ournal of Medicinal Chemistry, 2003, Vol. 46, No. 8 1295
(iii) The methyl group occupies the S2-pocket.
(iv) In the aryl binding site, the phenyl group of the
inhibitor takes part in an edge-to-face aromatic stacking
interaction with Trp-215 and is flanked by a number of
other hydrophobic residues (S4-pocket).
The hydrogen bond of the pyridine ring NH group
with Asp-189 was not included in the pharmacophore
definition, but the centroid of the pyridine ring (common
to the inhibitors studied) defined as a hydrophobic (S1-
pocket) aromatic center instead since the major interac-
tions in the S1-pocket are not simply driven by electro-
statics. Noncovalent thrombin inhibitors that have
nonbasic groups in the P1 position and that exhibit high
affinity exist,26 and it has been demonstrated that the
4-aminopyridine moiety contributes only negligibly to
the total binding energy of such ligands to thrombin.27
Consequently, all molecules were constructed with an
unprotonated pyridine moiety. The centroids of the two
remaining aromatic rings of the inhibitor were likewise
used to define hydrophobic aromatic features. Thus,
such a pharmacophore definition allows for alternative
binding modes by not biasing the pyridine ring to occupy
the S1-pocket. A precedent for unexpected “inverse”
binding modes with peptidic thrombin inhibitors is
known, where the interaction of an N-terminal trifluo-
roacetyl group with the enzyme catalytic triad was
predominant.28 The methyl carbon atom was defined as
a hydrophobic feature (S2-pocket), and the Ser-214
carbonyl oxygen atom as a hydrogen bond donor feature.
The other end of the vector was removed. The spherical
tolerance for all features was 1.5 Å. BM14.1248 (com-
pound 1) was extracted from the enzyme/inhibitor
complex, and this crystallographic conformation was
stored. BM14.1248 was able to successfully map to all
features of the described pharmacophore. Subsequently,
the remaining atoms delimiting the active site were
represented as excluded volumes (space which the
inhibitor is not allowed to occupy) defined within a cut
off of 6 Å from any atom of the inhibitor using Insight
II.29 One pharmacophore model was constructed in
which the excluded volumes were scaled at 30% of their
respective atomic van der Waals radii.30
F igu r e 1. Catalyst mapping of the crystal structure confor-
mation of compound 1 with thrombin to the structure-based
pharmacophore. Three aromatic-hydrophobic features (dark
blue spheres) correspond to the three aromatic rings of
compound 1 and a general hydrophobic feature (light blue) to
its methyl group. A hydrogen-bond donating feature (cyan) is
derived from the interaction of the 4-aminopyridine group with
the carbonyl oxygen atom of amino acid residue Ser-214
(chymotrysin numbering scheme). The excluded (ligand-inac-
cessible) volumes (black spheres) correspond to the atoms
delimiting the active site of thrombin. The important regions
of the enzyme active site, the specificity pocket of thrombin,
S1, and the two hydrophobic pockets S2 and S4 are labeled.
tions of molecules in aqueous solution have been used
to establish a pharmacophore model. In our previous
study,24 conformations were generated in continuum
solvent (MacroModel)17 for this data set (Table 1) in
which ligands contain up to 14 rotatable bonds. Each
conformational model typically consisted of several
thousand conformers, thus making the calculation of the
mapping mode to the pharmacophore model slow.
However, by clustering the results of a conformational
search into conformational families with the program
25
XCluster Version 1.7, the number of conformers has
now been reduced to a more tractable number. Hence,
in continuum solvent generated conformational models
are now computationally amenable to use for pharma-
cophore mapping of large data sets. A crystal structure
for a newly synthesized inhibitor cocrystallized with
thrombin is presented and it is checked whether the
highest scoring pharmacophore binding mode is indeed
consistent with available crystal structure data.
Tr a in in g a n d Test Sets. The training set (inhibitor
conformational models and Ki values) for the regression
of the structure-based pharmacophores consisted of 16
inhibitors taken from the data set of Bursi and Grooten-
huis13 (inhibitors with undefined stereochemistry were
excluded) (Table 1). Sixteen compounds represent the
minimum recommended number of molecules to be
included in a training set. Compound 17 (Table 1) was
selected as a test set inhibitor to examine the versatility
of the structure-based pharmacophore (see Results and
Discussion). The inhibitor/thrombin crystal structure
complex is available for compound 17 (PDB reference
code 1UVS). Note that care should be taken when
importing conformational models generated external to
Catalyst as the recalculation of conformational energies
may result in some conformers being given a conforma-
tional energy above the default 84 kJ /mol threshold such
that these conformers are excluded from the mapping
process. However, all conformers of the training set had
energies calculated by Catalyst to be less than 84 kJ /
mol above the lowest energy conformer for a given
compound. Three conformers of compound 17 were
Meth od s
Ca ta lyst P h a r m a cop h or e Con str u ction a n d 3D-
QSAR. Str u ctu r e-Ba sed P h a r m a cop h or es. The lo-
cation of features in the structure-based pharmaco-
phores were defined by the crystallographic coordinates
of atoms in the BM14.1248/thrombin complex (PDB
reference code 1UVT) in which water molecules had
been removed.
In ter a ction s of Com p ou n d 1 w ith Th r om bin .
(i) The 4-amino group is hydrogen bonded directly to
the Ser-214 (chymotrypsin numbering) carbonyl oxygen
atom.
(ii) The proton of the NH group of the pyridine ring
forms a hydrogen bond with a carboxylate oxygen atom
of Asp-189 in the S1-pocket.

1296 J ournal of Medicinal Chemistry, 2003, Vol. 46, No. 8
Greenidge et al.
considered to have conformational energies above the
threshold. These were stored as separate molecules by
using the “show conformational model” option, followed
by “add to view”. The conformer was then saved and
the necessary conformational model was made by
registering this conformer. The conformational model
giving the best fit score for compound 17 is discussed
in Results and Discussion. The catScramble script was
used to randomize (five trials) the experimental values
of the training set spreadsheet for regression with the
pharmacophore. All calculations were performed using
an Iris Indigo Elan, R4000, memory size 160 Mbytes,
100 MHZ IP20 processor.
the numbering of all other conformational families is
arbitrary with regard to conformational energies. By
means of comparison with “exhaustive” conformational
models (generated by a modified systematic search
method in which torsion angles are varied in specified
increments) and assessment with two quantitative
metrics, it has been shown that a small number of
conformations are sufficient to represent the low-energy
conformational spaces of small- to medium-sized mol-
ecules (using the method of conformation generation as
implemented in Catalyst).37 This led to the derivation
of a relationship between the number of rotatable bonds
possessed by a molecule, the number of conformers and
the resolution of a conformational model. The resolution
of a conformational model (subset) can be defined in
terms of the maximum expected rms distance of an
arbitrary low-energy conformer extracted from an ex-
haustive set (larger set), from a conformation in the
model.37 The number of rotatable bonds possessed by
molecules in this data set is in the main between 7 and
10, the exceptions being compound 2 with 14 (and
compound 17 with 12). Accordingly, a Catalyst gener-
ated conformational model size of 200 is appropriate for
the choice of tolerance sphere radius (1.5 Å). For this
reason, we also chose a MacroModel cluster size of 200.
(5) Store the leading conformation (conformation with
the lowest energy) of each conformational family. Ac-
cording to the previous point of this protocol the leading
conformation of the first conformational family is always
the GEM.
(6) For each compound write one multiple coordinate
file, which contains the 200 leading conformations in
ascending order of the numbering of the conformational
families. Conveniently the first conformation of such a
multiple coordinate file is therefore always the GEM
(see points 4 and 5 of this protocol). Each multiple
coordinate file represents the clustered in continuum
solvent conformational model of each compound. All
calculations were performed either with the Linux
version of MacroModel 7.217 on a 750 MHz Pentium III
processor or the Sun version on a Sun Ultra 2 machine.
Ma cr oMod el/XClu ster . The following protocol was
used to generate conformations for each compound:
(1) Execute an exhaustive conformational search
using the molecular modeling program MacroModel17-
(version 7.2). For calculations on all compounds the
latest development of the current best general-purpose
force-field for medicinal chemistry, MMFF94s,31 was
used in combination with the GB/SA solvation model.32
GB/SA treats solvent as analytical dielectric continuum
that starts near the van der Waals surface of the solute
and extends to infinity. The model includes both gen-
eralized Born-based (GB) solvent polarization terms and
surface area-based (SA)33 solvent displacement terms.
All nonbonded cutoffs were set to infinity for all calcula-
tions. Energy minimizations were performed with the
Truncated Newton-Raphson conjugate gradient (TN-
CG)34 method, which involves the use of second deriva-
tives; the derivative convergence criterion was set to
0.05 kJ /Å-mol. Conformational search was performed
by the Monte Carlo35 method for the random variation
of all of the rotatable bonds combined with the so-called
low mode conformational search (LMCS)36 algorithm.
For each calculation 10 000 Monte Carlo steps were
carried out.
(2) Sort all found conformations according to energy.
(3) Store only conformations whose molecular me-
chanics energy differences to the calculated global
energy minimum of a compound (GEM) are below 15
kJ /mol. The “exhaustive” conformational models (both
in vacuo and in continuum solvent) in our earlier study
were based on conformational energy thresholds of 42
kJ /mol. It was decided to decrease that energy window
to avoid rather unpopulated conformations (according
to the Boltzmann distribution) and distorted conforma-
tions being classified as conformational families, since
the number of conformational families is restricted (see
point 4 in this protocol). Additionally, this threshold is
in line with the results of a study by Bostro¨m et al.22
They found that for 70% of the protein-ligand com-
plexes that they investigated, the energies of the bio-
active conformations were within 12.5 kJ /mol of their
respective minimum energy conformation. Uncertainties
in the interpretation of the experimental data or limita-
tions of the computational methods accounted for the
higher calculated penalties in the remaining 30% of
complexes.
Resu lts a n d Discu ssion
Com p a r ison of “Exh a u stive” a n d “Clu ster ed ” In
Con tin u u m Solven t Con for m a tion a l Mod els: Im -
p a ct on 3D-QSAR Mod el. The crystal bound confor-
mation of compound 1 (BM14.1248) with thrombin (PDB
code 1UVT), and its interactions therewith, were used
to define a structure-based pharmacophore (Methods,
Figure 1). The excluded volumes (space which the
inhibitor is not allowed to penetrate and defined by
atoms delimiting the active site) were scaled to 30% of
their respective van der Waals radii to allow for direct
comparison with an earlier study.24 Scaling may be a
necessary consequence of the assumption of a rigid
enzyme structure as Murray et al.38 have shown that
for a given enzyme (the data set included thrombin) the
docking success of flexible ligands is reduced against
enzyme crystal structures derived from other ligands.
In that earlier study,24 “exhaustive” conformational
models for thrombin inhibitors (Table 1) both in vacuo
and in continuum solvent with the MacroModel17 soft-
ware were produced, as well as a conformational model
using Catalyst (“best” mode). Three criteria were uti-
(4) Cluster the remaining conformations in exactly
200 different conformational families (numbered from
1 to 200) based on atomic RMSD after rigid-body
superposition using XCluster.25 The conformational
family with the number 1 always contains the GEM;

Ligand Conformations with Protein Active Site Topology
J ournal of Medicinal Chemistry, 2003, Vol. 46, No. 8 1297
Ta ble 2. A Comparison of the Structure-Based Pharmacophore
(Figure 1) Featuresa Mapped by the Training Set Inhibitors
Whose Conformational Models Are Generated In Continuum
Solvent
lized to assess the quality of the resulting 3D-QSAR
models following regression of the structure-based phar-
macophore (Figure 1) with these collections of conform-
ers. On the basis of the correlation coefficient and root-
mean-square deviation (RMSD) between estimated and
observed Ki values, and the ability to reproduce crys-
tallographically observed binding modes, the continuum
solvent conformational model was judged to produce the
best 3D-QSAR model with the Catalyst generated
conformers performing least well.24 For instance, com-
pound 1 was calculated to bind in an inverse binding
mode. The S4-feature was not mapped and the benzene-
sulfone moiety occupied the S1 pocket instead of the
pyridine ring as observed in the crystal structure
complex (1UVT). However, the closely related compound
9 assumed the expected binding mode and mapped four
of the five pharmacophore features. Given the conserva-
tive nature of the training set the inconsistency of these
results was disappointing. Both the Macromodel17 “ex-
haustive” conformational models for compound 1 re-
sulted in a mapping mode such that all five features of
the pharmacophore were satisfied as in the crystal
structure complex (1UVT). Each “exhaustive” confor-
mational model consisted of up to several thousand
conformers, making them slow to import into Catalyst
and in performing calculations thereafter; however, the
size of the conformational models has now been reduced
by clustering them into conformational families (see
Methods).
compound
S1
HBD
S2
central phenyl ring
S4
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
-
-
-
-
-
+
+
+
+
+
+
+
-
-
+
+
+
+
+
+
+
+
-
+
-
-
-
-
-
-
-
-
+
+
+
+
+
+
+
+
+
+
-
+
+
+
+
+
+
+
+
+
+
-
-
-
+
-
+
+
+
-
+
+
a
S1, S2, and S4 refer to hydrophobic features corresponding to
the S1-, S2-, and S4-pockets respectively of thrombin. HBD refers
to the hydrogen bond donor feature defined by the Ser-214
carbonyl oxygen atom and the centroid of the central phenyl ring
of compound 1 (Table 1) is used to define a hydrophobic feature.
Note that no indication is given of how well a particular feature
is mapped by an inhibitor.
In the current work, after regression of the pharma-
cophore model (Figure 1) with the training set (“clus-
tered” in continuum solvent conformational model,
Table 1) the correlation coefficient and root-mean-square
deviation (RMSD) between estimated and observed Ki
values have been examined. As compared to the 3D-
QSAR model regressed with the “exhaustive” in con-
tinuum solvent conformational model the quality of the
present model is degraded with R values of 0.69 and
0.75 and RMSD values of 1.32 and 1.21, respectively.
Interestingly, the figures for the clustered in continuum
solvent conformational model still compare favorably
with those for the “exhaustive” in vacuo conformational
model for which R ) 0.57 and RMSD ) 1.52.24 See point
(vii) below for a discussion as to why this is so. In the
present study, compound 1 does not map to the hydro-
gen bond feature defined by Ser-214 (Table 2), although
it does map to all four of the hydrophobic features.
Figure 2 shows the superimposition of the crystal
structure of compound 1 extracted from the complex
with thrombin and the conformation that maps to the
pharmacophore with “fast” fit. A RMSD value of 1.97 Å
is obtained over 36 atoms using the centroids of the
aromatic rings for superimposition. The greatest con-
formational variation between the conformers occurs
about the torsion defined by the atoms CA2 NA1 C4 C3
(Figure 2). A value of 178.4° is obtained for the crystal
structure conformation and this compares with an
equivalent value of 2.74° for the conformation in con-
tinuum solvent. Hence, the inability of the generated
conformation of compound 1 to map to the hydrogen
bond donor feature as the 4-amino hydrogen atom points
away from the Ser-214 carbonyl oxygen atom. Although
a conformer of compound 1 that is able to map to all
five features of the pharmacophore model does exist, in
F igu r e 2. Superimposition of the crystal structure conforma-
tion of compound 1 (bold) extracted from the complex with
thrombin (1UVT) and the conformation calculated to have the
best fit score to the pharmacophore model (light shade).
doing so it does not obtain the best fit score. This finding
may perhaps be attributed to two factors. First, that
the current Catalyst fit score may not be a perfect
measure of conformational quality, and second the
choice of tolerance radii. It may well be the case that
the tolerance radii for the pharmacophore regressed
with the “clustered” conformational model should be
made larger as compared to the pharmacophore re-
gressed with the “exhaustive” conformational models.
Str u ctu r e-Ba sed P h a r m a cop h or e Mod el a n d 3D-
QSAR. Catalyst assumes that all chemical features
contribute equally in providing binding energy, and that
compounds are more or less active because they possess
or do not possess features that contribute positively to
activity. By use of Tables 1 and 2 the mapping mode of
the ligands and the consequent effect on calculated Ki
values can be assessed. The contribution of the different
moieties of which compounds such as 1 are comprised,
to binding energy with thrombin has been examined.
The inhibition constant increases considerably when the
arylsulfonamide moiety is removed, and compounds that
lack the central phenyl group do not inhibit thrombin
at all. As compared to benzamidine the pyridine ring

1298 J ournal of Medicinal Chemistry, 2003, Vol. 46, No. 8
Greenidge et al.
Ta ble 3. Randomization of the Experimental Ki Data to Test
the Robustness of the 3D-QSAR Model
continuum solvent the minimum energy conformer of
compound 9 is relatively extended, the equivalent
conformer of compound 14 is folded such that the
benzene-sulfonamide ring and central aromatic ring
engage in a face to face pi-stacking interaction. To give
an indication of the difference in conformation between
these two molecules, the distance between the pyridine
ring nitrogen atom and para-carbon atom of the benzene-
sulfonamide moiety is given. The values for compounds
9 and 14 are 15.6 and 10.5 Å, respectively. So what in
two dimensions appears to the eye to be a minor
structural difference between compounds 9 and 14
(Table 1) in fact has a major impact on their Ki values
(Table 1), perhaps explained by the inability of com-
pound 14 to occupy the S4-pocket (Table 2). Using a 42
kJ /mol energy threshold, a conformer of compound 14
exists in the in vacuo conformational model such that
data
Ra
RMSD
1.32
nonscrambled
scrambled
trial 1
trial 2
trial 3
0.69
-0.12
-0.04
0.50
0.1
-0.1
1.94
1.93
1.60
1.90
1.94
trial 4
trial 5
a
Experimental Ki values (Kiexptl, Table 1) were randomized
using catScramble (see Methods) and the table shows the results
of correlation with calculated Ki Values in five trials.
penetrates deeper into the S1-pocket, but is a less
significant source of binding energy.27
Given that the pharmacophore features are based on
the interactions of compound 1 with thrombin, it is not
surprising that the most potent compounds, 2, 3, and 4
are calculated to have affinities less than or approxi-
mately equal to that of compound 1 even though they
have the potential to make additional interactions with
thrombin (Table 2). Thus, an obvious shortcoming of the
Catalyst 3D-QSAR approach is that a predefined phar-
macophore description is required. To generalize the
pharmacophore, more features could be included to
represent interactions thought to be important for
inhibitors binding to thrombin,39 but not necessarily
made by this particular series of compounds. For
example, in a series analogous to the current one, but
which has a P1 guanidinoalkyl moiety, the sulfonamide
NH group forms a weak hydrogen bond to the Gly-216
carbonyl oxygen atom.40 However, the sulfonamide
moiety of the present series is not assumed to directly
interact with the protein.
the affinity of compound 14 is overestimated (Kiexptl
/
Kicalcd ) 43). This contributes to the lower R and higher
RMSD values of the resulting 3D-QSAR model as
compared to that resulting from regression of either of
the continuum solvent conformational models (exhaus-
tive or clustered) with the pharmacophore model.
However, the model fares less well with respect to
compound 16 which has a benzylamine group instead
of a sulfonamide group as possessed by a majority of
the other molecules. A possible enhancement to the
pharmacophore might include describing the S4-pocket
as an aromatic feature instead of as a hydrophobe, thus
creating a more realistic description of the geometric
relationship between the phenyl and Trp-215 rings (see
Method point (iv)).
To test the robustness of the 3D-QSAR model, the
pharmacophore was regressed several times with
scrambled activities (Table 3). The correlation coef-
ficients for the randomized experimental data were less
than for the nonscrambled data and the RMSD values
higher. Thus, the best 3D-QSAR model is obtained from
correlating the calculated Ki values for the compounds
with their corresponding experimental Ki values. In
addition, before deciding if the model should prove
useful in guiding further synthesis a check should be
made whether compounds external to the training set
map to the pharmacophore in chemically reasonable
ways41 as it has been argued that validation of a QSAR
model can only be accomplished by use of an external
set of compounds.42 For this reason, the literature
compound 17 (Table 2) which although related to
compound 1 does show some structural variance was
chosen as part of the test set. In the crystal structure
with thrombin, compound 17 adopts an “Argatroban-
like” conformation such that the N-carboxypiperidyl (S2-
pocket) and cyclohexyl sulfonyl (S4-site) rings are in
close proximity and thus able to make a number of van
der Waals interactions with each other.43 The pyridine
ring occupies the S1-site, but the 4-amino group is
unable to form a hydrogen bond with the Ser-214 amino
acid residue. However, the mapping of the continuum
solvent generated conformation of compound 17 to the
pharmacophore predicts an “inverse” binding mode
relative to that observed in the crystallographic complex
with thrombin. The pyridine ring maps to the S4-
hydrophobe, the N-carboxypiperidyl ring occupies a
region close to the S1-hydrophobe, allowing the sulfona-
The pharmacophore model does seem to reflect the
trends of compound 9 versus:
(i) Compound 1. Compound 9 lacks a group capable
of mapping to the hydrophobe corresponding to the S2-
pocket (Table 2).
(ii) Compound 5, which has as an R2-subtituent as a
chlorine atom (Table 1), and in which the sulfonamide
nitrogen atom is methylated.
(iii) Compounds 6, 8, and 10. Given that the phar-
macophore is defined using compound 1 which has a
sulfonamide linker as compared to the shorter sulfone
linker common to these three compounds (Table 1), it
is not surprising that they are unable to map to the S4-
feature (Table 2).
(iv) Compound 7 which is in good agreement with our
measured Ki value (Table 3).
(v) Compound 11, which has the same Ki observed as
compound 9 (Table 1).
(vi) Compounds 12 and 15 for which excluded volume
problems may play a role (increasing the size of the
excluded volumes may increase the sensitivity of the
model).
(vii) Compound 13, which is unable to make a
hydrogen bond with Ser-214 as both its sulfonamide and
4-amino nitrogen atoms are methylated (Tables 1 and
2), and compound 14 in which the sulfonamide group
is reversed relative to compound 9. As compared to
compound 9 the sulfonamide linker of compound 14 is
reversed (Table 1), and this is conformationally signifi-
cant as the sulfone group is coplanar to the aromatic
ring to which it is directly attached. Whereas in

Ligand Conformations with Protein Active Site Topology
J ournal of Medicinal Chemistry, 2003, Vol. 46, No. 8 1299
Sch em e 1a
a
Reagents and conditions: (i) (a) ethylene glycol di-p-tosylate, DMF, NaH, (b) 3-aminophenol, NaH; (ii) benzenesulphonyl chloride
TEA, DCM; (iii) 50% TFA in DCM.
Ta ble 4. Structures and Predicted and Experimental (exptl) Ki
Values for Compounds Synthesized In-House
F igu r e 3. Superimposition of the crystal structure conforma-
tion of compound 17 (bold) extracted from the complex with
thrombin (1UVS) and the conformation calculated to have the
best fit score to the pharmacophore model (light shade).
mide group to map to the Ser-214 hydrogen bond donor
feature and the N-carboxy piperidyl to the S2-pocket.
The RMSD between the highest scoring conformer of
compound 17 that maps to the pharmacophore model
and the crystallographic conformation in complex with
thrombin (1UVS) is 1.89 Å, indicating an essentially
correct conformation but incorrect binding mode (Figure
3). The centroids of the crystal structure pyridine,
piperidyl, and cyclohexane rings respectively, were
superimposed on the centroids of the cyclohexane,
piperidyl and pyridine rings respectively of the con-
tinuum solvent conformation.
Syn th esized Test Set. Ch em istr y. To facilitate
parallel synthesis of the test set compounds (7 and 15)
as well as new analogues (20-24, Table 4) a novel route
a
Solubility problems.
was designed in four steps, via N-alkylation of the Boc
protected aminopyridine with a bis-tosylate, O-alkyla-
tion with the appropriate aminophenol and sulfonyla-
tion (Scheme 1). The Boc-protected 4-aminopyridine 18
was readily prepared in 65% yield by treating 4-ami-
nopyridine with di-tert-butyl dicarbonate in THF. N-
Alkylation of 18, using ethylene glycol di-p-tosylate,
followed by nucleophilic displacement of the tosylate
intermediate by 3-aminophenol, gave the intermediate
21 in 35% overall yield. The reaction of 19 with the
appropriate sulfonyl chloride, in the presence of base,
led to a mixture of the mono- and the di-sulfonamides
20-22 which were separated by flash chromatography.
Disulfonylation was unexpected since the more electron
rich 3-amino-5-methylphenol has been reported to react
cleanly in good yield. Removal of the Boc group, using
50% TFA in DCM, led to the extensive decomposition
of the sulfonamides, especially in the case of the
benzenesulfonamides. The toluenesulfonamides 15 and
24 were obtained in very low yields (5%).

1300 J ournal of Medicinal Chemistry, 2003, Vol. 46, No. 8
Greenidge et al.
Sch em e 2a
a
Reagents and conditions: (i) (a) ethyl chloroacetate, potassium carbonate, DMF; (ii) H2, Pt(IV)O2, EtOH; (iii) benzenesulfonyl chloride,
pyridine, DCM; (iv) methyl iodide, potassium carbonate, DMF; (v) potassium hydroxide, MeOH; (vi) (a) thionyl chloride, (b) 4-aminopyridine,
NMM, DMF; (vi) LiAlH4, THF.
Due to these poor yields, and to allow unsymmetrical
substitution of the aminophenol nitrogen, a new seven-
step route (Scheme 2) was investigated in which the
N-alkyl sulfonamides were prepared via: (i) O-alkyla-
tion of 3-nitrophenol, (ii) reduction of the nitro group to
afford the amino derivative, and 3) reaction of the latter
with benzenesulfonyl chloride, to form the desired
sulfonamide 25 in 47% overall yield. To avoid disulfo-
nylation the reaction with benzenesulfonyl chloride (one
equivalent) and the mild base, pyridine, was followed
by TLC and quenched at 0 °C to give 25. N-methylation
of the sulfonamide was carried out using methyl iodide
in the presence of base in DMF to produce 26 in good
yield (88%). However, hydrolysis of the ethyl ester group
of the sulfonamides gave the acids in poor yields (31-
54%). The acid 26 was converted into an acid chloride
before being reacted with 4-aminopyridine to form the
amides 30 in 53% yield. Reduction of 30 using lithium
aluminum hydride in THF gave the desired sulfonamide
7 in 51% yield.
Mod elin g. Due to the difficulty of the synthesis of
substituted aminophenol derivatives, this limited the
diversity of the synthesized test set. Table 4 shows the
structures and observed and predicted Ki values for the
five new compounds synthesized in house (20-24) plus
two additional compounds (7 and 15) which were
prepared to provide a reference for our measured Ki
values versus those in the literature. The Ki values for
compounds 20 and 21 are somewhat under predicted
but the poor solubility of the compounds is likely to be
a contributory factor. In the pharmacophore model, the
Boc group shared by compounds 20 and 21 and one tosyl
group belonging to the di-tosyl compound 21 point into
the solvent. Neither the hydrogen bond donor or the S2-
features are mapped by the molecules, but the S1-pocket
is occupied by the pyridine ring common to them. The
S4-feature and hydrophobe defined by the central
aromatic ring of compound 1 are also mapped. Com-
pounds 22 and 24 are predicted to map to the pharma-
cophore in a manner similar to each other, though
compound 24 satisfies four out of five features as
compared to only three by compound 22, the presence
of the Boc group preventing the participation of its
4-amino group in hydrogen bonding. As expected, the
S2-feature is not mapped by either of these molecules.
The predicted Ki values for compounds 22 and 24 are
in good agreement with the observed ones, and it will
be discussed below how the calculated binding modes
fair with respect to the experimental data. Compound
23 is predicted to have too high an affinity and calcu-
lated to map to the pharmacophore in an inverse
binding mode.
A Com p a r ison of th e Cr ysta l Str u ctu r e Bin d in g
Mod es of Com p ou n d s 1 a n d 24. As in the crystal
structure complex of compound 1 with thrombin, the
hydrogen bonds are conserved between the 4-amino
group with the Ser-214 carbonyl oxygen atom (Figure
4; N-O distance 2.77 Å for 1UVT and 3.28 Å for
compound 24, respectively), and the pyridine proton
with Asp-189 (distances from the pyridyl nitrogen atom
to Asp-189 OD1 and OD2 are 3.01 and 2.95 Å, respec-
tively for 1UVT as compared to 3.1 and 2.83 Å for
compound 24). The C3 methyl group of the central aryl
ring of 1UVT occupies the S2-site more fully than the
unsubstituted phenyl ring of compound 24 and this
results in some movement of the side-chain of Trp-60D.

Ligand Conformations with Protein Active Site Topology
J ournal of Medicinal Chemistry, 2003, Vol. 46, No. 8 1301
F igu r e 4. Superimposition of the crystal structures of compound 1 (bold) and compound 24 in complex with thrombin (light
shade).
However, after structural alignment the RMSD between
the alpha-carbons of the two crystal structure complexes
is only 0.51 Å and the overall difference between the
two complexes is small. One tosyl group of compound
24 engages in an edge-to-face pi-stacking interaction
with Trp-215, but is displaced relative to the sulfona-
mide-benzene moiety of compound 1. This is necessary
to avoid a clash of the methyl group with the alpha-
carbon atom of Asn-98. The separation of the Asn-98
alpha-carbon atom from the tosyl methyl of compound
24 is 4.05 Å while for the para position of the phenyl
ring of compound 1 the distance is 3.48 Å. The overall
conformation of the molecule is however maintained
such that the dihedral angle subtended by the atoms
phenylC-N-S-arylC is 64.13° in 1UVT and 83.57° for
compound 24. The second tosyl group of compound 24
is solvent exposed, and in the crystal structure is
F igu r e 5. Superimposition of the crystal structure conforma-
tion of compound 24 (bold) extracted from the complex with
observed to form van der Waals interactions with the
side chain methylenes of Glu-217 stabilizing an interac-
thrombin and the conformation calculated to have the best fit
tion of the Glu-217 carboxylate oxygen atom with the
sulfur atom (3.07 Å for compound 24 as compared to
5.85 Å for compound 1) of the other, inward facing, tosyl
group. This interaction is possibly an artifact of the
hydrophobic environment of the crystal since it is not
observed in the 1UVT complex or that of the monotosyl,
N-methyl compound 7 (unpublished results), and would
require disruption of a favorable surface salt bridge
between Glu-217 and Lys224 (2.90 Å). Thus, the calcu-
lated binding mode of compound 24 in complex with
thrombin is consistent with the crystallographic one.
Upon the basis of the superimposition of the centroids
of the pyridine and central rings, and a tosyl group of
compound 24 (pharmacophore and crystal structure
conformation), the RMSD is 2.23 Å over 36 atoms
(Figure 5). Even though no pharmacophore feature
existed to describe the location of the second tosyl group,
there is an obvious similarity in the spatial disposition
of this moiety in the crystal structure and in the
pharmacophore conformation.
score to the pharmacophore model (light shade).
as a close correspondence between the solution and
bound conformations is entropically advantageous for
the ligand.45 Clustering conformers generated by an
exhaustive search in continuum solvent into conforma-
tional families allowed pharmacophore mapping calcu-
lations to be more quickly performed. The quality of the
3D-QSAR model after regression with these conforma-
tional models is considered to be reduced relative to the
exhaustive in continuum solvent conformational model
based on the RMSD and correlation coefficient between
calculated and observed affinities. However, using these
two criteria the clustered in continuum solvent confor-
mational model out-performs the 3D-QSAR model re-
gressed with the exhaustive in vacuo conformational
model, but this is largely attributable to the difference
in the value of Ki calculated for one compound, com-
pound 14. Importantly, it was possible to rationalize
differences in the binding affinities of the inhibitors
based on their calculated mappings to the pharmaco-
phore model; randomizing the experimental data led to
a worse 3D-QSAR model. The binding modes of inhibi-
tors external to the training set were well produced
upon comparison with conformations when cocrystal-
lized with thrombin. MacroModel17 in continuum sol-
vent conformational models offer an alternative to
default Catalyst16 conformer generation, and may result
Con clu sion s
The suitability of the combination of a structure-based
pharmacophore model and solvent generated thrombin
inhibitor conformations as a 3D-QSAR method requiring
limited structural information was investigated. It has
long been considered that the preferred conformation
of a ligand in solution is the relevant one for binding44

1302 J ournal of Medicinal Chemistry, 2003, Vol. 46, No. 8
Greenidge et al.
Ta ble 5. Accurate Mass Measurements for Compounds 21-24
in calculated binding modes which are more consistent
with experimental data and thus provide reliable start-
ing points for lead optimization during a medicinal
chemistry program.
molecular
formula
theoretical
mass
measured
mass
error
(ppm)
compound
21
22
23
24
C
C
C
C
25H29N3O5S
484.19056
484.19056
610.16676
470.17437
538.14614
0.1
2.3
1.23
1.65
30H31N3O7S2 610.16815
24H27N3O5S 470.17495
27H27N3O5S2 538.14703
Exp er im en ta l Section
Sta n d a r d Meth od of Deter m in a tion of IC50 for Test
Com p ou n d s. Measurements were obtained on a Molecular
Devices Corp. Thermomax plate reader using Softmax soft-
ware. The chromogenic substrate S-2238 (H-D-Phe-Pip-Arg-
pNA, for Thr) was obtained from Quadratech. The buffer was
0.1 M sodium phosphate; 0.2 M sodium chloride; 0.5% PEG6000
and 0.02% sodium azide. Human R-thrombin was obtained
from Haematologics inc.
A solution of human R-thrombin (50 mL of a dilution of
1/10000 of stock at 1 mg/mL) was added after 2 min to a series
of 10-fold dilutions of each inhibitor (100 µL of dilution in
buffer of 1/20 from stock in DMSO) in the presence of a fixed
concentration of the appropriate chromogenic substrate (50 µL
of a dilution of 1/100 of 2.4 mg/mL stock) at 37 °C, to determine
the inhibitor concentration needed to give ∼50% inhibition. A
series of concentrations of inhibitor either side of this ap-
proximate 50% value (of at least five different inhibitor
concentrations, each point measured in duplicate) were plotted
graphically to determine the exact inhibitor concentration
needed to give 50% inhibition (IC50).
Ch em istr y. Gen er a l Meth od s. 1H and 13C NMR were
obtained on a Bruker Avance DPX400 instrument at 400.13
and 100.16 MHz, respectively, relative to TMS internal
standard. Chemical shifts are given in ppm and peak multi-
plicities are designated as follows: s, singlet; d, doublet; t,
triplet; m, multiplet. Coupling constants are given in Hz. Mass
spectra were determined on a Finnigan SSQ 710C spectrom-
eter. Elemental analyses were performed by the Elemental
Analysis Department, Department of Biological and Applied
Sciences of the University of North London. Results were
within 0.4% of theoretical values, unless otherwise stated. All
compounds exhibited NMR and MS analyses consistent with
the proposed structures. All solvents and reagents were
purchased from Aldrich except dimethylformamide (from
Rathburn), dichloromethane, methanol and ethanol (Romil),
potassium hydroxide and potassium carbonate (BDH) and
hydrogen cylinders (Boc gases). The abbreviations used are
as follows: DCM: dichloromethane, DMF: N,N-dimethylfor-
mamide, HCl: hydrochloric acid, MeCN: acetonitrile, NaOH:
sodium hydroxide, NMM: N-methylmorpholine, TFA: trifluo-
roacetic acid, THF: tetrahydrofuran, TLC: thin-layer chro-
matography.
ethyl acetate/hexane), 0.3 g (35%) of the pure product was
collected; 1H NMR δ 1.26 (m, 2H, CH2), 1.52 (s, 9H, t-Bu), 3.70
(br s, 2H), 4.12 (q, 2H, CH2), 6.23 (m, 2H), 6.8 (s, 1H), 6.98 (m,
1H), 7.34 (d, J ) 5.3 Hz, 2H), 8.44 (d, J ) 5.3 Hz, 2H); 13C
NMR δ 28.2, 60.4, 81.8, 102.3, 105.8, 107.5, 112.4, 130.3, 145.9,
149.9, 150.1, 151.8, 157.3; ES-MS (m/e) 330.0 (M+H).
3,3-Dit osyla m in o[2′-(ter t-b u t oxyca r b on ylp yr id in -4-
yla m in o)eth yl]p h en yl eth er (20) a n d 3-tosyla m in o[2′-
(ter t-b u t oxyca r b on ylp yr id in -4-yla m in o)et h yl]p h en yl-
eth er (21). p-Toluenesulfonyl chloride (0.25 g, 1.5 equiv) in
DCM was added dropwise to a stirred solution of 19 (0.25 g,
0.9 mmol) and triethylamine (0.14 mL) in dry dichloromethane
(5 mL). The mixture was stirred at room temperature and the
reaction was followed by TLC. DCM was added and the organic
phase was washed with aqueous HCl (0.1 M), water and brine,
and dried over magnesium sulfate. After filtration of the
desiccant and evaporation of the solvent, a yellow oil (0.39 g)
was obtained. The mixture was separated by flash chroma-
tography (silica, ethyl acetate/hexane); 20: 0.262 g (55%); 1H
NMR δ 1.49 (s, 9H, t- Bu), 2.40 (s, 6H, 2 CH3), 4.11 (d, 4H, 2
CH2), 6.49 (m, 1H), 6.50 (m, 1H), 7.15 (m, 1H), 7.25 (m, 4H),
7.72 (m, 6H), 8.54 (d, J ) 6.8 Hz, 2H); 13C NMR δ 22.1, 28.5,
49.2, 66.4, 84.9, 103.2, 117.3, 117.4, 119.1, 125.1, 129.0, 130.0,
130.3, 135.9, 136.8, 145.6, 152.5, 158.5, 207.4; ES-MS (m/z)
638.3 (M+H).
21: 0.213 g (29%); 1H NMR δ 1.26 (m, 2H, CH2), 1.49 (s,
3H, CH3), 1.53 (s, 9H, t-Bu), 4.1 (m, 2H, CH2), 6.54 (dd, J )
1.6 and 8.7 Hz, 1H), 6.66 (d, J ) 1.6 Hz, 1H), 7.08 (d, J ) 8.7
Hz, 1H), 7.20 (d, J ) 8.2 Hz, 2H), 7.31 (d, J ) 8.5 Hz, 1H),
7.33 (d, J ) 6.4 Hz, 2H), 7.65 (d, J ) 8.3 Hz, 2H), 8.44 (d, J )
6.2 Hz, 2H); 13C NMR δ 21.5, 28.2, 48.7, 65.9, 81.6, 107.3,
110.7, 112.4, 113.5, 119.7, 127.2, 129.6, 130.1, 136.4, 138.3,
143.7, 145.8, 150.1, 152.0, 153.3, 158.9; ES-MS (m/z) 484.2
(M+H).
3,3-Diben zylsu lfon yla m in o[2′-(ter t-bu toxyca r bon ylp y-
r id in -4-yla m in o)eth yl] p h en yl eth er (22) a n d 3-ben zyl-
su lfon yla m in o[2′-(ter t-b u t oxyca r b on ylp yr id in -4-yla m i-
n o) eth yl] p h en yl eth er (23). These were synthesized using
the same method as for 20 and 21. 22: 0.0037 g (14%); 1H
NMR δ 0.80 (m, 2H, CH2), 1.20 (s, 9H, t-Bu), 4.15 (m, 2H, CH2),
6.40 (s, 1H), 6.48 (d, J ) 7.3 Hz, 1H), 6.78 (dd, J ) 2 and 7.3
Hz, 1H), 7.20 (m, 1H), 7.50 (m, 4H), 7.62 (m, 2H), 7.81 (d, J )
6.1 Hz, 2H), 7.91 (d, J ) 7 Hz, 4H), 8.57 (d, J ) 6.1 Hz, 2H);
ES-MS (m/e) 610.3.
4-[N-(ter t-bu tyloxycar bon yl)am in o]pyr idin e (18). 4-Ami-
nopyridine (10 g, 0.106 mol) was added slowly, in portions, to
a solution of di-tert-butyl dicarbonate (25.5 g, 1 equiv) in THF,
under argon. The reaction mixture was stirred at room
temperature for 90 min, the solvent evaporated under reduced
pressure and the product crystallized with diethyl ether to give
1
23: 0.0046 g (17%); H NMR δ 0.81 (m, 2H, CH2), 1.23 (s,
9H, t-Bu), 4.15 (m, 2H, CH2), 6.2 (dd, J ) 1.8 and 7.1 Hz, 1H),
6.33 (d, J ) 1.8 Hz, 1H), 6.48 (dd, J ) 1.8 and 7.1 Hz, 1H),
6.95 (m, 1H), 7.45 (m, 2H), 7.58 (m, 1H), 7.68 (d, 2H, J ) 5.4
Hz), 7.79 (d, 2H, J ) 7.1 Hz), 8.56 (d, 2H, J ) 5.4 Hz); ES-MS
(m/e) 470.3.
1
a white solid (13 g, 65%, mp ) 144-145 °C); H NMR δ 1.53
(s, 9H, t-Bu), 7.30 (d, J ) 6 Hz, 2H), 8.43 (d, J ) 6 Hz, 2H);
13C NMR δ 28.2, 81.8, 112.3, 150.3, 151.8, 157.3.
3-Am in o[2-(ter t-bu toxycar bon ylpyr idin -4-yl-am in o)eth -
oxy]p h en ol (19). Sodium hydride (0.11 g) was added to a
stirred solution of 18 (0.5 g, 2.5 mmol) and ethylene glycol di-
p-tosylate (0.93 g, 2.5 mmol) in dry DMF under argon. The
reaction mixture was stirred for a further 7 h. 3-Aminophenol
(0.27 g, 2.5 mmol) was added, followed by sodium hydride (0.08
g) and the mixture was stirred for a further 72 h. After cooling
to room temperature, water (10-15 mL) was added and the
mixture was evaporated under reduced pressure. The residue
was partitioned between ethyl acetate and HCl (2 M). The
acidic layer was exhaustively extracted with ethyl acetate,
basified with NaOH pellets and extracted again with ethyl
acetate. The organic phase was washed with water and brine,
and dried over magnesium sulfate. After filtration of the
desiccant and evaporation of the solvent a yellow oil was
obtained (Rf 0.2, ethyl acetate 100%). After purification (silica,
3,3-Dit osyla m in o[2′-(p yr id in -4-yla m in o)et h yl]p h en yl
eth er (24). A mixture of 1% anisole in TFA (2 mL) was added
slowly to a stirring solution of 20 (0.1 g, 0.2 mmol) in DCM (2
mL). After 45 min at room temperature, the solvent was
removed under reduced pressure and the residue obtained was
dissolved in DCM, washed twice with water and brine, and
dried over magnesium sulfate. After filtration of the desiccant
and evaporation of the solvent, an oil was obtained which was
purified by RP-HPLC (MeCN, H2O, TFA). 24 (0.005 g) was
1
recovered. H NMR δ 1.37 (m, 2H, CH2), 2.43 (s, 6H, 2 CH3),
4.12 (m, 2H, CH2), 6.52 (m, 2H), 7.19 (m, 2H), 7.27 (m, 4H),
7.78 (m, 6H), 8.53 (d, J ) 7.1 Hz, 2H); ES-MS (m/z) 538.2
(M+H) (Table 5).
3-Tosylam in o[2′-(pyr idin -4-ylam in o)eth yl]ph en yl eth er
(15). This was synthesized from 21 following the same
procedure as for 24 to give 15. 0.075 g (95%).18

Ligand Conformations with Protein Active Site Topology
J ournal of Medicinal Chemistry, 2003, Vol. 46, No. 8 1303
Ta ble 6. Data Obtained from Crystallographic Analysis of
Compound 24 with Human R-Thrombin
2-(3-Nitr op h en oxy)-a cetic a cid eth yl ester . meta-Nitro-
phenol (5 g, 8.6 mmol) and potassium carbonate (4.97 g, 1
equiv) were stirred in DMF at room temperature while ethyl
chloroacetate (3.8 mL, 1 equiv) was added. The mixture was
heated under reflux at 70 °C for 5 h, and 130 °C for 48 h. After
cooling to room temperature, water was added and the solution
was concentrated and taken up in ethyl acetate. The organic
phase was washed with NaOH (0.2 M), water, and brine and
dried over magnesium sulfate. After filtering the desiccant,
the solvent was evaporated under vacuum to give an oil (4.5
space group
unit cell
C2
A ) 69.747 Å, b ) 71.417 Å,
C ) 72.528 Å, â ) 100.363°
resolution range Å
no. of unique reflections
overall completeness
R-merge (%)
R-factor (R-working) (%)
R-factor (free) (%)
20-1.8 (1.85-1.80)a
32519
99.1% [98.0]a (20-1.80 Å)
5.5 (32)
18.19
22.21
0.012
1.9
1
g, 56%); H NMR δ 1.31 (t, 3H, CH3), 4.28 (q, 2H, CH2), 4.72
rms dev. bonds (Å)
rms dev angles (deg)
(s, 2H, CH2), 7.26 (d, J ) 2.5 Hz, 1H), 7.46 (m, 1H), 7.72 (s,
1H), 7.87 (d, J ) 1.9 Hz, 1H); 13C NMR δ 14.2, 61.7, 65.5, 109.1,
116.8, 121.8, 130.2, 142.2, 162.6, 167.7; FAB-MS (m/z) 226
(M+H). Anal. (C10H11NO5) C, H, N.
a
Figures in brackets refer to the highest resolution bin data.
2-(3-Am in op h en oxy)-a cetic Acid Eth yl Ester . A balloon
of hydrogen was placed over a degassed stirred mixture of 2-(3-
nitrophenoxy)-acetic acid ethyl ester (2 g, 8.9 mmol) and
platinum (IV) oxide (0.10 g, 5% w/w) in ethanol (60 mL). The
mixture was stirred at room-temperature overnight, filtered
over Celite and concentrated down to give a colorless oil (1.66
g, 96%); 1H NMR δ 1.25 (t, 3Η, CΗ3), 3.72 (broad s, 2H, NH2),
4.26 (q, 2H, CH2), 4.57 (s, 2H, CH2), 6.26-6.34 (m, 3H), 7.05
(m, 1H); 13C NMR δ 14.2, 61.3, 65.4, 102.0, 104.2, 108.9, 130.2,
was added quickly to 27 (0.44 g, 1.4 mmol) dissolved in DMF
(1 mL), and the solution heated under reflux at 70 °C. After 5
min, a solution of 4-aminopyridine (0.16 g, 1 equiv) with NMM
(0.18 mL) in DMF (0.5 mL) was added slowly to the reaction
mixture, which was stirred for an additional 12 h at 70 °C.
After cooling to room temperature, water was added and the
aqueous phase was extracted with ethyl acetate. The organic
phase was washed with NaOH (0.1 M), water and brine and
dried over magnesium sulfate. After filtration of the desiccant
and evaporation of the solvent, an orange oil (0.29 g, 53%) was
obtained; 1H NMR δ 3.18 (s, 3H, CH3), 4.59 (s, 2H, CH2), 6.66
(dd, J ) 1.7 and 7.8 Hz, 1H), 6.88 (dd, J ) 2.4 and 8.2 Hz,
1H), 6.94 (s, 1H), 7.25 (m, 1H), 7.46 (m, 2H), 7.55-7.61 (m,
147.9, 159.0, 169.1; FAB-MS (m/z) 196 (M+H). Anal. (C10H13
-
NO3) C, H, N.
2-[3-(P h en ylsu lfon yl)a m in op h en oxy]-a cetic Acid Eth -
yl Ester (25). Benzene sulfonyl chloride (1.51 g, 1.09 mL, 8.5
mmol) was added slowly, at 0 °C, to a stirred solution of 2-(3-
aminophenoxy)-acetic acid ethyl ester (1.66 g, 8.5 mmol) and
pyridine (0.69 mL, 1 equiv) in DCM (30 mL). The mixture was
left to stir at room-temperature overnight, diluted with DCM
and the organic phase was washed with HCl (0.2 M), water
and brine and dried over magnesium sulfate. After filtering
the desiccant and evaporation of the solvent, an orange solid
(2.52 g, 88%) was obtained; 1H NMR δ 1.28 (t, 3Η, CH3), 4.26
(q, 2H, CH2), 4.55 (s, 2H, CH2), 6.61 (dd, J ) 2.3 and 8 Hz,
1H), 6.68 (dd, J ) 1.3 and 8 Hz, 1H), 6.74 (d, J ) 2 Hz, 1H),
7.09 (m, 1H), 7.41 (m, 2H), 7.51 (m, 1H), 7.79 (d, J ) 8.6 Hz,
2H); 13C NMR δ 14.1, 61.5, 65.3, 107.6, 111.4, 114.3, 127.2,
129.1, 130.2, 133.1, 137.9, 138.9, 158.4, 168.8; FAB-MS (m/z)
336 (M+H), 358 (M+Na). Anal. (C16H17NO5S) C, H, N.
5H), 8.45 (broad s, 1H), 8.56 (dd, J ) 1.5 and 4.2 Hz, 2H); 13
C
NMR δ 38.1, 67.6, 113.3, 113.9, 114.6, 119.7, 127.9, 128.9,
130.0, 133.0, 136.3, 143.1, 143.8, 150.1, 156.9, 166.6; ESI-MS
(m/e) 399.6 (M+2H).
N-Meth yl-3-[2-(p yr id in -4-yla m in o)eth oxy-p h en yl]ben -
zen esu lfon a m id e (7). A solution of 28 (0.1 g, 0.25 mmol) in
dry THF (10 mL) was added slowly to a refluxing solution of
lithium aluminum hydride (0.03 g, 0.8 mmol) in THF. The
reaction mixture was stirred overnight at 80 °C. After cooling
to room temperature, a few drops of water were added and
the solvent evaporated under reduced pressure. The residue
was taken up in ethyl acetate and washed with NaOH (0.1
M), water, brine and dried over magnesium sulfate. After
filtering the desiccant and evaporating the solvent, an oil (0.05
1
2-[3-(P h en ylsu lfon yl)-N-m et h yla m in o-p h en oxy]-a ce-
tic Acid Eth yl Ester (26). Methyl iodide (0.42 g, 0.19 mL)
was added to a mixture of 25 (1 g, 3 mmol) and potassium
carbonate (0.41 g, 1 equiv) in DMF (20 mL) and the mixture
was stirred for 1 h. The mixture was concentrated down and
the residue was dissolved in ethyl acetate, washed with water
and brine, and dried with magnesium sulfate. After filtering
the desiccant and evaporating the solvents, a yellow oil (1.04
g, 86%) was obtained; 1H NMR δ 1.28 (t, 3H, CH3), 3.15 (s,
3H, CH3), 4.27 (q, 2H, CH2), 4.58 (s, 2H, CH2), 6.68 (dd, J )
1.9 and 8 Hz, 1H), 6.73 (d, J ) 0.6 Hz, 1H), 6.82 (dd, J ) 2.5
g, 51%) was obtained; H NMR δ 2.82 (m, 2H, CH2), 3.14 (s,
3H, CH3), 3.54 (m, 2H, CH2), 4.65 (br s, NH, 1H), 6.26 (dd, J
) 1.5 and 8.0 Hz, 1H), 6.50 (d, J ) 4.0 Hz, 2H), 6.82 (d, J )
6.1 Hz, 1H), 7.09 (m, 2H), 7.19 (m, 1H), 7.45 (m, 2H), 7.56 (d,
J ) 6.1 Hz, 2H), 8.21 (d, J ) 4.0 Hz, 2H); 13C NMR δ 38.1,
42.0, 65.8, 102.6, 106.4, 107.8, 113.6, 118.5, 127.8, 128.8, 129.6,
130.1, 132.9, 149.9, 150.9, 153.3; ESI-MS (m/e) 385.5 (M+2H).
An a lysis of th e In ter a ction of Com p ou n d 24 w ith
Th r om bin by Cr ysta llogr a p h y. Human R-thrombin was
obtained from Haematologic Technologies Inc., and N-Ac-
Hirugen55-64 was obtained from Bachem.
and 8.3 Hz, 1H), 7.19 (m, 1H), 7.45 (m, 2H), 7.56 (m, 3H); 13
C
Crystals of the thrombin-hirugen-compound 24 complex
were grown by vapor diffusion at 4 °C using the hanging drop
method with 25% (w/v) PEG 8000, 0.05 M ammonium phos-
phate (pH 7.20) and 0.05 M sodium azide. Crystal, typically
of dimensions 0.25 × 0.15 × 0.1 mm3, appeared in 4-6 weeks.
Crystals were flash cooled by soaking in the crystallization
buffer made up with 25% (v/v) PEG 400.
Data sets were collected using an R-axis II image plate
mounted on a Rigaku RU200 anode generator to a maximum
Bragg spacing of 1.8 Å, and were processed using DENZO46
and Scalepack.46
Initial molecular replacement, based upon the PDB set
1H8D, was determined with AMORE.47 Initial coordinates
were obtained with CCP4I and the complex was refined using
Refmac,46 with water molecules added using ARP/WARP.48
Refinement converged to an R-factor of 18% (Rfree ) 22%, using
5% reflections) (Table 6).
NMR δ 14.2, 38.1, 61.5, 65.4, 113.3, 113.7, 119.3, 127.8, 128.6,
129.5, 132.8, 136.3, 142.7, 158.0, 168.6; FAB-MS (m/z) 350
(M+H).
2-[3-(P h en ylsu lfon yl)-N-m et h yla m in o-p h en oxy]-a ce-
tic Acid (27). Powdered potassium hydroxide (0.29 g, 0.005
mol) was added to 26 (0.89 g, 2.6 mmol) and then dissolved in
methanol (20 mL). The resulting mixture was stirred for 24
h. The solvent was evaporated under reduced pressure and
the residue was dissolved in water and acidified to pH 4-5
with citric acid (2 M). The acidic phase was extracted with
ethyl acetate, which was washed with water and brine over
magnesium sulfate. The desiccant was filtered and the solvent
1
was evaporated to give a colorless oil (0.44 g, 54%); H NMR
δ 3.16 (s, 3H, CH3), 4.64 (s, 2H, CH2), 5.05 (broad s, 1H, OH),
6.68 (d, J ) 7.9 Hz, 1H), 6.76 (s, 1H), 6.83 (d, J ) 8.3 Hz, 1H),
7.20 (m, 1H), 7.45 (m, 2H), 7.56 (m, 3H); 13C NMR δ 38.1, 64.9,
113.5, 113.7, 119.6, 127.8, 128.8, 129.7, 132.9, 132.9, 136.2,
142.8, 157.6; FAB-MS (m/z) 322 (M+H).
Ack n ow led gm en t. We would like to thank Andrew
Burd for technical support and the referees for their
insightful comments and suggestions.
2-([3-(P h en ylsu lfon yl)-N-m eth yla m in o-p h en oxy]-4-p y-
r id yl) Aceta m id e (28). Thionyl chloride (0.12 mL, 1.7 mmol)

1304 J ournal of Medicinal Chemistry, 2003, Vol. 46, No. 8
Greenidge et al.
(20) Vieth, M.; Hirst, J . D.; Brooks, C. L., III Do Active Site
Conformations of Small Ligands Correspond to Low Free Energy
Solution Structures ? J . Comput. -Aided Mol. Design 1998, 12,
563-572.
(21) Gund P. In: Progress in Molecular and Subcellular Biology,
Hahn F. E., Ed., Springer-Verlag: New York, 1977; Vol. 5, pp
117-143.
(22) Bostro¨m, J .; Norrby, P.-O.; Liljefors, T. Conformational Energy
Penalties of Protein-Bound Ligands. J . Comput.-Aided Mol. Des.
1998, 12, 383-396.
(23) (a) Wielert-Badt S.; Lin, J . T.; Lorenz, M.; Fritz, S.; Kinne, R.
K. Probing the Conformation of the Sugar Transport Inhibitor
Phlorizin by 2D-NMR, Molecular Dynamics Studies, and Phar-
macophore Analysis. J . Med. Chem. 2000, 43, 1692-1698. (b)
Koerber S. C.; Rizo, J .; Struthers R. S.; Rivier, J . E. Consensus
Bioactive Confromation of Cyclic GnRH Anatagonists Defined
by NMR and Molecular Modeling. J . Med. Chem. 2000, 43, 819-
828.
(24) Greenidge, P. A.; Weiser J . A Comparison of Methods for
Pharmacophore Generation With the Catalyst Software and
Their Use for 3D-QSAR: Application to a Set of 4-Aminopyridine
Thrombin Inhibitors. Mini-Rev. Med. Chem. 2001, 1, 79-
87.
(25) Shenkin, P. S.; McDonald, D. Q. Cluster Analysis of Molecular
Conformations, J . Comput. Chem. 1994, 15, 899-916. XCluster
1.7 is part of the MacroModel package, Schro¨dinger, Portland,
OR.
(26) Tucker, T. J .; Brady, S. F.; Lumma, W. C.; Lewis, S. D.; Gardell,
S. J .; Naylor-Olsen, A. L.; Yan, Y.; Sisko, J . T.; Stauffer, K. J .;
Lucaas, 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.; 1 Vacca, J . 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.
Refer en ces
(1) Steinmetzer, T.; Hauptmann, J .; Sturzebecher, J . Advances In
the Development of Thrombin Inhibitors. Expert Opin. Invest.
Drugs 2001, 10, 845-864.
(2) Fenton, J . W., II; Ofosu, F. A.; Moon, D. G.; Maragonore, J . M.
Thrombin Structure and Function: Why Thrombin Is the
Primary Target for Antithrombotics. Blood Coagul. Fibrin. 1990,
2, 69-75.
(3) Hauel, N. H.; Nar, H.; Priepke, H.; Ries, U.; Stassen, J .-M.;
Wienen, W.; Structure-Based Design of Novel Potent Nonpeptide
Thrombin Inhibitors. J . Med. Chem. 2002, 45, 1757-1766.
(4) Gustafsson, D.; Nystrom, J .; Carlsson, S.; Bredberg, U.; Eriks-
son, U.; Gyzander, E.; Elg, M.; Antonsson, T.; Hoffmann, K.;
Ungell, A.; Sorensen, H.; Nagard, S.; Abrahamsson, A. The
Direct Thrombin Inhibitor Melagatran and Its Oral Prodrug H
375/95: Intenstinal Absorption Properties, Biochemical and
Pharmacodynamic Effects. Thromb. Res. 2001, 101, 171-
181.
(5) Li, C. Q.; Vindigni, A.; Sadler, J . E.; Wardell, M. R. Platelet
Glycoprotein Ib alpha Binds to Thrombin Anion-Binding Exosite
II Inducing Allosteric Changes in the Activity of Thrombin. J .
Biol. Chem. 2001, 276, 6161-6168.
(6) De Candia, E.; Hall, S. W.; Rutella, S.; Landolfi, R.; Andrews,
R. K.; Cristofaro, R. Binding of Thrombin to Glycoprotein 1b
Accelerates the Hydrolysis of PAR-1n on Intact Platelets. J . Biol.
Chem. 2001, 276, 4692-4698.
(7) Davie, E. W.; Fujikawa, K.; Kisiel, W. The Coagulation cas-
cade: Initiation, Maintenance and Regulation. Biochemistry
1991, 30, 10363-10369.
(8) Guinto, E. R.; Vindigni, A.; Ayala, Y. M.; Dang, Q. D.; Di Cera,
E. Identification of Residues Linked to The Slow to Fast
Transistion of Thrombin. Proc. Natl. Acad. Sci. U.S.A. 1995, 92,
11185-11189.
(9) Goodwin, C. A.; Deadman, J . J .; Le Bonniec, B. F.; Elgendy, S.;
Kakkar, V. V.; Scully, M. F. Heparin Enhances the Catalytic
Activity of des-ETW-Thrombin, Biochem. J . 1996, 315, 77-
83.
(27) von der Saal, W.; Kucznierz, R.; Leinert, H.; Engh, R. Derivatives
of 4-Amino-Pyridine As Selective Thrombin Inhibitors. Bioorg.
Med. Chem. Lett. 1997, 7, 1283-1288.
(28) Deadman, J . J .; Elgendy, S.; Goodwin, C. A.; Green, D.; Baban,
J . A.; Patel, G.; Skordalakes, E.; Chino, N.; Claeson, G.; Kakkar,
V. V.; Scully, M. F. Characterization of a Class of Peptide
Boronates With Neutral P1 Side Chains As Highly Selective
Inhibitors of Thrombin. J . Med. Chem. 1995, 38, 1511-
1522.
(29) Insight II User Guide, September 1997. San Diego: MSI,
1997.
(30) Pauling L. The Nature of the Chemical Bond; Cornell University
Press: New York, 1960.
(10) Sanschagrin, P. C.; Kuhn, L. A. Cluster analysis of consensus
water sites in thrombin and trypsin shows conservation between
serine proteases and contribution to ligand specificity. Protein
Sci. 1998, 7, 2054-2064.
(11) Matter, H.; Defossa, E.; Heinelt, U.; Blohm, P.-M.; Schneider,
D.; Mu¨ller, A.; Herok, S.; Schreuder, H.; Liesum, A.; Brachvogel,
V.; Lo¨nze, P.; Walser, A.; Al-Obeidi, F.; Wildgoose, P. Design and
Quantitative Structure-Activity Relationship of 3-Amidinoben-
zyl-1H-indole-2-carboxamides as Potent Nonchiral, and Selective
Inhibitors of Blood Coagulation Factor Xa. J . Med. Chem. 2002,
45, 2749-2769.
(31) Halgren, T. A. MMFF VI. MMFF94s Option for Energy Mini-
mization Studies. J . Comput Chem. 1999, 20, 720-729.
(32) Still, W. C.; Tempczyk, A.; Hawley, R. C.; Hendrickson, T.
Semianalytical Treatment of Solvation for Molecular Mechanics
and Dynamics. J . Am. Chem. Soc. 1990, 112, 6127-6129.
(33) (a) Weiser, J .; Shenkin, P. S.; Still, W. C. Approximate Atomic
Surfaces from Linear Combinations of Pairwise Overlaps (LCPO).
J . Comput. Chem. 1999, 20, 217-230. (b) Weiser, J .; Weiser, A.
A.; Shenkin, P. S.; Still, W. C. Neighbor-List Reduction: Opti-
mization for Computation of Molecular van der Waals and
Solvent-Accessible Surface Areas. J . Comput. Chem. 1998, 19,
797-808. (c) Weiser, J .; Shenkin, P. S.; Still, W. C. Fast,
Approximate Algorithm for Detection of Solvent-Inaccessible
Atoms. J . Comput. Chem. 1999, 20, 586-596.
(34) Ponder, J . W.; Richards, F. M. An Efficient Newton-like Method
for Molecular Mechanics Energy Minimization of Large Mol-
ecules. J . Comput. Chem. 1987, 8, 1016-1024.
(35) Chang, G.; Guida, W. C.; Still, W. C. An Internal Coordinate
Monte Carlo Method for Searching Conformational Space. J . Am.
Chem. Soc. 1989, 111, 4379-4386.
(36) Kolossvary, I.; Guida, W. C. Low Mode Search. An Efficient,
Automated Computational Method for Conformational Analy-
sis: Application to Cyclic and Acyclic Alkanes and Cyclic
Peptides. J . Am. Chem. Soc. 1996, 118, 5011-5019.
(37) (a) Smellie, A.; Kahn, S. D.; Teig, S. L. Analysis of Conforma-
tional Coverage 1. Validation and Estimation of Coverage. J .
Chem. Inf. Comput. Sci. 1995, 35, 285-294. (b) Smellie, A.; Teig,
S. L.; Towbin, P. Poling: Promoting Conformational Variation.
J . Comput. Chem. 1995, 16, 171-187.
(38) Murray, C. W.; Baxter, C. A.; Frenkel, A. D.; The Sensitivity of
the Results of Molecular Docking to Induced Fit Effects: Ap-
plication to Thrombin, Thermolysin and Neuraminidase. J .
Comput.-Aided Mol. Des. 1999, 13, 547-562.
(12) J ones-Hertzog, D. K.; J orgensen, W. L. Binding Affinities for
Sulfonamide Inhibitors with Human Thrombin Using Monte
Carlo Simulations with a Linear Response Methodol. J . Med.
Chem. 1997, 40, 1539-1549.
(13) Bursi, R.; Grootenhuis, P. D. Comparative Molecular Field
Analysis and Energy Interaction Studies of Thrombin-Inhibitor
Complexes. J . Comput. -Aided Mol. Design 1999, 13, 221-
232.
(14) Cramer, R. D., III; Patterson, D. E.; Bunce, J . D. Comparative
Molecular Field Analysis (CoMFA). 1. Effect of Shape on Binding
of Steroids to Carrier Proteins. J . Am. Chem. Soc. 1988, 110,
5959-5967.
(15) (a) Greenidge, P. A.; Carlsson, B.; Bladh, L.-G.; Gillner, M.
Pharmacophores Incorporating Numerous Excluded Volumes
Defined by X-ray Crystallographic Structure in Three-Dimen-
sional Database Searching: Application to the Thyroid Hormone
Receptor. J . Med. Chem. 1998, 41, 2503-2512; (b) Gillner, M.;
Greenidge, P. A. In Pharmacophore Perception, Development and
Use in Drug Design; International University Line: La J olla,
1999; pp 373-384.
(16) Catalyst 4.0 Tutorials, August 1998, San Diego: Molecular
Simulations Inc.
(17) Mohamadi, F.; Richards, N. G. J .; Guida, W. C.; Liskamp, R.;
Lipton, M.; Caufield, C.; Chang, G.; Hendrickson, T.; Still, W.
C. MacroModel-An Integrated Software System for Modeling
Organic and Bioorganic Molecules Using Molecular Mechanics.
J . Comput. Chem. 1990, 11, 440-467. MacroModel 7.2, Schro¨-
dinger, Portland, OR.
(18) Engh, R. A.; Brandstetter, H.; Sucher, G.; Eichinger, A.; Bau-
mann, U.; Bode, W.; Huber, R.; Poll, T.; Rudolph, R.; von der
Saal, W. Enzyme Flexibility, Solvent and “Weak” Interactions
Characterize Thrombin-Ligand Interactions: Implications For
Drug Design. Structure 1996, 4, 1353-1362.
(39) Coombs, G. S.; Rao, M. S.; Olson, A. J .; Dawson, P. E.; Madison,
E. L. Revisiting Catalysis by Chymotrypsin Family Serine
Proteases Using Peptide Substrates and Inhibitors With Un-
natural Main Chains. J . Biol. Chem. 1999, 274, 24074-
24079.
(19) Schwarzl, S. J .; Tschopp, B. T.; Smith, J . C.; Fischer, S. Can
the Calculation of Ligand Binding Free Energies Be Improved
with Continuum Solvent Electrostatics and an Ideal-Gas En-
tropy Correction? J . Comput. Chem. 2002, 23, 1145-
1149.

Ligand Conformations with Protein Active Site Topology
J ournal of Medicinal Chemistry, 2003, Vol. 46, No. 8 1305
(40) Weber, I. R.; Neidlein, R.; von der Saal, W.; Grams, F.; Leinhert,
H.; Strein, K.; Engh, R. A.; Kucznierz, R. Diarylsulfonamides
As Selective, Non-Peptidic Thrombin Inhibitors. Bioorg. Med.
Chem. Lett. 1998, 8, 1613-1618.
(41) http://www.msi.com/support/catalyst/hypogen.html
(42) Golbraikh, A.; Tropsha A. Beware of q2. J . Mol. Graphics and
Modelling 2002, 20, 269-276.
(43) (a) Hilpert, K.; Ackermann, J .; Banner, D. W.; Gast, A.; Klaus,
G.; Schmid, G.; Tschopp, T. B.; van de Waterbeemd, H. The
Development of Potent and Highly Selective Thrombin Inhibi-
tors. Eur. J . Med. Chem. Suppl. 1995, 30, 131S-138S. (b) Banner,
D. W.; Hadvary, P. Crystallographic Analysis at 3.0-A Resolution
of the Binding to Human Thrombin of Four Active Site-Directed
Inhibitors. J . Biol. Chem. 1991, 266, 20085-20093.
(44) Roberts, G. C. K., Molecular and Quantum Pharmacology, Eds.
E. D. Bergmann and B. Pullman, D. Reidel Publishing Co.,
Dordecht, Holland, 1974; pp 88-90.
(45) Lim, M. S.; J ohnston, E. R.; Kettner, C. A. The solution
conformation of D-Phe-Pro-containing peptides - Implications
on the activity of Ac-(D)Phe-Pro-OH, a potent thrombin inhibi-
tor. J . Med. Chem. 1993, 36, 1831-1838.
(46) Otwinowski, Z. In Data Collection and Processing, 56-61 (SERC
Daresbury Laboratory, 1993).
(47) Baileys, S.. The CCP4 suite: programs for protein crystal-
lography. Acta Crystallogr. 1994, D50, 760-763.
(48) Lamzin, V. S.; Wilson, K. S. Automated refinement of protein
models. Acta Crystallogr. 1993, D49, 129-147.
J M021028J