Probing the ATP-Binding Pocket of Protein Kinase DYRK1A with Benzothiazole Fragment Molecules

Article abstract of DOI:10.1021/acs.jmedchem.6b01086

DYRK1A has emerged as a potential target for therapies of Alzheimer's disease using small molecules. On the basis of the observation of selective DYRK1A inhibition by firefly d-luciferin, we have explored static and dynamic structural properties of fragment sized variants of the benzothiazole scaffold with respect to DYRK1A using X-ray crystallography and NMR techniques. The compounds have excellent ligand efficiencies and show a remarkable diversity of binding modes in dynamic equilibrium. Binding geometries are determined in part by interactions often considered "weak", including "orthogonal multipolar" types represented by, for example, F-CO, sulfur-aromatic, and halogen-aromatic interactions, together with hydrogen bonds that are modulated by variation of electron withdrawing groups. These studies show how the benzothiazole scaffold is highly promising for the development of therapeutic DYRK1A inhibitors. In addition, the subtleties of the binding interactions, including dynamics, show how full structural studies are required to fully interpret the essential physical determinants of binding.

Full text of DOI:10.1021/acs.jmedchem.6b01086

Subscriber access provided by CORNELL UNIVERSITY LIBRARY
Article
Probing the ATP-binding pocket of protein kinase
DYRK1A with benzothiazole fragment molecules
Ulli Rothweiler, Wenche Stensen, Bjorn-Olav Brandsdal, Johan Isaksson,
Frederick Alan Leeson, Richard Alan Engh, and John S. Svendsen
J. Med. Chem., Just Accepted Manuscript • DOI: 10.1021/acs.jmedchem.6b01086 • Publication Date (Web): 13 Oct 2016
Downloaded from http://pubs.acs.org on October 14, 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 41
Journal of Medicinal Chemistry
1
2
3
4
5
6
7
8
9
Probing the ATPꢀbinding pocket of protein kinase
DYRK1A with benzothiazole fragment molecules
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
Ulli Rothweiler1,2, Wenche Stensen2,4, Bjørn Olav Brandsdal3, Johan Isaksson4, Frederick
Alan Leeson2,4, Richard Alan Engh1, John S. Mjøen Svendsen2,4*
1The Norwegian Structural Biology Centre, Department of Chemistry, UiT The Arctic
University of Norway, Nꢀ9037 Tromsø, Norway,
2Pharmasum Therapeutics AS, Smørblomstvn. 36, Nꢀ9102 Kvaløysletta, Norway.
3The Centre of Theoretical and Computational Chemistry, Department of Chemistry, UiT The
Arctic University of Norway, Nꢀ9037 Tromsø, Norway,
4Department of Chemistry, UiT The Arctic University of Norway, Nꢀ9037 Tromsø, Norway.
AUTHOR INFORMATION
*Corresponding author:
johnꢀsigurd.svendsen@uit.no
Keywords: Alzheimer’s disease, DYRK1A, protein kinase, modeling, NMR, Xꢀray
crystallography, inhibitor design, structure guided drug design
ACS Paragon Plus Environment

Journal of Medicinal Chemistry
Page 2 of 41
1
2
3
4
ABSTRACT
5
6
7
8
9
DYRK1A has emerged as a potential target for therapies of Alzheimer’s disease using small
molecules. Based on the observation of selective DYRK1A inhibition by firefly Dꢀluciferin,
we have explored static and dynamic structural properties of fragment sized variants of the
benzothiazole scaffold with respect to DYRK1A using Xꢀray crystallography and NMR
techniques. The compounds have excellent ligand efficiencies and show a remarkable
diversity of binding modes in dynamic equilibrium. Binding geometries are determined in
part by interactions often considered “weak”, including “orthogonal multipolar” types
represented by e.g. FꢀCO, sulfurꢀaromat, and halogenꢀaromat interactions, together with
hydrogen bonds that are modulated by variation of electron withdrawing groups. These
studies show how the benzothiazole scaffold is highly promising for the development of
therapeutic DYRK1A inhibitors. In addition, the subtleties of the binding interactions,
including dynamics, show how full structural studies are required to fully interpret the
essential physical determinants of binding.
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
.
ACS Paragon Plus Environment

Page 3 of 41
Journal of Medicinal Chemistry
1
2
3
4
INTRODUCTION
5
6
7
8
9
Never in the history of man has the world’s population been older. Advanced age has
traditionally been regarded as a blessing, but a less welcome correlate of aging is dementia:
rates of dementia rise sharply with age, with a prevalence of <5 % for populations younger
than 65, and 15 – 40 % for those over 85.1 Three out of four dementia patients have the
dreaded Alzheimer’s disease (AD),1 first described in 1906 by the Bavarian neuropathologist
Alois Alzheimer.2 first described in 1906 by the Bavarian neuropathologist Alois Alzheimer.
In 1992 Hardy and Higgins presented the amyloid cascade hypothesis3 to explain the
molecular pathology of AD. The hypothesis posits that deposition of amyloidꢀβ (Aβ) after
secretase proteolysis of the amyloid precursor protein (APP) in the brain forms toxic insoluble
fibrils (Aβ−plaques) that contribute to neuronal death and dementia. More recently, additional
focus has been directed towards the role of the protein tau in AD.4ꢀ10 Tau is a highly soluble
protein that binds to and stabilizes microtubules in the axon of the nerve cells. In diseases
where tau is involved in the pathology, hyperphosphorylation of tau prevents binding to
microtubules, leading to inhibited axonal flow and thereby directly to neurodegeneration.11,12
Hyperphosphorylated tau also selfꢀassembles into aggregates that ultimately form the
neurofibrillary tangles (NFTs)13 initially observed by Alois Alzheimer himself as one of the
histological hallmarks of AD.
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
The phosphorylation of protein tau in AD is a complex process whereby protein kinases
such as GSK3β and CDK5/p25 play important roles. Recently it has been established that
another protein kinase, DYRK1A, may play an even more central function, by orchestrating a
wide variety of events that cause Alzheimer’s disease.14ꢀ16 DYRK1A can phosphorylate tau
directly at the Thr212 site;17 a process that inhibit the ability of tau to promote microtubule
assembly, but also promotes additional tau phosphorylation by the protein kinase GSK3β,4,18ꢀ
20
which in turn increases tau hyperphosphorylation and subsequent loss of microtubule
ACS Paragon Plus Environment

Journal of Medicinal Chemistry
Page 4 of 41
1
2
3
4
5
6
7
8
9
stabilization further. In addition, DYRK1A phosphorylates the "alternate splicing factor"
(ASF), leading to an increased amount of 3R tau relative to 4R tau, further facilitating tau
aggregation.16 DYRK1A also phosphorylates the "regulator of calcineurin" (RCAN), downꢀ
regulating calcineurin (Caln) phosphatase activity, thereby increasing levels of
phosphorylated tau and reducing Caln/NFAT transcriptional activity.21,22
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
DYRK1A also plays important roles in amyloid plaque formation. DYRK1A increases
secretaseꢀmediated cleavage of APP into A peptides both via APP phosphorylation
directly,19,23 and by phosphorylation of presenelin 1 (PSEN1),24 which even further increases
secretase activity. Moreover, there is evidence that Aβ peptides stimulate DYRK1A
expression in a positive feedback loop.25
The bulk of recent Alzheimer research has not only expanded the disease etiology from
betaꢀamyloid plaques into tau and NFT’s, but has also identified new potential targets for the
development of therapeutic drugs, including the protein kinase DYRK1A.26ꢀ28 There are
however significant gaps in our knowledge of essential DYRK1A properties, including details
of structure, plasticity, inhibitor binding potential and also critically biological consequences
of activity modulation, before DYRK1A can be finally validated as a promising target for
treatment of AD.
A significant proportion of the current literature has been concerned with identifying
ligands that may inhibit DYRK1A, and a recent review summarizes the different chemotypes,
all comprising Type 1 ligands that target the ATPꢀbinding site.15 We have recently discovered
that luciferin and derivatives can act as selective DYRK1A inhibitors.29 Little work however
has surfaced that systematically explore the ATPꢀbinding site of DYRK1A using a library of
ligands. In the present study, a library of small benzothiazole fragment molecules was
designed as ligands to systematically explore the repertoire of binding interactions between
the ligand and the active site of the DYRK1A enzyme. In particular, hydrogen bonding
ACS Paragon Plus Environment

Page 5 of 41
Journal of Medicinal Chemistry
1
2
3
4
5
6
7
8
9
interaction and electronic effects are studied. The binding interactions between the ligands
and the protein kinase were studied using a variety of biological, thermodynamic, structural
and computational methods. The effectiveness of the ligands was measured in inhibition
assays. Structural studies were performed by a combination of Xꢀray crystallography,
dynamic NMR and computational simulations. Put together, these studies provide important
new insights in potential interaction sites and design rules for de novo DYRK1A inhibitor
construction.
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
ACS Paragon Plus Environment

Journal of Medicinal Chemistry
Page 6 of 41
1
2
3
4
RESULTS
5
6
7
Focused benzothiazole fragment library design
8
9
DꢀLuciferin and derivatives have recently been described as inhibitors of several protein
kinases, particularly within the CMGCꢀfamily.29 The inhibitory activity of the luciferin
derivatives is in particular sensitive to modifications of the 6ꢀhydroxy group on the
benzothiazole system of luciferin, where Oꢀmethylation greatly diminishes inhibitory activity.
Alterations in the dihydrothiazole ring of luciferin affects the inhibition of DYRKꢀfamily
protein kinases only minimally. In the design of the current library it was decided to reduce
the complexity of the luciferin ring system by replacing the dihydrothiazole moiety with a
simple acetamide group, and greatly increase the repertoire of substituents in the
benzothiazole ring, thus creating a focused fragment library design to explore the active site
of DYRK kinases.
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
Choice of benzothiazole substituents
The substituents were selected according to their ability to donate or accept hydrogen bonds
(or both) and their ability to affect the πꢀelectron system in the aromatic benzothiazole ring
system. All substituents chosen, with the exception of chloride and bromide, have hydrogen
bond accepting properties varying from strong to weak. The hydroxyl group is the premier
hydrogen bond donor (and/or acceptor) substituent, while the carboxamide moiety also has
dual hydrogenꢀbonding capacities. The substituents have also been chosen to represent
diversity in electronic effects spanning from activating to deactivating groups with
predominantly mesomeric or inductive effects.
Geometry
The inhibitors were designed with the substituents placed in the 5ꢀ or 6ꢀposition (“end”ꢀ
positions) of the benzothiazole nucleus, however, two fragments with either an hydroxyꢀ or a
ACS Paragon Plus Environment

Page 7 of 41
Journal of Medicinal Chemistry
1
2
3
4
5
6
7
8
9
methoxy substituent in the 4ꢀposition (“edgeꢀposition”) were also prepared. From the study on
luciferin derived inhibitors29 it was observed that the 4ꢀhydroxy substituted analog occupied a
region different from luciferin itself, so for the present study “end”ꢀtype were favored over
“edge” type substitutions in order to address the same region of the protein kinase as luciferin.
The 5ꢀ and 6ꢀsubstituents are geometrically related through a pseudo 2ꢀfold rotational axis
that exchanges the positions of the sulfur and the nitrogen atoms in the thiazole ring (Figure
1). Besides this geometrical difference, properties that differ between equivalent 5ꢀ or 6ꢀ
substituted molecules will also reflect differences of mesomeric or inductive effects.
Substitution at the 5ꢀposition can influence the sulphur atom through mesomeric effects and
the nitrogen atom through inductive effects, and vice versa for substitutions at the 6ꢀposition.
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. Design considerations for the benzothiazole library
Preparation of the inhibitors
The inhibitors of the library were prepared by microwaveꢀassisted acetylation of the
corresponding 2ꢀaminobenzothiazoles. Several 2ꢀaminobenzothiazoles are commercially
available; those that were not were prepared by a modified twoꢀstep Stuckwisch procedure30
according to Scheme 1. The 6ꢀsubsituted benzothiazoles were prepared from the
corresponding paraꢀanilines, analogously; the 4ꢀsubstituted derivatives were prepared from
the orthoꢀaniline. The metaꢀanilines can in principle give both 5ꢀ and 7ꢀsubstituted
benzothiazoles upon cyclization, however in practice, the 5ꢀisomer dominates, and only in the
ACS Paragon Plus Environment

Journal of Medicinal Chemistry
Page 8 of 41
1
2
3
4
5
case of 3ꢀtrifluoromethylanisole substantial amounts of the 7ꢀisomer was formed (the ratio
between the 5ꢀ and 7ꢀtrifluoromethyl isomer was 65:35).
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 1. a) NaSCN (1.5 eq.), TFA (2.5 eq.), Isopropyl acetate (reflux, 16h), b) Br2 (1 eq.),
LiBr (1.5 eq), AcOH, 40 °C, c) Acetic anhydride (2.2 eq.), DIPEA (3 eq.), DMF, Microvawe.
Protein kinase profiling
The protein kinase selectivity pattern was measured as a Gini coefficient31 calculated on the
basis of one of two kinase panels; either a large panel with 140 protein kinases or a smaller
panel with 50 protein kinases (Table S1 lists the protein kinases included in the panels). The
Gini coefficients is based on the Lorenz curve, which plots the proportion of the total
inhibition of the full kinase panel that is cumulatively inhibited by each protein kinase. A Gini
coefficient can thus vary from 0 (when all protein kinases are inhibited equally) to 1 (where
the inhibitor inhibits only one protein kinase) and is regarded as a robust measure of kinase
selectivity and is generally not greatly affected by panel size.31 The fragments were profiled at
a concentration of 100 ꢁM to compensate for the expected low binding expected for
fragments, and also to be able to pick up protein kinase targets with weaker binding affinity.
The results from the protein kinase profiling are presented in Table 1 as a Gini coefficient
for each tested fragment molecule. With few exceptions, the observed Gini indices indicated
that the fragments were quite selective (Gini index > 0.4). Members of the CMGCꢀ
superfamily (the DYRKꢀfamily, GSK3β, the CLKꢀfamily) as well as CK1 were among the
targets commonly inhibited by the fragments.
The hydroxylꢀsubstituted molecules 1 and 2 were more promiscuous inhibitors than the
other members of the library. The halogenꢀsubstituted fragments were more selective, but the
ACS Paragon Plus Environment

Page 9 of 41
Journal of Medicinal Chemistry
1
2
3
4
5
6
7
8
highest selectivities were found for the methoxyꢀ, trifluoromethoxyꢀ and trifluoromethylꢀ
compounds. There was also a clear trend that the 6ꢀsubstituted molecules were less selective
inhibitors than the corresponding 5ꢀsubstituted molecules.
9
Inhibition of DYRK1A
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
The primary target for this study is the ADꢀrelevant protein kinase DYRK1A. Inhibition of
DYRK1A activity was measured as IC50 values determined using a radioactive filterꢀbinding
assay provided by the International Centre of Kinase Profiling at the University of Dundee,
UK. The results are compiled in Table 1. The inhibitory activity of the fragmentꢀlike
molecules ranged from 0.4 to 36.0 ꢁM, corresponding to high ligand efficiencies (LE)32 with
LE values ranging from 0.369 – 0.600 kcal/mol per nonꢀhydrogen atom. The most effective
inhibitors were substituted with hydroxy or methoxy groups, halogen, or classical electron
withdrawing groups (EWG) such as trifluoromethyl, cyano or nitro groups. In the majority of
the molecules, the position of the substituent was also significant for inhibitory activity; for
the hydroxyꢀ and methoxyꢀcompounds the 5ꢀsubstituted compounds were 35 or 31 times more
potent, respectively, than the 6ꢀsubstituted isomers. A similar but weaker trend was observed
for the 5ꢀfluoroꢀ and 5ꢀbromobenzothiazoles that were 6 and 2 times more potent than the 6ꢀ
substituted analogs, respectively. On the other hand, for other types of substituents, the 6ꢀ
isomer was most effective; e.g. the 6ꢀtrifluoromethyl derivative was 24 times more efficient
as a DYRK1A inhibitor than the 5ꢀisomer. Despite low activity, the methyl derivatives
showed a similar pattern with the 6ꢀmethyl derivative 4 times more potent than the 5ꢀisomer.
Judged by the reported % residual activity against DYRK1A, the benzothiazole fragments
in this work showed a similar level of inhibitory activity as Dꢀluciferin.29 However, both
hydroxyꢀ and methoxyꢀderivatives of the current molecules were potent DYRK1A inhibitors
whereas methoxyꢀluciferin showed diminished activity relative to luciferin.29
ACS Paragon Plus Environment

Journal of Medicinal Chemistry
Page 10 of 41
1
2
3
4
5
6
7
8
9
Table 1. Inhibitory activity of benzothiazole derived fragments against protein kinase
DYRK1A measured as residual activity (%) at an inhibitor concentration of 100 ꢁM, IC50
values (ꢁM) against DYRK1A, protein kinase selectivity calculated as a Gini coefficient
profiled against a 50 member or a 140 member protein kinase panel at 100 ꢁM and structure
cluster as determined by Xꢀray crystallography.
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
Compound
X
% residual acitivity
(DYRK1A)
IC50
GINI
Structural
cluster
(DYRK1A) coefficient
6ꢀOH
20
5
28.1
0.8
0.291
0.348
0.489
0.473
0.618
0.668
0.473
0.527
0.439
ꢀ
1
2
5ꢀOH
4ꢀOH
6ꢀOCH3
5ꢀOCH3
4ꢀOCH3
6ꢀF
1
1
1
2
25
14
2
35.0
12.3
0.4
3
4
5
15
12
7
10.5
23.6
3.9
6
7
5ꢀF
8
6ꢀBr
14
ND*
6
2.3
9
5ꢀBr
1.1
10
11
12
13
14
15
16
17
18
19
20
6ꢀCl
2.9
0.475
0.480
0.617
0.414
0.500
ꢀ
6ꢀCF3
5ꢀCF3
6ꢀCN
6ꢀNO2
6ꢀCONH2
6ꢀOCF3
6ꢀCH3
5ꢀCH3
H
8
1.1
22
9
26.5
1.3
2
2
6
0.7
ND*
28
ND*
23
43
15.7
3.2
0.552
ꢀ
7.3
29.3
36.0
0.546
0.551
*ND: Not determined
ACS Paragon Plus Environment

Page 11 of 41
Journal of Medicinal Chemistry
1
2
3
4
5
6
Structure of the complexes
7
8
9
In the present work, six complexes between the benzothiazole fragment molecules and the
DYRK1A protein could be successfully coꢀcrystallized. The Xꢀray structures show the overall
crystal packing to be similar to the DYRK1A structures previously published,33,34 with
tetramers (labeled monomer A, B, C and D) constituting the asymmetric unit. In general, the
best electron density fit is found for monomer A. The greatest disorder is seen in monomer C,
where especially the CMGC insert35 shows a poor fit to the electron density (as indicated e.g.
by low real space Rꢀfactors). In alignment with the previously described structures, several
chains in the monomers show disulfide bridge formation between the HCD motif and the
activation loop cysteine.34 However, in approximately half of the monomers, the cysteine
residues are reduced and/or in a mixed state, with additional evidence of radiation damage due
to low amplitude electron density for cysteine C312. Nevertheless, the oxidation state of the
cysteines C286 and C312 does not seem to influence the geometry of binding of the
inhibitors.
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
The presence of the six inhibitors is clearly shown by omit difference density maps (Figure
S1). However, the lack of strongly asymmetric features in the ligand, along with the degree of
mesomery available to the inhibitors, creates ambiguity regarding binding poses for some
inhibitors. Furthermore, the omit maps suggest that that there are binding poses in monomer
D (and to a lesser degree B) that are not completely overlapping with the omit maps of the
more densely packed monomer A (Figure S2).
The sulfur atom of the thiazole ring provides the strongest and usually unequivocal signal
for interpretation of the difference density maps. Although the benzothiazole ring system is
rigid, the acetamide moiety has two rotatable bonds with structural consequences for ligandꢀ
protein interactions. Both the bond linking the thiazole ring with the amide nitrogen atom and
ACS Paragon Plus Environment

Journal of Medicinal Chemistry
Page 12 of 41
1
2
3
4
5
6
7
8
9
the bond between the amide and the acetamide carbonyl are restricted to interconvertible
torsional angles close to 0° and 180°. The first torsion can be defined by S1ꢀC2ꢀN(amide)ꢀ
C(carbonyl) and be labeled syn (0° torsion angle) or anti (180° torsion angle), while the amide
bond, defined by C2ꢀN(amide)ꢀC(carbonyl)ꢀC(methyl) can be regarded as trans (180° torsion)
or cis (0° torsion). The diffraction data do not resolve whether the amide torsion is cis or
trans.
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
The six inhibitors can be subdivided into two structural clusters based on their poses in the
ATP binding pocket. Cluster 1 inhibitors bind via hinge interactions with the 5ꢀsubstitution
moiety, while cluster 2 inhibitors bind "edge"ꢀwise via hinge interactions with the
benzothiazole and amide nitrogen atoms.
Structural Cluster 1
Cluster one comprises inhibitors 2, 5 and 8, with 5ꢀOH, 5ꢀOMe and 5ꢀF substituents,
respectively. The hydroxyꢀcompound 2 can be placed in an unequivocally unique pose based
on the thiazole sulfur atom and bind to DYRK1A with the 5ꢀhydroxy group as the hinge
binder, and the acetamide group forming Hꢀbonds, either in a synꢀtrans conformation to the
catalytic lysine K188 (Figure 2A), or alternatively in a synꢀcis conformation to D307 from the
DFG motif(dotted gray line in Figure 2A). Compound 2 can make two hydrogen bonds with
the hinge, one with the hydroxyl oxygen as a hydrogen bond acceptor to the amide of L241,
and one as a hydrogen donor to the carbonyl of L241. In comparison, the methoxyꢀcompound
5 in crystallographic monomer A is slightly skewed inward into the lipophilic pocket forming
the acetamido interactions to D307 in an antiꢀcis conformation (Figure 2B), whereas in
crystallographic monomer D (Figure 2B’), the inhibitor shows an even larger inward skew,
and the interactions between the acetamido moiety and D307 is now taking place through a
synꢀcis ligand conformation. The hinge interaction of compound 5 takes place through a
hydrogen bond between the L241 amide and the methoxyꢀoxygen atom. Thus, in contrast to
ACS Paragon Plus Environment

Page 13 of 41
Journal of Medicinal Chemistry
1
2
3
4
5
6
7
8
9
the 5ꢀhydroxyꢀcompound 2, where the hinge binding involves two hydrogen bonds, the
methoxyꢀcompound 5 can only form a single Hꢀbond. The gray dashed line in Figure 2A
represents an alternative Hꢀbond to D307 when the inhibitor 2 binds with its alternate syn/cis
isomer.
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 2. Binding poses of the individual inhibitors in the ATP pocket. A: 5ꢀ
hydroxybenzothiazole 2 in its syn/trans conformation (dotted gray line represents the
ACS Paragon Plus Environment

Journal of Medicinal Chemistry
Page 14 of 41
1
2
3
4
5
6
7
8
9
interaction between the carbonyl oxygen and D307 in the alternative syn/cis conformation),
B: Overlay of the two binding modes of 5ꢀmethoxybenzothiazole 5 in crystallographic
monomer A, B’: Overlay of the two binding modes of 5ꢀmethoxybenzothiazole 5 in monomer
D, C: Overlay of the two binding modes of 5ꢀfluorobenzothiazole 8 in monomer A, D:
Overlay of 6ꢀchlorobenzothiazole 11 and 5ꢀhydroxybenzothiazole 2.
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
The 5ꢀfluoro compound 8 also uses the substituent as a hinge binder, however through an
unusual orthogonal multipolar interaction with a hinge carbonyl oxygen atom (Figure 2C).36,37
The distance between CꢀF and O=C varies among the four fragments bound to the four chains
AꢀD in the asymmetric unit, with distances between 3.1Å ꢀ 4.0 Å and an average of 3.6 Å.
With the exception of the 4.0 Å distance, the fragments bind at distances typical for Cꢀ
F···O=C interactions found in small molecule crystal structures38 and protein ligand
interactions.36,37 Similar to the 5ꢀmethoxyꢀcompound (in binding pose 1), the acetamide
moiety of inhibitor 8 is anchored to the DYRK1A enzyme via a hydrogen bond to both D307
and K188 in an anti/cis acetamide conformation.
Additional binding poses in Cluster 1
For the methoxyꢀcompound 5, the difference electron density map also reveals an additional
sulfur position corresponding to a binding mode where the inhibitor is rotated by 180 degrees
around its short axis, effectively swapping the positions of the acetamide carbonyl oxygen
atom and the methoxy oxygen atom compared with the principal binding pose (Figure 2B and
B’). In the alternate binding mode the L241 amide forms a Hꢀbond to the anti/cis acetamide
carbonyl oxygen. Away from the hinge, 5ꢀmethoxy group is oriented towards the catalytic
lysine K188, which remains ordered and hydrogen bonded to its active state partner E203
from helix C. The distance between the methoxyꢀoxygen atom and the terminal amine
ACS Paragon Plus Environment

Page 15 of 41
Journal of Medicinal Chemistry
1
2
3
4
5
nitrogen of K188 is ~4 Å, too far for a hydrogen bond, but clearly an important local
electrostatic interaction.
6
7
8
9
In all binding poses identified so far the ligand orients its thiazole sulfur atom towards the
gatekeeper phenylalanine, forming a sulfurꢀaromat interaction. The thiazole sulfur–
phenylalanine phenyl distance varies among the inhibitors, and the shortest distance to the
phenyl ring (3.5 Å) is found for the 5ꢀfluoro inhibitor 8, while the other two inhibitors show
distances that range between 3.9 and 4.5 Å.
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
In monomer A of the crystallographic tetramer of the compound 8 complex, the 5ꢀfluoro
compound can also adopt a binding pose atypical for Cluster 1. The electron density for 5ꢀ
fluoro molecule 8 in monomer A of pdb code 5A4L is consistent with two distinct poses, and
both poses are present in monomer A (Figure S1C). In the alternate binding pose (Figure 2C)
the 5ꢀfluoro inhibitor is oriented along the hinge with the thiazole sulfur atom oriented
towards L241, and a hydrogen bond linking the amide of S242 and the acetamide carbonyl,
which is in an antiꢀconformation. The distance between the thiazole sulfur atom and the L241
nitrogen atom is 3.1 Å, suggestive of a hydrogen bond. In the alternate binding pose the
fluorine atom is located in a similar position as occupied by the sulfur atom in the normal
binding pose (Figure 2C), however the distances between the fluorine atom and the carbon
atoms in the phenyl ring are between 5.2 and 6.0 Å, too long for a specific fluorineꢀaromat
interaction. The finding, however, that the binding poses of compound 5 in the B, C and D
monomers of the asymmetric unit are unambiguously aligned with the normal Cluster 1
binding mode indicates that the alternate binding pose found only in monomer A is a less
preferred binding pose for the 5ꢀfluoro inhibitor.
ACS Paragon Plus Environment

Journal of Medicinal Chemistry
Page 16 of 41
1
2
Structural Cluster 2
3
4
5
6
7
8
Cluster two (represented in Figure 2D by the 6ꢀchloroꢀinhibitor 11), contains the binding
poses adopted by benzothiazoles with 6ꢀchloroꢀ, 6ꢀnitroꢀ and 6ꢀcyanoꢀsubstituents (11, 14 and
15). The binding poses of Cluster 2 are uniquely identified by the thiazole sulfur atom and the
substituent in the 6ꢀposition, and show the thiazoleꢀ and acetamideꢀnitrogen atoms to bind at
the hinge (the thiazole nitrogen atom to the L241 amide and the acetamide nitrogen to L241
carbonyl moiety), while the substituents are oriented toward the gatekeeper F238 as shown for
compound 11 in Figure 2D. All inhibitors in structural cluster two bind with the acetamido
group in the synꢀconformation required for the second hydrogen bond to L241, but the
acetamide carbonyl group is not involved in any hydrogen bond formation. The 6ꢀchloroꢀ
substituent of fragment 11 superimposes with the thiazole sulfur atom of the 5ꢀhydroxyꢀ
substituted fragment 2 (Figure 2D) replacing the sulfur – aromat with a halogen – aromat
interaction. The chlorine atom in 11 interacts with F238 via one edge of the aromatic ring
system. While the distance to the center of the aromatic π system is approximately 4.4 Å, the
closest contact to the phenyl ring edge is only 3.6 Å. These distances are in agreement with
typical halogen phenyl interactions.37,39 The angle between the plane of the aromatic ring and
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
the chlorine atom is around 80°. Thus in this case the chlorine atom interacts with the
π
system itself rather than forming the more common CH···Cl interaction.37 Similar interactions
of the substituent with the πꢀsystem of the gatekeeper residue can be observed for 6ꢀcyanoꢀ
(14) and 6ꢀnitroꢀbenzothiazole (15). The 6ꢀcyanoꢀgroup is nearly parallel to the edge of the
phenylalanine, at a distance of ~4.2 Å. The 6ꢀnitroꢀcompound 15 participates in a nitroꢀπꢀ
stacking interaction. One of the oxygen atoms in the 6ꢀnitroꢀsubstituent is placed below the
center of the π system at a distance of 3.3 Å; the nitrogen itself is at a distance of 4.3 Å. The
substituents of the inhibitors in structural cluster two do not form hydrogen bonds to D307
ACS Paragon Plus Environment

Page 17 of 41
Journal of Medicinal Chemistry
1
2
3
4
5
nor to the catalytic lysine K188, which is ordered and hydrogen bonded to its partner E203
from helix C.
6
7
8
9
Dynamics of ligand binding
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
The Xꢀray structure shows that the 5ꢀmethoxyꢀcompound 5 probably binds to the active site
of DYRK1A in either of two distinct binding poses related by a 180° rotation about an axis
perpendicular to the benzothiazole ring (Figure 2B). To further probe this structural
ambiguity, the binding mode of 5 was investigated using NMR techniques. A slight line
1
broadening of the Hꢀsignals of ligand 5 was observed in the presence of DYRK1A (100:1
ligand:protein in D2O) (Figure S3), characteristic of a rapid equilibrium between bound and
unbound states, consistent with the low µM binding constant. The presence of rapid binding
equilibrium between free and protein bound ligand was further corroborated in a Saturation
Transfer Difference (STD) experiment (Figure S4).
The multiple binding modes of 5 to DYRK1A were examined by a 2D transferred NOE
(trNOESY) experiment on the same sample as above. The trNOESY experiment revealed an
unexpected longꢀrange correlation between the methyl group of the acetamide moiety and the
Hꢀ6 hydrogen of the benzothiazole ring (Figure S5A). This correlation disappears in an
experiment performed in the absence of protein, showing directly that compound 5 interacts
with the DYRK1A protein in two distinct binding poses. The correlation arises when the Hꢀ6
atom of the benzothiazole ring in one binding mode occupies the same space in the protein as
the acetamide methyl group in the alternate binding mode (corresponding to a 180° flip of the
ligand between the binding poses). This observation is fully consistent with the two binding
modes proposed by the Xꢀray crystallographic data (Figures 2B and B’) existing in rapid
equilibrium on the mixing time scale (milliseconds). Lacking a better term, this will be
referred to as "molecular autoꢀcorrelation" in the discussion below – i.e. there is a correlation
ACS Paragon Plus Environment

Journal of Medicinal Chemistry
Page 18 of 41
1
2
3
4
5
between the position of one proton with the position of another proton in the same molecule at
an earlier point in time.
6
7
8
9
The binding mode of the 5ꢀhydroxyꢀligand, 2 to DYRK1a was also investigated by
trNOESY. The experiment revealed a weak “autoꢀcorrelation” between the acetamido methyl
group and the Hꢀ4 hydrogen at the benzothiazole moiety (Figure S6). As with compound 5,
this correlation is consistent with two binding poses related by a 180° flip between the
binding modes, however for the hydroxyl compound 2, one pose is significantly more
populated than the other. Xꢀray crystallography showed only a single wellꢀresolved binding
mode for 2. Interestingly, while the 5ꢀhydroxy compound 2 shows an “autoꢀcorrelation”
between methylꢀ3’ and Hꢀ4, the methoxy compound 5 instead shows a correlation between
methylꢀ3’ and Hꢀ6, indicating that the binding modes of ligands 2 and 5 are slightly different.
This observation is consistent with the binding modes in the Xꢀray structures for 5 (monomer
D) and 2 (Figures 2B/B’ and 2A, respectively) where compound 5 is slightly more skewed
inward than compound 2.
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
MD simulations were carried out for compound 2 and 5 in two binding modes to determine
the relative binding affinities of the respective binding modes observed in Xꢀray and NMR
structures. Consistent with the experimental data, the free energy simulations show that the
difference in binding free energy between the two orientations of compound 2 is 2.4
±0.4
kcal/mol in favor of the Xꢀray conformation. For compound 5, the binding free energy
difference between the two orientations is 0.7±0.3 kcal/mol, indicating that the two modes are
equally populated.
A trNOE competition experiment between ligands 2 and 5 further revealed that the 5ꢀ
methoxy group in ligand 5 displayed interligand trNOEs (ILOE) to the aromatic 4ꢀ, 6ꢀ, and 7ꢀ
hydrogens of the hydroxy ligand, 2, implying that the dominant binding orientations of 2 and
5 are opposite (Figure S7). As ligand 5 binds to DYRK1A with two significantly populated
ACS Paragon Plus Environment

Page 19 of 41
Journal of Medicinal Chemistry
1
2
3
4
5
binding modes, correlations from the near overlapping acetamide methyls of 2 and 5 produced
ILOe’s to the Hꢀ6 and Hꢀ7 hydrogen of 2 and the Hꢀ4 and H7 of 5 .
6
7
8
Because the acetamido methyl groups of compound 2 and 5 were insufficiently resolved in
9
1
the 2D trNOESY experiment, an additional 1D dpfgseꢀNOESY experiment with higher H
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
resolution was performed. Selective irradiation of the methoxy group in 5 produced a clear
ILOE to the acetamido methyl group in ligand 2, in addition to an “autoꢀcorrelation” to its
own acetamido methyl group (Figure S8).
The findings from the trNOE competition experiments of ligands 2 and 5 are in good
agreement with the results obtained by Xꢀray crystallography. Ligand 5 binds in two well
populated different poses differing by a 180° flip, whereas ligand 2 predominantly binds in
one binding pose similar, but slightly skewed compared to one of the binding poses of 5
(Figure 2 A, B and B’).
A similar trNOE competition experiment was also performed between the 5ꢀhydroxy (2)
and 6ꢀhydroxy (1) ligands (Figures S9). In this case the ILOE’s are edge to edge pseudoꢀ
contacts between the aromatic protons, a finding that implies that 1 and 2 bind in similar
poses, where the hydroxylꢀ and the acetamido groups occupy the same space, respectively,
but the benzothiazole ring of 1 is flipped around its long axis, allowing the 6ꢀhydroxyl groups
to occupy the same space as the 5ꢀhydroxyl group in 2. As a consequence of the
benzothiazole swap, compound 1 binds to DYRK1A with the sulfur atom facing the outside
of the active site while compounds 2 and 5 bind with their sulfur atom facing inward in the
active site in all their binding modes.
ACS Paragon Plus Environment

Journal of Medicinal Chemistry
Page 20 of 41
1
2
3
4
DISCUSSION
5
6
7
8
9
From the finding that luciferin acts as an inhibitor of DYRK1A as well as other members of
the CMGCꢀsuperfamily (the DYRKꢀfamily, GSK3β, the CLKꢀfamily) as well as CK1, a 20
member focused fragment library based on the benzothiazole scaffold was designed, prepared,
profiled and tested for inhibitory activity against DYRK1A. The results showed that several
members of this library displayed high DYRK1A selectivity (with GINI indices > 0.500) and
very high binding efficacy (LE up to 0.600 kcal/mol). Furthermore, clear trends in fragment
structureꢀDYRK1A inhibitory activity and fragment structureꢀprotein kinase selectivity could
be discerned.
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
With two exceptions, the 4ꢀhydroxy and the 4ꢀmethoxy isomers (compounds 3 and 6)
respectively) that both showed weak inhibitory activity against DYRK1A, the benzothiazole
fragments were designed to be pseudoꢀsymmetrical, i.e. the 5ꢀ and 6ꢀsubstituents are related
by a 180° rotation along the long molecular axis, effectively swapping the positions of the
thiazole Nꢀ and Sꢀatoms (Figure 1). Despite the pseudosymmetric nature of the fragments,
large regioisomeric differences in the inhibitory activities against DYRK1A (determined as
IC50 values) could be observed (Table 1). Inspection of Table 1 reveals that the 5ꢀ and 6ꢀ
substituted fragments could be classified according to their activity, strong (< 5 ꢁM) vs weak
(> 5 ꢁM), and regional selectivity (activity of 5ꢀisomer vs the corresponding 6ꢀisomer). Based
on these criteria, four classes of inhibitors can conveniently be constructed. The first class
consists of the fragments where substitution in the 5ꢀposition produces more effective
inhibitors than the corresponding substitution in the 6ꢀposition with a high degree of
selectivity (> 5 times). Class 1 consisted of inhibitors with methoxyꢀ, hydroxyꢀ and fluoroꢀ
substituents, and the regional selectivity was up to 35 fold for the hydroxylꢀsubstituent. The
second class consisted of the bromoꢀsubstututed benzothiazoles (9 and 10) where both
isomers produced strong inhibitors (albeit with a weak preference for the 5ꢀisomer). The third
ACS Paragon Plus Environment

Page 21 of 41
Journal of Medicinal Chemistry
1
2
3
4
5
6
7
8
9
class contained compounds where substitution in the 6ꢀposition produced strong inhibitors.
Class three consists of compounds with chloro, cyano, nitro, trifluoromethoxy and
trifluoromethyl substituents. For the fragments with a trifluoromethylꢀsubstituent (compounds
12 and 13), the 6ꢀisomer was 24 times more potent than the 5ꢀisomer. The last class consisted
of inhibitors with weaker activity against DYRK1A (unsubstituted benzothiazole as well as
carboxamido and methyl substituted benzothiazole derived fragments), although some
selectivity (towards the 6ꢀposition) for the methyl substituted isomers was evident.
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. Overlay of all binding poses from the Xꢀray crystallographic structures for the
DYRK1A complexes with compounds 2, 5, 8, 11, 14 and 15.
The Xꢀray and NMR structural studies reveal that the present benzothiazole fragment
library quite extensively explores the ATPꢀbinding pocket. This is depicted in Figure 3 which
shows a superposition of all binding poses observed in the Xꢀray crystallography structures.
The effective exploration of the binding pocket also allows many of the trends in structure –
DYRK1A inhibitory activity to be understood and interpreted from the Xꢀray crystallographic
ACS Paragon Plus Environment

Journal of Medicinal Chemistry
Page 22 of 41
1
2
3
4
5
6
7
8
and NMR structural studies of the targetꢀligand complexes. Furthermore, the Xꢀray studies
also suggest that some ligands may bind in different binding poses, a feature confirmed by the
transient NOE NMR experiments.
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
A
B
C
D
E
F
ACS Paragon Plus Environment

Page 23 of 41
Journal of Medicinal Chemistry
1
2
3
4
5
6
7
8
9
Figure 4. Schematic representations of the binding poses for representative members in
Cluster 1 and Cluster 2. Panel A shows the 5ꢀhydroxyꢀcompound 2, Panels B and D shows the
two alternative binding poses of the 5ꢀmethoxyꢀcompound 5, Panels C and E shows two
binding poses of the 5ꢀfluoroꢀcompound 8, and Panel F shows the binding pose of the 6ꢀ
chloroꢀcompound 11 as an example of the binding mode in Cluster 2.
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
The Xꢀray crystal structures fall broadly into two clusters; the first cluster consists of the
Class 1ꢀinhibitors with high positional selectivity for the 5ꢀ position. These are represented in
the Xꢀray structures with the 5ꢀhydroxy derivative 2, the 5ꢀmethoxyꢀderivative 5 and the 5ꢀ
fluoroꢀderivative 8. The overall position of the inhibitors from this cluster in the active site of
DYRK1A is determined by the interactions between the 5ꢀsubstituent to the hinge residue
L241 and the acetamide moiety to the D307 from the DFG motif or K188, as illustrated in
Figure 4AꢀC. The 5ꢀF compound 8 is able to form a C=O···F bond with a geometry suitable
for productive “orthogonal multipolar interactions” to the hinge L241. This interaction,
however, is weak compared to the hydrogen bonds seen for the 5ꢀOH and 5ꢀOMe compounds;
as a consequence, the 5ꢀF fragment shows a 5ꢀfold reduction in affinity compared to the other
two.
Noteworthy for the Cluster 1 Xꢀray structures is the presence of a second binding pose
defined by a flip about the short axis for the 5ꢀOMe compound, switching the hinge binding
interaction from the methoxy group to the acetamide carbonyl (e.g. Figure 4B vs. Figure 4D).
This second binding pose for compound 5 is confirmed by the trNOE experiments. The
trNOE NMR also indicates that this flip along the short axis is possible for the 5ꢀOH
compound 2 as well, however, the alternate binding pose in compound 2 is much less
populated in the trNOE experiments, and is not observed in the Xꢀray structure, showing that
the hydroxyl group at the hinge with the potential of two hydrogen bonds is clearly preferred
ACS Paragon Plus Environment

Journal of Medicinal Chemistry
Page 24 of 41
1
2
3
4
5
compared to the alternate binding pose with its single hydrogen acceptor (the carboxamide
carbonyl group).
6
7
8
9
A significant feature of all Xꢀray and NMR structures of these two binding poses is that the
thiazole sulfur atom is oriented towards the interior of the ATP binding pocket in an
interaction with the gatekeeper phenylalanine F238. The distance between the thiazole sulfur
atom and the face of the F238 phenyl group varies between 3.5 and 4.5 Å, compatible with
favorable sulfur – π interaction energies.40 Due to the pseudorotational relationship between
the 5ꢀsubstituted and the corresponding 6ꢀsubstituted inhibitors, the 6ꢀsubstituted isomers
could adopt the same hinge and acetamidoꢀD307/K188 interactions as the 5ꢀisomers;
however, the 6ꢀisomers would then have to orient the thiazole nitrogen atom towards the
inside of the pocket and not the thiazole sulfur. This allows the conclusion that the regional
selectivity for the 5ꢀisomers of the Class 1 inhibitors is to a large degree attributable to the
thiazole sulfur – gatekeeper phenylalanine interaction.
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 direct insight into the importance of the sulfur – aromat interaction is provided by trNOE
NMR competition experiments of the 5ꢀhydroxy fragment 2 and the 6ꢀhydroxy fragment 1.
The experiments show that the hydroxyl groups of 1 and 2 occupy the same space of
DYRK1A, just as their acetamido groups do. This finding illustrates the governing influence
of the hinge and D307 interactions on the overall binding pose of these fragments. However
as discussed above, the 5ꢀhydroxy compound 2 is bound with its thiazole sulfur facing the
phenyl group of F238, whereas the same binding pose in 1 would leave the thiazole sulfur
atom exposed to the solvent. The influence of sulfurꢀaromat interaction is thus evident when
comparing the inhibitory activities of 1 and 2 (Table 1), where the IC50 value of compound 2
is 35 times lower than for compound 1. A similar interaction pattern for the methoxyꢀ
substituted pair 4 and 5 can also explain the 30ꢀfold difference favoring the 5ꢀmethoxyꢀ
compound. The fluoroꢀsubstituted compounds bind less efficiently (the 5ꢀfluoroꢀcompound 8
ACS Paragon Plus Environment

Page 25 of 41
Journal of Medicinal Chemistry
1
2
3
4
5
6
7
8
9
inhibits 10 times less efficiently than the 5ꢀmethoxyꢀcompound 11), resulting from the weaker
orthogonal multipolar interaction linking the fluorine atom to the carbonyl of L241. However,
only the 5ꢀfluoroꢀisomer is capable of a sulfur – aromat interaction with F238, as described
above, leading to an expected regioisomeric selectivity for the 5ꢀisomer. The observed
selectivity is only 6ꢀfold and is thus not as prominent as for the hydroxyl and methoxy
substituted compounds. However, because the 5ꢀfluoroꢀcompound displays a binding mode
along the hinge (Figure 4E) with a moderate strength hydrogen bond to the thiazole sulfur, the
6ꢀisomer may as well adopt a similar binding mode, decreasing the sensitivity to the
regioisomeric differences in the overall orientation of the benzothiazole ring system.
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
Structural cluster two consists of the Class 3 fragment molecules with classical electron
withdrawing substituents in the 6ꢀposition, including nitroꢀ (14) and cyanoꢀgroups (15), as
well as a chlorine substituent (compound 11) with a combined inductive withdrawing, but
mesomeric electron donating ability. The structures within cluster two are very similar to each
other as depicted in Figure 4F, but are very different from the structures in cluster one. The
binding mode in cluster two is most likely governed by the electron withdrawing ability of the
6ꢀsubstituent that is directed towards the gatekeeper phenylalanine, allowing the formation of
a
σꢀhole –
π
interaction37 (for the 6ꢀchloro inhibitor 11), a nitroꢀ
π interaction (for the 6ꢀnitro
inhibitor 14) or a
π
– π
interaction (for the 6ꢀcyano inhibitor 15) between the substituent and
the aromatic ring system of F238. The amide group of the acetamide moiety forms a hydrogen
bond to the carbonyl group of L241. Furthermore, close contacts ranging from 3.1 Å (for the
6ꢀchloro derivative) to 3.3 Å for the (6ꢀcyano and 6ꢀnitro derivatives) between the amide
nitrogen of L241 and the nitrogen atom of the thiazole ring are observed. These distances
correspond to hydrogen bonds, either “moderate and mostly electrostatic” or “weak,
electrostatic” according to the definition of Jeffrey.41 The mesomeric electron withdrawing
effect coupling the 6ꢀsubstituent directly to the thiazole nitrogen atom through resonance
ACS Paragon Plus Environment

Journal of Medicinal Chemistry
Page 26 of 41
1
2
3
4
5
6
7
8
9
would be expected to weaken the hydrogen bond acceptor ability for the thiazole nitrogen
atom relative to the corresponding 5ꢀisomer, or to compounds with a mesomeric electron
donating 6ꢀsubstituent, hence the observed binding mode is surprising, and highlights the
importance of the “weak” interaction between the electron withdrawing 6ꢀsubstituent and the
gatekeeper Phe238.
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
The bromine substituted compounds 9 and 10 take an intermediate position, where both the
5ꢀ and the 6ꢀisomer show substantial inhibition. It is not unreasonable to expect that the 6ꢀ
bromo compound would binds to DYRK1A with a pose similar to that of the 6ꢀchloro
compound 11. The main interactions in the binding of 11 would thus be a hydrogen bond
between the acetamidoꢀNH and the carbonyl of L241, as well as a halogen σꢀhole πꢀ
interaction with F238, and a weak hydrogen bond between the L241 amide and the thiazoleꢀ
nitrogen atom. If bound in the same pose, the 5ꢀbromoꢀisomer would be able to enjoy the
same strong interactions as the 6ꢀisomer (i.e. σꢀhole πꢀinteraction and the acetamide hydrogen
bond), but would replace the weak thiazoleꢀnitrogen hydrogen bond with a weak thiazole
sulfur hydrogen bond. The 5ꢀbromo position is preferred, but the difference in IC50 values
(twoꢀfold) between the 5ꢀ and 6ꢀbromo isomers is marginal compared to the other cases where
a clear regiochemical trend is observed.
The protein kinase inhibition promiscuity of the studied inhibitors expressed as Giniꢀ
coefficients was also influenced by the type and regiochemistry of the benzothiazole
substituent. Generally, the inhibitors in the library showed high selectivity for DYRK1A, with
Giniꢀindices above 0.500. On the other hand, some of the strong inhibitors, in particular the
hydroxylꢀsubstituted fragments 1 and 2, were much less selective for DYRK1A than the other
members of the library. The kinase selectivity is significantly higher for the halogenꢀ
substituted fragments, and highest for the methoxyꢀ, trifluoromethoxyꢀ and trifluoromethylꢀ
substituted fragments. There is also a clear trend that substitutions in the 6ꢀposition of the
ACS Paragon Plus Environment

Page 27 of 41
Journal of Medicinal Chemistry
1
2
3
fragments gives less selective inhibition than fragments with substituents in the 5ꢀposition.
4
5
6
The profiling data, performed at a ligand concentration of 100 ꢀM, indeed, reveals that the
7
8
9
hydroxy compounds are more promiscuous than the other members of the fragment library,
inhibiting an array of protein kinases at a moderate level. The reason for this difference in
promiscuity is likely related to the dual hydrogen bond donor acceptor properties of the
phenolic hydroxyl group that provide generic hingeꢀbinding properties to the hydroxyꢀ
compounds 1 and 2. It is important to note that the 5ꢀisomer 2 is an effective DYRK1A
inhibitor, whereas the 6ꢀisomer 1 is not. The present structural studies show that the
difference can be attributed to the additional binding interaction between the thiazole sulfur
atom and the gatekeeper phenylalanine residue. The 5ꢀisomers of methoxyꢀ and the fluoroꢀ
substituted compounds are far more selective DYRK1A inhibitors, in large part because their
"generic" hinge binding capacity is substantially lower than that for the 5ꢀhydroxyꢀcompound
2, and thus are more dependent on more specific kinase interactions. The major difference in
protein kinase profiles between the 5ꢀmethoxyꢀ and the 6ꢀmethoxyꢀisomer is that the 5ꢀisomer
inhibits DYRK1A particularly well, whereas the profile outside the DYRK family is rather
similar for both compounds. The sulfur – aromat interaction is thus not only responsible for
higher binding affinities, but also for the higher 5ꢀisomer selectivity of these compounds. For
the 6ꢀisomer selective compounds with electron withdrawing substituents, profiling data are
available only for the trifluoromethylꢀsubstituted compounds 12 and 13. These compounds
belongs to the structural class for which the interaction between the ligand and DYRK1A
takes place via two hydrogen bonds; one between the acetamideꢀH atom and a hinge
carbonyl, the other between a hinge amide and the thiazole nitrogen atom. This interaction is
complemented by interactions between the electron withdrawing substituent and the
gatekeeper phenylalanine. The 5ꢀtrifluoromethyl compound 13 is a modest DYRK1A
inhibitor, but is a selective inhibitor for the PIM1 and PIM3 kinases.
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
ACS Paragon Plus Environment

Journal of Medicinal Chemistry
Page 28 of 41
1
2
3
4
5
CONCLUSION
6
7
8
9
The protein kinase DYRK1A has emerged as a potential target for small molecule therapy
of Alzheimer’s disease. In the present study we have explored static and dynamic structural
properties of fragment sized variants of the benzothiazole scaffold with respect to DYRK1A
using Xꢀray crystallography and NMR techniques. Not only do several of these small
inhibitors have excellent ligand efficiencies, their DYRK1A complex structures show a
remarkable diversity of binding modes in dynamic equilibrium. Our studies have shown that
the overall binding modes of these fragmentꢀsized inhibitors are mainly determined by
classical interactions such as hydrogen bonding and electrostatic interactions. However, the
detailed binding geometries are apparently driven by additional types of interactions often
considered “weak”, including “orthogonal multipolar interactions”, e.g. FꢀCO, sulfurꢀaromat,
halogenꢀaromat. The most stable binding mode or modes in equilibrium are determined by the
sum of these interactions, together with the effects of electron withdrawing groups that
modulate hydrogenꢀbonding propensities, but without taking the “weak” interactions into
account no reliable model for structure activity relationship (SAR) can be derived.
Furthermore, several of the small benzothiazole inhibitors display dynamic behavior by
binding in multiple binding poses in rapid (NMRꢀtime scale) exchange, a feature that adds a
further layer of complexity to the SAR model. On the other hand, by taking both the detailed
structural interactions and the dynamic behavior into account a reliable SAR model of
DYRK1A inhibition by small, fragment sized, benzothiazole inhibitors can be derived, a
model which can even provide insight into protein kinase selectivity. The present work shows
that even fragmentꢀsized benzothiazole derivatives can be effective and selective inhibitors of
the protein kinase DYRK1A, however, the fragment nature of the compounds would require
significant effort in translating the molecules into drug lead compounds. On the other hand,
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
ACS Paragon Plus Environment

Page 29 of 41
Journal of Medicinal Chemistry
1
2
3
4
5
6
7
8
9
the identification of multiple “weak” binding interactions in addition to the classical hinge
and conserved lysine interactions provides guidance for further optimization. In particular, the
interactions with the gatekeeper F238 residue of DYRK1A may be modulated by design for
enhancing both potency and selectivity for future inhibitors.
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
Furthermore, the present study also illustrates the benefit of combining “static” Xꢀray
structural determination with “dynamic”ꢀNMR experiments for a more comprehensive
understanding of fragment binding, providing a foundation for the SBDD effort to prepare
selective and effective DYRK1A inhibitors that is underway in our laboratories.
ACS Paragon Plus Environment