Synthesis of novel nonpeptidic thrombin inhibitors

Article abstract of DOI:10.1002/(SICI)1522-2675(20000510)83:5<855::AID-HLCA855>3.0.CO;2-9

A novel class of nonpeptidic, active, and selective thrombin inhibitors has resulted from X-ray-structure-based design and subsequent improvement of the initial lead molecules. These inhibitors possess a bi- or tricyclic central core structure with attached side chains to reach the three binding pockets (selectivity S1 pocket, distal D pocket, and proximal P pocket) present in the active site of the enzyme. The key step in the preparation of these compounds is the 1,3-dipolar cycloaddition between an azomethine ylide, prepared in situ by the decarboxylative method from an aromatic aldehyde and an α-amino acid, with an N-substituted malcimide (e.g., see Schemes 1 and 2). All potent inhibitors contain an amidinium residue in the side chain for incorporation into the S1 pocket, which was introduced in the last step of the synthesis by a Pinner reaction. The compounds were tested in biological assays for activity against thrombin and the related serine protease trypsin. The first-generation lead compounds (±)-11 and (±)-19 (Scheme 1) with a bicyclic central scaffold showed K(i) values for thrombin inhibition of 18 μM and 0.67 μM, respectively. Conformationally more restricted second- generation analogs (Scheme 2) were more active ((±)-22i: K(i) = 90 nM (Table 1)); yet the selectivity for thrombin over trypsin remained weak. In the third-generation compounds, a small lipophilic side chain for incorporation into the hydrophobic P pocket was introduced (Schemes 7 and 8). Since this pocket is present in thrombin but not in trypsin, an increase in binding affinity was accompanied by an increase in selectivity for thrombin over trypsin. The most selective inhibitor (K(i) = 13 nM, 760-fold selectivity for thrombin over trypsin; Table 2) was (±)-1 with an i-Pr group for incorporation into the P pocket. Optical resolution of (±)-1 (Scheme 9) provided (+)-1 with a K(i) value of 7 nM and a 740-fold selectivity, whereas (-)-1 was 800-fold less active (K(i) = 5.6 μM, 21-fold selectivity). The absolute configuration of the stronger-binding enantiomer was assigned based on the X-ray crystal structure of the complex formed between thrombin and this inhibitor. Compound (+)-1 mimics the natural thrombin substrate, fibrinogen, which binds to the enzyme with the Ph group of a phenylalanine (piperonyl in (+)-1) in the distal D pocket, with the i-Pr group of a valine (i-Pr in (+)-1) in the proximal P pocket, and with a guanidinium side chain of an arginine residue (phenylamidinium group in (+)-1) in the selectivity S1 pocket of thrombin. A series of analogs of (±)-1 with the phenylamidinium group replaced by aromatic and aliphatic rings bearing OH or NH₂ groups (Schemes 10-14) were not effectively bound by thrombin. A number of X-ray crystal-structure analyses of free inhibitors confirmed the high degree of preorganization of these compounds in the unbound state. Since all inhibitors prefer similar modes of association with thrombin, detailed information on the strength of individual intermolecular bonding interactions and their incremental contribution to the overall free energy of complexation was generated in correlative binding and X-ray studies. The present study demonstrates that defined mutations in highly preorganized inhibitors provide an attractive alternative to site-directed mutagenesis in exploring molecular-recognition phenomena at enzyme active sites.

Full text of DOI:10.1002/(SICI)1522-2675(20000510)83:5<855::AID-HLCA855>3.0.CO;2-9

Helvetica Chimica Acta ± Vol. 83 (2000)
855
Synthesis of Novel Nonpeptidic Thrombin Inhibitors
by Ulrike Obst, Patrick Betschmann, Christian Lerner, Paul Seiler, and François Diederich*
Laboratorium für Organische Chemie der Eidgenössischen Technischen Hochschule, ETH-Zentrum,
Universitätstrasse 16, CH-8092 Zürich
and Volker Gramlich
Laboratorium für Kristallographie der Eidgenössischen Technischen Hochschule, ETH-Zentrum,
Sonneggstrasse 5, CH-8092 Zürich
and Lutz Weber, David W. Banner, and Peter Schönholzer
Pharma Division, Präklinische Forschung, F. Hoffmann-La Roche AG, CH-4070 Basel
A novel class of nonpeptidic, active, and selective thrombin inhibitors has resulted from X-ray-structure-
based design and subsequent improvement of the initial lead molecules. These inhibitors possess a bi- or tricyclic
central core structure with attached side chains to reach the three binding pockets (selectivity S1 pocket, distal D
pocket, and proximal P pocket) present in the active site of the enzyme. The key step in the preparation of these
compounds is the 1,3-dipolar cycloaddition between an azomethine ylide, prepared in situ by the
decarboxylative method from an aromatic aldehyde and an a-amino acid, with an N-substituted maleimide
(
e.g., see Schemes 1 and 2). All potent inhibitors contain an amidinium residue in the side chain for
incorporation into the S1 pocket, which was introduced in the last step of the synthesis by a Pinner reaction. The
compounds were tested in biological assays for activity against thrombin and the related serine protease trypsin.
The first-generation lead compounds (Æ)-11 and (Æ)-19 (Scheme 1) with a bicyclic central scaffold showed K
i
values for thrombin inhibition of 18 mm and 0.67 mm, respectively. Conformationally more restricted second-
generation analogs (Scheme 2) were more active ((Æ)-22i: K ˆ 90 nm (Table 1)); yet the selectivity for thrombin
i
over trypsin remained weak. In the third-generation compounds, a small lipophilic side chain for incorporation
into the hydrophobic P pocket was introduced (Schemes 7 and 8). Since this pocket is present in thrombin but
not in trypsin, an increase in binding affinity was accompanied by an increase in selectivity for thrombin over
trypsin. The most selective inhibitor (K
i
ˆ 13 nm, 760-fold selectivity for thrombin over trypsin; Table 2) was
(
(
Æ)-1 with an i-Pr group for incorporation into the P pocket. Optical resolution of (Æ)-1 (Scheme 9) provided
‡)-1 with a K
i
value of 7 nm and a 740-fold selectivity, whereas (À)-1 was 800-fold less active (K ˆ 5.6 mm, 21-
i
fold selectivity). The absolute configuration of the stronger-binding enantiomer was assigned based on the X-
ray crystal structure of the complex formed between thrombin and this inhibitor. Compound (‡)-1 mimics the
natural thrombin substrate, fibrinogen, which binds to the enzyme with the Ph group of a phenylalanine
(
piperonyl in (‡)-1) in the distal D pocket, with the i-Pr group of a valine (i-Pr in (‡)-1) in the proximal
P pocket, and with a guanidinium side chain of an arginine residue (phenylamidinium group in (‡)-1) in the
selectivity S1 pocket of thrombin. A series of analogs of (Æ)-1 with the phenylamidinium group replaced by
2
aromatic and aliphatic rings bearing OH or NH groups (Schemes 10 ± 14) were not effectively bound by
thrombin. A number of X-ray crystal-structure analyses of free inhibitors confirmed the high degree of
preorganization of these compounds in the unbound state. Since all inhibitors prefer similar modes of
association with thrombin, detailed information on the strength of individual intermolecular bonding
interactions and their incremental contribution to the overall free energy of complexation was generated in
correlative binding and X-ray studies. The present study demonstrates that defined mutations in highly
preorganized inhibitors provide an attractive alternative to site-directed mutagenesis in exploring molecular-
recognition phenomena at enzyme active sites.

8
56
Helvetica Chimica Acta ± Vol. 83 (2000)
1. Introduction. ± Thrombin is a trypsin-like serine protease that is central in the
process of hemostasis and thrombosis. In the blood-coagulation cascade, it cleaves the
protein fibrinogen after specific Arg residues to give polymerizable fibrin, which is a
major constituent of blood clots. Moreover, thrombin is the main activator of platelet-
aggregation and other coagulation factors [1][2]. It is therefore an important
pharmaceutical target for prevention and treatment of thrombotic diseases. Natural
inhibitors of the enzyme include the protein hirudin from the leech Hirudo medicinalis.
Hirudin is currently the most active and selective thrombin inhibitor (inhibition
constant K ˆ 0.27 pm) [3]. In the past two decades, an intensive search for small,
i
potentially orally bioavailable synthetic inhibitors has been conducted [4] (for recent
reviews on thrombin inhibition, see [5]). X-Ray crystal structures of thrombin-
inhibitor complexes show that the enzyme is a rather rigid protein with well-defined
binding pockets in the active site [6][7], and, accordingly, structure-based design has
been intensively applied to the development of new inhibitors. Fig. 1 shows a small
selection of potent synthetic thrombin inhibitors [6][8 ± 15].
The active site of thrombin features three major binding pockets in addition to the
catalytic site with the oxy-anion hole. This is schematically shown in Fig. 2 for the
complexation of the most potent inhibitor (‡)-1 (K ˆ 7 nm) described in this report
i
[
16]. i) The specificity S1 pocket contains Asp 189 for interaction with Arg or Lys side
chains of peptidic substrates. ii) The large hydrophobic D pocket (D ˆ distal to the
catalytic site [7]) is preferentially occupied by aromatic rings that can undergo CH ´´´ p
aromatic interactions with the indole ring of Trp 215. iii) An additional smaller P pocket
(
P ˆ proximal) is positioned close to the catalytic site. This pocket is absent in trypsin;
therefore, its filling with parts of an inhibitor not only enhances the binding affinity but
also the selectivity for thrombin against trypsin. Other sub-binding sites (not shown in
Fig. 2) whose occupancy by appropriate inhibitor moieties contributes to binding
strength and selectivity were identified in recent work by researchers at Merck [17].
Among the compounds shown in Fig. 1, 2 [6] is an irreversible inhibitor that is
attacked by Ser 195 of the catalytic triad (for the classification of thrombin inhibitors,
see [4a]). Boronic acid 3 [8] is a so-called ꢀreversible covalent inhibitorꢁ that reacts with
Ser 195 to form a tetrahedral anionic adduct that resembles the transition state of
peptide hydrolysis. All other compounds in Fig. 1 are non-covalent inhibitors.
Compounds 4 ± 7 [9 ± 12] bind with their positively charged side chains into the S1

Helvetica Chimica Acta ± Vol. 83 (2000)
857
i
Fig. 1. Examples of synthetic thrombin inhibitors. Shown are also the K values for the inhibition of thrombin.
pocket. They all interact to different degrees with the D and P pockets of the enzyme
and form H-bonds with Gly 216. Since positively charged aliphatic ammonium,
guanidinium, or amidinium side chains tend to strongly limit the bioavailablity of

8
58
Helvetica Chimica Acta ± Vol. 83 (2000)
Fig. 2. Schematic representation of the bonding interactions in the complex between (‡)-1 and thrombin as
revealed by X-ray crystal-structure analysis [16]. S1 ˆ specificity pocket; D ˆ hydrophobic distal pocket; P ˆ
hydrophobic proximal pocket.
thrombin inhibitors, an intensive search for active compounds lacking such side chains
has been conducted in several laboratories. The high affinity of compounds 8 ± 10 [13 ±
15] demonstrates that thrombin can be efficiently complexed by inhibitors that direct
neutral groups, such as the 6-hydroxybenzo[b]thiophene residue in 10, into the S1
pocket.
We started in 1993 a program aimed at the de novo structure-based design of novel
nonpeptidic inhibitors of thrombin, featuring a central bi- or tricyclic (see (‡)-1 in
Fig. 2) scaffold from which side chains are directed into the S1, D, and P pockets of the
active site. The central scaffold (sometimes called a ꢀcyclic templateꢁ in medicinal
chemistry, see [18]) is readily assembled by azomethine ylide 1,3-dipolar cycloaddition
[
19 ± 21]. In two previous papers, we described in detail the design of these inhibitors
and their biological affinity [16][22]. The molecular-recognition characteristics of these
compounds at the enzyme-active site were revealed in three X-ray crystal-structure
analyses of thrombin-inhibitor complexes. Here, we describe the synthesis and
structural characterization of this new class of thrombin inhibitors. We believe that
their bi- and tricyclic scaffolds also represent attractive, multi-functional central
platforms for inhibitors of other classes of enzymes. To provide a better structural
appreciation of these new molecular shapes, the X-ray crystallographic character-
ization of several inhibitors or precursors, and their often quite intriguing crystal
packing in the absence of thrombin are described. In new developments, we describe
the synthesis and biological results for inhibitors, in which the phenylamidinium residue
in 1 is replaced by a phenolic group or aliphatic or aromatic amines.
2
. Results and Discussion. ± 2.1. Synthesis of First-Generation Inhibitors. Using
computer simulations, initially with the molecular-modeling program Insight II [23],

Helvetica Chimica Acta ± Vol. 83 (2000)
859
then in most parts of the work the F. Hoffmann-La Roche in-house program MOLOC
[
24], we identified (Æ)-11 (Scheme 1) as our first target compound. It was anticipated
that (Æ)-11 would bind to the active site of thrombin in a way similar to (Æ)-1 (Fig. 1),
with the side chains that depart from the central scaffold filling the S1, P, and D pockets
and Gly 216 of the enzyme, forming a H-bond to one of the imide O-atoms of the
scaffold [22]. For the synthesis of (Æ)-11 (Scheme 1), an azomethine ylide was
generated in situ from 4-bromophenylalanine (12) and acetone under decarboxylation
[
20][21] (for a review, see [21c]) and then treated with N-benzylmaleimide to give the
desired endo-adduct (Æ)-13 besides exo-derivative (Æ)-14.
The endo-geometry of (Æ)-13 was confirmed by an X-ray crystal-structure analysis
(
Fig. 3). The compound crystallized as a racemate in the space group P2 /n, and the two
1
enantiomers are arranged in an extended orientation to allow for efficient CÀH ´´´ p
aromatic interactions [25] between their aromatic termini. Additional stabilizing
interactions are short CÀH ´´´ O contacts (C ´´´ O distance 3.51 Š) between HÀC(19)
of one enantiomer and the imide O-atom O(9) of the second.
Scheme 1. Synthesis of First-Generation Inhibitors
a) PhMe, D, 68 h; 25% ((Æ)-13), 14% ((Æ)-14). b) HCHO, HCO
CuCN, DMF, D, 56 h; 63% ((Æ)-16), 85% ((Æ)-21). d) HCl (g), MeOH, 48, 48 h. e) NaHCO
MeOH/H
O, 658, 3 h; 57% ((Æ)-11), 70% ((Æ)-19) (steps d) ± f)). g) DMF, 908, 7 h; 47%.
2
H, 1008, 10 h; 81% ((Æ)-15), 96% ((Æ)-20). c)
3
. f) NH Cl,
4
2

8
60
Helvetica Chimica Acta ± Vol. 83 (2000)
Fig. 3. X-Ray crystalstructure (top) and crystalpacking (bottom) of (Æ)-13. Arbitrary numbering. Atomic
displacement parameters obtained at 100 K are drawn at the 30% probability level.
Eschweiler-Clarke methylation [26] of (Æ)-13 gave (Æ)-15, and Br! CN exchange
led to nitrile (Æ)-16. The conversion of nitrile (Æ)-16 to the amidinium salt (Æ)-11 was
achieved with a Pinner reaction, using NH Cl to transform the intermediate imido-ester
4
hydrochloride into the amidinium salt [27].

Helvetica Chimica Acta ± Vol. 83 (2000)
861
In a model reaction aimed at improving the yield of the 1,3-dipolar cycloaddition
reaction, we prepared the azomethine-ylide intermediate starting from 4-bromoben-
zaldehyde (17) and l-alanine and obtained the endo-trans-cycloadduct (Æ)-18 in 49%
yield (endo refers to the orientation of the bromophenyl ring and trans to the position
of this ring and the Me group on the pyrrolidine ring; Scheme 1). Compound (Æ)-18 is
formed from an anti-ylide through an exo-transition state. It is well-established that the
1,3-dipolar cycloadditions with highly reactive dipolarophiles (such as maleimides) and
azomethine ylides generated by the decarboxylative method preferentially yield trans-
substituted pyrrolidines via kinetically favored anti-ylides [21c] (see also Sect. 2.2). By
the synthetic sequence described for (Æ)-11, aminidinium salt (Æ)-19 was obtained via
methylamine (Æ)-20 and nitrile (Æ)-21. The latter compound crystallized as a racemate
Å
in the space group P1. The X-ray crystal-structure analysis (Fig. 4) showed that this
molecule adopts a more folded U-shape (compared to the more extended conforma-
tion of crystalline (Æ)-13), with the terminal aromatic rings participating in intra-
molecular CÀH ´´´ p interactions (C(13) ´´´ C(19) 3.67 Š; C(14) ´´´ C(17) 3.73 Š).
Fig. 4. X-Ray crystalstructure of ( Æ)-21. Arbitrary numbering. Atomic displacement parameters obtained at
293 K are drawn at the 50% probability level.
Enzyme assays [11][22][27] gave for (Æ)-11 a K value for thrombin inhibition of
i
1
8 mm and a selectivity for thrombin over trypsin Tr/Th (ˆ [K (trypsin)/K (thrombin)])
i
i
of 0.14, whereas (Æ)-19 showed a much higher affinity with K ˆ 0.67 mm and also a
i
better selectivity (Tr/Th ˆ 7.3). Computer modeling revealed that (Æ)-19, with the
phenylamidinium ꢀneedleꢁ directly attached to the bicyclic scaffold, had a much higher
complementarity to the active site than (Æ)-11, with an extra CH group between
2
needle and scaffold. In particular, the modeling showed that only one enantiomer ±
with the same configuration at the stereogenic centers as in (‡)-1 ± should be able to

8
62
Helvetica Chimica Acta ± Vol. 83 (2000)
occupy the active site (this was proven in later studies; see below). Based on these
findings, more extensive structure-activity relationships were explored with a second
inhibitor generation.
2
.2. Synthesis of Second-Generation Inhibitors. At this stage of the study, we were
interested in exploring changes in the substituents at the two N-atoms of the cyclic
scaffold, as well as further expansions of this central platform by transformation into a
tricyclic system. The synthesis of the phenylamidinium salts (Æ)-22a ± l is shown in
1
2
3
Scheme 2; the different residues R , R , and R in (Æ)-22a ± l and their precursors in the
synthetic route are given in Table 1 together with the biological activities [22]. The
bicyclic scaffolds (Æ)-23 were prepared by the azomethine ylide 1,3-dipolar cyclo-
Scheme 2. Synthesis of the endo-trans-Phenylamidinium Chlorides (Æ)-22a ± l as Second-Generation Thrombin
Inhibitors
a) DMF, 808, 5 h. b) HCHO, HCO
2
H, 1008, 10 h. c) Ac
2
O, pyridine, 08 ! 208, 2 h. d) HCl (g), MeOH, 48, 24 h.
1 3
e) NH , MeOH, 658, 3 h. For the definition of residues R ± R , see Table 1.
3

Helvetica Chimica Acta ± Vol. 83 (2000)
863
Table 1. Activities of the Thrombin Inhibitors (Æ)-22a ± l and Selectivities with Respect to Trypsin [22]
Inhibitor
R1
R2
R3
K
i
[mm]
Selectivity
Tr/Th )
a
(
(
(
(
(
(
(
(
(
(
(
(
Æ)-22a
Æ)-22b
Æ)-22c
Æ)-22d
Æ)-22e
Æ)-22f
Æ)-22g
Æ)-22h
Æ)-22i
Æ)-22j
Æ)-22k
Æ)-22l
Me
Me
Me
Me
Me
i-Bu
i-Bu
H
Ac
H
Me
Ac
H
Bn
Bn
Bu
Bu
Bu
Bn
Bn
Bn
1.1
170
11
2.9
160
1.5
4.6
2.5
1.3
7.6
2.3
3.4
2.8
12
7.8
15
0.6
20
Me
1.0
À(CH
À(CH
À(CH
À(CH
2
2
2
2
)
)
)
)
3
3
3
4
À
À
À
À
0.22
0.09
0.35
10
0.44
b
Piperonyl )
Cyclohexylmethyl
Bn
ÀCH
2
SCH
2
À
Piperonyl
a)
K
(Trypsin)/K
(Thrombin). b) Piperonyl (ˆ(1,3-benzodioxol-5-yl)methyl) ˆ
i
i
addition starting from 4-formylbenzonitrile (24), commercial amino acids 25 (in either
optically pure or racemic form), and maleimides 26. The desired endo-trans-adducts
(see Sect. 2.1 for definitions of endo and trans) were obtained in 17 ± 71% yield. Larger
amounts of exo-trans-products were obtained in the cycloaddition reactions with the
cyclic amino acids proline (25h), piperidine-2-carboxylic acid (25k), and thiazolidine-4-
carboxylic acid (25l). The reduction of the endo-selectivity in the cycloadditions of
azomethine ylides formed from aldehydes and N-substituted amino acids by the
decarboxylation method had previously been noted in the literature [21c].
The endo-trans-geometry of the tricyclic cycloadduct (Æ)-23h obtained from proline
(
25h), 4-formylbenzonitrile (24), and N-benzylmaleimide (26a) was clearly established
by X-ray crystal-structure analysis (Fig. 5). In the solid state, (Æ)-23h adopts a folded
conformation resembling that observed in crystals of the bicyclic cycloadduct (Æ)-21
(see Fig. 4).
Eschweiler-Clarke methylation afforded compounds (Æ)-27, whereas acylation with
Ac O gave amides (Æ)-28. Starting from (Æ)-23, (Æ)-27, and (Æ)-28, the targeted
2
amidinium salts were obtained by the Pinner reaction, this time using methanolic NH
3
to transform the intermediate imidoester hydrochloride into the amidinium salt. The
amidinium salts (Æ)-22a ± l were purified either by repeated dissolution in EtOH
followed by fractional precipitation with Et O or by reversed-phase chromatography.
2
By the same methodology, some of the isolated exo-trans-cycloadducts (Æ)-29i, j, k, l
were transformed into amidinium salts ((Æ)-30i, k, l) (Scheme 3).
Biological studies with the endo-derivatives (Æ)-22a ± l (Table 1) provided the
following results:
i) Compounds with a methylated N-atom ((Æ)-22d, g and also (Æ)-19 (Scheme 1))
showed increased activity relative to the nonmethylated ones ((Æ)-22a, c, f), whereas a
dramatic reduction in binding affinity was observed with acetylated compounds ((Æ)-
22b, e). Although geometric differences cannot be completely excluded, we believe
that the pyrrolidine N-atom is preferentially bound when protonated.

8
64
Helvetica Chimica Acta ± Vol. 83 (2000)
Fig. 5. X-Ray crystalstructure of ( Æ)-23h. Arbitrary numbering. Atomic displacement parameters obtained at
73 K are drawn at the 50% probability level.
1
Scheme 3. Synthesis of the exo-trans-Phenylamidinium Salts (Æ)-30i, k, l
3
a) HCl (g), MeOH, 48, 24 ± 48 h. b) NH , MeOH, 658, 3 h.
ii) The best affinities for thrombin (lowest K values) were achieved with the
i
addition of a five-membered ring on the right side of the molecule ((Æ)-22h ± j). A
larger six-membered ring as in (Æ)-22k strongly reduces the inhibitory activity,
probably because of the unfavorable displacement of a H O molecule localized in the
2
oxy-anion hole close to Ser 195. Thus, the change from five- to six-membered ring
fusion leads to a remarkably large, unfavorable change in binding free enthalpy of
À1
D(DG) ((Æ)-22h ! (Æ)-22k) ˆ 2.2 kcal mol . A fused thiazolidine ring (in (Æ)-22l) is
probably also slightly too big (length of a CÀS single bond: 1.82 Š, length of a CÀC
single bond: 1.53 Š [28]).

Helvetica Chimica Acta ± Vol. 83 (2000)
865
iii) Changes in the substituent R3 (Bu, PhCH , cyclohexylmethyl) on the
2
succinimide N-atom for incorporation of the D pocket have only a minor effect on
the binding affinity. Interestingly, similar K values were measured for the benzyl- ((Æ)-
i
2
2h) and the cyclohexylmethyl-substituted ((Æ)-22j) compounds, both of which
presumably adopt an edge-to-face geometry with respect to Trp 215 in the D pocket.
This finding suggests that specific electrostatic arene-arene CÀH ´´´ p interactions in
the complex of (Æ)-22h are not of particular importance; a result in good agreement
with other recent findings [29]. The highest activity in the series was achieved with a
(
1,3-benzodioxol-5-yl)methyl (ˆ piperonyl) substituent ((Æ)-22i; K ˆ 90 nm).
i
The X-ray crystal structure of the complex between thrombin and (Æ)-22i at 1.9-Š
resolution [22] confirmed the predictions from computer modeling (Fig. 6). Only the
(
3aS,4R,8aS,8bR)-enantiomer of the racemic mixture was found in the crystal structure,
demonstrating a high degree of chiral recognition by the enzyme. The overall binding
geometry is very similar to that shown for (‡)-1 in Fig. 2, with the large tricyclic scaffold
occupying the middle of the active site and the S1 and D pockets being filled by the
piperonyl and phenylamidinium residues, respectively. Whereas one imide CˆO group
undergoes H-bonding with Gly 216, the other one is found in the hydrophobic P pocket
of the enzyme pointing onto the p-surface of Tyr 60A. Such an orthogonal orientation
between CˆO group and aromatic ring is quite disfavored (compared to the location of
the CˆO group in the plane of an aromatic ring), as had been shown by protein X-ray
crystal-structure analyses [30]. Furthermore, the CˆO group must become desolvated
upon transfer from aqueous solution into the P pocket, which does not contain H-bond
donors. This is energetically quite unfavorable, and CˆO groups in the interior of
proteins almost always (> 98%) participate in H-bonding interactions [31][32]. This
loss in desolvation energy is not required for the binding of the inhibitor to trypsin,
because it lacks the loop that forms the P pocket in thrombin. This could explain why
Fig. 6. Schematic representation of the interactions of (3aS,4R,8aS,8bR)-22i with thrombin, according to the
X-ray crystalstructure [22]

8
66
Helvetica Chimica Acta ± Vol. 83 (2000)
this second generation class of inhibitors is only slightly more selective (up to 20-fold;
Table 1) for thrombin than for trypsin.
Whereas the exo-derivative (Æ)-30i (K ˆ 19 mm) was a much weaker thrombin
i
inhibitor than the endo-isomer (Æ)-22i (K ˆ 90 nm), the opposite selectivity was
i
observed for the pair of endo/exo-diastereoisomers (Æ)-22k and (Æ)-30k (Table 1 and
Scheme 3). Computer modeling was not of much help in proposing a satisfactory
binding mode for the exo-isomers.
2
.3. Synthesis of Racemic Third-Generation Inhibitors with Strongly Enhanced
Binding Selectivity for Thrombin over Trypsin. Based on structure-activity data
(
Sect. 2.2), we had identified the unfavorable binding of the ꢀupperꢁ imide CˆO group
into the hydrophobic P pocket as the origin of the weak selectivity of the second-
generation inhibitors (see Sect. 2.2). Therefore, we targeted a new series of compounds
in which this CˆO group was either reduced to a CH moiety or substituted by a
2
suitably sized lipophilic group to fill the P pocket.
2
.3.1. Transformation of Imide into Lactam Inhibitors. Initial attempts to reduce the
imide to a g-hydroxy lactam with NaBH at pH 8 ± 10, as described in the literature for
4
succinimide [33], failed. Much more successful was the reduction to the g-hydroxy
lactam with Li[Et BH] (ꢀsuper hydrideꢁ) [34]. Subsequent reduction to the corre-
3
sponding lactam was readily accomplished with Na[BH CN] in CF COOH. The latter
3
3
reaction presumably involves protonation of the OH group and elimination to an
acyliminium ion, and subsequent reduction to the lactam [35]. Starting from the bromo
derivatives (Æ)-20 and (Æ)-31, lactams (Æ)-32 and (Æ)-33, respectively, could be
prepared with a remarkably high regioselectivity (for reduction of the upper CˆO
group); however, the subsequent Br! CN exchange led to rapid aromatization of the
pyrrolidine to a pyrrole ring (Scheme 4).
Scheme 4. Attempted Synthesis of Phenylamidinium Inhibitors with a Lactam Ring
a) Li[Et
3
BH], THF, À788 ! 08, 20 min. b) Na[BH
3 3
CN], CF COOH, 208, 1.5 h. c) CuCN, DMF, D.

Helvetica Chimica Acta ± Vol. 83 (2000)
867
To prevent aromatization, we introduced an additional Me group. 1,3-Dipolar
cycloaddition gave (Æ)-35, and subsequent N-methylation yielded imide (Æ)-34. The
introduction of the additional Me group in (Æ)-34 (as compared to (Æ)-20), however,
led to a loss in regioselectivity in the reduction of the imide moiety, giving the
regioisomeric lactams (Æ)-36 and (Æ)-37 in 49% and 31% yield, respectively.
Compounds (Æ)-36 and (Æ)-37 were separated by chromatography and transformed
via nitriles (Æ)-38 and (Æ)-39 into the amidinium salts (Æ)-40 and (Æ)-41 (Scheme 5).
For comparison purposes (see below), the phenylamidinium salt (Æ)-42 was also
prepared from (Æ)-34 via (Æ)-43.
The constitution of lactam (Æ)-38 was established by X-ray crystal-structure
analysis (Fig. 7,a). A superimposition with the crystal structure of imide (Æ)-21
(
Fig. 7,b) showed that the conformation of the bicyclic scaffold does not change
significantly upon reduction of the imide. Thus, differences in binding affinity between
inhibitors containing imide and lactam scaffolds should mostly originate from the
additional CˆO group in the imide and not be due to structural changes in the bicyclic
skeleton. Compound (Æ)-38 crystallized as a racemate in the space group P2 /c; the
1
arrangement of the enantiomers in the crystal leads to a specific overlap of their
piperonyl rings (Fig. 7,c), presumably as a result of favorable compensation of the
C(aryl)ÀO dipoles.
2
.3.2. Preparation of Inhibitors with Lipophilic Residues for Incorporation into the P
Pocket of Thrombin. We first attempted the replacement of the upper imide CˆO
group in (Æ)-34 by an alkyl group using a Grignard reagent (MeMgCl) [33b]. The
crude product was subsequently reduced with Na[BH CN] in CF COOH [36] to
3
3
provide, in excellent yield and complete regioselectivity, the methylated lactam (Æ)-44
Scheme 6). This specific nucleophilic attack at the ꢀlowerꢁ CˆO group of imide (Æ)-34
is in contrast to the preferred reduction of the ꢀupperꢁ CˆO group with ꢀsuper hydrideꢁ
(
(
Schemes 4 and 5). This could suggest that the ꢀlowerꢁ imide CˆO group is the
intrinsically more reactive one, and that steric factors may be irrelevant in the attack of
the small MeMgCl reagent. Lactam (Æ)-44 was subsequently transformed via nitrile
(
Æ)-45 into the amidinium salt (Æ)-46.
The configuration of (Æ)-44 with the endo-Me group at C(6) was established by a
nuclear Overhauser effect (NOE) observed between HÀC(6) and HÀC(3a)
(
300 MHz, CDCl ). It is conceivable that addition of the Grignard reagent is followed
3
by protonation of the resulting tertiary alcohol and elimination of H O to the
2
acyliminium ion, which is then reduced from the sterically less hindered exo-face.
Grignard addition to (Æ)-20 with only one Me group near the ꢀupperꢁ CˆO group,
followed by reduction, also provided the endo-methylated lactam. Utilization of MeLi
instead of the Grignard reagent expectedly did not affect the stereochemical outcome
of the transformation of (Æ)-34.
Since organometallic reagents preferentially added to the ꢀlowerꢁ imide CˆO
group, we turned to further transformations of g-hydroxy lactam (Æ)-47 (Scheme 7) in
order to introduce the lipophilic residue for incorporation into the P pocket.
Compound (Æ)-47 was readily obtained as a mixture of a major and a minor
configurational isomer (no assignment was made) by reduction of (Æ)-34 with
Li[Et BH] (Sect. 2.3.1). Attempts to form the corresponding acyliminium ion with a
3
Lewis acid (BF ´ OEt [37], SnCl4 [38], or CF COOH [39]) and to add allyl
3
2
3

8
68
Helvetica Chimica Acta ± Vol. 83 (2000)
Scheme 5. Preparation of the Lactam Inhibitors (Æ)-40 and (Æ)-41
a) HCHO, HCO
2
H, 1008, 10 h; 92%. b) Li[Et
3
BH], THF, À788 ! 08, 1 h. c) Na[BH
3 3
CN], CF COOH, 208, 5 h.
, MeOH, 658, 3 h.
d) CuCN, DMF, D, 17 ± 30 h. e) HCl (g), MeOH, 48, 24 h. f) NH
3
trimethylsilane to give the allyl-substituted lactam were unsuccessful. Similarly,
transformation of (Æ)-47 into the methanesulfonate and subsequent reaction with
MeLi did not provide the methylated lactam.
Finally, hydroxy lactam (Æ)-47 could be transformed in good yield with 4-
toluenesulfinic acid in the presence of CaCl into sulfone (Æ)-48 (Scheme 7). Starting
2

Helvetica Chimica Acta ± Vol. 83 (2000)
869
Fig. 7. a) X-Ray crystalstructure of ( Æ)-38. Arbitrary numbering. Atomic displacement parameters obtained at
95 K are drawn at the 30% probability level. b) Superimposition of the X-ray crystalstructures of ( Æ)-38 and
Æ)-21. c) View onto (left) and parallel (right) to the stacking piperonylrings of the two enantiomers of ( Æ)-38.
2
(
from this versatile intermediate, we took advantage of a protocol by Ley and co-
workers [40], who described that the conversion of Grignard reagents with ZnCl2
generates organometallic reagents suitable for the nucleophilic substitution of the
arylsulfone group. The exact nature of the reagent is unknown; however, the presence
of Mg salts is essential for the successful displacement of the sulfone by an alkyl residue.
By this methodology, the alkylated derivatives (Æ)-49a ± d were prepared in 75%
(attack of primary alkyl groups) and 30 ± 35% yield (attack of bulkier secondary alkyl
groups). In each case, the reaction proceeded with retention of configuration,
providing the exo-stereoisomer. The configuration at the newly created stereogenic
center was readily established by NOE difference spectroscopy. A plausible explan-

8
70
Helvetica Chimica Acta ± Vol. 83 (2000)
Scheme 6. Preparation of the Lactam Inhibitor (Æ)-46
a) MeMgCl, THF, 208, 13 h, b) Na[BH
3
CN], MeOH, CF
3
COOH, 208, 2 h. c) CuCN, DMF, D, 30 h. d) HCl (g),
, MeOH, 658, 3.5 h.
MeOH, 48, 24 h. e) NH
3
Scheme 7. Synthesis of the Alkylated Inhibitors (Æ)-51a ± d
a) Li[Et
3
BH], THF, À788 ! 08, 2 h. b) CH
2
Cl
2
, CaCl
2
, 208, 19 h. c) ZnCl
2
, RMgCl or RMgBr, CH
2
Cl
2
, 08 ! 208,
13 ± 36 h. d) CuCN, DMF, D, 17 ± 30 h. e) MeOH, HCl (g), CHCl
3
, 48, 24 h. f) NH
3
, MeOH, 658, 3.5 h.
ation for this stereochemical outcome involves the intermediate formation of the
acyliminium salt, which is attacked from the exo-face. The alkylated derivatives (Æ)-
4
9a ± d were subsequently transformed into nitriles (Æ)-50a ± d and eventually into the
targeted amidinium salts (Æ)-51a ± d.

Helvetica Chimica Acta ± Vol. 83 (2000)
871
Crystallization of amidinium salt (Æ)-51c from H O afforded crystals suitable for X-
2
Å
ray analysis. The compound crystallized as a racemate in the space group P1 together
with 1 equiv. of H O. Two similar conformers I and II ± each in racemic form ± were
2
present in the crystal; they differed mainly in the orientation of their piperonyl rings
(
Fig. 8). The dihedral angle between the planes of the phenyl ring and the attached
amidinium moiety amounted to 298 in both conformers.
Fig. 8. X-Ray crystalstructure of ( Æ)-51c. Shown are the two conformers I and II, each present in racemic form
in the crystal. Arbitrary numbering. Atomic displacement parameters obtained at 295 K are drawn at the 30%
probability level.

8
72
Helvetica Chimica Acta ± Vol. 83 (2000)
The crystal packing was quite complicated in view of the presence of two
conformers. Basically, the piperonyl rings arrange in hydrophobic layers whereas the
À
amidinium moieties form hydrophilic layers which also contain the Cl counterions and
H O molecules (Fig. 9). A network of ionic H-bonds, involving the amidinium groups,
2
À
the Cl counterion, the H O molecules, and the lactam CˆO groups extends
2
throughout the crystal.
Fig. 9. View onto the crystalpacking of ( Æ)-51c featuring the H-bonding network within the hydrophilic layer.
À
The lighter spheres in the network are Cl ions, the darker ones are H
2
O molecules. H-Bonding distances
(
X ´´´ Y) are given in Š.
We then turned to the substitution of the ꢀupperꢁ imide CˆO group by alkyl
residues in the tricyclic, higher-affinity series. Reduction of (Æ)-31 proceeded
selectively, yielding only one of the four possible isomeric g-hydroxy lactams, which
was transformed directly ± without further purification ± into sulfone (Æ)-52
(
Scheme 8). Larger quantities of this compound are conveniently purified by simple
washing of the crystalline crude product obtained after workup with AcOEt. In this
solvent, (Æ)-52 is quite insoluble, whereas the impurities dissolve readily. Lactams (Æ)-
53a ± e were formed from (Æ)-52 with the corresponding Grignard reagents in the

Helvetica Chimica Acta ± Vol. 83 (2000)
873
Scheme 8. Synthesis of Inhibitors (Æ)-55a,b,d,e and 1 with a Tricyclic Central Scaffold
a) Li[Et
RMgBr, CH
3
BH], THF, À788 ! 08, 30 min. b) 4-Toluenesulfinic acid, CH
Cl
, 08 ! 208, 13 ± 36 h. d) CuCN, DMF, D, 17 ± 30 h. e) MeOH, HCl (g), CHCl
MeOH, 658, 3.5 h.
2
Cl
2
, CaCl
2
, 208, 24 h. c) ZnCl
2
, RMgCl or
2
2
3 3
, 48, 24 h. f) NH ,
presence of ZnCl , with a particularly high yield (90%) being obtained with PhMgBr.
2
The conversions with the larger Grignard reagents exclusively provided the exo-products
(
Æ)-53b ± e, whereas conversion with EtMgBr afforded an isomeric mixture from which
the desired exo-isomer could be separated only after conversion to the nitrile. Bromides
Æ)-53a ±e were transformed without significant aromatization (see Scheme 4) into
(
nitriles (Æ)-54a ±e and finally to the amidinium salts (Æ)-55a,b,d,e and (Æ)-1. The exo-
orientation of the i-Pr group in nitrile (Æ)-54c was supported by NOE difference
spectroscopy; furthermore, this configuration was also confirmed for phenylamidinium
salt (Æ)-1 by the X-ray crystal structure of its complex with thrombin [16].
With their rigid bi- and tricyclic cores, these inhibitors are all highly preorganized
for binding (as shown by the X-ray crystal structures and computer modeling), and it is
reasonable to assume that they all prefer similar modes of association to thrombin.
Measured changes in the binding free enthalpy were, therefore, expected to correlate
directly with the bonding contributions of individual substituents of the inhibitors,
thereby providing valuable information on the magnitudes of individual intermolecular
interactions. The affinities of the lactam inhibitors for thrombin and the selectivities for
thrombin over trypsin are given in Table 2.

8
74
Helvetica Chimica Acta ± Vol. 83 (2000)
Table 2. Activities of the Lactam-Based Thrombin Inhibitors and Selectivities with Respect to Trypsin [16]. The
activities of the two imide inhibitors (Æ)-42 and (Æ)-22i are given for comparison. Also shown is the stabilization
or destabilization DDG of the thrombin-inhibitor complex by replacement of one CˆO group in the two imide
inhibitors.
Inhibitor
K
i
[mm]
Selectivity
Tr/Th )
a
DDGb)
(
(
(
(
(
(
(
(
Æ)-42
0.5
2.0
9.2
6.2
2.0
0.095
0.030
0.86
2.2
32
4.3
5
32
77
Æ)-41
‡ 0.8
‡ 1.7
‡ 1.5
‡ 0.8
À 1.0
À 1.7
‡ 0.3
Æ)-46
Æ)-40
Æ)-51a
Æ)-51b
Æ)-51c
Æ)-51d
223
38
DDGc)
(
(
(
(
(
(
Æ)-22i
Æ)-55a
Æ)-55b
Æ)-1
0.09
0.0081
0.010
0.013
1.7
7.8
210
230
760
2.6
À 1.4
À 1.3
À 1.1
‡ 1.7
‡ 1.6
Æ)-55d
Æ)-55e
1.4
3.6
a)
)
K
(Trypsin)/K
(Thrombin). b) DG((Æ)-42) À DG(x); x: other inhibitors with a bicyclic core in Table 2.
i
i
c
DG((Æ)-22i) À DG(x); x: other inhibitors with a tricyclic core in Table 2.
Inhibitors lacking the ꢀlowerꢁ CˆO group ((Æ)-41, (Æ)-46) are less active than imide
(
Æ)-42, since they are unable to form a H-bond to the backbone NH of Gly 216 (see
Fig. 2). A comparison between the inhibitory affinities of imide (Æ)-42 and lactam (Æ)-
À1
4
1 (Table 2) shows that this H-bond contributes only 0.8 kcal mol to the overall
binding free enthalpy. This comparison was validated by X-ray crystallography. The
crystal structures of the thrombin complexes with (Æ)-22i and (Æ)-41 (for details, see
[
16]) showed that the removal of the CˆO group and, consequently, of the H-bond to
Gly 216 did not induce a significant positional shift of the bicyclic core structure within
the protein. In both structures, the inhibitors adopt a similar position in the active site,
analogous to that shown for (Æ)-1 in Fig. 2. The free energy increment assigned to this
H-bond is in good agreement with data from Fersht et al. [41], who observed losses in
À1
binding free energy of 0.5 ± 1.5 kcal mol upon deleting uncharged H-bonds in
enzyme-substrate complexes using site-directed mutagenesis (for a more indepth
discussion of the weakness of this H-bond, see [16]). Lactam (Æ)-46 (Scheme 6), with
an endo-Me group at C(6) still showed some affinity towards thrombin and trypsin; a
different binding mode can, however, not be ruled out.
A comparison of the affinities of (Æ)-42 and (Æ)-40 indicated, quite surprisingly,
À1
that the imide bound by 1.5 kcal mol better than the lactam. Apparently, it is better to
fill the hydrophobic P pocket with an imide CˆO group ± despite its polar character ±
than to leave it empty for solvation by H O. Introduction of an exo-Me group at C(4) in
2
(
Æ)-51a led to only a modest improvement (it was known from literature that a Me
group does not sufficiently fill the P pocket [11]) in both binding affinity and selectivity,

Helvetica Chimica Acta ± Vol. 83 (2000)
875
compared to lactam (Æ)-40 (Table 2). A large improvement in both affinity and
selectivity resulted from the introduction of an Et group and, in particular, an i-Pr
group. With a K of 30 nm and a 223-fold higher affinity for thrombin over trypsin, (Æ)-
i
51c was the most potent and most selective among the inhibitors with a bicyclic core. A
cyclohexyl substituent ((Æ)-51d) is too large for optimal binding in the P pocket and
binding affinity and selectivity decrease again.
The inhibitors with a tricyclic core structure are much more active and selective
than those with a bicyclic one (Table 2). The three lactams with an exo-Et ((Æ)-55a),
exo-cyclopropyl ((Æ)-55b), and an exo-(i-Pr) group ((Æ)-1) showed high binding
affinities and selectivities. Compound (Æ)-1 (K ˆ 13 nm) displayed the highest
i
selectivity for thrombin over trypsin (760-fold) and was only a very weak binder of
the other coagulation factors Xa (K ˆ 140 mm) and VIIa (K > 500 mm). It is noticeable
i
i
that the natural thrombin substrate fibrinogen also binds to the P pocket with an i-Pr
group of a valine residue. Cyclohexyl (in (Æ)-55d) and phenyl rings (in (Æ)-55e) are
again too voluminous for energetically favorable incorporation into the P pocket.
Obviously, the loop in thrombin that forms the P pocket is rather rigid and cannot adapt
well to larger hydrophobic groups of the inhibitor. As mentioned in the introduction,
the X-ray crystal structure of the complex between thrombin and (‡)-1 was solved, and
the observed interactions between enzyme and inhibitors are schematically depicted in
Fig. 2; for a detailed discussion of this crystal structure, the reader is referred to [16].
2
.4. OpticalReso lu tion of ( Æ)-1. In crystals obtained from thrombin and (Æ)-1 (for
the experimental protocol, see [16]), the (‡)-(1R,3aS,4R,8aS,8bR)-enantiomer was
found exclusively in the enzyme-active site. To quantify the degree of enantioselectivity
in thrombin recognition, which had also been predicted by computer modeling, we
carried out the optical resolution of this most selective inhibitor.
Initial attempts to resolve (Æ)-1 by crystallization of diastereoisomeric salts formed
with a diversity of non-racemic acids such as (‡)-(1S)-camphor-10-sulfonic acid, (À)-d-
mandelic acid, or (‡)-l-tartaric acid and derivatives in a variety of solvents all failed.
Attempts to form esters from hydroxy lactam (Æ)-56 (Scheme 9) and acyl halides of
some of the mentioned non-racemic acids and to separate the diastereoisomers by
chromatography also remained unsuccessful, presumably due to the instability of the
esters. In contrast, etherification of (Æ)-56 was found to yield relatively stable products
that could be chromatographed without decomposition. Thus, crude (Æ)-56, obtained
by reduction of imide (Æ)-31, was treated with methyl (À)-d-mandelate in refluxing
toluene in the presence of a catalytic amount of pyridinium toluene-4-sulfonate (PPTS)
[
42] and the resulting diastereoisomeric ethers 57 and 58 were separated by
chromatography on SiO (Scheme 9). Since these compounds decompose slowly on
2
SiO , small quantities of intensely yellow-colored impurities, which eluted together
2
with the separated diastereoisomers, were not completely removed. All attempts to
obtain X-ray-quality crystals of 57 or 58 failed, which prevented a direct determination
of their absolute configurations. Therefore, the ethers were directly converted with
toluene-4-sulfinic acid into sulfones (‡)-52 and (À)-52, which are intermediates on the
way to the amidinium salts (‡)-1 and (À)-1, respectively. Configurational assignments
in Scheme 9 were made based on the strongly different binding affinities of (‡)-1 and
(
À)-1 (see below), and the assumption that the better-binding enantiomer is the one
seen in the X-ray crystal structure (Fig. 2).

8
76
Helvetica Chimica Acta ± Vol. 83 (2000)
Scheme 9. Synthesis of Optically Pure (‡)-1
a) Li[Et
3
BH], THF, À788 ! 08, 30 min. b) PPTS, PhMe, D, 6 h; 61% (57), 64% (58), yields starting from (Æ)-
3
1. c) Toluene-4-sulfinic acid, CaCl
2
, CH
2
Cl
2
, 208, 12 h; 91% ((‡)-52), 79% ((À)-52). d) Me
2
CHMgBr, ZnCl
2
,
,
CH
2
Cl
2
, 208, 24 h. e) CuCN, DMF, D, 27 h, (51% from (‡)-52). f) MeOH, HCl (g), CHCl
3
, 48, 24 h. g) NH
3
MeOH, 658, 3.5 h (75% from (‡)-54c). The synthesis of (À)-1 from (À)-52 followed the same protocol.
The purity of the non-racemic sulfones was determined by high performance liquid
chromatography (HPLC) on a (S,S)-Whelk-O-1 chiral stationary phase (CSP) [43]
with AcOEt/hexane 1:1 as eluent (Fig. 10). Enantiomer (À)-52 was found to be of high
purity (> 99.5%), whereas (‡)-52 (96.5%) was substantially contaminated with its
antipode (3.5%).
Since an enantiomeric purity of > 99 :1 was required for meaningful biological
assays at the stage of the amidinium salts, a further purification of (‡)-52 was necessary.
This was achieved by addition of a small quantity of racemic (Æ)-52 to a solution of (‡)-
52 in AcOEt. Upon very slow addition of hexane, nearly pure racemate crystallized out,
leaving the remaining (‡)-52 in solution with an enantiomeric purity higher than 99 :1
(
Fig. 11).
An in-depth insight into the high propensity of (Æ)-52 to crystallize even in the
presence of a large excess of the (‡)-enantiomer, was provided by X-ray crystal-
structure analysis. Sulfone (Æ)-52 crystallized as a racemate in the space group P2 /c
1
(
Fig. 12). An analysis of the crystal lattice revealed a much larger amount of favorable
contacts between opposite than between identical enantiomers. A double-layer

Helvetica Chimica Acta ± Vol. 83 (2000)
877
Fig. 10. Analytical HPLC of sulfones (À)-52 (right top, t
R
ˆ 5.5 min) and (‡)-52 (right bottom, t
R
ˆ 7.5 min) on
À1
a (S,S)-Whelk-O-1 CSP. Eluent: AcOEt/hexane, flow rate 1 ml min , detection at 254 nm; retention time of
non-retained solute: t
0
ˆ 3.0 min; separation factor a ˆ (tR' À t
0
)/(t
R
À t ) ˆ 1.8. The chromatogram on the left
0
corresponds to racemic (Æ)-52.
structure (Fig. 13,a) features segregated stacks of the (‡)- and the (À)-enantiomer,
with the Ts groups providing the boundaries of each double layer. In this layered
structure, each molecule is surrounded by four molecules of opposite configuration
(
Fig. 13,b).
A closer analysis reveals the nature of the intermolecular contacts in the crystal
lattice. Fig. 14,a, shows the segregated stacks of the enantiomers in a double layer.
Between two identical enantiomers, aromatic CÀH ´´´ p interactions exist between the
Ts and the piperonyl groups, with distances between C-atoms of the two rings of 3.66 Š
(
C(33) ´´´ C(23)) and 3.75 Š (C(14) ´´´ C(22); (for the numbering, see Fig. 12). Between
two opposite enantiomers, a very short CÀH ´´´ p interaction (distance C(18) ´´´ C(23):
.45 Š) is observed between the methyl group of the Ts residue and the piperonyl
group.
Several additional contacts exist between opposite enantiomers in adjacent double
layers (Fig. 14,b). There is an aromatic CÀH ´´´ p interaction (distance C(17) ´´´ C(22):
.72 Š) between Ts and piperonyl rings of opposite enantiomers in adjacent double
layers. A remarkably short contact (distance 3.34 Š) is observed between the Ts
3
3

8
78
Helvetica Chimica Acta ± Vol. 83 (2000)
Fig. 11. Analytical HPLC showing the purification of (‡)-52 by crystallization of (Æ)-2 as an insoluble racemate.
For the conditions, see caption to Fig. 10.
Fig. 12. X-Ray crystalstructure of ( Æ)-52. Arbitrary numbering. Atomic displacement parameters obtained at
295 K are drawn at the 50% probability level.

Helvetica Chimica Acta ± Vol. 83 (2000)
879
Fig. 13. a) Crystalpacking of ( Æ)-52. The two enantiomers form segregated stacks (perpendicular to the paper
plane) that arrange in a double-layer structure. The (‡)-enantiomer is shown in darker shading. b) Schematic
representation of the contacts between opposite enantiomers in the crystalpacking.

Fig. 14. a) Intermolecular contacts in the segregated stacks of one double layer of the crystal packing of (Æ)-52. b) Intermolecular contacts between opposite enantiomers
of the same and of adjacent double layers. View a) corresponds to a view from the left side onto the crystal packing shown in Fig. 13,a; view b) is identical to that in Fig. 13,a.

Helvetica Chimica Acta ± Vol. 83 (2000)
881
C-atom C(16) and the piperonyl O-atom O(4), and an additional contact (distance
3
.36 Š) exists between O-atoms of the SO moieties. Aromatic CÀH ´´´ p interactions
2
are also seen between bromophenyl and piperonyl rings of opposite enantiomers within
a double layer (distances C(25) ´´´ C(54): 3.61 Š and C(25) ´´´ C(55): 3.64 Š).
Overall, these numerous contacts result in a very dense crystal packing with a
À3
remarkably high calculated density of 1.493 g cm . This explains the high crystal-
lization tendency of the racemate, and it can be assumed that a similarly favorable
packing cannot be reached in crystals containing identical enantiomers.
The enantiomerically pure sulfones (‡)-52 and (À)-52 were subsequently trans-
formed into the amidinium salts (‡)-1 and (À)-1, respectively, by the route applied for
the preparation of (Æ)-1 (Scheme 9). Purification of the intermediate nitriles (‡)-53c
and (À)-53c proved to be more difficult than that for the synthesis of racemic material.
An intensively yellow-colored impurity could not be separated by crystallization, but
required filtration through neutral Al O and activated charcoal. The amidinium salts
2
3
were conveniently purified by chromatography (SiO ; CHCl /MeOH 7:1).
2
3
Biological studies showed that the affinity of enantiomer (‡)-1 (K ˆ 7 nm) to
i
thrombin was nearly twice that of racemic (Æ)-2 and showed a 740-fold selectivity for
thrombin over trypsin. It was, therefore, 800 times more active than (À)-1 (K ˆ 5.6 mm,
i
selectivity 21). There is little doubt that the more potent enantiomer is the one found in
the crystal structure of the complex formed between thrombin and (Æ)-1 and,
consequently, the (1R,3aS,4R,8aS,8bR)-configuration is assigned to (‡)-1 (for the
numbering, see Fig. 6 ).
The enantioselectivity in the molecular recognition by thrombin has also been
investigated in other studies. Racemic 4 (ꢀNAPAPꢁ) gave a K value of 6 nm, the more
i
active d-enantiomer a K value of 2.1 nm, whereas the l-enantiomer was 670 times less
i
active (1.4 mm) [44], the extent of chiral recognition, therefore, being of similar
magnitude as the one observed for 1 in this study.
2
.5. Substitution of the Phenylamidinium Side Chain. The high affinity of inhibitors
8
± 10 (Fig. 1) demonstrates that an amidinium or guanidinium side chain is not
necessary for efficient complexation to thrombin. Therefore, we decided to substitute
the phenylamidinium side chain in the tricyclic inhibitor (Æ)-1 by a series of other
residues with a geometry complementary to the S1 pocket and undertook the
preparation of compounds (Æ)-59a ± e, (Æ)-60, and (Æ)-61.
For the preparation of phenol (Æ)-59a, we took advantage of the recent advances in
the Pd-catalyzed formation of bonds between aromatic C-atoms and heteroatoms [45 ±
4
7]. Reaction of bromide (Æ)-53c with bis(4,4,5,5-tetramethyl-1,3,2-dioxaborolanyl)
(
62) [48] in the presence of [PdCl (dppf)] (dppf ˆ 1,1-bis(diphenylphosphino)ferro-
2
cene) afforded crude boronic ester (Æ)-63, which was converted without further
purification with H O [49] to phenol (Æ)-59a (Scheme 10).
2
2
Carboxamide (Æ)-59b was readily obtained by partial hydrolysis of nitrile (Æ)-54c
with H O (Scheme 11). Also, the chemoselective reduction of (Æ)-54c with Li[Et BH]
2
2
3
[
50] afforded benzylamine (Æ)-59c in very high yield (93%).
The synthesis of acetamide (Æ)-59d started with the 1,3-dipolar cycloaddition using
N-(4-formylphenyl)acetamide to give the endo ((Æ)-64) and exo ((Æ)-65) cycloadducts,
and was completed via the sequence (Æ)-64 ! (Æ)-66 ! (Æ)-67 ! (Æ)-59d
(
Scheme 12). Subsequent acidic hydrolysis provided the aniline (Æ)-59e. Alternatively,

8
82
Helvetica Chimica Acta ± Vol. 83 (2000)
Scheme 10. Synthesis of Phenol( Æ)-59a
a) [PdCl
2
(dppf)], KOAc, Me
2
SO, 808, 12 h. b) H
2
O
2
, H
2
O, Et
2
O, 208, 3 h; 48% (from (Æ)-53c).
Scheme 11. Synthesis of Carboxamide (Æ)-59b and Benzylamine (Æ)-59c
a) 1. H
2
O
2
, KOH, Me
2
CO, H
2
O, 208, 1 h; 2. Na
2
S
2
O
3
, H
2
O; 61%. b) Li[Et
3
BH], CH
2
Cl
2
, À788, 2 h; 93%.
(
Æ)-59e was prepared by Pd-catalyzed amination of bromide (Æ)-53c via imine
intermediate (Æ)-67 [51] (Scheme 13). Despite much experimentation and the
application of other protocols [52], the yield of this conversion could not be raised
above 17%.

Helvetica Chimica Acta ± Vol. 83 (2000)
883
Scheme 12. Synthesis of Acetamide (Æ)-59d and Aniline (Æ)-59e
a) MeCN, D, 14 h. b) Li[Et
3%. d) Me
3
BH], CH
CHMgCl, ZnCl
2
Cl
2
, À788, 2 h; 96%. c) 4-Toluenesulfinic acid, CaCl
2
, CH
2
Cl
2
, 208, 3 d;
9
2
2
, CH Cl , 208, 40 h, 60%. e) 2n HCl, D, 16 h; 60%.
2
2
Scheme 13. Alternative Synthesis of Aniline (Æ)-59e
a) [Pd(OAc)
2
], (‡)-(R)-2,2'-bis(diphenylphosphino)-1,1'-binaphthalene ((‡)-(R)-BINAP), Cs
008, 5 d. b) 2n HCl, THF, 208, 30 min; 17% (from (Æ)-53c).
2 3
CO , PhMe,
1
For the preparation of the cis- and trans-cyclohexylamines (Æ)-60 and (Æ)-61,
aniline (Æ)-59e was hydrogenated in the presence of PtO as catalyst [53][54]. The
2
reaction was highly chemoselective, since neither hydrogenolysis of the N-piperonyl
residue nor hydrogenation of its aromatic ring were observed, and the mixture of (Æ)-
6
0 and (Æ)-61 was obtained in 66% yield (Scheme 14). Separation of the two
diastereoisomeric pairs of enantiomers, however, was exceedingly difficult. Similarly,

8
84
Helvetica Chimica Acta ± Vol. 83 (2000)
Scheme 14. Synthesis of cis- and trans-Cyclohexylamines (Æ)-60 and (Æ)-61
a) H
2
(2 ! 4 bar) PtO
2
, 4 d; 67%. b) (CF
3
CO)
2
O, CH
2
Cl
2
, 208, 2 h, then chromatography (SiO
2 2 2
; CH Cl /MeOH
97 :3); 25% (Æ68) and 32% ((Æ)-69). c) K
2
CO , MeOH, H
3
2
O, D, 3 h; 99% ((Æ)-60) and 68% ((Æ)-61).
attempts to separate an N-Boc-protected (Boc ˆ (tert-butoxy)carbonyl) mixture of
(
Æ)-60 and (Æ)-61 ((Boc) O, Et N, MeOH; 93% [55]) remained unsuccessful.
2
3
Eventually, the trifluoroacetamides (Æ)-68 and (Æ)-69 were prepared [56], which
could be separated chromatographically by using a very large excess of SiO .
2
Subsequent hydrolysis afforded the desired pure cyclohexylamines (Æ)-60 and
(
Æ)-61.
1
The configuration of (Æ)-60 and (Æ)-61 was first assigned with the help of H-NMR
1
1
spectroscopy, including { H, H}-COSY-2D-NMR spectra. Eventually, crystals of
trifluoroacetamide (Æ)-69 suitable for X-ray analysis were obtained by diffusion of
pentane vapors into a solution of the compound in CHCl . The compound crystallized
3
as a racemate in the space group P2 /c, and the X-ray crystal-structure analysis
1
(
Fig. 15,a) confirmed the trans-configuration, which had been predicted by the
1
H-NMR spectroscopic analysis. The crystal lattice shows a layered structure with two
opposite enantiomers forming two H-bonds (N ´´´ O distance ˆ 2.86 Š) between the
NÀH groups of their trifluoroacetamide moieties and the CˆO groups of their
pyrrolidone rings (Fig. 15,b).