Discovery of a Selective Aurora A Kinase Inhibitor by Virtual Screening

Article abstract of DOI:10.1021/acs.jmedchem.6b00709

Here we report the discovery of a selective inhibitor of Aurora A, a key regulator of cell division and potential anticancer target. We used the atom category extended ligand overlap score (xLOS), a 3D ligand-based virtual screening method recently developed in our group, to select 437 shape and pharmacophore analogs of reference kinase inhibitors. Biochemical screening uncovered two inhibitor series with scaffolds unprecedented among kinase inhibitors. One of them was successfully optimized by structure-based design to a potent Aurora A inhibitor (IC₅₀ = 2 nM) with very high kinome selectivity for Aurora kinases. This inhibitor locks Aurora A in an inactive conformation and disrupts binding to its activator protein TPX2, which impairs Aurora A localization at the mitotic spindle and induces cell division defects. This phenotype can be rescued by inhibitor-resistant Aurora A mutants. The inhibitor furthermore does not induce Aurora B specific effects in cells.

Full text of DOI:10.1021/acs.jmedchem.6b00709

Subscriber access provided by University of Sussex Library
Article
Discovery of a Selective Aurora A Kinase Inhibitor by Virtual Screening
Falco Kilchmann, Maria Jose Marcaida, Sachin Kotak, Thomas Schick,
Silvan D Boss, Mahendra Awale, Pierre Gönczy, and Jean-Louis Reymond
J. Med. Chem., Just Accepted Manuscript • DOI: 10.1021/acs.jmedchem.6b00709 • Publication Date (Web): 08 Jul 2016
Downloaded from http://pubs.acs.org on July 12, 2016
Just Accepted
“Just Accepted” manuscripts have been peer-reviewed and accepted for publication. They are posted
online prior to technical editing, formatting for publication and author proofing. The American Chemical
Society provides “Just Accepted” as a free service to the research community to expedite the
dissemination of scientific material as soon as possible after acceptance. “Just Accepted” manuscripts
appear in full in PDF format accompanied by an HTML abstract. “Just Accepted” manuscripts have been
fully peer reviewed, but should not be considered the official version of record. They are accessible to all
readers and citable by the Digital Object Identifier (DOI®). “Just Accepted” is an optional service offered
to authors. Therefore, the “Just Accepted” Web site may not include all articles that will be published
in the journal. After a manuscript is technically edited and formatted, it will be removed from the “Just
Accepted” Web site and published as an ASAP article. Note that technical editing may introduce minor
changes to the manuscript text and/or graphics which could affect content, and all legal disclaimers
and ethical guidelines that apply to the journal pertain. ACS cannot be held responsible for errors
or consequences arising from the use of information contained in these “Just Accepted” manuscripts.
Journal of Medicinal Chemistry is published by the American Chemical Society. 1155
Sixteenth Street N.W., Washington, DC 20036
Published by American Chemical Society. Copyright © American Chemical Society.
However, no copyright claim is made to original U.S. Government works, or works
produced by employees of any Commonwealth realm Crown government in the course
of their duties.

Page 1 of 81
Journal of Medicinal Chemistry
1
2
3
4
5
6
7
8
9
Discovery of a Selective Aurora A Kinase Inhibitor by Virtual
Screening
a)§
a)c)§
b)d)
a)
a)
Falco Kilchmann, Maria J. Marcaida,
Sachin Kotak, Thomas Schick, Silvan D. Boss,
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
a)
b)
a)
Mahendra Awale, Pierre Gönczy * and JeanꢀLouis Reymond *
a)
Department of Chemistry and Biochemistry, National Center of Competence in Research NCCR
Chemical Biology and NCCR TransCure, University of Berne, Freiestrasse 3, 3012 Berne
b)
Switzerland; eꢀmail: jeanꢀlouis.reymond@dcb.unibe.ch; Swiss Institute for Experimental Cancer
Research (ISREC), School of Life Sciences, National Center of Competence in Research NCCR
Chemical Biology, Swiss Federal Institute of Technology (EPFL), CH-1015 Lausanne, Switzerland;
e-mail: pierre.gonczy@epfl.ch
c)
Present address: Institute of Bioengineering, School of Life Sciences, Swiss Federal Institute of
Technology (EPFL), CH-1015 Lausanne, Switzerland
d)
Present address: Department of Microbiology and Cell Biology (MCB), Indian Institute of
Science (IISc), Bangalore, India
§
These authors contributed equally to the work
ACS Paragon Plus Environment

Journal of Medicinal Chemistry
Page 2 of 81
2
1
2
3
4
5
6
7
8
9
1
1
1
1
1
1
1
1
1
1
2
2
2
2
2
2
2
2
2
2
3
3
3
3
3
3
3
3
3
3
4
4
4
4
4
4
4
4
4
4
5
5
5
5
5
5
5
5
5
5
6
Abstract
Here we report the discovery of a selective inhibitor of Aurora A, a key regulator of cell division
and potential antiꢀcancer target. We used atom category extended ligand overlap score (xLOS), a
3
D ligand based virtual screening method recently developed in our group, to select 437 shape and
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
pharmacophore analogs of reference kinase inhibitors. Biochemical screening uncovered two
inhibitor series with scaffolds unprecedented among kinase inhibitors. One of them was
successfully optimized by structureꢀbased design to a potent Aurora A inhibitor (IC50 = 2 nM) with
very high kinome selectivity for Aurora kinases. This inhibitor locks Aurora A in an inactive
conformation and disrupts binding to its activator protein TPX2, which impairs Aurora A
localization at the mitotic spindle and induces cell division defects. This phenotype can be rescued
by inhibitorꢀresistant Aurora A mutants. The inhibitor furthermore does not induce Aurora B
specific effects in cells.
ACS Paragon Plus Environment

Page 3 of 81
Journal of Medicinal Chemistry
3
1
2
3
4
5
6
7
8
9
Introduction
Computerꢀaided drug design exploits accumulated experimental and theoretical knowledge of drugꢀ
target interactions to preselect small sets of compounds for experimental evaluation enabling faster
1
ꢀ3
and more efficient drug discovery. While structureꢀbased methods exploit structural information
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
4
, 5
on the protein targets, ligandꢀbased virtual screening (LBVS) methods employ similarity to
6
previously known compounds to propose new analogs for testing. In LBVS substructure similarity
7
searches, for example using Tanimoto similarity with Daylightꢀtype substructure fingerprints, can
be used to find available analogs with varying substituents around conserved scaffolds. On the other
hand shape and pharmacophore similarity methods, in particular those comparing the 3D structure
8
9
10
11
of molecules either by overlays (e.g. FlexS, Topomer shape similarity searching, SurflexꢀSim,
12
13
14
15
16
pharmACOphore, ROCS ) or via 3Dꢀfingerprints (e.g. USRCAT, 3DXfp ) can suggest
analogs with entirely different scaffolds and thus open the way to new compound series. This
transition to new scaffolds while preserving shapes and pharmacophores is referred to as scaffoldꢀ
17
hopping.
Here we applied scaffold hopping LBVS to identify a new and selective inhibitor of Aurora
A, an evolutionary conserved Ser/Thr kinase essential for proper progression through mitosis and
18, 19
one of the most intensively researched kinase targets due to its significance in cancer.
Although
many potent inhibitors of this kinase are known, most of these also inhibit Aurora B, a
chromosomal passenger protein required notably for cytokinesis and which shares over 85%
20, 21
sequence homology with Aurora A in the kinase domain.
Alisertib (MLN8237, 1), one of the
22
most selective Aurora A inhibitors reported to date, is currently under advanced clinical
23
investigation for the treatment of various solid and hematological malignancies (Figure 1).
2
4
Interestingly 1 and its close relative MLN8054 (2) belong to a scaffold that is unique among the
2
5
4
,874 scaffolds (as defined by Bemis and Murcko) occurring in 11,286 kinase inhibitors with
26ꢀ32
potencies better than 50 nM listed in ChEMBL.
Furthermore the diversity of scaffolds is quite
ACS Paragon Plus Environment

Journal of Medicinal Chemistry
Page 4 of 81
4
1
2
3
4
5
6
7
8
9
1
1
1
1
1
1
1
1
1
1
2
2
2
2
2
2
2
2
2
2
3
3
3
3
3
3
3
3
3
3
4
4
4
4
4
4
4
4
4
4
5
5
5
5
5
5
5
5
5
5
6
high among Aurora A kinase inhibitors, with 174 different scaffolds occurring among 329
inhibitors with potencies better than 50 nM listed in ChEMBL. This suggests that each inhibitor
represents an already optimized compound for each scaffold, and that discovering further selective
inhibitors for this particular kinase might require identifying different scaffolds, for example using
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
23, 24
scaffold hopping LBVS.
To perform scaffold hopping LBVS we used the atom category extended Ligand Overlap
3
3
Score (xLOS), a 3D shape and pharmacophore similarity algorithm recently developed in our
34, 35
group. The xLOS algorithm compares molecules by maximizing the ligand overlap score
computed separately for three atom categories (hydrophobic, Hꢀbond acceptor and Hꢀbond donor
3
6, 37
atoms) previously found to be useful for pharmacophore fingerprint design.
The method shows
excellent scaffold hopping properties, which may be attributed to the fact that the algorithm
compares molecules by atom positions only without considering atom connectivity patterns.
Furthermore xLOS penalizes nonꢀoverlapping parts of seed and query molecule, and therefore gives
higher scores to molecules with a relatively small size, which are most suitable for initial activity
2
9, 30
screening since these may later be optimized by the addition of substituents.
In our initial
implementation we used xLOS to optimize weak and unselective inhibitors of the calcium channel
TRPV6 by early scaffold redesign, which enabled the identification of a new scaffold leading to the
33
first potent and selective inhibitor for this target. The method was also recently used for the
identification of a new nonꢀcompetitive inhibitor with an original scaffold for the iron transporter
38
DMT1, for which only very few inhibitors have been described to date.
As detailed below, we used xLOS to search commercial catalogs for shape and
pharmacophore analogs of 16 known kinase inhibitors and assembled a kinase screening set of 437
compounds featuring mostly different scaffolds compared to known kinase inhibitors and kinase
screening compounds. Biochemical screening against Aurora A gave 18 hits featuring 12 different
scaffolds, two of which were pursued further. While attempts to optimize benzimidazole 5 only led
to modest improvements with its hydroxyl analog 6, we succeeded in modifying the phenyliminoꢀ
ACS Paragon Plus Environment

Page 5 of 81
Journal of Medicinal Chemistry
5
1
2
3
4
5
6
7
8
9
thiazolidinone hit 7 and its close analog 8 identified in a second round of LBVS using structureꢀ
based drug design and obtained the potent inhibitor 9. Compound 9 is a type II kinase inhibitor
displaying strong and selective binding to Aurora kinases against the entire kinome. Inhibitor 9
induces a selective Aurora A phenotype in cells without any Aurora B specific effects.
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
Figure 1. Structures of reference kinase inhibitors and new inhibitors discussed in this paper.
ACS Paragon Plus Environment

Journal of Medicinal Chemistry
Page 6 of 81
6
1
2
3
4
5
6
7
8
9
1
1
1
1
1
1
1
1
1
1
2
2
2
2
2
2
2
2
2
2
3
3
3
3
3
3
3
3
3
3
4
4
4
4
4
4
4
4
4
4
5
5
5
5
5
5
5
5
5
5
6
Results and Discussion
Virtual Screening
We assembled a kinase discovery set by searching the ZINC database comprising ~12 million
3
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
compounds for shape and pharmacophore analogs of 16 known inhibitors of Aurora A and other
kinases (Supplementary Figure 1) using the xLOS algorithm described above. We focused our
selection on two vendor catalogs totaling ~1.2 million compounds and purchased 437 compounds
selected by clustering the 1000 top scoring compounds against each of the 16 references, taking into
account scaffold diversity, potential for derivatization, and absence of reactive functional groups
(aldehydes, quinones, chloroketones, imines, etc.). Pan assay interference compounds (PAINS)
40
were however not specifically excluded. Indeed the recently available automated tool at the
4
1
docking.org website flags 27 of the 437 compounds (6.2 %) as PAINS, which is significantly
higher than for DrugBank (67 PAINS from 1775 drugs, 3.8 %) or a random selection of ChEMBL
(86 PAINS from 2000 ChEMBL compounds, 4.3 %). The selected compounds had relatively high
xLOS scores indicating significant shape and pharmacophore similarity to the reference inhibitors.
These compounds also showed relatively low substructure similarity to these reference inhibitors as
measured by the Tanimoto similarity coefficient relative to a 1024ꢀbit Daylight type substructure
fingerprint (TSfp < 0.7), although this similarity was still higher than the average Tanimoto value of
0
.23 for all compounds used in LBVS (Figure 2a). Our screening set was comparable to
commercially available kinase discovery sets in terms of molecular properties (molecular size,
number of Hꢀbond donor and acceptor atoms, number of rotatable bonds). Both of these discovery
sets contained compounds that were on average smaller and with fewer Hꢀbond donor atoms and
rotatable bonds compared to kinase inhibitors reported in ChEMBL, in particular the more potent
ones, reflecting the difference between screening compounds and optimized inhibitors (Figure 2c).
Most importantly, our screening set showed very low overlap with known kinase inhibitors and
discovery sets in terms of molecules and scaffolds, highlighting its potential for identifying
scaffolds not previously used for kinase inhibition (Table 1).
ACS Paragon Plus Environment

Page 7 of 81
Journal of Medicinal Chemistry
7
1
2
3
4
5
6
7
8
9
a
b
xLOS vs Sfp for tested cpds
xLOS vs Sfp for hit cpds
Average Tanimoto
xLOS vs Sfp for tested cpds
xLOS vs Sfp for hit cpds
Average Tanimoto
1
0
0
0
.00
.80
.60
.40
350000
1.00
0.80
0.60
0.40
0.20
0.00
300000
50000
200000
xLOS score histogram
300000
250000
200000
xLOS score histogram
2
1
1
5
0
50000
00000
0000
150000
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
1
5
0
00000
0.20
0000
0
.00
0
.00
0.50
1.00
xLOS
1.50
2.00
2.50
0.00
0.50
1.00
1.50
xLOS
2.00
2.50
c
1
1
1
4%
2%
0%
35%
30%
25%
20%
15%
10%
5%
70%
60%
50%
40%
30%
20%
10%
0%
25%
Kinase ChEMBL
Screening Com.
Kinase < 50nM
Screening xLOS
20%
15%
8
6
4
2
0
%
%
%
%
%
1
0%
5%
0%
0%
1
0
15 20 25 30 35 40 45 50
0
1
2
3
4
5
6
7
8
9
10
0
1
2
3
4
5
6
0
2
4
6
8
10 12 14
HAC
HBA
HBD
RBC
Figure 2. Kinase library design and screening. a. Combined frequency histogram as a function of
xLOS score (black line, ~1.2 M commercial cpds from two catalogs, right vertical axis), and scatter
plot of substructure similarity (TSfp: Tanimoto coefficient of a 1024ꢀbit Daylight type substructure
fingerprint) versus xLOS score for the selected 437 screening compounds (black dots) and active
Aurora A hits (red dots) against their most similar reference inhibitor. b. Combined frequency
nd
histogram and similarity scatter plot as a for 212 compounds selected in the 2 round of virtual
st
screening compared to their reference 1 round hit. c. Property profile of kinase screening sets and
inhibitors in terms of heavy atom count (HAC), hydrogen bond acceptor atoms (HBA), hydrogen
bond donor atoms (HBD), and rotatable bond count (RBC). See Table 1 for details of the
composition of the sets.
ACS Paragon Plus Environment

Journal of Medicinal Chemistry
Page 8 of 81
8
1
2
3
4
5
6
7
8
9
1
1
1
1
1
1
1
1
1
1
2
2
2
2
2
2
2
2
2
2
3
3
3
3
3
3
3
3
3
3
4
4
4
4
4
4
4
4
4
4
5
5
5
5
5
5
5
5
5
5
6
a)
Table 1. Compound and scaffold overlap between different kinase inhibitor and screening sets.
Kinase
Screening
Kinase
Aurora
50nM
Screening
xLOS
Hits
xLOS
Shared compounds
Unique
ChEMBL Commercial 50nM
Kinase ChEMBL
51643
139
11286
329
0
3
0
40222
Screening Commercial 139
23979
7
9
0
23831
Kinase 50nM
11286
7
0
9
0
11286
329
329
0
0
0
0
Aurora 50nM
329
3
329
0
0
0
0
Screening xLOS
Hits xLOS
437
15
15
24
410
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
0
0
0
Shared Scaffolds
Kinase ChEMBL
23400
836
13447
61
4874
61
174
1
66
58
9
4
17719
12584
0
Screening Commercial 836
5
Kinase 50nM
Aurora 50nM
Screening xLOS
Hits xLOS
4874
174
66
4874
174
9
174
174
1
1
1
1
0
0
58
247
15
15
18
145
b)
4
5
1
0
3 (12)
a)
Kinase ChEMBL: all kinase inhibitors found in ChEMBL; Screening commercial: aggregated series of commercially available screening
compounds for kinases; Kinase 50 nM: ChEMBL kinase inhibitors reported with IC50 < 50 nM; Aurora 50 nM: ChEMBL Aurora inhibitors with IC50
50 nM; Screening xLOS: screening set in the present study. Scaffolds are defined according to Bemis and Murcko. Scaffolds unique to the hits not
b)
<
considering overlap with the xLOS screening set.
Hit identification
42
We conducted a timeꢀresolved FRET assay to test the impact of our kinase screening set on
Aurora A activity and found that 15 compounds inhibited Aurora A by more than 50 % at 10 µM
(Supplementary Figure 2), corresponding to a 3.4 % hit rate comparable to other LBVS guided
43
screens. None of these hits were flagged as PAINS. An additional 205 analogs of these 15 hits
were selected by further similarity searching using xLOS, which featured mostly very close analogs
of the hits displaying both high xLOS and substructure similarity scores to these hits (Figure 2b). In
this case a subsequent check at docking.org flagged six compounds as PAINS (see supporting
nd
information), none of which appeared as hits. Biochemical testing of these 2 round LBVS
compounds provided an additional 9 hits with comparable inhibition (Supplementary Figure 3). The
compendium of 24 hits featured 18 scaffolds, 12 of which did not appear among commercial kinase
screening compounds or known kinase inhibitors. Two of these hits were investigated closer, which
were the benzimidazole 5 (IC50 = 6.5 µM), with a xLOS score of 1.1 relative to the known CDK
4
4
and Aurora A/B inhibitor 3 (JNJꢀ7706621), and the Nꢀsubstituted thiazolidinone 7 (IC50 = 2.0
45
µM), with a xLOS score of 1.01 relative to the known CDK inhibitor 4 (ATꢀ7519) (Figure 1). Hit
ACS Paragon Plus Environment

Page 9 of 81
Journal of Medicinal Chemistry
9
1
2
3
4
5
6
7
8
9
7
was confirmed in the second round of virtual screening by the identification of a more potent
analog 8 (IC50 = 0.39 µM). Both series represented scaffolds previously unknown among kinase
inhibitors. Indeed only a single derivative of 5 and no bioactivity have been reported for this
46
scaffold, whereas the closest relative of 5 is the natural product indirubin (10), which is known to
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
inhibit cyclin dependent kinases (CDKs), but with which 5 only shares the cyclic skeleton (scaffold
47, 48
not considering heteroatoms and double bonds).
Furthermore, although a variety of bioactive
49ꢀ52
compounds with Nꢀsubstituted thiazolidinones are known,
only the Nꢀunsubstituted
thiazolidinone scaffold has been used in relationship with kinases in the case of compound 12 and
53, 54
analogs developed as PIM kinase and CDK2 inhibitors.
Note that hit 7, 8, our optimized
49ꢀ52
inhibitor 9, the known inhibitor 12 as well as the various reported bioactive thiazolidinones
are
not flagged by the PAINS tool at docking.org, and that their thiazolidinone core is allowed in the
41
ZINC clean set. In the present context the steep structureꢀactivity relationship profile observed
during the optimization of hit 7 and the relatively low glutathione reactivity observed for 9 (see
below Figure 8) confirm that this structural type, although listed as “rhodanineꢀlike” PAINS in the
4
0
original publication, does not behave as a typical PAINS.
Synthesis
Benzimidazole hit 5 was resynthesized from 1,2ꢀdiaminobenzene by condensation with carbonyl
diimidazole to form the corresponding cyclic urea 13, reaction with phosphorus oxychloride to
form chlorobenzimidazole 19, coupling with monoꢀboc protected 1,2ꢀdiaminobenzene to
intermediate 25 and finally cyclization of the Boc group under basic conditions to form the cyclic
urea group. This synthetic route was used to prepare 5 and its analogs 6 and 33-42 (Scheme 1).
Coupling of the monoꢀboc protected 4ꢀmethylꢀ1,2ꢀdiaminobenzene 43 with chlorobenzimidazole 52
and cyclization gave benzimidazole 44 as an isomeric mixture. Oxidation of the methyl groups
afforded the carboxylic acids 45 and 46 as a separable pair of isomers, which were subsequently
converted to the corresponding Nꢀmethyl amide 47 and 48. Finally syntheses involving Nꢀmethyl
ACS Paragon Plus Environment

Journal of Medicinal Chemistry
Page 10 of 81
1
0
1
2
3
4
5
6
7
8
9
1
1
1
1
1
1
1
1
1
1
2
2
2
2
2
2
2
2
2
2
3
3
3
3
3
3
3
3
3
3
4
4
4
4
4
4
4
4
4
4
5
5
5
5
5
5
5
5
5
5
6
chlorobenzimidazole 49 on the one hand and boc protected Nꢀmethylꢀ1,2ꢀdiaminobenzene 52 on the
other hand afforded the selectively Nꢀmethylated analogs 51 and 53 (Scheme 2).
Analogs of phenyliminoꢀthiazolidinone hit 7 were prepared by addition of various anilines
to alkyl isothiocyanates, followed by regioselective cyclization with ethyl chloroacetate to form the
thiazolidinone nucleus, and finally Knoevenagel condensation with aromatic aldehydes (Scheme
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
5
5
3
). The 2ꢀaminopyridine analog 79, prepared by condensation of intermediate 54 with Bocꢀ
protected 2ꢀaminoꢀ4ꢀformylꢀpyridine followed by acidic deprotection (see supporting information),
was further elaborated by coupling to carboxylic acids by amide bond formation to analogs 80-83,
or to aromatic bromides by palladium coupling to analogs 84-88 and 9 (Scheme 4). Alternatively,
the exocyclic double bond of 79 was either cyclopropanated to yield diastereoisomeric
cyclopropanes racꢀ89/racꢀ90 or reduced by hydrogenation to yield racꢀ91, which was subsequently
extended by coupling to 6ꢀbromoꢀnicotinic acid to yield racꢀ92.
ArNH2, KH2PO4,
n-BuOH, 90 °C
H
N
POCl3, 100 °C
H
N
6
5
6
5
O
Cl
N
H
N
4
R
4
R
R
H
R
19
H
1
3
4
1
4-Me
20 4-Me
21 5-Me
22 5-Br
15 5-Me
1
1
1
6
7
8
5-Br
23 5-CF3
24 5,6-Cl2
5-CF3
5,6-Cl2
X = Boc: K2CO3,
DMF
110 °C
O
H
H
N
6
N
6
NH
NH
NHX
5
N
N
5
4
N
4
R
X = H: CDI,
DMF, 25°C
6
R
4
7
5R'
R'
6
4
5
X
R
H
R'
R (R' = H)
R' (R = H)
7-OH 38
2
5
Boc
H
H
H
H
H
H
H
5
2
2
2
2
3
3
3
6
7
8
9
0
1
2
Boc 4-Me
Boc 5-Me
Boc 5-Br
4-Me 33
5-Me 34
6-Cl
39
6-OH 40
5-Br
35
5-Cl
41
6
-CF3 36
,6-Cl2 37
5-OH
Boc 5-CF3
Boc 5,6-Cl2
5
5
4-OH 42
H
H
H
4-Cl
5-Cl
Boc
Scheme 1. Synthesis of benzimidazole hit 5 and derivatives.
ACS Paragon Plus Environment

Page 11 of 81
Journal of Medicinal Chemistry
1
1
1
2
3
4
5
6
7
8
9
1
) 19, KH2PO4,
n-BuOH, 90 °C
2) K2CO3, DMF,
10°C
O
H
N
NHBoc
NH2
1
) CrO3
1
NH
Me
2) H5IO6
N
N
Me
4
3
44
O
O
H
H
N
N
NH
NH
N
N
+
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
N
N
COX
XOC
5 (X = OH)
4
4
46 (X = OH)
48 (X = NHMe)
MeNH2, BOP
7 (X = NHMe)
BocHN
ArNH2, KH2PO4,
K2CO3, DMF,
110 °C
Cl
N
Me
N
O
n-BuOH, 90 °C
Me
HN
N
NH
Me
N
N
N
N
4
9
50
51
O
NMeBoc
1
9, KH PO , n-Butanol
H
2
4
Me
N
H2N
N
9
0 °C
N
N
5
2
53
Scheme 2. Synthesis of further analogs of benzimidazole 5.
ACS Paragon Plus Environment

Journal of Medicinal Chemistry
Page 12 of 81
1
2
1
2
3
4
5
6
7
8
9
1
1
1
1
1
1
1
1
1
1
2
2
2
2
2
2
2
2
2
2
3
3
3
3
3
3
3
3
3
3
4
4
4
4
4
4
4
4
4
4
5
5
5
5
5
5
5
5
5
5
6
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
Scheme 3. Synthesis of phenyliminoꢀthiazolidinones.
ACS Paragon Plus Environment

Page 13 of 81
Journal of Medicinal Chemistry
1
3
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
Scheme 4. Synthesis of further thiazolidinone analogs.
ACS Paragon Plus Environment

Journal of Medicinal Chemistry
Page 14 of 81
1
4
1
2
3
4
5
6
7
8
9
1
1
1
1
1
1
1
1
1
1
2
2
2
2
2
2
2
2
2
2
3
3
3
3
3
3
3
3
3
3
4
4
4
4
4
4
4
4
4
4
5
5
5
5
5
5
5
5
5
5
6
The indirubin-like benzimidazole 5 is a type I inhibitor
The benzimidazole derivative 5 was investigated first. Analogs 6, 33-42, 45-48, 51 and 52 provided
an initial structureꢀactivity relationship profile. Many of these derivatives lost potency or only
provided modest improvements compared to the initial hit compound. In particular Nꢀmethylation
of the benzimidazole nitrogen atoms resulted in the inactive analogs 51 and 52, suggesting that
these hydrogen bond donor atoms were essential for activity and might therefore interact with the
hinge region of the kinase. Furthermore a 3ꢀfold gain in activity was observed upon introduction of
a hydroxyl group at position 5 with 6 (IC50 = 2.3 µM, Figure 3a). The 5ꢀcarboxy analog 46 was
almost as active, but activity returned to lower levels with the corresponding Nꢀmethyl carboxamide
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
4
8.
To understand the inhibition mechanism of the best inhibitor 6 at the atomic level, we solved
the crystal structure of 6 bound to Aurora A, revealing a typical Type I mode of inhibition including
the activation loop in the active, DFGꢀin conformation (PDB code 4ZS0, 3.0ꢀ3.1 Å resolution,
56, 57
Supplementary Table 1 and Figure 3b).
Compound 6 engaged in two hydrogen bonds between
its benzimidazole and Ala213 in the hinge region of Aurora A, as suggested by the loss of activity
observed upon Nꢀmethylation of the benzimidazole with analogs 51 and 52. These interactions were
analogous to those between the related 6ꢀbromoindirubin (11) and its target CDK2 (PDB code
58
2
BHE). Moreover, 6 made additional hydrophobic contacts with Val147, Leu139, Leu194 and
Leu263, and a hydrogen bond between its 6ꢀhydroxyl group and Asp274 presumably explaining its
stronger binding compared to 5 (Figure 3c). Although no breakthrough derivative (IC50 < 100 nM)
could be identified in this series, the small size and high ligand efficiency of 6 (LE = 0.39) suggest
that its rather original scaffold might be amenable to further optimization for Aurora A or other
kinases.
ACS Paragon Plus Environment

Page 15 of 81
Journal of Medicinal Chemistry
1
5
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
Figure 3. The indirubinꢀlike scaffold 1. (a) Structureꢀactivity relationship profile. Aurora A
inhibition was measured using an HTRF assay (see methods). (b) Crystal structure of the Aurora A
ATPꢀbinding site with bound 6 (PDB 4ZS0, Supplementary Table S1), with highlighted salt bridge
between residues Lys162 and Glu181 (blue), DFG motif (pink) and HRD motif (yellow)
characteristic of an active kinase conformation. (c) Schematic of contact residues between 6 and
Aurora A. Dashed lines indicate hydrogen bonds and red arcs represent hydrophobic interactions.
ACS Paragon Plus Environment

Journal of Medicinal Chemistry
Page 16 of 81
1
6
1
2
3
4
5
6
7
8
9
1
1
1
1
1
1
1
1
1
1
2
2
2
2
2
2
2
2
2
2
3
3
3
3
3
3
3
3
3
3
4
4
4
4
4
4
4
4
4
4
5
5
5
5
5
5
5
5
5
5
6
The phenylimino-thiazolidinone 7 can be optimized to the potent and selective inhibitor 9
The phenyliminoꢀthiazolidinones 7 was investigated next. Initial SAR profiling indicated potential
for optimization upon variation of the Nꢀphenyl substituents with 8 and pinpointed to the essential
role of the 4ꢀpyridine ring since several analogs of 7 with alkyl, halogen, hydroxy or methoxy
substituted phenyl rings or a 3ꢀpyridyl ring at that position were inactive (data not shown).
Additional analogs were investigated to test if the initial gain in activity with 8 could be improved
further (Figure 4a). Activity profiling confirmed the optimal 4ꢀethylphenylimino substituent of 8
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
(63-70 and 93) and the need for a small alkyl substituent on the endocyclic nitrogen atom of the
thiazolidinone (75-78). Furthermore, crystal structures of 76 and 78 established the
Zꢀ
stereochemistry of both exocyclic double bonds of the thiazolidinone core (Supplementary Figure
4).
A further crystal structure of 77 in complex with Aurora A (see below) suggested that
59
placing an amino group at the 2ꢀpyridyl position substituted with acyl or aromatic groups might
enhance hingeꢀbinding interactions. Significant potency gains were indeed obtained by this
approach, reaching low nanomolar values with 81-83 and 87, 9 and 88 displaying substituents with
a carboxylate or tetrazole group presumably interacting with Arg137 (Figure 4b).
These inhibitors were remarkably selective for Aurora A against Aurora B used in complex
60
with its physiological activator INCENP (Figure 4c). The most potent inhibitor 9 (IC50 = 2 nM)
was selected for further in depth analysis because this compound was also the most potent when
tested on cells (see below). A kinome scan of 456 kinases using an active site directed binding
6
1
competition assay followed by determination of binding affinities showed that 9 was remarkably
selective and bound tightly only to Aurora A ( = 5 nM), Aurora B (without INCENP, = 10
= 11 nM) (Figure 4d). Note that the biochemical inhibition and binding
K
D
K
D
6
2
nM), and Aurora C (K
D
affinity measurements were both carried out with free Aurora A, which is autophosphorylated at
T288, and gave comparable values. By contrast and due to the inactivity of the kinase alone the
ACS Paragon Plus Environment

Page 17 of 81
Journal of Medicinal Chemistry
1
7
1
2
3
4
5
6
7
8
9
biochemical inhibition of Aurora B was measured for its complex with its activator protein
INCENP, to which it is always bound in cells, giving a 15ꢀfold weaker value for inhibition of
catalysis compared to the K value for binding measured with free Aurora B.
D
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
Figure 4. Optimization of the phenyliminoꢀthiazolidinone hit 7 to the selective Aurora A inhibitor
1
2
9. (a) Structure and activities of analogs with variations of R and R . (b) Optimization of the 2ꢀ
pyridyl substituent. SI = selectivity index: IC50(Aurora B/INCENP)/IC50(Aurora A). (c) Structure
and IC curves for Aurora A and Aurora B/INCENP of the most potent inhibitor 9. (d) Kinome
50
profiling and K
D
values for 9 towards its 9 best target kinases.
ACS Paragon Plus Environment

Journal of Medicinal Chemistry
Page 18 of 81
1
8
1
2
3
4
5
6
7
8
9
1
1
1
1
1
1
1
1
1
1
2
2
2
2
2
2
2
2
2
2
3
3
3
3
3
3
3
3
3
3
4
4
4
4
4
4
4
4
4
4
5
5
5
5
5
5
5
5
5
5
6
Binding of 9 to Aurora A excludes the activator protein TPX2
To understand the inhibition mechanism at the atomic level, we solved the crystal structures of
Aurora A in complex with 77 from the initial SAR study, with the highest affinity inhibitor 9, and
with its phenyl tetrazole analog 88 (PDB codes 4ZTQ, 4ZTR and 4ZTS, resolution 2.8ꢀ2.9Å,
Supplementary Table 1, Figure 5). All three inhibitors bind to the ATP pocket in the adenine
binding region (Leu139, Val147, Ala160, Leu263) and occupy the hydrophobic back pocket
(Leu194, Arg195, Leu196, Leu210, Phe275). The inhibitors induce an inactive DFG motif
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
conformation also found in other Aurora A inhibitor complexes (PDB codes 4JBQ, 4JAI, 2J5O,
56, 57
2
BMC, 3P9J, 3R22, 3FDN, 3K5U, and 4BOG) with four characteristic features:
(1) disruption
of the Rꢀspine (residues Leu196, Gln185, Phe275, His254, and Asp311), (2) an outward
displacement of the αC helix compared to the active state (approximately 2.5 Å), (3) absence of the
salt bridge between Lys162 and Glu181, and (4) disruption of the Asp256–Thr292–Lys258
hydrogen bond network. In this conformation Asp274 is pointing away from the ATP pocket and
Phe275 is rotated upwards thereby contacting the phenylimino moiety of the inhibitors.
Furthermore, Trp277 (or Arg255 of the HRD motif in the case of 9) forms Hꢀbonds with Gln185
6
3,
and Asp274, an arrangement clearly different from classical DFGꢀin or DFGꢀout conformations,
64
65, 66
and which is specific of Aurora kinases, thus probably contributing to inhibitor selectivity.
The pyridine/aminopyridine group of the inhibitors engages in one or two hydrogen bonds with
Ala213 in the hinge region. Finally, the terminal carboxylate of 9 forms a salt bridge with Arg137
6
7
and Arg220, analogous to 1 in the structure bound to Aurora A (PDB code 2X81) , thus explaining
its stronger affinity compared to 77.
Tryptophan fluorescence experiments showed that binding of 9 and of its analogs 77 and 88
perturbed the environment of Trp277, reflecting the rearrangement of the DFG loop and Trp277
interactions with Gln185 occurring upon binding of these inhibitors, as suggested by inspection of
the crystal structures. By contrast the type I inhibitor 6, which does not induce rearrangement of the
DFG loop, had no effect on tryptophan fluorescence (Supplementary Figure 5). Further analysis of
ACS Paragon Plus Environment

Page 19 of 81
Journal of Medicinal Chemistry
1
9
1
2
3
4
5
6
7
8
9
these crystal structures indicated that this rearrangement should be incompatible with the rotation of
the αCꢀhelix and the subsequent salt bridge formation between Lys162 and Glu181 that take place
upon binding to the microtubuleꢀassociated activator protein TPX2 (PDB code 1OL5 or 4C3P,
68, 69
Supplementary Figure 6).
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
1
ꢀ43
To test this hypothesis, we measured the binding of Aurora A to labeled TPX2 using
microscale thermophoresis (K ~800 nM). As reported in Figure 6a, we found that this binding was
D
indeed abolished in the presence of excess 9. The incompatibility between 9 and TPX2 was further
evidenced by the fact that TPX2 binding, which induced a 3ꢀfold increase in the activity of Aurora
70
A similar to other reports (Figure 6b), reduced inhibition by 9 from IC50 = 2.0 nM to IC50 = 1.4
µM (Figure 6c). In control human cells, TPX2 helps in the localization of Aurora A to the spindle
70, 71
microtubules, but not to the centrosomes.
Our findings raised the possibility that addition of 9
7
1
should prevent localization from taking place strictly on the spindle microtubules. Indeed, we
found that Aurora A remained localized at centrosomes in the presence of 9 but was notably
72
displaced from the spindle microtubules, as is the case for 1 (Figure 6d).
ACS Paragon Plus Environment

Journal of Medicinal Chemistry
Page 20 of 81
2
0
1
2
3
4
5
6
7
8
9
1
1
1
1
1
1
1
1
1
1
2
2
2
2
2
2
2
2
2
2
3
3
3
3
3
3
3
3
3
3
4
4
4
4
4
4
4
4
4
4
5
5
5
5
5
5
5
5
5
5
6
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
Figure 5. Crystal structures of the 4ꢀthiazolidinone derivatives bound to Aurora A. The 2F ꢀF
o
c
electron density map contoured at 1 sigma around the compounds is shown to highlight their
unambiguous position in the pocket. Lys162 and Glu181 (blue), DFG motif (pink) and HRD motif
yellow). The catalytic pocket of Aurora A acquires an inactive conformation upon binding to 77
(
(a), 9 (b) and 88 (c). The lower panels show schematic representations of the interactions between
the compounds and the protein. Dashed lines indicate hydrogen bonds, in particular interactions
between the aminopyridine portion of the inhibitors and the hinge region (A213) and between the
anionic carboxyl or tetrazole group and the guanidinium groups of R137 and R220. Residues
involved in hydrophobic interactions are highlighted with a red arc.
ACS Paragon Plus Environment

Page 21 of 81
Journal of Medicinal Chemistry
2
1
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
Figure 6. Binding of Aurora A to 9 excludes TPX2. (a) Microscale thermophoresis (MST) analysis
of TPX2 binding to Aurora A kinase in presence and absence of compound 9. Three independent
measurements were performed; error bars represent the standard deviation (SD). (b) Aurora A
kinase activity in the presence of TPX2 or control protein BSA. (c) IC curves for 9 in the presence
5
0
of different TPX2 concentrations. (d) 9 impairs localization of Aurora A at the mitotic spindle.
Representative images of HeLa Kyoto cells treated with compound 9 overnight and stained for
DNA (blue), Aurora A (green) and TPX2 (red). TPX2 coꢀlocalizes with Aurora A in control cells
(top row). 9 impairs Aurora A localization at the spindle microtubules but does not affect
centrosomal Aurora A. Aurora A is also present at centrosomes but not on spindle microtubules in
cells treated with 1 (MLN8237).
ACS Paragon Plus Environment

Journal of Medicinal Chemistry
Page 22 of 81
2
2
1
2
3
4
5
6
7
8
9
1
1
1
1
1
1
1
1
1
1
2
2
2
2
2
2
2
2
2
2
3
3
3
3
3
3
3
3
3
3
4
4
4
4
4
4
4
4
4
4
5
5
5
5
5
5
5
5
5
5
6
Compound 9 selectively inhibits Aurora A in cells
We next set out to test the impact of 9 on mitotic progression of human tissue culture cells.
Treatment of HeLa cells with 9 resulted in a doseꢀdependent increase in the mitotic index, as
observed also with the reference inhibitor 1 and as expected from the role of Aurora A for timely
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
24
progression through mitosis (Figure 7a). Inhibitor 9 also decreased phosphorylation of Aurora A
Thr288, an autophosphorylation site that marks Aurora A kinase activity (IC = 760 nM, Figure
50
73
7b). Furthermore, 9 induced defective chromosome alignment during metaphase, with a
phenotypic EC50 value of 6 µM, in line with the requirement of Aurora A activity for proper spindle
18, 24, 74
dynamics (Figure 7c/d).
Many Aurora A kinase inhibitors also target Aurora B in the cellular context. To address
whether this may be the case also for 9, we examined the distribution of phosphoꢀHistone H3 Ser10,
7
5
a histone modification that is imparted during early mitosis by Aurora B. We found that this
feature remained unchanged in cells treated with 9, whereas it was absent in cells treated with the
76
Aurora B inhibitor ZM447439 (94) (Figure 7f/g). Furthermore the massive accumulation of cells
with 8N and 16N DNA contents observed by flow cytometry upon treatment with 94, which is a
hallmark of defective cytokinesis following Aurora B inactivation, also did not occur with 9 even
76
when provided at 10 µM (Figure 7h). We conclude that in contrast to other Aurora A inhibitors,
23, 77
including the reference inhibitor 1,
our inhibitor 9 had no effect on Aurora B activity in cells.
To unequivocally test whether the effect of 9 on human cells was specific to Aurora A
inhibition we performed phenotypic rescue experiments using cells expressing GFPꢀAurora Aꢀ
Arg137Ala or GFPꢀAurora AꢀTrp277Ala mutants, which were predicted by examination of the
crystal structure to be insensitive to 9. As shown in Figure 7e, we found that these proteins were
insensitive to the addition of 9, demonstrating that the phenotype normally induced by this
65, 67
compound was indeed caused by Aurora A inhibition.
ACS Paragon Plus Environment

Page 23 of 81
Journal of Medicinal Chemistry
2
3
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
Figure 7. Inhibitor 9 yields an Aurora A specific inhibition phenotype in cells without impairing
Aurora B function. (a) Mitotic index of HeLa Kyoto cells after overnight incubation with indicated
compounds. The experiment was repeated twice and error bars represent SD. (b) Centrosomal
pThr288 Aurora A levels in HeLa Kyoto cells after overnight incubation with indicated
concentrations of 9. More than 10 cells were analyzed in every condition. The experiment was
repeated twice and the error bars represent SD. (c) Chromosome misalignment phenotype of HeLa
Kyoto cells incubated with indicated compounds. The experiment was repeated twice and error bars
represent SD. (d) Representative images of HeLa Kyoto cells after overnight incubation with
indicated compounds. The cells were stained for pThr288 Aurora A (red), αꢀtubulin (green), and
DNA (blue). Impaired alignment of chromosomes on the metaphase plate is indicated by white
arrows. (e) Chromosome misalignment phenotype of HeLa Kyoto cells. The cells were transfected
ACS Paragon Plus Environment

Journal of Medicinal Chemistry
Page 24 of 81
2
4
1
2
3
4
5
6
7
8
9
1
1
1
1
1
1
1
1
1
1
2
2
2
2
2
2
2
2
2
2
3
3
3
3
3
3
3
3
3
3
4
4
4
4
4
4
4
4
4
4
5
5
5
5
5
5
5
5
5
5
6
with the indicated GFPꢀAurora A constructs before incubation with 9 overnight. The bars represent
mean phenotype ± SD of two independent experiments. At least 100 cells were analyzed in every
condition. Note that cells expressing the highest amount of GFPꢀAurora A, GFPꢀAurora A
Arg137Ala and GFPꢀAurora A Trp277Ala were considered for the analysis; cells expressing lower
levels of the mutant transgenes exhibited less or no rescue (data not shown). (f) Representative
images of HeLa Kyoto cells after overnight incubation with indicated compounds and staining for
pHisꢀH3 (red) and DNA (blue). (g) Western blot of lysates from cells synchronized in
prometaphase with 100 nM nocodazole and treated with the indicated inhibitors. βꢀactin was used
as loading control. (h) DNA content analysis of cells treated for 48 h with the indicated
compounds. The cells were stained with propidium iodide and analyzed using flow cytometry.
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
Compound 9 shows moderate reactivity with glutathione
Despite of its high selectivity towards Aurora A the cellular activity of 9 (EC50 > 1 µM) was much
weaker than its biochemical activity (IC50 = 2 nM). To understand whether this lower activity might
be caused by covalent reaction of the electrophilic double bond of 9 we measured its reactivity
towards the intracellular nucleophile glutathione (GSH) under physiological conditions. Conversion
to a GSH adduct was indeed detected, however only to a limited extent (50 % conversion after 24 h
with 5 mM GSH, pH 7.4, 37°C, Figure 8), suggesting that a significant portion of the inhibitor
remained unreacted within the cell and was available for inhibition of Aurora A. Nevertheless
analogs of 9 lacking the electrophilic double bond were prepared as described above (Scheme 4)
and investigated as alternatives. Diastereomeric cyclopropanes racꢀ89/racꢀ90 were more than 400ꢀ
fold less active than their precursor 79. On the other hand its reduced double bond analog racꢀ91
was only 7ꢀfold less active, and the reduced double bond analog of 9, racꢀ92, was also still quite
potent (IC50 = 24 nM). However this derivative did not show any activity in cells despite the fact
that it cannot form GSH adducts.
ACS Paragon Plus Environment

Page 25 of 81
Journal of Medicinal Chemistry
2
5
1
2
3
4
5
6
7
8
9
Taken together, these data suggest that GSH reactivity is insufficient to explain the
discrepancy between the IC50 values of cellular versus biochemical activity of 9 on Aurora A and B
presented above (Aurora A: biochemical IC50 = 2.0 nM, cellular IC50 (pT288) = 760 nM, Aurora
B/INCENP: biochemical IC50 = 149 nM, cellular IC50 (pH3) > 25 µM). This activity difference is
probably a consequence of the higher ATP concentration in cells (1 mM) compared to in vitro
assays (20 µM) and the presence of competing ligands such as TPX2 (Figure 6d), as well as the
presence of a carboxylate group which might reduce cellular uptake. The same effects probably
explain the weaker cellular versus biochemical activities reported for inhibitor 1 (Aurora A:
biochemical IC50 = 0.04 nM, cellular IC50 (pT288) = 6.7 nM, Aurora B/INCENP: biochemical IC50
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
=
1.1 nM, cellular IC50 (pH3) = 1.5 µM), as well as for MKꢀ5108, a further selective Aurora A
inhibitor (Aurora A: biochemical IC = 0.064 nM, cellular IC (pT288) = 300 nM, Aurora
50
50
22, 23, 78
B/INCENP: biochemical IC50 = 1.49 nM, cellular IC50 (pH3) > 10 µM).
ACS Paragon Plus Environment

Journal of Medicinal Chemistry
Page 26 of 81
2
6
1
2
3
4
5
6
7
8
9
1
1
1
1
1
1
1
1
1
1
2
2
2
2
2
2
2
2
2
2
3
3
3
3
3
3
3
3
3
3
4
4
4
4
4
4
4
4
4
4
5
5
5
5
5
5
5
5
5
5
6
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
Figure 8. In vitro reactivity of 9 towards glutathione (GSH). Inhibitor 9 (30 µM) was incubated
with GSH (5 mM) in H O/CH CN (4:1) at pH = 7.0, and the reaction was followed by LCꢀMS. a.
2
3
Reaction scheme and structures of modified analogs. b. LC chromatogram of the reaction mixture
after t = 0 h and t = 14 h. The product formed with a retention time t = 1.29 min is identified by
ret
MS as the GSHꢀadduct of 9. 9ꢀmethylcarbazole (t = 2.44 min) was used as internal control. c.
ret
Concentration of compound 9 during the course of the reaction as determined by integration of the
absorption peaks (at 310 nm) of compound 9 at t = 1.78 min.The average of two independent
ret
experiments is shown, SD is < 5 % for all data points. d. Aurora A inhibition in the HTRF
biochemical assay. e. Cellular pThr288 Aurora A levels of cells incubated with 9 or racꢀ92
overnight at the indicated concentrations. The experiment was repeated twice and the error bars
represent SD. Derivative racꢀ92 does not reduce pThr288 levels in cells, although having an IC of
50
2
4 nM in the biochemical assay. Numbers represent average IC ± SD of two independent
50
experiments.
ACS Paragon Plus Environment

Page 27 of 81
Journal of Medicinal Chemistry
2
7
1
2
3
4
5
6
7
8
9
Conclusion
To identify new inhibitors of Aurora A kinase we performed scaffold hopping LBVS using our
recently reported method xLOS, a 3D shape and pharmacophore similarity algorithm, and
assembled a kinase discovery set with very low compound and scaffold overlap with known kinase
inhibitors and screening compounds. We then performed biochemical activity screening against
Aurora A and optimization for two screening hits, benzimidazole 5 and thiazolidinone 7. While
benzimidazole 5 did not yield potent inhibitors, we successfully optimized thiazolidinone 7 by
elaboration of the hingeꢀbinding motif using structureꢀbased drug design and discovered the potent
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
inhibitor 9 with IC50 = 2 nM. This type II inhibitor locks the kinase in an inactive conformation and
competes with the activator protein TPX2. Inhibitor 9 shows strong binding selectivity for Aurora
kinases against the entire kinome, displays a highly specific Aurora A inhibition phenotype without
Aurora B specific effects.
Experimental Section
Virtual screening. Virtual screening was performed using xLOS, an inꢀhouse developed 3Dꢀligand
based virtual screening method. Two rounds of virtual screening were performed. In the first round,
39
the ZINC database containing ~12M commercial compounds was screened using each of the 16
known inhibitors of Aurora A and other kinases as seed molecule. For each seed (one stereoisomer)
and database (all possible stereoisomers), a single lowest energy 3Dꢀmodel was generated using
CORINA program available from Molecular Networks Pvt. Ltd. These 3Dꢀmodels were used by
xLOS to calculate the xLOS similarity score between seed and database molecule as described
33
previously. For database molecules only best scoring stereoisomer was retained in the list. The
search was limited to catalogs from Princeton Molecular Research Inc. (~800,000 cpds) and Otava
Chemicals (~300,000 cpds). For each of the 16 reference molecules the top 1000 scoring
ACS Paragon Plus Environment

Journal of Medicinal Chemistry
Page 28 of 81
2
8
1
2
3
4
5
6
7
8
9
1
1
1
1
1
1
1
1
1
1
2
2
2
2
2
2
2
2
2
2
3
3
3
3
3
3
3
3
3
3
4
4
4
4
4
4
4
4
4
4
5
5
5
5
5
5
5
5
5
5
6
compounds were selected. The resulting 16,000 compounds were further analyzed by clustering and
visual analysis to select structurally diverse compounds. In the second round 15 hits identified in
first round of virtual screening were used in the same manner.
Scaffold analysis. The Kinase ChEMBL was downloaded from the ChEMBL database website
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
(https://www.ebi.ac.uk/chembl/). The screening commercial set was created by combining
commercially available screening compounds for kinases from three different vendors namely
ChemBridge (http://www.chembridge.com/), Life Chemicals (http://www.lifechemicals.com/) and
Otava (http://www.otavachemicals.com/). The Kinase 50 nM and Aurora 50 nM sets were created
by selecting the compounds from ChEMBL database with reported IC50 of ≤50 nM against any
kinases and Aurora kinase targets respectively. Molecules were processed using inꢀhouse written
java program utilizing Java Chemistry library (JChem) from ChemAxon, Pvt. Ltd as starting point.
Molecules were checked for valence error, counter ions were removed and ionization states were
adjusted to pH 7.4. Molecules were stored in nonꢀstereo smiles format and duplicate molecules
were removed from each database. Within each database, each molecule was converted to Bemisꢀ
Murcko Scaffold (BMS) using the StructuralFrameworksPlugin available in the JChem library from
ChemAxon Pvt. Ltd. For each database, the list of unique BMS was generated, followed by
calculation of the number of compounds represented by each BMS. For the shared scaffold analysis
the unique SMILES of scaffolds were used for comparing molecules.
Chemistry. All reagents were purchased from commercial sources and were used without further
purification. Flash chromatography purifications were performed with silica Gel 60 (Fluka, 0.040–
0
.063 nm, 230–400 mesh ASTM). Low resolution mass spectra were obtained by electron spray
ionization (ESIꢀMS) in the positive mode on a Thermo Scientific LCQ Fleet. High resolution mass
spectra were obtained by electron spray ionization (HRꢀESIꢀMS) in the positive mode recorded on a
1
13
Thermo Scientific LTQ Orbitrap XL. H and CꢀNMR spectra were measured on a Bruker Avance
300 spectrometer (at 300 MHz and 75 MHz, respectively) or on a Bruker AVANCE II 400
1
13
spectrometer (at 400 MHz and 101 MHz, respectively). H and C chemical shifts are quoted
ACS Paragon Plus Environment

Page 29 of 81
Journal of Medicinal Chemistry
2
9
1
2
3
4
5
6
7
8
9
relative to solvent signals, and resonance multiplicities are reported as s (singlet), d (doublet), t
(triplet), q (quartet), p (pentet), and m (multiplet); br = broad peak. Compound purities were
assessed by analytical reversed phase HPLC (RPꢀHPLC) at a detection wavelength of 254 nm or
3
2
10 nm. The purity of tested compounds was > 95 %, unless otherwise stated (Supplementary Table
). Analytical RPꢀHPLC was performed on a Dionex Ultimate 3000 RSLC System (DADꢀ3000 RS
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
Photodiode Array Detector) using a Dionex Acclaim RSLC 120 column (C18, 3.0 x 50 mm,
ꢀ1
particle size 2.2 ꢁm, 120 Å pore size) and a flow rate of 1.2 mL min . Data were recorded and
processed with Dionex Chromeleon Management System (version 6.8) and Xcalibur (version 2.2,
Thermo Scientific). Eluents for analytical RPꢀHPLC were as follows: (A) milliQꢀdeionized water
containing 0.05 % TFA, and (D) HPLCꢀgrade acetonitrile/milliQꢀdeionized water (9:1) containing
0.05 % TFA. Conditions for analytical RPꢀHPLC were as follows: From A/D (7:3) to 100 % D (2.2
min) followed by 100 % D (1 min). Preparative RPꢀHPLC was performed with a Waters Prep
LC4000 Chromatography System using a Reprospher 100 column (Dr. Maisch GmbH, C18ꢀDE,
1
00 x 30 mm, particle size 5 µM, pore size 100 Å) and a Waters 489 Tunable Absorbance Detector
operating at 214 nm. Eluents for preparative RPꢀHPLC were as follow: (A) milliQꢀdeionized water
containing 0.1 % TFA, and (D) HPLCꢀgrade acetonitrile/milliQꢀdeionized water (9:1) containing
0
.1 % TFA.
1
'
H
-[1,2'-bibenzo[d]imidazol]-2(3H)-one (5). A solution of compound 25 (150 mg, 0.463 mmol, 1
eq.) and K
2
CO (194 mg, 1.39 mmol, 3 eq.) in DMF (4 mL) was stirred at 110 °C for 14 h. DMF
3
was evaporated in vacuo, and EtOAc (20 mL) and brine (20 mL) were added. The organic phase
was separated and washed with H O (3 x 20 mL). EtOAc was removed in vacuo. The crude was
2
purified by FC (SiO
2
; hexane/EtOAc 1:2) to afford 5 as gray powder (46 mg, 0.19 mmol, 40 %).
1
m.p. > 300°C (dec.). H NMR (300 MHz, DMSOꢀd
6
) δ 11.69 (s, 1H), 8.59 – 8.18 (m, 1H), 7.87 –
13
7.44 (m, 2H), 7.50 – 7.08 (m, 5H); C NMR (75 MHz, DMSOꢀd
6
) δ 152.42, 143.91, 128.56,
+
127.02, 123.41, 121.68, 121.61, 113.41, 109.45; HRꢀESIꢀMS (m/z): calculated for [M + H]
C H ON 251.0927, found 251.0923.
14
11
4
ACS Paragon Plus Environment