Generation of highly potent DYRK1A-dependent inducers of human β-Cell replication via Multi-Dimensional compound optimization

Article abstract of DOI:10.1016/j.bmc.2019.115193

Small molecule stimulation of β-cell regeneration has emerged as a promising therapeutic strategy for diabetes. Although chemical inhibition of dual specificity tyrosine-phosphorylation-regulated kinase 1A (DYRK1A) is sufficient to enhance β-cell replicat

Full text of DOI:10.1016/j.bmc.2019.115193

Journal Pre-proofs
Generation of Highly Potent DYRK1A-Dependent Inducers of Human β-Cell
Replication via Multi-Dimensional Compound Optimization
Paul A. Allegretti, Timothy M. Horton, Yassan Abdolazimi, Hannah P.
Moeller, Benjamin Yeh, Matthew Caffet, Guillermina Michel, Mark Smith,
Justin P. Annes
PII:
S0968-0896(19)31138-1
https://doi.org/10.1016/j.bmc.2019.115193
BMC 115193
DOI:
Reference:
To appear in:
Bioorganic & Medicinal Chemistry
Received Date:
Revised Date:
Accepted Date:
12 July 2019
20 September 2019
29 October 2019
Please cite this article as: P.A. Allegretti, T.M. Horton, Y. Abdolazimi, H.P. Moeller, B. Yeh, M. Caffet, G.
Michel, M. Smith, J.P. Annes, Generation of Highly Potent DYRK1A-Dependent Inducers of Human β-Cell
Replication via Multi-Dimensional Compound Optimization, Bioorganic & Medicinal Chemistry (2019), doi:
https://doi.org/10.1016/j.bmc.2019.115193
This is a PDF file of an article that has undergone enhancements after acceptance, such as the addition of a cover
page and metadata, and formatting for readability, but it is not yet the definitive version of record. This version
will undergo additional copyediting, typesetting and review before it is published in its final form, but we are
providing this version to give early visibility of the article. Please note that, during the production process, errors
may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.
©
2019 Elsevier Ltd. All rights reserved.

Generation of Highly Potent DYRK1A-Dependent
Inducers of Human β-Cell Replication via Multi-
Dimensional Compound Optimization
Paul A. Allegretti†,#,‡ Timothy M. Horton, †,#,+,‡ Yassan Abdolazimi, Hannah P. Moeller, $,†
†
†
#
†
#
*,†, #,
Benjamin Yeh, Matthew Caffet, Guillermina Michel, Mark Smith and Justin P. Annes
†
#
+
Department of Medicine and Division of Endocrinology, Stanford ChEM-H, Department of
$

Chemistry, Department of Chemical and Systems Biology, Stanford Diabetes Research Center,
Stanford University, Stanford, CA 94305, USA.
KEYWORDS β-cell replication, β-cell regeneration, Dual-specificity tyrosine-regulated kinase
1
A (DYRK1A), diabetes, medicinal chemistry.
1

Abstract
Small molecule stimulation of β-cell regeneration has emerged as a promising therapeutic strategy
for diabetes. Although chemical inhibition of dual specificity tyrosine-phosphorylation-regulated
kinase 1A (DYRK1A) is sufficient to enhance β-cell replication, current lead compounds have
inadequate cellular potency for in vivo application. Herein, we report the clinical stage anti-cancer
kinase inhibitor OTS167 as a structurally novel, remarkably potent DYRK1A inhibitor and inducer
of human β-cell replication. Unfortunately, OTS167’s target promiscuity and cytotoxicity curtails
utility. To tailor kinase selectivity towards DYRK1A and reduce cytotoxicity we designed a library
of fifty-one OTS167 derivatives based upon a modeled structure of the DYRK1A-OTS167
complex. Indeed, derivative characterization yielded several leads with exceptional DYRK1A
inhibition and human β-cell replication promoting potencies but substantially reduced cytotoxicity.
These compounds are the most potent human β-cell replication-promoting compounds yet
described and exemplify the potential to purposefully leverage off-target activities of advanced
stage
compounds
for
a
desired
application.
2

1
. Introduction
Diabetes is a disorder of glucose homeostasis characterized by a mismatch between insulin
supply and demand.1, 2 In normal physiology, insulin is secreted by pancreatic β-cells in
proportional response to elevations in blood glucose; however, diabetes ensues when insulin
secretion is insufficient to lower blood glucose, generally in conjunction with reduced β-cell
mass.3, 4 In adult animals, new β-cells generated to increase insulin production capacity may arise
from a variety of cell types including partially defined facultative progenitor cells, other islet cells
and, most convincingly, previously existing β-cells.5-9 Among the variety of strategies for
replacing an individual’s β-cell mass, expansion of endogenous β-cells remains paramount.
Although mature human β-cells do not typically exhibit vigorous expansion, human β-cell mass
is flexible.7, 10, 11 β-cell number increases through early adulthood and modest compensatory β-cell
mass expansion is an adaptive mechanism for avoiding hyperglycemia.6, 12-15 In T1D, up to 88%
of individuals have a residual, functional, β-cell mass despite ongoing β-cell destruction.16-18 In
pancreata of T1D patients, β-cells exhibit increased replication, suggesting an ongoing
regenerative effort that could be pharmacologically supported.16, 17, 19 Importantly, replacement of
β-cell mass with islet/pancreas transplantation can restore glucose homeostasis in T1D and T2D
patients.20-22 While islet transplantation, whether cadaveric- or stem-cell-derived, may play an
increased role in future T1D treatment, the expense and complexity of these procedures will limit
widespread use. Consequently, developing a safe regenerative medication to expand residual β-
cell mass would be transformative.
3

Figure 1. Representative chemical structures of small
molecule inducers of β-cell replication.
Historically, the major challenge to expanding human β-cell mass was the absence of human β-
cell replication-promoting compounds. To address this gap, we developed a high-content primary
β-cell replication-screening platform.23-26 Using this platform, we discovered chemically diverse
small molecule inducers of β-cell replication including CC-401 1, 5-iodotubercidin (5-IT) 2, and
JAK3 inhibitor VI, while others have identified harmine 3 (and derivatives) and aminopyrazines
such as GNF-4877 4 (Figure 1).23, 26-31 Critically, most (if not all) compounds that exhibit human
β-cell replication-promoting activity inhibit the dual-specificity tyrosine phosphorylation-
regulated kinase 1A (DYRK1A), indicating a central role for DYRK1A in controlling human β-
cell entry into cell cycle. Mechanistically, DYRK1A inhibition promotes replication via enabling
nuclear localization of nuclear factor of activated T-cells (NFAT), 26-29, 32 repression of the cell-
cycle inhibitor p27kip1 and destabilization the multi-protein DREAM complex (DP, RBL1,
RBL2, E2F4 and the MuvB core), which maintains cellular quiescence through repression of cell-
cycle-promoting genes.23, 33, 34 Given the global role of DYRK1A in maintaining cellular
3
4
quiescence, the immediate use of DYRK1A inhibitors in humans to promote β-cell expansion is
limited by concerns for off-target growth-promoting activity.
4

To employ DYRK1A inhibitors as a regenerative therapy for diabetes, a strategy for selective
delivery to β-cells in situ is necessary. Currently, a variety of β-cell targeting approaches are being
explored including peptide- or antibody drug conjugation35-39 and conjugation to small molecules,
including zinc chelator drug conjugation that leverages the uniquely high zinc content of β-cells40
4
1
and VMAT2 antagonists. Primary limitations of β-cell targeting strategies are (a) compound
delivery capacity (ligand surface expression level and compound internalization/release kinetics)
and (b) β-cell selectivity of compound delivery (off-target ligand expression and/or non-selective
uptake/compound release).42 Notably, experience with antibody-drug conjugate systems for
targeted delivery has demonstrated that compounds must exhibit exceptional potency in
biochemical assays (IC in the 10–100 pM range) and cellular assays (<low nM range) to be
5
0
4
2
efficacious. Unfortunately, candidate β-cell replication-promoting compounds have insufficient
intrinsic DYRK1A inhibition [IC ] and/or human β-cell replication induction-potency [EC ],
5
0
min
respectively: 5-IT 2 (~10 nM; ~100 nM),26, 29 Harmine 3/Harmine-derivatives (~50 nM; ~3,000
nM)28, 30, 43 and GNF-4877 4 (6 nM; ~100 nM). Hence, there is a need to develop more potent
DYRK1A inhibitors to realize the promise of a regenerative therapy for diabetes.44, 45
27
2
2
. Results
.1 Discovery of OTS167 as A Potent Inducer of Human β-Cell Replication
In prior work, we identified CC-401 1 as a DYRK1A inhibitor that promotes human β-cell
replication.23 To gain insight into the mechanism of CC-401 1-dependent β-cell replication
induction, we performed differential gene expression analysis on fluorescence activated cell-sorted
(FACS) vehicle- and CC-401-treated β-cells. Interestingly, expression of the cell-cycle regulator
2
3
maternal embryonic leucine zipper kinase (MELK) was induced (2.6-fold) by CC-401 1. Given
5

4
6
47,
MELK’s role in activating forkhead box M1 (FOXM1), a master regulator of β-cell replication,
4
8
we hypothesized that inhibition of MELK would abrogate CC-401 1-dependent induction of
human β-cell replication. Unexpectedly, OTS167 5, a chemotherapeutic MELK-inhibitor
(
OncoTherapy Science, Inc, Japan)49, 50 induced rather than inhibited human β-cell replication
Figure 2).
(
Figure 2. Structure and biological activity of OTS167 5. (A) Chemical structure of OTS167 5. (B) Relative human β-
+
+
+
cell replication (Ki67 Insulin / Insulin ) in OTS167 5-treated wells (n = 5 per condition) compared to the vehicle-
treated wells. This experiment was repeated with 3 independent donors with similar results (results from a single donor
shown; standard deviation is indicated; *, p<0.05).
Strikingly, OTS167 5 was an exceptionally potent (ECmin = 5 nM) inducer of human β-cell
replication, demonstrating efficacy at ~50-fold lower concentration than the most potent known β-
2
7
cell replication promoting compound (GNF4877 4, ECmin ≈ 100 nM). Consistent with OTS167’s
clinical application as a cytotoxic agent, β-cell replication only occurred in a narrow dose range
~5–40 nM). Hence, OTS167 5 is a uniquely potent inducer of human β-cell replication with
5
(
limited utility because of concomitant cytotoxicity. We hypothesized that the β-cell replication-
promoting and cytotoxic activities of OTS167 5 resulted from inhibition of distinct (separable)
kinase targets.
6

2
.2. Evaluation of OTS167 Kinome Inhibition
To investigate our hypothesis that OTS167 5 promoted replication and cytotoxicity via inhibition
of distinct kinase targets, we performed a kinome inhibition scan (468 kinases, DiscoverX) (Figure
3
; Supporting Information Table 4S). OTS167 5 [100 nM] exhibited remarkable promiscuity,
inhibiting 189/403 kinases to less than 35% of baseline activity. More striking, 69/403 of kinases
showed <1% residual activity, including DYRK1A. Notably, inhibition of DYRK1A is sufficient
to promote human β-cell replication,23, 27-30 indicating this was the likely mechanism of OTS167
5
-induced human β-cell replication. By contrast, the basis of OTS167 5 cytotoxicity was less
intuitive as it inhibited numerous targets with the potential to confer cytotoxicity (Figure 3).
Although OTS167 5 was advanced through clinical trials (Phase I/II), the purported cytotoxic
target (MELK) has been challenged.51-54 Indeed, the promiscuity of OTS167 5 indicated
cytotoxicity was a consequence of inhibiting multiple pro-survival kinases.
Figure 3. Kinase selectivity profile of OTS167 5. Kinome scan of 468 kinases with OTS167 5 at 100 nM. Circles
indicate percent residual activity for an individual kinase. DYRK1A is highlighted (blue) as it was among the most
inhibited kinases.
7

Given the likely role of DYRK1A inhibition in mediating OTS167 5-dependent replication
induction, we modeled the OTS167 5-DYRK1A complex (Glide, Schrödinger) to identify key
molecular interactions (Figure 4). Among these were H-bonds between the phenoxide (head
group) and the side chains of the catalytic triad consisting of Lys188, Glu203, and Asp307, a H-
bond between N of the 1,5-naphthyridine core and the hinge residue Leu241, and several Van der
1
Waals interactions between the protein and the amino-cyclohexane group (tail group). The binding
model revealed two regions of OTS167 5, the 3-position of the 1,5-naphthyridine ring (core group)
and the dimethylamino group (tail group), that appeared amenable to non-disruptive modification.
A third position, the dichlorophenol (head group) sat tightly within the ATP-binding pocket of
DYRK1A and was likely to tolerate only small modifications. However, assuming a similar pose
in other kinase targets, head group modifications were anticipated to substantially influence kinase
selectivity. Therefore, to reduce cytotoxicity while retaining human β-cell replication-promoting
activity (increased DYRK1A selectivity) we designed and synthesized a library of 1,5-
naphthyridine analogs with modifications in these three regions. From a library design standpoint,
we disregarded MELK as the primary target for cytotoxicity; this approach proved prudent, as
cytotoxicity was independent of MELK inhibition activity (see below).
8

A
B
Figure 4. Computational modeling of OTS167 5. (A) Modeled structure of OTS167 5 docked into the DYRK1A ATP-
binding site. (B) Associated ligand interaction diagram between OTS167 5 and predicted key binding residues that
are within 4 Å of the ligand. The gray spheres on atoms of OTS167 5 indicate regions that are solvent exposed. Arrows
indicate key hydrogen bonds between the protein backbone (Leu241) and side-chain (Lys188) residues and the ligand.
Negatively charged residues are indicated in red, positively charged residues are highlighted in blue, polar residues
are highlighted in light blue, and hydrophobic residues are highlighted in green. Docking was performed using
Schrödinger’s Glide software and PDB code 4MQ2 as the template for DYRK1A.55
Scheme 1. General synthesis of 1,5-naphthyridine based DYRK1A inhibitors.
9

Scheme 2. General synthesis of 3-carboxamide-1,5-naphthyridines
2
.3. Chemistry
The synthetic route for 1,5-naphthyridine-based DYRK1A inhibitors facilitated generation of
diverse compounds from common intermediates (Scheme 1). Construction of the 4, 6-dichloro-
1
,5-naphthyridine cores (10) was accomplished via the condensation of substituted aminopyridines
with the desired 2-ethoxy methylene malonate derivatives, enabling the incorporation of several
electron withdrawing groups (CO Et, CN, C(O)Me) at what will be the 3-position of the
2
®
naphthyridine systems. Heating of the condensation products in refluxing Dowtherm A followed
by chlorination delivered the desired core fragments. Exposure to a variety of elaborated amines
under standard S Ar conditions introduced substituents at the 4-position of the naphthyridine ring.
N
A final Suzuki coupling completed the desired DYRK1A inhibitors. Analogs containing amides
at the 3-position (14) were accessed via ester hydrolysis and HATU coupling with the desired
amine (Scheme 2).
2
.4. Primary Screen of DYRK1A Inhibition and Cytotoxicity
To perform primary screening of the OTS167-derived compound library, we simultaneously
assessed cellular DYRK1A inhibition and cytotoxicity using a dual luciferase reporter system
1
0

(
Figure 5).23, 40 Determination of DYRK1A inhibition was based upon de-repression of an NFAT-
dependent reporter (DYRK-Luc) activity. In this assay, OTS167 demonstrated potent DYRK1A
inhibition (DYRK-Luc EC ≈ 5 nM; Table 1). To assess compound toxicity, we determined the
5
0
concentration at which constitutive Renilla Luciferase (R-Luc) activity dropped by 50% (R-Luc
CC ). While R-Luc activity is commonly used to normalize transfection efficiency, we observed
5
0
a concentration-dependent loss of R-Luc activity in OTS167-treated HEK293T cells (CC ≈ 64
5
0
nM; Table 1) that was independent of transfection efficiency (the same population of transfected
cells were treated with different compound concentrations). Further, the CC ’s based on R-Luc
5
0
2
(
R-Luc CC ) linearly correlated (R = 0.78) with CC ’s obtained by direct assessment of cell
50 50
number (Supporting Information Figure 1S). Hence, loss of R-Luc served as a useful proxy for
compound cytotoxicity (R-Luc CC ; Table 1).
5
0
Figure 5. (A) Schematic representation of the cellular (HEK293) DYRK1A- and cytotoxicity-reporter assays. The
potency of DYRK1A inhibitors (DYRK-Luc) was assessed based upon de-repression of the NFAT-dependent
luciferase reporter construct, which is silenced by DYRK1A kinase activity. Co-transfection of a constitutively active
CMV-Renilla expression construct allowed concurrent assessment of cellular viability in response to compound
1
1

treatment (R-Luc). (B) A representative result for OTS167 5, a dose curve of which was included as a positive control
in every assay, with DMSO (n=8) included in every plate as a negative control.
1
2

Table 1. Chemical structures of the 1,5-naphthyridines library with measured DYRK1A inhibition potency
DYRK1A-Luc EC50), cytotoxicity (R-Luc CC50) and calculated cytotoxicity to DYRK1A inhibition ratio (Index) in
(
HEK 293 cells are shown.
DYRK1A
Luc
Index
R-Luc
-
Compound
R1
R2
R3
CC50
(CC50
/
EC50
EC )
b
5
0
(nM)
(
nM)a
5
5
64
13
1
1
1
1
5
6
7
8
7
45
15
6
186
26
>10,000c >220
69
36
4
6
1
3

1
2
2
2
2
2
2
2
9
0
1
2
3
4
5
6
6
6
39
38
7
7
3
51
17
4
5
19
6
16
3
3
21
8
641
354
>10,000c
>10,000c
>16
>28
1
4

2
6
3,794
165
146
20
2
2
7
8
8
2
9
15
3,344
1,141
2,673
217
71
1
6
3
3
0
1
100
27
3
3
3
3
2
3
4
5
>10,000d >10,000c
n/a
5
9
6
44
1,022
3,465
170
17
203
1
5

3
3
3
3
4
4
4
4
4
6
7
8
9
0
1
2
3
4
123
105
73
>10,000c
3,678
>82
35
>10,000c >137
7
491
79
71
18
4
1,252
50
>10,000c
5,902
>8
119
39
175
678
6,854
>10,000c
>15
1
6

4
4
4
4
4
5
5
5
5
5
6
7
8
9
0
1
2
3
11
97
56
74
43
1,662
3,006
2,119
1,974
1,893
153
21
38
27
44
>10,000d >10,000c
n/a
>33
300
12
>10,000c
>10,000c >842
2,095
>10,000c
>5
1
7

5
5
5
5
5
5
6
6
6
4
5
6
7
8
9
0
1
2
990
4,646
223
>10,000c
>10,000c
8,051
>10
>2
36
1,530
>10,000c
>7
n/a
19
>10,000d >10,000c
59
46
1,129
>10,000c >216
383
1,336
2,361
6
>10,000c
>7.5
1
8

6
6
3
4
37
2,962
1,347
756
79
12
7
114
1
08
6
5
aCalculated by 4-parameter agonist dose response (Graphpad) of the induction of DYRK1A-
responsive firefly luciferase to test compound in a HEK293T reporter cell assay (11-point dose
curve, n = 4 per concentration).
bCalculated by 4-parameter inhibitor dose response (Graphpad) to test compound of
constitutively-active renilla luciferase in a HEK293T reporter cell assay (11-point dose curve, n =
4
per concentration).
cCompound did not inhibit renilla luciferase expression by >50% at 10 μM.
dCompound did not stimulate firefly luciferase expression at 10 μM.
We determined the DYRK1A inhibition potency and cytotoxicity of fifty-one 1,5-naphthyridine
derivatives with structural modifications at three different positions of the naphthyridine scaffold
(Table 1). Generally, modifications to the electron-withdrawing group at the 3-position of the 1,5-
1
naphthyridine system (R 15-26) resulted in minimal change to DYRK1A inhibition potency or
;
cytotoxicity. However, introduction of an ethyl ester (15) at this position decreased cytotoxicity
(
~3-fold) compared to OTSS167 (R-Luc CC = 186 nM (15) vs. 64 nM (5)) with little effect on
50
1
DYRK1A inhibition potency [DYRK1A-Luc EC = 7 nM (15) vs. 5 nM (5)]. Similarly, R
5
0
substitutions with various carboxamides (17-24) minimally affected cytotoxicity or DYRK1A
inhibition, consistent with our OTS167-DYRK1A interaction model. Carboxamides derived from
1
9

secondary amines, like the morpholino (25) and dimethyl amide (26), exhibited no cytotoxicity
R-Luc CC = >10,000 nM, both); however these compounds also showed decreased DYRK1A
(
50
inhibition (DYRK1A-Luc EC = 641 nM (25) and 354 nM (26)). Remarkably, introduction of a
5
0
cyano-group (16) at the 3-position rendered the compound virtually non-toxic with only a ~9-fold
reduction in DYRK1A inhibition potency. Overall the structure-activity relationships exhibited by
substitution at the 3-position were flat with respect to DYRK1A inhibition, but in the case of
compound 16 revealed significant reductions in toxicity. These results demonstrated the feasibility
of separating the DYRK1A inhibition- and cytotoxic-activities of OTS167.
Next we assessed the activities of amine substituents that were introduced at the 4-position (R2;
2
7–49). Unexpectedly, substitution of an aromatic analog at this position (27) maintained
DYRK1A inhibition potency, but reduced cytotoxicity by ~60-fold. The basic dimethyl amino
group also played a remarkable role in modulating the toxicity of the compounds. Des-methyl
compound 28 had activities (DYRK1A-Luc EC = 8 nM, R-Luc CC = 165 nM) comparable to
5
0
50
the dimethyl analog (15), however the acetylated version (29) exhibited significantly reduced
toxicity (R-Luc CC = 3,344 nM) while retaining potency against DYRK1A (DYRK1A-Luc EC
50
5
0
=
15 nM). Notably, reducing the ester of compound 29 to an alcohol (30) maintained the reduced
cytotoxicity (R-Luc CC = 1,142 nM) and potent DYRK1A inhibition (DYRK1A-Luc EC = 16
5
0
50
nM). Exchange of the amino group for a hydroxyl (31) was relatively tolerated; however, removal
of a hydrogen bond donor/acceptor group in this position eliminated cellular activity (32). Moving
the basic nitrogen atom into the ring in the 4-aminopiperidine analog (34) reduced cytotoxicity
~
16-fold (R-Luc CC = 1,022 nM) without affecting DYRK1A inhibition potency (DYRK1A-
50
Luc EC = 6 nM). Hence, compound 34 demonstrated a dramatic (~13-fold) increase in the
5
0
Cytotoxicity / Activity Index. Thus, 34 demonstrated the feasibility separating the cytotoxic and
2
0

DYRK1A-inhibition activities of OTS167 5 through substitutions at the 4-position of the 1,5-
naphthyridine ring.
Additionally, we evaluated the activity of several cyclic secondary amines substituted at the 4-
2
position of the 1,5-naphthyridine ring (R ). Piperidine analogs 35 and 36, morpholino analog 37,
and piperazine analog 38 all exhibited reduced cytotoxicity while also demonstrating reduced
DYRK1A inhibition potency. However, the presence of a basic amine (piperazine) in the R2
position (38) minimized the loss in DYRK1A inhibition potency (DYRK1A-Luc EC = 73 nM),
5
0
while significantly reducing cytotoxicity (R-Luc CC = ≥10,000 nM); resulting in a desirable
5
0
cytotoxicity / Activity Index (Index (CC /EC ) =137).
5
0
50
Several compound pairs with substitutions at both the 3- and 4-positions produced activities that
were divergent from what was observed when only a single position was modified. The 4-
(
morpholinomethyl)cyclohexanamine analogs 39 and 40 had similar DYRK1A potencies
DYRK1A-Luc EC = 7 nM vs 4 nM respectively), yet the ester analog 38 was 7-fold less toxic
(
5
0
than its butyl amide counterpart 39 (R-Luc CC = 491 nM vs 79 nM). The morpholino amide
5
0
1
1
analogs 46 (R = ethyl ester) and 47 (R = butyl amide) had similar cellular activities and toxicities.
The 4-aminocyclohexane carboxamide derived compounds 42 and 43 had similar toxicities (R-
Luc CC = 5,912 nM vs 6,854 nM, respectively), yet the ethyl ester 42 was more potent against
5
0
DYRK1A than the butyl amide 43 (DYRK1A-Luc EC = 50 nM vs 175 nM respectively),
5
0
1
indicating that substitutions at the 3-position (R ) of the 1,5-naphthyridines that reduce cytotoxicity
are context-dependent.
3
The last region of 5 we explored was the head group (R ) that makes a key interaction with the
catalytic Lys188 of DYRK1A. Although modeling indicated that modifications in this region
would affect DYRK1A potency, effects on cytotoxicity were less predictable without knowledge
2
1

of the relevant kinase target(s). As expected, removal of the 4-hydroxyl (50) abrogated DYRK1A-
inhibition, while removal of the Cl-atoms in analog 51 reduced the DYRK1A-Luc EC to 300
5
0
nM, and rendered the compound nontoxic (R-Luc CC = >10,000 nM). Our docking model
5
0
indicated the two Cl-atoms were filling small hydrophobic pockets adjacent to the catalytic triad
residues. Interestingly, similar interactions were maintained by methyl groups (52) (DYRK1A-
Luc EC = 12 nM) while profoundly reducing cytotoxicity (R-Luc CC = >10,000 nM), resulting
5
0
50
in the best Index score (>842) of the library. The 4-pyridyl 56 was also relatively tolerated, though
this compound’s potency against DYRK1A decreased to 223 nM. Presence of adjacent methyl
groups (57 and 58) reduced DYRK1A binding to >10,000 nM while the presence of an amino
group adjacent to the pyridine nitrogen (59) retained DYRK1A inhibition potency (DYRK1A-Luc
EC = 59 nM). Exchange of the ethyl ester in 59 with and ethyl amide 60 maintained good
5
0
DYRK1A activity, but dramatically reduced cytotoxicity (R-Luc CC = >10,000 nM).
5
0
2
.5. Characterization of the OTS167 Cytotoxic Mechanism
Although MELK inhibition is the purported mechanism of OTS167 cytotoxicity, several
publications have challenged this conclusion.51-54 To assess the contribution of MELK inhibition
to cytotoxicity, we evaluated the relationship between MELK binding affinity (MELK K ) and
d
cellular cytotoxicity (R-Luc CC ) among a small collection of compounds (5, 15, 21, 28 and 42).
5
0
These compounds were selected based upon a linear relationship between DYRK1A binding
2
affinity (DYRK1A K ) and cellular DYRK1A inhibition (DYRK1A-Luc EC ) (R = 0.98, p-value
d
50
of slope = 0.0016) (Figure 6) to exclude confounding impacts of cell permeability. As suggested
2
by others, a correlation between MELK binding affinity and cytotoxicity was not observed (R =
0
.05, p-value of slope=0.71) (Table 2, Figure 6). Although these compounds demonstrated potent
2
2

MELK binding affinity (K ≤15 nM), cytotoxicity (R-Luc CC ) ranged from 51 nM to 5,900 nM.
d
50
Hence, inhibition of MELK was not sufficient for cytotoxicity.
Figure 6. DYRK1A binding correlates to DYRK-NFAT EC50 and β-cell replication, while MELK binding does not
correlate to cytotoxicity. (A) Selected compounds (5, 15, 21, 28, and 42) for which DYRK-NFAT EC50 was linear
2
with DYRK1A K
d
(R =0.98) plotted as DYRK-NFAT EC50 vs DYRk1A K
d
. (B) The same compounds (5, 15, 21, 28,
2
and 42) with R-Luc CC50 plotted versus MELK K
d
(R =0.05). (C) The same compounds (5, 15, 21, 28, and 42) with
2
human β-cell replication ECmin plotted versus DYRK1A K
d
(R =0.99).
Table 2: Comparisons of DYRK1A K
d
vs DYRK1A-Luc EC50 and MELK K
d
versus cytotoxicity (R-Luc CC50) for
select 1,5-naphthyridines.
DYRK1A-
Luc
DYRK1A
K (nM)
d
MELK Kd
(nM)
R-Luc
Compound
CC (nM)
5
0
EC (nM)
5
0
5
0.12
0.16
0.075
0.22
0.6
5
15
11
12
9
64
1
5
1
8
2
7
3
186
51
2
2
4
8
165
5900
50
13
To identify kinase targets relevant to OTS167 5 cytotoxicity, we compared the kinome
selectivity of compound (56) (R-Luc CC = 8,051 nM) to OTS167 5 (R-Luc CC = 64 nM)
5
0
50
2
3

(Figure 7). Compound 56 was prioritized for further characterization based on <30 nM
human beta-cell replication potency and >1000 nM R7T1 toxicity as discussed below.
Importantly, these compounds have equivalent MELK binding affinity (MELK K = 15 nM).
d
Remarkably, the simple substitution of the dichlorophenol group with a pyridyl resulted in a
dramatic improvement in kinase selectivity. By filtering for targets robustly inhibited by OTS167
5
but not inhibited by compound 56, we obtained a narrowed list of potential cytotoxic targets. Of
the 468 kinases assayed, 40 were strongly inhibited (<2% residual activity) by OTS167 5 at a
concentration of 100 nM but not by compound 56 at a concentration of 500 nM. Although the
promiscuity of OTS167 5 indicated toxicity was likely to be multifactorial, we selected five
differentially inhibited kinases (AURKA/B, MAP2K1/2, and ABL1) with pro-survival activity for
further analysis.56-58
Figure 7. Kinase selectivity profile for compound 56. (A) Kinome scan with compound 56 at 500 nM. DYRK1A is
highlighted (blue). (B) Activity of select kinases (ABL1, DYRK1A, AURKB, AURKA, MAP2K1, MAP2K2, and
MELK) with OTS 167 5 (100 nM) or compound 56 (500 nM) as a percent of the control (DMSO-treated) activity.
To test our hypothesis that the inhibition of AURKA/B, MAP2K1/2 and/or ABL1 mediated the
OTS167 cytotoxicity, we identified relatively selective chemical inhibitors of each of these targets:
2
4

AURKA (MK-5108; MK), AURKB (AZD-1152-HQPA; AZD), AURKA/B (MLN-8054, MLN),
MAP2K1/2 (TAK-733 and AS-703026) and ABL1 (Dasatanib and Imatinib).59-67 We reasoned,
for example, that if OTS167 5 inhibition of AURKA/B was responsible for cytotoxicity, then (A)
AZD, MK, and/or MLN would be cytotoxic in our model system, and/or (B) the combination of
AURKA/B inhibition with compound 56 would recapitulate the cytotoxicity of OTS167 5.
Informatively, both results were observed (Figure 8, Supporting Information Figure 2S).
Figure 8. Combination of 56 with inhibitors of AURKB but not MAP2K1/2 or ABL1 reduces cell viability. HEK293T
cells were treated with (A) DMSO, 56 (30 nM, 90 nM, and 270 nM), AZD-1152-HQPA (40 nM) without and with
5
6, or 5 (300 nM); (B) DMSO, 56 (30 nM, 90 nM, and 270 nM), MLN-8054 (40 nM) without and with 56, or 5 (300
nM); (C) DMSO, 56 (30 nM, 90 nM, and 270 nM), MK-5108 (3 μM) without and with 56, or 5 (300 nM); (D) DMSO,
6 (90 nM), TAK-733 (10 μM) without and with 56, or 5 (300 nM); (E) DMSO, 56 (90 nM), AS-703026 (10 μM)
5
2
5

without and with 56, or 5 (300 nM); and (F) DMSO, 56 (90 nM), Imatinib (3 μM) without and with 56, or Dasatinib
1 μM) without and with 56. Cells were treated for 48 hrs, fixed, stained, counted, and cell number plotted as a
percentage of untreated wells. Error bars indicate standard deviation (n = 4-16, ****,p<0.0001; ***, 0.0001<p<0.001;
(
*
*, 0.001<p<0.01; *, 0.01<p<0.05; NS, p>0.05).
AZD (CC = 38.9 nM) and MLN (CC = 55.1 nM) were highly cytotoxic, while MK was not
5
0
50
(CC50 = >10 μM); indicating that AURKB inhibition, but not AURKA, contributed to the
cytotoxicity of OTS167 5. Furthermore, combining AURKB inhibitors with 56 enhanced
cytotoxicity. Compound 56 exhibited no cytotoxicity at 30 nM, 90 nM or 270 nM while AZD
reduced cell number by 41% at 40 nM (p=0.0012). However, combining of AZD (40 nM) with
-4
compound 56 [(90 nM] or (270 nM)] resulted in a 67% reduction in cell number (p=2.01x10 and
-5
-4
p=2.7 × 10 vs DMSO; p=0.0026 and p=6.5 × 10 vs AZD; Figure 8); similar to treatment with
OTS167 5 (300 nM). Likewise, MLN killed only 22% of cells at 40 nM (p=0.013), but was
-4
augmented to 50% when combined with compound 56 (90 nM or 270 nM; p=8.9 × 10 and
-4
-4
p=0.0011 vs DMSO; p=1.2 × 10 and p=1.7 × 10 vs MLN; Figure 8). These results are
consistent with reports that inhibition of AURKB by OTS167 5 is at least partially responsible for
5
4
its cancer-killing effects. The AURKA-selective inhibitor MK [3 µM] alone killed 24% of cells
(p=0.013) while supplementation with 56 further reduced cell viability (31%, 36%, and 34%;
p=0.0018, p=0.0010, p=0.0014 vs DMSO; p=0.029, p=0.008, p=0.018 vs MK). The relatively poor
potency of this compound may indicate that residual AURKB inhibition (200-fold higher IC than
50
5
9
AURKA), rather than its primary effect of AURKA inhibition, is responsible for the cytotoxicity
in this case. By contrast, the MAP2K1/2 inhibitor TAK-733 (10 uM) did not exhibit cytotoxicity
when tested alone or in combination with 56 (30 nM, 90 nM, or 270 nM). The MAP2K1/2 inhibitor
AS-703026 (10 uM) only killed 10% of cells (p=0.006), and combination with 56 (90 nM) killed
2
6

6
% of cells (p=0.048), which were not significantly different (p=0.071). Similarly, ABL1
inhibition with Dasatinib (1 μM) or Imatinib (3 μM) exhibited some cytotoxicity when tested alone
-4
(
21% and 12.5%, p=0.019 and p=4.0 × 10 versus DMSO), and combination with 56 did not cause
additional cell death (p=0.39 and p=0.46). Taken together, OTS167 5 cytotoxicity was mediated
in part via inhibition of AURKB but not AURKA, MAP2K1/2 or ABL1. However, the increased
cytotoxicity observed with combined AURKB inhibitor and compound 56 treatment indicated
cytotoxicity was multifactorial and did not exclude a contributing role of MELK inhibition.
2
.6. Secondary Screening for β-Cell Cytotoxicity and Human β-Cell Replication Induction
The objective of the primary screening was to identify 1,5-naphthyridine analogs that retained
DYRK1A inhibition potency and reduced cytotoxicity. Accordingly, we selected 18 compounds
with a DYRK1A-Luc EC ≤100 nM and an R-Luc CC ≥500 nM for further characterization,
5
0
50
including β-cell cytotoxicity (R7T1 cells) and human β-cell replication-promoting activity (Table
, Figure 9, Supporting Information Figure 3S, Supporting Information Table 5S,
3
6
8
Supporting Information Table 7S). As anticipated, R7T1 β-cells were generally more sensitive
to compound cytotoxicity than HEK293T cells, with the majority of compounds exhibiting a ~3-
fold increase (Tables 1 and 3). Surprisingly, the 3-cyano substituted naphthyridine 16 and the 3-
amino-4-pyridyl analog 60 exhibited cytotoxicity towards R7T1 β-cells (CC = 160 nM, 300 nM
5
0
respectively) but not HEK293 cells (R-Luc CC = >10,000 nM). Compounds, 34 and 63 exhibited
5
0
similar human β-cell replication-promoting potency (ECmin = 3.3 nM) to OTS167 5, with a greatly
improved toxicity/replication Index (R7T1 CC /Islet EC of 117 and 259, respectively).
5
0
min
Interestingly, the human β-cell replication-promoting activity of compounds 30 and 52 was less
than anticipated (ECmin = 500 nM and 250 nM respectively) from their DYRK1A-Luc assay
2
7

potency (EC = 16 nM and 12 nM respectively). We also observed a discordance between
5
0
DYRK1A-Luc and human β-cell replication-promoting activities with compounds 59 and 60.
Structurally, these compounds are highly similar, differing only in the naphthyridine ring 3-
substitution. While these compounds had comparable activities in the DYRK1A-Luc assay [EC50
=
59 nM (59), EC = 46 nM (60)], the ethyl ester 59 was a significantly more potent inducer of β-
50
cell replication (ECmin = 22 nM) than the ethyl amide 60 (ECmin = >500 nM). Although the basis
of these discrepancies is unknown, these data indicate that the relationship between DYRK1A
inhibition and β-cell replication-promoting potencies are not necessarily direct.
2
8

Figure 9. Human β-cell replication. Representative images of dispersed human islets treated with (A) DMSO or (B)
compound 56 (125 nM) and stained with Hoechst 33342 (blue), anti-Insulin (red), and anti-Ki67 (green). Replicating
cells not definitively insulin positive (white arrows) and replicating β-cells (magenta arrows) are indicated. Images
+
+
+
were processed in parallel and cropped. (C) Relative human β-cell replication (Ki67 Insulin / Insulin ) in compound-
treated treated wells (n = 3 per condition). Asterisks indicate significance testing compared to vehicle-treated wells.
Error bars indicate standard deviation (*, 0.01<p<0.05; **, 0.001<p<0.01; ***, 0.0001<p<0.001; ****,p<0.0001).
Table 3. Human β-cell replication and R7T1 β-cell cytotoxicity of selected compounds.
2
9