Exploring the chemical space of multitarget ligands using aligned self-organizing maps

Article abstract of DOI:10.1021/ml4002562

Design of multitarget drugs and polypharmacological compounds has become popular during the past decade. However, the main approach to design such compounds is to link two selective ligands via a flexible linker. Although such chimeric ligands often have reasonable potency in vitro, the in vivo efficacy is low due to high molecular weight, low ligand efficiency, and poor pharmacokinetic profile. We developed an unprecedented in silico approach for fragment-based design of multitarget ligands. It relies on superposition of the chemical spaces related to the affinity on single targets represented by self-organizing maps. We used this approach for screening of molecular fragments, which bind to the enzymes 5-lipoxygenase (5-LO) and soluble epoxide hydrolase (sEH). Using STD-NMR and activity-based assays, we were able to identify fragments binding to both targets. Furthermore, we were able to expand one of the fragments to a potent dual inhibitor bearing a reasonable molecular weight (MW = 446) and high affinity to both targets (IC₅₀ of 0.03 μM toward 5-LO and 0.17 μM toward sEH).

Full text of DOI:10.1021/ml4002562

Letter
pubs.acs.org/acsmedchemlett
Exploring the Chemical Space of Multitarget Ligands Using Aligned
Self-Organizing Maps
Janosch Achenbach,† Franca-Maria Klingler,† Rene Blocher,† Daniel Moser,† Ann-Kathrin Hafner,†
́
̈
̈
Carmen B. Rodl,† Simon Kretschmer,† Bjorn Kruger,‡ Frank Lohr,§ Holger Stark,† Bettina Hofmann,†
̈
̈
̈
̈
Dieter Steinhilber,† and Ewgenij Proschak*,†
†Institute of Pharmaceutical Chemistry, ZAFES/OSF, Goethe University, Max-von-Laue-Strasse 9, D-60438 Frankfurt am Main,
Germany
‡Chemical R&DDrug Design, Merz Pharmaceuticals GmbH, Eckenheimer Landstrasse 100, D-60318 Frankfurt, Germany
§Institute of Biophysical Chemistry, Goethe University, Max-von-Laue Strasse 9, D-60438 Frankfurt am Main, Germany
S
* Supporting Information
ABSTRACT: Design of multitarget drugs and polypharmacological compounds has become popular during the past decade.
However, the main approach to design such compounds is to link two selective ligands via a flexible linker. Although such
chimeric ligands often have reasonable potency in vitro, the in vivo efficacy is low due to high molecular weight, low ligand
efficiency, and poor pharmacokinetic profile. We developed an unprecedented in silico approach for fragment-based design of
multitarget ligands. It relies on superposition of the chemical spaces related to the affinity on single targets represented by self-
organizing maps. We used this approach for screening of molecular fragments, which bind to the enzymes 5-lipoxygenase (5-LO)
and soluble epoxide hydrolase (sEH). Using STD-NMR and activity-based assays, we were able to identify fragments binding to
both targets. Furthermore, we were able to expand one of the fragments to a potent dual inhibitor bearing a reasonable molecular
weight (MW = 446) and high affinity to both targets (IC50 of 0.03 μM toward 5-LO and 0.17 μM toward sEH).
KEYWORDS: Aligned self-organizing maps, 5-lipoxygenase, soluble epoxide hydrolase, structure−activity relationships
he one drug−one target−one disease paradigm in drug
Tdiscovery has been reconsidered during the last decade.
This paradigm change was mainly caused by high attrition rates
in drug approvals due to toxicity and lack of efficacy. On top of
that, the results of post-genomic and network biology showed
that putative drug targets rarely act within isolated systems but
rather as a part of a highly connected network.1 Furthermore,
the efficacy of several approved drugs has been traced back to
the interaction with multiple targets.2 Inhibition of a single
target in such a network might not lead to the desired
therapeutic effect, which explains a large proportion of clinical
study failures. A paradigm shift towards designed polypharma-
cology should overcome the lack of efficacy;3 however, the
design of selective multi-target drugs is still challenging. In
particular, sufficient affinity to each target as well as an adequate
pharmacokinetic profile has to be considered.4 Early design
strategies comprise linking the relevant pharmacophores of
known selective ligands, but these methods often lead to
compounds exhibiting high molecular weight and low ligand
efficiency.3,4 The application of computer-aided techniques
could be beneficial for the development of multi-target drugs5
as it was recently shown by Besnard et al.6
Fragment-based techniques have been broadly applied to
design potent and selective drugs.7,8 In this study, we extend
this successful rational design strategy to multitarget ligands.
We present an in silico approach based on alignment of self-
organizing maps (SOM),9,10 which we call multiSOM, for the
identification of multitarget fragments with low molecular
weight bearing space for further optimization. Our workflow
Received: July 9, 2013
Accepted: October 23, 2013
Published: October 23, 2013
© 2013 American Chemical Society
1169
dx.doi.org/10.1021/ml4002562 | ACS Med. Chem. Lett. 2013, 4, 1169−1172

ACS Medicinal Chemistry Letters
Letter
comprises the search for common substructures of known
ligands for each target followed by the identification of
multitarget relevant substructures among the target-specific
substructures. We used our multiSOM approach to retrieve
molecular fragments, which target 5-lipoxygenase (5-LO)11 and
soluble epoxide hydrolase (sEH).12,13 We validated these
findings by saturation transfer difference (STD)-NMR14,15
and functional in vitro assay systems. Both enzymes are part of
the arachidonic acid cascade and involved in inflammatory
processes, pain, and cardiovascular diseases.16,17 The simulta-
neous inhibition of both enzymatic pathways was shown to
have synergistic effects in treatment of inflammation in vivo.18
The initial step in our screening workflow was the
identification of characteristical molecular substructures of
known active compounds for each target (5-LO, sEH). For the
identification of the relevant substructures for each target we
created two compound sets derived from the ChEMBLdb
(v. 12) with annotated affinity data for each target.19 We
considered only ligands with IC50 and Ki values equal or less
than 10 μM. The DrugBank (v. 3.0) database served as a
background distribution of the drug-like chemical space for the
multiSOM training.20 We generated a virtual fragment library as
a subset of all compounds available from Specs (v. May 2011)
by applying the “Astex Rule of 3” filter.21 All computational
procedures described were implemented as KNIME nodes or
workflows.22
We used the Molecular Substructure Miner (MoSS) KNIME
node,23 a maximum common substructure based approach to
search for frequent substructures in each active compound set.
A substructure has been defined as frequent for one of the
targets if it contains more than seven heavy atoms and occurs in
at least 5% of the known active compounds of the
corresponding target and only at most in 1% of the DrugBank
compounds. We performed that kind of search for both known
active sets and additionally (with swapped active/inactive
definition) for the DrugBank compounds to generate a set of
prevalent frequent substructures as background distribution.
We could find 173 substructures that were characteristic for
sEH inhibitors, 150 for 5-LO inhibitors, and 312 that were
common for DrugBank compounds representing the drug-like
chemical space. The subsequent step was the identification of
potential dual-target substructures, i.e., the intersection of both
sets of frequent substructures derived from sEH and 5-LO
ligands that are most similar to each other. For the
identification of this intersection, we used our multiSOM
approach, which is based on the alignment of multiple self-
organizing maps (Figure 1).
Figure 1. (a) Schematically depicted principle of the multiSOM
approach. Two chemical subspaces described by the ligands of two
distinct targets are aligned to determine the shared chemical space. (b)
Primary maps of the multiSOM approach trained on the 5-LO/
DrugBank and sEH/DrugBank frequent substructures (top). The
resulting multiSOM map represents the similarity of the underlying,
aligned neurons by its third dimension (bottom); a higher lying
neuron means a higher similarity of the aligned, original neurons.
inactive substructure sets, we used the same multiSOM
approach for the virtual screening. Because of the difference
in size between the Specs fragment library and the frequent
substructures collection, we decided to perform multiple
multiSOM calculations in parallel. Each of these multiSOMs
was trained on the dual substructures, the inactive DrugBank
substructures and an equally sized subset of the Specs
fragments library. Again, multiSOM neurons, which aggregated
only dual relevant substructures and Specs fragments were
further analyzed and served as a source for the final list of
fragments purchased. We performed this procedure with three
different fingerprint types, including the RDKit FeatMorgan,
the RDKit layered fingerprint25, and the 2D CATS descriptor,26
to maximize the chemical diversity of the purchased fragments.
All three screenings yielded a total of 274 fragments of which
we purchased 24 compounds regarding manual inspection and
availability.
We had to determine the activity of the purchased fragments
on two separate targets, which led to the decision to use a
combination of two complementary in vitro assay approaches
consisting of STD-NMR as binding assay under the same
conditions for both targets, followed by functional assays for
5‑LO and sEH. STD-NMR allows the testing of multiple
fragments in a single experiment. Because of this fact, we
clustered all 24 compounds in five subsets leading to a total of
ten STD-NMR experiments for both targets. The composition
of each subset was aimed to maximize the difference between
the chemical shifts of the included fragments to facilitate the
discrimination of the subsequent STD-NMR results. All
fragments were tested at a final concentration of 400 μM.
Despite these preparations, we were not able to determine an
utterly clear discrimination between all fragments on the STD-
NMR spectra; thus, we had to classify the fragment in binders,
nonbinders, and potential binders. We additionally determined
the inhibitory potency of all fragments at a single concentration
of 100 μM in the particular functional in vitro assays for both
recombinant enzymes. In the case of the 5-LO, we used a
HPLC-based assay to determine the remaining 5-LO product
In this study we used the RDKit FeatMorgan fingerprint, a
functional-class, generalized extended-connectivity fingerprint24
to encode the identified frequent substructures. The multiSOM
approach yielded three self-organized maps containing 42
neurons: two primary maps, each trained on one of both active
sets and the inactive set, respectively, and additionally one
aligned multiSOM (Figure 1b). High-lying neurons that
contain similar substructures only from both active sets
(colored) and not from the background set (gray) have been
considered to be relevant for both targets. All substructures
contained on those neurons were used for the following virtual
screening procedure.
The resulting collection containing 229 dual relevant
substructures was used to search for similar fragments in the
prefiltered Specs database containing 8417 fragment-like small
molecules. On the basis of the previously prepared active and
1170
dx.doi.org/10.1021/ml4002562 | ACS Med. Chem. Lett. 2013, 4, 1169−1172

ACS Medicinal Chemistry Letters
Letter
formation.27 The sEH activity was determined in a
fluorescence-based assay, which uses (3-phenyl-oxiranyl)-acetic
acid cyano-(6-methoxy-naphthalen-2-yl)-methyl ester
(PHOME) as substrate (for all STD-NMR data and inhibition
at 100 μM, see Supporting Information)
Afterward, we determined an IC50 of all fragments, which
showed a potential dual binding in the STD-NMR and an
inhibition of at least 25% in the functional assays. The results of
the 11 compounds matching these criteria are shown in Table
1. We were able to identify fragments with a functional IC50
value in the range of 3 to 379 μM for sEH and in the range of 7
to 237 μM for 5-LO. For both targets, we identified known,
already described active scaffolds like benzimidazole, urea, and
amide (l, q, o, and r) for the sEH or imidazo-[1,2-a]-pyridine,
aminothiazole, and benzoxazole (g, w, and x) for the 5-LO as
well as new, so far not described scaffolds like compounds m
and n.
Furthermore, we were able to identify five fragments i, m, n,
w, and x, which exhibit inhibitory activity on both targets.
Especially compound n showed inhibitions in a low micromolar
concentration range for both sEH and 5-LO. To investigate the
optimization potential of at least one of the fragments, we
searched our in-house library for compounds containing the
dual hit fragments. We found a derivative of fragment w, an
enlarged aminothiazole 1 (ST-1366, Figure 2).
Table 1. IC50 Values (μM) and STD-NMR Binding Data of
the Most Promising Candidates
Figure 2. Substructure search in our in-house library based on
compound w lead to compound 1.
Compound 1 was subsequently tested in both assay systems
yielding IC50 values of 0.03 μM at 5-LO and 0.17 μM at sEH
(Figure 3). Thus, compound 1 is a good starting point for lead
optimization.
Figure 3. Dose−response curves for the IC50 determination of
compound 1 on 5-LO (green) and sEH (blue).
In conclusion, our study presents a novel approach for the
development of multitarget drugs. We show that fragment-
based techniques are applicable to design multitarget ligands, as
postulated in theoretical papers by Morphy and Rancovic29,30
and Bottegoni et al.5 We suggest an in silico technique for
recognition of molecular fragments suitable for multitarget drug
design, which led to enrichment of dual fragments targeting
sEH and 5-LO in a prospective study. An exemplary testing of
an enlarged fragment yields a potent lead structure for further
optimization. Further studies following this multiSOM strategy
are needed to demonstrate the broad applicability of diverse
fragment-based design.
ASSOCIATED CONTENT
* Supporting Information
More detailed description of the multiSOM approach, the
identified relevant substructures for each target, all purchased
compounds and assay setups, and synthesis of compound 1.
■
S
a
IC50 values in μM; each experiment was performed at least three
b
times. STD-NMR data: checkmark = binder; × = nonbinder; ∼ =
ambiguous; and n.s. = not soluble under STD conditions.
1171
dx.doi.org/10.1021/ml4002562 | ACS Med. Chem. Lett. 2013, 4, 1169−1172

ACS Medicinal Chemistry Letters
Letter
(16) Hwang, S. H.; Wecksler, A. T.; Wagner, K.; Hammock, B. D.
Rationally designed multitarget agents against inflammation and pain.
Curr. Med. Chem. 2013, 20, 1783−99.
This material is available free of charge via the Internet at
http://pubs.acs.org.
(17) Meirer, K.; Rodl, C. B.; Wisniewska, J. M.; George, S.; Hafner,
̈
̈
AUTHOR INFORMATION
■
A.-K.; la Buscato, E.; Klingler, F.-M.; Hahn, S.; Berressem, D.;
́
Wittmann, S. K.; Steinhilber, D.; Hofmann, B.; Proschak, E. Synthesis
and structure activity relationship studies of novel dual inhibitors of
soluble epoxide hydrolase and 5-lipoxygenase. J. Med. Chem. 2013, 56,
1777−1781.
Corresponding Author
*(E.P.) Tel: +49 69 798 29301. Fax: +49 69 798 29258. E-mail:
proschak@pharmchem.uni-frankfurt.de.
Notes
(18) Liu, J.-Y.; Yang, J.; Inceoglu, B.; Qiu, H.; Ulu, A.; Hwang, S.-H.;
Chiamvimonvat, N.; Hammock, B. D. Inhibition of soluble epoxide
hydrolase enhances the anti-inflammatory effects of aspirin and 5-
lipoxygenase activation protein inhibitor in a murine model. Biochem.
Pharmacol. 2010, 79, 880−7.
(19) Gaulton, A.; Bellis, L. J.; Bento, A. P.; Chambers, J.; Davies, M.;
Hersey, A.; Light, Y.; McGlinchey, S.; Michalovich, D.; Al-Lazikani, B.;
Overington, J. P. ChEMBL: a large-scale bioactivity database for drug
discovery. Nucleic Acids Res. 2012, 40, D1100−7.
The authors declare no competing financial interest.
ACKNOWLEDGMENTS
■
This work was supported by the Deutsche Forschungsgemein-
schaft (Sachbeihilfe PR 1405/2-1 and SFB 1039, Teilprojekt
A07), Oncogenic Signaling Frankfurt (OSF), Deutsches
Konsortium fur Translationale Krebsforschung (DKTK), and
̈
(20) Knox, C.; Law, V.; Jewison, T.; Liu, P.; Ly, S.; Frolkis, A.; Pon,
A.; Banco, K.; Mak, C.; Neveu, V.; Djoumbou, Y.; Eisner, R.; Guo, A.
C.; Wishart, D. S. DrugBank 3.0: a comprehensive resource for
“omics” research on drugs. Nucleic Acids Res. 2011, 39, D1035−41.
(21) Congreve, M.; Carr, R.; Murray, C.; Jhoti, H. A “rule of three”
for fragment-based lead discovery? Drug Discovery Today 2003, 8,
876−7.
LOEWE-Schwerpunkt: Anwendungsorientierte Arzneimittel-
forschung. J.A. thanks Merz Pharmaceuticals for a fellowship.
J.A. and R.B. thank Else Kroner-Fresenius Foundation (EKFS),
̈
Research Training Group Translational Research Innovation−
Pharma (TRIP) for a fellowship.
(22) Berthold, M. R.; Cebron, N.; Dill, F.; Gabriel, T. R.; Kotter, T.;
̈
REFERENCES
(1) Jeong, H.; Mason, S. P.; Barabas
■
Meinl, T.; Ohl, P.; Thiel, K.; Wiswedel, B. KNIME: the Konstanz
information miner. ACM SIGKDD Explorations Newsletter 2009, 11,
26.
́
i, A. L.; Oltvai, Z. N. Lethality
and centrality in protein networks. Nature 2001, 411, 41−2.
́
(2) Zimmermann, G. R.; Lehar, J.; Keith, C. T. Multi-target
(23) Borgelt, C.; Berthold, M. R. In Proceedings 2002 IEEE
International Conference on Data Mining. ICDM; IEEE Comput. Soc.:
Washington, DC, 2002; pp 51−58.
therapeutics: when the whole is greater than the sum of the parts.
Drug Discovery Today 2007, 12, 34−42.
(3) Morphy, R.; Kay, C.; Rankovic, Z. From magic bullets to
designed multiple ligands. Drug Discovery Today 2004, 9, 641−51.
(4) Morphy, R.; Rankovic, Z. Designed multiple ligands. An
emerging drug discovery paradigm. J. Med. Chem. 2005, 48, 6523−43.
(5) Bottegoni, G.; Favia, A. D.; Recanatini, M.; Cavalli, A. The role of
fragment-based and computational methods in polypharmacology.
Drug Discovery Today 2011, 17, 23−34.
(6) Besnard, J.; Ruda, G. F.; Setola, V.; Abecassis, K.; Rodriguiz, R.
M.; Huang, X.-P.; Norval, S.; Sassano, M. F.; Shin, A. I.; Webster, L.
A.; Simeons, F. R. C.; Stojanovski, L.; Prat, A.; Seidah, N. G.; Constam,
D. B.; Bickerton, G. R.; Read, K. D.; Wetsel, W. C.; Gilbert, I. H.;
Roth, B. L.; Hopkins, A. L. Automated design of ligands to
polypharmacological profiles. Nature 2012, 492, 215−20.
(7) Hajduk, P. J.; Greer, J. A decade of fragment-based drug design:
strategic advances and lessons learned. Nat. Rev. Drug Discovery 2007,
6, 211−9.
(24) Rogers, D.; Hahn, M. Extended-connectivity fingerprints. J.
Chem. Inf. Model. 2010, 50, 742−54.
(25) Landrum, G. RDKit: Open-source cheminformatics. http://
www.rdkit.org.
(26) Schneider, G.; Neidhart, W.; Giller, T.; Schmid, G. Scaffold-
Hopping” by Topological Pharmacophore Search: A Contribution to
Virtual Screening. Angew. Chem., Int. Ed. 1999, 38, 2894−2896.
(27) Brungs, M.; Radmark, O.; Samuelsson, B.; Steinhilber, D.
̊
Sequential induction of 5-lipoxygenase gene expression and activity in
Mono Mac 6 cells by transforming growth factor beta and 1,25-
dihydroxyvitamin D3. Proc. Natl. Acad. Sci. U.S.A. 1995, 92, 107−111.
(28) Morisseau, C.; Hammock, B. D. In Current Protocols in
Toxicology; Bus, J. S., Costa, L. G., Hodgson, E., Lawrence, D. A., Reed,
D. J., Eds.; John Wiley & Sons, Inc.: New York, 2007; pp 33:4.23.1−
4.23.18.
(29) Morphy, R.; Rankovic, Z. Fragments, network biology and
designing multiple ligands. Drug Discovery Today 2007, 12, 156−60.
(30) Stark, H. Turning from monogamy to strategic promiscuity.
Drug Discovery Today 2004, 9, 736−7.
(8) Congreve, M.; Chessari, G.; Tisi, D.; Woodhead, A. J. Recent
developments in fragment-based drug discovery. J. Med. Chem. 2008,
51, 3661−80.
(9) Kohonen, T. Self-organized formation of topologically correct
feature maps. Biol. Cybern. 1982, 43, 59−69.
(10) Digles, D.; Ecker, G. F. Self-organizing maps for in silico
screening and data visualization. Mol. Inf. 2011, 30, 838−846.
(11) Werz, O.; Steinhilber, D. Therapeutic options for 5-lipoxygenase
inhibitors. Pharmacol. Ther. 2006, 112, 701−18.
(12) Imig, J. D.; Hammock, B. D. Soluble epoxide hydrolase as a
therapeutic target for cardiovascular diseases. Nat. Rev. Drug Discovery
2009, 8, 794−805.
(13) Shen, H. C.; Hammock, B. D. Discovery of inhibitors of soluble
epoxide hydrolase: a target with multiple potential therapeutic
indications. J. Med. Chem. 2012, 55, 1789−808.
(14) Mayer, M.; Meyer, B. Group epitope mapping by saturation
transfer difference NMR to identify segments of a ligand in direct
contact with a protein receptor. J. Am. Chem. Soc. 2001, 123, 6108−17.
(15) Wagstaff, J. L.; Taylor, S. L.; Howard, M. J. Recent
developments and applications of saturation transfer difference nuclear
magnetic resonance (STD NMR) spectroscopy. Mol. Biosyst. 2013, 9,
571−7.
1172
dx.doi.org/10.1021/ml4002562 | ACS Med. Chem. Lett. 2013, 4, 1169−1172