Limiting the Number of Potential Binding Modes by Introducing Symmetry into Ligands: Structure-Based Design of Inhibitors for Trypsin-Like Serine Proteases

Article abstract of DOI:10.1002/chem.201503534

In the absence of X-ray data, the exploration of compound binding modes continues to be a challenging task. For structure-based design, specific features of active sites in different targets play a major role in rationalizing ligand binding characteristic

Full text of DOI:10.1002/chem.201503534

DOI: 10.1002/chem.201503534
Full Paper
&
Medicinal Chemistry
Limiting the Number of Potential Binding Modes by Introducing
Symmetry into Ligands: Structure-Based Design of Inhibitors for
Trypsin-Like Serine Proteases
Norbert Furtmann,[a, b] Daniela Häußler,[a] Tamara Scheidt,[a] Marit Stirnberg,[a]
Torsten Steinmetzer,[c] Jürgen Bajorath,[b] and Michael Gütschow*[a]
Abstract: In the absence of X-ray data, the exploration of
compound binding modes continues to be a challenging
task. For structure-based design, specific features of active
sites in different targets play a major role in rationalizing
ligand binding characteristics. For example, dibasic com-
pounds have been reported as potent inhibitors of various
trypsin-like serine proteases, the active sites of which con-
tain several binding pockets that can be targeted by cationic
moieties. This results in several possible orientations within
the active site, complicating the binding mode prediction of
such compounds by docking tools. Therefore, we introduced
symmetry in bi- and tribasic compounds to reduce confor-
mational space in docking calculations and to simplify bind-
ing mode selection by limiting the number of possible
pocket occupations. Asymmetric bisbenzamidines were used
as starting points for a multistage and structure-guided opti-
mization. A series of 24 final compounds with either two or
three benzamidine substructures was ultimately synthesized
and evaluated as inhibitors of five serine proteases, leading
to potent symmetric inhibitors for the pharmaceutical drug
targets matriptase, matriptase-2, thrombin and factor Xa.
This study underlines the relevance of ligand symmetry for
chemical biology.
are available,[6] docking represents a fast and cost-efficient ap-
proach for the identification and optimization of bioactive
compounds, compared to high-throughput screenings or the
synthetic generation and biological evaluation of compound
libraries.
Introduction
Structure-based design is a popular strategy in rational drug
development and constitutes a growing research area within
the last decades in academia as well as in industrial settings.
Computational methods in chemical biology and structure-
based drug design are widely used in all stages of drug discov-
ery, including hit-identification and lead-optimization.[1,2]
Among them, molecular docking is one of the most used tech-
niques and requires the availability of a three-dimensional
target structure. It has been reported that docking is more fre-
quently applied than ligand-based methods in virtual screen-
ing investigations, and increasing numbers of docking studies
have been published within recent years.[3,4] Although the
amount of publicly available X-ray structures is increasing rap-
idly,[5] and high-throughput X-ray crystallography techniques
Within the docking process, the first step comprises the
sampling of possible ligand conformations, owning to the
ligand conformational flexibility and its orientation in the
target active site (posing). In a second step, scoring functions
evaluate the predicted ligand conformations, and selected
poses are ranked according to their estimated free-energy
scores.[1] Despite being a complex and challenging process,
considering the rotational freedom of even small organic mole-
cules, docking programs were found to predict compound ori-
entations similar to crystallographically determined binding
modes.[7] However, scoring functions are not capable of pre-
dicting reliable binding affinities and therefore the selection of
“true” binding modes from a pool of possible conformations
and the ranking of compounds still constitutes a major limita-
tion of docking calculations.[1,2,4,7,8] Therefore chemical intu-
ition, detailed knowledge about the target and the availability
of experimentally determined binding modes of related com-
pounds are crucial parameters for the evaluation of docking
results.
[a] N. Furtmann, D. Häußler, T. Scheidt, Dr. M. Stirnberg, Prof. Dr. M. Gütschow
Pharmaceutical Institute, Pharmaceutical Chemistry I
University of Bonn, An der Immenburg 4, 53121 Bonn (Germany)
E-mail: guetschow@uni-bonn.de
[b] N. Furtmann, Prof. Dr. J. Bajorath
Department of Life Science Informatics, B-IT
LIMES Program Unit Chemical Biology and Medicinal Chemistry
University of Bonn, Dahlmannstr. 2, 53113 Bonn (Germany)
[c] Prof. Dr. T. Steinmetzer
When dealing with targets having open and largely solvent
exposed binding sites the identification of meaningful ligand
conformations is even more difficult because the conforma-
tional space (with respect to the binding site) is less restricted
Institute of Pharmaceutical Chemistry
Philipps University Marburg, Marbacher Weg 6, 35032 Marburg (Germany)
Supporting information for this article is available on the WWW under
http://dx.doi.org/10.1002/chem.201503534.
Chem. Eur. J. 2016, 22, 610 – 625
610
ꢀ 2016 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim

Full Paper
and often allows the formation of several plausible complexes.
The situation is even more complex when distinct pockets
within a given binding site have comparable features, leading
to the assumption that a certain inhibitor substructure might
address each of those subsites with a comparable probability.
In this scenario, not only the binding affinities predicted by
docking tools for possible ligand conformations are of ques-
tionable value, also knowledge-driven interpretations reach
their limitations.
flat and fused with the S4 subsite forming a distal binding
area, also referred to as S3/S4 pocket. In case of matriptase
and matriptase-2, this distal pocket is relatively open, whereas
due to the side chains of the residues Ile174 (thrombin) and
Phe174 (factor Xa) it is more closed and well defined in both
clotting proteases. Especially factor Xa has a deep and hydro-
phobic S4 pocket formed by residues Tyr99, Phe174 above the
highly conserved Trp215. In thrombin, Tyr99 and Phe174 are
exchanged by the aliphatic residues Leu99 and Ile174. Resi-
dues Trp215 and Phe99 (matriptase) or His99 (matriptase-2)
shape the lower part of the S3/S4 subsite in the matriptases.
The upper part of the S3/S4 pocket constitutes a negatively
charged environment created by residue Asp97 in matriptase-2
while a phenylalanine residue at the same position in matrip-
tase closes the pocket and makes it more hydrophobic and
less solvent exposed.[9,20,21]
For example, the family of trypsin-like serine proteases con-
sists of several enzymes characterized by relatively open and
solvent exposed binding sites, among them the coagulation
proteases thrombin and factor Xa and the type II transmem-
brane serine proteases matriptase and matriptase-2. Thrombin
and factor Xa are targeted by anticoagulants for the prophylax-
is and treatment of thromboembolic complications such as
heart attack, angina pectoris and stroke.[9] The thrombin inhibi-
tor drugs melagatran and dabigatran as well as the factor Xa
inhibitors rivaroxaban, apixaban and edoxaban have been ap-
proved as first orally available drugs addressing trypsin-like
serine proteases.[9–12]
Bisbenzamidines are known to bind to the minor grooves of
adenine/thymine-rich double-strand DNA.[22,23] Moreover, benz-
amidine moieties present prominent arginine mimetics and
have often been used to address the S1 binding site of tryp-
sin-like serine proteases.[24–26] In particular, dibasic compounds
containing at least one benzamidine substructure have been
reported as potent inhibitors of trypsin, thrombin, factor Xa,
matriptase and matriptase-2, indicating that several binding
pockets within the active sites of these enzymes can be ad-
dressed by basic moieties.[14,15,20,27–39] Thus, when evaluating
docked binding conformations of dibasic inhibitors, in particu-
lar of bisbenzamidines, it is difficult to assign the respective
benzamidine to its appropriate binding subsite. This problem
cannot be definitely solved by relying on computed docking
scores even after sophisticated calculations. Noteworthy, X-ray
structures of matriptase–bisbenzamidine complexes indicated
different possible orientations of di- and tribasic com-
pounds,[34,35] further complicating knowledge-driven ap-
proaches for the selection of preferred poses. However, the ac-
curate prediction of binding modes remains an essential pre-
requisite for structure-based design and optimization strat-
egies. Therefore, we introduced symmetry in bi- and tribasic
compounds to reduce conformational space available for
design and simplify binding mode selection by limiting the
number of possible pocket occupations. Thus, the symmetry
feature of ligands was explored with respect to the chemical
biology of protein–ligand binding determinants. Bisbenzami-
dines were used as starting points for a multistage and struc-
ture-guided optimization leading to branched inhibitors for
trypsin-like serine proteases.
Matriptase is involved in the activation of several substrates
which are implicated with tumorigenesis, for example, hepato-
cyte growth factor/scatter factor, urokinase-type plasminogen
activator and protease activated receptor 2. Therefore, inhibi-
tors of matriptase might be possible anticancer agents.[13–15]
The closely related enzyme matriptase-2 plays a critical role in
human iron homeostasis by cleaving hemojuvelin, the bone
morphogenetic co-receptor at the cell surface of hepato-
cytes.[16] This leads to an interruption of the BMP/SMAD signal-
ing cascade resulting in a down-regulated expression of hepci-
din, a hepatic peptide hormone that suppresses the iron trans-
porter ferroportin, inhibits resorption of iron and the release of
iron from macrophages. Accordingly, inhibiting matriptase-2
was considered to be a promising strategy for the treatment
of iron overload diseases, including b-thalassemia.[17–20]
In the aforementioned enzymes the catalytic triad is formed
by residues His57, Asp102 and Ser195. These trypsin-like serine
proteases share a primary substrate specificity for basic amino
acids. At the bottom of their S1 pocket, a negatively charged
aspartyl side chain (Asp189, according to the chymotrypsino-
gen numbering) can interact through charge-assisted hydro-
gen bonds with positively charged moieties such as the side
chain of arginine or arginine mimetics. The S1 specificity
pocket is very similar with the exception that residue Ala190 in
thrombin, factor Xa and matriptase-2 is replaced by serine in
matriptase and trypsin. The size of the proximal S2 pocket is
determined by residue 99. The sterically more demanding
Tyr99, Phe99 and His99 contribute to a more restricted S2
pocket found in factor Xa, matriptase and matriptase-2, respec-
tively. Due to the presence of the less demanding Leu99 in
thrombin and trypsin, both proteases possess a better accessi-
ble S2 subsite. Moreover, the S2 pocket of thrombin is covered
by the Tyr60A-Pro60B-Pro60C-Trp60D segment of its character-
istic “60-insertion loop”, which results in a preference for sub-
strates containing medium-sized hydrophobic P2 residues such
as proline, valine and isoleucine. The S3 pockets are relatively
Results and Discussion
We considered a hypothetical scenario with: 1) a target bind-
ing site constituting of three similar pockets with an equal af-
finity for a certain moiety, and 2) a ligand designed around
a central fragment having several branching moieties with
complementary features to interact with these binding pock-
ets. Accordingly, multiple plausible binding orientations of this
ligand are conceivable in which each branching moiety would
potentially bind to each subsite of the target structure.
Chem. Eur. J. 2016, 22, 610 – 625
www.chemeurj.org
611
ꢀ 2016 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim

Full Paper
For example, a tetrahedral carbon atom associated with two
or three different basic residues in complex with a target struc-
ture which can accommodate basic moieties in distinct bind-
ing pockets, results in a variety of possible orientations as ex-
emplified in Table 1. The general model used to illustrate such
theoretical binding events includes a potentially chiral ligand
the four residues of which are depicted as spheres (represent-
ing R1 to R4) or H atoms. The equilibrium (Table 1, top) de-
scribes the binding event of the ligand to a target with three
similar pockets. All possible arrangements of such hypothetical
ligand–target complexes are delineated. A related “three-point
attachment” model was established to discuss the stereospeci-
ficity of enzyme–ligand interactions.[40–42]
Well knowing that such a target with three completely simi-
lar binding sites is uncommon, we nevertheless constructed
this scenario based on our experiences when docking bisbenz-
amidines to trypsin-like serine proteases, in particular to ma-
triptase and matriptase-2. Most likely and supported by several
X-ray structures, one benzamidine moiety occupies the S1
pocket forming ionic interactions with the negatively charged
aspartic acid residue at the bottom of the pocket, while the
second benzamidine group extends into the S3 and S4 sub-
sites. However, when dealing with asymmetric compounds (for
example compounds 1–4 with a peptidic backbone, Table 2) it
is difficult to conclude which of the two benzamidine residues
is directed into the S1 and the S3/S4 pockets, giving rise to dif-
ferent compound orientations.
Furthermore, our docking studies suggested an area located
above the S2 and close to the S1’ pockets of both matriptases
which is also likely to accommodate basic residues. In theory,
this leads to six possible orientations of such inhibitors within
the active site as exemplified in Table 1 (bottom, entry R1 ¼ R2;
the ligand contains two non-identical residues and two hydro-
gen atoms bound to a central carbon atom and is not chiral).
All of those six different binding modes were detected in our
docking studies on matriptase and matriptase-2 with asymmet-
ric bisbenzamidines as exemplarily shown in Figure 1A for in-
hibitor 1 (docked to a comparative model of matriptase-2).
To provide a sound basis for further structure-guided com-
pound optimization, unambiguous binding conformations are
of great importance as in silico modeling templates. Accord-
ingly, symmetric bisbenzamidines would halve the possible
binding modes to three (Table 1, entry R1 =R2).
Figure 1. A) Six different pocket occupations of the asymmetric bisbenzami-
dine 1 within the active site of a comparative model of matriptase-2, which
were observed in docking calculations. The active site is rendered using
a nontransparent surface representation (grey) and the pockets are annotat-
ed according to the Schechter and Berger nomenclature. The inset in the
upper left corner depicts a scheme of all possible orientations taken from
Table 1. Accordingly, B) depicts three different possible orientations of the
symmetric bisbenzamidine 5 docked to the same model of matriptase-2.
Before introducing symmetry into bisbenzamidine inhibitors
it was considered that such candidates could only constitute
of para/para or meta/meta substitution patterns regarding the
position of the two amidine moieties. To investigate if symmet-
ric substitution does not significantly decrease biological activi-
ty compared to mixed positions of the benzamidine residues
(i.e., meta/para or para/meta) we first synthesized bisbenzami-
dines 1–4 representing all possible substitution patterns
(Table 2). For this purpose, Boc-protected glycine was treated
with 3- or 4-cyanobenzylamine and, after deprotection, cou-
pled to 3- or 4-cyanophenylacetic acid using HATU for amide
coupling in both steps. The corresponding bisbenzonitriles
(27–30) were subsequently converted into bisbenzamidines
following the Judkins procedure,[43] by treatment with hydrox-
ylamine to obtain the corresponding hydroxyamidines 51–54,
which were acetylated with acetic anhydride, followed by a pal-
ladium-catalyzed hydrogenolysis to bisbenzamidines 1–4. To
suppress the formation of 1,2,4-oxdiazole byproducts,[44] the re-
action conditions were individually adjusted. The bisbenzami-
dines were purified using preparative HPLC and collected as
bis-hydrochlorides after lyophilization.
The biological activity of the compounds towards human
matriptase, matriptase-2 and thrombin as well as bovine factor
Xa and trypsin was evaluated in duplicate measurements with
five different inhibitor concentrations using fluorogenic sub-
Chem. Eur. J. 2016, 22, 610 – 625
www.chemeurj.org
612
ꢀ 2016 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim

Full Paper
strates. At human matriptase, compound 3 with mixed para/
meta substitution showed the highest inhibitory activity with
a Ki value of 0.945 mm followed by the fourfold less active
para/para substituted inhibitor 1, which was the strongest in-
hibitor for matriptase-2, thrombin and trypsin with Ki values of
7.92, 11.8 and 0.242 mm, respectively. Highest potency toward
factor Xa was observed for compounds 2 (meta/para) and 4
(meta/meta) with Ki values of 4.87 and 5.04 mm, respectively.
Overall, the results show that either para/para (1) or meta/
meta (4) compounds showed superior inhibition compared to
those with mixed substitution patterns (2 and 3) in four of the
five targets and still comparable inhibition in case of matrip-
tase, supporting the design of symmetric molecules.
Table 1. Possible ligand binding poses in a binding site composed of
three similar pockets.
[b]
Substituents
Chiral[a]
no
#
Arrangements
R1 =R2 =R3 =R4
1
R1 =R2 =R3
no
no
2
6
To construct symmetric bisbenzamidines, malonic acid was
selected as central core fragment. The carboxylate groups can
be converted into amides resembling a peptidomimetic back-
bone and the acidity of the central methylene structure ena-
bles the introduction of branching residues in subsequent op-
timization steps. As discussed above, unbranched symmetric
bisbenzamidines (5–12, Table 3 and Table 4) would theoretical-
ly only adopt three distinct pocket occupations compared to
the asymmetric analogues. Again, all three possible orienta-
tions were frequently detected in our docking calculations for
both matriptases as exemplarily shown for compound 5 in Fig-
ure 1B. Nevertheless, careful analysis of a high number of well-
characterized experimental complexes of trypsin-like serine
proteases indicated that several basic and dibasic inhibitors ad-
dress the S1 and the S3/S4 binding cavities,[28,34,35,45,46] strongly
supporting only one of the three detected binding modes for
a given symmetric ligand (magenta colored binding orienta-
tion in Figure 1B) and thus providing a solid template for fur-
ther structure-guided optimization efforts.
R1 =R2 ¼ R3
R1 ¼ R2 ¼ R3
yes
6
enantiomer
To investigate whether the exchange of an asymmetric pep-
tide backbone to a symmetric, malonic-acid-derived scaffold is
disadvantageous for protease inhibition, prototypic com-
pounds 5 and 6 were synthesized and their Ki values have
been determined (Table 3). Since 5 and 6 retained the activity
of their asymmetric isomers 1 and 4 (Table 2), different linker
lengths were evaluated in a next optimization step. The meth-
ylene linker in 5 and 6 was extended to ethylene in 7 and 8 or
oxyethylene in 9 and 10. All aforementioned symmetric bis-
benzamidines were synthesized by amide bond formation be-
tween malonic acid and the corresponding benzonitrile build-
ing blocks, using EDC/HOBt for bisbenzonitriles 31–33 and 35,
4-nitrophenylmalonate for 34 and HATU for 36. The bisbenzo-
nitriles were converted into corresponding bisbenzamidines 5–
10 as described above for 1–4. In general, the inhibitory activi-
ties were in the low micromolar range with the exception of
compound 5 with para substitution and a methylene linker
showing a Ki value of 0.447 mm for trypsin as well as the meta-
substituted, oxyethylene-connected inhibitor 10 with a Ki value
of 0.751 mm for factor Xa. This combination of 10 also resulted
in the strongest inhibition for matriptase and thrombin with Ki
values of 3.39 and 2.50 mm, respectively. For matriptase-2, all
para-substituted compounds (5, 7 and 9) performed equally
well irrespective of the length and linker type, while meta-sub-
stitution seemed to be less suitable.
R1 ¼ R2 ¼ R3
yes
12
racemate
R1 =R2
no
no
3
6
R1 ¼ R2
[a] Indication whether the ligand is chiral or not. [b] Total number of pos-
sible arrangements.
Chem. Eur. J. 2016, 22, 610 – 625
www.chemeurj.org
613
ꢀ 2016 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim

Full Paper
Table 2. Synthesis and biological evaluation of asymmetric bisbenzamidines 1–4.
Cpd
Arom.
subst. (left/right)
Human
matriptase Ki [mm][a]
Human
matriptase-2 Ki [mm][a]
Human
thrombin Ki [mm][a]
Bovine
factor Xa Ki [mm][a]
Bovine
trypsin Ki [mm][a]
1
2
3
4
para/para
meta/para
para/meta
meta/meta
3.76Æ0.18
8.30Æ0.28
0.945Æ0.026
9.81Æ0.50
7.92Æ0.79
14.0Æ0.95
10.7Æ0.5
57.3Æ1.8
11.8Æ0.3
19.0Æ0.4
20.1Æ1.4
37.5Æ2.6
12.2Æ1.0
4.87Æ0.14
18.9Æ1.6
5.04Æ0.19
0.242Æ0.011
0.359Æ0.012
2.03Æ0.03
12.7Æ0.3
[a] Rates were obtained from progress curves by linear regression. IC50 values were determined by duplicate measurements using five different inhibitor
concentrations and were determined by nonlinear regression using equation v=v0/(1+[I]/IC50), were v is the rate of product formation in the presence of
different inhibitor concentrations and v0 is the rate in the absence of the inhibitor. Ki values were calculated using the equation Ki =IC50/(1+[S]/Km) with
assay substrate concentrations of 10, 40, 40, 100 and 40 mm and Km values of 7.98, 32.2, 40.2, 59.0 and 22.1 mm for human matriptase, human matriptase-2,
human thrombin, bovine factor Xa and bovine trypsin, respectively.
In order to vary the type of linker with respect to different
hydrogen bond donor/acceptor patterns, inhibitors 11 and 12
were generated starting from malonic acid hydrazide and 3-
and 4-cyanobenzoylchloride, following the route to bisbenza-
midines (Table 4). Because of their unimproved activity, 11 and
12, representing aza-analogues of 7 and 8 (Table 3), were not
considered for further optimization.
quiring enantiopure materials, racemization-free syntheses or
separations of enantiomers. According to Table 1 (entry R1 =
R2 ¼ R3), the situation can be simplified by branching the cen-
tral carbon atom of a symmetric two-winged compound (e.g.
5–12) generating an achiral molecule. In our scenario, achiral
compounds with congeneric branching moieties could adopt
six different occupations within a binding site and, at the same
time, the synthetic efforts can be facilitated.
After probing the active site of the trypsin-like serine pro-
teases with several combinations of para- and meta-substitut-
ed bisbenzamidines connected by linkers of different length
and type, potential potency and selectivity improvements
should be evaluated by converting the dipterous compounds
into branched three-winged inhibitors. Following our hypo-
thetical scenario, those compounds could address all three
subsites of the target with their corresponding branching resi-
dues at the same time, while the previously described com-
pounds could occupy only two of the three pockets. However,
the number of possible pocket occupations would increase in
case of three-winged inhibitors as shown in Table 1 (entry R1 ¼
R2 ¼ R3, racemate and entry R1 ¼ R2 ¼ R3, enantiomer). Attaching
a branching residue to the central carbon atom of an asym-
metric compound, for example bisbenzamidines 1–4, results in
a chiral compound. A racemic mixture of such a compound
could adopt a total number of 12 different occupations within
the binding site, involving six different orientations of each
enantiomer. Furthermore, introducing chirality into bioactive
molecules is not only inconvenient in terms of binding mode
prediction, it also complicates the synthetic accessibility by re-
Nevertheless, before considering any knowledge-driven eval-
uations of docking results, six different orientations within
a binding site still constitute an ambiguous template for fur-
ther structure-based optimization. Thus, taking the idea of
symmetric molecules one step further, a compound carrying
three equal branching moieties at a central carbon atom,
could only adopt two different binding conformations (Table 1,
entry R1 =R2 =R3). Following our model, this can then only be
exceeded by adding a fourth equal residue to the central
carbon atom, resulting in only one pocket occupation (Table 1,
entry R1 =R2 =R3 =R4). However, a structural rationale in form
of a corresponding subsite is missing, most likely leading to
a high solvent exposure of a fourth moiety without significant
contributions to target binding. As previously discussed and
shown in Figure 1, several different pocket occupations were
observed when docking bisbenzamidines to matriptase and
matriptase-2. Frequently observed poses in our docking stud-
ies and in-depth analysis of the active site indicated that, be-
sides the S1 and S3/S4 pockets, a third area above the S2 and
S1’ subsites is likely to accommodate basic moieties. This ob-
Chem. Eur. J. 2016, 22, 610 – 625
www.chemeurj.org
614
ꢀ 2016 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim

Full Paper
Table 3. Synthesis and biological evaluation of unbranched symmetric bisbenzamidines 5–10.
Cpd
Arom.
subst.
Linker (X)
Human
matriptase Ki [mm][a]
Human
matriptase-2 Ki [mm][a]
Human
thrombin Ki [mm][a]
Bovine
factor Xa Ki [mm][a]
Bovine
trypsin Ki [mm][a]
5
6
7
8
9
10
para
meta
para
meta
para
meta
CH2
CH2
CH2CH2
CH2CH2
OCH2CH2
OCH2CH2
6.62Æ0.30
6.74Æ0.25
4.23Æ0.15
10.0Æ0.6
24.6Æ1.2
3.39Æ0.19
8.01Æ0.26
45.3Æ1.4
8.24Æ0.25
39.4Æ1.2
9.09Æ0.67
12.3Æ1.0
14.6Æ0.3
41.3Æ1.0
35.2Æ0.7
9.04Æ0.64
17.1Æ1.3
2.50Æ0.11
8.74Æ0.42
5.26Æ0.39
15.3Æ0.5
0.447Æ0.031
16.2Æ0.7
11.1Æ0.5
5.98Æ0.35
21.3Æ1.7
9.90Æ0.51
6.89Æ0.38
1.86Æ0.13
[b]
[b]
0.751Æ0.029
[a] For assay details, see Table 2. [b] X as part of the following substructure: Ar-O-CH2-CH2-NH.
Table 4. Synthesis and biological evaluation of unbranched symmetric bisbenzamidines 11 and 12.
Cpd
Arom.
subst.
Human
matriptase Ki [mm][a]
Human
matriptase-2 Ki [mm][a]
Human
thrombin Ki [mm][a]
Bovine
factor Xa Ki [mm][a]
Bovine
trypsin Ki [mm][a]
11
12
para
meta
39.6Æ3.4
30.0Æ1.7
54.3Æ3.0
13.6Æ0.6
107Æ6
9.78Æ0.80
20.2Æ1.1
9.86Æ0.35
11.1Æ0.6
16.8Æ0.5
[a] For assay details, see Table 2.
servation was recently supported by experimental complex
structures of tribasic inhibitors bound to matriptase, occupying
the described regions within the active site of the enzyme.[34,35]
Hence, we designed three-winged compounds with equal
benzamidine branching moieties and malonic acid as core
fragment. The resulting triacylmethane derivatives 13–15
(Table 5) were synthesized by a-acylation of diphenyl malonate
with phenyl chloroformate in the presence of magnesium chlo-
ride.[47] The resulting triphenyl methanetricarboxylate was then
coupled to 4-(aminomethyl)benzonitrile, 4-(2-aminoethyl)-
benzonitrile or 4-(2-aminoethoxy)benzonitrile to generate the
corresponding benzonitrile precursors, which were converted
into the trisbenzamidines 13–15.
tase, Figure 2 illustrates both possible orientations of such
a compound.
In Figure 2A the hydrogen atom at the central carbon atom
is directed away from the binding site, while in Figure 2B the
hydrogen atom points to the enzyme. In both orientations,
one benzamidine moiety occupies the S1 binding site of ma-
triptase, forming ionic interactions to the side chain of residue
Asp189 at the bottom of the pocket. The second benzamidine
residue addresses the S3/S4 pocket and might form p-interac-
tions to residues Phe99 and Trp215 as well as hydrogen bonds
to residues Phe99 and Gln175. In Figure 2A the third benzami-
dine residue is in hydrogen bonding distance to residues His57
and Ile60, while in the other orientation (Figure 2B) just one
potential hydrogen bond to the backbone carbonyl of residue
Asp60A can be observed. Both predicted binding modes repre-
sent plausible interaction patterns and all three pockets are oc-
cupied. Comparable results were obtained for the related in-
Docking 13–15 to matriptase and matripase-2 suggested
that all three binding sites (i.e., S1, S3/S4 and a third pocket
above the S2 area) can be accommodated by the benzamidine
residues. As exemplarily shown for 14 in complex with matrip-
Chem. Eur. J. 2016, 22, 610 – 625
www.chemeurj.org
615
ꢀ 2016 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim

Full Paper
Table 5. Synthesis and biological evaluation of symmetric trisbenzamidines 13–15.
Cpd
Linker (X)
Human
matriptase Ki [mm][a]
Human
matriptase-2 Ki [mm][a]
Human
thrombin Ki [mm][a]
Bovine
factor Xa Ki [mm][a]
Bovine
trypsin Ki [mm][a]
13
14
15
CH2
CH2CH2
OCH2CH2
3.11Æ0.16
0.393Æ0.004
4.01Æ0.11
3.86Æ0.27
1.46Æ0.10
4.59Æ0.28
18.8Æ0.8
10.3Æ0.2
18.9Æ1.7
5.24Æ0.34
14.7Æ0.5
15.7Æ0.8
2.18Æ0.14
4.45Æ0.14
6.98Æ0.56
[b]
[a] For assay details, see Table 2. [b] X as part of the following substructure: Ar-O-CH2-CH2-NH.
hibitors 13 and 15. However, in case of compound 14, Fig-
ure 2B shows a more relaxed fit and higher shape complemen-
tarity within the active site and therefore might represent the
preferred orientation.
sidered to be arginine mimetics for the S1 pocket and several
X-ray structures of the investigated enzymes support the occu-
pation of the S1 pocket by this residue, we introduced this cri-
terion to reduce the number of docking poses for visual in-
spection.
Primarily designed for the matriptases, the para-substituted
trisbenzamidines 13–15 showed improved inhibition at both
enzymes compared to their two-winged analogs 5, 7 and 9.
Still most Ki values were in the low micromolar range between
1.46 and 4.59 mm (for matriptase and matriptase-2), with the
exception of 14, the best benzamidine inhibitor for matriptase
(Ki =0.393 mm) of our series so far. Encouraged by the potency
improvements of branched trisbenzamidines 13–15, additional
three winged compounds were conceived carrying varying
branching residues on a bisbenzamidine backbone.
The docking results were visually analyzed for favorable
short-range interactions of the branching residues with the
corresponding active sites and their shape complementarity.
As expected, basic branching residues seemed to be preferable
in case of the matriptases, while hydrophobic residues gave
promising poses for thrombin and factor Xa. On the basis of
our visual inspection and the commercial availability of the
corresponding building blocks, we chose six branched bisbenz-
amidines for synthesis and determination of their protease-in-
hibiting properties (Table 6).
In order to evaluate diverse branching residues, a virtual li-
brary of branched compounds was created in silico and
docked into the active sites of the X-ray structures of human
matriptase, thrombin and factor Xa and into a previously de-
scribed comparative model of matriptase-2,[20] as no experi-
mental structure was accessible. For factor Xa, a human tem-
plate was selected, because no bovine X-ray structure in com-
plex with a small molecule inhibitor was publicly available and
the enzymes from both organisms share a high similarity
within the active site region. For the introduction of branching
residues, the symmetric bisbenzamidine 5 was selected for sev-
eral reasons: 1) it possessed good overall activity against all
targeted enzymes, 2) it contains a short linker, leading to less
rotatable bonds, which is an advantage for docking calcula-
tions and their interpretation, 3) the required building block, 4-
cyanobenzylamin, is easily commercially available, simplifying
the synthetic efforts in subsequent evaluation steps. The dock-
ing results for all targets were prefiltered only considering
docking poses in which the S1 pocket was occupied by a benz-
amidine moiety. Because benzamidine substructures are con-
The bisbenzonitrile precursor 31 of the symmetric bisbenz-
amidine 5 was alkylated at the acidic a-carbon atom of the
malonic acid backbone with the appropriate alkyl bromides in
the presence of sodium hydride to give the branched benzoni-
triles 42–47 (Table 6), which still constitute symmetric and achi-
ral molecules. These intermediates were then converted into
bis- and trisbenzamidines 16–21 by a route that included de-
protection in case of 18 and 21. Compound 18 was designed
as a surrogate for unfavorable interaction patterns as suggest-
ed by the docking analysis, because negatively charged and
acidic substructures seemed to be disadvantageous in each of
the three subsites of all investigated targets. As expected, the
inhibitory activity of this compound against all enzymes was
significantly decreased compared to the unbranched precursor
5 (Table 3). Furthermore, compound 16 with a benzyl side
chain could only retain the activity of 5 for factor Xa and
showed reduced inhibition of the remaining targets. Although
we predicted basic moieties as suitable branching residues to
address the S3/S4 or S2 area of the matriptases, the introduc-
Chem. Eur. J. 2016, 22, 610 – 625
www.chemeurj.org
616
ꢀ 2016 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim

Full Paper
seemed to be well-suited in our docking calculations to fit into
the characteristic S2 pocket of thrombin as well as to favorably
interact with the lower part of the hydrophobic S3 and S4
binding cavities of the matriptases and factor Xa. While com-
pound 20 carrying the 2-naphthyl substituent did not show
improved activity compared to its unbranched parent com-
pound 5, the 1-naphthyl isomer 19 comprises a particular suit-
able branching residue for thrombin and factor Xa with Ki
values of 9.78 and 1.46 mm, respectively.
In a subsequent optimization step, three new compounds
were constructed by assembling the preferred linker, benzami-
dine substitution (para or meta) and branching residue. The 2-
(piperidin-4-yl)ethyl branching residue of compound 21 was
combined with ethylene linkers attached to para-substituted
benzamidine moieties leading to 22, or with oxyethylene link-
ers and meta-substitutions providing 23 (Table 7). For both
compounds, we expected strong matriptase inhibition. Toward
matriptase-2, 22 and 23 retained the inhibitory activity of com-
pound 21. As these variations in linker type and aromatic sub-
stitution did neither cause activity differences in the un-
branched compounds (5, 7, 10; Table 3) it could be concluded,
that the linker length is not an important determinant for ac-
tivity against matriptase-2. By contrast, in case of matriptase,
the ethylene and oxyethylene derivatives 22 and 23 improved
the affinity compared to the methylene compound 21. In par-
ticular, 23 with oxyethylene linker and meta-substitution signif-
icantly increased the matriptase-inhibiting activity by more
than 30-fold (23: Ki =0.0385 mm versus 21: Ki =1.32 mm). As
most potent inhibitor for matriptase within our series, 23 also
showed a 94- and 38-fold selectivity over matriptase-2 and
thrombin as well as a twelvefold selectivity over factor Xa and
trypsin. While the only moderate activity against thrombin can
be explained by the capped S2 pocket of thrombin, which is
less favorable for basic moieties attached by extended linkers,
the high selectivity of 23 over matriptase-2 remains unclear on
the basis of our modeling studies.
However, as discussed above, binding mode prediction of
three-winged symmetric compounds carrying three congeneric
basic moieties still remains a complex task. Figure 3A and B
depict two plausible orientations of 23 within the active site of
matriptase, following the assumption that one of the two
benzamidine moieties binds to the S1 subsite, interacting with
residue Asp189 at the bottom of the pocket. According to Fig-
ure 3A, the 2-(piperidin-4-yl)ethyl branching residue occupies
the S3/S4 pocket and forms a possible hydrogen bonding in-
teraction to the backbone carbonyl of residue Phe99, while
being in a suitable position for simultaneous cation–p interac-
tions to the side chains of Phe99 and Trp215. The second
benzamidine residue is directed into the third binding area
above the S2 pocket and might be hydrogen bonded to the
backbone carbonyl of residue Ile60 as well as being well-posi-
tioned for p-stacking interactions to the side chain of Tyr60G.
In a further putative binding mode illustrated in Figure 3B, the
second benzamidine moiety of 23 occupies the S3/S4 subsite,
where it is in hydrogen bonding distance to the side chain of
residue Gln175 and might form p-interactions to residues
Phe99 and Trp215. The 2-(piperidin-4-yl)ethyl residue occupies
Figure 2. Two different possible orientations of the trisymmetric triacylme-
thane derivate 14 (cyan and magenta) in complex with matriptase are de-
picted. The surface is rendered transparent (grey) and active site residues
are shown in orange color. The insets in the upper left corners depict
a scheme of the respective orientations taken from Table 1. A) An orientation
where the hydrogen atom (white) at the central carbon atom of 14 is direct-
ed towards the solvent, while in B) this hydrogen atom faces the surface of
the enzyme.
tion of a 4-amidinobenzyl side chain did not significantly im-
prove the inhibition of both enzymes by 17 with Ki values of
6.79 and 4.98 mm for matriptase and matriptase-2, respectively.
However, the other tribasic inhibitor 21 showed a five- and
twofold enhanced inhibitory activity at matriptase and matrip-
tase-2, compared to 5, with Ki values of 1.32 and 3.71 mm, re-
spectively. Not surprisingly, the affinity of both tribasic com-
pounds (17 and 21) to the coagulation proteases thrombin
and factor Xa was weaker than that of 5. The large and lipo-
philic naphthyl branching residues of compounds 19 and 20
Chem. Eur. J. 2016, 22, 610 – 625
www.chemeurj.org
617
ꢀ 2016 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim

Full Paper
Table 6. Synthesis and biological evaluation of branched symmetric bisbenzamidines 16–21.
Cpd
Branching
residue (R’’)
Human
matriptase Ki [mm][a]
Human
matriptase-2 Ki [mm][a]
Human
thrombin Ki [mm][a]
Bovine
factor Xa Ki [mm][a]
Bovine
trypsin Ki [mm][a]
16[b]
17[c]
18[d]
12.3Æ0.2
6.79Æ0.20
287Æ17
15.1Æ0.3
4.98Æ0.37
134Æ3
21.6Æ1.4
38.3Æ1.6
210Æ16
10.1Æ0.6
4.63Æ0.24
33.4Æ2.2
81.7Æ5.9
4.80Æ0.29
35.8Æ0.6
19[b]
9.92Æ0.27
5.73Æ0.20
9.78Æ0.58
1.46Æ0.13
3.05Æ0.21
20[b]
21[e]
17.0Æ1.4
15.6Æ0.9
14.0Æ0.7
57.8Æ2.7
13.8Æ0.4
33.7Æ3.1
5.94Æ0.49
1.32Æ0.05
3.71Æ0.40
0.820Æ0.021
[a] For assay details, see Table 2. [b] R=R’=R’’. [c] R=CH2-p-C6H4-CN, R’=CH2-p-C6H4-C(NOH)NH2. [d] R=R’=CH2-p-C6H4-CO2tBu. [e] R=R’=(2-(1-tert-butox-
ycarbonyl)piperidin-4-yl)ethyl.
the third subsite, most likely forming cation-p interactions to
residue Tyr60G.
benzamidine residue of bisbenzamidines when a third basic
moiety can address the remaining pocket. However, in our pre-
dictions, the first binding mode (Figure 3A) might be preferred
because of higher shape complementarity within the binding
site and more well-defined short-range interactions compared
to the second orientation in Figure 3B.
To further illustrate the complicated situation when analyz-
ing the predicted binding modes of such compounds in com-
plex with the matriptases, superpositions with two crystallo-
graphic tribasic bisbenzamidine inhibitors of matriptase were
performed and are shown in Figure 3C and D. The experimen-
tally determined binding mode of the pyridyl bis(oxy)dibenz-
imidamide inhibitor (taken from PDB ID: 4JZI)[34] in Figure 3C
shows a similar pocket occupation as compared to the predict-
ed binding mode of 23 in Figure 3A. The two para-substituted
benzamidine moieties occupy the S1 and S2/S1’ binding sites,
while the third basic moiety, a trans-4-aminocyclohexane resi-
due, is oriented into the S3/S4 pocket. However, in the overlay
shown in Figure 3D, a co-crystallized 3,5-bis(4-carbamimidoyl-
benzyl)oxy representative of 1,3,5-trisubstituted benzene deriv-
atives (taken from PDB ID: 4O97)[35] adopts an orientation com-
parable to our prediction depicted in Figure 3B. In this case,
the meta-substituted bisbenzamidine moieties occupy the S1
and S3/S4 subsites, while the third basic residue, again a trans-
4-aminocyclohexane group, addresses the S2/S1’ area. In con-
clusion, besides the S1 pocket, both the S3/S4 as well as the
S2/S1’ subsites, have been shown to accommodate the second
Both experimentally determined binding modes of closely
related three-winged bisbenzamidines further illustrate the
complexity in binding mode prediction of such compounds
and support our strategy of designing symmetric inhibitors,
which decreases the number of possible orientations and pro-
hibits the explosion of conformational space.
The third optimized compound 24 (Table 7) represents the
best thrombin inhibitor of our series, comprising meta-substi-
tuted benzamidines, oxyethylene linker substructures and a 1-
naphthyl branching residue. It inhibited thrombin with a Ki
value of 0.0515 mm and factor Xa with 0.440 mm, while being
16- and 64-fold selective for thrombin over matriptase and ma-
triptase-2, respectively. The design of 24 as thrombin and
factor Xa inhibitor originated from a hybridization of the struc-
tures 10 (Table 3) and 19 (Table 6) which provided the pre-
ferred linker and substitution pattern as well as the favored
branching residue.
Chem. Eur. J. 2016, 22, 610 – 625
www.chemeurj.org
618
ꢀ 2016 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim

Full Paper
Table 7. Synthesis and biological evaluation of branched symmetric bisbenzamidines 22–24.
Cpd
Arom. Linker (X)
subst.
Branching
residue (R’’)
Human
Human
Human
Bovine
Bovine
matriptase Ki [mm][a] matriptase-2 Ki [mm][a] thrombin Ki [mm][a] factor Xa Ki [mm][a] trypsin Ki [mm][a]
22[b] para
23[b] meta
CH2CH2
0.743Æ0.031
5.26Æ0.24
3.61Æ0.24
45.9Æ0.9
90.9Æ1.9
8.38Æ0.63
[c]
[c]
OCH2CH2
0.0385Æ0.0001
1.46Æ0.05
0.473Æ0.013
0.491Æ0.049
24
meta
OCH2CH2
0.813Æ0.025
3.32Æ0.19
0.0515Æ0.0009
0.440Æ0.021
0.536Æ0.026
[a] For assay details, see Table 2. [b] R=R’=(2-(1-tert-butoxycarbonyl)piperidin-4-yl)ethyl. [c] X as part of the following substructure: Ar-O-CH2-CH2-NH.
Compared to its precursor 19 with para-benzamidines and
methylene linkers the thrombin-inhibiting activity of 24 was
improved by 190-fold, which can be explained by their mod-
eled binding modes (Figure 4A and B). In Figure 4A, one benz-
amidine moiety of 19 occupies the S1 pocket forming ionic in-
teractions to residue Asp189 while the second benzamidine
residue addresses the S3 and S4 binding sites and forms a po-
tential hydrogen bond to residue Glu97A. The S2 pocket ac-
commodates the lipophilic naphthyl branching residue, which
is surrounded by the aromatic side chains of residues His57,
Tyr60A and Trp60D. Furthermore, one carbonyl group of the
malonic acid core of 19 is in hydrogen bonding distance to
residue Gly216. A comparison to the putative binding mode of
24 in Figure 4B reveals a possible structural rationale for the
significant increase in potency. The oxyethylene linker and the
meta-substituted benzamidines lead to an improved fit of the
naphthyl moiety in the S2 pocket and enable the formation of
additional hydrogen bonding interactions. The malonic acid
backbone of 24 is involved in two well-defined hydrogen
bonds to residues Gly216 and Gly219. Both residues have been
described as important interaction hotspots of thrombin,
which are frequently addressed by highly active compounds.[46]
Moreover, one benzamidine is perfectly orientated in the S4
subsite. The aromatic substructure fills a hydrophobic cavity
formed by residues Leu99, Ile174, Trp215 and the amidine
function is involved in a network of hydrogen bonding interac-
tions to residues Tyr60A, Arg97 and Glu97A. Thus, the addi-
tional interactions result in an improved fit into the binding
site and provide a solid explanation for the superior potency.
Nevertheless, we detected a second plausible binding mode
for 24, where the naphthyl moiety is directed into the S3 and
S4 pockets, as observed for several thrombin inhibitors and
shown in an overlay with a crystallographic structure of
NAPAP[48] (taken from PDB ID 1DWD)[49] in Figure 4C. However,
in this orientation of 24, the S2 pocket is not occupied and the
characteristic hydrogen bonds to Gly216 and Gly219 were not
observed. The second benzamidine moiety is directed into the
S1’ pocket, but favorable interactions were not present in this
area of the model. Therefore, the first prediction in Figure 4B
seems to be more likely, as it includes several well-defined
short-range interactions, which are frequently detected in com-
plexes of related inhibitors for trypsin-like serine proteases.
In Figure 4D, a putative binding mode of 24 in complex
with factor Xa is shown, which supports the rationalization of
the reduced activity against factor Xa. Again, the S1 pocket ac-
commodates one benzamidine group. The 1-naphthyl branch-
ing residue nicely addresses the S3 and S4 subsites, being
sandwiched between the aromatic side chains of residues
Tyr99, Phe174 and Trp215. Residue Gly216 interacts via a hydro-
gen-bonding interaction with the malonic acid backbone. Be-
sides those well-defined interaction patterns, no favorable
short-range contacts of the second benzamidine moiety within
the S1’ pocket could be detected, which might explain the
ninefold weaker activity for factor Xa than for thrombin.
It is important to note that we are aware of the limitations
of binding mode predictions by docking approaches. Even
after extensive docking calculations, we did not rely on dock-
ing scores and subjected the predicted poses to careful visual
inspection. By comparing our models to experimentally deter-
mined binding modes of related inhibitors and subsequent
synthesis and biological evaluation of our compounds, we
tried to ensure high reliability for our structure-based optimiza-
Chem. Eur. J. 2016, 22, 610 – 625
www.chemeurj.org
619
ꢀ 2016 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim

Full Paper
Figure 3. A,B) Two possible orientations of compound 23 (magenta and cyan) within the active site of matriptase. The surface is rendered transparent (grey)
and active site residues are shown in orange color. C) Overlay of the first putative binding mode of 23 (magenta) with a related crystallographic inhibitor
(yellow, taken from PDB ID: 4JZI). D) Overlay of the second predicted orientation of 23 (cyan) with an experimentally determined binding mode of an related
bisbenzamidine inhibitor (green, taken from PDB ID: 4O97).
tion efforts. Symmetric inhibitors were designed to decrease
the number of possible pocket occupations and facilitate bind-
ing mode selection. However, in retrospect it has to be admit-
ted, that the malonic acid backbone in combination with in-
creasing linker length resulted in highly flexible compounds,
a drawback for the structure-based docking strategy as it ex-
tended the conformational space. In future efforts, more rigid
core fragments and linker elements could be considered to fur-
ther facilitate compound design and potentially increase the
affinity of the inhibitors.
proteases as an exemplary target class. It was proposed that
achiral, symmetric ligand structures reduce the number of pos-
sible binding poses and aid in structure-based ligand design.
Therefore, bis- and trisbenzamidines were synthesized and ki-
netically characterized as serine protease inhibitors. In a struc-
ture-guided approach with multiple optimization steps, the tri-
symmetric triacylmethane derivative 14 with three equal benz-
amidine moieties was identified as submicromolar and low mi-
cromolar inhibitor for matriptase and matriptase-2, respective-
ly. The assembling of the favorable aromatic substitution,
linker and branching patterns resulted in the symmetric,
branched bisbenzamidine 23, a nanomolar inhibitor of matrip-
tase with high selectivity over the closely related proteases
matriptase-2 and thrombin, and an at least tenfold selectivity
over factor Xa and trypsin. Moreover, compound 24 was devel-
oped for dual inhibition of thrombin (Ki =0.0515 mm) and
Conclusions
We have conceptualized a model to address ligand-target
binding events under the assumption that the target offers
three essentially equivalent subsites using trypsin-like serine
Chem. Eur. J. 2016, 22, 610 – 625
www.chemeurj.org
620
ꢀ 2016 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim

Full Paper
factor Xa (Ki =0.440 mm). In most cases, the inhibitory activities
of the compounds studied herein confirmed the structure–ac-
tivity relationships extracted on the basis of the predicted
binding modes, which serve as sound templates for further po-
tency and selectivity optimizations. The inhibitors presented
herein can be used as starting points for drug development
and are considered to be valuable tools to characterize the ad-
dressed enzymes which mainly constitute important pharma-
ceutical targets. Even though not all binding sites might be
promising templates for our strategy of symmetric compound
design, this study is expected to motivate future investigations
on the rare species of polyfunctionalized symmetric molecules
as bioactive compounds. Moreover, it supports the importance
of symmetry features for chemical biology efforts focused on
protein–ligand binding.
Experimental Section
General
For thin-layer chromatography Merck aluminium sheets, silica
gel 60 F254 were used and detection was carried out with UV light
at 254 nm. Merck silica gel (0.06–0.20 mm, 60 Š) was used for prep-
arative column chromatography and the respective eluents are
noted for each compound. Preparative HPLC was performed on
a Knauer system using a C18 column (250”20 mm, particle size
10 mm, Eurospher 100) with a mixture of either methanol and H2O
or methanol and H2O (0.1% TFA) as eluents at a flow rate of
20 mLminÀ1 (detailed methods are given for each compound).
Melting points were measured on a Büchi melting point apparatus
1
and are uncorrected. H NMR and 13C NMR spectra were recorded
on a Bruker Avance DRX-500 spectrometer at 308C with 500 MHz
1
for H and 125 MHz for 13C or on a Bruker Avance III 600 at 308C
with 600 MHz for 1H and 150 MHz for 13C (listed for each spec-
Figure 4. A) The predicted binding mode of the branched inhibitor 19 (green) within the binding site of thrombin. The surface is rendered transparent (grey)
and active site residues are shown in orange color. B) A preferred orientation of compound 24 (cyan) within the active site of thrombin is depicted; C) a
second possible binding mode of 24 (magenta) superposed with the experimental orientation of the related NAPAP inhibitor (yellow, taken from PDB ID
1DWD). D) A putative binding mode of 24 (purple) docked to factor Xa (active site residues are colored in dark green).
Chem. Eur. J. 2016, 22, 610 – 625
www.chemeurj.org
621
ꢀ 2016 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim

Full Paper
trum). Chemical shifts (d) are given in ppm referring to the signal
center using the solvent peaks for reference: [D6]DMSO 2.49/
39.7 ppm. For the determination of purity, LC-DAD chromatograms
and ESI-MS spectra recorded on an Agilent 1100 HPLC system cou-
pled with an Applied Biosystems API-2000 mass spectrometer were
used. UV absorption was detected from 220 to 400 nm using
a diode array detector. HRMS was recorded on a microOTOF-Q
mass spectrometer (Bruker) with ESI-source coupled with a HPLC
Dionex Ultimate 3000 (Thermo Scientific). Solvents and reagents
were obtained from Acros (Geel, Belgium), TCI Deutschland GmbH
(Eschborn, Germany), Bachem (Bubendorf, Switzerland), Sigma–Al-
drich (Steinheim, Germany), Alfa Aesar (Karlsruhe, Germany), Fluo-
rochem (Derbyshire, Great Britain), Activate Scientific (Prien, Germa-
ny) or Ellanova Laboratories (Hamden, USA). Enzymatic reactions of
human matriptase, human matriptase-2, human thrombin, bovine
factor Xa and bovine trypsin were monitored in 96-well plates with
flat-bottom and lid (Sarstedt, Newton, USA) on a FLUOstar Optima
fluorimeter (BMG LABTECH, Offenburg, Germany). The substrates
Boc-Gln-Ala-Arg-AMC (human matriptase-2, bovine trypsin), Cbz-
Gly-Gly-Arg-AMC (human thrombin) and Boc-Ile-Glu-Gly-Arg-AMC
(bovine factor Xa) were obtained from Bachem (Bubendorf, Swit-
zerland). Mes-d-Arg-Pro-Arg-AMC (human matriptase) was ob-
tained as described elsewhere.[33] Bovine trypsin and bovine factor
Xa were purchased from Sigma–Aldrich (Steinheim, Germany) and
human thrombin from Calbiochem (Darmstadt, Germany). The con-
ditioned medium of HEK cells expressing matriptase-2 was used as
source for human matriptase-2.[19] Human matriptase was obtained
as described elsewhere.[14]
General procedure for amide formation of malonic acid with
benzonitriles (31–33, 35): Malonic acid (1 equiv), EDC (2.4 equiv),
HOBt (2.4 equiv) and triethylamine (4.4 equiv for 31, 33 and 35,
5 equiv for 32) were dissolved in dry DMF and cooled to 08C. After
15 min 4-(aminomethyl)benzonitrile hydrochloride (2 equiv; for 31),
3-(aminomethyl)benzonitrile (2.2 equiv; for 32), 4-(aminoethyl)ben-
zonitrile hydrochloride (2 equiv; for 33) or 4-(2-aminoethoxy)ben-
zonitrile methanesulfonate (2 equiv; for 35) was added to the solu-
tion. The reaction mixture was allowed to warm to RT and was
stirred for 6 h. After evaporation of the solvent, the residue was
suspended in H2O (100 mL) and extracted with methylene chloride
(3”100 mL). The combined organic layers were washed with sat.
NaHCO3 (150 mL), 10% KHSO4 (150 mL), H2O (150 mL) and brine
(150 mL). The solvent was dried (Na2SO4) and evaporated. The resi-
due was suspended in ethyl acetate, heated to reflux and the solid
was filtered off and washed with ethyl acetate.
General procedure for amide formation of malonic acid hydra-
zide with cyanobenzoyl chlorides (37, 38): Malonic acid dihydra-
zide (1 equiv) and triethylamine (2.5 equiv for 37, 2.2 equiv for 38)
were dissolved in dry THF and cooled to 08C. 4-Cyanobenzoyl
chloride (2 equiv; for 37) or 3-cyanobenzoyl chloride (2 equiv; for
38) was dissolved in dry methylene chloride (20 mL) and added
dropwise to the cooled reaction mixture. The solvent was evapo-
rated after 1 h. The residue was suspended in ethanol (for 37) or
ethyl acetate (for 38), heated to reflux and the solid was filtered
off and washed with the corresponding solvent.
Preparation of triphenyl methanetricarboxylate and general
procedure for amide formation with benzonitriles (39, 41): Tri-
phenyl methanetricarboxylate was synthesized following a literature
procedure.[47] Diphenyl malonate (1.50 g, 5.85 mmol) and magnesi-
um chloride (0.56 g, 5.85 mmol) were suspended in acetonitril
(10 mL) and cooled to 08C. Triethylamine (1.18 g, 1.64 mL,
11.7 mmol) was added and after 10 min, phenyl chloroformate
(0.92 g, 0.74 mL, 5.85 mmol) was given to the reaction mixture. The
mixture was stirred at 08C for 1 h and afterwards for 24 h at RT.
After evaporation of the solvent, the residue was purified by
column chromatography on silica gel using petroleum ether and
ethyl acetate (3:1). The corresponding fractions were evaporated
and the residue was recrystallized from petroleum ether/ethyl ace-
tate to obtain colorless crystals (1.64 g, 75%); m.p. 164–1668C. Tri-
phenyl methanetricarboxylate (1 equiv), 4-(aminomethyl)benzoni-
trile hydrochloride (5 equiv; for 39), 4-(2-aminoethyl)benzonitrile
hydrochloride (4 equiv; for 40) or 4-(2-aminoethoxy)benzonitrile
(4 equiv; for 41) and DIPEA (5 equiv; for 39 or 4 equiv; for 40)
were dissolved in dry THF and stirred at RT for 24 h. After evapora-
tion of the solvent, the residue was purified by column chromatog-
raphy on silica gel using methylene chloride/methanol 29:1 (for 39
and 41) or 19:1 (for 40). The corresponding fractions were evapo-
rated in vacuo and the residue was suspended in methanol, fil-
tered off and washed with methanol.
Chemistry
General procedure for amide formation and Boc-deprotection of
Boc-glycin with 3- and 4-(aminomethyl)benzonitrile (25, 26):
Boc-Gly-OH (1 equiv), HATU (1 equiv) and DIPEA (2 equiv) were dis-
solved in dry DMF and stirred at RT. After 15 min, 4-(aminomethyl)-
benzonitrile hydrochloride (1 equiv; for 25) or 3-(aminomethyl)ben-
zonitrile acetate (1 equiv; for 26) was added and the mixture was
stirred, overnight, at RT. After evaporation of the solvent, the resi-
due was suspended in H2O (50 mL) and extracted with methylene
chloride (3”50 mL). The combined organic layers were washed
with sat. NaHCO3 (150 mL), 10% KHSO4 (150 mL), H2O (150 mL) and
brine (150 mL). The solvent was dried (Na2SO4) and evaporated.
The residue was purified by column chromatography on silica gel
using methylene chloride/methanol 9:1 (for 25) or 19:1 (for 26).
After evaporation of the corresponding fractions, the residue was
dissolved in ethyl acetate (10 mL) and added to a prepared solu-
tion of dry methanol (0.96 g, 1.21 mL, 30 mmol) and acetyl chloride
(2.35 g, 2.13 mL, 30 mmol) in ethyl acetate (10 mL). The solution
was stirred for 2 h and the precipitate was filtered off.
General procedure for amide formation with 25 or 26 and 3- or
4-cyanophenylacetic acid (27–30): Compounds 25 (1 equiv; for
27, 28) or 26 (1 equiv; for 29, 30), HATU (1.2 equiv) and DIPEA
(2 equiv) were dissolved in dry DMF and stirred at RT. After 15 min,
4-cyanophenylacetic acid (1 equiv; for 27 and 29) or 3-cyanophe-
nylacetic acid (1 equiv; for 28 and 30) was added and the mixture
was stirred, overnight, at RT. After evaporation of the solvent, the
residue was suspended in H2O (50 mL) and extracted with methyl-
ene chloride (3”50 mL). The combined organic layers were
washed with sat. NaHCO3 (150 mL), 10% KHSO4 (150 mL), H2O
(150 mL) and brine (150 mL). The solvent was dried (Na2SO4) and
evaporated. The residue was suspended in ethyl acetate, filtered
off and washed with ethyl acetate.
General procedure for the alkylation of bisbenzonitriles (42–50):
Bisbenzontrile 31, 33 or 36 (1.1–1.2 equiv) and sodium hydride
(1.1–1.2 equiv) were dissolved in dry THF and the mixture was
heated to 708C. One equivalent of benzyl bromide (for 42), 4-(bro-
momethyl)benzonitrile (43), tert-butyl 4-(bromomethyl)benzoate
(for 44), 1-(bromomethyl)-naphthalene (for 45 and 50), 2-(bromo-
methyl)naphthalene (for 46) or tert-butyl 4-(2-bromoethyl)piperi-
dine-1-carboxylate (47–49) was slowly added dropwise and the re-
action mixture was stirred at 708C for 4 h. After evaporation of the
solvent in vacuo, 45–50 were purified by column chromatography
as indicated. Compounds 47–49 were additionally purified using
preparative HPLC. Compounds 42–46 and 50 were suspended in
methanol, filtered off and washed with methanol. Compounds 47–
Chem. Eur. J. 2016, 22, 610 – 625
www.chemeurj.org
622
ꢀ 2016 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim

Full Paper
48 were suspended in a mixture of petroleum ether and ethyl ace-
tate (1:1) and filtered off.
monitored with an excitation wavelength of 340 nm and emission
wavelength of 460 nm.
Human matriptase-2 inhibition assay:[30] The assay buffer was
50 mm Tris-HCl, 150 mm NaCl, pH 8.0. The conditioned medium of
stably transfected HEK cells (HEK-MT2 cells) was used as a source
of matriptase-2 activity. The conditioned medium was collected
and concentrated as described previously,[19] and stored at À208C.
After thawing, it was diluted with assay buffer (1:50) and kept at
08C. A 10 mm stock solution of fluorogenic substrate Boc-Gln-Ala-
Arg-AMC in DMSO was diluted with assay buffer. The final concen-
tration of the substrate was 40 mm and of DMSO was 6%. Into
each well a varying amount of buffer (153.8–163.8 mL), a varying
amount of an inhibitor solution (0–10 mL) in H2O or in DMSO (only
for compounds 19 and 24), 11.2 mL DMSO and 10 mL of a substrate
solution (800 mm) were added and thoroughly mixed. In case of
compounds 19 and 24 the amount of buffer was kept constant
(163.8 mL) and DMSO was varied (1.2–11.2 mL). At 378C the reaction
was initiated by adding 15 mL of diluted conditioned medium and
was followed over 10 min (lex =340 nm, lem =460 nm).
General procedure for the conversion of benzonitriles to hy-
droxyamidines (51–74): The corresponding benzonitrile (1 equiv),
hydroxylamine hydrochloride (8–12 equiv) and DIPEA (8–12 equiv)
were suspended or dissolved in dry ethanol (5–20 mL) or 1-butanol
(5 mL) and heated to reflux. The reaction mixture turned into
a clear solution. After 2 h the reaction mixture was cooled to RT. If
a precipitate was formed, it was filtered off and washed with etha-
nol and the material was subjected to NMR spectroscopy and LC/
MS analysis. Otherwise, the solvent was evaporated in vacuo and
the compound was used without further purification or was puri-
fied by column chromatography on silica gel and was, in both
cases, subjected to LC/MS characterization. The corresponding
yields were not calculated because of the occurrence of insepara-
ble impurities which were not detectable in the LC/MS analysis.
General procedure for the conversion of hydroxyamidines to
benzamidines (1–12): The corresponding hydroxyamidine
(1 equiv; 51–62) was dissolved in acetic acid (5 mL). Acetic anhy-
dride (3–10 equiv) was added and the solution was stirred at RT.
After 2 h 10% Pd/C was added and the reduction was performed
under hydrogen atmosphere (3 atm) at RT for 2 h. After filtration
and evaporation of the solvent, the residue was purified by prepa-
rative HPLC, and several runs were performed. The corresponding
fractions were evaporated, dissolved in aqua dest. (10–40 mL), and
one drop of 2m HCl was added before lyophilization.
Human thrombin inhibition assay:[30] Assay buffer was 50 mm
Tris-HCl, 150 mm NaCl, pH 8.0. The enzyme stock solution (1 UmLÀ1
)
was kept at À208C, prepared in water, and diluted with assay
buffer (1:50) and kept at 08C. A 10 mm stock solution of fluorogen-
ic substrate Cbz-Gly-Gly-Arg-AMC in DMSO was diluted with assay
buffer. The final concentration of the substrate was 40 mm, of
DMSO was 6% and of human thrombin was 1.5 UmLÀ1. Into each
well a varying amount of buffer (153.8–163.8 mL), a varying amount
of an inhibitor solution (0–10 mL) in H2O or in DMSO (only for com-
pounds 19 and 24), 11.2 mL DMSO and 10 mL of a substrate solu-
tion (800 mm) were added and thoroughly mixed. In case of com-
pounds 19 and 24 the amount of buffer was kept constant
(163.8 mL) and DMSO was varied (1.2–11.2 mL). The reaction was
performed at 258C, initiated by adding 15 mL of enzyme solution
and followed over 10 min (lex =340 nm, lem =460 nm).
General procedure for the conversion of hydroxyamidines to
benzamidines (13–24): The corresponding hydroxyamidine
(1 equiv; 63–74) was dissolved in acetic acid (5 mL). Under ice-
cooling, acetic anhydride (4–8 equiv) and 10% Pd/C was added
and the reduction was performed under hydrogen atmosphere
(3 atm) for 2 h. The mixture was allowed to warm to RT during the
reaction time. After filtration and evaporation of the solvent, the
residue was purified by preparative HPLC, were several runs were
performed. The corresponding fractions were evaporated, dis-
solved in aqua dest. (10–40 mL), and one drop of 2m HCl was
added before lyophilization. For compounds 18 and 21–23, after
preparative HPLC and evaporation, the corresponding residues
were dissolved in an ice-cooled mixture of 5 mL H2O and 5 mL TFA.
After 1 h the solution was allowed to warm to RT and stirred for
another 3 h. The mixture was evaporated and the residue was dis-
solved in aqua dest. (10–40 mL), and one drop of 2m HCl was
added before lyophilization.
Bovine factor Xa inhibition assay: Assay buffer was 50 mm Tris-
HCl, 100 mm NaCl, 10 mm CaCl2, pH 8.0. The enzyme stock solution
(1 UmLÀ1) was prepared in water, stored at À808C and diluted with
assay buffer (1:50) and kept at 08C. A 20 mm stock solution of fluo-
rogenic substrate Boc-Ile-Glu-Gly-Arg-AMC·AcOH in DMSO was di-
luted with assay buffer. The final concentration of the substrate
was 100 mm, of DMSO was 6% and of bovine factor Xa was
1 UmLÀ1. Into each well a varying amount of buffer (159–169 mL),
a varying amount of an inhibitor solution (0–10 mL) in H2O or in
DMSO (only for compounds 19 and 24), 11 mL DMSO and 10 mL of
a substrate solution (2 mm) were added and thoroughly mixed. In
case of compounds 19 and 24 the amount of buffer was kept con-
stant (169 mL) and DMSO was varied (1–11 mL). The reaction was
performed at 378C, initiated by adding 10 mL of enzyme solution
and followed over 10 min (lex =340 nm, lem =460 nm).
Enzymatic assays
Human matriptase inhibition assay: Assay buffer was 50 mm Tris-
HCl and 150 mm NaCl, pH 8.0. A solution of 2 g bovine serum albu-
min in 10 mL of a 154 mm NaCl solution was diluted 1:200 with
154 mm NaCl solution. The so-obtained solution was used for a 1:3
dilution of an enzyme stock solution (0.1 mgmLÀ1). This diluted
enzyme solution was stored at 08C and used for the measure-
ments. A 500 mm stock solution of fluorogenic substrate Mes-d-
Pro-Arg-AMC·2TFA in DMSO was diluted 1:5 with water. The final
concentration of the substrate was 10 mm, of DMSO was 6% and
of human matriptase was 0.00495 mgmLÀ1. Into each well a varying
amount of buffer (132–142 mL), a varying amount of an inhibitor
solution (0–10 mL) in H2O or in DMSO (only for compounds 19 and
24), 8 mL DMSO and 20 mL of a substrate solution (100 mm) were
added and thoroughly mixed. In case of compounds 19 and 24
the amount of buffer was kept constant (142 mL) and DMSO was
varied (0–8 mL). At 378C the reaction was initiated by adding 30 mL
of enzyme solution and followed over 10 min. Reactions were
Bovine trypsin inhibition assay:[30] Assay buffer was 50 mm Tris-
HCl, 150 mm NaCl, pH 8.0. The enzyme stock solution (1 mgmLÀ1
)
prepared in 1 mm HCl was kept at À208C, and diluted with 1 mm
HCl (1:100), followed by a dilution with assay buffer (1:40) and
kept at 08C. A 10 mm stock solution of the fluorogenic substrate
Boc-Gln-Ala-Arg-AMC in DMSO was diluted with assay buffer. The
final concentration of the substrate was 40 mm, of DMSO was 6%
and of bovine trypsin was 10 mgmLÀ1. Into each well a varying
amount of buffer (148.8–158.8 mL), a varying amount of an inhibi-
tor solution (0–10 mL) in H2O or in DMSO (only for compounds 19
and 24), 11.2 mL DMSO and 10 mL of a substrate solution (800 mm)
were added and thoroughly mixed. In case of compounds 19 and
24 the amount of buffer was kept constant (158.8 mL) and DMSO
was varied (1.2–11.2 mL). The reaction was performed at 258C, initi-
Chem. Eur. J. 2016, 22, 610 – 625
www.chemeurj.org
623
ꢀ 2016 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim

Full Paper
[12] B. I. Eriksson, D. J. Quinlan, J. W. Eikelboom, Annu. Rev. Med. 2011, 62,
41–57.
[13] K. List, Future Oncol. 2009, 5, 97–104.
ated by adding 20 mL of enzyme solution and followed over
10 min (lex =340 nm, lem =460 nm).
[14] T. Steinmetzer, A. Schweinitz, A. Stürzebecher, D. Dçnnecke, K. Uhland,
O. Schuster, P. Steinmetzer, F. Müller, R. Friedrich, M. E. Than, W. Bode, J.
Stürzebecher, J. Med. Chem. 2006, 49, 4116–4126.
[15] M. Hammami, E. Rühmann, E. Maurer, A. Heine, M. Gütschow, G. Klebe,
T. Steinmetzer, MedChemComm 2012, 3, 807–813.
Molecular docking
The X-ray structures of human matriptase (PDB ID: 4JZI),[34] human
thrombin (PDB ID: 2ANM),[51] and human factor Xa (PDB ID:
1LPG)[29] were obtained from the RCSB Protein Bank.[5] An X-ray
structure of matriptase-2 is publicly not available and therefore
a previously described comparative model was selected for dock-
ing calculations.[20] The X-ray structures and the comparative model
were used as templates for flexible ligand docking with Auto-
Dock 4.2.[52] Prior to the docking calculations, the template struc-
tures were prepared by removing bound inhibitors and crystallo-
graphic water molecules. Hydrogen atoms were assigned using the
Molecular Operating Environment (MOE 2012.10, MOE 2013.0801
and MOE 2014.09)[53] and partial charges were calculated with
AutoDock Tools.[52] For the docking calculations the number of ge-
netic algorithm runs and the number of energy evaluations were
varied while for all other parameters AutoDock 4.2 standard param-
eter settings were used. Putative compound binding modes were
selected after detailed visual inspection of high-scoring docking
poses and if necessary, preferred docking poses were further opti-
mized by subsequent energy minimizations with MOE. Co-crystal-
lized inhibitors for the superposition with predicted binding
modes were also taken from RCSB Protein Bank[5] (PBD IDs 4JZI[34]
and 4O97[35] for matriptase and PDB ID 1DWD[48] for thrombin). For
the visual evaluation of branched compounds an automatic filter-
ing procedure was used, only considering docking poses where
the S1 pocket was occupied by a benzamidine residue.
[16] L. Silvestri, A. Pagani, A. Nai, I. De Domenico, J. Kaplan, C. Camaschella,
Cell Metab. 2008, 8, 502–511.
[17] M. Stirnberg, M. Gütschow, Curr. Pharm. Des. 2013, 19, 1052–1061.
[18] C. Y. Wang, D. Meynard, H. Y. Lin, Front. Pharmacol. 2014, 5, 114.
[19] M. Stirnberg, E. Maurer, A. Horstmeyer, S. Kolp, S. Frank, T. Bald, K.
Arenz, A. Janzer, K. Prager, P. Wunderlich, J. Walter, M. Gütschow, Bio-
chem. J. 2010, 430, 87–95.
[20] M. T. Sisay, T. Steinmetzer, M. Stirnberg, E. Maurer, M. Hammami, J. Ba-
jorath, M. Gütschow, J. Med. Chem. 2010, 53, 5523–5535.
[21] E. Maurer, M. T. Sisay, M. Stirnberg, T. Steinmetzer, J. Bajorath, M. Güt-
schow, ChemMedChem 2012, 7, 68–72.
[22] J. Mosquera, J. Rodríguez, M. E. Vµzquez, J. L. MascareÇas, ChemBio-
Chem 2014, 15, 1092–1095.
[23] M. I. Sµnchez, J. Martínez-Costas, J. L. MascareÇas, M. E. Vµzquez, ACS
Chem. Biol. 2014, 9, 2742–2747.
[24] B. Gabriel, M. T. Stubbs, A. Bergner, J. Hauptmann, W. Bode, J. Stürze-
becher, L. Moroder, J. Med. Chem. 1998, 41, 4240–4250.
[25] O. Klingler, H. Matter, M. Schudok, S. P. Bajaj, J. Czech, M. Lorenz, H. P.
Nestler, H. Schreuder, P. Wildgoose, Bioorg. Med. Chem. Lett. 2003, 13,
1463–1467.
[26] R. J. de Beer, B. Bçgels, G. Schaftenaar, B. Zarzycka, P. J. Quaedflieg, F. L.
van Delft, S. B. Nabuurs, F. P. Rutjes, ChemBioChem 2012, 13, 1785–
1790.
[27] W. J. Guilford, K. J. Shaw, J. L. Dallas, S. Koovakkat, W. Lee, A. Liang, D. R.
Light, M. A. McCarrick, M. Whitlow, B. Ye, M. M. Morrissey, J. Med. Chem.
1999, 42, 5415–5425.
[28] I. J. Enyedy, S. L. Lee, A. H. Kuo, R. B. Dickson, C. Y. Lin, S. Wang, J. Med.
Chem. 2001, 44, 1349–1355.
Acknowledgements
[29] H. Matter, E. Defossa, U. Heinelt, P. M. Blohm, D. Schneider, A. Müller, S.
Herok, H. Schreuder, A. Liesum, B. Brachvogel, P. Lçnze, A. Walser, F. Al-
Obeidi, P. Wildgoose, J. Med. Chem. 2002, 45, 2749–2769.
[30] S. Dosa, M. Stirnberg, V. Lülsdorff, D. Häußler, E. Maurer, M. Gütschow,
Bioorg. Med. Chem. 2012, 20, 6489–6505.
[31] M. NazarØ, H. Matter, D. W. Will, M. Wagner, M. Urmann, J. Czech, H.
Schreuder, A. Bauer, K. Ritter, V. Wehner, Angew. Chem. Int. Ed. 2012, 51,
905–911; Angew. Chem. 2012, 124, 929–935.
[32] E. Colombo, A. DØsilets, D. DuchÞne, F. Chagnon, R. Najmanovich, R.
Leduc, E. Marsault, ACS Med. Chem. Lett. 2012, 3, 530–534.
[33] D. Meyer, F. Sielaff, M. Hammami, E. Bçttcher-Friebertshäuser, W. Garten,
T. Steinmetzer, Biochem. J. 2013, 452, 331–343.
N.F. was supported by a fellowship from the Jürgen Manchot
Foundation, Düsseldorf, Germany, D.H. by a fellowship from
the Bonn International Graduate School of Drug Sciences (BIGS
DrugS). M.S. is supported by the German Research Foundation
(STI 660/1-1) and the Maria von Linden-Program of the Univer-
sity of Bonn. The authors thank Dr. Matthias D. Mertens for
support in building block synthesis and Dr. Eva Maurer and
Anna-Madeleine Beckmann for providing matriptase-2.
[34] R. Goswami, S. Mukherjee, G. Wohlfahrt, C. Ghadiyaram, J. Nagaraj, B. R.
Chandra, R. K. Sistla, L. K. Satyam, D. S. Samiulla, A. Moilanen, H. S. Sub-
ramanya, M. Ramachandra, ACS Med. Chem. Lett. 2013, 4, 1152–1157.
[35] R. Goswami, S. Mukherjee, C. Ghadiyaram, G. Wohlfahrt, R. Sistla, J. Na-
garaj, L. K. Satyam, K. Subbarao, R. K. Palakurthy, S. Gopinath, N. R. Krish-
namurthy, T. Ikonen, A. Moilanen, H. S. Subramanya, P. Kallio, M. Rama-
chandra, Bioorg. Med. Chem. 2014, 22, 3187–3203.
Keywords: benzamidines · chirality · peptidomimetics · serine
proteases · symmetry
[1] D. B. Kitchen, H. Decornez, J. R. Furr, J. Bajorath, Nat. Rev. Drug Discovery
2004, 3, 935–949.
[36] A. Tziridis, D. Rauh, P. Neumann, P. Kolenko, A. Menzel, U. Bräuer, C.
Ursel, P. Steinmetzer, J. Stürzebecher, A. Schweinitz, T. Steinmetzer, M. T.
Stubbs, Biol. Chem. 2014, 395, 891–903.
[37] Z. Han, P. K. Harris, D. E. Jones, R. Chugani, T. Kim, M. Agarwal, W. Shen,
S. A. Wildman, J. W. Janetka, ACS Med. Chem. Lett. 2014, 5, 1219–1224.
[38] D. DuchÞne, E. Colombo, A. DØsilets, P. L. Boudreault, R. Leduc, E. Mar-
sault, R. Najmanovich, J. Med. Chem. 2014, 57, 10198–10204.
[39] F. M. Franco, D. E. Jones, P. K. Harris, Z. Han, S. A. Wildman, C. M. Jarvis,
J. W. Janetka, Bioorg. Med. Chem. 2015, 23, 2328–2343.
[40] L. H. Easson, E. Stedman, Biochem. J. 1933, 27, 1257–1266.
[41] A. G. Ogston, Nature 1948, 162, 963.
[42] A. D. Mesecar, D. E. Koshland, Jr., Nature 2000, 403, 614–615.
[43] B. D. Judkins, D. G. Allen, T. A. Cook, B. Evans, T. E. Sardharwala, Synth.
Commun. 1996, 26, 4351–4367.
[2] G. Sliwoski, S. Kothiwale, J. Meiler, E. W. Lowe, Jr., Lowe, Pharmacol. Rev.
2014, 66, 334–395.
[3] P. Ripphausen, B. Nisius, L. Peltason, J. Bajorath, J. Med. Chem. 2010, 53,
8461–8467.
[4] Y.-C. Chen, Trends Pharmacol. Sci. 2015, 36, 78–95.
[5] RCSB Protein Data Bank: http://www.rcsb.org.
[6] T. L. Blundell, H. Jhoti, C. Abell, Nat. Rev. Drug Discovery 2002, 1, 45–54.
[7] G. L. Warren, C. W. Andrews, A. M. Capelli, B. Clarke, J. LaLonde, M. H.
Lambert, M. Lindvall, N. Nevins, S. F. Semus, S. Senger, J. Med. Chem.
2006, 49, 5912–5931.
[8] S. Z. Grinter, X. Zou, Molecules 2014, 19, 10150–10176.
[9] A. Straub, S. Roehrig, A. Hillisch, Angew. Chem. Int. Ed. 2011, 50, 4574–
4590; Angew. Chem. 2011, 123, 4670–4686.
[10] D. Kikelj, Pathophysiol. Haemost. Thromb. 2003, 33, 487–491.
[11] D. J. P. Pinto, J. M. Smallheer, D. L. Cheney, R. M. Knabb, R. R. Wexler, J.
Med. Chem. 2010, 53, 6243–6274.
[44] S. Dosa, J. Daniels, M. Gütschow, J. Heterocycl. Chem. 2011, 48, 407–
413.
Chem. Eur. J. 2016, 22, 610 – 625
www.chemeurj.org
624
ꢀ 2016 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim

Full Paper
[45] N. Furtmann, Y. Hu, M. Gütschow, J. Bajorath, RSC Adv. 2015, 5, 43660–
43668.
[51] U. E. Lange, D. Baucke, W. Hornberger, H. Mack, W. Seitz, H. W. Hçffken,
Bioorg. Med. Chem. Lett. 2006, 16, 2648–2653.
[46] N. Furtmann, Y. Hu, M. Gütschow, J. Bajorath, Chem. Biol. Drug Des.
2015, 86, 1458–1465.
[52] G. M. Morris, R. Huey, W. Lindstrom, M. F. Sanner, R. K. Belew, D. S. Good-
sell, A. J. Olson, J. Comput. Chem. 2009, 30, 2785–2791.
[47] M. W. Rathke, P. J. Cowan, J. Org. Chem. 1985, 50, 2622–2624.
[48] J. Stürzebecher, D. Prasa, J. Hauptmann, H. Vieweg, P. Wikstrçm, J. Med.
Chem. 1997, 40, 3091–3099.
[53] Molecular Operating Environment (MOE), 2014.09, Chemical Computing
Group Inc., Montreal, Quebec, Canada.
[49] D. W. Banner, P. Hadvµry, J. Biol. Chem. 1991, 266, 20085–20093.
[50] I. Jabin, G. Revial, N. Monnier-Benoit, P. Netchita¨ılo, J. Org. Chem. 2001,
66, 256–261.
Received: September 4, 2015
Published online on December 2, 2015
Chem. Eur. J. 2016, 22, 610 – 625
www.chemeurj.org
625
ꢀ 2016 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim