PRO_SELECT: Combining structure-based drug design and array-based chemistry for rapid lead discovery. 2. The development of a series of highly potent and selective factor Xa inhibitors

Article abstract of DOI:10.1021/jm010944e

In silico screening of combinatorial libraries prior to synthesis promises to be a valuable aid to lead discovery. PRO_SELECT, a tool for the virtual screening of libraries for fit to a protein active site, has been used to find novel leads against the se

Full text of DOI:10.1021/jm010944e

J . Med. Chem. 2002, 45, 1221-1232
1221
P RO_SELECT: Com bin in g Str u ctu r e-Ba sed Dr u g Design a n d Ar r a y-Ba sed
Ch em istr y for Ra p id Lea d Discover y. 2. Th e Develop m en t of a Ser ies of High ly
P oten t a n d Selective F a ctor Xa In h ibitor s
J ohn W. Liebeschuetz,*,† Stuart D. J ones,† Phillip J . Morgan,† Chris W. Murray,† Andrew D. Rimmer,†
J onathan M. E. Roscoe,† Bohdan Waszkowycz,† Pauline M. Welsh,† William A. Wylie,† Stephen C. Young,†
Harry Martin,† J acqui Mahler,† Leo Brady,‡ and Kay Wilkinson‡
Protherics Molecular Design, Beechfield House, Lyme Green Business Park, Macclesfield SK11 0J L, U.K., and Department of
Biochemistry, University of Bristol, Bristol BS8 1TD, U.K.
Received J une 6, 2001
In silico screening of combinatorial libraries prior to synthesis promises to be a valuable aid to
lead discovery. PRO_SELECT, a tool for the virtual screening of libraries for fit to a protein
active site, has been used to find novel leads against the serine protease factor Xa. A small
seed template was built upon using three iterations of library design, virtual screening,
synthesis, and biological testing. Highly potent molecules with selectivity for factor Xa over
other serine proteases were rapidly obtained.
In tr od u ction
Serine proteases represent a class of enzymes of great
therapeutic importance. Members of the class which
have been targeted for drug design include tryptase and
urokinase and, in the blood coagulation cascade, throm-
bin, factor VIIa, and factor Xa. Factor Xa lies at the
junction of the intrinsic and extrinsic pathways of the
coagulation cascade. It is the active enzyme present in
the prothrombinase complex, which converts prothrom-
bin into thrombin. Thrombin is the final enzymatic
product of the blood coagulation cascade and is respon-
sible for the conversion of fibrinogen into fibrin. Much
effort has been spent targeting thrombin, in particular,
F igu r e 1. Bis-amidine factor Xa inhibitors DX-9065a and YM-
60828.
and, more recently, factor Xa, with the aim of designing
antithrombotic drugs which are orally available and
which show a reduced potential for bleeding as a side
effect.1 Current therapies include the heparins, which
are not orally available, and the coumarins, which have
a narrow therapeutic window with regard to bleeding.
Factor Xa has been claimed to be a better antithrom-
botic target than thrombin because there are indications
that factor Xa inhibitors may have less propensity to
show bleeding side effects.2,3 Additionally, a rebound
effect has been observed following cessation of therapy
with direct thrombin inhibitors.4 Potential indications
are for deep vein and arterial thrombosis, post operative
prophylactic use, myocardial infarction, and stroke.
Crystal derived structural models are available for
quite a number of serine proteases. Their mode of
catalytic action is well understood, and the structural
features that give rise to substrate selectivity have in
many cases been elucidated. Recently, for both thrombin
and factor Xa, structures have been published which
have a variety of competitive inhibitors bound in the
active sites. Despite this wealth of structural informa-
tion, the role of structure-based design, to date, has
generally been to help suggest analogues of an existing
lead and to post-rationalize the activity data, rather
than as a tool for the de novo design of inhibitors.5,6,7
The first crystal structure of the factor Xa enzyme
was published by Bode’s group in 1993 (Brookhaven
code 1HCG).8 This crystal structure is missing the Gla
(γ-carboxyglutamic acid) domain (N terminal residues
1-45) and also residues Glu 146-Gln 151, close to the
active site, which are apparently autocleaved during
crystallization. The S1 pocket of the active site is
occupied by the A-chain terminal Arg 439 of a neighbor-
ing factor Xa molecule, which hydrogen bonds to Asp
189 in the standard bidentate fashion. The active site
is similar to trypsin, but differs from many other serine
proteases in having a large S4 pocket. The S1 pocket
differs from trypsin in that Ser 190 is replaced with
hydrophobic Ala 190. These features suggest that selec-
tive small molecule inhibitors for factor Xa can be
obtained and, indeed, many are already known. For
example, DX-9065a is a dicationic inhibitor with a Ki
of 41 nM against factor Xa and a Ki of 630 nM against
trypsin and >2000 nM against thrombin (Figure 1.).9
YM-60828 is another dicationic inhibitor with a Ki of
2.3 nM against factor Xa, 159 nM against trypsin, and
* Corresponding author: J ohn W. Liebeschuetz, Tularik Ltd, Beech-
field House, Lyme Green Business Park, Macclesfield SK11 0J L, U.K.
Tel: (44) 1625 427369. Fax: (44) 1625 612311. E-mail: jliebeschuetz@
tularik.com.
† Protherics Molecular Design.
‡ University of Bristol.
10.1021/jm010944e CCC: $22.00 © 2002 American Chemical Society
Published on Web 02/16/2002

1222 J ournal of Medicinal Chemistry, 2002, Vol. 45, No. 6
Liebeschuetz et al.
chemicals. The selection is normally carried out by
searching according to a simple pharmacophore ap-
propriate to the target pocket. Each list may contain
several thousand different chemicals.
Each substituent from a list is computationally
screened for fit to the target pocket. This was done in
two stages in the work presented here. First each
substituent is attached to the template and then a user
defined number of conformations are roughly assessed
for goodness of fit and favorable interactions. This was
done by assessing the substituent for complementarity
to an ‘interaction site model’ in the manner of Klebe.15
The number of attempts at finding graph matches is of
the order of a thousand. A match may not represent a
viable conformation for a substituent, and this is
checked by carrying out a local conformationally flexible
fitting of the substituent onto the interaction sites using
a directed tweak algorithm. The number of attempts at
finding a viable conformation from a given match is of
the order 30 to 120. Many substituents may be rejected
at this stage. Accepted substituents can be further
refined in a second stage, using molecular mechanics
to optimize internal geometries and substituent:receptor
interactions, via an implementation of the ‘CLEAN’
force field.16 Each substituent is scored using an empiri-
cal scoring function to find and preserve the best match
for each substituent. More recent versions of PRO_SE-
LECT use a docking protocol on both template and sub-
stituent to obtain a good binding conformation in which
the template position can be adjusted.17 The empirical
scoring function represents binding energy and is
derived by regression analysis of measured binding
affinity to terms known to be important in determining
affinity and calculable from existing crystal structures.
A number of different such functions have been pub-
lished. The empirical scoring function used in this work
is that of Bo¨hm, although we have subsequently derived
our own scoring function, ChemScore.18-20
F igu r e 2. Flow diagram demonstrating the PRO_SELECT
approach.
>10 000 nM against thrombin.10 Both these compounds
are currently in preclinical and clinical trials and show
some promise as oral antithrombotics without bleeding
side effects.11,12 A crystal derived structure for DX-9065a
bound to factor Xa (Brookhaven code 1FAX) exists. This
indicates that the naphthamidine portion sits in the S1
pocket, making a single hydrogen bond to Asp 189, while
the acetamidinopyrrolidine portion sits in the electron
rich S4 pocket and makes a hydrogen bond to the
pendant Glu 97 side chain.13
Vir tu a l Scr een in g Meth od ology
The score is thus used to drive placement of the
substituent. It is also used by the molecular designer
to differentiate between substituents in the design of
the final sublibrary. It is generally used only as a cutoff
filter to pick out those substituents which have the best
chance of making good binding interactions. It is not
expected that the Bo¨hm score will correlate well with
actual binding affinities of the library members, as the
Bo¨hm score is derived from complexes in which only
favorable ligand:protein interactions are made. Unfa-
vorable interactions which are not penalised by the
Bo¨hm score, for instance polar:lipophilic contacts, are
liable to frequently arise in the PRO_SELECT place-
ments. For this reason, other criteria are also used in
the selection process. These can include strain energy
estimation, a diversity metric or calculated physical
chemical properties. Manual inspection of the predicted
binding modes also plays an important role. The process
may be repeated for separate substituent lists attached
to different points on the template. Thus the final
library will consist of an array of substituents for a
single attachment point or a combinatorial array con-
structed from two or more lists each corresponding to a
separate attachment point.
We have recently published a methodology for the
computational or ‘virtual’ screening of combinatorial
libraries against the active site of an enzyme.14 The
program central to this methodology is called PRO_SE-
LECT. A flow diagram illustrating the methodology is
given in Figure 2. The idea is to generate relatively
small candidate libraries (10-1000 compounds) which
have a high ratio of hits to inactives. These libraries
are generally based on a template chemistry and are
designed to be synthetically accessible in a timely and
cost-effective manner. The catholic nature of the screen-
ing procedure allows a wide diversity of structures to
be explored within the constraints imposed by the
template.
We start with a template structure. This is designed
to fit into part of the active site, normally in a central
position, and make favorable interactions with the
enzyme. The template has attachment points to which
substituents can be affixed using simple chemical reac-
tions. Each attachment point is ideally directed toward
a different pocket in the binding site, in which a suitable
substituent may find favorable interactions. The sub-
stituents must be readily accessible to allow rapid
chemical synthesis. Therefore the substituents are
directly derived from lists of appropriate reagents, each
selected from a directory of commercially available
PRO_SELECT was first validated using thrombin as
a target.21 We now report on the use of this methodology

PRO_SELECT: A Tool for Rapid Lead Discovery
J ournal of Medicinal Chemistry, 2002, Vol. 45, No. 6 1223
exploit other specificity pockets within the active site.
The S1 pocket was chosen as the most suitable pocket
in which to place a template. Inhibitors of the trypsin
class of serine proteases generally have a cationic moiety
in this pocket which can make a single or bidentate
H-bond with Asp 189. This provides a good anchor and
a predictable binding orientation for a variety of ligands.
The strategy of iterative library design to grow into
different regions of the factor Xa active site is demon-
strated in Figure 3, with reference to the 1HCG struc-
ture. This was the only structure openly available at
the start of this work and is therefore the one that was
used. Initial placement of the template in S1 (blue) was
to be followed by development of the first library to
probe the central (red) region. This was to be followed
up by second and third libraries to occupy the green and
purple pockets. The red region incorporates the Ser 214
to Gly 218 backbone, the position of which is well
conserved in many serine proteases and which is known
to be capable of providing H-bonding recognition inter-
actions with natural substrates. It also partially con-
tains the characteristic ‘aromatic box’ of factor Xa,
constructed from the Tyr 99, Trp 215, and Phe 174 side
chains. This ‘box’ constitutes the majority of the S4
pocket. The purple region represents the back of the S4
pocket and is characterized by three backbone carbonyls
from Thr 98, Glu 97, and Lys 96 and the anionic Glu
97 side chain which can overhang the pocket. In theory
these groups are available to hydrogen bond strongly
to electropositive and cationic groups. The green region
represents a hydrophobic pocket that has as its base
the Cys 191-Cys 220 disulfide bridge, Gln 192 and Arg
143 side chains as the left-hand wall, and the Gly 218
backbone as the right-hand wall. This pocket is well
conserved in many serine proteases but has not been
exploited frequently in the design of inhibitors, perhaps
because the pocket can often be occupied by the mobile
side chain of residue 192. The structure of the potent
anti-Xa protein, tick anticoagulant peptide, bound to
factor Xa, shows that this ligand can make use of this
region, albeit with considerable reorganization of the
active site.24 We were of the opinion, after consideration
of the 1HCG structure, that this pocket could be a prime
target area for substituent placement in factor Xa.
F igu r e 3. Factor Xa active site VdW surface (A) and sche-
matic (B) illustrating the strategy used in the iterative design
process. The initial template, 3-benzamidinecarbonyl, and the
interaction site model for the first library are shown (A).
to design factor Xa inhibitors as the primary stage in
an ongoing program to discover novel antithrombotic
drugs. This has led to a new class of chemistry that is
highly active and specific for factor Xa. Several groups
have described similar ‘Virtual Screening’ approaches
that have lead to active inhibitors against other tar-
gets.22,23 To our knowledge this is one of the first
examples where such an approach has led to molecules
sufficiently active to show therapeutic effect in relevant
animal disease models. It is also the first published
example of the de novo design of inhibitors for factor
Xa.
Resu lts a n d Discu ssion
Design Str a tegy
Tem p la te Selection a n d F ir st Libr a r y Design . It
was decided to employ an amidino group to anchor the
S1 template via a bidentate hydrogen bond to Asp 189.
We were aware that such a group has in the past lead
to problems in oral availability and rapid clearance.
Nevertheless it was felt important to use a template
that had reasonable base activity of its own so that
structure/activity trends would become immediately
apparent.
It was envisaged that the synthetically facile linkage
of the template to the substituent would be via an amide
bond. It was also envisaged that this amide bond might,
itself, be able to make hydrogen bonding interactions
with the active site. Several possible template candi-
dates were considered. It was decided to use PRO_SE-
LECT to design a library with each template and to
compare the quality of the libraries in order to select
the best template. Three of the templates examined are
shown in Table 1.
It is usual, when considering the design of a combi-
natorial library, to build the library around a central
template to which two or more substituents are attached
or incorporated via facile chemistry. The placement of
such a template in an active site would necessarily be
central. Examination of the factor Xa 1HCG structure
reveals that the central region of the active site is very
broad, and the disposition of polar ‘anchoring’ sites is
sparse (Figure 3A). Therefore we felt no confidence that
a central template could be designed which would be
guaranteed to bind in a single predictable manner. An
alternative strategy was chosen in which a template
would be placed in one of the specificity pockets and
PRO_SELECT would be used to find potential substit-
uents to fit the central portion of the active site.
Synthesis and screening of this initial library would
then allow a lead molecule to be selected. This could be
further elaborated, again by use of PRO_SELECT, to

1224 J ournal of Medicinal Chemistry, 2002, Vol. 45, No. 6
Liebeschuetz et al.
Ta ble 1. Hit Rate and Hit Quality of Libraries Designed
around Three Different S1 Templates and a Single Substituent
List, Using PRO-SELECT
The initial stage in using PRO_SELECT was to create
a ‘Design Model’. This is a simple ‘interaction site model”
of the active cleft.14 The Design Model for the region
targeted for the first library is included in Figure 3A.
The dark/light blue vectors represent hydrogen bond
donor sites, the blue/purple vectors represent hydrogen
bond acceptor sites, and lipophilic point sites are
represented as orange crosses. The first step in sub-
stituent evaluation was to find matches between inter-
action sites in the substituent, attached to template, and
complementary sites in the design model. Also il-
lustrated in Figure 3A is one of the templates bound in
the S1 pocket in the binding orientation employed in
the PRO_SELECT job.
The choice of substituent lists and the virtual screen-
ing protocol used are given in the Experimental Section.
A summary of the output performance is given in Table
1 for each of the three templates. The number of
substituents that passed the matching stage was roughly
the same for each template. However, the number of
substituents binding reasonably well (Bo¨hm score < -3)
was much less for the 4-benzamidinocarbonyl template
than for the other two templates. The arginine template
had marginally more high quality hits than the 3-ben-
F igu r e 4. Lead compounds arising out of library 1.
namide substituent was found to be unobtainable, and
the presence of the highly carcinogenic â-naphthylamine
substructure mitigated against its synthesis. Neverthe-
less the motif looked of sufficient interest to be inves-
tigated further. Accordingly it was decided to append
the glycine to the 3-amidinobenzoyl group and use this
molecule as a larger template. PRO_SELECT was used
to search for substituents which, when attached via an
amide bond to the glycine carbonyl, would probe deeper
into the S4 pocket. It was envisaged that it might be
possible to use the carbonyl groups at the back of the
S4 pocket for hydrogen bonding. For this reason, the
library of substituents chosen for virtual screening was
selected to be bisamines separated by a hydrophobic
group. PRO_SELECT offers the ability to interconvert
functional groups in silico prior to virtual screening.
This ‘deprotection’ facility mimics a synthetically facile
functional group transforming reaction.13 This was
exploited here to broaden the diversity of possible
substituents by inclusion of a list of bis-nitriles, con-
verted on the fly to bis-amines.
Further details of the virtual screening protocol are
given in the Experimental Section. Eight substituents
were chosen which had Bo¨hm contributions of better
zamidinocarbonyl template. However, the Bo¨hm score
-1
for the former was 10 kJ mol
higher than that for
the latter (-5.3 versus -16.4), corresponding to a
shortfall of roughly 2 orders of magnitude in terms of
binding affinity. It was concluded therefore that the
3-benzamidinocarbonyl template was the most ap-
propriate template to use.
than -10 and strain energies of lower than 25 kJ mol
-1 25
.
Good hits from this second search were incorpo-
rated with those of the first, and 14 compounds in total
were selected for synthesis (library 1a).
Addition of bulky and partially hydrophobic substit-
uents to a small template could arguably be expected
to lead to increased efficacy through nonspecific lipo-
philic interactions. We wanted to be confident that any
increase in activity in the initial library was not
generated this way. Therefore it was decided to prepare
a second library using some of the same substituents
selected for the 3-amidinobenzoyl template but attached
to the ‘wrong’ 4-amidinobenzoyl template. Eight sub-
stituents were chosen (library 1b).
A disappointing feature, even in the case of the
3-benzamidinocarbonyl template, was the paucity of
high scoring substituents (score < -10). Substituents
could be found which made either the desired polar
interactions or hydrophobic interactions but rarely both.
The number of highly promising substituents therefore
appeared limited. The reason for this is likely to be the
restriction on substituent diversity, placed upon us by
using solely the Available Chemicals Directory as a
source. Six targets from this first PRO_SELECT run
were selected for synthesis.
One substituent that did score exceptionally well was
the glycine-2-naphthylamide, 1 (Figure 4). The naphthyl
group was found to sit well in the hydrophobic S4
pocket, and the glycine CdO was able to H-bond with
Gly 218. Both of these effects led to a good Bo¨hm score.
When modeled against the active site, it was found that
the NH from the benzamide amide group also could
make a hydrogen bond to Gly 216. This naphthylglyci-
The synthetic routes used for the preparation of these
compounds are shown in Schemes 1 and 2. Simple
amine substituents, as exemplified by amino acids and
their esters, were prepared by coupling with 3-cy-
anobenzoic acid (using TBTU in DMF), convertion to
the imidate (HCl in ethanol), and then to the amidine
(using ammonia in ethanol). Hydrolysis of the ester was
accomplished using aqueous sodium hydroxide in etha-
nol, Scheme 1.

PRO_SELECT: A Tool for Rapid Lead Discovery
J ournal of Medicinal Chemistry, 2002, Vol. 45, No. 6 1225
Ta ble 2. Comparison of Activities against Factor Xa, Trypsin, and Thrombin for Benzamidine and Libraries 1a-c, 2a-e, and 3aa
factor Xa
trypsin
mean
(SD Kib (µM) Ki (µM)
thrombin
mean
mean
mean
best
Ki (µM)
mean
pKi
best
mean
pKi
best
library
n
pKi
(SD Kib (µM)
(SD Kib (µM) Ki (µM)
ben za m id in e
1a
1a (subset)
1b
1a + 1c
2a
2b
2c
3.7
4.3
4.2
3.1
4.4
5.5
4.7
4.7
4.5
5.7
6.5
200
50
58
780
36
3.5
21
21
29
1.9
0.34
200
8.5
8.5
162
2.1
0.22
1.0
2.9
3.0
0.063
0.016
4.8
4.4
4.2
3.7
4.2
4.6
4.3
4.3
4.3
4.9
5.4
17
38
61
180
61
26
44
43
47
11
4.1
17
6
6
2.3
6.5
9
21
8
15
0.79
0.040
4.6
4.5
4.6
3.6
4.3
c
25
31
23
230
51
c
c
c
c
18
7
25
5
5
89
3.0
c
c
c
c
14
7
7
36
6
6
(0.6
(0.8
(0.3
(0.8
(0.8
(0.7
(0.4
(0.5
(0.8
(0.8
(0.8
(0.9
(0.9
(0.8
(0.6
(0.4
(0.4
(0.4
(0.7
(0.6
(0.7
(0.8
(0.5
(0.9
c
c
c
9
2d
2a + 2e
3a
11
34
106
4.7d
5.2e
(0.5
(0.6
1.4
0.023
a
b
Figures not in bold represent benchmark libraries. Ki figures are geometric means calculated as the reciprocal antilog of the pKi
mean. c Insufficient compounds tested against thrombin. Only 23 compounds tested against thrombin. e Only 99 compounds tested against
d
thrombin.
Sch em e 1. Solution-Phase Synthesis of Inhibitorsa
substituents to library 1b are also given. The designed
library 1a with the 3-amidinobenzoyl template demon-
strates on average markedly improved activity over
benzamidine (Ki of 200 µM, pKi of 3.7). Moreover, the
‘control’ library with the 4-amidinobenzoyl template, 1b,
shows no such improvement and is, on average, more
than an order of magnitude less active than the corre-
sponding subset of compounds from library 1a. The
elevation of activity in library 1a is therefore not simply
a function of increasing the size of the molecule by
addition of a lipophilic fragment.
It was decided to broaden the library 1a by making
some simple structure-predicated modifications of some
of the hits. Thus a variety of more lipophilic esters of
the naphthylalanine and tryptophan analogues 2 and
3 were prepared, with the idea they might better fill
the S4 pocket (library 1c). This led to the low micromolar
lead 4. This compound is selective for factor Xa over
trypsin (Ki of 2.0 µM vs Ki of 12.5 µM) despite the fact
that benzamidine itself is 10 fold better for trypsin (Ki
of 20 µM vs trypsin).
Another one of the most active compounds in library
1a was 5 (Ki, 14 µM), and this was selected as a second
possible lead compound. The more active of the two
leads, 4, did not readily lend itself to modification via
easily accessible chemistry. Compound 5, on the other
hand, looked ideally set up for quick variation, both by
replacement of the glycine with other amino acids and
by replacement of the 1,4-bis-aminomethylcyclohexane
fragments. This was therefore chosen as the lead
molecule from which to develop the second library.
a
Conditions: (a) TBTU in DMF; (b) HCl in ethanol; (c) ammonia
in ethanol; (d) NaOH in ethanol.
Where the S4 unit was a bis-amine a solid-phase route
was preferred, Scheme 2. The bis-amine was attached
to 2-chlorotrityl polystyrene resin and coupled to an
Fmoc protected amino acid using TBTU in DMF. The
Fmoc protection was removed with piperidine in DMF
and free amine coupled to 3-amidinobenzoic acid TFA
salt using DIPCI and HOBt. The product was then
cleaved from the resin using TFA/triethylsilane.
Compounds were tested in chromogenic assays and
Ki’s calculated against a range of serine proteases. These
included factor Xa, trypsin, and thrombin.
The mean, standard deviation, and best activities of
libraries 1a and 1b against factor Xa, trypsin, and
thrombin are given in Table 2. The corresponding
activities for the subset of library 1a with common
Sch em e 2. Solid-Phase Synthesis of Inhibitorsa
a
Conditions: (a) TBTU/DIPEA and Fmoc-amino acid in DMF; (b) 20% piperidine in DMF; (c) 3-amidinobenzoic acid TFA salt and
DIPCI/HOBt in DMF; (d) 10% triethylsilane in TFA.

1226 J ournal of Medicinal Chemistry, 2002, Vol. 45, No. 6
Liebeschuetz et al.
F igu r e 5. Compound 5 modeled in the 1HCG factor Xa
structure.
Secon d Lib r a r y Design . Figure 5 illustrates 5
docked into the 1HCG structure. The terminal nitrogen
of the S4 binding portion has been modeled as proto-
nated. The glycine in 5 can be elaborated from either of
the prochiral hydrogens. Elaboration involving a D-
amino acid or analogue thereof (i.e., coming off the
hydrogen marked purple in Figure 5) appeared to have
a good chance of exploiting the lipophilic disulfide pocket
(green area, Figure 3, disulfide in yellow in Figure 5)
according to the model. The enantiomeric L-amino acids
appear not to be able to access good binding sites in the
vicinity. PRO_SELECT jobs were carried out using the
list of available R-amino acids. Both L and D configura-
tions were examined, and both polar and hydrophobic
interactions were sought. The list of amino acids that
resulted consisted mainly of D-amino acids, which
docked into the disulfide pocket. A library of seven
compounds from this list, all D-enantiomers, was syn-
thesized (sublibrary 2a), all with the 1,4-bis-aminom-
ethylcyclohexane S4 fragment. The corresponding L-
enantiomers were also made (sublibrary 2b). A variety
of other D-amino acids (9 in all, sublibrary 2c) and
L-amino acids (11 in all, sublibrary 2d) was also utilized.
Thus libraries 2b, 2c, and 2d represent benchmark
libraries to compare with 2a. The solid-phase synthetic
route in Scheme 2, was applicable to all the library 2
compounds.
F igu r e 6. Lead molecule for library 3 (6) and examples of
library 3 compounds.
logues were designed, using both medicinal chemistry
principles and modeling (library 2e). Only modest
increases in activity were achieved. The reasons for this
remained unclear until the crystal structure of DX-
9065a bound to factor Xa was published (1FAX).12 This
structure retains the autolysis loop, missing in the
1HCG structure. This loop sits at the back of the
disulfide pocket area, severely curtailing its size and
depth.
Th ir d Libr a r y Design a n d Activity. Only a limited
number of substituents designed to access the S4 pocket
was tried in libraries 1 and 2. Many of these were
diamines. However, it was accepted that there was
considerable scope to find better S4 pocket binders and
that these could either contain cations or hydrogen bond
donors or, alternatively, they could be hydrophobic in
nature and interact strongly with the aromatic ‘box’.
Therefore it was decided to carry out PRO_SELECT jobs
using a docked conformation of 6 as the template, with
the aim of replacing the terminal diamine with a
hydrophobic primary or secondary monoamine. Several
jobs were run. The list of available monoamines was
large, and the pocket to be filled is also sizable.
Therefore, a much bigger list of quality substituents was
found from these jobs than from those run previously.
Approximately 100 targets were selected (library 3a).
The vast majority of these compounds contained a
lipophilic S4 binder. Where the S4 binder was lipophilic,
a solution-phase synthetic route was used (Scheme 3).
Boc-D-phenylglycine was coupled to an S4 component
in DMF using EDC/HOBt or HOAt. The products thus
obtained were deprotected, using TFA in DCM, and then
coupled to 3-amidinobenzoic acid TFA salt. Where the
S4 binder was a diamine, the solid-phase route (Scheme
2) could be used.
Table 2 gives the mean, standard deviation, and best
activities for all four sublibraries against factor Xa,
trypsin, and thrombin.
The figures in Table 2 indicate that designed subli-
brary 2a is, on average, almost an order of magnitude
more active against factor Xa than any of the three
comparison libraries, 2b, 2c, and 2d. However, this is
not mirrored in the average trypsin activities. The
compound within the library that showed the highest
activity was 6 (Figure 6), with a Ki of 220 nM against
factor Xa and 7.4 µM against trypsin. Analysis of the
PRO_SELECT run and further modeling revealed that
the phenyl group appeared able to sit well into the
disulfide pocket. It also appeared able to make an edge-
on interaction with the disulfide bridge. This compound
was selected as the lead molecule for the third cycle of
PRO_SELECT driven optimization.
Summary activity data for library 3a is given in Table
2. Activity, in relation to the lead compound 6, was
generally retained and, in many cases, improved upon
in this library, despite the fact that there was usually
no possibility of obtaining a hydrogen bond to the back
of the S4 pocket in the new series. The most active
targets out of this set of analogues, 7, Ki 16 nM, and 8,
Ki 16 nM (Figure 6), showed a greater than 10-fold
increase in activity over 6 and retained selectivity over
other serine proteases (Ki’s of 980 and 1700 nM against
Analysis of the 1HCG structure suggested that there
was plenty of room left in the disulfide pocket for further
hydrophobic elaboration, especially in the vicinity of the
3 and 4 positions of the phenyl ring in 6. Accordingly,
to exploit this extra binding possibility, further ana-

PRO_SELECT: A Tool for Rapid Lead Discovery
J ournal of Medicinal Chemistry, 2002, Vol. 45, No. 6 1227
Sch em e 3. Solution-Phase Synthesis of Inhibitorsa
the disulfide pocket, which can also be accessed in
trypsin, and the benzoyl piperidine group sits centrally
in the S4-pocket with the carbonyl pointing upward.
Both amide bonds in the ligand make the predicted
hydrogen bonds. The first H-bonds Gly 216 through NH,
the second, Gly 218, through CdO. The bound ligands
appear least well superimposed in the S4 region.
However, trypsin and factor Xa differ in the S4 pocket,
most noticeably at residues 99 and 174 (Leu and Gln
in trypsin). Leu 99 allows more room for the terminal
phenyl group to sit on the right-hand side of the pocket,
than does Tyr 99 in factor Xa.
The third library contained a wide diversity of active
chemistries. This gave rise to a number of structurally
different lead molecules that could be exploited further
using either classical medicinal chemistry or a structure
informed approach. Compound 9 represents one ex-
ample where structure-based modification of a PRO_SE-
LECT lead gave rise to other chemistries with potent
activity and selectivity. The synthesis of this type of
compound is outlined in Scheme 4. The Boc-N-protected
S4 intermediates were first prepared as shown in
Scheme 3. The Boc protection was removed using TFA/
DCM, and the resulting amine reacted with 1,3-bis-tert-
butoxycarbonyl methyl pseudothiourea in the presence
of mercury II chloride to give the bis-butoxycarbonyl
guanidines. Final treatment with TFA/DCM and puri-
fication by preparative HPLC gave the final products
isolated as TFA salts. Compound 9 has a Ki of 26 nM
against factor Xa but only 1.6 µM against trypsin and
8 µM against thrombin. Other related compounds had
similar activity and selectivity factors of 250.
a
Conditions: (a) coupling agentssee text; (b) 25% TFA in DCM;
(c) 3-amidinobenzoic acid TFA salt and DIPCI/HOBt in DMF.
F igu r e 7. Compound 7 bound in trypsin (green) and super-
imposed with the predicted binding mode in Factor Xa (blue).
trypsin, 1100 and 1600 nM against thrombin). These
molecules are both amenable to further optimization at
the S4 end of the molecule and thus represent good third
cycle leads. Selectivity for the library as a whole against
thrombin and trypsin was slightly improved, despite the
fact that both these enzymes have sizable lipophilic
pockets that correspond to the Xa S4 pocket.
A cocrystal of compound 7 bound in trypsin, was
successfully obtained. Figure 7 compares the predicted
binding mode in factor Xa (blue) with that found in
trypsin (green). The benzamidine is found in S1, hydro-
gen bonding in a bidentate fashion to Asp189; the
phenyl portion of the phenylglycine linker is found in
Over view . Figure 8 illustrates how the activity of the
series evolved. The benzamidine template and lead
molecules from each library are indicated. Each library
contains, as well as the original PRO_SELECT mem-
bers, additional structure-based and medicinal-chemistry-
based analogues, some of which are mentioned above.
To obtain predicted binding affinities for all compounds
under a consistent set of docking conditions, all mol-
ecules were subsequently redocked, keeping the ap-
propriate template portion of each molecule rigid as in
the original PRO_SELECT run. Predicted binding af-
finity was calculated from the docking score for each
molecule. Predicted binding affinity is plotted against
Sch em e 4. Preparation of S4 Guanidino Compoundsa
a
Conditions: (a) TFA/DCM; (b) 1,3-bis-tert-butoxycarbonyl methyl pseudothiourea/HgCl2; (c) TFA/DCM.

1228 J ournal of Medicinal Chemistry, 2002, Vol. 45, No. 6
Liebeschuetz et al.
important to have good compound integrity in each
library. Impure compounds gave rise to data that
confounded early stage SAR development in some cases.
Several other points are worth making. First, the lead
molecule chosen at the beginning of each cycle was not
necessarily the most active molecule in the previous
library. It is more important that the lead be reasonably
easy to chemically modify and also be likely to have the
least pharmacokinetic problems.
Second, this approach is synergistic with modern
combinatorial chemical techniques and it could be used
with benefit alongside them, to design large focused
libraries. However, in such an approach only one
synthetic route is generally followed, and it is accepted
that a fraction of library members will not be success-
fully made. All compounds in a PRO_SELECT designed
library are to some extent ‘cherished’, and therefore
there is more incentive to make all members of the
library than is usual in combinatorial chemical exer-
cises. If medium throughput array chemistry is em-
ployed and the libraries are relatively small, then it is
practical and useful to develop and employ more than
one synthetic route, as was done in this study.
F igu r e 8. Activity progression in the benzamidine series from
the first library (circles) to the second (squares) and third
(triangles) libraries. Original template and lead molecules for
subsequent library development are marked.
Con clu sion s
We have described the successful application of
‘virtual’ screening in the rational design of potent and
selective factor Xa inhibitors. The starting point for the
program was a simple template, benzamidine, placed
in the S1 pocket. Iterative library design built upon the
template in order to access other pockets in the active
site. The chemistry involved in the synthesis of these
libraries was designed to be straightforward, allowing
rapid access of targets. Libraries using the same chem-
istry were also synthesized which were not designed
through use of PRO_SELECT but which were reason-
able from a medicinal chemistry standpoint. These
consistently showed poorer average activity against (by
roughly a factor of 10) and selectivity for factor Xa than
those designed by ‘virtual’ screening.
F igu r e 9. Activity against factor Xa in the benzamidine series
versus predicted activity calculated from docking score. Com-
pounds are from both the PRO_SELECT (b) and benchmark
(4) libraries.
factor Xa activity in Figure 9. Both those compounds
selected through virtual screening and those not so
selected (libraries 1b and 2b,c,d) are plotted. There is
not a tight correlation. Nevertheless, out of the selected
compounds, there is only a small proportion which score
well but which show poor activity. In addition, those
compounds not selected through virtual screening gen-
erally show both poor predicted and actual activity.
These things are what we would hope to see as the
primary aim of the virtual screening approach is to
concentrate synthetic resource in the areas where
reasonable activity is most likely to reside. One of the
reasons the correlation of predicted and measured
activity is not better is because the empirical scoring
function is only able to describe positive features of the
binding mode and not negative ones other than rota-
tional entropy. In addition, current scoring functions
ignore subtle electrostatic effects such as π stacking and
so on, which can greatly influence activity. So quite a
spread of activities is obtained in the final data set. This
spread of activity can be very useful, however, as it often
envelops a rudimentary structure-activity relationship
among subgroups of similar chemistry within the li-
brary. This provides a springboard for a classical lead
optimization approach. For this reason, we found it
Several lead molecules with diverse structure and Ki
’s in the range of 10 to 50 nM were obtained, represent-
ing 4 orders of magnitude increase with regard to the
binding affinity of the starting template. Further testing
established that some of these compounds showed
antithrombotic activity when given intraperitoneally in
the Wessler stasis model of venous thrombosis in rat
and therefore have potential therapeutic use as an
injectable treatment. None of the compounds showed
an effect when given orally, however. It was felt this
was likely to be because of the highly basic benzamidine
moiety. Further work has since been carried out to
replace this group with a group of moderate basicity and
to optimize potency by modification elsewhere in the
molecule. Highly potent and selective compounds with
strong oral antithrombotic activity were found. These
will be described in later publications.
The design of libraries through structure-based ‘vir-
tual screening’ is a methodology of drug design that is
currently of high interest. We have demonstrated that
it can be an efficient method for the generation of potent
and selective lead molecules, in cases where a target
protein structure exists.

PRO_SELECT: A Tool for Rapid Lead Discovery
J ournal of Medicinal Chemistry, 2002, Vol. 45, No. 6 1229
F igu r e 11. Template and pharmacophore for library 1b: (A)
template, (B) pharmacophore for bis-amine substituents, (C)
pharmacophore for bis-nitrile list.
F igu r e 10. Pharmacophores used in finding substituent lists
for library 1a: (A) pharmacophore for H-bond acceptor sub-
stituents, (B) pharmacophore for lipophilic substituents.
sites). Hydrophobic sites were assigned to carbons in five- and
six-membered carbocyclic rings. Passes from the interaction
site matching stage were minimized using the Clean force field,
keeping the template rigid, and scored for binding affinity
using the empirical scoring function of Bo¨hm.18 The best
scoring conformation per substituent was retained. The final
list of substituents was filtered to pick out only those which
had Bo¨hm contributions of -3 kJ mol-1 or better.
Exp er im en ta l Section
Com p u ta tion a l Deta ils. Manipulation and inspection of
receptor, template, and ligands before and after minimization
or simulation was carried out using InsightII 95.0.25 All
molecular mechanics minimizations and molecular dynamics
simulations were carried out using the Discover 2.97 program26
with the CFF95 force field. The Discover calculations were
carried out on a Convex Exemplar (16 × HP7100s) running
SPP-UX 3.1. The pharmacophore searches of the Available
Chemical Directory (ACD)27 and substituent list generation
were performed using ISIS/Base 1.228 and ISIS/Draw 1.28
Searches were carried out allowing full conformational flex-
ibility. Inspection of the structures and associated numerical
data generated by PRO_SELECT was carried out using in-
house graphics software (XMOLBROWSE). Substituent list
generation and graphical inspection was carried out on SGI
Indigo R3000 workstations running IRIX 4.0.5.
P r otocol for th e Design of Libr a r y 1a Mem ber s Ba sed
on Th r ee S1 Tem p la tes. The template positioning for the
arginine template (Table 1) was taken from a minimized
structure of PPACK in FXa. The docked conformation of
PPACK was derived from the 1PPB (PDB designation) PPACK/
thrombin structure. The template positioning of the benza-
midines was derived from a docking of DX-9056a into the
1HCG (PDB designation) factor Xa structure. The carbonyl
group was placed at the ring position proximal to the lip of S4
in the case of the 3-amidinobenzoyl template (Table 1). Both
possible orientations of the carbonyl group planar to the ring
were treated as valid, and PRO_SELECT runs were carried
out on each. Only one orientation was deemed useful for the
4-amidinobenzoyl template.
P r otocol for th e F u r th er Design of Libr a r y 1a Mem -
ber s Ba sed a r ou n d th e 3-Am id in oben zoylglycin e Tem -
p la te. Molecular dynamics simulations were carried out on
ligands 1 and 4, among others, docked into the 1HCG factor
Xa structure. Snapshots were taken from the simulations at
1 ps intervals and each snapshot minimized. Analysis of these
simulations allowed the selection of two significantly different
template geometries, both of which gave reasonably good Bo¨hm
scores. PRO_SELECT runs were carried out using each
geometry. The template structure is given in Figure 11A.
The substituent list was prepared by searching the ACD
using the pharmacophore in Figure 11B. The list contained
422 substituents. A second bis-nitrile list was prepared ac-
cording to the pharmacophore in Figure 11C. PRO_SELECT
runs were carried out separately on this list which contained
656 substituents. Substituents for library 1a were selected out
of that set of substituents with Bo¨hm contributions of -9 kJ
mol-1 or better, and strain energies of 27 kJ mol-1 or lower.25
The hits were clustered using the J arvis-Patrick method
within XMOLBROWSE, and substituents were selected from
those that passed the criteria via manual inspection of each
ligand docking. It was found that the two different template
positions gave rise to quite different sets of high scoring
substituents.
P r otocol for th e Design of Libr a r y 2. The origin of the
templates for this library was an “annealed” geometry of 5
manually docked into the 1HCG FXa X-ray structure. The
receptor geometry was held rigid while the ligand was
subjected to molecular dynamics at temperatures up to 300 K
with subsequent slow cooling followed by minimization. This
final geometry then provided an appropriate ligand template
geometry. The disconnection point for this template was taken
to be either of the prochiral hydrogens of the ligand glycine
methylene. Each of these hydrogen atoms was substituted
during separate PRO_SELECT runs as it was desired to look
at both L- and D-amino acids.
The list of potential “amino acid” substituents was obtained
by searching the ACD for all free R-amino acids. There were
1230 possible amino acid substituents after removal of high
molecular weight compounds (MW > 250) filtering of undesir-
able chemistries and conversion into 3D. The substituent
disconnection point employed here was the amino acid side
chain f C-R bond.
Two lists of substituents were prepared using the ACD as
a source. Pharmacophores for the substituent selection were
calculated from the crystal structure assuming reasonable
placement of the template and are given in Figure 10. One
pharmacophore was targeted toward the polar functionality
at the lip of S4, with the primary aim of picking up interactions
with the N-H of Gly 218 (Figure 10A). The other was targeted
at the hydrophobic ‘aromatic box’ (Figure 10B). The amino
group common to both is the linking group to the template.
The initial aim was to only probe the central region of the
active site. Therefore a molecular weight limit of 250 was set.
Conformationally flexible searching of the ACD with these
pharmacophores using ISIS/Base software generated 2D struc-
ture lists which were than converted to 3D using Converter26
(Molecular Simulations Inc.).
The final ‘polar’ list numbered 1534 substituents, the
‘nonpolar’ one numbered 797. Each list of substituents was
evaluated separately by PRO_SELECT. The polar substituents
were assigned hydrogen bonding acceptor sites on double
bonded oxygen, nitrogen, and sulfur (this was found as
effective at finding hits as assigning both acceptor and donor
Several SELECT jobs were then carried out to focus on
lipophilic or polar interactions arising from the amino acid

1230 J ournal of Medicinal Chemistry, 2002, Vol. 45, No. 6
Liebeschuetz et al.
Purification was by gradient reverse-phase HPLC on a
Waters Deltaprep 4000 at a flow rate of 50 mL/min using a
Deltapak C18 radial compression column (40 mm × 210 mm,
10-15 mm particle size) and solvent mixtures consisting of
eluant A (0.1% aq TFA) and eluant B (90% MeCN in 0.1% aq
TFA) with gradient elution.
Analytical HPLC was on a Shimadzu LC6 gradient system
equipped with an autosampler, a variable wavelength detector
at flow rates of 0.4 mL/min. Eluents A and B as for preparative
HPLC used the following columns: Luna2 C18 2 × 150 mm 5
µm, Symmetry C8 4.6 × 30 mm 3.5 µm (Phenomenex). Purified
products were further analyzed by Maldi TOF and/or LCMS
and 1H NMR.
Compound libraries were prepared using both solid-phase
and solution-phase parallel synthetic methods as described
below.
Sim p le Am in e S4 Un its (Sch em e 1). Amino acid ester
hydrochlorides were either (i) coupled to 3-amidinobenzoic acid
TFA salt using DIPCI/HOBt in DMF containing 1 equiv of
DIPEA, purified by reverse-phase preparative HPLC and
isolated as the TFA salt, or (ii) coupled to 3-cyanobenzoic acid
using TBTU/HOBt in DMF containing 1 equiv of DIPEA and
the nitrile converted to the amidine by sequential treatment
with HCl gas in ethanol and ammonia gas in ethanol. The
products were purified by reverse-phase preparative HPLC
and isolated as the TFA salt. Compounds 3 and 4 were
prepared by these routes.
3-Am id in oben zoyl-D-tr yp top h a n TF A Sa lt, 3. To a solu-
tion of 3-cyanobenzoic acid (500 mg, 3.4 mmol), D-tryptophan
methyl ester hydrochloride (866 mg, 3.40 mmol), and HOBt
(459 mg, 3.4 mmol) in DMF (10 mL) were added TBTU (1.09
g, 3.40 mmol) and DIPEA (592 µL, 3.4 mmol). The reaction
was stirred until complete by TLC and then partitioned
between ethyl acetate and water. The organic solution was
evaporated in vacuo to give 3-cyanobenzoyl-D-tryptophan
methyl ester (954 mg, 81%).
F igu r e 12. Pharmacophores used in finding substituent lists
for library 3: (A) pharmacophore for hydrophobic primary
amines, (B) pharmacophore for hydrophobic cyclic secondary
amines, (C) pharmacophore for hydrophobic acyclic secondary
amines. Aromatic and fused bicyclic systems were allowed in
the lists generated by pharmacophores A, B, and C.
substituent with the receptor interaction sites. A list of
substituents suitable for synthesis was generated after re-
moval of duplicates, inappropriate substituents, e.g., side
chains from substituents available only in L form where the D
was preferred, and filtering of poor scoring substituents using
the criterion that the Bo¨hm score be less than -3.0 kJ mol-1
.
P r otocol for th e Design of Libr a r y 3. Compound 6 and
also an analogue of compound 6 which had the bis(aminom-
ethyl)cyclohexane replaced by a 1-adamantylamine, docked
into the 1HCG factor Xa structure, were simulated at 300 K.
The receptor was kept rigid, and snapshots were taken at 5
ps intervals and minimized. Low energy snapshots were
selected to provide two different template positionings. Three
pharmacophores were used to search the ACD for appropriate
hydrophobic amines.
These pharmacophores, given in Figure 12, represent
respectively primary amines, secondary cyclic amines, and
secondary acyclic amines. The hydrophobic parts of the phar-
macophores were designed so as to avoid linear hydrocarbons.
Hits that had several polar groups were excluded, as were
certain classes of reactive chemistry. A molecular weight limit
of 250 for free base was used. The primary amine list
numbered, after conversion to 3D and enumeration of enan-
tiomers and diastereomers, 1053 molecules, and the cyclic
amine list numbered 250 compounds. The secondary acyclic
amine list, which was restricted to substituents containing a
six-membered carbocycle, numbered 366 compounds. The link
site was chosen to be the N-H trans to the carbonyl of the
phenyl glycine, in the case of the primary amine list, but the
C(dO)-N bond for both the secondary and the cyclic lists.
Hydrophobic interaction sites were placed at carbons adjacent
to a hydrophobic branch point. Passes from the interaction site
matching stage were treated as described above. Priority
substituent lists were selected on the basis of having favorable
Bo¨hm contributions (generally -12 kJ mol-1 or better) and
low strain energies (generally less than -21 kJ mol-1). These
lists were clustered according to chemistry and processed by
manual inspection of the binding mode to generate synthetic
candidates.
Ch em istr y. Abbreviations used follow IUPAC-IUB no-
menclature. Additional abbreviations are HPLC, high perfor-
mance liquid chromatography; DMF, dimethylformamide;
DCM, dichloromethane; HATU, O-(7-azabenzotriazol-1-yl)-
1,1,3,3-tetramethyluronium hexafluorophosphate; HOBt, 1-hy-
droxybenzotriazole; TBTU, 2-(1H-(benzotriazol-1-yl)-1,1,3,3-
tetramethyluroniumtetrafluoroborate; DIPEA, diisopropylethyl-
amine; TEA, triethylamine; HOAt, 1-hydroxy-7-azabenzotria-
zole; Fmoc, 1-(9H-fluoren-9-yl)methoxycarbonyl; TFA, trifluo-
roacetic acid; MALDI-TOF, matrix assisted laser desorption
ionization-time-of-flight mass spectrometry. Unless otherwise
indicated, amino acid derivatives, resins, and coupling re-
agents were obtained from Novabiochem (Nottingham, U.K.)
and other solvents and reagents from Rathburn (Walkerburn,
U.K.) or Aldrich (Gillingham, U.K.) and were used without
further purification.
HCl gas was bubbled into a solution of 3-cyanobenzoyl-D-
tryptophan methyl ester (925 mg, 2.66 mmol) in ethanol, and
the mixture was left overnight before evaporating to dryness
in vacuo. The solid was taken up in ethanol, and the solution
was saturated with ammonia gas. After being stirred over-
night, the mixture was evaporated to dryness in vacuo and
the resulting solid purified by preparative HPLC to give a
mixture of 3-amidinobenzoyl-D-tryptophan methyl and ethyl
ester TFA salts.
To a solution of 3-amidinobenzoyl-D-tryptophan methyl and
ethyl ester TFA salts (50 mg) in ethanol (5 mL) was added 1
M aqueous sodium hydroxide (1 mL), and the mixture was
stirred overnight. The mixture was evaporated to dryness in
vacuo and purified by preparative HPLC to give 3-amidi-
1
nobenzoyl-D-tryptophan TFA salt (11 mg). H NMR (CD3CN)
δ 8.14 (1H, s, Ar); 8.08 (1H, d, Ar); 7.82 (1H, d, Ar); 7.68 (2H,
m, Ar); 7.45 (1H, d, Ar); 7.05-7.30 (3H, m, Ar); 4.96 (1H, m,
R-proton); 3.3-3.5 (d-ABq, â-proton). Homogeneous by HPLC
Luna C18, Symmetry C8. High resolution MS (M+1)+ found
351.14555 (C19H18N4O3 requires 351.14568).
3-Am id in o-D-2-n a p h th yla la n in e Eth yl Ester TF A Sa lt,
4. 3-Amidinobenzoic acid TFA salt (100 mg) was added to a
mixture of HOBT (48.6 mg) and DIPCI (57 µL) in DMF that
had been stirring for 10 min. To this mixture was added a
solution of 2-naphthylalanine ethyl ester hydrochloride (100.5
mg) and triethylamine (50 µL). After being stirred overnight,
the crude reaction mixture was purified by preparative HPLC
to give 3-amidinobenzoyl-D-2-naphthylalanine ethyl ester TFA
salt. 1H NMR (CD3CN) δ 7.98 (1H, s, Ar); 7.88 (1H, d, Ar);
7.67 (5H, m, Ar); 7.50 (1H, t, Ar); 7.3-7.4 (3H, m, Ar); 4.80
(1H, dd, R-proton); 4.0 (2H,q, Et); 3.35-3.1 (d-ABq, â-proton);
1.1 (3H, t, Et). Homogeneous by HPLC Luna C18, Symmetry
C8. LCMS 390 (M+1)+ , high resolution MS (M+1)+ found
390.18077 (C23H23N3O3 requires 390.181740).
Bis-a m in e S4 Un it s b y Solid -P h a se Met h od ology
(Sch em e 2). The S4 component (bis-1,4-aminomethylcyclo-
hexane) was supported on 2-chlorotrityl resin (1.2 mmol/g) and
coupled with an Fmoc protected amino acid using TBTU/

PRO_SELECT: A Tool for Rapid Lead Discovery
J ournal of Medicinal Chemistry, 2002, Vol. 45, No. 6 1231
DIPEA in DMF. The washed resin was treated with 20%
piperidine in DMF to remove the Fmoc protection and then
reacted with 3-amidinobenzoic acid TFA salt using DIPCI/
HOBt in DMF. The product was then cleaved off the washed
resin using 10% triethylsilane in TFA ,and the crude product
obtained was purified by preparative reverse-phase HPLC and
isolated as the TFA salt. Compounds 5 and 6 were made by
this route.
3-Am id in oben zoyl-glycin yl-(4-a m in om eth ylcycloh exy-
l)m eth yla m in e, 5. 1H NMR (CD3CN/D2O) mixture of cis/trans
isomers, major isomer only: δ 8.15 (s, 1H, Ar); 8.1 (d,1H, Ar);
7.9 (d, 1H, Ar); 7.66 (t, 1H, Ar); 4.97 (s, 2H, “Gly CH2”); 3.0 (d,
2H, amide CH2); 2.70 (d, 2H, amine CH2); 1.70 (m, 4H,
cyclohexyl); 1.40 (m, 3H, cyclohexyl; 0.90 (m, 3H, cyclohexyl).
Homogeneous by HPLC Luna C18, Symmetry C8. LCMS
346 (M+1)+, high resolution MS (M+1)+ found 346.22378
(C18H27N5O2 requires 346.22426).
3-Am id in oben zoyl-D-p h en ylglycin yl-(4-a m in om eth yl-
cycloh exyl)m eth yla m in e, 6. 1H NMR (D2O) mixture of
cyclohexyl cis and trans isomers 8.09 (1H, s); δ 8.05 (1H, d, J
) 7.5 Hz); 7.90 (1H, d, J ) 7.5 Hz); 7.66 (1H, t, J ) 7.5 Hz);
7.43 (5H, m); 5.47 (1H, s); 3.05 (2H, m); 2.78 (2H, m); 1.48
(7H, m); 0.86 (3H, m), Homogeneous by HPLC Luna C18,
Symmetry C8. LCMS 422 (M+1)+ , high resolution MS (M+1)+
found 422.25548 (C24H31N5O2 requires 422.25556).
(1H,m); 2.85 (1H,m); 2.35 (1H,m), Homogeneous by HPLC
Luna C18, Symmetry C8. LCMS 476 (M+1)+, high resolution
MS (M+1)+ ) 476.18340 (C26H26N5O2 requires 476.18529).
Amidino compounds such as 9 were prepared initially using
the solution-phase method II described above to give Boc-N-
protected S4 intermediates. The Boc protection was removed
using TFA/DCM, and the resulting amine reacted with 1,3-
bis-tert-butoxycarbonyl methyl pseudothiourea in the presence
of mercury II chloride to give the bis-butoxycarbonyl guanidines.
Final treatment with TFA/DCM and purification by prepara-
tive HPLC gave the final products isolated as TFA salts
(Scheme 4).
3-Am id in oben zoyl-D-p h en ylglycin e 1-Am id in op ip er i-
d in -4-yleth yl Ester , 9. 3-Amidinobenzoyl-D-phenylglycine
1-Boc-piperidin-4-ylethyl ester was prepared using the general
solution-phase method described above for compounds 7 and
8 and then treated with 25% TFA in DCM to remove the Boc
protec-
tion. Treatment with 1,3-bis-tert-butyloxycarbonylmethylthio-
pseudourea (1 equiv), TEA (3 equiv), and mercury(II) chloride
(1 equiv) in DMF overnight followed by extraction into ethyl
acetate, washing with 2 N aqueous sodium hydroxide and
water, and drying (MgSO4) gave 3-amidinobenzoylphenylgly-
cine 1-(1,3-bis-tert-butyloxycarbonyl amidine)-4-piperidin-4-
ylethanol ester.
Solu tion P h a se II (Sch em e 3). Boc-D-Phenylglycine was
coupled to an S4 component in DMF using HATU or TBTU/
DIPEA or, alternatively, EDCI or DIPCI with HOBt or HOAt
as an additive. When the S4 component was an alcohol,
catalytic DMAP was added. The products thus obtained were
deprotected using TFA in DCM and then coupled to 3-amidi-
nobenzoic acid TFA salt using DIPCI/HOBt in DMF. The
products were purified by reverse-phase preparative HPLC
and isolated as the TFA salts. Compounds 7 and 8 were made
by this route.
The above compound was treated with 25% TFA in DCM
until the Boc protection was removed and then evaporated in
vacuo. The residue was purified by preparative RPHPLC to
give 3-amidinobenzoyl-D-phenylglycine 1-amidinopiperidin-4-
ylethyl ester. 1H NMR (D2O) 8.17 (1H, m); δ 8.07 (1H, d); 7.93
(1H, d); 7.70 (1H, t); 7.45 (5H, m); 5.60 (1H,s); 4.25 (2H,m);
3.55 (2H,m); 2.75 (2H, m); 1.60 (4H, m); 1.25 (1H, m); 1.00
(2H, m). Homogeneous by HPLC Luna C18, Symmetry C8.
LCMS451 (M+1)+ , high resolution MS (M+1)+ ) 451.244439
(C24H30N6O3 requires 451.245725).
1-(3-Am id in oben zoyl-D-p h en ylglycin yl)-4-ben zoylp ip -
er id in e, 7. To a solution of Boc-D-phenyl glycine (251 mg, 1
mmol) and a mixture of DMF (1 mL) and DCM were added
4-benzoylpiperidine (339 mg 1.5 mmol), DIPEA (348 µL, 2
mmol), and TBTU (353 mg 1.1 mmol). After being stirred at
room temperature overnight, the mixture was partitioned
between ethyl acetate (6 mL) and 10% hydrochloric acid (2
mL). The organic layer was washed with 10% hydrochloric acid
(2 mL), saturated aqueous sodium bicarbonate, and then brine.
Evaporation of solvent gave the crude product which was taken
up in dichloromethane (2 mL) and treated with trifluoroacetic
acid (2 mL) until removal of the Boc group was complete.
Solvent was evaporated in vacuo, and the residue was taken
up in ethyl acetate and washed with saturated aqueous sodium
bicarbonate and then brine before evaporating to dryness. The
residue was dissolved in DMF (5 mL), and to this was added
a mixture of HOAt (150 mg, 1.1 mmol), 3-amidinobenzoic acid
TFA salt (300 mg, 1.08 mmol), and DIPCI (180 µL, 1.15 mmol),
and the mixture was stirred overnight. Any solids were
removed by filtration, and solvent was removed in vacuo. The
residue was taken up in ethyl acetate, washed with saturated
aqueous sodium bicarbonate, dried (MgSO4), and evaporated
in vacuo. The residue was converted to the TFA salt by
addition and evaporation of 25% TFA in acetonitrile and then
dissolved in a minimum of aqueous acetonitrile for purification
by preparative RPHPLC to give 1-(3-amidinobenzoyl-D-phe-
nylglycinyl)-4-benzoylpiperidine TFA salt (159 mg, 27% over
three steps). 1H NMR (DMSO-d6) δ 8.40 (2H, m); 8.10 (1H, d);
7.70 (1H,t); 7.50 (10H, m); 5.55 (1H, s); 3.60 (1H, m); 2.5 (2H,
m); 1.00 (6H,m). Homogeneous by HPLC Luna C18, Symmetry
Biology. Inhibition of factor Xa was assessed at room
temperature in 0.1 M phosphate buffer, pH 7.4, according to
the method of Tapparelli et al.29 Purified human factor Xa was
purchased from Alexis corporation, Nottingham, U.K. Chro-
mogenic substrate pefachrome-FXA was purchased from Pen-
tapharm AG, Basel, Switzerland. Product (4-nitroaniline) was
quantified by absorption at 405 nm in 96-well plates using a
Dynatech MR5000 reader (Dynex Ltd, Billingshurst, U.K.). Km
and Ki were calculated using SAS PROC NLIN (SAS Institute,
Cary, NC, Release 6.11). Km values were determined as 100.9
µM for factor Xa/pefachrome-FXA. Inhibitor stock solutions
were prepared at 40 mM in dimethylsulfoxide and tested at
500 µM, 50 µM, and 5 µM. Accuracy of Ki measurements was
confirmed by comparison with Ki values of known inhibitors
of factor Xa.
Cr ysta lliza tion , Da ta Collection , a n d Refin em en t.
Bovine trypsin (Sigma, Type III) was further purified by ion-
exchange chromatography (Mono S, 0.1 M sodium phosphate
pH 6.0, eluted with a 0-1 M sodium chloride gradient). A
complex of the purified trypsin with compound 7 was prepared
by incubating a 3-fold molar excess of the inhibitor with the
enzyme, which was then concentrated to 15 mg/mL in 0.05 M
Tris pH 8, 3 mM calcium chloride, 18% acetonitrile, and 5%
DMF. Crystals were grown by vapor diffusion against a well
containing 2.1 M ammonium sulfate and 0.05 M Tris pH 8.15.
Nucleation of crystal growth required streak seeding using
low-density crystals grown in the presence of benzamidine
according to the procedure of Batunik (1989).30 Crystals of the
bovine trypsin-compound 7 complex belonged to space group
P212121, with a ) 60.08 Å, b ) 63.83 Å, and c ) 70.04 Å, and
diffracted to beyond 2.0 Å. A complete native data set,
comprising 17 272 unique reflections in the range 30-2.0 Å
and with an average redundancy of 4.0, was collected at 100
K using station PX7.2 of the Daresbury SRS synchrotron
(wavelength 1.488 Å). These data were processed using
DENZO and SCALEPACK,31 and the structure solved by
molecular replacement using AMORE32 with the coordinates
from PDB entry 3PTN as search model. The structure was
refined using iterative cycles of simulated annealing refine-
C8. LCMS 469 (M+1)+ , high resolution MS (M+1)+
)
469.22282 (C28H28N4O3 requires 469.223935).
1-(3-Am id in ob en zoyl-D-p h en ylglycin yl)-4-ch lor op h e-
n ylp ip er a zin e TF A Sa lt, 8. 1-(3-Amidinobenzoyl-D-phenylg-
lycinyl)-4-chlorophenylpiperazine TFA salt was prepared from
4-chlorophenylpiperazine in a manner similar to that described
above. 1H NMR (CD3CN) δ 8.05 (1H, s); 8.00 (1H, d); 7.87 (1H,
d); 7.55(1H, t); 7.31 (5H, m); 7.08,(2H,d); 6.75,(2H,d); 5.95 (1H,
s); 3.70 (1H,m); 3.55 (2H,m); 3.45 (1H,m); 3.12 (1H,m); 3.00

1232 J ournal of Medicinal Chemistry, 2002, Vol. 45, No. 6
Liebeschuetz et al.
ment with X-PLOR33 and manual rebuilding using O.34 The
final model has good geometry and an Rcryst of 17.8% and Rfree
of 24.0% (calculated with data in the range 15-2.0 Å). The
coordinates have been deposited in the PDB (reference code
1eb2).
(13) Brandstetter, H.; Ku¨hne, A.; Bode. W.; Huber, R.; von der Saal,
W.; Wirtensohn, K.; Engh, R. A. X-ray Structure of Active Site-
inhibited Clotting Factor Xa. J . Biol. Chem. 1996, 271 (47),
29988-29992.
(14) Murray, C. W.; Clark, D. E.; Auton, T. R.; Firth, M. A.; Li, J .;
Sykes, R. A.; Waszkowycz, B.; Westhead, D. R.; Young, S. C.
PRO_SELECT: Combining structure-based drug design and
combinatorial chemistry for rapid lead discovery. 1. Technology.
J . Comput.-Aided Mol. Des. 1997, 11, 193.
(15) Klebe, G. J . The Use of Composite Crystal-field Environments
in Molecular recognition and the de Novo Design of Protein
Ligands. J . Mol. Biol. 1994, 237, 212.
(16) Hahn, M. Receptor Surface Models. 1. Definition and Construc-
tion. J . Med. Chem. 1995, 38, 2080.
Ack n ow led gm en t. The authors thank Allen Miller
for encouragement and for helpful suggestions during
the preparation of this manuscript.
Refer en ces
(1) (a) Wiley, M. R.; Fisher, M. J . Small-molecule direct thrombin
inhibitors. Expert Opin. Ther. Pat. 1997, 7 (11), 1265-1282. (b)
Menear, K. Progress towards the discovery of orally active
thrombin inhibitors. Curr. Med. Chem. 1998, 5, 457-468. (c)
Rewinkel, J . B. M.; Adang, A. E. P. Strategies and progress
towards the ideal orally active thrombin inhibitor. Curr. Pharm.
Des. 1999, 5, 1043-1075. (d) Zhu, B.-Y.; Scarborough, R. M.
Recent advances in inhibitors of factor Xa in the prothrombinase
complex. Curr. Opin. Cardiovasc., Pulm. Renal Invest. Drugs
1999, 1 (1), 63-88. (e) Walenga, J . M.; J eske, W. P.; Hoppen-
steadt, D.; Kaiser, B. Factor Xa inhibitors: Today and beyond.
Curr. Opin. Cardiovas., Pulm. Renal Invest. Drugs 1999, 1 (1),
13-27. (f) Al-Obeidi, F.; Ostrem, J . A. Factor Xa inhibitors.
Expert Opin. Ther. Pat. 1999, 9 (7), 931-953.
(17) Baxter, C. A.; Murray, C. W.; Clark, D. E.; Westhead, D. R.;
Eldridge, M. D. Flexible Docking using Tabu Search and an
Empirical Estimate of Binding Affinity. Proteins: Struct., Funct.,
Genet. 1998, 33, 367.
(18) Bo¨hm, H.-J . The development of a simple empirical scoring
function to estimate the binding constant for a protein-ligand
complex of known three-dimensional structure. J . Comput.-
Aided Mol. Des. 1994, 8, 243.
(19) Eldridge, D. E.; Murray, C. W.; Auton, T. R.; Paolini, G. V.; Mee,
R. P. Empirical scoring functions: I. The development of a fast
empirical scoring function to estimate the binding affinity of
ligands in receptor complexes. J . Comput.-Aided Mol. Des. 1997,
11, 425-445.
(20) Murray, C. W.; Auton, T. R.; Eldridge, M. D. Empirical Scoring
Functions. II. The testing of an empirical scoring function for
the prediction of ligand-receptor binding affinities and the use
of Bayesian Regression to improve the quality of the model. J .
Comput.-Aided Mol. Des. 1998, 12, 503.
(21) Li, J .; Murray, C. W.; Waszkowycz, B.; Young, S. C. Targeted
Molecular Diversity in Drug Discovery - Integration of Structure-
Based design and Combinatorial Chemistry. Drug Discovery
Today 1998, 3 (3), 105-112.
(22) Kick, E. K.; Roe, D. C.; Skillman, A. G.; Liu, G.; Ewing, T. J . A.;
Sun, Y.; Kuntz, I. D.; Ellman, J . A. Structure-based design and
combinatorial chemistry yield low nanomolar inhibitors of
cathepsin D. Chem. Biol. 1997, 4, 297-307.
(23) Bo¨hm, H.-J .; Banner, D. W.; Weber, L. Combinatorial docking
and combinatorial chemistry: Design of potent non-peptide
thrombin inhibitors. J . Comput.-Aided Mol. Des. 1999, 13, 51-
56.
(24) Wei, A.; Alexander, R. S.; Duke, J .; Ross, H.; Rosenfeld, S. A.;
Chang, C.-H. Unexpected Binding Mode of Tick Anticoagulant
Peptide Complexed to Bovine Factor Xa. J . Mol. Biol. 1998, 283,
147-154.
(25) The strain energies quoted here are generally much higher than
the associated calculated binding energy. This is because they
are calculated by different methods, the strain energy arising
out of estimates, derived using the ‘Clean’ force field. Therefore
they cannot be compared and are used independently from one
another in the process of ranking the substituents.
(26) Copyright 1995, BIOSYM/Molecular Simulations, San Diego.
(27) Copyright 1990-1994, MDL Information Systems, Inc. San
Leandro, CA. All Rights Reserved.
(28) MDL Information Systems, Inc. San Leandro, CA. All Rights
Reserved.
(29) Tapparelli, C.; Metternich, R.; Ehrardt, C.; Zurini, M.; Claeson,
G.; Scully, M. F.; Stone, S. R. In Vitro and In Vivo Characteriza-
tion of a Neutral Boron-containing Thrombin Inhibitor. J . Biol.
Chem. 1993, 268, 4734-4741.
(30) Bartunik, H. D.; Summers, L. J .; Bartsch, H. H. Crystal structure
of bovine b-trypsin at 1.5 Å resolution in a crystal form with
low molecular packing density. J . Mol. Biol. 1989, 210, 813-
828.
(31) Otwinoski, Z.; Minor, W. Processing of X-ray diffraction data
collected in oscillation mode. Methods Enzymol. 1996, 276, 307-
326.
(32) Collaborative Computational Project, Number 4. The CCP4
Suite: Programs for Protein Crystallography. Acta Crystallogr.
1994, D50, 760-763.
(33) Brunger, A. T. 1992 X-PLOR Manual Version 3.1.
(34) J ones, T. A.; Zou, J .-Y.; Cowan, S. W.; Kjeldgaard, M. Improved
methods for building protein structures in electron-density maps
and the location of errors in these models. Acta Crystallogr. 1991,
A47, 110-119.
(2) Chi, L.; Rogers, K. L.; Uprichard, A. C. G.; Gallagher, K. P. The
therapeutic potential of novel anticoagulants. Expert Opin.
Invest. Drugs 1997, 6 (11), 1591-1622.
(3) Morishima, Y.; Tanabe, K.; Terada, Y.; Hara, T.; Kunitada, S.
Antithrombotic and Hemorrhagic Effects of DX-9065a, a Direct
and Selective Factor Xa: Comparison of a Direct Thrombin
Inhibitor and Antithrombin III-Dependent Anticoagulants.
Thromb. Haemostasis 1997, 78, 1366-1371.
(4) Gold, H. K.; Torres, F. W.; Garabedian, H. D.; Werner, W.; J ang,
I.; Khan, A.; Hagstrom, J . N.; Yasuda, T.; Leinbach, R. C.;
Newell, J . B.; Bovill, E. G.; Stump, D. C.; Collen, D. Evidence
for a Rebound Coagulation Phenomenon after Cessation of a
4-hour Infusion of a Specific Thrombin Inhibitor in Patients with
Unstable Angina Pectoris. J . Am. Coll. Cardiol. 1993, 21, 1039-
1047.
(5) Galemmo, R. A.; Maduskuie, T. P.; Dominguez, C.; Rossi, K. A.;
Knabb, R. M.; Wexler, R. R.; Stouten, P. F. W. The de novo design
and synthesis of cyclic urea inhibitors of Factor Xa: Initial SAR
studies. Bioorg. Med. Chem. Lett. 1998, 8, 2705-2710.
(6) Klein, S. I.; Czekaj, M.; Gardner, C. J .; Guertin, K. R.; Cheney,
D. L.; Spada, A. P.; Bolton, S. A.; Brown, K.; Colussi, D.; Heran,
C. L.; Morgan, S. R.; Leadley, R. J .; Dunwiddie, C. T.; Perrone,
M. H.; Chu, V. Identification and Initial Structure-Activity
Relationships of a Novel Class of Nonpeptide Inhibitors of Blood
Coagulation. J . Med. Chem. 1998, 41, 437-450.
(7) Dominguez, C.; Duffy, D. E.; Han, Q.; Alexander, R. S.; Galemmo,
R. A.; Park, J . M.; Wong, P. C.; Amparo, E. C.; Knabb, R. M.;
Luettgen, J .; Wexler, R. R. Design and Synthesis of Potent and
Selective 5,6-fused Heterocyclic Thrombin Inhibitors. Bioorg.
Med. Chem. Lett. 1999, 9, 925-930.
(8) Padmanabhan, K. P.; Tulinsky, A.; Park, C. H.; Bode, W.; Huber,
R.; Blankenship, D. T.; Cardin, A. D.; Kiesel, W. Structure of
Human Des(1-45) Factor Xa at 2.2 Å Resolution. J . Mol. Biol.
1993, 232, 947-966.
(9) Hara, T.; Yokoyama, A.; Ishihara, H.; Yokoyama, Y.; Nagahara,
T.; Iwamoto, M. DX-9065a, a New Synthetic, Potent Anticoagu-
lant and Selective Inhibitor for Factor Xa. Thromb. Haemostasis
1994, 71 (3), 314-319.
(10) Hirayama, F.; Koshio, H.; Taniuchi, Y.; Sato, K.; Hisamichi, N.;
Sakai, Y.; Katayama, N.; Kawasaki, T.; Matsumoto, Y.; Yanag-
isawa, I. Abstracts of Papers, 214th National Meeting of the
American Chemical Society, Las Vegas, NV, 1997; American
Chemical Society: Washington, DC, 1997; MEDI049.
(11) Yamazaki, M.; Asakura, H.; Aoshima, K.; Saito, M.; J okaji, H.;
Uotani, C.; Kumbashiri, I.; Morishita, E.; Ikeda, T.; Matsuda,
T. Protective Effects of DX-9065a, an Orally Active Novel
Synthesized and Selective Inhibitor of Factor Xa, Against
Thromboplastin-Induced Experimental Disseminated Intravas-
cular Coagulation in Rats. Sem. Thromb. Hemostasis 1996, 22
(3), 255-259.
(12) Sato, K.; Taniuchi, Y.; Hirayama, T.; Koshio, H.; Matsumoto,
Y.; Iizumi, Y. Comparison of the Anticoagulant and Antithrom-
botic Effects of YM-75466, a novel orally-Active Factor Xa
Inhibitor and warfarin in Mice. J pn. J . Pharmacol. 1998, 78,
191-197.
J M010944E