Design and Analysis of the 4-Anilinoquin(az)oline Kinase Inhibition Profiles of GAK/SLK/STK10 Using Quantitative Structure-Activity Relationships

Article abstract of DOI:10.1002/cmdc.201900521

The 4-anilinoquinoline and 4-anilinoquinazoline ring systems have been the focus of significant efforts in prior kinase drug discovery programs, which have led to approved medicines. Broad kinome profiles of these compounds have now been assessed with the advent of advanced screening technologies. These ring systems, while originally designed for specific targets including epidermal growth factor receptor (EGFR), but actually display a number of potent collateral kinase targets, some of which have been associated with negative clinical outcomes. We have designed and synthesized a series of 4-anilinoquin(az)olines in order to better understand the structure-activity relationships of three main collateral kinase targets of quin(az)oline-based kinase inhibitors: cyclin G associated kinase (GAK), STE20-like serine/threonine-protein kinase (SLK) and serine/threonine-protein kinase 10 (STK10). This was achieved through a series of quantitative structure-activity relationship (QSAR) analysis, water mapping of the kinase ATP binding sites and extensive small-molecule X-ray structural analysis.

Full text of DOI:10.1002/cmdc.201900521

Accepted Article
Title: Design and analysis of the 4-anilino-quin(az)oline kinase
inhibition profiles of GAK/SLK/STK10 using quantitative structure
activity relationships
Authors: Christopher Robert Michael Asquith, Tuomo Laitinen, James
M. Bennett, Carrow I. Wells, Jonathan M. Elkins, William J.
Zuercher, Graham Tizzard, and Antti Poso
This manuscript has been accepted after peer review and appears as an
Accepted Article online prior to editing, proofing, and formal publication
of the final Version of Record (VoR). This work is currently citable by
using the Digital Object Identifier (DOI) given below. The VoR will be
published online in Early View as soon as possible and may be different
to this Accepted Article as a result of editing. Readers should obtain
the VoR from the journal website shown below when it is published
to ensure accuracy of information. The authors are responsible for the
content of this Accepted Article.
To be cited as: ChemMedChem 10.1002/cmdc.201900521
Link to VoR: http://dx.doi.org/10.1002/cmdc.201900521
A Journal of
www.chemmedchem.org

ChemMedChem
10.1002/cmdc.201900521
Design and analysis of the 4-anilino-quin(az)oline
kinase inhibition profiles of GAK/SLK/STK10 using
quantitative structure activity relationships
*
[a,b]
[c]
[d]
[b]
Christopher R. M. Asquith
, Tuomo Laitinen , James M. Bennett , Carrow I. Wells ,
[
d,e]
[b,f]
[g]
[c]
Jonathan M. Elkins , William J. Zuercher , Graham Tizzard
,
Antti Poso
[
a] Dr. C. R. M. Asquith, Department of Pharmacology, School of Medicine, University of North
Carolina at Chapel Hill, Chapel Hill, NC 27599 (USA)
E-mail: chris.asquith@unc.edu
[
b] Dr. C. R. M. Asquith, C. I. Wells, Dr. W. J. Zuercher
Structural Genomics Consortium, UNC Eshelman School of Pharmacy, University of North Carolina
at Chapel Hill, Chapel Hill, NC 27599 (USA)
[
b] Dr. T. Laitinen, Prof. A. Poso
School of Pharmacy, Faculty of Health Sciences, University of Eastern Finland, 70211 Kuopio
(
[
Finland)
c] J. M. Bennett, Dr. J. M. Elkins
Structural Genomics Consortium and Target Discovery Institute, Nuffield Department
of Clinical Medicine, University of Oxford, Old Road Campus Research
Building, Oxford, OX3 7DQ (UK)
[
d] Dr. J. M. Elkins
Structural Genomics Consortium, Universidade Estadual de Campinas-UNICAMP, Campinas, São
Paulo, 13083-886 (Brazil)
[
e] Dr. W. J. Zuercher
Lineberger Comprehensive Cancer Center, University of North Carolina at Chapel Hill, NC 27599
(
[
USA)
f] Dr. G. J. Tizzard
UK National Crystallography Service, School of Chemistry, University of Southampton, Highfield
Campus, Southampton, SO17 1BJ (UK)
http://orcid.org/0000-0001-5871-3458 (Christopher Asquith)
https://orcid.org/0000-0003-1539-2142 (Tuomo Laitinen
http://orcid.org/0000-0003-4799-6792 (Carrow Wells)
http://orcid.org/0000-0003-2858-8929 (Jonathan Elkins)
http://orcid.org/0000-0002-9836-0068 (William Zuercher)
http://orcid.org/0000-0002-1577-5779 (Graham Tizzard)
https://orcid.org/0000-0003-4196-4204 (Antti Poso)
Abstract: The 4-anilinoquinoline and 4-anilinoquinazoline ring systems have been the focus
of significant efforts in prior kinase drug discovery programs, which have led to approved
medicines. Broad kinome profiles of these compounds have now been assessed with the advent
of advanced screening technologies. These ring systems, while originally designed for specific
targets including epidermal growth factor receptor (EGFR), actually display a number of potent
collateral kinase targets, some of which have been associated with negative clinical outcomes.
We have designed and synthesized a series of 4-anilino-quin(az)olines in order to better
understand the structure activity relationships of three main collateral kinase targets of
quin(az)oline-based kinase inhibitors: cyclin G associated kinase (GAK), STE20-like
serine/threonine-protein kinase (SLK) and serine/threonine-protein kinase 10 (STK10). This
was achieved through a series of quantitative structure activity relationship (QSAR) analysis,
water mapping of the kinase ATP binding sites and extensive small molecule x-ray structural
analysis.
This article is protected by copyright. All rights reserved.

ChemMedChem
10.1002/cmdc.201900521
INTRODUCTION
Kinases have been successfully utilized as drug targets for the past 30 years, with 47 kinase inhibitors
approved by the FDA to date. These drugs are predominantly multi-targeted tyrosine kinase inhibitors
for the treatment of cancer.[1,2] A large number of additional inhibitors are currently being progressed
[2]
[3]
through preclinical and clinical development, but with over 500 kinases in the human genome, there
remains significant untapped potential within the kinome for the treatment a wide range of human
disease.[4]
To advance the understanding of kinome structure activity relationships, large scale kinome-wide
profiling of ATP-competitive kinase inhibitors have started to uncover the preferred chemotypes for the
inhibition of both well studied and under studied kinases.[5-8] The first- and second-generation
GlaxoSmithKline Published Kinase Inhibitor Sets (PKIS and PKIS2) are collections of well
characterized ATP-competitive kinase inhibitors with broad coverage of the human kinome.[9-10]
Together these screens have provided a more clear view of the kinome inhibitor landscape and the
kinome promiscuity profiles of these inhibitors. These kinome-wide profiling efforts have also started
to uncover the preferred chemotypes for the inhibition of different families or sub-families of kinases.[5-
1
0]
The 4-anilinoquin(az)oline is a common kinase inhibitor chemotype.[5-10] Several quinazoline-based
clinical kinase inhibitors show cross reactivity with cyclin G associated kinase (GAK), STE20-like
serine/threonine-protein kinase (SLK) and serine/threonine-protein kinase 10 (STK10), including the
approved drugs gefitinib, vandetanib, lapatinib, neratinib, afatinib, erlotinib, and other candidate
quin(az)olines such as pelitinib and dacomitinib (Figure 1).[2,6-7] These drugs were originally designed
as inhibitors of either epidermal growth factor receptor (EGFR) or Src family kinases, and all show
GAK, SLK, and STK10 as common collateral targets (Figure 1).[2,6-7] GAK activity in this chemotype
has already been observed in the erlotinib quinoline derivative (1)[11] and a selective chemical probe for
GAK, SGC-GAK-1 (2).[12] This GAK probe was developed from a series of quinolines exemplified by
3
(Figure 1).[13] The SLK/STK10 activity is less consistent, but several examples display potent profiles
with erlotinib demonstrating single digit nanomolar potency on both SLK and STK10 (Figure 1).[6-7]
This article is protected by copyright. All rights reserved.

ChemMedChem
10.1002/cmdc.201900521
Figure 1. Previously reported EGFR 4-anilinoquin(az)olines with GAK/SLK/STK10 activity.
GAK is a 160 kDa serine/threonine kinase originally identified as a direct association partner of
cyclin G and a member of the numb-associated kinase (NAK) family.[3,14-15] In addition to its kinase
domain, the GAK C-terminus has high homology with auxilin and tensin, with broad expression and
localization to the Golgi complex, cytoplasm, and nucleus.[16-17] GAK has been implicated in diverse
biological processes, and genome-wide association studies have identified single nucleotide
polymorphisms in the gak gene associated with susceptibility to Parkinson’s disease.[18] GAK is also
involved in membrane trafficking and uncoating of clathrin-coated vesicles in the cytoplasm,[19-20] and
is required for maintenance of centrosome maturation and progression through mitosis.[21] GAK is over-
expressed in osteosarcoma, and prostate cancer, where GAK is implicated in progression to androgen
independence.[12,22] GAK inhibition has been associated with clinical toxicity due to pulmonary alveolar
dysfunction based upon the phenotype of transgenic mice expressing catalytically inactive GAK, but
this hypothesis has not yet been tested with a selective small molecule GAK inhibitor.[23-24]
STK10/SLK are members of the Ste20 family of serine/threonine protein kinase sub-family. SLK
has just one amino acid difference in the ATP active site of STK10.[3,25-26] The expression and activity
of SLK is increased during kidney development and recovery from ischemia-reperfusion injury.[27-29]
SLK mediates apoptosis in various cells, and can regulate cell cycle progression and cytoskeletal
remodelling.[30] STK10 is closely related to several polo-like kinase (PLK) kinases.[3,26] The protein can
This article is protected by copyright. All rights reserved.

ChemMedChem
10.1002/cmdc.201900521
associate with and phosphorylate polo-like kinase 1, and over-expression of a kinase-dead version of
the protein interferes with normal cell cycle progression.[26] The kinase can also negatively regulate
interleukin 2 expression in T-cells via the mitogen activated protein kinase kinase 1 pathway.[31]
STK10 is also known as lymphocyte-oriented kinase (LOK), and has been identified to prevent
ezrin, radixin, and moesin (ERM) phosphorylation during apoptosis. These proteins play a crucial role
linking the plasma membrane to actin filaments.[32] In lymphocytes, ERM phosphorylation contributes
to cell rigidity and to the maintenance of microvilli.[33] This causes restricted formation of microvilli at
the apical surface and disrupted cell function.[34] Furthermore, broad siRNA knock down screening of
four cell lines representative of Ewing's sarcoma, an aggressive musculoskeletal tumour, has
demonstrated that STK10 is a key kinase for tumour progression.[35] STK10 was not only shown to have
a key role in proliferation of the tumours, but siRNA knock down of STK10 showed increased apoptosis
of the tumour cells.[36]
The 4-anilinoquin(az)oline shows differing kinome wide inhibition profiles in the literature. The
primary EGFR target is often accompanied by a number of secondary kinases targets including GAK,
STK10 and SLK.[5-10] We wanted to explore the possibility of chemical probe development for SLK
and STK10 in addition to GAK.[12] We now describe through a series of quantitative structure activity
relationship (QSAR) and rational design a series of diverse explorations and kinase profiles on the 4-
anilinoquin(az)oline scaffold.
RESULTS
We first selected a set of diverse literature 4-anilinoquin(az)oline based kinase inhibitors that
covered a cross section of chemical space (Figure 1). The compounds were screened for activity on the
kinase domains of the NAK sub-family (GAK, AAK1, BMP2K, and STK16) and Ste20 family of
serine/threonine protein kinase (STK10 and SLK) using a time-resolved fluorescence energy transfer
(
TR-FRET) binding displacement assay in a 16-point dose-response format to determine the inhibition
) (Table 1 and Table S1).[13] The compounds were then evaluated to assess their in-cell
constant (K
i
activity on GAK in a nanoBRET cellular target engagement assay. This was done in an eight-point
dose–response format to determine the cellular binding potency (IC50). The nanoBRET evaluated the
compounds ability to displace a fluorescent tracer molecule from the ATP binding site of a transiently
This article is protected by copyright. All rights reserved.

ChemMedChem
10.1002/cmdc.201900521
expressed full length GAK protein N-terminally fused with NanoLuc luciferase (Nluc) in HEK293 cells.
In the absence of compound, the tracer molecule and fusion protein were in proximity and able to
generate an observable bioluminescence resonance energy transfer (BRET) signal.[37]
Screening of literature inhibitors to assess NAK and SLK/STK10 activity
Table 1. Screening results of literature inhibitors.
Binding displacement assay K
(μM)a
NanoBRETb
i
Name
GAK
0.29
SLK
0.21
6.9
STK10
0.063
2.8
0.59
0.22
0.34
0.086
2.0
GAK IC50 (μM)
Neratinib
Afatinib
Dacomitinib
Pelitinib
Gefitinib
Vandetanib
Lapatinib
Erlotinib
1
2.3
0.27
0.049
0.012
0.41
3.8
>5
0.88
4.2
0.061
0.0070
0.0035
0.0065
0.13
1.1
0.28
0.59
0.16
6.0
0.98
0.0067
0.10
0.17
1.2
0.098
0.72
a
b
TR-FRET-based ligand binding displacement assay (n = 1); NanoBRET assay IC50 (μM) (n = 2, all values
within 2 standard deviations of the mean average); TR-FRET for NAK family other members (AAK1, BMP2K,
STK16) see Table S1.
We found neratinib to be relatively potent on our main targets of interest GAK, SLK and STK10,
with a 3-fold preference towards STK10. There was micromolar activity on the other NAK family
members (Table S1). Despite the low GAK K = 290 nM, the GAK nanoBRET data demonstrated good
i
potency at IC50 = 2.3 μM, less than a 10-fold right shift. Afatinib, another covalent EGFR inhibitor, was
found to be a potent GAK inhibitor with a 3-fold increase and showed some selectivity over the other
5
kinases screened. Dacomitinib was found to be single digit nanomolar inhibitor of GAK with a potent
GAK nanoBRET (IC50 = 49 nM). Pelitinib, from the same class of covalent EGFR inhibitors with a 3-
cyano quinoline hinge binder, was the most potent GAK inhibitor screened from the literature
compounds, with an in-cell target engagement of IC50 = 12 nM. Pelitinib was also a good inhibitor of
other NAK family members BMPK2 (K
SLK/STK10 K = 280 nM and 220 nM respectfully. Interestingly gefitinib was shown to have near
equipotency in GAK K compared with pelitinib, but a corresponding 34-fold drop off in the GAK
i
= 200 nM) (Table S1), AKK1 (K = 570 nM) (Table S1) and
i
i
i
nanoBRET compared to 3-fold with pelitinib. This highlights the important of combining both binding
assay data and in-cell data to get a more complete compound profile. Gefitinib did however show the
This article is protected by copyright. All rights reserved.

ChemMedChem
10.1002/cmdc.201900521
same off targets as with the covalent EGFR inhibitors, with GAK/SLK/STK10 all inhibited below K
μM. However, gefitinib showed only limited activity on AAK1/BMPK2 despite having the same
substitution pattern on the aniline portion of the compound as pelitinib (Table S1).
i
=
1
Vandetanib has a similar profile to neratinib, showing a preference for STK10 over GAK with
potencies in the triple digit nanomolar range for SLK/STK10/GAK and limited activity on the other
targets screened (Table S1). Lapatinib showed a drop off in activity on all 6 targets; this is likely due
to the para-substituted benzyl on the aniline portion of the molecule accessing a deeper pocket within
the ATP binding site of EGFR.[38] Interestingly this is not the case with neratinib that has a similar
benzyl substitution but with a 2-pyridyl instead of a meta-fluoro and a meta-chloro on the parent aniline
ring system (Figure 1). Erlotinib was shown to be a potent inhibitor of nearly all 6 targets, only STK16
showing very limited binding (K
i
= 42 μM). We observed potent inhibition of GAK at K
i
= 6.7 nM,
with SLK/STK10/AAK1 and BMPK2 all below K
i
= 400 nM. Erlotinib and gefitinib both showed a
drop off in the GAK nanoBRET, highlighting the need for comprehensive data collection. Surprisingly
the erlotinib quinoline (1) showed reduced GAK activity over erlotinib, highlighting the importance of
the solvent exposed 6,7-substitution pattern within this scaffold.
Synthesis of analogs to probe NAK and SLK/STK10
We synthesized the compounds 1-113 to probe the NAK/SLK/STK10 inhibitor landscape through
nucleophilic aromatic displacement of 4-chloroquin(az)olines (Scheme 1).[11-13] This enabled us to
furnish the products in good-excellent overall yields (52-95 %). Compounds with the powerful electron
withdrawing trifluoromethyl group on the quinazoline (30, 35, 61 & 66), all had lower yields (21-23
%
). The 6,7-difluoro (57) was also low yielding (21 %), in addition to the 7-carbonitrile (67) (21 %)
and the 6-carbonitrile (68) (20 %) analogs. This was rounded off by two other lower yielding analogs
07 (45 %) and 63 (44 %) both having a sub-optimal yields, likely due to the electronics of the ring
systems involved.
1
This article is protected by copyright. All rights reserved.

ChemMedChem
10.1002/cmdc.201900521
Scheme 1. General synthetic procedure.
We first screened three quinolines (4-6) with differing meta-substituted head groups and a fixed 6,7-
dimethoxyquinoline core (Table 2, Table S2). We found the GAK data to be consistent with previous
reports,[13] but with the new cross-screening against SLK/STK10 we observed that kinase selectivity
was able to be modulated towards SLK/STK10. The meta-alkyne (4) showed increased selectivity
towards SLK/STK10 over the corresponding bromo (5) and methoxy (6) substitution.
Table 2. Initial screening results of meta-substituted head groups.
Binding displacement assay K
i
(μM)a
Nano-BRETb
GAK IC50 (μM)
0.042
Name
R
Ratioc
GAK
0.0035
0.0025
0.0094
SLK
0.38
5.8
STK10
0.18
4
5
6
C≡C
Br
OMe
51
920
1100
2.3
10
b
0.51
0.077
32
a
TR-FRET-based ligand binding displacement assay (n = 1); NanoBRET assay IC50 (μM) (n = 2, all values
c
within 2 standard deviations of the mean average); Selectivity ratio of NAK family members (AAK1, BMP2K,
GAK and STK16) over STK10/SLK (See Table S2 for TR-FRET of AAK1, BMP2K, STK16).
One of the major kinase targets of the 4-anilinoquin(az)oline scaffold is GAK. If we were going to
get selectivity over GAK for SLK/STK10, we first sought to investigate a known GAK inhibitor
quinoline template. This enabled the development of an inhibition landscape and allowed for design
deviations on the 4-anilinoquin(az)oline template (Table 3, Table S3).
This article is protected by copyright. All rights reserved.

ChemMedChem
10.1002/cmdc.201900521
Table 3. Screening results of GAK inhibitor N-(3,4,5-trimethoxyphenyl)quinolin-4-amine template.
Binding displacement assay K
i
(μM)a
NanoBRETb,c
Name
R1
R2
Ratiod
GAKa
0.040
SLK
>55
5.1
26
STK10
27.8
45.9
2.7
GAK IC50 (μM)
0.73
7
8
9
H
F
H
H
F
700
900
390
1500
1500
1500
1200
1100
920
130
670
160
78
0.0057
0.0069
0.00068
0.0031
0.0038
0.0039
0.0015
0.013
0.27
F
0.14
1
0
Cl
Br
I
H
H
H
H
H
H
H
F
11
1.0
0.21
2
34
4.6
0.048
0.11
1
1
1
1
1
1
1
1
1
2
2
2
1
2
3
4
5
6
7
8
9
0
1
2
5.8
33
24.3
4.6
CF
3
0.17
CN
19
1.7
0.36
OMe
>55
23
12
0.36
SO
2
Me
0.018
2.4
2.9
H
0.0024
0.0013
0.0049
0.0044
0.0062
0.0027
0.0026
0.00054
17
1.6
0.54
H
Cl
Br
I
2.1
2.7
1.7
3.2
1.1
13
0.21
0.38
0.21
0.37
0.13
1.3
0.25
H
0.27
H
0.074
0.52
48
H
CF
3
60
H
CN
0.35
48
H
OMe
OMe
0.44
500
2000
3
OMe
7.1
1.1
0.026
a
b
c
Consistent with previous reports, TR-FRET-based ligand binding displacement assay (n = 1); NanoBRET
d
assay IC50 (μM) (n = 2, all values within 2 standard deviations of the mean average); Selectivity ratio of NAK
family members (AAK1, BMP2K, GAK and STK16) over STK10/SLK (See Table S3 for TR-FRET of AAK1,
BMP2K, STK16).
The simple unsubstituted quinoline (7) showed only limited STK10 binding (K = 28 μM) and no
i
SLK binding. The introduction of a fluorine to the 6-position of the quinoline (8) provided a boost in
binding to SLK, but also led to a corresponding increase in GAK binding so the ratio of
(
NAK)/(SLK/STK10) increased. The introduction of a second fluoro in the 7-position (9) switched the
binding preference between SLK and STK10 but did lead to an almost 3-fold reduction in the ratio
between NAK and SLK/STK10. However, the progression of increasingly large mono-substituted 6-
position halogens: chloro (10), bromo (2), iodo (11) and trifluoromethyl (12) yielded a static ratio at
around 1500-fold selective towards the NAK family. The switch to the 6-position cyano (13) and
methoxy (14) showed minimal improvements to the selectivity ratio. The larger methylsulfone (15) was
significantly weaker on GAK compared to the rest of the series, but still maintained a similar
SLK/STK10 binding ratio.
This article is protected by copyright. All rights reserved.

ChemMedChem
10.1002/cmdc.201900521
The 7-position halogens (16-20) demonstrated an improved ratio over the 6-position with the
halogens larger than the chlorine, showing consistent double-digit ratios. This was also the case with
the 7-position cyano (21). The 7-position methoxy (22) showed a 10-fold drop off for STK10 compared
with 21 which pushed the ratio of NAK vs SLK/STK10 above 500. The 6,7-dimethoxy quinoline (3),
showed the worst ratio in the series (>2000), due in part to the picomolar potency on GAK (K
pM).
i
= 540
The previous NAK family data on the 4-anilinoquinoline and 4-anilinoquinazoline cores suggested
1
3
that the quinazoline was a weaker binder of GAK with certain anilines. We observed that this was not
necessarily the case with the literature compounds (Table 1, Table S1). We decided to explore this as
an option to maintain SLK/STK10 potency while dialing back the GAK binding (Table 4, Table S4).
While the unsubstituted N-(3,4,5-trimethoxyphenyl)quinazolin-4-amine (23) and 6-position methyl
(
24) demonstrated almost equi-potent GAK binding; however the ratio between NAK vs SLK/STK10
did show some widening. The introduction of a 6-position fluoro (25) had only a limited effect on the
activity profile, but the introduction of a second fluorine at the 7-position (26) did reduce the GAK
activity by 3-fold and led to an almost 4-fold increase in the selectivity ratio; and interestingly removed
all SLK affinity. The introduction of larger halogens in the 6-position (27-30) show potent GAK
profiles, with only the trifluoromethyl (30) showing a similar profile to 6-fluoro (25). In addition, 30-
3
2 showed increased binding to STK16 (K = 2-3 μM) (Table S4).
i
The switch to the 7-position halogen substitution (31-35) yielded a slightly weaker GAK profile,
consistent with the quinolines 7-position halogens (16-20). The 7-fluoro (31) appeared too small to
impact the binding, but the 7-chloro (32) to 7-iodo (34) showed progressively better NAK vs
SLK/STK10 ratios down to 10. The 7-cyano substitution (36) proved to be a further improvement due
in part to the significant reduction in GAK binding to K = 210 nM and complete obliteration of other
i
NAK family members (Table S4). However, 36 did not have the same potency against STK10 observed
with the 7-position halogens (32-34).
This article is protected by copyright. All rights reserved.

ChemMedChem
10.1002/cmdc.201900521
The switch to the 6-cyano (37) yielded an increase in GAK binding by 2-fold, with the 6-
methylsulfone (38) showing a similar overall 6 kinase profile but with a 6-fold drop compared with 37.
While the 6-methoxy (39) showed a jump of 5-fold GAK binding compared with 37, the 6,7-dimethoxy
(
40) showed a 15-fold increase in GAK binding and a jump in GAK target engagement. However, with
a K = 200 nM on STK10 the overall ratio showed only a small improvement, but significantly better
than the quinoline (3). The 7-methoxy (41) reversed the trend observed with the 7-position halogens
34-37) with an improved GAK binding profile. The trimethoxyaniline erlotinib quinazoline (42)
i
(
showed a 10-fold drop compared with 40 and a 4-fold drop compared with erlotinib itself. However,
the ratio between NAK and SLK/STK10 was an almost 100-fold improvement towards STK10/SLK
compared with 3.
Table 4. Screening results of trimethoxyphenylquinazoline template.
Binding displacement assay K
i
(μM)a
NanoBRETb
Name
R1
R2
Ratioc
GAK
0.034
0.022
0.033
0.093
0.017
0.010
0.011
0.037
0.055
0.030
0.025
0.021
0.079
0.21
SLK
50
STK10
6.0
GAK IC50 (μM)
2
2
2
2
2
2
2
3
3
3
3
3
3
3
3
3
3
4
4
4
3
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
H
Me
F
H
H
H
F
4.6
2.1
3.6
>5
180
230
280
76
25
5.1
9.4
>55
47
11
F
7.1
Cl
Br
I
H
H
H
H
F
7.0
2.4
1.8
0.89
4.5
>5
410
380
370
110
120
27
21
3.8
18
4.1
CF
3
25
4.0
H
44
6.7
H
Cl
Br
I
3.8
2.5
1.8
8.2
10
0.81
0.43
0.21
1.2
3.2
4.2
2.4
>5
H
17
H
10
H
CF
CN
H
3
15
H
1.6
>5
7.6
73
CN
0.099
0.60
44
7.2
>5
SO
2
Me
H
44
7.4
>5
12
OMe
OMe
H
H
0.022
0.0014
0.012
0.026
18
4.3
3.4
0.35
1.6
3
200
140
130
22
OMe
OMe
2.1
7.6
6.8
0.2
1.5
6,7-(OCH
2
CH
2
OMe)
2
0.58
a
b
TR-FRET-based ligand binding displacement assay (n = 1); NanoBRET assay IC50 (μM) (n = 2, all values
within 2 standard deviations of the mean average); Selectivity ratio of NAK family members (AAK1, BMP2K,
c
GAK and STK16) over STK10/SLK (See Table S4 for TR-FRET of AAK1, BMP2K, STK16).
This article is protected by copyright. All rights reserved.

ChemMedChem
10.1002/cmdc.201900521
The cone angle of the quinoline/aniline amine bond is around 60 degrees while the quinazoline is
almost flat.[13] The 3-cyanoquinoline with the projected cyano in the arched region of the molecule
should push the aniline into an almost perpendicular position with respect to the quinoline. This
combined with the new electronic conformation provides for a new chemical space to seek the
separation of activities between the three main kinases (GAK, SLK & STK10) of interest.
The unsubstituted trimethoxyphenylquinoline-3-carbonitrile (43) showed a halving in activity on
GAK compared with quinazoline (23) and quinoline (7) (Table 5). There was no improvement on
SLK/STK binding, meaning the ratio between GAK and SLK/STK10 was static. The introduction of a
larger halogen in the 6-position (44-46) boosted GAK binding and GAK in-cell target engagement, with
negligible SLK/STK10 activity. The 6-methoxy (47) showed a 3-fold boost in the GAK nanoBRET
compared to 43, but no selectivity ratio improvement.
The 6,7-dimethoxy (48) is the most potent GAK nanoBRET result observed to date. Improvement
was observed across all kinases screened, with SLK/STK10 and AAK1 all showing binding below K
i
=
600 nM. The 6-methylsulfone (49) showed a drop on GAK binding, but also a reducing of binding
across all six kinases screened. 49 was still the lowest ratio observed between NAK vs SLK/STK10.
There was no difference when switching to the 7-position halogens (50-52) on the GAK binding
observed. However other NAK family members (AAK1, BMPK2 & STK16) were inactive, while
STK10/SLK showed activity mostly below 1 μM. The progression of increasing size from fluorine to
iodine did, however, demonstrate an increase in selectivity towards SLK/STK10 over the NAK family.
This highlighted that selectivity maybe achievable on this template. The fragility of this selectivity was
demonstrated by the switch between 7-iodo (52) to 7-methoxy (53) with a reversal of selectivity towards
GAK and a reduction in selectivity ratio to above 500.
This article is protected by copyright. All rights reserved.

ChemMedChem
10.1002/cmdc.201900521
Table 5. Screening results of trimethoxyphenylquinoline-3-carbonitrile template.
Binding displacement assay K
i
(μM)a
NanoBRETb
Name
R1
R2
Ratioc
GAK
0.0088
0.0038
0.0023
0.0048
0.0024
0.00085
0.21
SLK
7.7
53
9.4
9.7
3.2
0.57
1.2
STK10
4.7
26
11
4.2
1.9
0.42
1.5
GAK IC50 (μM)
0.33
4
4
4
4
4
4
4
5
5
5
5
3
4
5
6
7
8
9
0
1
2
3
H
Cl
Br
H
H
H
H
H
OMe
H
Cl
Br
I
OMe
530
6700
4100
880
790
140
6
120
150
38
0.24
0.20
0.26
0.092
0.006
>5
0.47
0.14
I
OMe
OMe
SO
2
Me
H
H
H
H
0.0066
0.0029
0.0052
0.0020
1.4
0.79
0.44
0.20
1.2
0.71
0.49
2.1
0.20
0.078
580
a
b
TR-FRET-based ligand binding displacement assay (n = 1); NanoBRET assay IC50 (μM) (n = 2, all values
within 2 standard deviations of the mean average); Selectivity ratio of NAK family members (AAK1, BMP2K,
c
GAK and STK16) over STK10/SLK (See Table S5 for TR-FRET of AAK1, BMP2K, STK16).
Development of meta-alkyne substituted aniline analogs
In order to find more SLK/STK10 selective compounds we switch back to the original meta-alkyne
substituted aniline which demonstrated the original selectivity preference towards SLK/STK10 in 4.
Combining this with the screening knowledge gained with 4-53, we decided to pursue the quinazoline
core. This is due to the consistent 5-10-fold drop off in GAK binding/activity observed compared with
the quinoline core.
The unsubstituted analog (54) showed a reduction in GAK binding that was 3-5-fold greater that the
trimethoxy counterparts (7, 23, 43), with limited activity on SLK/STK10 (Table 6). The addition of a
methyl in the 6-position (55) provided a 4-fold improvement in GAK binding with a limited
improvement in SLK/STK10 binding. A single fluorine in the 6-position (56) showed no significant
change in profile across the kinases screened. However, addition of a second fluorine in the 7-position
(
57) did boost the SLK/STK10 while maintaining GAK binding; bringing the ratio of GAK vs
SLK/STK10 down to 34. Increasing the size of the 6-position halogen (58-60) above fluorine increased
the GAK binding while maintaining the SLK/STK10 in place, producing an improvement in selectvity.
This article is protected by copyright. All rights reserved.

ChemMedChem
10.1002/cmdc.201900521
Increasing to the trifluoromethyl (61) reduced the GAK binding by more than 5-fold, maintaining the
SLK/STK10 activity and pushing the ratio down to just above 20.
Switching to the 7-position halogens (62-66) yielded an improvement in the ratio of the NAK vs
SLK/STK10. The 7-fluoro (62) was too small to significantly impact the profile, however increasing
the size improved the GAK vs SLK/STK10 profile to single digit ratios. The 7-iodo (65) provided the
best STK10 binding with a K
i
= 150 nM with a ratio to GAK of 9-fold. The 6-cyano (67) provided
= 100
potential route to a selective SLK/STK10 inhibitor with a ratio of just 2.6 and GAK binding of K
i
nM. Switching to the 7-position cyano (68) proved to be good at reducing GAK binding but also
knocked SLK/STK10 back to the low micromolar range.
Table 6. Screening results of meta-acetylene quinazoline template.
Binding displacement assay K
i
(μM)a
NanoBRETb
Name
R1
R2
Ratioc
GAK
0.17
SLK
23
STK10
10
GAK IC50 (μM)
5
5
5
5
5
5
6
6
6
6
6
6
6
6
6
6
7
7
7
7
7
4
5
6
7
8
9
0
1
2
3
4
5
6
7
8
9
0
1
2
3
4
H
Me
F
H
H
H
F
H
H
H
H
F
4.5
3.6
2.0
>5
>5
>5
>5
>5
4.1
>5
>5
61
60
79
34
93
89
84
21
55
10
8.4
9
11
2.6
28
3.1
92
120
52
79
49
15
0.043
0.070
0.085
0.030
0.028
0.044
0.150
0.022
0.035
0.031
0.020
0.088
0.10
4.3
8.5
3.8
3.9
3.4
4.3
3.2
1.2
0.36
0.26
0.18
1.2
0.26
7.9
2.6
5.5
2.9
2.8
2.5
3.7
3.4
1.2
0.44
0.29
0.15
1.0
0.31
4.7
F
Cl
Br
I
CF
H
3
H
H
H
H
H
CN
Cl
Br
I
>5
>5
>5
>5
CF
CN
H
3
0.17
SO
2
Me
H
H
OMe
OMe
0.42
1.3
1.3
>5
OMe
OMe
H
OCH
CH OCH
6,7-(OCH CH
0.024
0.0010
0.0088
0.019
0.035
0.0067
2.8
2.2
2.9
0.079
1.2
1.7
3.2
0.88
0.14
0.67
2.7
0.12
0.46
1.5
1.7
0.10
2
O
2
2
3.6
Erlotinib
2
2
OMe)
2
0.17
a
b
TR-FRET-based ligand binding displacement assay (n = 1); NanoBRET assay IC50 (μM) (n = 2, all values
c
within 2 standard deviations of the mean average); Selectivity ratio of NAK family members (AAK1, BMP2K,
GAK and STK16) over STK10/SLK (See Table S6 for TR-FRET of AAK1, BMP2K, STK16).
This article is protected by copyright. All rights reserved.

ChemMedChem
10.1002/cmdc.201900521
The 6-methylsulfone (69) reduced the GAK binding to K
the affinity to just above 1 μM for both SLK/STK10. This brought the ratio down to the same level as
7 while reducing the GAK binding selectively. The switch to the 6-methoxy (70) increased GAK
binding by 18-fold while maintaining SLK/STK10 binding at similar levels to 69; 70 also showed
increased AAK1 binding with a K = 930 nM (Table S6). The introduction of a second methoxy in the
-position (71) provided a strong template for GAK consistent with 3, 40 and 48 with binding at K = 1
nM. However, the ratio was not as high as previous 6,7-dimethoxy examples and potency of SLK and
STK10 was K = 140 nM and 120 nM respectively, with a corresponding increase in AAK1 and BMP2K
at K = 200 nM and 600 nM respectively. The removal of the 6-position methoxy from 71 highlights
i
= 420 nM, while simultaneously increasing
6
i
7
i
i
i
that the methoxy groups are synergetic and not additive as the 7-methoxy analog (72) was 2.5-fold more
potent on GAK than 70, with a similar drop on SLK/STK10 and AAK1 binding. The closing of the 6,7-
dimethoxy to form a 5-membered ring (73) or a 6-membered ring (74) reduced the GAK binding but
also reduced binding to other the other kinases screened. Erlotinib was not equi-potent to the 6,7-
dimethoxy (71) with an increase in selectively towards SLK/STK10 but still more potent on GAK. The
in-cell target engagement on GAK is surprisingly low considering the potent GAK K .
i
To increase selectivity towards SLK/STK we explored the 3-cyanoquinoline hinge binder in order
to manipulate the electronics and sterics of the template (Table 7). The unsubstituted analog (75)
showed good GAK binding with relatively potent SLK/STK10 potency. The introduction of
progressively larger 6-position halogens (76-79) eroded the SLK/STK10 potency, while GAK binding
was maintained. The increase in halogen size also appeared to correlate with a decrease in-cell target
engagement despite only a limited change in binding. The introduction of a 6-methylsulfone (80) lead
to an overall decrease in binding with a drop of 10-fold in GAK binding compared to 75.
An affinity well was observed around the methoxy substitutions (81-83). These analogs showed a
high potency across most of the kinases except STK16 (Table S7). The 6,7-dimethoxy (82) was the
most potent GAK binder observed to date with picomolar affinity (K
to an equivalent high GAK in-cell target engagement value, despite 82 still being highly active (IC50
0 nM). The introduction of 7-position halogens (84-86) boosted the SLK/STK activity while keeping
i
= 320 pM), this did not translate
=
4
This article is protected by copyright. All rights reserved.

ChemMedChem
10.1002/cmdc.201900521
the GAK binding static. This culminated in the 7-iodo compound (86) having a potency on SLK/STK10
of K = 75 nM and 80 nM respectively and a NAK vs SLK/STK10 ratio of just under 4.
i
Table 7. Screening results of meta-acetylene quinazoline template.
Binding displacement assay K
i
(μM)a
NanoBRETb
Name
R1
R2
Ratioc
GAK
0.018
0.026
0.016
0.017
0.020
0.18
SLK
0.35
0.54
0.71
1.2
STK10
0.34
0.58
0.72
1.1
GAK IC50 (μM)
7
7
7
7
7
8
8
8
8
8
8
8
5
6
7
8
9
0
1
2
3
4
5
6
H
F
H
H
1.5
3.8
3.2
4.4
>5
19
21
Cl
Br
I
H
44
H
65
H
1.1
1.0
50
SO
2
OMe
OMe
H
Me
H
>55
42
>5
230
43
H
0.0063
0.00032
0.0035
0.028
0.027
0.019
0.27
0.018
0.062
0.18
0.16
0.075
0.31
0.027
0.073
0.21
0.13
0.080
0.59
0.040
0.29
>5
OMe
OMe
Cl
Br
I
56
18
H
6.4
4.9
3.9
H
>5
H
>5
a
b
TR-FRET-based ligand binding displacement assay (n = 1); NanoBRET assay IC50 (μM) (n = 2, all values
c
within 2 standard deviations of the mean average); Selectivity ratio of NAK family members (AAK1, BMP2K,
GAK and STK16) over STK10/SLK (See Table S7 for TR-FRET of AAK1, BMP2K, STK16).
Optimization of the aniline head group
Four compounds (60, 65, 67, 71) with the some of the best potency and selectivity towards
SLK/STK10 were chosen for further optimization within the meta-acetylene aniline head group. GAK
is known to be sensitive to aniline modifications particularly in the ortho- and para-position.[38] We first
selected, the 7-cyano quinazoline (67), and installed several small halogens (87-90) on the aniline
portion of the molecule. The three fluorine derivatives (87-89) showed a selectivity ratio improvement
whereas the 4-chloro analog (91) did not. The addition of a fluoro in the 6-position of the aniline ring
(
87) led to a 120-fold reduction in GAK binding, with maintenance of activity of both SLK and STK10.
Compound 87 had the highest selectivity ratio of any compound synthesized in the current work with a
ratio below 1, where GAK binding was K = 1.2 μM and SLK/STK10 binding was K = 560 and 670
i
i
nM respectively. The analogs with the fluoro in the 2-position (88) and the 4-position (89) showed
slightly reduced overall binding to the kinases screened. There was very limited activity in the wider
NAK family for any of these analogs (67, 87-90) (Table S8).
This article is protected by copyright. All rights reserved.

ChemMedChem
10.1002/cmdc.201900521
Table 8. Screening results of optimized meta-acetylene quinazolines (60, 65, 67, 71).
Binding displacement assay K
i
(μM)a
NanoBRETb
Ratioc
Name
R1
R2
R3
R4
R5
GAK
0.010
1.2
SLK
0.26
0.56
1.1
STK10
0.31
0.67
0.50
0.87
1.6
GAK IC50 (μM)
6
8
8
8
9
6
9
9
9
9
6
9
9
9
9
7
9
7
7
8
9
0
5
1
2
3
4
0
5
6
7
8
1
9
H
CN
CN
CN
CN
CN
I
H
F
H
H
F
H
H
H
F
>5
>5
2.6
0.47
1.5
1.2
5
H
H
H
H
H
H
F
0.33
>5
H
H
H
H
H
F
0.55
0.69
6.3
>5
H
Cl
H
H
H
F
0.32
>5
H
0.020
0.11
0.18
0.39
0.86
0.11
0.93
4.3
0.15
0.25
0.27
0.11
0.24
3.7
>5
9
H
I
>5
2.3
2.9
3.1
2.6
84
H
I
H
H
H
H
F
0.093
0.035
0.092
0.044
0.19
>5
H
I
H
H
H
H
F
>5
H
I
Cl
H
H
H
F
>5
I
I
H
>5
H
4.5
4.9
>5
24
I
H
H
H
H
H
F
0.29
18
6.7
>5
23
I
H
H
H
H
H
F
0.068
0.14
1.6
0.87
1.0
>5
13
I
H
Cl
H
H
H
F
7.1
>5
7.1
120
38
OMe
OMe
OMe
OMe
OMe
OMe
OMe
OMe
OMe
OMe
0.0010
0.0058
0.0063
0.0024
0.0034
0.14
0.28
0.67
0.072
0.33
0.12
0.22
0.25
0.047
0.075
0.079
0.39
0.40
0.14
0.39
1
1
1
00
01
02
H
H
H
40
H
H
20
Cl
22
a
b
TR-FRET-based ligand binding displacement assay (n = 1); NanoBRET assay IC50 (μM) (n = 2, all values
c
within 2 standard deviations of the mean average); Selectivity ratio of NAK family members (AAK1, BMP2K,
GAK and STK16) over STK10/SLK (See Table S8 for TR-FRET of AAK1, BMP2K, STK16).
The 7-iodo quinazoline (65) showed some potency towards SLK/STK10 with a promising selectivity
ratio. The small halogen modifications (91-94) yielded compounds with improved selectivity ratios
primarily modulated through the GAK binding affinity. Interestingly, the 4-chloro (94) was still
relatively active against GAK, unlike with 90. There was no improvement in SLK/STK10 potency and
no activity in the wider NAK family (Table S8). The corresponding 6-iodo derivative (60) had a larger
selectivity deficit towards SLK/STK10, however this was still able to be reduced by the introduction of
halogens into the aniline portion of the molecule (95-98). The 4-fluoro (97) was able to drive down
affinity for STK10 to K
i
= 870 nM with a ratio to NAK of 13. The 4-chloro (98) was able to improve
= 1.0 μM.
that ratio with a slight drop in potency on STK10 to K
i
This article is protected by copyright. All rights reserved.

ChemMedChem
10.1002/cmdc.201900521
The 6,7-dimethoxyquinazoline (71) is one of the better ATP competitive kinase inhibitor substitution
patterns on this template. Despite the potent GAK profile and triple digit NAK vs SLK/STK10 ratio,
the halogen substitutions (99-102) proved to be successful in driving down the selectivity and potency
on SLK/STK10. The ratios were improved by increasing potency on SLK/STK10 with the 4-fluoro
(
101) being the most potent binder on SLK/STK10 with a K = 72 nM and 47 nM respectively with a
i
ratio of 20.
Halogenated aniline analogs
Finally, to map the chemical space and explore the halogen substitutions further we synthesized a series
of 6-bromo quinoline and quinazolines derivatives with differing halogenated substitution patterns
(
103-113). While binding was maintained almost consistently on GAK, affinity towards SLK/STK10
was limited (Table 9). The bulkier substitutions, 3,4,5-trifluoro (103), 3,4-difluoro (104) and 3,5-
difluoro (105) all showed between K = 10-25 nM binding to GAK but no activity on the other five
i
kinases (Table S9). Movement of the second fluorine around the ring (106-107) maintaining one
fluorine in the 6-position meant a small increase in binding across the NAK-family but limited
improvement for SLK/STK10. The mono-substituted fluoro compounds (108-111) had surprisingly
similar properties, but the 4-fluoro quinazoline did show some reduction in GAK binding and an
increase in AAK1 binding to K = 490 nM (Table S9). Switching to a bromo in the 3-position (112) did
i
not immediately improve the NAK vs SLK/STK10 selectivity ratio, but it did remove all activity across
the wider NAK family. A switch to the quinazoline (113) increased binding to SLK/STK10, brought
the ratio down to double digits. However, the increases in affinity across the NAK family made this
ratio not a true reflection of overall selectivity improvement in this case.
This article is protected by copyright. All rights reserved.

ChemMedChem
10.1002/cmdc.201900521
Table 9. Screening results of halogenated quinolines and quinazolines.
Binding displacement assay K
i
(μM)a
NanoBRET
Ratioc
Name
R1
R2
R3
R4
X
GAK
0.023
0.011
0.012
0.035
0.015
0.015
0.12
SLK
>55
>55
>55
38
STK10
>55
>55
>55
11
GAK IC50 (μM)b
1
1
1
1
1
1
1
1
1
1
1
03
04
05
06
07
08
09
10
11
12
13
H
H
H
F
F
H
F
F
F
F
F
C-H
C-H
C-H
C-H
C-H
C-H
N
4.4
>5
>2400
>5000
>4600
310
H
F
F
>5
H
H
H
H
F
H
F
2.7
0.87
0.94
>5
F
H
F
>55
36
>55
21
3700
1400
460
H
H
H
F
H
H
H
H
H
H
F
>55
>55
>55
>55
9.4
>55
>55
30
H
H
H
H
C-H
C-H
C-H
N
0.019
0.049
0.072
0.084
1.9
2.6
>5
2900
610
H
Br
Br
H
H
>55
6.0
>760
71
>5
a
b
TR-FRET-based ligand binding displacement assay (n = 1); NanoBRET assay IC50 (μM) (n = 2, all values
c
within 2 standard deviations of the mean average); Selectivity ratio of NAK family members (AAK1, BMP2K,
GAK and STK16) over STK10/SLK (See Table S9 for TR-FRET of AAK1, BMP2K, STK16).
GAK K vs GAK nanoBRET
i
We measured GAK affinity of the 4-anilinoquin(az)oline analogues with a wide range of nanoBRET
values and affinity over several orders of magnitude in the TR-FRET binding assay. The two binding
2
assays were highly correlated, as shown by a plot of IC50 (nM) versus log K
i
(nM) which has R =0.62
(
Figure 2). There are always exceptions, but the effective cut off for GAK target engagement in the
nanoBRET (IC50 <5 μM) appears roughly correlated to a GAK K = 20 nM or below.
i
This article is protected by copyright. All rights reserved.

ChemMedChem
10.1002/cmdc.201900521
Figure 2. Comparison of GAK K
Small Molecule Crystal Structures
i
vs GAK IC50 nanoBRET (all compounds IC50 <5μM) .
In order to better understand how the structural characteristics of the compounds might influence the
kinase activity profile we solved a series of small molecule crystal structures (Table S10. Figure 3).
All the compounds apart from 44, 50, 53 and 67 crystallized as chloride salts. Each of the compounds
have two strong hydrogen bond donors so in the solid-state typically form hydrogen bonded tapes with
the chloride ion as the acceptor. The structures which include solvent water (7, 11, 15-17, 20, 24, 27,
34, 36, 39-40, 43, 45-46, 53) display more complex hydrogen bond networks. We observed that the
o
o
quinolines (7, 11, 15-20, 22) had a narrow range of C-N-C-C torsion angles of 46 -71 . This resulted in
o
o
two narrow distributions of plane angles between 48 -74 and 115-131 arising from the restricted
rotation of the 4-anilinoquinoline, which is generally out of plane due to the steric interaction between
the C-3 C-H and the aniline ring. The quinolines (7, 15-17, and 22) generally packed in the triclinic P-
1
space group, although 11, 18 and 20 were monoclinic.
This article is protected by copyright. All rights reserved.

ChemMedChem
10.1002/cmdc.201900521
Figure 3. x-ray crystal structures of a set of quin(az)olines (7, 17, 23, 32, 43, 50) exemplifying the
plane angles observed.
The quinazolines (24, 27-29, 31-34, 37, 67) generally packed in the triclinic crystal system, however
3, 25, 36, 40 and 68 were monoclinic and 39 was orthorhombic. The flat planarity of the ring system
2
allows for a wider variance in crystal packing. Interestingly, 23 had a bigger twist in structure at 38o,
o
but this only resulted in a plane angle of 40 . However, 39 demonstrated a disruption that pushed the
aniline ring system out of plane like the quinoline structures (7, 11, 15-20, 22) with a C-N-C-C torsion
o
o
angle of 50 degrees and a plane angle of 132 . Additionally the meta-acetyl quinazoline, 67, crystallizes
with a C-N-C-C torsion angle of 29 and a plane angle of 29. The rest of the quinazolines 23-25, 27-
o
o
2
9, 31-37, 39-40 were near planar (1.4 -7.9 ). The 3-cyanoquinoline derivatives (43, 45, 46, 50, 53)
were monoclinic and 44 orthorhombic, with no triclinic packing examples observed. The C-N-C-C
o
o
o
o
torsion angle range between 44 -51 with plane angles ranging between 60 -122 . The projected C-3
cyano group projects out pushing the aniline into an orthogonal orientation.
GAK QSAR Model
We constructed an explanatory field-based 3D QSAR model, which would agree with structure
activity relationships of reported structures, as well as with recently released x-ray structures and
This article is protected by copyright. All rights reserved.

ChemMedChem
10.1002/cmdc.201900521
molecular docking simulations.[39-40] The 3D-QSAR model was constructed so that the main fields could
be overlaid with co-crystallized ligands and favorable docking poses. WaterMap simulations were
computed for active site of GAK (PDB:5Y80) as solvated and with presence of selected active ligands
(
co-crystal and 82).[40] Docking suggests that the overall shape of the current ligand series aligns well
with the kinase back pocket, and fully fills typical hydrogen bond interaction with hinge residues
Figure 4).
(
Figure 4. Docking of quinoline scaffold to GAK active site (PDB:5Y80, compound 3).
The quinoline scaffold is able to form a hydrogen bond interaction to main chain amide of C126 and
two additional aromatic hydrogen bond contacts to main chain carbonyl of C126 and L125. The co-
crystalline structures of the GAK ATP binding site show an additional lipophilic space towards the beta
sheet, next to the mainchain of L68, K69 and T123, where small substituents like methoxy or ethyne
are able to orient.[40] This pocket also contains a high energy water that appears to be displaced with
certain orientations and substituent patterns.[13,41]
We constructed 3D-QSAR models for GAK, SLK and STK10 kinases using the field based QSAR
functionality of Schrödinger Maestro (Table 10). Flexible ligand alignments were performed using a
Bemis-Mucko method,[42] enabling recognition of the largest common scaffolds using a favorable
This article is protected by copyright. All rights reserved.

ChemMedChem
10.1002/cmdc.201900521
docking pose of structurally representative derivative as alignment template. Torsion angles of larger
substituents, such as methoxy groups, were manually adjusted to fit template structures when needed.
The models were constructed by defining 80 % of compounds randomly as training sets and rest of the
compounds were considered as test set. Grid spacing of 1.0 Å extended 3.0 beyond training set limits
was used to the calculate field values. Subsequently, predicted versus measured affinities were plotted
to identify outliers (Figure 5). Compounds 80, 81 and 83 were omitted from the SLK and STK10
models as potential outliers, which increased Q2-value and decreased standard error of prediction. The
statistical metrics of models: R-square values with four PLS components were 0.7402, 0.8064 and
0.7900, for GAK, SLK and STK10 models respectively. All three models show acceptable internal
predictivities (Q-square values were 0.5967, 0.7303, 0.7631), determined by leave-one-out (LOO)
cross-validation. Stability values above 0.97 with four partial least squares regression (PLS)
components demonstrate the quality and predictivity of the model inside the studied applicability
domain was robust.
Table 10. Validation of 3D-QSAR models for GAK.
R^2
CV
2
#
Factors
SD
R
R^2 Scramble Stability
F
P
RMSE
Q^2
Pearson-r
1
2
3
4
5
6
0
.5999
.5162
.4432
.3809
.3563
.3419
0.3337
0.2685
0.4160
0.5188
0.6201
0.6624
0.6613
0.0486
0.0907
0.1265
0.1619
0.1973
0.2238
0.995
0.989
0.975
0.971
0.977
0.967
45.6
47.2
53.7
62.7
60.0
55.7
1.34e-009
9.46e-015
6.43e-020
5.91e-025
9.63e-027
1.47e-027
0.49
0.49
0.46
0.45
0.44
0.46
0.5254
0.5244
0.5662
0.5967
0.6036
0.5746
0.7440
0.7301
0.7537
0.7973
0.7971
0.7963
0
0
0
0
0
0.5121
0.6443
0.7402
0.7753
0.7954
Table 11. Validation of 3D-QSAR models for SLK.
R^2
Scramble
2
#
Factors
SD
R
R^2 CV
Stability
F
P
RMSE
Q^2
Pearson-r
1
2
3
4
5
6
0.5877
0.4520
0.3797
0.3448
0.3337
0.3175
0.4180
0.6595
0.7625
0.8064
0.8208
0.8397
0.3563
0.6164
0.7229
0.7284
0.7423
0.7471
0.0448
0.995
0.997
0.997
0.987
0.985
0.974
63.9
85.2
93.1
89.6
77.8
73.3
4.46e-012
0.63
0.54
0.46
0.41
0.45
0.48
0.3679
0.5467
0.6588
0.7303
0.6843
0.6427
0.6149
0.7532
0.8133
0.8569
0.8276
0.8024
0.0824
0.1125
0.1433
0.1669
0.1918
2.58e-021
4.54e-027
7.75e-030
3.05e-030
2.73e-031
Table 12. Validation of 3D-QSAR models for STK10.
R^2
RMS
E
2
#
Factors
SD
R
R^2 CV
Stability
F
P
Q^2
Pearson-r
Scramble
1
2
0.5556
0.4327
0.5875
0.7525
0.5509
0.7178
0.0627
0.998
0.998
128.2
135.3
5.36e-019
1.03e-027
0.45
0.42
0.7312
0.7604
0.8643
0.8764
0.0861
This article is protected by copyright. All rights reserved.

ChemMedChem
10.1002/cmdc.201900521
3
4
5
6
0.4092
0.4032
0.3798
0.3769
0.7813
0.7900
0.8158
0.8207
0.7409
0.7398
0.7186
0.7182
0.1226
0.1637
0.1850
0.2127
0.997
0.995
0.978
0.976
104.8
81.8
76.2
64.8
6.06e-029
1.16e-028
4.23e-030
1.25e-029
0.44
0.42
0.39
0.37
0.7469
0.8660
0.8765
0.8968
0.9075
0.7631
0.7993
0.8174
Figure 5. Scatter plots of field based 3D-QSAR models showing predicted versus measured binding
affinities (nM) of active compounds printed at logarithmic scale, from right to left GAK, STK10 and
SLK.
Placement of positive steric QSAR fields supports the role of the methoxy groups pointing to the
back pocket of the solvent exposed region and providing perfect shape complementarity (Figure 6-9).
WaterMap simulations of the GAK protein structures highlighted a high energy water (marked with
W1, Figure 10) at the top of back pocket, positioned against beta-sheet.[40] The replacement of this
water is common among high affinity GAK inhibitors compounds. This gains additional support from
hydrophobic QSAR fields (Figure 8), showing the meta-position on the aniline ring should be occupied
with this substituent. Interestingly, negative hydrophobic field is in good agreement with oxygen
positions of methoxy substituents. Taken together positive steric field (Figure 7) and negative
hydrophobic field in (Figure 8) can further explain why triple methoxy compounds are good GAK
inhibitors.
Electrostatic fields from QSAR (Figure 9) confirms that the quinoline scaffold is a better inhibitor
than the quinazoline. The quinazoline provides a negatively charged area, where a positive field is
required according to the QSAR model of GAK and leaves a favorable steric field empty (Figure 7).
The hydrogen bond acceptor fields shown in figure 10 indicate that the space next to position 7 of the
This article is protected by copyright. All rights reserved.

ChemMedChem
10.1002/cmdc.201900521
quin(az)oline would also not result in an affinity improvement, if filled with hydrogen bond acceptor,
however a nitrile group which is significantly bulkier could be tolerated.
WaterMap analysis of the GAK ATP binding site with 82 shows that water occupying site (marked
as W3, Figure 11) has less a favorable position by means of ΔG. Waters at site W3 serve as a hydrogen
bond bridge between nitrile group of 82. The water W3 has hydrogen bonding (HB) contact to water
bound close to DFG residues, but it is not able to make hydrogen bonds to the bottom back pocket.
Thus, it appears that water W3 is trapped between ligand and hydrophobic pocket forming a bridged
interplay between these residues (Figure 12). The ideal fit of 82 as ligand results from optimal shape
complementarity, high rigidity (favorable entropy upon binding) and replacement of some high energy
waters including W1.
Figure 6. Mains steric fields where green is positive and yellow negative.
Figure 7. Main Hydrophobic fields where brown is positive and gray negative.
Figure 8. Main electrostatic fields where red stands for negative and blue positive.
This article is protected by copyright. All rights reserved.

ChemMedChem
10.1002/cmdc.201900521
Figure 9. At top row HB acceptor fields where cyan is positive and magenta negative. At the bottom
row HB donor fields where blue is positive and green negative.
Figure 10. WaterMap analysis of GAK (PDB:5Y80): Showing high energy water W1 at the top of
hydrophobic ATP pocket bound with and without erlotinib.
Figure 11. WaterMap analysis of GAK (PDB:5Y80) with preserved co-crytallized ligand (gefitinib)
shows that bridged water between ligand and protein marked with W2 is stabilizing binging of
quinazoline scaffold. On right selected bridging waters taken from hydrogen bond optimized x-ray
This article is protected by copyright. All rights reserved.

ChemMedChem
10.1002/cmdc.201900521
structure shown. Interestingly, water in site W3 becomes higher in energy compared to empty binding
site (Figure 10).
Figure 12. Watermap analysis of active site of GAK (PDB:5Y80) with 82 showing that water in site
W3 is less favorable for binding. Water W3 serves as hydrogen bond bridged between nitrile group of
82 and water bound close to DFG residues, but it is not able to make hydrogen bonds patterns to the
bottom back pocket and is trapped between ligand and hydrophobic pocket.
Interestingly comparison of WaterMap sites between GAK, STK10 and SLK with representative 4-
anilinoquin(az)oline ligands suggests that the high energy water is not able to be accessed easily in the
case of SLK and STK10 (Figure 13). Indeed, according to the docking of the 4-anilinoquin(az)olines
into SLK the molecules prefer a flipped binding mode, preventing effective manipulation of the water
network seen in GAK (Figure 13).
Figure 13. Comparison of WaterMap sites between GAK, STK10 and SLK with selected ligands.
According to docking SLK binds quinoline scaffold in flipped geometry.
DISCUSSION
More than 40 drugs have been approved to target the ATP binding site, mainly in oncology. Although
these were developed with specific targets, in practice most of these drugs were multi-kinase inhibitors
that leverage the ATP pocket across multiple kinases to increase their clinical efficacy. Outside of
This article is protected by copyright. All rights reserved.

ChemMedChem
10.1002/cmdc.201900521
oncology, selective kinase inhibition profiles are required for treatment of chronic diseases with kinase
targets, including inflammation and neurodegeneration. These compounds require enhanced selectivity
profiles in order to avoid potential toxicity indications associated with broad kinome inhibition.[43-44]
New chemical approaches and molecular insight into the development of highly selective kinase
inhibitors that target the conserved ATP binding site are essential if we are to facilitate this unmet
clinical need. Binding assays are an accurate and robust method to measure potency and selectivity of
ATP-competitive kinase inhibitors.[45] Binding assays are available at several commercial vendors and
are routinely used to profile kinase inhibitors for their selectivity across the human kinome.[7] Ligand
binding displacement assays provide an accepted method for direct measurement of kinase inhibition
in drug optimization of ATP binding site inhibitors.[46] This point is particularly relevant for neglected
kinases including such as GAK, where there are currently no robust and validated enzyme activity
assays.
Despite the original target for the 4-anilino-quin(az)oline scaffold being EGFR, SLK/STK10/GAK
have been shown to be consistent tractable collateral targets for inhibition.[6-8] The levels of activity
vary for compound to compound and there is not a linear relationship. Erlotinib and vandetanib kinome
profiles suggest that there is a potential to find a route to separate these activities. We found that the
structural features of the scaffold were key: 1) The cone angle; 2) Displacement of the high energy
water and 3) The electronics and sterics of the aniline substitution.
Conformational restriction is an effective tactic in follow-on drug discovery and rational design.[47]
It has been used in several medicinal chemistry programs including the development of a selective CB2
cannabinoid receptor,[48] and on the human histamine H4 receptor (hH4R).[49] This methodology has
shown promise in kinase design by pre-organising the inhibitors in a preferred conformation. This can
be used to increase potency on target and enhance selectivity over other kinases/targets. Cardiac
troponin I-interacting kinase (TNNI3K)[50-51] and Spleen tyrosine kinase (Syk)[52] have both utilised this
to increase potency (by 60-160-fold) and selectivity by the addition of a nitrogen (Figure S1).
Adding a nitrogen has been proven to have a significant effect on activities across different systems,
not only by altered conformation, but potentially reduced de-solvation energy potential.[53] In the case
of the 4-anilinoquin(az)oline ring system with GAK, we found the opposite observation; where going
This article is protected by copyright. All rights reserved.

ChemMedChem
10.1002/cmdc.201900521
from C-H to nitrogen increases activity, highlighting that importance of both conformation and
electronics. (Figure 14). The addition of the nitrogen has allowed for a planar conformation, but the
o
GAK ATP competitive active site prefers a 60-90 4-anilinoquin(az)oline. The favorable cone angle of
o
the 4-anilinoquinoline laying between 45-60 is needed for optimal GAK binding and results in
increased rigidity along amine linkage with a hierarchy of 3-cyanoquinoline > quinoline > quinazoline.
Figure 14. GAK inhibitor conformational sensitive to increased ring rigidity.
There are extensive examples of quinolines and quinazolines in the literature.[6-8] Our modelling
highlights that the quinoline scaffolds seems to have favourable binding to the hydrophobic back
pocket, whereas the quinazoline scaffold seems to require favourable interaction to bridging structural
water inside the binding cavity to reach high affinity. The 3-cyanoquinoline requires more space than
quinoline/quinazoline, and according to several experimental co-crystal structures, including bosutinib
in Src and Pelitinib in PKMYT1,[54-55] the bulky cyano group is able form directional hydrogen bond
contacts with waters or polar side chains at the hinge binder region. Docking simulations of GAK
suggest that favourable effects of 3-cyanoquinoline allow it to completely fill the available space at
back pocket; in addition to providing a conformational rigidity to the scaffold. The properties of the
scaffold are all interrelated, as the addition of a nitrogen/cyano at the 3-positon of the quin(az)oline can
potentially form an additional hinge contact within other kinases.
The small molecule crystal structures (Figure 4 & Table S10) support the theoretical observations
with the GAK ATP binding site. The quinazoline is consistently >10-fold less potent for GAK with a
maximum of short of 80-fold for 21 vs 36. The 3-cyanoquinoline is consistently 5-15-fold more potent
than the quinoline for GAK. The switch from trimethoxy to 3-actylene show similar trends with regards
to GAK activity. This is likely due to the ability of both substitution patterns being able to displace the
high energy water W1 in the GAK hydrophobic pocket (Figure 10).[13] Comparison of WaterMap sites
between GAK, STK10 and SLK suggests that the high energy water is not able to be accessed easily in
This article is protected by copyright. All rights reserved.

ChemMedChem
10.1002/cmdc.201900521
the case of SLK and STK10 (Figure 13). SLK appears to prefer a flipped binding mode when docked
with a range of 4-anilinoquin(az)oline; this prevents effective control of the water network as seen in
GAK (Figure 13). Water networks within kinase inhibitor design and drug discovery more widely, are
generally poorly understood. The successful targeting of these networks prospectively, has great
promise for all facets of inhibitor design, for both potency and selectivity on target.
Conclusions
While literature 4-anilinoquin(az)oline inhibitors showed consistently potent cellular target
engagement for GAK except for lapatinib. We have demonstrated an ability to modulate the GAK
activity profile in vitro and in cells on the same scaffold. We were also able to design and synthesize a
compound (87) that was 0.47-fold SLK/STK10 selective over GAK on a scaffold that has shown
consistently high potency towards GAK. The landscape of kinase drug discovery is complex, and the
identification of highly selective inhibitors is still a significant challenge.[56] Our QSAR and water
network models for GAK, SLK and STK10 provide new insights into the development of selective
chemical probes for SLK and STK10 utilising the 4-anilinoquin(az)oline scaffold. The comprehensive
data sets also afford a more complete level of knowledge of the 4-anilinoquin(az)oline previously
unavailable to the medicinal chemists toolbox.
Molecular Modelling
Structures of small molecules: were prepared using the LigPrep module of Schrodinger suite
(
Schrödinger Release 2018-4: Schrödinger, LLC, New York, NY, 2018) employing OPLS3e force field.
Prior to field bases QSAR studies the ligand structures were overlaid with co-crystallized ligands or
favorable docking poses using Flexible ligand alignment tool employing a Bemis-Mucko method,[42]
Torsion angles of larger substituents, such as methoxy groups, were manually adjusted to fit template
structures when needed.
3D-QSAR models: were constructed using the field based QSAR functionality of Schrödinger Maestro
2018-4.
This article is protected by copyright. All rights reserved.