Between two worlds: A comparative study on in vitro and in silico inhibition of trypsin and matriptase by redox-stable SFTI-1 variants at near physiological pH

Article abstract of DOI:10.1039/c2ob26162f

A comparative study on in vitro and in silico inhibition of trypsin and matriptase by derivatives of the sunflower trypsin inhibitor-1 at near physiological pH is reported. Besides wild-type bicyclic SFTI-1, monocyclic variants possessing native cystine as well as redox-stable triazolyl side-chain macrocyclization motifs were studied for the first time in matriptase inhibition assays. Interestingly, monocyclic SFTI-1[1,14] demonstrated higher potency against this pharmacologically relevant protease compared to its bicyclic counterpart. Structural analysis of binding/inhibition of investigated SFTI-1 derivatives was performed using a combination of molecular dynamics simulations and docking experiments. In silico data were in good accordance with in vitro results, indicating the importance of the terminal inhibitor regions for the affinity towards matriptase. Presented work gives new perspectives for the optimization of the SFTI-1 framework towards in vivo applications. The Royal Society of Chemistry 2012.

Full text of DOI:10.1039/c2ob26162f

View Article Online / Journal Homepage / Table of Contents for this issue
Dynamic Article Links
Organic &
Biomolecular
Chemistry
Cite this: Org. Biomol. Chem., 2012, 10, 7753
www.rsc.org/obc
PAPER
Between two worlds: a comparative study on in vitro and in silico inhibition of
trypsin and matriptase by redox-stable SFTI-1 variants at near physiological
pH†
Olga Avrutina, Heiko Fittler, Bernhard Glotzbach, Harald Kolmar and Martin Empting*
Received 18th June 2012, Accepted 3rd August 2012
DOI: 10.1039/c2ob26162f
A comparative study on in vitro and in silico inhibition of trypsin and matriptase by derivatives of the
sunflower trypsin inhibitor-1 at near physiological pH is reported. Besides wild-type bicyclic SFTI-1,
monocyclic variants possessing native cystine as well as redox-stable triazolyl side-chain
macrocyclization motifs were studied for the first time in matriptase inhibition assays. Interestingly,
monocyclic SFTI-1[1,14] demonstrated higher potency against this pharmacologically relevant protease
compared to its bicyclic counterpart. Structural analysis of binding/inhibition of investigated SFTI-1
derivatives was performed using a combination of molecular dynamics simulations and docking
experiments. In silico data were in good accordance with in vitro results, indicating the importance of the
terminal inhibitor regions for the affinity towards matriptase. Presented work gives new perspectives for
the optimization of the SFTI-1 framework towards in vivo applications.
higher negative polarization (Fig. 1). This provides a better
overall charge complementarity with SFTI-1. It has been pro-
Introduction
Very often interesting scientific findings arise from unexpected
or even counterintuitive results. Recently, a comparison between
the crystal structures of the sunflower trypsin inhibitor-1
(SFTI-1), a bicyclic tetradecapeptide, in complex with two
closely related serine proteases revealed a surprising biochemical
peculiarity.1,2 By analyzing the binding geometry of
SFTI-1 matriptase and trypsin aggregates, Yuan et al. confirmed
a high similarity between both structures.1,2 Though the homolo-
gous interaction profile of these serine proteases with SFTI-1
was expected, it appeared rather astonishing that this smallest
naturally occurring Bowman–Birk Inhibitor (BBI)3,4 favors
trypsin over matriptase in retrospect.1–3,5 Under the basic con-
ditions (pH 8.2–8.5) commonly applied to in vitro inhibition
assays for these enzymes, the electrostatic surface potential
around the active site of matriptase shows, compared to the
environment around the catalytic center of trypsin, a significantly
posed that this anticipated enthalpic preference is overcompen-
sated by an entropic penalty predominantly originating from the
secondary binding loop of SFTI-1.1 This might explain the sur-
prising difference between its inhibitory activity against trypsin
(Ki 0.1 nM) and matriptase (Ki 1–100 nM).1–3,5
However, selective binding and/or inhibition of matriptase
under physiological conditions is of great pharmacological inter-
est as this type II transmembrane serine protease (TTSP) is
involved in a variety of pathological processes like progression
and metastasis of epithelial tumors, as well as osteoarthritis and
atherosclerosis.6–10 A number of potent inhibitors of this serine
protease based either on small organic molecules or large anti-
body fragments and exhibiting single digit nanomolar or
even picomolar inhibition constants, respectively, have already
been reported.11,12 Nevertheless, attempts to tune the peptidic
SFTI-1 framework towards matriptase resulted in inhibitors with
enhanced selectivity but dramatically decreased potency.5
Besides monocyclic variants lacking the disulfide connectivity
and possessing substantially worsened kinetic properties, matrip-
tase inhibition by side-chain cyclized SFTI-1 derivatives having
free N- and C-termini has not been reported to date.
Clemens-Schöpf Institute of Organic Chemistry and Biochemistry,
Technische Universität Darmstadt, Petersenstr. 22, 64287 Darmstadt,
Germany. E-mail: Empting@Biochemie-TUD.de
†Electronic supplementary information (ESI) available: Detailed syn-
thetic procedures of compounds 1 and 3, experimental details of enzyme
assays, Fig. S1–S9 (SDS-Page analysis of inclusion body production,
refolding/autoactivation, and purification of recombinant matriptase, as
well as RP-HPLC traces, ESI-MS spectra, and dose–response curves
against trypsin for previously uncharacterized inhibitors 1 and 3),
and Tables S1–S7 (determination of Ki values using alternative
equations and the full set of used force field parameters). See DOI:
10.1039/c2ob26162f
Recently, we developed potent inhibitors of trypsin based on
monocyclic SFTI-1 derivatives containing redox-stable triazole
bridges of different length and shape.13 The impact of the macro-
cyclization motif in their bioactivity was demonstrated as well.13
Herein, we report an unexpected change of the relative potency
within the set of matriptase inhibitors 1–6 (Fig. 2) and evaluate
This journal is © The Royal Society of Chemistry 2012
Org. Biomol. Chem., 2012, 10, 7753–7762 | 7753

View Article Online
Fig. 2 Structures of investigated peptides and peptidomimetics. Bicyc-
lic SFTI-1 (1), monocyclic SFTI-1[1,14] (2), monocyclic peptidomi-
metics containing either a 1,5-disubstituted 1,2,3-triazole (3, 4) or a
1,4-disubstituted 1,2,3-triazole (5, 6) as the macrocyclization motif.
each peptidic ligand in solution and upon binding to both trypsin
and matriptase. In silico binding affinities were calculated and
compared with the in vitro results. Thus, the role of the second-
ary loop as well as the impact of different macrocyclization
motifs in matriptase inhibition of variants 1–6 were elucidated.
Results and discussion
In vitro inhibition of trypsin and matriptase
Trypsin is a prominent digestive protease commonly found in
the small intestine of vertebrates19 and operating most effectively
under the mild basic conditions of its native environment (pH ∼
8).20 Membrane-anchored matriptase, on the other hand, is
usually expressed in epithelial cells and regulates cellular
adhesion and growth by hydrolyzing proteins bound to cell sur-
faces or present in the extracellular matrix.21 Thus, it functions
in an environment controlled by homeostasis in healthy tissues
(pH 7.4). Furthermore, tumors in general are known to generate
hypoxic regions leading even to acidic microenvironments.22–24
However, cancer-related matriptase possesses a pH optimum of
about 9, therefore, the majority of reported inhibition assays
have been carried out at a pH range between 8 and 9.3,5,11,12,25,26
We evaluated the potency of synthesized inhibitors through
the determination of the fractional activity of the respective
enzyme (v/v0) in the presence of an inhibitor at various concen-
trations [I]. A number of mathematical methods are commonly
used for the calculation of apparent inhibition constants Kiapp or
IC50 values (concentration of the inhibitor to achieve a half-
maximal degree of inhibition) for competitive inhibitors like
SFTI-1.27–32
Fig. 1 Molecular surfaces of bicyclic SFTI-1 (transparent surface and
yellow sticks) in complex with (A) trypsin (PDB ID: 1SFI) and (B)
matriptase (PDB ID: 3P8F). Color gradients (red to blue) indicate elec-
trostatic surface potentials (negative to positive) at pH 8.5 calculated via
the Particle Mesh Ewald method with a maximum value of 500 kJ
mol−1 14
(C) Stick representation of SFTI-1 structure. Blue: nitrogen,
.
red: oxygen, green: sulfur, white: carbon, hydrogen omitted for clarity.
this finding on a structural basis using a two-step in silico
method.
Recently, Swedberg and coworkers have used molecular
dynamics (MD) simulations to predict beneficial amino acid
exchanges for SFTI-1-derived inhibitors of kallikrein-related
peptidase 4 (KLK4).15 Our computational approach combines an
MD calculation with subsequent semi-flexible and rigid docking
experiments using a customized AMBER-derived force field
(YASARA2) with embedded parameters for the calculation of
non-natural triazolyl moieties within peptides.16–18 This pro-
cedure enabled us to investigate the structural characteristics of
7754 | Org. Biomol. Chem., 2012, 10, 7753–7762
This journal is © The Royal Society of Chemistry 2012

View Article Online
Table 1 Determined Ki for compounds 1–6
Matriptase/μM
Therefore, we used the following equations for the non-linear
fit of observed dose–response curves ([E] meaning active con-
centration of the enzyme).
Trypsin/nM
pH 7.6
Entry
pH 7.6
pH 8.5
v
1
¼
ð1Þ
½Iꢀ
1
2
3
4
5
6
0.07 0.01
0.21 0.03a
5.1 1.0
1.1 0.14
0.7 0.09
1.24 0.16
12.9 1.6
285 37
0.15 0.03
0.1 0.02
n.d.
n.d.
n.d.
v0
1 þ IC
50
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
0.34 0.05a
273 51a
106 18a
ð½Eꢀ ꢁ ½Iꢀ ꢁ KiappÞ þ ð½Iꢀ þ Kiapp ꢁ ½EꢀÞ þ 4Kiapp½Eꢀ
2½Eꢀ
2
v
¼
94.1 12
n.d.
v0
All experiments were performed in triplicate. n.d. means not
determined.a Values calculated from previously reported Kiapp 13
ð2Þ
.
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ð½Eꢀ ꢁ ½Iꢀ þ KiappÞ ꢁ ð½Eꢀ þ ½Iꢀ þ KiappÞ ꢁ 4½Eꢀ½Iꢀ
2
v
¼ 1 ꢁ
v0
2½Eꢀ
ð3Þ
In the case of reversible competitive inhibition IC50 values
calculated using eqn (1) can be converted into the substrate-inde-
pendent Ki using eqn (4) ([S] is the concentration of the chromo-
genic substrate used within the enzyme assay and KM is the
corresponding Michaelis–Menten constant).27
IC50
Ki ¼
ð4Þ
½Sꢀ
1 þ K
M
In the case of tight-binding inhibition, eqn (5) has to be used
instead of eqn (4).28
E
IC50 ꢁ 2
Ki ¼
ð5Þ
½Sꢀ
1 þ K
M
Eqn (2) and (3) are both adaptations of the Morrison equation
for the determination of the substrate-dependent (apparent) inhi-
bition constants for tight-binding inhibitors.29–32 They can be
transformed into each other via simple arithmetic conversions
(see ESI†). Substrate-independent inhibition constant Ki can be
calculated from Kiapp through eqn (6).30
K1app
Ki ¼
ð6Þ
½Sꢀ
1 þ K
M
However, all used methods yielded essentially the same Ki
values (see ESI†). For the sake of comparability with our pre-
vious studies, the results generated via eqn (3) and (6) are sum-
marized in Table 1.
To investigate the effect of near physiological pH on matrip-
tase inhibition by SFTI-1 (1) and SFTI-1[1,14] (2), we deter-
mined Ki values for both compounds at pH 7.6 and 8.5 (Fig. 3A,
Table 1). A significant decrease in inhibitory activity for 1 and 2
at lower pH was observed. However, the detected inhibitory
potency of bicyclic SFTI-1 under more basic conditions was in
accordance with reported data (0.1 μM – Li et al., 0.15 μM –
this work).5 Interestingly, the monocyclic variant SFTI-1[1,14]
(2) possessing only a disulfide bond as the macrocyclization
motif showed a better inhibition of matriptase compared to its
backbone-cyclized counterpart 1. To our knowledge, this unex-
pected finding has not been reported to date.
Fig. 3 Dose–response curves for the inhibition of matriptase-catalyzed
proteolysis of chromogenic substrate Boc-QAR-pNA with the X-axis on
a logarithmic scale. (A) Comparison of bicyclic SFTI-1 (1) with mono-
cyclic SFTI-1[1,14] (2) at pH 7.6 (white and black circles, respectively)
and 8.5 (white and black triangles, respectively). (B) Comparison of
disulfide-bridged inhibitors 1 (white circles) and 2 (black circles) with
triazole-bridged peptidomimetics 3 (grey triangles), 4 (black triangles),
5 (grey squares), and 6 (black squares) at pH 7.6. Data points are arith-
metic means of three experiments and error bars are given as the stan-
dard deviation.
In the case of trypsin, the highly constrained bicyclic wild-
type peptide (1) was the most potent inhibitor within the set of
This journal is © The Royal Society of Chemistry 2012
Org. Biomol. Chem., 2012, 10, 7753–7762 | 7755

View Article Online
tested compounds 1–6. Each modification of the SFTI-1 scaffold
resulted in the loss of bioactivity against trypsin (Table 1).
Nevertheless, the decrease in potency was quite marginal for
open-chain variants 2 and 4. The substitution pattern within the
macrocyclization motif of peptidomimetics 3–6 had a dramatic
influence on their bioactivity.13 Inhibitors 3 and 4 containing a
1,5-disubstituted 1,2,3-triazole were significantly more potent
compared to variants 5 and 6 possessing the 1,4-isomer.
However, novel compound 3 lacking one methylene unit
within the side chain of the third residue (Fig. 2) was not as
active as variant 4.
To study the effect of the described SFTI-1 framework modifi-
cations on matriptase inhibition, we determined Ki values for all
compounds at pH 7.6 (Fig. 3B, Table 1). Similarly to trypsin
inhibition, peptidomimetics 3 and 4 showed better performance
than 5 and 6 in matriptase assays as well. Surprisingly, compari-
son of variants 3 and 4 showed that 3 was more active against
matriptase, while 4 was the better trypsin inhibitor.
The applied simulation procedure resulted in significantly
larger distances between the α-carbons of residues 3 and 11 for
both compounds (5 and 6) compared to our previous study
where only an energy minimization step was used.13 Thus, the
two opposing β-strands within the respective peptide frameworks
are more remote from each other than within the structures of
inhibitors 1–4. Consequently, the possibility to form intramole-
cular hydrogen bonds is reduced for peptides 5 and 6 (Table 2).
Since the hydrogen bond network is a structural prerequisite for
the stability of the BBI loop conformation and, thus, biological
activity,15 these findings could explain the observed loss of
potency for SFTI-1 variants possessing the 1,4-substitution
pattern within their macrocyclization motif.
Bicyclic wild type 1 and monocyclic compound 2, as well as
1,5-disubstituted 1,2,3-triazole containing peptidomimetics 3
and 4, did not show a significant disorder of the inhibitor loop
(Fig. 5B). Nevertheless, considerable structural changes within
the N- and C-terminal regions of compounds 2–6 compared to
the secondary loop of SFTI-1 occurred (Fig. 5C). This result,
however, was highly anticipated as the conformational con-
straints caused by the additional backbone macrocyclization of
peptide 1 are missing within the open-chain variants. The struc-
tural deviation of residues 1, 2 and 12–14 within highly potent
trypsin inhibitor 4 was the most prominent one, while the values
for compounds 3, 5, and 6 were in the range of monocyclic
SFTI-1[1,14] (Fig. 5C). Considering the decreased relative
activity of 4 against matriptase, our structural data support the
notion that the N- and C-terminal regions are responsible for
enzyme specificity rather than for the general capability to
inhibit trypsin-like proteases. Thus, the observed structural
changes and increased flexibility of these backbone segments in
open-chain variants most likely are the reason for the improved
activity of monocyclic compound 2 compared to bicyclic 1
against matriptase.
The collected in vitro data shown in Table 1 emphasize that
the native SFTI-1 scaffold has a general preference for trypsin
over matriptase, especially at near physiological pH. The dis-
crimination between trypsin and matriptase upon changing the
target enzyme must have intrinsic structural reasons. Moreover,
the unexpected inversion of relative affinities (1 versus 2, and 3
versus 4) has to be explained.
MD simulation of SFTI-1 (1), SFTI-1[1,14] (2) and
peptidomimetic compounds 3–6
To elucidate in atomic detail the impact of every single modifi-
cation, we simulated compounds 1–6 as shown in Fig. 4 (see
also the Experimental section).
A significant amount of structural data on SFTI-1 in solution
as well as in complex with relevant proteases is available in the
protein data base.1,2,33 Thus, for the simulation of monocyclic
variants 3–6 initial coordinates were derived from the reported
ones (1JBN). However, preliminary studies revealed that the
extensive hydrogen bond network of the SFTI-1 framework is
able to compensate the steric strain on the peptide backbone gen-
erated by rigid, planar triazole bridges. This resulted in unnatu-
rally bent heterocyclic structures within compounds 5 and 6
(ESI†). Thus, we decreased the scaling factor for non-bonded
interactions and applied increased temperature during the initial
phase of simulation (Fig. 4). This procedure allowed respective
structures to leave the local minimum of the template coordi-
nates. After incremental rescaling of parameters within 1 ns and
an additional relaxation period of 8 ns no significant fluctuations
in backbone root mean square deviations (RMSD) were observed
(Fig. 5A). Hence, stable trajectories were analyzed in detail
revealing fundamental structural differences between compounds
1–6 (Fig. 5B, C, and Table 2). The most dramatic effect on the
peptide structure is caused by the 1,4-disubstituted 1,2,3-triazole
Docking of SFTI-1 (1), SFTI-1[1,14] (2) and peptidomimetic
compounds 3–6 to trypsin and matriptase
To investigate the impact of described framework modifications
in the binding geometries of inhibitor–protease complexes,
docking experiments were performed (see the Experimental
section, Fig. 4). The collected data are summarized in Table 3.
As all of the investigated compounds 1–6 inhibit trypsin and
matriptase in a concentration-dependent manner (Fig. 3 and
Table 1), we rationalized that interaction of open-chain variant 2
and peptidomimetics 3–6 with these proteases must follow a
similar binding geometry to wild-type SFTI-1. This postulation
is supported by recently solved inhibitor–trypsin complex struc-
tures of monocyclic SFTI-based peptidomimetics.34 Thus, in
order to exhibit observed biological activities, each compound
must penetrate the S1 site of a respective protease with the side
chain of Lys5. Consequently, we screened the set of generated
inhibitor–protease complexes for structures that fulfilled this
requirement. The ratio of counted aggregates matching the
expected binding geometry to the number of all calculated struc-
tures (100 each) is given as Phit (Table 3). Interestingly, in our
docking experiments this value was significantly higher for
trypsin than for matriptase. Furthermore, all potent inhibitors
bridge within variant 5. Fig. 5B clearly illustrates
a
significant distortion of the inhibitory loop compared to crystal
structures of wild-type SFTI-1 in complex with trypsin and
matriptase. A similar, although not so dramatic, effect was
observed for compound 6 which contained the 1,4-substitution
pattern as well.
7756 | Org. Biomol. Chem., 2012, 10, 7753–7762
This journal is © The Royal Society of Chemistry 2012

View Article Online
Fig. 4 Design of the two-step in silico experiment. Illustrative images (left) show a simulation cell containing water and a peptidic ligand (top left),
as well as a number of ligand poses generated in a local docking experiment restricted by a simulation cell around the active site of the receptor
(bottom left). The MD simulation procedure (top) is depicted as time-lapsed absolute temperature (dashed, white) and scaling factor curves for non-
bonded interactions (black, solid). Both parameters are given in percent to 298 K and 1, respectively. The color gradient (red to blue) qualitatively
illustrates the resulting stress applied to the simulated structure; red means harsh and blue means normalized conditions. The 8 ns period for relaxation
is shortened. Snapshots between nanosecond nine and ten are analyzed using a docking procedure (bottom). Free energies of binding ΔBG are calcu-
lated by sequential semi-flexible and rigid docking steps. Receptor molecules trypsin and matriptase are shown as white surfaces. The ligand is
depicted as yellow tubes and sticks. The star symbol indicates an energy minimization step.
1–4 consistently showed a higher probability Phit to interact with
both proteases in a mode similar to the reported complexes of
SFTI-1, compared to weak binders 5 and 6.
interactions. Indeed, only a moderate increase of the backbone
RMSD values was observed for compounds 3, 4, and 6. Similar
data were extracted from docking experiments with trypsin
(Fig. S9, ESI†). However, a more prominent deviation between
predicted binding geometries and the crystal structure was
measured for the residues corresponding to the secondary loop
of SFTI-1. The stability of the investigated aggregates was evalu-
ated by averaging calculated ΔBG for every structure with
expected binding geometry resulting in ΔBGdock (Table 3).
In general, the calculated mean values indicated a significantly
more favorable interaction between matriptase and inhibitors 1–6
compared to respective complexes with trypsin. This finding cor-
roborated the expected influence of charge complementarity
between the SFTI-1 scaffold and the surface around the active
site of matriptase on overall complex stability.
The most prominent deviations upon docking to trypsin are
observed in the N- and C-terminal regions of compounds 1–6.
This indicates less conformational constraints compared to the
respective matriptase complexes. Thus, an increased flexibility of
bound SFTI ligands is provided, while the amount of attractive
electrostatic interactions is reduced. A tighter, but also more con-
strained binding of SFTI-1 to matriptase had already been postu-
lated by Yuan et al. and was corroborated by our experiments.1
Surprisingly, the ability to bind similarly to the reported
crystal structures did not necessarily lead to high values of mean
ΔBGdock. Thus, for variant 6 the best docking complexes were
characterized by rather low backbone RMSD values (Fig. 6F and
We also investigated the capability of compounds 1–6 to form
intermolecular hydrogen bonds with each of the target proteases.
Thus, the number of corresponding interactions was counted
for each complex with the ligand docked following the described
binding mode. As expected, peptidomimetic 5 showed the
lowest number of hydrogen bonds within the set of examined
structures. However, the amount of measured intermolecular
proton donor–acceptor interactions of compound 6 with trypsin
and matriptase was in the same range as for inhibitors 1–4.
The analysis of the structural and thermodynamic properties
of the generated docking results was of particular importance.
The applied procedure enabled us to determine the free energy
of binding ΔBG for each ligand–receptor complex (Table 3).
Calculated structures of compounds 1–6 bound to matriptase and
possessing the highest absolute value of ΔBG are shown in
Fig. 6. These models exhibited the most favorable interaction
profiles observed, therefore were the best predicted structures for
the actual binding geometry within our in silico experiment.
Complexes 1 and 2 were found to interact with matriptase in a
very similar manner compared to the reported crystal structure
(Fig. 6A, B). Due to the highly distorted BBI loop of peptidomi-
metic 5, this variant showed the most deviating docking geome-
try (Fig. 6E). The N- and C-terminal regions were drastically
displaced, considerably reducing the possibility for attractive
This journal is © The Royal Society of Chemistry 2012
Org. Biomol. Chem., 2012, 10, 7753–7762 | 7757

View Article Online
Table 2 Data collected from the simulation experiments for
compounds 1–6
Backbone
Distance
Intramol.
Entry
RMSDa,b/Å
Cα3–Cα11b/Å
H-bondsb,c
1
2
3
4
5
6
0.87 0.09
1.13 0.07
1.72 0.15
1.74 0.06
3.02 0.10
1.79 0.15
3.93 0.08
3.95 0.07
3.90 0.06
4.29 0.10
5.66 0.12
5.16 0.14
6.08 1.08
6.12 1.14
5.85 1.15
5.67 0.96
2.63 0.91
4.36 0.82
a Compared to solution structure 1JBL. b Values are arithmetical means
of 100 structures taken from the last nanosecond of simulation.
Errors are given as standard deviation. c Measured using a threshold of
6.25 kJ mol−1
.
simulation (Fig. 5B). Nevertheless, the applied two-step docking
procedure allowed a number of the in silico samples of peptido-
mimetic 6 to reconfigure the backbone conformation to the orig-
inal BBI geometry upon interaction with the target enzymes.
In our docking experiments, compound 4 showed average
values of ΔBGdock very similar to 6 (Table 3). In this case, the
pronounced deviation within the N- and C-terminal regions can
be considered as the reason for the reduced complex stability
(Fig. 5C). Thus, within the predicted inhibitor 4–matriptase
complex the aromatic side chain of Phe12 is positioned consider-
ably different to inhibitors 1–3 (Fig. 6D). In the corresponding
crystal structure this residue is highly constrained, being flanked
by two benzyl groups of Phe97 and Phe99 of the protease mole-
cule.1 Thus, affinity of peptidomimetic 4 towards matriptase
might be reduced due to a loss of favorable hydrophobic inter-
actions within this region. Upon docking to trypsin, the C-term-
inal residues were even more displaced (Fig. S9D†).
Nevertheless, the observed loss of free binding energy was still
only of moderate magnitude for compound 4. Interestingly, this
inhibitor demonstrated the highest probability to interact with
both proteases in the BBI geometry within the set of peptido-
mimetics 3–6 (Table 3).
Compound 3 showed significantly lower Phit values compared
to 4. However, if bound in the BBI modality, 3 formed far more
stable complexes. The determined absolute values of ΔBGdock
were even higher than those of monocyclic 2. As, compared to
4, peptidomimetic 3 lacks one methylene unit within the macro-
cyclization motif (Fig. 2), this structure possesses less degrees of
conformational freedom. As a result, the proximity of α-carbons
between residues 3 and 11 in compound 3 matches the corre-
sponding distance within the wild-type peptides 1 and 2 almost
perfectly (Table 2). Nevertheless, the discussed structural con-
straints also caused the loss of backbone flexibility. Thus, com-
pound 3 is held within a defined range of conformations
displaying high stabilities of complexes with both proteases,
once the inhibitor is positioned accordingly. However, the
applied docking algorithm did not always succeed to fit com-
pound 3 into the BBI geometry. This indicates that the latitude
between perfect fit and steric clashes for SFTI-1 variants in
complex with matriptase is rather limited.
Fig. 5 (A) Backbone RMSD values of 1 (top, black), 2 (top, grey),
3 (middle, black), 4 (middle, grey), 5 (bottom, black), and 6 (bottom,
grey) compared to SFTI-1 solution structure 1JBL plotted against the
simulation time. Dot plots of backbone RMSD values of the inhibitory
loop (B) and the secondary loop/N- and C-terminal regions (C) of com-
pounds 1–6 compared to SFTI-1 in complexes with trypsin (black) and
matriptase (grey) showing 100 values determined every 10 ps during the
last nanosecond of simulation.
S9F). Nevertheless, more distorted structures with significantly
lower affinities were also observed. This heterogeneity of inter-
action profiles was reflected by the highest errors (standard devi-
ations) of corresponding ΔBGdock within the set of investigated
compounds (Table 3). This probably originates from the
described moderate distortion of the inhibitory loop during
A similar effect was observed for wild type 1. This very
constrained bicyclic inhibitor consistently showed the highest
7758 | Org. Biomol. Chem., 2012, 10, 7753–7762
This journal is © The Royal Society of Chemistry 2012

View Article Online
Table 3 Data collected from the docking experiments for compounds 1–6 against trypsin and matriptase and corresponding in vitro free energies of
binding ΔBGexp
Trypsin
Phit
Matriptase
/
Intermol.
ΔBGdock a
/
ΔBGcalc c
(kJ mol−1
/
ΔBGexp d
(kJ mol−1
/
Phit
/
Intermol.
H-bondsa
ΔBGdock a
/
ΔBGcalc c
(kJ mol−1
/
ΔBGexp d
(kJ mol−1
)
/
Entry
(%)
H-bondsa,b
(kJ mol−1
)
)
)
(%)
(kJ mol−1
)
)
1
2
3
4
5
6
96
99
87
100
69
81
10.3 1.9
11.3 1.5
10.9 1.4
10.0 1.7
8.9 1.4
9.8 1.6
61.3 6.3
51.6 4.9
54.8 6.9
48.6 5.4
33.6 3.9
43.8 9.8
58.9 6.1
51.1 4.8
47.7 6.0
48.6 5.4
23.2 2.7
35.4 8.0
58.0 0.4
55.2 0.4
47.3 0.5
54.0 0.3
37.5 0.5
39.8 0.4
51
67
51
61
47
43
9.2 1.8
10.1 2.1
11.0 1.6
9.1 1.9
8.4 1.5
9.3 2.2
71.0 8.1
63.2 4.9
69.0 6.9
56.4 5.4
42.6 3.9
56.1 9.8
36.2 4.1
42.3 5.8
35.2 4.4
34.4 7.1
20.0 3.0
24.1 4.8
34.0 0.3
35.1 0.3
33.7 0.3
27.9 0.3
20.2 0.3
23.0 0.3
a Values are arithmetical means of all inhibitor–protease complexes following BBI modality and errors are standard deviations. b Measured using a
threshold of 6.25 kJ mol−1 c Calculated using ΔBGcalc = Phit × ΔBGdock d Calculated using eqn (7). Errors of ΔBGexp were calculated by propagation
. .
of errors (ESI†).
Fig. 6 Predicted structures of compounds 1 (A), 2 (B), 3 (C), 4 (D), 5 (E), and 6 (F) in complex with matriptase (grey surface) as an overlay with
reported crystal structure 3P8F (white sticks).1 Blue: nitrogen, green: sulfur, red: oxygen, yellow: carbon, hydrogen is omitted for clarity. Measured
RMSD values for inhibitor backbones compared to 3P8F are given in Å.
values of ΔBGdock upon binding to both enzymes. Never-
Comparison of in vitro and in silico data
theless, the probability of 1 to actually interact under the
mode revealed by crystal structures was significantly lower
than for compounds 2 and 4. Thus, flexibility within the
region of the secondary loop might be crucial for the inhibitor
To take into account the reduced probabilities for SFTI-1 deriva-
tives to bind to matriptase according to elucidated crystal struc-
tures, we directly weighted all ΔBGdock by corresponding Phit to
calculate in silico binding affinities ΔBGcalc (Table 3). This pro-
to be effectively transferred from solution (simulation) into
cedure allowed evaluating relative potencies of the investigated
the active site of the proteases (docking). This effect is far
compounds towards examined proteases. Additionally, experi-
more pronounced in the case of matriptase, probably due to
mental inhibition constants Ki were used to calculate free ener-
the mentioned reduced space around residue 12 of inhibitors
1–6.
Monocyclic inhibitor 2 seems to provide a good compromise
gies of binding ΔBGexp using eqn (7) (R meaning the universal
gas constant and T the absolute temperature).
between flexibility and proper backbone arrangement. Hence, it
forms very stable complexes with both proteases at a high
probability.
ΔBGexp ¼ ꢁRT ln Ki
ð7Þ
This journal is © The Royal Society of Chemistry 2012
Org. Biomol. Chem., 2012, 10, 7753–7762 | 7759

View Article Online
Conclusion
Our in vitro data show that the in vivo application of the wild-
type SFTI-1 as a therapeutic agent targeting matriptase is limited
due to its drastically reduced potency at physiological pH.
Nevertheless, as reported in previous works, the SFTI-1 frame-
work provides an appropriate starting point for lead compound
optimization towards inhibition of pharmacologically relevant
proteases.15,36–38 With the focus on matriptase as the target
enzyme, we suggest the open-chain variant SFTI-1[1,14] as a
promising scaffold rather than its bicyclic ancestor. Based on
these findings, we developed SFTI-1[1,14] derivatives with sig-
nificantly improved affinity and selectivity towards the addressed
TTSP; these results will be published elsewhere.
Additionally, triazole-bridged matriptase inhibitor 3 provides
the best kinetic properties within the set of peptidomimetics 3–6.
Its highly constrained side-chain macrocyclic motif might also
be a valuable asset when redox stability is of particular impor-
tance within the pharmaceutical context.
Our in silico experiments provided a plausible explanation for
the potencies observed in vitro. Despite the intrinsic limitations
and simplifications of the force field approach, computed
binding affinities ΔBGcalc represented satisfactory qualitative esti-
mates of values determined by enzyme assays. Structural analy-
sis of the simulation and docking experiments indicated the
importance of the terminal regions of SFTI-1 derivatives upon
matriptase binding/inhibition. Higher conformational constraints
lead to better interaction profiles while lowering the overall prob-
ability to bind matriptase following the expected modality.
Nevertheless, changes within the structure of the BBI macro-
cycle affected inhibitory potency in the most severe manner.
However, further refinement of force field parameters for triazo-
lyl moieties within peptidic molecules using additional X-ray
and/or NMR experiments might be necessary for more precise
calculations. Additionally, to confirm the predicted entropic
effects on inhibitory potency, isothermal calorimetry could be
conducted.39
Fig. 7 Bar chart of in vitro ΔBGexp (black) and in silico ΔBGcalc (grey)
for complexes of compounds 1–6 with trypsin (A) and matriptase (B).
Rankings of binding affinities for both in vitro and in silico experiments
are indicated by attic numerals (best to worst: I.–VI.).
Thus, the collected in silico data were compared with corre-
sponding in vitro results (Fig. 7).
The applied simulation/docking procedure enabled to fully
reproduce the outcome of the enzyme assays qualitatively. Thus,
ranking investigated inhibitors 1–6 according to their affinity
(ΔBGcalc or ΔBGexp) resulted in the same relative order for both
proteases (Fig. 7). Generally, absolute values of ΔBGcalc for the
inhibition of matriptase were higher compared to those deter-
mined by in vitro experiments. Since calculated binding affinities
can only be regarded as rough estimates, both ΔBGcalc and
ΔBGexp are in astonishingly good agreement.
Experimental
A similar result was achieved for trypsin inhibition. However,
the most significant difference between in vitro and in silico data
was observed for the weakest inhibitor 5 (Fig. 7A). Thus, it is
probable that the applied docking procedure overestimates the
influence of the massive distortion of the BBI loop in solution
on trypsin affinity. The dramatic displacement of the peptide
backbone may actually be attenuated through dynamic inter-
actions of 5 with the surface of the protease. Nevertheless, the
possibility for minor backbone readjustments was provided by
energy minimization between the semi-flexible and the rigid
docking steps. An additional time-intensive MD simulation pro-
cedure might be required to cover the dynamic nature of ligand–
receptor interactions as well as the influence of solvent mol-
ecules on the investigated complex structures. Furthermore,
effects of transient or permanent breakage and formation of
covalent bonds have to be ignored within the classical force field
approach. Such reactions might have a significant impact in the
investigated system, as dynamic hydrolysis/recyclization equili-
brium has been reported for trypsin inhibition by SFTI-1.35
Chemicals
All Fmoc-protected amino acids, resins for solid phase peptide
synthesis (SPPS), solvents, and other chemicals were purchased
from Agilent (Varian), Bachem, Fischer Chemicals, Iris Biotech,
Novabiochem, Roth and Sigma-Aldrich.
Peptide synthesis
Linear peptide precursors were assembled by automated micro-
wave-assisted SPPS using the Fmoc strategy.40,41
Bicyclic SFTI-1 (1) was synthesized via solution-phase back-
bone macrocyclization of linear precursor H-Arg(Pbf)-Cys(Trt)-
Thr(tBu)-Lys(Boc)-Ser(tBu)-Ile-Pro-Pro-Ile-Cys(Trt)-Phe-Pro-Asp-
(tBu)-Gly-OH using C-terminal PyBOP/HOBt activation followed
by acidolytic side-chain deprotection, oxidative disulfide bond
formation and chromatographic isolation.
Synthesis and characterization of monocyclic SFTI-1[1,14]
(2) and triazole-containing peptidomimetics 4–6 have been
7760 | Org. Biomol. Chem., 2012, 10, 7753–7762
This journal is © The Royal Society of Chemistry 2012

View Article Online
previously reported.13 New peptidomimetic compound 3 posses-
sing a 1,5-disubstituted 1,2,3-triazolyl side chain macrocycliza-
tion motif was synthesized using a similar approach.13 First, the
peptide resin containing unnatural amino acids azidoalanine
(Aza) and propargylglycine (Pra) (Fmoc-Aza-Thr(tBu)-Lys
(Boc)-Ser(tBu)-Ile-Pro-Pro-Ile-Pra-Phe-Pro-Asp(tBu)-resin) was
synthesized. Then, the triazole bridge was installed via on-
support ruthenium(II)-catalyzed azide–alkyne cycloaddition
(RuAAC) followed by coupling of N-terminal residues Gly1 and
Arg2. Acidolytic cleavage and chromatographic isolation
followed.
Force field customization
All in silico experiments were performed with the YASARA
structure package (YASARA Biosciences). For the precise hand-
ling of triazole-bridged peptides during in silico calculations,
new topology entries and atom types for the involved carbon and
nitrogen species have been implemented into the YASARA2
force field description file. Thus, amino acid residues with 1,4-
and 1,5-disubstituted 1,2,3-triazoles in the side chain of com-
pounds 3–6 were covered. Corresponding equilibrium bond
lengths, bond angles and dihedrals have been deduced from
reported crystal structures.46,47 Values for force constants and
scaling factors were adopted from similar aromatic and hetero-
aromatic structures (e.g. imidazole) within the original force
field. The full set of parameters used to calculate disubstituted
1,2,3-triazoles is given in the ESI.†
For experimental details and analytical data on bicyclic
SFTI-1 (1) and new triazole-bridged compound 3 refer to ESI.†
Recombinant production of matriptase
Molecular dynamics simulation and docking
BL21-CodonPlus (DE3)-RP competent E. coli cells (Stratagene)
were transformed with expression vector pET42dest-His-hMatI
(cd)596–855. This vector was constructed for overexpression of
the zymogen sequence of the catalytic domain of matriptase
as the following fusion: Met-Ser-Tyr-Tyr-His6-aa596–855.
Inclusion body production, protein denaturation, purification and
refolding/autoprotolytic activation were conducted according to
reported procedures (see ESI, Fig. S1 and S2† for details).11,25,42
Substrate-specific catalytic activity of the purified enzyme
was verified through the hydrolysis of chromogenic substrate
Boc-QAR-pNA and quantified via active-site titration using
4-methylumbelliferyl-p-guanidinobenzoate (MUGB) (see ESI,
Fig. S3†).43,44
Solution NMR structures of SFTI-1 (PDB ID: 1JBL) and SFTI-1
[1,14] (PDB ID: 1JBN) were used as initial templates for com-
pounds 1–6.33 Triazole bridges of peptidomimetics 3–6 were
modelled into 1JBN to match the topology entry within the cus-
tomized force field. Each compound was then subjected to a
two-step simulation and docking procedure performed via a self-
written command sequence (macro). The design of the in silico
experiment is sketched in Fig. 4.
Following point charge assignment via semi-quantitative
quantum mechanics,48 an energy minimization was conducted
with initial structures of compounds 1–6 in 0.9% (m/v) NaCl
(aq.) at pH 7.6. To prevent structures from being trapped in the
local minimum of the template coordinates, all compounds were
then simulated at 373 K and the scaling factor for non-bonded
interactions was reduced to 10%. This parameter was stepwise
re-adjusted to 100% within 500 ps. After 1 ns of simulation, the
temperature was set to 298 K and energy-minimized structures
were simulated for another 9 ns. Trajectories (snapshots) were
saved every 10 ps. Thus, 100 models for each individual com-
pound 1–6 from the last nanosecond of the simulation were
extracted and used for the docking experiment. Semi-flexible
docking to the active sites of trypsin and matriptase with fixed
ligand backbones and adjusted side-chain atoms of Lys5 was
performed to identify the most probable binding geometry using
the AutoDock 4 algorithm with 999 runs.18 The highest ranked
complex structure was subjected to an energy minimization with
fixed receptor atoms followed by another energy minimization
step with a fully unrestrained ligand–receptor complex. The
resulting structures were then used for the rigid docking experi-
ment (additional 999 runs) to compute virtual free energies of
binding ΔBG with AutoDock.
Inhibition assays
Apparent (substrate-dependent) inhibition constants Kiapp of
SFTI-1-derivatives 1–6 against active-site-titrated bovine trypsin
([E] = 0.5 nM) and matriptase ([E] = 0.9 nM) were determined
by recording kinetic curves of the enzymatic hydrolysis of
chromogenic substrate Boc-QAR-pNA ([S] = 250 μM) at differ-
ent concentrations of inhibitor [I] (pH 7.6 or pH 8.5) using a
Tecan GENios microplate reader.13 Calculated relative initial
velocities of the residual proteolytic reaction (v/v0) were plotted
against [I]. Eqn (3) for tight-binding inhibitors was fitted by
non-linear regression to the resulting dose–response curves using
the Marquardt–Levenberg algorithm of Sigma Plot 11.29
Substrate-independent inhibition constants Ki were calculated
from Kiapp and the Michaelis–Menten constants KM for the cor-
responding enzyme–substrate complex using eqn (6).25,34
KM for Boc-QAR-pNA in complex with matriptase at pH 7.6
and pH 8.5 were calculated by Lineweaver–Burk plots as 236.8
56.1 μM and 66.6 16.0 μM, respectively. KM for trypsin as
well as the method for the estimation of errors for each individ-
ual Ki have been recently reported.34
Acknowledgements
This work was supported by the Deutsche Forschungs-
gemeinschaft through grant Ko 1390/9-1 and by BMBF. We
thank Prof. Dr Kay Hamacher (AG Computational Biology &
Simulation, Technische Universität Darmstadt) for discussion
concerning in silico experiments and for reading the manuscript.
In the case of reversible competitive inhibition Ki equals the
dissociation constant Kd of the inhibitor–enzyme complex.45
Thus, the free energy of binding ΔBG can be calculated from Ki
using eqn (7).
This journal is © The Royal Society of Chemistry 2012
Org. Biomol. Chem., 2012, 10, 7753–7762 | 7761

View Article Online
22 J. L. Wike-Hooley, J. Haveman and H. S. Reinhold, Radiother. Oncol.,
1984, 2, 343–366.
References
23 I. F. Tannock and D. Rotin, Cancer Res., 1989, 49, 4373–4384.
24 P. Montcourrier, I. Silver, R. Farnoud, I. Bird and H. Rochefort, Clin.
Exp. Metastasis, 1997, 15, 382–392.
25 A. Desilets, J. M. Longpre, M. E. Beaulieu and R. Leduc, FEBS Lett.,
2006, 580, 2227–2232.
26 F. Beliveau, A. Desilets and R. Leduc, FEBS J., 2009, 276, 2213–2226.
27 Y. Cheng and W. H. Prusoff, Biochem. Pharmacol., 1973, 22, 3099–
3108.
28 R. A. Copeland, D. Lombardo, J. Giannaras and C. P. Decicco, Bioorg.
Med. Chem. Lett., 1995, 5, 1947–1952.
1 C. Yuan, L. Q. Chen, E. J. Meehan, N. Daly, D. J. Craik, M. D. Huang
and J. C. Ngo, BMC Struct. Biol., 2011, 11.
2 S. Luckett, R. S. Garcia, J. J. Barker, A. V. Konarev, P. R. Shewry,
A. R. Clarke and R. L. Brady, J. Mol. Biol., 1999, 290, 525–533.
3 Y. Q. Long, S. L. Lee, C. Y. Lin, I. J. Enyedy, S. M. Wang, P. Li,
R. B. Dickson and P. P. Roller, Bioorg. Med. Chem. Lett., 2001, 11,
2515–2519.
4 M. L. J. Korsinczky, R. J. Clark and D. J. Craik, Biochemistry, 2005, 44,
1145–1153.
5 P. Li, S. Jiang, S. L. Lee, C. Y. Lin, M. D. Johnson, R. B. Dickson,
C. J. Michejda and P. P. Roller, J. Med. Chem., 2007, 50, 5976–5983.
6 Y. E. Shi, J. Torri, L. Yieh, A. Wellstein, M. E. Lippman and
R. B. Dickson, Cancer Res., 1993, 53, 1409–1415.
29 J. F. Morrison, Biochim. Biophys. Acta, Enzymol., 1969, 185, 269–286.
30 J. W. Williams and J. F. Morrison, Methods Enzymol., 1979, 63, 437–
467.
31 W. R. Greco and M. T. Hakala, J. Biol. Chem., 1979, 254, 12104–12109.
32 D. J. Murphy, Anal. Biochem., 2004, 327, 61–67.
33 M. L. J. Korsinczky, H. J. Schirra, K. J. Rosengren, J. West,
B. A. Condie, L. Otvos, M. A. Anderson and D. J. Craik, J. Mol. Biol.,
2001, 311, 579–591.
34 M. Tischler, D. Nasu, M. Empting, S. Schmelz, D. W. Heinz,
P. Rottmann, H. Kolmar, G. Buntkowsky, D. Tietze and O. Avrutina,
Angew. Chem., Int. Ed., 2012, 51, 3708–3712.
35 U. C. Marx, M. L. J. Korsinczky, H. J. Schirra, A. Jones, B. Condie,
L. Otvos and D. J. Craik, J. Biol. Chem., 2003, 278, 21782–21789.
36 A. Lesner, A. Legowska, M. Wysocka and K. Rolka, Curr. Pharm. Des.,
2011, 17, 4308–4317.
37 E. Zablotna, A. Jaskiewicz, A. Legowska, H. Miecznikowska, A. Lesner
and K. Rolka, J. Pept. Sci., 2007, 13, 749–755.
7 K. Uhland, Cell. Mol. Life Sci., 2006, 63, 2968–2978.
8 K. List, Future Oncol., 2009, 5, 97–104.
9 J. M. Milner, A. Patel, R. K. Davidson, T. E. Swingler, A. Desilets,
D. A. Young, E. B. Kelso, S. T. Donell, T. E. Cawston, I. M. Clark,
W. R. Ferrell, R. Plevin, J. C. Lockhart, R. Leduc and A. D. Rowan,
Arthritis Rheum, 2010, 62, 1955–1966.
10 I. Seitz, S. Hess, H. Schulz, R. Eckl, G. Busch, H. P. Montens, R. Brandl,
S. Seidl, A. Schomig and I. Ott, Arterioscler., Thromb., Vasc. Biol., 2007,
27, 769–775.
11 T. Steinmetzer, A. Schweinitz, A. Sturzebecher, D. Donnecke, K. Uhland,
O. Schuster, P. Steinmetzer, F. Muller, R. Friedrich, M. E. Than, W. Bode
and J. Sturzbecher, J. Med. Chem., 2006, 49, 4116–4126.
12 C. J. Farady, J. Sun, M. R. Darragh, S. M. Miller and C. S. Craik, J. Mol.
Biol., 2007, 369, 1041–1051.
38 J. D. McBride, H. N. M. Freeman and R. J. Leatherbarrow,
Eur. J. Biochem., 1999, 266, 403–412.
39 S. Leavitt and E. Freire, Curr. Opin. Struct. Biol., 2001, 11, 560–566.
40 M. Amblard, J. A. Fehrentz, J. Martinez and G. Subra, Mol. Biotechnol.,
2006, 33, 239–254.
13 M. Empting, O. Avrutina, R. Meusinger, S. Fabritz, M. Reinwarth,
M. Biesalski, S. Voigt, G. Buntkowsky and H. Kolmar, Angew. Chem.,
Int. Ed., 2011, 50, 5207–5211.
14 U. Essmann, L. Perera, M. L. Berkowitz, T. Darden, H. Lee and
L. G. Pedersen, J. Chem. Phys., 1995, 103, 8577–8593.
15 J. E. Swedberg, S. J. de Veer, K. C. Sit, C. F. Reboul, A. M. Buckle and
J. M. Harris, PLOS One, 2011, 6.
41 S. Park, S. Gunasekera, T. L. Aboye and U. Goransson, Int. J. Pept. Res.
Ther., 2010, 16, 167–176.
42 T. Takeuchi, M. A. Shuman and C. S. Craik, Proc. Natl. Acad.
Sci. U. S. A., 1999, 96, 11054–11061.
43 T. Chase and E. Shaw, Biochemistry, 1969, 8, 2212–2224.
44 G. W. Jameson, D. V. Roberts, R. W. Adams, W. S. A. Kyle and
D. T. Elmore, Biochem. J., 1973, 131, 107–117.
16 E. Krieger, K. Joo, J. Lee, J. Lee, S. Raman, J. Thompson, M. Tyka,
D. Baker and K. Karplus, Proteins: Struct., Funct., Bioinf., 2009, 77,
114–122.
17 W. Brandt, T. Herberg and L. Wessjohann, Biopolymers, 2011, 96, 651–
668.
45 W. R. Greco and M. T. Hakala, J. Biol. Chem., 1979, 254, 2104–2109.
46 M. R. Krause, R. Goddard and S. Kubik, J. Org. Chem., 2011, 76, 7084–
7095.
47 A. Brik, J. Alexandratos, Y. C. Lin, J. H. Elder, A. J. Olson,
A. Wlodawer, D. S. Goodsell and C. H. Wong, ChemBioChem, 2005, 6,
1167–1169.
18 G. M. Morris, D. S. Goodsell, R. S. Halliday, R. Huey, W. E. Hart,
R. K. Belew and A. J. Olson, J. Comput. Chem., 1998, 19, 1639–1662.
19 W. R. Rypniewski, A. Perrakis, C. E. Vorgias and K. S. Wilson, Protein
Eng., Des. Sel., 1994, 7, 57–64.
20 S. N. S. Murthy, J. Kostman and V. P. Dinoso, Dig. Dis. Sci., 1980, 25,
289–294.
48 A. Jakalian, D. B. Jack and C. I. Bayly, J. Comput. Chem., 2002, 23,
1623–1641.
21 T. H. Bugge, T. M. Antalis and Q. Y. Wu, J. Biol. Chem., 2009, 284,
23177–23181.
7762 | Org. Biomol. Chem., 2012, 10, 7753–7762
This journal is © The Royal Society of Chemistry 2012