Computer-Assisted Selective Optimization of Side-Activities—from Cinalukast to a PPARα Modulator

Article abstract of DOI:10.1002/cmdc.201900286

Automated computational analogue design and scoring can speed up hit-to-lead optimization and appears particularly promising in selective optimization of side-activities (SOSA) where possible analogue diversity is confined. Probing this concept, we employ

Full text of DOI:10.1002/cmdc.201900286

Accepted Article
Title: Computer-assisted selective optimization of side-activities - from
cinalukast to a PPARα modulator
Authors: Julius Pollinger, Simone Schierle, Sebastian Neumann, Julia
Ohrndorf, Astrid Kaiser, and Daniel Merk
This manuscript has been accepted after peer review and appears as an
Accepted Article online prior to editing, proofing, and formal publication
of the final Version of Record (VoR). This work is currently citable by
using the Digital Object Identifier (DOI) given below. The VoR will be
published online in Early View as soon as possible and may be different
to this Accepted Article as a result of editing. Readers should obtain
the VoR from the journal website shown below when it is published
to ensure accuracy of information. The authors are responsible for the
content of this Accepted Article.
To be cited as: ChemMedChem 10.1002/cmdc.201900286
Link to VoR: http://dx.doi.org/10.1002/cmdc.201900286
A Journal of
www.chemmedchem.org

10.1002/cmdc.201900286
ChemMedChem
FULL PAPER
Computer-assisted selective optimization of side-activities - from
cinalukast to a PPARα modulator
Julius Pollinger[a]#, Simone Schierle[a]#, Sebastian Neumann[a], Julia Ohrndorf[a], Astrid Kaiser[a], and
Daniel Merk*[a]
[a]
#
Julius Pollinger, Simone Schierle, Sebastian Neumann, Julia Ohrndorf, Astrid Kaiser and Dr. Daniel Merk (ORCID: 0000-0002-5359-8128)
Institute of Pharmaceutical Chemistry
Goethe University Frankfurt
Max-von-Laue-Str. 9, D-60438 Frankfurt, Germany
E-mail: merk@pharmchem.uni-frankfurt.de
Julius Pollinger and Simone Schierle contributed equally to this work and are listed in alphabetical order
Supporting information for this article is given via a link at the end of the document.
Abstract: Automated computational analogue design and scoring
can speed up hit-to-lead optimization and appears particularly
promising in selective optimization of side-activities (SOSA) where
possible analogue diversity is confined. Probing this concept, we
employed the cysteinyl leukotriene receptor 1 (CysLT1R) antagonist
cinalukast as lead for which we discovered peroxisome proliferator-
activated receptor α (PPARα) modulatory activity. We automatically
generated a virtual library of close analogues and classified these
approx. 8000 compounds for PPARα agonism and CysLT1R
antagonism using automated affinity scoring and machine learning. A
computationally preferred analogue for SOSA was synthesized and in
vitro characterization indeed revealed a marked activity shift towards
enhanced PPARα activation and diminished CysLT1R antagonism.
Thereby, this prospective application study highlights the potential of
automating SOSA.
drug’s structural analogues for predicted activity on the desired
side-target appears very promising.
To probe this concept of computer-assisted SOSA, we have
selected the cysteinyl leukotriene receptor
1 (CysLT1R)
antagonist cinalukast (1) as lead compound for which we have
discovered weak partial agonistic activity on peroxisome
proliferator-activated receptor α (PPARα)[4,5] in a systematic
screening campaign. The fatty acid mimetic[6] cinalukast (1)[7,8]
has favorable physicochemical properties and its modular
architecture allows various structural modifications. Thus, 1
appears well suitable for SOSA but its low-yielding 5-step
synthesis renders a systematic structure activity relationship
(SAR) study of 1 an elaborate task. We employed 1 as lead for
computer-assisted SOSA aiming to structurally optimize the
drug’s PPARα agonism. We automated the design of cinalukast
analogues and their activity prediction on PPAR and CysLT1R.
After successful proof-of-concept evaluations, a computationally
preferred analogue was synthesized and biologically
characterized. It comprised higher activity on PPARα than 1 and
simultaneously revealed markedly reduced CysLT1R antagonism
confirming the potential of this computer-assisted structural
optimization approach.
Introduction
Computer-assisted drug discovery increasingly strives to
automate structural optimization of bioactive small molecules in
order to minimize experimental efforts.[1] Repeated design-
synthesize-test cycles of typical hit-to-lead expansion in medicinal
chemistry may considerably profit from modern computational
approaches to compound prioritization. A particular promising
application of such virtual compound optimization lies in its
combination with the concept of selective optimization of side-
activities[2,3] (SOSA) where (weak) off-target activities of approved
or experimental drugs are turned into the main activity of a new,
closely related structural analogue while diminishing the activity
on the original target. A great advantage of this concept are the
superior properties of drugs being used as lead compounds.
Since drugs have already been optimized for favorable
physicochemical properties, bioavailability and safety, the starting
point of SOSA-based drug discovery is by definition drug-like.
However, in order to conserve the favorable drug-like properties
of the lead compound selected for a SOSA campaign, structural
modifications usually need to be kept small during optimization.
Within such confined chemical space, virtual prioritization of a
Results and Discussion
Synthesis
Cinalukast derivatives 2-4 were synthesized in five to six steps
according to Scheme 1 starting from bromomethylketones 5-6
which were first cyclized with thioacetamide (7) to 2-
methylthiazoles 8 and 9. 4-tert-Butyl-2-methylthiazole 10 was
commercially available. Condensation of 8-10 with 3-
nitrobenzaldehyde (11) to nitrostyrylthiazoles 12-14 followed by
reduction with SnCl2/HCl then produced aminostyrylthiazoles 15-
17. Further reduction of the styryl moiety in 15 with H2/Pd(C)
afforded phenethylthiazole 18. 16-18 were then coupled with
dicarboxylic acid monoethylesters 19 and 20 to obtain esters 21-
23 using EDC*HCl and 4-DMAP, and saponification of 21-23
yielded test compounds 2-4. E/Z-isomerism was observed for 2
1
This article is protected by copyright. All rights reserved.

10.1002/cmdc.201900286
ChemMedChem
FULL PAPER
Scheme 1. Synthesis of cinalukast derivatives 2-4. Reagents and conditions: (a) thioacetamide, DMF, reflux, 4 h, 74-82%; (b) NaOAc, HOAc, reflux, 10 h,
14-24%; (c) SnCl2, EtOH, 65°C, 2-3 h, 55-66%; (d) H2 (1 bar), Pd(C), rt, 18 h, 94%. (e) EDC*HCl, 4-DMAP, CHCl3, 60°C, 12-16 h, 14-28%; (f) LiOH,
THF/H2O, rt, 12 h, 21-71%.
and 4 and their E-isomers (E/Z > 95%) were isolated for in vitro
pharmacological characterization by preparative HPLC. The
commercial sample of 1 contained the pure E-isomer. The
observed E/Z-isomerism turned out to be light dependent but was
not observed under the conditions of the in vitro test system used
in this study confirming that it does not affect the activity data (see
Supporting Information for details).
computational score and biological potency for some targets[17,18]
.
Especially for PPARs[18] scoring may be applied successfully
when the co-crystallized ligand in the template used for scoring
sufficiently resembles the studied molecules.
Aiming to evaluate the suitability of HYDE for our approach, we
first studied the correlation between the HYDE score for
analogues of 1 concerning affinity to PPARα and their activity on
the nuclear receptor. For this proof-of-concept, we selected
simple building blocks that were available in house to minimize
synthesis efforts and costs. We manually designed a small library
comprising 27 derivatives of 1 with variations in the thiazole
substituent, in the acidic side chain as well as in the geometry and
saturation degree of the central styrylthiazole moiety (Table S1).
To prioritize compounds for synthesis, we docked all 27 cinalukast
analogues into the PPARα ligand binding site using FlexX[19] and
calculated the HYDE scores for the top-ranking poses. The
PPARα-LBD complex X-ray 4CI4[20] served as structural template
as it contains a ligand with similar linear three-ring structure as 1.
The results suggested that both the thiazole substituent and the
acidic side chain length had marked impact on potency (Table S1).
Amongst the simplified derivatives with no variations in the central
styrylthiazole moiety, compound 2 (Scheme 1) comprising a 4-
phenylthiazole moiety and a 3-oxopropanoic acid side chain
appeared most favored (-63 kJ/mol).
To computationally assess the importance of the carboxylic acid
for HYDE scores on PPARα, we studied the influence of replacing
it in 2 with an amide, a nitrile, an alcohol, or a methyl group (Table
S2). As expected for the fatty acid sensor PPARα[5], all four
replacements were predicted as significantly less active by HYDE.
The scores also suggested that a hydrogen bond donor (amide,
alcohol) is essential while the scores for the nitrile and the terminal
methyl moiety were markedly lower. Based on these observations,
we selected carboxylic acid derivative 2 for synthesis and in vitro
characterization.
Biological evaluation
PPARα modulatory activity of 1 and derivatives 2-4 was
determined in a specific PPARα-Gal4-hybrid reporter gene
assay[9,10] relying on a chimeric receptor composed of the human
PPARα ligand binding domain (LBD) and the DNA binding domain
of the receptor Gal4 from yeast to govern reporter gene
expression. A Gal4-inducible firefly luciferase served as reporter
and a constitutively expressed renilla luciferase was used to
monitor test compound toxicity and transfection efficiency.
CysLT1R antagonism of 1 and 4 was assessed in a cell-based
Ca2+-flux assay in competition with 0.1 nM leukotriene D4[11]
.
Computer-assisted structural optimization
Characterization of cinalukast (1) on therapeutically relevant
nuclear receptors revealed partial agonistic activity on PPARα
(EC50 = 10±2 µM, 5.3±0.6-fold activation). With this attractive
side-activity, 1 was chosen as lead for SOSA-based optimization
towards PPARα agonism with computational support. To predict
the potency of 1 and analogues on the nuclear receptor PPARα,
we have chosen the HYDE scoring function[12]. HYDE estimates
free energies of binding for ligand-protein complexes focusing on
hydrogen bond formation between ligand and protein as well as
dehydration of binding sites[12]. Therein, it considers ligand
geometry and interaction angles. HYDE was successfully applied
on hydrophobic ligand binding sites previously[13–16] and appeared
suitable for predicting the interaction of 1 and analogues with the
highly lipophilic PPARα ligand binding site. Although scoring
functions for computational ranking of protein-ligand interactions
in many cases are error-prone, it was shown that scoring can
have predictive power and provide reliable correlation between
Concerning variations in the styrylthiazole residue, the HYDE
results suggested that changes in the geometry would be
detrimental (Table S1, row a) for potency on PPARα but that
reduction of the styryl residue might be tolerated (Table S1, rows
b and c). The 4-tert-butylthiazole derivative 3 (Scheme 1) with
reduced styryl residue and 3-oxopropanoic acid side chain was
2
This article is protected by copyright. All rights reserved.

10.1002/cmdc.201900286
ChemMedChem
FULL PAPER
predicted as weak PPARα modulating cinalukast analogue (-55
kJ/mol) and to cover a broader structural variation in this proof-of-
concept evaluation, we selected 3 for synthesis and in vitro
characterization.
Table 1. Computational activity predictions for top-5-ranked entries of the
combinatorial cinalukast analogue library. Activity on PPARα predicted by
HYDE-based affinity prediction (the respective top-scored pose was considered
for each molecule). CysLT1R antagonism was computationally assessed with a
random forest classification model to assign candidates to high (class 1) or low
(class 2) CysLT1R antagonistic potency. Cinalukast (1) for comparison. The top-
3 compounds 4a, 4b and 4 differed marginally in their predicted activities and 4
was selected for synthesis and in vitro characterization based on building block
availability.
Proof-of-concept cinalukast derivatives 2 and 3 showed partial
agonistic activity on PPARα with lower potency and activation
efficacy than 1 (2: EC50 = 11.7±0.5 µM, 3.0±0.1-fold act.; 3: EC50
= 73±17 µM, 3.3±0.5-fold act.; Table S3). No direct correlation
between the in vitro potencies of cinalukast (1) and derivatives 2
and 3 with their HYDE scores was observed but the prediction
agreed well with the rank order of potency, particularly when
considering EC50 value and activation efficacy. This result
sufficiently validated the HYDE-based compound prioritization for
further computer-assisted optimization of 1 towards selective
PPARα agonism.
ID
Structure
PPARα
HYDE
score [a]
CysLT1R
random
forest [b]
1
-83
class 1
As the prospective proof-of-concept study confirmed suitability of
HYDE to predict PPARα modulatory potency of 1 and derivatives,
we aimed to use this tool in a markedly expanded chemical space
of cinalukast analogues considered for structural optimization. For
this, we automatically generated a combinatorial library from all
suitable building blocks that were commercially available from
typical vendors. To retain the basic molecular architecture of drug
1, consider the observations from the proof-of-concept evaluation,
and keep structural changes small according to the SOSA
concept, we only varied the thiazole substituent and the acidic
side chain. The virtual combinatorial library was generated from
bromomethylketones (A) for the generation of the 4-substituted
styrylthiazole residue as well as haloalkyl carboxylic acid esters
(B) respectively dicarboxylic acid monoesters (C) to introduce the
acidic side chain (Scheme 2). The latter (B&C) covered linear and
branched alkyl chains as well as aromatic moieties for broader
structural variety. The resulting library contained 7922 cinalukast
analogues.
4a
-85
class 2
4b
-84
-81
-76
class 2
class 2
class 2
4
4c
4d
-69
class 2
[a] PPARα affinity prediction (HYDE): kJ/mol. [b] CysLT1R activity prediction:
class 1 - predicted IC50 < 1 µM; class 2 - predicted IC50 > 1 µM.
Scheme 2. Virtual combinatorial library design. The thiazole substituent
covered aliphatic and aromatic residues with varying size and substitution
patterns, the acidic side chain covered aromatic and linear or branched aliphatic
carboxylic acid moieties linked to the aminostyryl scaffold via an amine or amide
bond.
We then computationally assessed the potential of these PPARα-
favored cinalukast analogues to interact with CysLT1R. Due to the
lack of X-ray data of this G-protein coupled receptor as structural
template, the computational estimation of CysLT1R affinity had to
follow a ligand-based strategy. We retrieved all available
compounds with annotated activity on CysLT1R from ChEMBL[21]
(215 potent antagonists with IC50 < 1 µM, 904 weak antagonists
with IC50 > 1 µM or inactive examples) and used this collection of
compounds with reliable activity data to train a random forest
model for high/low activity classification on CysLT1R. Known
compounds were assigned to two classes with an activity
threshold of 1 µM. Molecules were represented by various
fingerprints (MACCS[22], Morgan[23], AtomPair[24]) for individually
training random forest models. Stratified 50/50% train-test
splitting with cross-validation (Table S5) revealed high prediction
accuracy for all models. All three models predicted cinalukast as
highly active (class 1) and agreed in the classification of
compounds 4 and 4a-d which were assigned to class 2 with high
In an automated workflow, the structures of this virtual
combinatorial library were docked with FlexX before the top-
scored docking poses were assessed with HYDE for their
predicted interaction with PPARα using 4CI4[20] as structural
template. According to the SOSA concept, the molecules were
also filtered for low lipophilicity (clogP ≤ 4). Among the five
computationally favored compounds for interaction with PPARα
(Table 1, top-30 in Table S4), the top-3 (4a, 4b, 4) hardly differed
in their scores for PPARα, whereas 4c and 4d comprised slightly
lower predicted affinities.
3
This article is protected by copyright. All rights reserved.

10.1002/cmdc.201900286
ChemMedChem
FULL PAPER
confidence (Table 1). In light of the close structural similarity of 4
and 4a-d to the known CysLT1R antagonist 1, this ligand-based
activity prediction seems reasonable.
and in vitro characterization based on its favorable activity
prediction profile and building block availability. In vitro
characterization of 4 revealed robust PPARα activation (13.7±0.1-
fold) with an EC50 value of 8.1±0.1 µM and markedly reduced
activity on CysLT1R compared to lead compound 1 (Figure 1b).
Thus, the structural modifications caused a remarkable activity
shift with enhanced PPARα activation efficacy (4: 13.7-fold vs. 1:
5.3-fold) and strongly reduced antagonistic activity on CysLT1R
(4: IC50 >> 100 nM vs. 1: IC50 ~ 1 nM). Moreover, with an aqueous
solubility of 21.7 mg/L (48 µM) and preferable lipophilicity (logP
1.6), 4 even exceeded the favorable properties of lead compound
1 (1.1 mg/L, 2.7 µM; logP 2.2) further confirming successful SOSA.
Inspection of the predicted binding mode (Figure 1c) of 4 in the
PPARα ligand binding site compared to 1 revealed only minor
differences which agrees with the compounds’ very similar HYDE
scores. Participation in the canonical hydrogen bond network with
Ser280, Tyr314, Tyr464 and His440 was observed for both
compounds. Due to its shortened acidic chain, the styrylthiazole
moiety of 4 was slightly shifted to this region of the binding site.
As a consequence, the benzene ring was bound in closer
proximity to Phe318. The large lipophilic cavity at the end of the
PPARα ligand binding site formed by Ile241, Leu247, Leu254,
Cys275 and Val332 accommodated the thiazole 4-substituents of
both 1 and 4 but appeared more favorably occupied by the
dimethoxyphenyl residue of 4.
Moreover, the predictions agreed with the limited data on the SAR
of cinalukast (1) and derivatives as CysLT1R antagonists
available in literature[25] which suggest importance of the 2,2-
diethyl-4-oxobutanoic acid residue both in terms of chain length
and ethyl substituents as well as of a small aliphatic ring as
thiazole substituent for high potency on CysLT1R. The latter is
also reflected by SAR data for other CysLT1R antagonists such
as zafirlukast[6,26,27] where small terminal aliphatic rings
(cyclobutyl, cyclopentyl) were essential for antagonism while
larger rings were markedly less active.
Conclusions
Combining virtual activity prediction and selective optimization of
side-activities appears very promising for refining side-target
activities of approved drugs towards bioactive new chemical
entities with minimized experimental efforts. Since the basic
concept of SOSA often demands that structural modifications are
confined during optimization in order to conserve the favorable
profile of the lead drug, focused virtual libraries of suitable
analogues for optimization can be generated. We hypothesized
that their computational prioritization is then capable of reducing
time and cost intensive design-synthesize-test cycles and can
speed up structural optimization for selective activity on the
desired target.
Following this concept, we have employed the CysLT1R
antagonist cinalukast (1) as lead for computer-assisted SOSA
towards PPARα agonism. HYDE was chosen as computational
scoring approach for PPARα activity as it was successfully
applied to highly lipophilic binding sites as found in PPARs,
previously, and yielded reliable results in a proof-of-concept
evaluation. CysLT1R antagonism was computationally predicted
using a random forest model as classifier between high and low
CysLT1R antagonistic potency. From a virtual combinatorial
library of close cinalukast analogues, 4 was computationally
favored both in terms of high predicted affinity to PPARα and low
estimated CysLT1R antagonism, and was consequently selected
for synthesis and characterization. In vitro profiling of 4 confirmed
the predicted activity shift towards higher activation efficacy on
PPARα and markedly improved selectivity over CysLT1R. Thus,
our straightforward computational approach to analogue selection
for SOSA successfully predicted the activity profile of 4.
Figure 1. (a) Automated docking in the PPARα ligand binding site (PDB-ID:
4CI4[20]) and HYDE scoring combined with a random forest model trained on
fingerprint representations of known CysLT1R antagonists successfully
classified cinalukast analogue 4 for enhanced PPARα agonism and diminished
CysLT1R antagonistic potency. (b) In vitro characterization of the
computationally favored analogue 4 revealed enhanced PPARα activation
efficacy and strongly reduced antagonism on CysLT1R (inhibition of CysLT1R
activation by 0.1 nM leukotriene D4). Results are the mean ± S.E.M., n≥3 for
PPARα, n=2 for CysLT1R. (c) Predicted binding modes of 1 (green) and 4 (blue)
in the PPARα ligand binding site (PDB-ID: 4CI4[20], co-crystallized ligand as
yellow wire; docking was performed with FlexX and visualized with UCSF
Chimera[28]). 1 and 4 form a very similar binding mode. The carboxylic acid
residues of 1 and 4 participate in the canonical hydrogen bond network with
Ser280, Tyr314, Tyr464 and His440 of the PPARα LBD. The styrylthiazole of 4
is slightly shifted towards the polar end of the binding site due to its shorter
carboxylic acid side chain. The cyclobutyl (1) and dimethoxyphenyl (4)
substituents occupy the lipophilic cavity at the end of the pocket, where the latter
moiety seems to fill the available lipophilic space more favorably.
The strong activity shift achieved by the computationally selected
cinalukast analogue 4 corroborates the potential of applying
modern activity prediction techniques on structural compound
Cinalukast analogue 4 comprising a shortened 2,2-dimethyl-3-
oxopropionic acid side chain and a bulky dimethoxyphenyl
substituent on the thiazole was selected for synthesis (Scheme 1)
4
This article is protected by copyright. All rights reserved.

10.1002/cmdc.201900286
ChemMedChem
FULL PAPER
optimization in early drug discovery. It suggests that experimental
efforts in (SOSA-based) optimization of drug molecules towards
desired activity profiles can be markedly reduced by automated
analogue design and scoring.
Aided. Mol. Des. 2013, 27, 15–29.
[13]
H. A. Carlson, R. D. Smith, K. L. Damm-Ganamet, J. A. Stuckey, A.
Ahmed, M. A. Convery, D. O. Somers, M. Kranz, P. A. Elkins, G.
Cui, et al., J. Chem. Inf. Model. 2016, 56, 1063–1077.
C. Ruggeri, N. Drinkwater, K. K. Sivaraman, R. S. Bamert, S.
McGowan, A. Paiardini, PLoS One 2015, 10, e0138957.
Z. Gaieb, S. Liu, S. Gathiaka, M. Chiu, H. Yang, C. Shao, V. A.
Feher, W. P. Walters, B. Kuhn, M. G. Rudolph, et al., J. Comput.
Aided. Mol. Des. 2018, 32, 1–20.
[14]
[15]
Acknowledgements
The authors thank Dr. Francesca Grisoni and Dr. Franca Klingler
for scientific advice and technical support. The free use of the
BioSolveIT chemoinformatics software suite as part of the
scientific challenge is gratefully acknowledged. Molecular
graphics and analyses performed with UCSF Chimera, developed
by the Resource for Biocomputing, Visualization, and Informatics
at the University of California, San Francisco, with support from
NIH P41-GM103311.
[16]
[17]
P. da Silva Figueiredo Celestino Gomes, F. Da Silva, G. Bret, D.
Rognan, J. Comput. Aided. Mol. Des. 2018, 32, 75–87.
M. Su, Q. Yang, Y. Du, G. Feng, Z. Liu, Y. Li, R. Wang, J. Chem.
Inf. Model. 2019, 59, 895–913.
[18]
[19]
J. Weber, M. Rupp, E. Proschak, Mol. Inform. 2012, 31, 631–633.
M. Rarey, B. Kramer, T. Lengauer, G. Klebe, J. Mol. Biol. 1996, 261,
470–89.
[20]
[21]
[22]
J. C. dos Santos, A. Bernardes, L. Giampietro, A. Ammazzalorso, B.
De Filippis, R. Amoroso, I. Polikarpov, J. Struct. Biol. 2015, 191,
332–340.
Keywords: automation • peroxisome proliferator-activated
receptor • nuclear receptor • virtual combinatorial library
A. P. Bento, A. Gaulton, A. Hersey, L. J. Bellis, J. Chambers, M.
Davies, F. A. Krüger, Y. Light, L. Mak, S. McGlinchey, et al., Nucleic
Acids Res. 2014, 42, D1083–D1090.
Associated Content
J. L. Durant, B. A. Leland, D. R. Henry, J. G. Nourse, J. Chem. Inf.
Comput. Sci. 2002, 42, 1273–1280.
Supporting Information available. The Supporting Information
contains Supplementary Tables S1-S5, computational methods
and model validation, synthetic procedures and analytical
characterization of 2-4 and their precursors, data on evaluation of
E/Z-isomerism, as well as methods for in vitro characterization.
[23]
[24]
H. L. Morgan, J. Chem. Doc. 1965, 5, 107–113.
R. E. Carhart, D. H. Smith, R. Venkataraghavan, J. Chem. Inf.
Model. 1985, 25, 64–73.
[25]
[26]
[27]
G. F. Field, J. R. Vermeulen, W. J. Zally, Cycloalkylthiazole
Derivatives, 1990, EP0355353A2.
Compound datasets used in this study and predicted binding
F. J. Brown, Y. K. Yee, L. A. Cronk, K. C. Hebbel, R. D. Krell, D. W.
Snyder, J. Med. Chem. 1990, 33, 1771–1781.
poses for HYDE assessment are provided as sdf or csv files.
S. Schierle, C. Flauaus, P. Heitel, S. Willems, J. Schmidt, A. Kaiser,
L. Weizel, T. Goebel, A. S. Kahnt, G. Geisslinger, et al., J. Med.
Chem. 2018, 61, 5758–5764.
References:
[1]
[2]
[3]
G. Schneider, Nat. Rev. Drug Discov. 2017, 17, 97–113.
C. G. Wermuth, J. Med. Chem. 2004, 47, 1303–14.
T. Langer, C. G. Wermuth, in Polypharmacology Drug Discov., John
Wiley & Sons, Inc., Hoboken, NJ, USA, 2012, pp. 227–243.
L. Michalik, J. Auwerx, J. P. Berger, V. K. Chatterjee, C. K. Glass, F.
J. Gonzalez, P. A. Grimaldi, T. Kadowaki, M. A. Lazar, S. O’Rahilly,
et al., Pharmacol. Rev. 2006, 58, 726–741.
[28]
E. F. Pettersen, T. D. Goddard, C. C. Huang, G. S. Couch, D. M.
Greenblatt, E. C. Meng, T. E. Ferrin, J. Comput. Chem. 2004, 25,
1605–1612.
[4]
[5]
[6]
C. Lamers, M. Schubert-Zsilavecz, D. Merk, Expert Opin. Ther. Pat.
2012, 22, 803–841.
E. Proschak, P. Heitel, L. Kalinowsky, D. Merk, J. Med. Chem.
2017, 60, 5235–5266.
[7]
[8]
M. O’Donnell, Ann. N. Y. Acad. Sci. 1991, 629, 413–5.
M. O’Donnell, H. J. Crowley, B. Yaremko, N. O’Neill, A. F. Welton, J.
Pharmacol. Exp. Ther. 1991, 259, 751–8.
[9]
J. Schmidt, M. Rotter, T. Weiser, S. Wittmann, L. Weizel, A. Kaiser,
J. Heering, T. Goebel, C. Angioni, M. Wurglics, et al., J. Med. Chem.
2017, 60, 7703–7724.
[10]
[11]
[12]
D. Flesch, S.-Y. Cheung, J. Schmidt, M. Gabler, P. Heitel, J. S.
Kramer, A. Kaiser, M. Hartmann, M. Lindner, K. Lüddens-Dämgen,
et al., J. Med. Chem. 2017, 60, 7199–7205.
H. M. Sarau, R. S. Ames, J. Chambers, C. Ellis, N. Elshourbagy, J.
J. Foley, D. B. Schmidt, R. M. Muccitelli, O. Jenkins, P. R. Murdock,
et al., Mol. Pharmacol. 1999, 56, 657–63.
N. Schneider, G. Lange, S. Hindle, R. Klein, M. Rarey, J. Comput.
5
This article is protected by copyright. All rights reserved.

10.1002/cmdc.201900286
ChemMedChem
FULL PAPER
Entry for the Table of Contents
Selective optimization of side-activities (SOSA) aims to invert activity profiles of drug molecules. We have automated this approach
using the CysLT1R antagonist cinalukast as lead which has a weak side-activity on PPARα. Automated analogue design and scoring
produced a descendant of the drug that was experimentally confirmed as more active and markedly more selective towards the side-
target corroborating the concept of computer-assisted SOSA.
6
This article is protected by copyright. All rights reserved.