An integrative study to identify novel scaffolds for sphingosine kinase 1 inhibitors

Article abstract of DOI:10.1016/j.ejmech.2017.08.017

Sphingosine kinase 1 (SphK1), the enzyme that produces the bioactive sphingolipid metabolite, sphingosine-1-phosphate, is a promising new molecular target for therapeutic intervention in cancer and inflammatory diseases. In view of its importance, the main objective of this work was to find new and more potent inhibitors for this enzyme possessing different structural scaffolds than those of the known inhibitors. Our theoretical and experimental study has allowed us to identify two new structural scaffolds (three new compounds), which could be used as starting structures for the design and then the development of new inhibitors of SphK1. Our study was carried out in different steps: virtual screening, synthesis, bioassays and molecular modelling. From our results, we propose a new dihydrobenzo[b]pyrimido[5,4-f]azepine and two alkyl{3-/4-[1-hydroxy-2-(4-arylpiperazin-1-yl)ethyl]phenyl}carbamates as initial structures for the development of new inhibitors. In addition, our molecular modelling study using QTAIM calculations, allowed us to describe in detail the molecular interactions that stabilize the different Ligand-Receptor complexes. Such analyses indicate that the cationic head of the different compounds must be refined in order to obtain an increase in the binding affinity of these ligands.

Full text of DOI:10.1016/j.ejmech.2017.08.017

Accepted Manuscript
An integrative study to identify novel scaffolds for sphingosine kinase 1 inhibitors
Marcela Vettorazzi, Emilio Angelina, Santiago Lima, Tomas Gonec, Jan Otevrel,
Pavlina Marvanova, Tereza Padrtova, Petr Mokry, Pavel Bobal, Lina M. Acosta, Alirio
Palma, Justo Cobo, Janette Bobalova, Jozef Csollei, Ivan Malik, Sergio Alvarez,
Sarah Spiegel, Josef Jampilek, Ricardo D. Enriz
PII:
S0223-5234(17)30620-7
10.1016/j.ejmech.2017.08.017
EJMECH 9661
DOI:
Reference:
To appear in: European Journal of Medicinal Chemistry
Received Date: 7 April 2017
Revised Date: 2 August 2017
Accepted Date: 5 August 2017
Please cite this article as: M. Vettorazzi, E. Angelina, S. Lima, T. Gonec, J. Otevrel, P. Marvanova,
T. Padrtova, P. Mokry, P. Bobal, L.M. Acosta, A. Palma, J. Cobo, J. Bobalova, J. Csollei, I. Malik,
S. Alvarez, S. Spiegel, J. Jampilek, R.D. Enriz, An integrative study to identify novel scaffolds for
sphingosine kinase 1 inhibitors, European Journal of Medicinal Chemistry (2017), doi: 10.1016/
j.ejmech.2017.08.017.
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
An Integrative Study to Identify Novel Scaffolds for Sphingosine
Kinase 1 Inhibitors
a
b
c
d
d
Marcela Vettorazzi , Emilio Angelina , Santiago Lima , Tomas Gonec , Jan Otevrel ,
d
d
d
d
e
Pavlina Marvanova , Tereza Padrtova , Petr Mokry , Pavel Bobal , Lina M. Acosta , Alirio
e
f
g
d h
h
a
Palma , Justo Cobo , Janette Bobalova , Jozef Csollei , , Ivan Malik , Sergio Alvarez ,
c
h
a*
Sarah Spiegel , Josef Jampilek and Ricardo D. Enriz
a
Facultad de Química, Bioquímica y Farmacia, Universidad Nacional de San Luis; Instituto
Multidisciplinario de Investigaciones Biológicas (IMIBIO-SL).Chacabuco 915, 5700 San Luis,
Argentina; e-mail: denriz@unsl.edu.ar
b
Laboratorio de Estructura Molecular y Propiedades, Área de Química Física, Departamento de
Química, Facultad de Ciencias Exactas y Naturales y Agrimensura, Universidad Nacional del
Nordeste, Avda. Libertad 5460, (3400) Corrientes, Argentina
c
Department of Biochemistry and Molecular Biology, Virginia Commonwealth University School of
Medicine, Richmond, VA 23298 USA
d
Department of Chemical Drugs, Faculty of Pharmacy, University of Veterinary and
Pharmaceutical Sciences Brno, Palackeho 1/3, 612 42 Brno, Czech Republic
e
Laboratorio de Síntesis Orgánica, Escuela de Química, Universidad Industrial de
Santander, Carrera 27, Calle 9, A.A 678, Bucaramanga, Colombia
f
Inorganic and Organic Department, University of Jaén, Campus Las Lagunillas s/n, 23071,
Jaén, Spain
g
Institute of Analytical Chemistry of the Czech Academy of Sciences, Veveri 97, 602 00
Brno, Czech Republic
h
Department of Pharmaceutical Chemistry, Faculty of Pharmacy, Comenius University,
Odbojarov 10, 83232 Bratislava, Slovakia
Abstract
Sphingosine kinase 1 (SphK1), the enzyme that produces the bioactive sphingolipid
metabolite, sphingosine-1-phosphate, is a promising new molecular target for therapeutic
intervention in cancer and inflammatory diseases. In view of its importance, the main
objective of this work was to find new and more potent inhibitors for this enzyme possessing
1

ACCEPTED MANUSCRIPT
different structural scaffolds than those of the known inhibitors. Our theoretical and
experimental study has allowed us to identify two new structural scaffolds (three new
compounds), which could be used as starting structures for the design and then the
development of new inhibitors of SphK1. Our study was carried out in different steps:
virtual screening, synthesis, bioassays and molecular modelling. From our results, we
propose a new dihydrobenzo[b]pyrimido[5,4-f]azepines and two alkyl{3-/4-[1-hydroxy-2-
(4-arylpiperazin-1-yl)ethyl]phenyl}carbamates as initial structures for the development of
new inhibitors. In addition, our molecular modelling study using QTAIM calculations,
allowed us to describe in detail the molecular interactions that stabilize the different Ligand-
Receptor complexes. Such analyses indicate that the cationic head of the different
compounds must be refined in order to obtain an increase in the binding affinity of these
ligands.
Keywords: sphingosine kinase 1 inhibitors, virtual screening, synthesis, bioassays,
molecular modelling.
1
. Introduction
Sphingosine-1-phosphate (S1P) is a bioactive sphingolipid metabolite that regulates many
physiological functions [1–3]; however, it also has a pathological role in autoimmune
dysfunction, inflammation, cancer and many other diseases [4–6]. S1P is generated
intracellularly by the action of two sphingosine kinases (SphKs) named SphK1 and SphK2,
which catalyze the ATP-dependent phosphorylation of sphingosine on its primary hydroxyl
group.
It has been demonstrated that S1P levels and SphK1 expression and/or activity are
increased in distinct cancer types, including solid tumors of the breast, colon, lung, ovary,
stomach, uterus, kidney, liver, and melanoma and in leukemia, among others [7–9].
2

ACCEPTED MANUSCRIPT
Moreover, upregulation of SphK1 has been associated with tumor angiogenesis and
lymphangiogenesis [10–13] and correlates with poor prognosis in cancer patients [14,15].
Indeed, substantial evidence indicates that pharmacological or molecular inhibition of
SphK1 has anti-cancer effects [11,16–25]. Thus, it is clear that SphK1 is a promising novel
molecular target for therapeutic intervention in cancer and inflammatory diseases [26,27].
The first crystal structure of SphK1 was reported in 2013 [28]. More recently two SphK1-
co-crystal structures (4L02 and 4V24) with potent inhibitory compounds have also been
published [29,30]. These structures provide useful structural information on the interactions
of ligands at the active site of SphK1.
Previously, various compounds have been reported with inhibitory activity on SphK1
[
29,31–37]. The structural scaffolding of the best known compounds is shown in Figure 1,
with their respective references. Among these compounds, compound SLP7111228 has been
recently reported [37]. Recently, the crystal structure of SphK1 with PF-543, the most potent
and selective SphK1 inhibitor, was also revealed [30,36]. While numerous non-specific and
several isozyme-specific inhibitors of SphK1 have been studied, considering the important
role of this enzyme in inflammatory processes related to cancer [38–41], an important task is
to develop new and more potent SphK1 inhibitors based on different structural scaffolds
than those of the known inhibitors. Thus, the main objective of this work is to obtain new
SphK1 inhibitors having a different structural basis than the well-known inhibitors. In the
first step of our study, which was based on the crystal structures of SphK1 alone [28] and
complexed with an inhibitor [29,30], we have carried out virtual screening that allowed us to
identify potential new inhibitors. Next, we synthesized and tested the inhibitory activity of
these novel compounds, and finally, we conducted a molecular modelling study that allowed
us to understand interactions at the molecular level that stabilize the formation of different
3

ACCEPTED MANUSCRIPT
Ligand-Receptor (L-R) complexes. Figure 2 shows in schematic form the various steps that
have been carried out in this study.
Figure 1. Main structural scaffolds previously reported for inhibitors of SphK1.
Figure 2. Flow chart showing the various steps and techniques carried out in our study. The
numbers inside the boxes indicate the number of compounds evaluated.
4

ACCEPTED MANUSCRIPT
2
2
. Results and Discussion
.1. Structure-based virtual screening
Structure-Based Virtual Screening (SBVS) strategies rely on the three dimensional
structure of a target and on the ability of docking algorithms to predict the binding mode and
the binding affinities of different compounds obtained from libraries [42]. The docking
process is usually divided into two major steps: first, the correct placement of the ligand at
the protein binding-site; and then estimation of the ligand affinity by a scoring function [43].
Before undertaking the prospective Virtual Screening (VS) campaigns to search for novel
inhibitors of SphK1, we considered it prudent to evaluate the performance of the docking
algorithm in retrieving known inhibitors of SphK1 from a decoys library. There are two
treatments for constructing these libraries. In the most common approach a few active
compounds are seed in a larger database of randomly selected (and supposedly inactive)
molecules with similar physico-chemical properties but dissimilar 2-D topologies. However,
in real medicinal chemistry pipelines, molecules are often congeneric as they come from
parallel synthesis and, consequently, are structurally more similar to one another. It would
therefore be more reasonable to select decoys according to their similarity to the active
molecules [44]. We constructed the decoy library to evaluate docking performance by
following this second approach. Since in this approach decoys are more likely to be actives
than the randomly selected molecules, one needs to be sure that they are truly inactive.
Therefore, their biological activities against the target of interest must be known.
Accordingly, we have compiled a dataset of 54 compounds from CHEMBL
(www.ebi.ac.uk/chembl/) [45] for which their experimentally measured dissociation
constants (Ki values) against SphK1 are known. The compounds were sorted from the most
actives to the marginally active ones according to their Ki values. The cutoff was set to
5

ACCEPTED MANUSCRIPT
1
0000 nM, compounds above (below) this threshold were flagged as active (inactive). The
chosen cutoff value ensures a balanced population of active / inactive compounds.
After screening the library with AutoDock Vina (AD Vina) [46] we constructed a
Receiver Operating Characteristic Curve (ROC) to asses the quality of the docking results
[
44]. Then the ROC curve was constructed by plotting the True Positive Rate (or Sensitivity)
versus the False Positive Rate (or 1-Specificity) calculated at intervals over the ordered list
of docking scores. The ROC is depicted in Figure 3 for VS results on one of the
experimentally solved SphK1 protein conformers, 3vzd_A_chainA (see cross-docking
section below for details on protein conformers naming). In this graphic, the 45° diagonal
(
gray line) represents a random classification of the database with area under the curve
AUC) for the random case of 0.5. Any model with an AUC > 0.5 performs better than
(
random in discriminating the most active compounds from less active ones. As observed in
the figure, the docking model performs very well in the classification of the dataset (AUC =
0
.87). Similar results were obtained for the remaining available SphK1 structures.
Figure 3. Performance for the docking of the 54 known inhibitors of SphK1 on the
conformer 3vzd_A_chainB.
6

ACCEPTED MANUSCRIPT
Figure 4a shows the best score pose for the 54 known inhibitors of SphK1 as docked in
the same protein conformer 3vzd_A_chainB. While there is no structural information on
how these inhibitors actually bind to SphK1, they share the same structural scaffolds as the
four ligands in the solved structures and therefore, it is assumed that they will bind also in
the same binding mode. As depicted in the Figure 4, most of the inhibitors adopt the
characteristic J-shaped binding mode observed for the crystallographic ligands. However,
some outliers can be observed among the poses: some of them do not enter entirely into the
J-shaped cavity overlapping with the ADP binding site and some other ligands are turned
over, i.e. with their polar heads anchored at the bottom and the hydrophobic tail near the
entrance of the J-shaped cavity.
Therefore, although the AD Vina scoring function performs well in discriminating
between more active and less active ligands, the ability of the docking algorithm in
retrieving the correct ligand poses is somewhat deficient.
7

ACCEPTED MANUSCRIPT
Figure 4. Binding modes of the 54 known inhibitors of SphK1 before (a) and after (b) the
filtering out of the non-structure-like poses.
2
.1.1. Cross-docking
Many studies have assessed the success rate of programs in self-docking tests, whereby a
ligand is docked into the protein structure from which it was extracted (native docking).
Cross-docking, or using a protein structure from a complex containing a different ligand,
provides a more realistic assessment of a docking program's ability to reproduce X-ray
results [47].
To evaluate the ability of the docking algorithm to retrieve the correct poses of known
binders of SphK1, we performed a cross-docking analysis of the crystallographic complexes
of SphK1 available to date. Currently, there are five SphK1 crystal structures deposited in
the Protein Data Bank shown in Figure 5a. In one, SphK1 was crystallized both in apo form
(code 3vzb, chain C) and in complex with a substrate sphingosine-like lipid SQS (code
8

ACCEPTED MANUSCRIPT
3
vzb, chains A, B). In the remaining structures, SphK1 was solved in complex with a
moderate, non-isozyme-specific inhibitor SKI-II in presence of ADP (3vzd) and without
ADP (code 3vzc) and complexed with two potent inhibitors, PF543 (code 4v24) and 1V2
(code 4l02).
Four of the structures, 3vzb, 3vzd, 4l02 and 4v24 displayed both “A” and “B” alternative
conformations for some residues. Since none of the residues with alternate conformations
were close to the substrate cavity, only conformation “A” of those structures was kept.
Moreover, while SphK1 is monomeric (biological assembly), the asymmetric unit in the
crystal structures contains from 2 to 5 molecules (chains) depending on the structure (see
Figure 5a). Since the asymmetric unit constitutes the smallest repetitive portion of the
crystal and due to the different stoichiometry of the chains within the same structure (i.e. apo
form, bound to ligand, bound to ligand + ADP, see Figure 5a), structural differences can be
observed among the chains in a single crystal structure. In 19 out of the 20 SphK1 chains or
conformers, there was a ligand bound to the substrate cavity in the C-terminal domain.
Therefore, we performed cross-docking of the 19 ligands against the 20 conformations of
proteins from the five crystal structures.
For the naming of the different SphK1 conformers, we adopted the following convention:
“PDB id_X_chainY”, where PDB id is the four digit code with which the 3D structure is
deposited in the Protein Data Bank; X the alternate conformation (i.e: A, B, etc); and Y
corresponds to the protein chain (chain A, chain B, etc).
Previous to the cross-docking calculations, the twenty SphK1 conformers were
superimposed by their protein alpha carbons. Then, each ligand from each chain in the
crystal structures was cross-docked against all of the protein conformers. After docking, the
RMSD was calculated between the experimental binding mode of each ligand conformer
9

ACCEPTED MANUSCRIPT
and its docked pose in each one of the twenty protein conformers. The binary heatmap in
Figure 5b summarizes the results of the cross-docking experiment.
Figure 5. (a) Schematic representation of the 5 crystal structures available for SphK1. This
scheme is intended to illustrate the number and arrangement of the chains within each
structure as well as the composition stoichiometry of each chain (ligand-bound/unbound,
ADP-bound/unbound states). In structures 3vzb, 4l02 and 4v24 the ligand binding sites from
each chain are packed against each other, while in structures 3vzc and 3vzd, the binding
sites point outward from the asymmetric unit. Chains labeled with a star lack some residues
in the loop that connect helices α7 and α8 (see below). (b) Performance of the cross-docking
experiment. Top-ranked docking poses with a RMSD < 2 with respect to the
crystallographic binding mode in each chain are depicted in green. Red and yellow stars
indicate the best scored protein conformer for each ligand before and after filtering out the
"non X-ray like", respectively.
1
0

ACCEPTED MANUSCRIPT
Figure 6. (a) Crystal structure of SphK1 bound to SQS (in orange) superimposed to the apo
form of the enzyme (in cyan). (b) Crystal structure of the SQS bound SphK1 (in orange)
superimposed to the X-ray structure of the SKI-II bound enzyme (in green). SKI-II atoms
are represented with spheres to highlight the steric hindrance with V177 from the SQS-
bound protein conformer.
As can be seen, PF543 and 1V2 are docked in the proposed experimental binding pose in
most of the protein conformers. On the other hand, SKI-II and SQS in general reproduces
the experimental pose only in its native conformers and/or in a smaller subset of protein
conformers.
The poor performance of the cross-docking of SQS and SKI-II into their non-native
conformers suggests that conformational changes might take place in the enzyme that are
driven by the ligand characteristics (induced fit effects). This is also evidenced by the fact
that none of the ligands dock well in the apo form of the enzyme (molecule C from 3vzb).
Examination of the apo form (molecule C of 3vzb) revealed a similarly shaped cavity
with a comparable entrance around the head group in comparison with the holo enzyme, i.e.
the J-shaped cavity already exists in the apo form. However, its solvent-accessible volume
decreased markedly on going from the apo to the holo form due to the inward movement of
helices α7 and α8 (Figure 6a). As suggested by Wang et al [28], the α7-α8 segment in
SphKs might act like a lipid gate that controls the in-and-out of lipid substrate and product.
The SQS-bound enzyme superimposed to the SKI-II-bound, which illustrates the
importance of the conformational changes of SphK1, is shown in Figure 6b. As can be seen,
the “α7-α8 gate” shows a slightly different conformation in SKI-II-bound (green) and SQS-
bound conformers (orange). As a consequence of this conformational difference, when SKI-
II is docked into the SQS-bound protein conformer, it cannot approach close enough to the
α8 helix to form the N-H···O H-bond with Thr196 because Val177 would be located too
close to the aminophenol ring of SKI-II leading to steric hindrance. The inability of SKI-II
1
1

ACCEPTED MANUSCRIPT
to form that H-bond in the SQS-bound protein conformer, which is present in the SKI-II-
bound native structure, might explain in part the poor performance of the cross-docking of
SKI-II. Thus, the conformational changes in the α7-α8 segment, and the fact that the
docking performance is very sensitive to those changes, highlight the necessity of employing
an ensemble of protein conformers in the virtual screening campaigns against SphK1.
2
.1.2. Ensemble docking
In ensemble docking, the ligand is docked against a number of conformations of the
protein. The highest scoring binding mode is then selected from the ensemble of dockings
against all protein conformers. Accordingly, we selected the highest scoring pose obtained
for each one of the 19 ligands, which are labeled with a red star on the binary heat map in
Figure 5b. As can be seen in that figure, only in 8 of the 19 ligands (42 % of the cases) was
a correct pose (i.e. a pose with RMSD < 2Å) selected by the standard scoring function of
AD Vina [46]. Thus, it is evident that an ensemble docking campaign to find novel
inhibitors of SphK1 would likely fail if the “non X-ray like” (i.e. the poses with RMSD >
2
Å) are not first filtered out from the ensemble.
2
.1.3. Model to filter out the “non X-ray like”
Taking advantage of the structural information available for SphK1, we trained a
classification model that was able to distinguish the docking “X-ray like” from the “non X-
ray like” according to a 2Å RMSD cutoff, with ~ 86% accuracy on the training set. Figure
7
a shows the form of the logistic regression hypothesis h (X) where X and θ are vectors
θ
containing the independent variables (or features) and the adjustable parameters of the
model, respectively. The model, once the parameters were adjusted, estimated the
probability (between 0 and 1) of the docking pose being a crystal structure-like pose or a “X-
1
2

ACCEPTED MANUSCRIPT
ray like” pose, based on the ligand binding modes from the experimentally resolved
structures of SphK1.
Figure 7. Construction of the logistic regression model to classify docking poses in
crystal structure-like / non crystal structure-like poses.
The features that fit the model were selected from the conformation-dependent terms
defined in the AD Vina scoring function. The conformation-dependent terms in the default
AD Vina scoring function consist of three steric terms (gauss , gauss and repulsion), a
1
2
hydrogen bond term, and a hydrophobic term [46].
To make the features independent of the ligand size, each term was normalized by the
number of heavy atoms of the ligand or heavy atom counts (HAC) so that the model might
generalize to ligands other than the ones in the training set.
1
3

ACCEPTED MANUSCRIPT
By plotting the size-independent AD Vina terms against each other, we found that the
hydrophobic term was irrelevant for discriminating between “X-ray like” and “non X-ray
like”, so it was removed from the set of potential features for the classification model. The
hydrophobic term accounts for the desolvation penalty due to ligand binding. Since both “X-
ray like” and “non X-ray like” are quite buried into the protein core (see Figure 4a), the
hydrophobic term does not differ appreciably between both kinds of poses. From the
remaining four conformation-dependent terms, a new set of features was constructed by
taking into account the physical meaning of these terms and the knowledge gained from the
experimentally solved structures of SphK1.
Figure 7a shows the new set of features constructed from the original AD Vina terms.
Thus, the three steric terms (previously normalized by the HAC) were added together in a
single term x based on the observation that these three terms together (i.e. the two gauss
1
attractive terms for dispersion and the repulsive term) roughly behave like the van der Walls
terms in force field based scoring functions (i.e. the 12-6 Lennard-Jones potential). The
combined steric term (x ), similarly to the van der Waals terms, might be considered as a
1
measure of the shape complementarity between the ligand and the receptor binding cavity.
In addition to x , x represents the AD Vina hydrogen bonding term normalized by the
1
2
heavy atom count. Finally, alternative hydrogen bonding term x was constructed. Unlike x ,
3
2
x3 discriminates between those interactions where the ligand acts as hydrogen bond donor
and as hydrogen bond acceptor. Thus, x computes the AD Vina hydrogen bonding function
3
for the interactions involving ligand donor atoms (HBD) versus ligand acceptors atoms
(HBA), each one normalized by the total number of ligand H-bond donor and acceptor
atoms of the ligand, respectively.
In Figure 7b, x is plotted against x with the “X-ray like” (< 2Å RMSD) and “non X-ray
1
2
like” (> 2Å RMSD) depicted with black crosses and yellow circles, respectively, and the
1
4

ACCEPTED MANUSCRIPT
decision boundary between them is depicted in blue. This decision boundary discriminates
between both kinds of poses with 69% accuracy. On the other hand, when x is used to fit
3
the model instead of x (Figure 7c), the resulting decision boundary performs much better in
2
discriminating between crystal-structure like and non-crystal structure-like poses. This
improvement in the model performance is related to the simple observation that the ligands
in the experimentally solved structures act almost exclusively as hydrogen bond donors
against the enzyme residues. This results in a very high value of x (and a very high
3
HBD/HBA ratio) in the crystal structure-like poses and a very low value of x , allowing a
3
better discrimination between both kinds of poses.
Figure 7d depicts the J-shaped tunnel occupied by the ligand in the enzyme binding site.
As can be seen, the borders of the J-shaped tunnel are defined mostly by hydrophobic
residues and also by a set of negatively charged residues near the entrance of the tunnel,
including Asp81 and Asp178 among others. Thus, the molecular recognition event involves
shape complementarity between the ligand and SphK1 which is evidenced by the overall
higher values of x in the crystal structure-like poses and also charge complementarity
1
which is better recovered by x3.
The yellow stars in Figure 5b show the highest scoring poses for each one of the 19
ligands after filtering out the “non X-ray like” with the trained logistic model. In 18 out of
the 19 ligands a crystal structure-like pose was selected that represented a clear
improvement as compared with the ensemble docking performance without applying the
filter.
2
.1.4. Model validation
In order to validate the trained classification model we applied it to the docked poses of
the 54 known inhibitors of SphK1 from CHEMBL database to filter out the "non X-ray
1
5

ACCEPTED MANUSCRIPT
like". Figure 4b shows the best scored poses for the protein conformer 3vzd_A_chainB,
after applying the trained model. By visual inspection of Figure 4b, one can actually see
that most of the outlier poses were removed from the test set after filtering out the non-
crystal structure-like poses.
2
.1.5. Pre-selection of candidate compounds and results of the virtual screening
From the point of view of the structure, the compounds selected for further testing should
bear some structural resemblance to the known ligands; i.e. have a polar head and a
hydrophobic tail, as well as some conformational flexibility. Based on this, we selected a
group of 147 compounds including chalcones, acetogenins, protoberberines,
nitrosopyrimidines and carbamates among others. The fingerprints of these 147 compounds
were calculated using Open Babel [48], a chemical expert system mainly used for converting
chemical file formats, and the results were compared with fingerprints obtained for the 54
compounds used previously to test the model. Those compounds having a Tanimoto index
(Tc) [43] greater than 0.7 were discarded. Note that the fingerprints are a way to encode the
structure of a molecule in order to compare it with other compounds; whereas the Tc is a
measure of the similarity between two structures ranging from 0 to 1. Based on the Tc index,
we selected 114 structurally diverse compounds which possess structural differences with
respect to the known binders from the test set, justifying their further study.
.
2.2. Chemistry
Our model predicted sixteen different compounds as potential inhibitors of SphK1 (Table
1
1
) that were tested in vitro for inhibition of recombinant SphK1. Synthesis of compounds 2–
2, 14 and 15 have not been previously reported in the literature, and thus their
characterization is now described (compounds 7–9 were reported only as moderate
1
6

ACCEPTED MANUSCRIPT
antimicrobial and/or antioxidant agents, but without their synthesis and analytical data [49-
5
1]). Synthesis and characterization of compounds 1, 13 and 16 have been previously
reported [52-54].
Table 1. Structural features of compounds evaluated as SphK1 inhibitors.
Com
p.
Com
p.
Structure
Structure
1
2
3
4
5
6
9
10
11
12
13
14
1
7

ACCEPTED MANUSCRIPT
7
8
15
16
Compound 2 was synthesized in two steps from synthetically available 4-chloro-6,11-
dimethyl-6,11-dihydro-5H-benzo[b]pyrimido[5,4-f]azepine [52] by treatment with excess of
hydrazine monohydrate and subsequent condensation of the resulted hydrazinyl derivative
with p-chlorobenzaldehyde. Both the above precursor of compound 1 and compound 2 were
prepared following the general procedure shown in scheme 1, which starts from the
corresponding 5-allyl-4,6-dichloropyrimidine that suffers aminolysis by reaction with a
substituted N-methylaniline to afford the corresponding 5-allyl-4-arylamino-4-
chloropyrimidine which under a strong acid media provokes intramolecular Friedel-Crafts
cyclization to afford the corresponding 6,11–dihydro–5H–benzo[b]pyrimido[5,4–f]azepine;
in the case of starting from p-methoxy-N-methylaniline, the demethylation of the methoxy
1
8

ACCEPTED MANUSCRIPT
group is observed in the cyclization step and so rendering the compound 1.
Scheme 1: Synthesis of 4-chloro-6,11-dimethyl-6,11-dihydro-8-hydroxy-2-phenyl-5H-
1
benzo[b]pyrimido[5,4-f]azepine, (1) (R=Ph; R =OH) [49] and (E)-4-(2-(4-
chlorobenzylidene)hydrazinyl)-6,11-dimethyl-6,11-dihydro-5H-benzo[b]pyrimido[5,4-
1
f]azepine (2) (R = R = H).
Reagents and conditions: a) DIPEA, EtOH, reflux, 48-72 h; b) CH SO H, 115-120 °C, 10-
3
3
2
0 min; c) NH NH ·H O, EtOH, reflux, 24 h; d) p-chlorobenzaldehyde, EtOH, AcOH (cat.)
2 2 2
reflux, 4 h.
Reaction of 3- or 4-aminoacetophenone and a suitable alkyl chloroformiate gave alkyl (3-/4-
acetylphenyl)carbamates 3a–6a, which reacted with bromine in chloroform to yield alkyl [3-
/
4-(bromoacetyl)phenyl]carbamates 3b–6b; these compounds by treatment with different N-
monosubstituted aryl(heteroaryl)piperazines provided the corresponding alkyl {3-/4-[(4-
aryl(heteroaryl)piperazin-1-yl)acetyl]phenyl}carbamates 3c–6c. Subsequent reduction of
keto
group
using
NaBH4
provided
target
alkyl
{3-/4-[1-hydroxy-2-(4-
aryl(heteroaryl)piperazin-1-yl)ethyl] phenyl}carbamates 3-6, see Scheme 2.
1
9

ACCEPTED MANUSCRIPT
Scheme 2: Synthesis1 of alkyl {3-/4-[1-hydroxy-2-(4-arylpiperazin-1-yl)ethyl]
2
phenyl}carbamates 3–6. R : 3-Me, 3-Bu, 4-Bu; R : pyridine-2-yl, pyridine-4-yl, pyrimidine-
2
-yl.
1
Reagents and conditions: a) ClCOOR , pyridine, acetone, reflux 3h; b) Br , CHCl , ambient
2
3
temperature 3 h; c) N-(hetero)arylpiperazine, TEA, THF, ambient temperature 3 h; d)
NaBH , MeOH, reflux 3 h.
4
Studied 1-(3-{4-[(alkoxycarbonyl)amino]benzoyloxy}-2-hydroxypropyl)-4-phenylpiperazin-
1
-ium chlorides 7–9 were prepared by multiple-step reaction described in Scheme 3.
Epoxides 7a–9a and 13a (see below) were prepared from 4-aminobenzoic acid through
reaction with methyl, ethyl, propyl and butyl chloroformiates giving appropriate 4-
[
(alkoxycarbonyl)amino]benzoic acids. Chlorides of these acids formed by thionyl chloride
treatment gave desired epoxides 7a–9a and 13a after reaction with 2,3-epoxypropan-1-ol
53]. In the last step final compounds 7–9 were prepared by a reaction of the epoxides with
-(4-phenyl)piperazine and then converted to the hydrochloride salts using ethereal HCl to
enhance their solubility in water [55].
[
1
2
0

ACCEPTED MANUSCRIPT
Scheme 3: Synthesis of studied 1-(3-{4-[(alkoxycarbonyl)amino]benzoyloxy}-2-
hydroxypropyl)-4-phenylpiperazin-1-ium chlorides 7–9. R: Me, Et, Bu and Pr (13a).
Reagents and conditions: a) ClCOOR, pyridine, acetone, reflux 3 h; b) SOCl , toluene,
2
reflux; c) 2,3-epoxypropan-1-ol, THF, TEA, 0°C then ambient temperature; d) 1-
phenylpiperazine, i-PrOH, reflux; e) HCl, Et O, ambient temperature.
2
(
1-Butylpiperidin-4-yl)methanamine (10b=11b) and (1-propylpiperidin-4-yl)methanamine
12b) were prepared according to the literature from piperidine-4-carboxamide by alkylation
(
followed by reduction [56,57]. The preparation of target products 10–12 was conducted as
follows: the commercially available 4-aminophenol was treated with alkyl chloroformates
1
(
R = Me, Bu) followed by epichlorohydrin addition to give requisite methyl [4-(oxiran-2-
ylmethoxy)phenyl]carbamate
(10a=12a)
and
butyl
[4-(oxiran-2-
ylmethoxy)phenyl]carbamate (11a) [58]. The carbamates were then coupled to previously
prepared methanamines in MeOH at ambient temperature. The prepared final products as
free bases were tranformed to target alkyl [4-(3-{[(1-alkylpiperidin-4-yl)methyl]amino}-2-
hydroxypropoxy)phenyl]carbamate dimethanesulfonates 10–12, see Scheme 4.
2
1

ACCEPTED MANUSCRIPT
Scheme 4: Synthesis of studied alkyl [4-(3-{[(1-alkylpiperidin-4-yl)methyl]amino}-2-
1
2
hydroxypropoxy)phenyl]carbamate dimethanesulfonates 10–12. R : Me, Bu; R : Pr, Bu.
1
Reagents and conditions: a) ClCOOR , Et O, pyridine, ambient temperature; b) 2-
2
(
chloromethyl)oxirane, Et O, KOH, ambient temperature; c) alkylhalogenides; d) LiALH ;
2
4
e) MeOH, ambient temperature; f) CH SO H, acetone.
3
3
Synthesis
of
2-hydroxy-3-[2-(4-methoxyphenoxy)ethylamino]propyl
4-
(propoxycarbonylamino)benzoate hydrochloride (13) was described in Tengler et al. [53].
The first main intermediate, oxiran-2-ylmethyl 4-[(propoxycarbonyl)amino]benzoate (13a)
was synthesized according to Scheme 3. The epoxide ring was opened by addition of the
second main intermediate, 2-(4-methoxyphenoxy)ethanamine (13b) prepared by Gabriel
synthesis from 4-methoxyphenol via 1-(2-bromoethoxy)-4-methoxybenzene and 2-[2-(4-
methoxyphenoxy)ethyl]-1H-isoindole-1,3(2H)-dione, see Scheme 5. The higher aqueous-
soluble hydrochloride salt was prepared from the acquired base using ethereal HCl, as
described in Tengler et al. [53].
2
2

ACCEPTED MANUSCRIPT
Scheme 5: Synthesis of 2-hydroxy-3-[2-(4-methoxyphenoxy)ethylamino]propyl 4-
(propoxycarbonylamino)benzoate hydrochloride (13).
Reagents and conditions: 13a see Scheme 3; a) 1,2-dibromoethane, NaOH; b) potassium
phtalimide, KI, DMF; c) NH NH ·H O, EtOH; d) i-PrOH; e) HCl, Et O.
2
2
2
2
Final product 14 was prepared via multi-step synthesis as described in Scheme 6. Tosylate
intermediates 2-methoxyethyl 4-methylbenzenesulfonate and oxiran-2-ylmethyl 4-
methylbenzenesulfonate were prepared according to the published procedure by reaction of
4
-methylbenzenesulfonyl chloride with 2-methoxyethanol and oxiran-2-ylmethanol,
respectively, in dichloromethane [59,60]. 2-(2,6-Dimethoxyphenoxy)ethanamine was
prepared via Gabriel synthesis (see Scheme 5) from 2-[2-(2,6-dimethoxyphenoxy)ethyl]-
1
H-isoindole-1,3(2H)-dione in two-step synthesis starting from 2,6-dimethoxyphenol that
gave 2-(2-bromoethoxy)-1,3-dimethoxybenzene with 1,2-dibromoethane. Then the 2-(2-
bromoethoxy)-1,3-dimethoxybenzene was reacted with potassium phthalimide [61]. The
main synthesis was carried out as described by Marvanova et al. [62]. 4-(2-
Methoxy)ethoxybenzoic acid (14a) was prepared from ethyl 4-hydroxybenzoate via reaction
with 2-methoxyethyl 4-methylbenzenesulfonate and following hydrolysis of the ester. The
final product, fumarate salt (14), was prepared from the potassium salt of acid 14a that gave
(oxiran-2-yl)methyl 4-(2-methoxyethoxy)benzoate (14b) after reaction with oxiran-2-
ylmethyl 4-methylbenzenesulfonate. The oxirane ring was then opened by 2-(2,6-
2
3

ACCEPTED MANUSCRIPT
dimethoxyphenoxy)ethanamine. The obtained base was finally converted to its fumarate salt
using ethereal solution of fumaric acid to enhance the solubility of the compounds in water.
Scheme 6: Synthesis of target [2-(2,6-dimethoxyfenoxy)ethyl]{2-hydroxy-3-[4-(2-
methoxyethoxy)benzoyloxy]propyl} ammonium fumarate (14).
Reagents and conditions: a) 2-methoxyethyl 4-methylbenzenesulfonate, K CO , acetone,
2
3
ambient temperature; b) i) NaOH, ii) HCl, CHCl ; c) MeOH, i-PrOH, KOH; d) oxiran-2-
3
ylmethyl
4-methylbenzenesulfonate,
DMF,
7 h,
70
°C;
e)
2-(2,6-
dimethoxyphenoxy)ethanamine, i-PrOH, 1 h at 85 °C and for 72 h at ambient temperature; f)
fumaric acid, Et O.
2
1
-Chloro-2-methoxyethane by reaction with methyl 2-hydroxybenzoate gave methyl 2-(2-
methoxyethoxy)benzoate (15a) that yielded acid 15b after hydrolysis. A reaction of the acid
with SOCl provided 2-(2-methoxyethoxy)benzoyl chloride that with (2S)-oxiran-2-
2
ylmethanol in the presence of 4-dimethylaminopyridine formed (2R)-oxiran-2-ylmethyl 2-
(2-methoxyethoxy)benzoate (15c), a reaction of which with tert-butylamine resulted in the
formation of (2R)-3-(tert-butylamino)-2-hydroxypropyl 2-(2-methoxyethoxy)benzoate that
was converted to final fumarate salt 15, see Scheme 7.
Scheme 7: Synthesis of target [2-(2,6-dimethoxyfenoxy)ethyl]{2-hydroxy-3-[4-(2-
methoxyethoxy)benzoyloxy]propyl} ammonium fumarate (15).
Reagents and conditions: a) 1-chloro-2-methoxyethane, K CO , KI, DMF, reflux, 6h; b) i)
2
3
NaOH, ii) HCl, CH Cl; c) SOCl , toluene; d) (2S)-oxiran-2-ylmethanol, DMAP, CH Cl ; e)
3
2
2
2
tert-butylamine, i-PrOH; f) fumaric acid, Et O.
2
2
4

ACCEPTED MANUSCRIPT
Target compound 16 was prepared by a multiple-step synthesis according to Scheme 8, as
described in Tengler et al. [54]. Starting 4-butoxybenzoic acid was transformed to acid
chloride that formed a suitable epoxide with 2,3-epoxypropan-1-ol. The epoxide was opened
by reaction with tert-butylamine and gave 3-tert-butylamino-2-hydroxypropyl 4-
butoxybenzoate. The obtained base was transformed to hydrochloride salt 16 with increased
water solubility using ethereal HCl [54].
Scheme 8: Synthesis of 3-tert-butylamino-2-hydroxypropyl 4-butoxybenzoate
hydrochloride (16).
Reagents and conditions: a) toluene, PCl ; b) 2,3-epoxypropan-1-ol, CH Cl ; c) tert-
5
2
2
butylamine, i-PrOH; d) HCl, Et O.
2
2
.3. Binding affinities for sphingosine kinase 1 (Bioassays)
Putative SphK1 inhibitors were evaluated in 384-well high-throughput format as
described [63].
2
5

ACCEPTED MANUSCRIPT
Figure 8. SphK1 percent inhibition versus concentration plot for the compounds 2, 3 and 4.
Only 3 of the 17 compounds predicted to be potential inhibitors of SphK1 using virtual
screening showed activity as inhibitors (Figure 8). One dihydrobenzo[b]pyrimido[5,4-
f]azepine
(2)
and
two
alkyl
{3-/4-[1-hydroxy-2-(4-arylpiperazin-1-yl)ethyl]
phenyl}carbamates (3 and 4) were the compounds possessing significant inhibitory activities
against SphK1; compound 2 had relatively strong inhibitory activity with an IC of 12 µM.
5
0
While this inhibition was only moderate, it was considered significant for exploratory pre-
screening.
2
.4. Molecular modelling
In order to examine why these compounds were much weaker inhibitors than PF543, one
of the most potent SphK1 inhibitors reported so far [30,36], we next conducted a molecular
modelling study in which we simulated the molecular interactions of the active compounds
(2, 3 and 4) with SphK1 to analyze the different molecular interactions involved in the
complexes of the new compounds with SphK1.
The main objective was to assess the molecular interactions that can stabilize and
destabilize the different ligand-receptor complexes. We also included compound PF543 in
our molecular modelling study, which therefore allowed us to perform a comparative
analysis of the different activities displayed for these molecules in relation to their structural
differences.
The molecular modelling study was conducted in three different stages. In the first step
we performed a docking analysis using the Autodock program [64]. In the second stage of
this study, we carried out molecular dynamics (MD) simulations using the AMBER software
package [65]. From the trajectories obtained with the MD simulations, we performed an
2
6

ACCEPTED MANUSCRIPT
analysis per residue for the different compounds. Finally, to better understand the molecular
interactions involved in the different L-R complexes, a Quantum Theory of Atoms In
Molecules (QTAIM) study was carried out for the most representative structures of each
complex. In previous work, we have demonstrated the importance of these QTAIM studies
for understanding of the details of the different molecular interactions that stabilize or
destabilize the various complexes [66–70].
From docking studies (although they might be considered exploratory and preliminary), it
was already possible to see significant differences between PF543 and some of the new
compounds reported here. For PF543, the docking studies suggest that it binds in the well-
known J-shaped pocket (polar head in the polar zone of the receptor and the hydrophobic tail
at the hydrophobic areas) (Figure 9a). The same result was obtained for compound 2
(Figure 9b). In contrast, the docking analysis of compounds 4, and 3 suggests that these
compounds can bind in two ways: the known J form and in an inverted form in which the
polar head interacts with the hydrophobic region of the acceptor and the hydrophobic tail is
located at the polar region of the binding pocket (Figure 9b and 9c). It is interesting to note
that compound 2, which possessed the strongest inhibitory effect among the compounds
obtained here, binds to the active site only in the so-called correct form, the same as PF543.
2
7

ACCEPTED MANUSCRIPT
Figure 9. Spatial view of the different ligands bonded in the binding pocket of SphK1. The
blue and orange zones represent the cationic and hydrophobic portions of the active site,
respectively. a) PF543 (in green), b) PF543 (in green) superimposed on compound 2 (in
yellow). c) and d) compound 3 (in magenta) bonded in the two different ways.
Interesting results were also obtained from the molecular dynamics simulations. Analysis
per residue obtained from such simulations allowed us to define the main interactions that
stabilize the different complexes among the simulations (Figure 10). In general, the active
compounds studied here displayed their pharmacophoric portions in a closely related spatial
form to that displayed by PF543 [30] and other well-known inhibitors of SphK1 [28,29].
2
8