Synthesis, evaluation and structure-activity relationship of new 3-carboxamide coumarins as FXIIa inhibitors

Article abstract of DOI:10.1016/j.ejmech.2016.01.023

Inhibitors of the coagulation factor XIIa (FXIIa) are attractive to detail the roles of this protease in hemostasis and thrombosis, to suppress artifact due to contact pathway activation in blood coagulation assays, and they are promising as antithrombotic therapy. The 3-carboxamide coumarins have been previously described as small-molecular-weight FXIIa inhibitors. In this study, we report a structure-activity relationship (SAR) study around this scaffold with the aim to discover new selective FXIIa inhibitors with an improved physico-chemical profile. To better understand these SAR, docking experiments were undertaken. For this purpose, we built an original hybrid model of FXIIa. This model has the advantage to gather the best features from the recently published crystal structure of FXIIa in its zymogen form and a more classical homology model. Results with the hybrid model are encouraging as they help understanding the activity and selectivity of our best compounds.

Full text of DOI:10.1016/j.ejmech.2016.01.023

Accepted Manuscript
Synthesis, evaluation and structure-activity relationship of new 3-carboxamide
coumarins as FXIIa inhibitors
Charlotte Bouckaert, Silvia Serra, Grégoire Rondelet, Eduard Dolušić, Johan
Wouters, Jean-Michel Dogné, Raphaël Frédérick, Lionel Pochet
PII:
S0223-5234(16)30024-1
10.1016/j.ejmech.2016.01.023
EJMECH 8313
DOI:
Reference:
To appear in: European Journal of Medicinal Chemistry
Received Date: 28 August 2015
Revised Date: 15 December 2015
Accepted Date: 15 January 2016
Please cite this article as: C. Bouckaert, S. Serra, G. Rondelet, E. Dolušić, J. Wouters, J.-M. Dogné,
R. Frédérick, L. Pochet, Synthesis, evaluation and structure-activity relationship of new 3-carboxamide
coumarins as FXIIa inhibitors, European Journal of Medicinal Chemistry (2016), doi: 10.1016/
j.ejmech.2016.01.023.
This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to
our customers we are providing this early version of the manuscript. The manuscript will undergo
copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please
note that during the production process errors may be discovered which could affect the content, and all
legal disclaimers that apply to the journal pertain.

ACCEPTED MANUSCRIPT

ACCEPTED MANUSCRIPT
Synthesis, evaluation and structure-activity
relationship of new 3-carboxamide coumarins as
FXIIa inhibitors
1
1
2
3
2
Charlotte Bouckaert , Silvia Serra
,
Grégoire Rondelet
,
Eduard Dolušić
,
Johan Wouters
,
1
4
1*
Jean-Michel Dogné , Raphaël Frédérick
,
Lionel Pochet
1
Department of Pharmacy, Namur Medicine & Drug Innovation Center (NAMEDIC), Namur
Research Institute for Life Sciences (NARILIS), University of Namur, 61, Rue de Bruxelles,
2
5
000 Namur, Belgium. Department of Chemistry, Namur Medicine & Drug Innovation Center
(NAMEDIC), Namur Research Institute for Life Sciences (NARILIS), University of Namur, rue
3
de Bruxelles, 61, 5000 Namur, Belgium. Department of Organic Chemistry, Namur Medicine &
Drug Innovation Center (NAMEDIC), Namur Research Institute for Life Sciences (NARILIS),
4
University of Namur, 61, Rue de Bruxelles, 5000 Namur, Belgium. Medicinal Chemistry
Research Group (CMFA), Louvain Drug Research Institute (LDRI), Université Catholique de
Louvain, 73, Avenue E. Mounier, 1200 Brussels, Belgium
*
8
Corresponding author: Lionel Pochet. E-mail: lionel.pochet@unamur.be. Phone: +32 (0)
1/72.42.90
1

ACCEPTED MANUSCRIPT
ABSTRACT
Inhibitors of the coagulation factor XIIa (FXIIa) are attractive to detail the roles of this protease
in hemostasis and thrombosis, to suppress artifact due to contact pathway activation in blood
coagulation assays, and they are promising as antithrombotic therapy. The 3-carboxamide
coumarins have been previously described as small-molecular-weight FXIIa inhibitors. In this
study, we report a structure-activity relationship (SAR) study around this scaffold with the aim to
discover new selective FXIIa inhibitors with an improved physico-chemical profile. To better
understand these SAR, docking experiments were undertaken. For this purpose, we built an
original hybrid model of FXIIa. This model has the advantage to gather the best features from
the recently published crystal structure of FXIIa in its zymogen form and a more classical
homology model. Results with the hybrid model are encouraging as they help understanding the
activity and selectivity of our best compounds.
Keywords: Coumarins; Factor XIIa; Factor XII; FXII; antithrombotic agents; benzopyrans
2

ACCEPTED MANUSCRIPT
1. INTRODUCTION
*
The blood coagulation system is essential for limiting blood loss at sites of vascular injury
hemostasis), but excessive and uncontrolled activation leads to thrombotic complications such
(
as myocardial infarction or deep venous thrombosis. Since thrombotic disorders are a major
cause of morbidity and mortality [1], extensive research has been made in the field of
anticoagulation therapy with the hope to tackle these life-threatening diseases. Currently, the
most commonly used anticoagulants act by inhibiting thrombin or factor Xa (FXa), or by
lowering the plasma levels of the precursors of these key enzymes [2]. While having largely
proven its antithrombotic efficacy, the current anticoagulant therapy is associated with an
unavoidable bleeding risk. Indeed, existing drugs target key components of the coagulation
cascade and they do not distinguish between thrombin generation contributing to thrombosis
from thrombin generation required for hemostasis [2].
Coagulation factor XII (FXII) is a trypsin-like serine protease able to trigger the intrinsic
pathway of coagulation through the contact with negatively charged surfaces (reaction known as
the contact activation). In contrast with thrombin and FXa, FXII has long been neglected as an
attractive target since it was believed to have no function in hemostasis. This was supported by
the clinical observation showing that its deficiency does not lead to bleeding tendency in humans
[2, 3]. However, investigations with FXII-null mice completely challenged this dogma. Indeed, it
was shown that FXII-deficient mice are protected against thrombosis without suffering from
*
Abbreviations: ADME: absorption, distribution, metabolism and excretion, FXIIa: activated
coagulation factor XII, FXa: activated coagulation factor X, THR: thrombin, FVIIa: activated
factor VII, HPLC-UV: high performance liquid chromatography coupled with ultraviolet
detector, (plasma) kall: (plasma) kallikrein, PDB: Protein DataBase, p-NA: para-nitroaniline,
SAR: structure-activity relationship, TF: tissue factor, t-PA: tissue plasminogen activator, u-PA:
urokinase plasminogen activator.
3

ACCEPTED MANUSCRIPT
spontaneous or prolonged injury-related bleedings [4-9]. This raised the possibility to design
safer antithrombotic agents without hemostatic defects.
So far, different types of inhibitors of FXII or its activated form (FXIIa) have been developed
[10] including proteins [11-13], synthetic peptides [14], antibodies [15, 16] or antisense
nucleotides [17]. Of interest, all of them showed promising results in in vivo models of
thrombosis. Indeed, treated animals were protected against thrombosis without showing a
bleeding phenotype [2]. Given the diversity of inhibitors and thrombosis models investigated, it
can be assumed that inhibiting FXII or FXIIa is an innovative strategy to develop antithrombotic
agents that would offer dissociation between hemostasis and thrombosis.
Beyond a safer bleeding profile, such agents could afford new clinical benefits. Indeed, the
leading perspective for the FXII or FXIIa inhibition is the prevention of contact-mediated
thrombosis induced by medical devices such as catheter or cardiopulmonary bypass system [16-
1
8]. This statement is supported by studies with Corn Trypsin Inhibitor (CTI) [18], antisense
nucleotides [17] or the 3F7 antibody [16]. More recently, it has been stated that strategies that
target FXII or FXIIa would be of valuable interest in patients with mechanical heart valves since
thrombosis in these patients is driven by the contact pathway [19].
For acute or sub-chronic applications (i.e. catheter thrombosis, cardiopulmonary bypass, knee
or hip replacement surgery), the parenteral administration of proteins, peptides, antibodies or
RNA-derived products would be suitable. In contrast, for chronic indications as encountered in
patients exposed to mechanical heart valves or ventricular assisted-devices, the development of
small-molecular weight inhibitors that would be taken orally is preferable. Coumarin derivatives
were previously described by our group as serine proteases inhibitors [20-24] and the 3-
carboxamide coumarins were especially designed as small-molecular-weight FXIIa inhibitors
4

ACCEPTED MANUSCRIPT
[25]. Unfortunately, these compounds lacked of activity in an in vivo model of thrombosis [26]
probably due to poor pharmacokinetics and/or weak potency. Based on this, we aimed in the
present study to modulate the coumarin scaffold in an effort to obtain new selective FXIIa
inhibitors with optimized physico-chemical properties. Moreover, we analyzed the interactions
between the compounds and the target in order to better understand the structure-activity-
relationships. Since the three-dimensional structure of FXIIa is available in a zymogen
configuration, we built an original “hybrid model” of FXIIa to sustain our docking experiments.
2. RESULTS AND DISCUSSION
2.1 Chemistry
In our previous study, we identified compound 1 (Figure 1) as a promising starting point due
to its activity and selectivity [25]. This compound is a 3-phenylamide coumarin substituted by a
6
-chloro-8-bromo group. The amide linker between the coumarin and the side chain in the 3-
position and the absence of a chloromethyl function in the 6-position undoubtedly contribute to
the selectivity of this compound towards related serine proteases such as thrombin (THR), factor
VIIa (FVIIa)/tissue factor (TF) complex, factor Xa (FXa) and plasma kallikrein (Kall) that
require these substitution patterns for inhibition [25]. Based on these considerations, we
performed modulations on the 3-carboxamide coumarin scaffold in the 3,6-positions (Figure 1).
These latter aimed to afford selective FXIIa inhibitors with improved physico-chemical
properties. To achieve this, two series (series A and B) were designed. In the A-series, a basic
group (-NR ) was inserted on either positions while in the B-series, an oxygen-containing group
2
(-OH or -COOH) was introduced.
5

ACCEPTED MANUSCRIPT
Figure 1. The starting compound 1 and modulations on positions 3 or 6 of the 3-carboxamide
scaffold in series A and in series B.
Compounds of the A-series substituted with various amines in the 6-position were obtained
from the starting material 2 (Scheme 1). This compound was synthesized according to a
previously described procedure [25]. In this series, the phenyl function found in compound 1 was
kept in the 3-position. Compound 3 was obtained by a Delepine’s reaction [27] whereas
compounds 4-7 were synthesized following nucleophilic substitution reactions with the
appropriate amines.
6

ACCEPTED MANUSCRIPT
Scheme 1. Introduction of a basic group in position 6 of the 3-carboxamide coumarin
a
scaffold .
a
3
Reagents and conditions: (i) Hexamethylenetetramine/CHCl , 70°C, 18h, then HCl
6
M/ethanol/H O, 90°C, 90 min; (ii) R-NH , ethanol, microwave (100°C, 10 min).
2
2
Modulations of the 3-position with a basic group (16-22) are depicted on Scheme 2.
Compounds 8-10 were obtained as previously reported by a Knoevenagel-type reaction using the
appropriate and commercially available salicylaldehyde [28, 29]. Saponification of these
compounds afforded the 3-carboxylic acid derivatives (11-13). Treatment of compound 13 with
thionyl chloride provided the acyl chloride (14) that was further reacted with the mono N-Boc-
protected ethylenediamine to give 15. Deprotection of the latter in TFA afforded compound 16
[22, 28, 29]. For compounds 17-22, pivaloyl chloride was preferred as chlorinating agent [30]
because it speeded up the amide synthetic route. Briefly, the 6,8-halo-2-oxo-2H-1-benzopyran-3-
carboxylic acids (11-13) were reacted with triethylamine and pivaloyl chloride in
7

ACCEPTED MANUSCRIPT
dichloromethane to form the reactive acyl halide intermediates that were further reacted with the
appropriate amine to afford the desired target compounds 17-22.
Scheme 2. Introduction of a basic group (16-22) or an oxygen-based group in position 3 (23-
a
2
8) of the 3-carboxamide coumarin scaffold .
a
Reagents and conditions: (i) NaOH 10% (w/v), reflux, 30 min, then HCl 6M; (ii) SOCl2,
reflux, 3h; N-Boc-ethylenediamine/dioxane, room temperature, 30 min; (iv) TFA, CH Cl , room
2
2
temperature, 10 min; (v) pivaloyl chloride/CH Cl , Et N, from 0°C to room temperature, 30 min,
2
2
3
2 2 2 3
then R-NH , from 0°C to room temperature, 1-2h; (vi) pivaloyl chloride/CH Cl , Et N, from 0°C
to room temperature, 30 min, then R’-Aniline, from 0°C to room temperature, 2-7h.
8

ACCEPTED MANUSCRIPT
The same synthetic sequence with pivaloyl chloride as chlorinating reagent allowed the
synthesis of compounds 23-28 from compound 13 (Scheme2) that possess a hydroxyl or
carboxylic acid group in the 2-, 3-, 4-position with or without a chlorine atom in the 4- or 5-
position as substituents on the lateral phenylamide.
Compounds with a hydroxyl function in the 6-position were also prepared (Scheme 3). The
first step consisted in the synthesis of N,N’-diphenyl malonamide 30 by reaction of commercial
diethyl malonate 29 with aniline under microwaves irradiations [31]. Then, a Knoevenagel
reaction between compound 30 and the 5-(hydroxymethyl)-salicylaldehyde [22] 31 afforded
compound 32.
Scheme 3. Introduction of an oxygen-based group on position 6 of the 3-carboxamide coumarin
a
scaffold .
a
Reagents and conditions: (i) aniline, microwave (180°C, 30 min); (ii) N,N’-
diphenylmalonamide (30), piperidine, acetic acid, ethanol, reflux, 21h
9

ACCEPTED MANUSCRIPT
2
.2 Physico-chemical properties
Small-molecular-weight inhibitors of FXIIa that could be taken orally would be ideal for
thrombosis prevention in patients with mechanical heart valves. With that application in mind,
we decided to evaluate in silico the absorption profile of our compounds at the very early stage
of their development. To this end, we preconized an evaluation of the absorption characteristics
of our compounds using the Traffic Lights system elaborated by Lobell et al. [32] with a view to
prioritize our compounds with regard to potential ADME-related liabilities. The algorithm
developed by Lobell (see Supplementary materialfor more information) proposes a set of ‘Traffic
Lights’ (TLs) [32] based on five physico-chemical properties: (i) molecular size as assessed by
MW corrected for halogens, (ii) lipophilicity, solubility at pH 6.5, (iv) polarity as assessed by the
polar surface area (PSA), and (v) number of rotatable bonds. Taken together, the TL’s values
afford an in silico oral PhysChem score which ranges from 0 to 10. A low score means that
physico-chemical properties of the compound are more favorable for the development of an
orally administered drug [32, 33].
As observed in Table 1, introduction of a basic group (series A) on the 3-carboxamide-coumarin
scaffold improves the oral PhysChem score compared to compound 1 whose score is 3. Indeed,
nine compounds from the A-series (3, 6, 16-22) have an oral PhysChem score of 0 and three
have a score of 1 (4, 5, 7). The PhysChem score of 1 is explained by a moderate Log P and
solubility at intestinal pH (except for compound 17 which has an optimal Log P but a poor
solubility). Modulations in the B-series were focused on the introduction of an oxygen-
containing group on the coumarin scaffold. When a hydroxymethyl group is inserted in the 6-
position (32), the PhysChem score reaches a value of 0 (Table 2). For compounds 23-28, a
hydroxyl- or a carboxyl group was introduced as substituent on the phenyl side chain. This
10

ACCEPTED MANUSCRIPT
modulation does not improve the PhysChem score compared to compound 1 (Table 2).
Moderate Log P and poor solubility (except for compound 28 which has a high log P and a
moderate solubility) are responsible of the oral PhysChem score of 3.
Solubility is an important factor in the absorption process [34] and in our series it seems to be the
most influencing factor on the oral PhysChem score. However, taken together, these results are
consistent with a potential oral administration. Indeed, Lobell et al. observed that their hits (from
HTS) had an average oral PhysChem score of 4.1 while 90% of leads or marketed drugs showed
score lower than 5 [32].
11

ACCEPTED MANUSCRIPT
Table 1. PhysChem score calculation and inhibition percentage on FXIIa for Series A
a
Property predicted with the ACD/Labs software (module Phys-Chem predictions – version
b
c
d
1
2.01). MWcorr = molecular weight corrected for halogens. PSA = polar Surface Area. Solub.
pH 6.5 = solubility at pH 6.5. Rot. Bonds = rotable bonds. Results are expressed in mean ± SD.
e
f
12

ACCEPTED MANUSCRIPT
Table 2. PhysChem score calculation and inhibition percentage on FXIIa for Series B
a
Property predicted with the ACD/Labs software (module Phys-Chem predictions – version
b
c
d
1
2.01). MWcorr = molecular weight corrected for halogens. PSA = polar Surface Area. Solub.
pH 6.5 = solubility at pH 6.5. Rot. Bonds = rotable bonds. Results are expressed in mean ± SD.
e
f
13

ACCEPTED MANUSCRIPT
2.3 Biological profile
The newly synthesized compounds were first screened on FXIIa at a single concentration of
5
0 µM (Table 1 and Table 2) and the inhibitory potency (IC ) was determined for compounds
5
0
displaying at least 85% inhibition at this concentration. Compounds 23, 24 and 27 revealed to be
the three most active derivatives with IC of 5, 8 and 7 µM, respectively (Table 3). These data
5
0
suggest that modulation with an amino function (3-22) does not afford active compounds in
contrast with the introduction of an oxygen-based group which yields six active compounds (23-
2
7). Interestingly, the three most active compounds (23, 24, 27) all have a 2-hydroxyphenyl
substituent at the 3-position of the 3-carboxamide coumarin scaffold.
Then, the selectivity of the compounds was evaluated by investigating the compounds on
thrombin, activated factor X (FXa), complex tissue factor/activated factor VII (TF/FVIIa) and
plasma kallikrein at a concentration of 100 µM. All active compounds on FXIIa have no
inhibition at this concentration against other serine proteases suggesting they are selective (Table
3
).
Table 3. Potency and selectivity of promising compounds
FXIIa IC50
Plasma
kallikrein
Compound
a
Thrombin
TF/FVIIa
FXa
(µM)
b
c
1
16 (9-26)
5 (3-8)
NA
NA
NA
NA
NA
NA
NA
NA
NA
NA
NA
NA
NA
NA
NA
NA
NA
NA
NA
2
2
2
2
2
3
4
5
6
7
NA
NA
NA
NA
NA
8 (5-12)
41 (17-101)
34 (8-142)
7 (4-12)
14

ACCEPTED MANUSCRIPT
a
b
IC50 are expressed in mean and 95% confidence intervals. In literature [25], 4.4 (3.2-6.2) µM
c
in 10% DMSO instead of 5% in this study. NA = not active: inhibition < 10% at 100 µM.
2.4 Molecular modeling
To further explore the interactions between our coumarins and FXIIa, we carried out molecular
modeling experiments. This implied the elaboration of a 3D model of FXIIa to perform docking
experiments. Regarding the numbering of residues, it should be mentioned that in the following
sections, residues will be numbered according to the nomenclature of chymotrypsin
(Supplementary material).
2
.4.1 The « hybrid FXIIa model »
Although FXII is a protein known since almost fifty years, the crystal structure of its catalytic
light chain has only been recently published[35]. In fact, two structures, namely FXIIac and
FXIIc obtained using two different recombinant FXII proteins, were reported by Pathak et al.
(PDB: 4XE4 and 4XDE, respectively) [35].
In our study, as we aimed to investigate the interactions and structure-activity relationships
between FXIIa and our derivatives, we initially intent to use the structure of FXIIac (PDB:
XE4) [35] for docking experiments. However, Pathak et al. reported that their structure displays
4
an unexpected zymogen-like conformation [35]. Based on this consideration, we envisaged to
build a homology model of FXIIa. The homology model was built using the SWISS-MODEL
program [36, 37] and was generated from the crystal structure of the (active) hepatocyte growth
factor activator (HGFA- PDB: 2R0L; 43% similarity).
15

ACCEPTED MANUSCRIPT
Then, we compared this homology model to the crystal structure (PDB: 4XE4) and our
comparison was focused on the orientation of key amino acids. Our analysis (Supplementary
material) revealed that the homology model would be more suitable than the currently available
crystal structure for docking experiments but several adjustments would be needed since key
residues (Tyr99, Arg73, Leu33, Leu59, Leu64, Leu106 and Trp35) are not properly oriented in
this model (Supplementary material). We thus hypothesized that the construction of a
consensus model, that would join the best features from the homology model and the crystal
structure, would be more relevant for docking studies. The building of a consensus model, called
“
FXIIa hybrid model” in this publication, was inspired from the work of Bujnicki and coworkers
[38]. To achieve this, we superimposed the crystal structure and our homology model,
suppressed the loop containing the flipping residues (Tyr99, Arg73, Leu33, Leu59, Leu64,
Leu106 and Trp35) in the homology model and replaced it by the loop from the crystal structure
(Figure 2).
16

ACCEPTED MANUSCRIPT
Figure 2. Building of the FXIIa hybrid model. In this model, we replaced the loop containing the
flipping residues from the homology model by the loop from the crystal structure.
In the FXIIa hybrid model, residues are thus set in their putative active orientation
(Supplementary material). Also, the active site pockets are of particular interest since they
interact with inhibitors. First, the so-called specificity pocket S1, which is mainly characterized
by Asp189 and Tyr228, is conserved in all the trypsin-like serine proteases. In their work,
Emsley and coworkers [35] highlighted the presence of another pocket called S3/4 pocket. In this
pocket, Trp215 is a constant feature that is found in all the coagulation proteases while Tyr99 is
present in a similar orientation in FXII, t-PA, u-PA as well as in the more distant FXa [35].
Met180 is also a constant feature in these proteases (Figure 3). According to Pathak et al., FXIIa
is also characterized by a distinctive H1 pocket located in front of the S1 pocket. The side chains
of Leu33, Leu64, Leu59, Leu106 and Trp35 contribute to the highly hydrophobic character of
this particular pocket. (Figure 3).
17

ACCEPTED MANUSCRIPT
Figure 3. Surface structure of the FXIIa hybrid model. In the S1 pocket, Tyr228 is located at the
bottom. The S3/4 pocket described by Pathak et al.[35] is mainly delimited by Tyr99, Trp215
and Met180. The hydrophobic pocket, called the H1 pocket by Emsley and coworkers[35], is
drawn by the side chains of Leu33, Leu59, Leu64, Leu106 (not shown) and Trp35.
As a final step, the quality of the model was checked with the VERIFY3D tool [38-41]. As
stated by the wed-based server (http://services.mbi.ucla.edu/Verify_3D/), this method analyzes
the compatibility of an atomic model (3D) with its own amino acid sequence (1D). Each residue
is assigned a structural class based on its location and environment (alpha, beta, loop, polar,
apolar, etc). Then a database generated from good structures is used to obtain a score for each of
the 20 amino acids in this structural class. The scores range from -1 to +1 and the model passes
the test when at least 80% of the amino acids scored
http://services.mbi.ucla.edu/Verify_3D/). In our study, the hybrid model passed the test since
6.5% of the amino acids scored 0.2.
≥ 0.2 in the 3D/1D profile
(
9
≥
18

ACCEPTED MANUSCRIPT
2
.4.2 Docking study
To obtain information about target-inhibitor interactions that would help us to understand the
structural requirements for FXIIa inhibition, we performed a docking study using our hybrid
model. The docking simulations were carried out with all assessed compounds (1, 3-7, 16-28, 32)
using the automated GOLD program [42]. For each ligand, twenty solutions were generated and
ranked according to the ChemPLP scoring function. For each compound, the best pose was
retained.
First, the ability of our model to discriminate between active and non-active compounds was
evaluated. Interestingly, we observed that active derivatives (23-25 and 27) have the 6,8-
disubstituents on the coumarin ring pointing towards the Trp35 residue from the H1 pocket
(Figure 4A and 5B) whereas none of the inactive compounds displayed such orientation (Figure
4
B). For the active compounds 1 and 26, the best docking poses proposed a binding mode where
the coumarin ring was located in the S3/4 pocket and the 3-substituent in the S1 pocket.
However, poses from the second cluster (60% and 50% of occurrence for 1 and 26, respectively)
oriented the 6,8-halogen of the coumarin towards the H1 pocket which was in agreement with
the poses of the active compounds 23-25 and 27. Taken together, these results indicate that the
model may be considered as discriminating between active and inactive compounds within the
series presented in this study.
19

ACCEPTED MANUSCRIPT
Figure 4. Poses of active compounds (A) and inactive (B) compounds in the active site of the
hybrid model. (A) Active compounds have the 6,8-substituents of the coumarin ring pointing to
the H1 pocket. (B) None of the inactive compounds have their 6,8-halogens on the coumarin
scaffold directed towards the H1 pocket.
In the second step of our docking study, we aimed to investigate the interactions between the
active compounds and the enzyme in order to better explore the chemical features needed for the
inhibition. To achieve this, we first minimized the ligands in the active site and calculated the
binding energies before visualizing the interactions with the Discovery Studio 4.0 (Accelrys Inc¸
San Diego, California, USA) program. The calculation of the binding energies showed that the
most active compound (23) has the lowest binding energy (Supplementary material)
suggesting that this compound is better stabilized in the active site. On this basis, we decided to
further analyze the interactions between this compound and the enzyme.
As depicted in Figure 5A, the halogens from compound 23 are pointing to Trp35 from the
hydrophobic H1 pocket. Indeed, a
π-alkyl interaction (Figure 5B) was highlighted between the
bromine group in position 8 of the coumarin and the indole of Trp35 (8-BrꢀꢀꢀTrp35, distance =
20

ACCEPTED MANUSCRIPT
4
.69 Å). This bromine atom was also involved in two other hydrophobic interactions with
residues Phe41 (8-BrꢀꢀꢀPhe41, distance = 3.17 Å) and Cys42 (8-BrꢀꢀꢀCys42, distance = 4.54 Å).
The coumarin ring also interacted with the enzyme thanks to the lactone that formed hydrogen
bonds (COlactoneꢀꢀꢀNHGly193, distance = 1.92 Å, and COlactoneꢀꢀꢀCHGln192, distance = 2.76 Å). The
coumarin side chain was also engaged in the establishment of interactions. Indeed, hydrogen
bonds were found between the hydroxyl group and Ser195 (2’-OHꢀꢀꢀOHSer195, distance = 1.84 Å)
or the backbone of Cys220 (2’-OHꢀꢀꢀCOCys220, distance = 1.90 Å). The phenyl and its chlorine
atom both interact with Cys220 thanks to π-alkyl (PhenylꢀꢀꢀCys220, distance = 5.31 Å) and alkyl
interactions (5’-ClꢀꢀꢀCys220, distance = 4.46 Å).
Figure 5. Compound 23 stabilized FXIIa active site (A) and details of interactions (B).
Regarding the stabilization in pockets, compound 23 seems to interact at the entry of the S1
(through interactions with the disulfure bridge Cys191-Cys220) and H1 (through interactions
with Trp35) pockets. The fact that the molecule does not fully fill these pockets could explain its
micromolar range activity. Besides, the H1 pocket is a distinctive feature of FXIIa which could
account for the selectivity of compound 23.
21

ACCEPTED MANUSCRIPT
Finally, we decided to rationalize this selectivity hypothesis in silico. To achieve this, we
superimposed the FXIIa hybrid model-compound 23 complex with the crystal structures of
thrombin (PDB: 1TA2 [43]), FXa (PDB: 2W26 [44]), FVIIa (PDB: 4YT7 [45]) and plasma
kallikrein (PDB: 4OGX [46]).
The active site of thrombin gathers three pockets, namely the S, D and P pockets (Figure 6),
that can be targeted in order to develop inhibitors [24, 43, 47]. When superimposing FXIIa
hybrid model-compound 23 complex with thrombin (Figure 6), we noticed a clash of the
coumarin ring. This clash occurred in the area of the H1 pocket in FXIIa meaning that such
binding in thrombin is impossible. The H1 pocket in FXIIa is represented by Leu33, Leu59,
Leu64, Leu106 and Trp35. In thrombin, we found Leu33, Leu59, Leu64, Met106 and Arg35 but
these residues did not contribute to a pocket.
Figure 6. Superimposition of FXIIa hybrid model-compound 23 complex with thrombin. In the
picture, FXIIa is hidden and only compound 23 is shown for clarity purpose. Thrombin pockets
are the S (Asp189), D (Trp215, Ile174, Leu99) and P (Trp60D, Tyr60A) pockets. The orientation
22

ACCEPTED MANUSCRIPT
adopted by compound 23 due to the superimposition lead to a clash in the structure. Besides, we
previously mentioned that residues Leu33, Leu59, Leu64, Leu106 and Trp35 in FXIIa contribute
to the H1 pocket but in thrombin Leu33, Leu59, Leu64, Met106 and Arg35 are not involved in a
pocket.
The active site of FXa is restricted to the S1 and S4 pockets which implies the design of L-
shaped inhibitors [44]. Consequently, a clash was observed when we superimposed compound
2
3 docked in FXIIa with FXa (Figure 7). Also, as in thrombin, residues Leu33, Leu59, Phe64,
Leu106 and Asn35 did not draw a (unreachable) pocket (Figure 7).
Figure 7. Superimposition of FXIIa hybrid model-compound 23 complex with FXa. In the
picture, FXIIa is hidden and only compound 23 is shown for clarity purpose. Active site of FXa
is restricted to S1 (Asp189) and S4 (Tyr99, Trp215, Phe174) pockets and residues Leu33, Leu59,
23

ACCEPTED MANUSCRIPT
Phe64, Leu106 and Asn35 do not draw an unreachable pocket. This caused a clashed of
compound 23 after superimposition.
Coagulation FVIIa has 5 subsites (S1-S4 and S1’) in its active site [48]. The superimposition
of our FXIIa hybrid model-compound 23 complex with FVIIa crystal structure (PDB: 4YT7)
showed only a small clash in the S1 pocket (Figure 8). This may be not significant since the size
and shape of the FVIIa S1 pocket is very similar to the S1 pocket in thrombin, FXa and trypsin
[48]. As in thrombin and FXa, residues Leu33, Phe59, Leu64, Leu106 and Val35 in FVIIa did
not contribute to a pocket. However, FVIIa possesses a hydrophobic S1’ pocket (with Cys42-
Cys58 disulfide bridge, His57 and Leu41 delimiting the pocket) that could interact with a phenyl
moiety [45]. Considering the orientation of compound 23 due to the superimposition, it would be
likely that the coumarin ring interacts with this pocket. However, to effectively inhibit FVIIa,
inhibitors bind at least 3 subsites of the active site [48, 49]. This fact might explain why this
derivative did not show an activity at 50 µM on this protease.
24

ACCEPTED MANUSCRIPT
Figure 8. Superimposition of FXIIa hybrid model-compound 23 complex with FVIIa. In the
picture, FXIIa is hidden and only compound 23 is shown for clarity purpose. FVIIa has a S1
(Asp189), S2/4 (Trh99, Trp215, His57) and a S1’ pocket (Cys42-Cys58, His57, Leu41).
Residues Leu33, Phe59, Leu64, Leu106 and Val35 do not participate in the formation of a
pocket.
The Figure 9 illustrates the superposition of FXIIa hybrid model-compound 23 complex with
plasma kallikrein. This protease displays S1, S2, S3 and S1’ pockets in its active site [46, 50].
The orientation of compound 23 given by the superimposition shows that the molecule could
interact with S1 and S1’ pockets. This latter involves the side chains of Leu41, Leu60B and
Trp65D, as well as the backbone and side chain of Asp60. This pocket is located in the same area
than the H1 pocket in FXIIa but is different in size and shape. Therefore, it might be possible that
compound 23 inhibits plasma kallikrein but its binding may not be enough to display an activity
25

ACCEPTED MANUSCRIPT
at 50 µM. Here also, residues 33, 59, 64, 106 and 35 (meaning Leu33, Phe59, Trp65D, Ile 106
and Val35) do not draw a pocket.
Figure 9. Superimposition of FXIIa hybrid model-compound 23 complex with plasma kallikrein.
In the picture, FXIIa is hidden and only compound 23 is shown for clarity purpose. Plasma
kallikrein pockets are S1 (Asp189), S2 (Asp102), S3 (Trp215, Tyr174, Gly99) and S1’ (Leu41,
Leu60B and Trp65D, Asp60) pockets. Residues Leu33, Phe59, Trp65D, Ile 106 and Val35 do
not contribute to a pocket but the S1’ pocket looks like the H1 pocket of FXIIa. However, the
pockets are different in shape and size. Besides, taking into account the orientation of compound
2
3 in the protease suggests that interactions with the S1’pocket may be not excluded.
In summary for this in silico selectivity part, we found that residues contributing to the H1
pocket in FXIIa are different in nature and they do not participate in the formation of a pocket in
the assessed proteases (thrombin, FXa, FVIIa and plasma kallikrein). Among these proteases,
only the S1’ pocket of plasma kallikrein looks like the FXIIa H1 pocket but involves different
26

ACCEPTED MANUSCRIPT
residues leading to a difference in shape and size. These data thus suggest that selectivity could
be obtained by targeting the H1 pocket of FXIIa.
27

ACCEPTED MANUSCRIPT
3. CONCLUSION
The discovery and development of FXIIa inhibitors is of valuable interest for blood
coagulation assays as well as for clinical applications implying the exposure of patients to
artificial surfaces. In this publication, we have developed 3-carboxamide coumarins modulated
in position 3 or 6 with amino- or oxygen-based group. These modulations were introduced in the
perspective to afford selective FXIIa inhibitors with a better physico-chemical profile. Among
our series, the modulation of the carboxamide side chain with an oxygen-based group afforded
six active compounds with IC50 in the micromolar range. To better understand the structure-
activity relationship of our compounds, a docking study was performed. To this end, we
generated a hybrid model of FXIIa. This hybrid model was built from the published crystal
structure of FXIIa in its zymogen form and a homology model based on the X-ray structure of
the HGFA. Both initial 3D structures had limitations rendering docking investigations
unsuitable. Hence, the creation of a hybrid model was a means to keep the advantages while
overcoming the limitations of each structure. Interestingly, the molecular modeling study showed
that the hybrid model seems to discriminate active from inactive compounds within the series
described in this publication. Furthermore, the interactions highlighted for our most active
compound helped to understand its activity and selectivity.
28