Pyrido[2,3-b][1,5]benzoxazepin-5(6H)-one derivatives as CDK8 inhibitors

Article abstract of DOI:10.1016/j.ejmech.2020.112443

CDK8 is a cyclin-dependent kinase that forms part of the mediator complex, and modulates the transcriptional output from distinct transcription factors involved in oncogenic control. Overexpression of CDK8 has been observed in various cancers, representing a potential target for developing novel CDK8 inhibitors in cancer therapeutics. In the course of our investigations to discover new CDK8 inhibitors, we designed and synthesized tricyclic pyrido[2,3-b][1,5]benzoxazepin-5(6H)-one derivatives, by introduction of chemical complexity in the multi-kinase inhibitor Sorafenib taking into account the flexibility of the P-loop motif of CDK8 protein observed after analysis of structural information of co-crystallized CDK8 inhibitors. In vitro evaluation of the inhibitory activity of the prepared compounds against CDK8 led us to identify compound 2 as the most potent inhibitor of the series (IC₅₀ = 8.25 nM). Co-crystal studies and the remarkable selectivity profile of compound 2 are presented. Compound 2 showed moderate reduction of phosphorylation of CDK8 substrate STAT1 in cells, in line with other reported Type II CDK8 inhibitors. We propose herein an alternative to find a potential therapeutic use for this chemical series.

Full text of DOI:10.1016/j.ejmech.2020.112443

Journal Pre-proof
Pyrido[2,3-b][1,5]benzoxazepin-5(6H)-one derivatives as CDK8 inhibitors
Sonia Martínez-González, Ana Belén García, M. Isabel Albarrán, Antonio Cebriá,
Adrián Amezquita-Alves, Francisco Javier García-Campos, Jaime Martínez-Gago,
Jorge Martínez-Torrecuadrada, Inés Muñoz, Carmen Blanco-Aparicio, Joaquín Pastor
PII:
S0223-5234(20)30414-1
DOI:
https://doi.org/10.1016/j.ejmech.2020.112443
EJMECH 112443
Reference:
To appear in: European Journal of Medicinal Chemistry
Received Date: 11 March 2020
Revised Date: 7 May 2020
Accepted Date: 7 May 2020
Please cite this article as: S. Martínez-González, Ana.Belé. García, M.I. Albarrán, A. Cebriá,
Adriá. Amezquita-Alves, F.J. García-Campos, J. Martínez-Gago, J. Martínez-Torrecuadrada, Iné.
Muñoz, C. Blanco-Aparicio, Joaquí. Pastor, Pyrido[2,3-b][1,5]benzoxazepin-5(6H)-one derivatives
as CDK8 inhibitors, European Journal of Medicinal Chemistry (2020), doi: https://doi.org/10.1016/
j.ejmech.2020.112443.
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.
©
2020 Published by Elsevier Masson SAS.

H
N
N
O
H
N
O
O
H
N
O
N
N
H
N
O
O
HN
F3C
Cl
N
Sorafenib
CDK8 IC50 = 36.6 nM / P-STAT1 IC50 > 10 µM
2
F3C
Cl
CDK8 IC50 = 8.25 nM / P-STAT1 IC50 = 1.3 µM
P-Loop displacement
KINOMEscan®

Pyrido[2,3-b][1,5]benzoxazepin-5(6H)-one Derivatives as CDK8
Inhibitors
a
a
a
a
Sonia Martínez-González , Ana Belén García , Mª Isabel Albarrán , Antonio Cebriá ,
a
a
b
Adrián Amezquita-Alves , Francisco Javier García-Campos , Jaime Martínez-Gago , Jorge
b
b
a
a,*
Martínez-Torrecuadrada , Inés Muñoz , Carmen Blanco-Aparicio , Joaquín Pastor
a
Experimental Therapeutics Programme, Spanish National Cancer Research Centre
(
b
(
CNIO), C/ Melchor Fernández Almagro 3, E-28029 Madrid, Spain
Crystallography and Protein Engineering Unit, Spanish National Cancer Research Centre
CNIO), C/ Melchor Fernández Almagro 3, E-28029 Madrid, Spain
Corresponding Author
*
Joaquín Pastor. Phone: +34 917328000, fax: +34 917328051. E-mail address:
jpastor@cnio.es.
Present addresses

Abstract
CDK8 is a cyclin-dependent kinase that forms part of the mediator complex, and modulates
the transcriptional output from distinct transcription factors involved in oncogenic control.
Overexpression of CDK8 has been observed in various cancers, representing a potential
target for developing novel CDK8 inhibitors in cancer therapeutics. In the course of our
investigations to discover new CDK8 inhibitors, we designed and synthesized tricyclic
pyrido[2,3-b][1,5]benzoxazepin-5(6H)-one derivatives, by introduction of chemical
complexity in the multi-kinase inhibitor Sorafenib taking into account the flexibility of the
P-loop motif of CDK8 protein observed after analysis of structural information of co-
crystallized CDK8 inhibitors. In vitro evaluation of the inhibitory activity of the prepared
compounds against CDK8 led us to identify compound 2 as the most potent inhibitor of the
series (IC50 = 8.25 nM). Co-crystal studies and the remarkable selectivity profile of
compound 2 are presented. Compound 2 showed moderate reduction of phosphorylation of
CDK8 substrate STAT1 in cells, in line with other reported Type II CDK8 inhibitors. We
propose herein an alternative to find a potential therapeutic use for this chemical series.
Keywords: CDK8, Type II inhibitors, P-loop flexibility, Selectivity, pyrido[2,3-
b][1,5]benzoxazepin-5(6H)-one scaffold.

Highlights:
•
•
Design of novel type II tricyclic CDK8 inhibitors based on Sorafenib.
Binding of inhibitors to CDK8 by filling an unexplored cavity via displacement of
its flexible P-loop.
•
Potent and highly selective pyrido[2,3-b][1,5]benzoxazepin-5(6H)-one based CDK8
inhibitors.
Graphical Abstract

1
. Introduction
CDK8 is a cyclin-dependent kinase that forms part of the mediator complex [1], which
itself regulates the transcriptional activity of RNA polymerase II. There are numerous
studies showing that CDK8 modulates the transcriptional output from distinct transcription
factors involved in oncogenic control. These factors include the Wnt/β-catenin pathway,
Notch, p53, and TGF-β [2,3]. The means by which CDK8 activity accomplishes the
regulation of these various pathways remains an active area of investigation. However, it
has been found that CDK8 acts as a colon cancer oncogene [4] driving Wnt pathway
activity by mediating β-catenin transcriptional output. Moreover, CDK8 has been related to
colorectal cancer [5], melanoma [6], and it has been demonstrated that CDK8 and its
paralog CDK19 restrain increased activation of key super-enhancer-associated genes in
acute myeloid leukaemia (AML) cells. Thus, CDK8/CDK19 inhibitor Cortistatin A (CA)
has anti-leukaemic activity in vitro and in vivo, and induces upregulation of super-
enhancers in CA-sensitive AML cell lines but not in CA-insensitive cell lines [7].
Furthermore, the phosphorylation of STAT1 at serine 727 by CDK8 is also suppressive of
natural Killers (NK) cells activity. STAT1-S727-mutated NK cells display increased
release of cytotoxic proteins such as granzyme B and perforin and higher cytotoxicity
toward cancer cells. It suggests that CDK8 inhibition could play a role in cancer
immunotherapy [8].
Several CDK8 modulators have been reported so far, among them Sorafenib [9], Senexin
A and B [10], CCT251545 [11] and Cortistatin A [12]. During the last few years our group
has been immersed in this field applying several approaches for hit generation towards the
discovery of novel CDK8 inhibitors. Here we describe our efforts to obtain highly selective

type II CDK8 inhibitors based on Sorafenib, a known type II promiscuous inhibitor. The
use of Sorafenib to design type II CDK8 inhibitors has been recently reported by other
group although following a different strategy to the described herein [13]. Type II mode of
inhibition, generally shows a displaced Asp-Phe-Gly (DFG) motif from a buried position in
the protein to a DFG-out position, opening a “deep pocket” that can be occupied by
inhibitors. In the case of CDK8, these three residues are Asp-Met-Gly (DMG) as exception
to all other kinase family members [14]. In contrast to type I inhibitors that bind an
activated kinase conformation, the application of inhibitors with a deep pocket binding
mode is meant to target the inactive state of the enzyme.
We used the structural information of Sorafenib-bound CDK8 structure [14] (PDB
3
RGF) as a basis to design our novel series of CDK8 type II inhibitors. In this X-ray
structure the distal 4-pyridyl ring (N-methylpicolinamide fragment) of Sorafenib occupies
the ATP adenine binding pocket of the kinase domain, interacting with the hinge region in a
bidentate fashion via hydrogen bonds between the N-methylpycolinamide moiety with
Ala100 backbone. The lipophilic 3-trifluoromethyl-4-chlorophenyl ring inserts into the
deep pocket and establishes hydrophobic interaction with Phe176. Moreover, the C=O
group of the urea functionality forms one hydrogen bond with Asp173 in the DMG triad
and the corresponding NH groups form two hydrogen bonds with the glutamate side chain
of Glu66, a residue that is absolutely conserved in the αC-helix of kinases. Additionally π-
interactions are mapped between Sorafenib and Phe97. This binding mode of Sorafenib to
CDK8/CycC is in accordance to the classical features of DFG-out binders. Unfortunately
the P-loop of CDK8 is not completely solved in this structure.

The analysis of the structural information from several type II inhibitors co-crystallized
with CDK8 [15], where the P-loop have been solved and defined, allowed us to identify a
cavity in the ATP catalytic site flanked by side-chains of the P-loop residues Tyr32 and
Val35 and those of aminoacids Ala50 and Lys52. Interestingly, the comparison of these
structures indicated the high flexibility of Tyr32 depending on the nature of the inhibitor as
well as some degree of flexibility of the entire P-loop (Figure S1). As a matter of example,
protein structures from PDBs 4F7L and 4F6W were taken and aligned together with
Sorafenib co-crystal structure 3RGF to illustrate these findings (Figure 1).
Figure 1. A close view of aligned structures 3RGF (light pink), including Sorafenib (pink
C atoms), 4F7L (gold) and 4F6W (green).
We envisioned that introduction of additional chemical functionality in Sorafenib could
be an interesting approach to fill such cavity in order to access unexplored and therefore
novel type II CDK8 inhibitors. We designed prototype compounds 1 and 2 based on the
tricyclic scaffold pyrido[2,3-b][1,5]benzoxazepin-5(6H)-one. Compound 1 contained the

hinge binding moiety N-methylpycolinamide of Sorafenib while it was replaced by
pyridine in analogue 2 (Figure 2).
N
H
O
O
N
H
N
H
N
O
O
N
H
N
N
O
O
HN
F3C
Cl
N
R
F3C
Cl
1
2
, R= CONHMe
, R = H
Sorafenib
Figure 2. Design of novel type II tricyclic CDK8 inhibitors based on Sorafenib structure.
The simple alignment of the minimal energy conformation [16] of designed compound 2
with Sorafenib pointed to potential steric clashes with Val35, Lys52 and Tyr32 (inner
conformation) in CDK8 structures (Figure 3). Therefore, the proposed inhibitor would need
to induce a conformational change in this area of CDK8 in order to be accommodated, in
particular at the level of the P-loop. As mentioned before, the observed flexibility of the P-
loop, Lys52 and Tyr32 side chains in several CDK8 X-ray structures, suggested the
possibility of binding of compound 2 in the catalytic site of the kinase by an induced fit
mechanism, and therefore supported an experimental attempt to prove such hypothesis.
Furthermore, if the induced binding of compound 2 was achieved, an additional H-bond
interaction could occur between the pyridine N of its scaffold and the Lys52 side chain,
which is not possible in Sorafenib.

Figure 3. Alignment of designed tricyclic prototype compound 2 (cyan C-atoms) with
Sorafenib (pink) using protein structures from 3RGF (light pink), 4F7L (gold) and 4F6W
(
green). Potential steric clashes are detected with residues Val35, Lys52 and Tyr32 (inner
conformation).
2
. Results and discussion
2
.1. Chemistry
We focused our efforts on the preparation of compounds 1 and 2 (Scheme 1 & 2). The
synthesis started from 2-hydroxynicotinic acid and 2-amino-5-nitrophenol as building
blocks. In the first place, 2-hydroxynicotinic acid was converted into the corresponding
acid chloride by treatment with oxalyl chloride. This intermediate was then reacted with 2-
amino-5-nitrophenol to render the benzamide 3 (70% yield for the two steps). Treatment
with NaOH in DMF led to the formation of the tricyclic derivative 4 with good yield
(76%). Subsequent reduction of nitro group either by treatment with Fe in AcOH or by

hydrogenation with Pd/C as catalyst gave the key amino intermediate 5 in a straightforward
manner with good overall yield. The amino intermediate 5 was then reacted with 4-chloro-
3
-(trifluoromethyl) phenyl isocyanate under standard conditions giving the urea 6a in
moderate yield (29%). Finally, coupling of cyclic amide 6a, with 4-iodopyridine under
copper catalyzed conditions rendered the desired prototype 2 albeit in low yield (9%).
Attempts to obtain compound 1 from 6a using similar coupling conditions using methyl 4-
bromo-N-methylpicolinamide failed in our hands. To avoid any potential interference of the
urea moiety of compound 6a in the coupling reaction step, we decided to perform it on the
intermediate 5. Thus, reaction of 5 with 4-bromo-N-methylpicolinamide afforded
compound 7a with good yield (45%) as depicted in Scheme 2. Subsequent urea formation
by reaction with 4-chloro-3-(trifluoromethyl) phenyl isocyanate gave the desired compound
1
(52% yield).

Cl
N
O
X
N
N
O N
OH
O N
OH
O N
O
2
2
2
b
c
+
X
NH2
N
H
N
H
O
O
3
4
X = OH
X = Cl
a
d
H
N
H
N
O
H
N
H
N
N
O
N
O
N
N
O
H N
O
R1
R1
2
f
e
O
O
N
H
N
H
O
6
a, R = 3-CF -4-Cl-Ph
1 3
5
N
6b, R = Ph
1
X
g
2
8
9
, R = 3-CF -4-Cl-Ph; X= H
1 3
, R = 3-CF -4-Cl-Ph; X= NH
2
1
3
H
H
N
O
, R = Ph; X = H
1
O
N
N
N
O
N
14
Scheme 1. Synthetic route for compound 2 and depicted analogues. Reagents and
conditions: a) oxalyl chloride, CH Cl , rt, 4 h, 100%; b) DIPEA, THF, 0 ºC, 2 h, 70%; c)
2
2
NaOH, DMF, 130 ºC, 90 min, 76%; d) Fe, AcOH, EtOH, H O, 80 ºC, 1 h, 72% or H , Pd/C
2
2
1
0 wt%, EtOH, rt, 16 h, 71%; e) R -isocyanate, 1,4-dioxane, 55 ºC, 2 h, ~ 30%; f) N,N’-
1
dimethylethylendiamine, CuI, K CO , 4-halopyridine, 1,4-dioxane, 120 ºC, 24 h, ~ 10%; g)
2
3
4
-(chloromethyl)pyridine hydrochloride, NaH, DMF, rt to 100 ºC, 2 h, 16%.
To the best of our knowledge, there are not previous reports of pyrido[2,3-b]
[1,5]benzoxazepine-5(6H)-ones with this pattern of substitution.
Next, we explored the fragment of designed compounds targeting the deep pocket of
CDK8. The urea derivatives (compounds 9−11), amides (16−19) and sulfonamide (20)

were prepared. Originally, compound 9 was first synthesized from 5 (Scheme 1) although
with rather low yield (3%). However the synthesis of 9 was significantly improved
following Scheme 2 where coupling step was performed prior to the final urea formation
reaction (70% yield from 5). Based on these results we decided to apply the same synthetic
procedure for the preparation of the other “deep pocket targeting analogues”.
Starting from compound 5, the 4-pyridyl analogue 7b was obtained with 90% yield using
the coupling conditions cited before. Subsequent reaction with several isocyanates rendered
the final ureas 9
formed by reaction of 7b with benzene sulfonyl chloride in good yield (88%) and amide
derivatives 16 19 were obtained in moderate yields under standard conditions with the
−11 in moderate to good yield (49-89%). Sulfonamide derivative 20 was
−
corresponding acid chlorides.
We also focused our efforts on the exploration of the pyridyl hinge targeting moiety. 2-
aminopyridinyl fragment was introduced by copper coupling reaction of 6a with 2-amino-
4
-bromopyridine to give compound 8 (Scheme 1) with low yield but in sufficient amount
for biological evaluation. Compound 14 was synthesized from 6b according to Scheme 1
by alkylation reaction with 4-(chloromethyl)pyridine in moderate yield. Other hinge
targeting heteroaromatic rings were prepared from intermediate 5 by coupling reactions
with 3-bromopyridine and 3-bromofurane to afford intermediate 7c-d which rendered final
compounds 12−13 after urea formation with phenyl isocyanate (Scheme 2).

N
O
H
H
N
O
N
O
H N
O
N
N
O
H N
O
2
2
R2
a
b
O
N
N
N
H
R1
R1
1
, R = 2-CONHMe-4-pyridyl;
7
7
7
7
a, R = 2-CONHMe-4-pyridyl, (Y: 45%)
1
1
R = 3-CF -4-Cl-Ph (Y: 52%)
2
3
b, R = 4-pyridyl (Y: 90%)
1
9, R = 4-pyridyl; R = Ph (Y: 78%)
1
2
c, R = 3-pyridyl (Y: 90%)
1
5
1
1
1
1
0, R = 4-pyridyl; R = 2-F-Bn (Y: 40%)
1
2
d, R = 3-furyl (Y: 80%)
1
1, R = 4-pyridyl; R = n-Pr (Y: 69%)
1
2
2, R = 3-pyridyl; R = Ph (Y: 89%)
3, R = 3-furyl; R = Ph (Y:65%)
1
2
1
2
c or d
e
7b
H
N
O
N
O
R3
N
O
N
O
H
N
O
N
S
O
O
N
N
1
6, R = Ph (Y: 48 %)
3
20
17, R = Bn (Y: 77 %)
3
F3C
18, R3=
(Y: 10%)
OH
19, R3=
(
Y: 18%)
OH
Scheme 2. Synthetic route for depicted analogues. Reagent and conditions. a) R Br or R I,
1
1
CuI, N,N’-dimethylethylendiamine, Cs CO , 1,4-dioxane, 120 ºC, 20 h; b) R2NCO, 1,4-
2
3
dioxane, 55 ºC, 20 h; c) R
3
COCl, K
2
CO
3
3 2
, 1,4-dioxane, rt, 2 h (for 16 and 17); d) R CO H,
thionyl chloride, THF, DMAc, rt, 90 min, then 7b added, 70 ºC, 3 h (for 18 and 19); e)
benzenesulfonyl chloride, pyridine, 0 ºC, 1 h, 89%.
2
.2. Biological evaluation
Next, we evaluated the inhibitory activity (IC50 value, half maximal inhibitory
concentration) of the prepared compounds against CDK8. We were pleased to find that
compound 2 displayed a potent inhibition of CDK8 with an IC50 in the single digit

nanomolar range (8.25 nM), about 4-fold better than Sorafenib (36.6 nM in our hands, 32.5
nM in ref. [13]). This result is the experimental proof that our hypothesis in designing this
tricyclic molecule as CDK8 inhibitor was correct. Moreover, it also supports the notion that
the tricyclic inhibitor 2 should induce a displacement of the P-loop in order to avoid steric
clashes mentioned above and to achieve its final accommodation in the proposed binding
site. Surprisingly, compound 1 displayed moderate CDK8 inhibition (IC = 361 nM). It
5
0
could be explained by minor changes in the orientation of these tricyclic inhibitors in
comparison with Sorafenib in the catalytic site of CDK8 rendering a not fully paralleled
SAR for both classes of compounds.
Furthermore, we generated a concise SAR around these novel tricyclic CDK8 inhibitors
by exploring the activity of deep pocket targeting modifications (compounds 9
−11) and
alternative hinge binding moieties (compounds 8, 12−14,). The results are summarized in
Table 1.
Compound 9, with a phenyl urea fragment, demonstrated to be a potent CDK8 inhibitor
IC = 9.96 nM) in the range of compound 2, indicating that the hydrophobic interactions
(
5
0
made by the 3-trifluoromethyl-4-chlorophenyl ring with Phe176 in Sorafenib are still
achieved with the non-substituted phenyl urea as well as the bidentate interaction of the NH
groups with Glu66.
To check the size of the lipophilic cavity, the compound 10 bearing an o-F-benzylic urea
was prepared resulting in a less potent compound with IC50 value of 324 nM. This result
suggested that the position of phenyl ring in these molecules is important to maintain a
strong inhibition against CDK8. The phenyl ring of compound 10 seems to be allocated in a
less optimal position to reach the lipophilic interaction with Phe176. The aliphatic urea 11

resulted in > 500-fold loss in potency corroborating the importance of the interactions
achieved by the phenyl ring.
Table 1
CDK8 IC data for compounds 8−15
5
0
X
H
H
O
N
N
N
R2
O
O
R1
CDK8 IC50
Cpd
X
N
N
R1
R2
(nM)
N
Cl
8
9
57.5
CF3
H N
2
N
9.96
F
N
N
10
11
12
N
N
N
324
5060
4080
N
13
N
593
O
1
4
5
N
C
N
1780
1150
N
1

Since phenyl urea 9 displayed potent activity, we next focused our efforts on the
replacement of its hinge binding moiety 4-pyridine by other heteroaryls (compounds 8,
1
2−
14, Table 1). The exchange of the distal 4-pyridyl by 3-pyridyl (12) or by a 4-pyridine
homologated with a methylene spacer (14) decreased significantly the activity to the
micromolar range (IC50 values of 4.08 M and 1.78 M respectively). The 3-furyl
derivative 13, although was better tolerated showed also a decrease in activity (IC = 593
µ
µ
5
0
nM). The replacement by a bidentate moiety such as 2-amino-4-pyridyl resulted in a potent
compound (8) with an IC value of 57.5 nM. However, the comparison with compound 2
5
0
(
IC =8.25 nM) demonstrated a ~ 7-fold loss of potency pointing to the impossibility of
50
compound 8 to establish an additional interaction with the C=O backbone group of Ala100.
Moreover the presence of the 2-amino group seems to weaken the interaction of the
acceptor N of the pyridine moiety with the NH backbone group of the same residue.
The increase in potency of compound 2 with regard to Sorafenib and the docking studies
suggested that a new interaction could be established through the N of the fused pyridine of
the tricyclic scaffold and the catalytic Lys52 in CDK8. To determine its contribution we
synthesized compound 15 (Scheme S1), where the pyridine was replaced by a fused phenyl
ring. Interestingly, the CDK8 activity dropped to the micromolar range (Table 1) indicating
that the pyridine N atom of the tricyclic scaffold is playing an important role for the CDK8
inhibitory activity in these pyrido[2,3-b][1,5]benzoxazepin-5(6H)-one derivatives and
pointed to the establishment of an additional interaction with the protein.
Finally, we decided to replace the urea functionality (responsible for a bidentate
interaction with Glu66 in Sorafenib) by other functionalities that could keep at least one of
the NH groups for interaction with Glu166. The amides 16−19 were synthesized covering a

small range of substituents (Table 2). The benzamide 16 exhibited an IC50 value of 2.4 µM
against CDK8, which was approximately 250 fold less potent than its urea counterpart (9).
Similar inhibitory activity was obtained with the amide 17 which was designed with the
hope that the benzylic moiety could place the phenyl ring at the same distance as it is in the
potent urea 9. We additionally incorporated a tertiary trifluoromethyl-methyl alcohol
moiety previously reported as urea isostere [17, 18] in compound 18. The corresponding
methylated analogue 19 was prepared as a control for comparison purposes. The
sulfonamide 20 was also synthesized. Unfortunately all these modifications (18−20) led to
compounds with inhibitory activity in the micromolar range. All these results could be
rationalized by the loss of the bidentate interaction of the urea group of inhibitor 9 with
Glu66 and/or the impossibility to achieve the lipophilic interactions made by its phenyl ring
inside the deep pocket.
Table 2
CDK8 IC activity of compounds 16
−20
5
0
N
H
O
N
N
R2
O
N
Cpd
R
2
IC50
(
µM)
Cpd
R
2
IC50 (µM)
O
1
6
2.4
17
2.5
O

O
S
O
1
2
8
0
CF3
>10
>10
19
5.1
HO
HO
O
O
2
.3. Crystal Structure
After this SAR exploration the most potent CDK8 inhibitor 2 was selected for co-
crystallization studies with CDK8-CyC [19]. The crystal structure at a resolution of 2.8 Å
confirmed the expected binding mode in CDK8 (Figure 4, PDB: 6TPA). The tricyclic
inhibitor occupied the ATP binding site of CDK8 and prolonged the urea side chain to the
typical allosteric pocket induced by classical type II inhibitors such as Sorafenib, where the
protein adopted the corresponding DMG-out conformation. CDK8 inhibitor 2 established a
key H-bond between the N of its 4-pyridyl moiety and the backbone NH of Ala110 in the
hinge area of CDK8. Additionally, the urea portion of this molecule interacted with CDK8
by the formation of H-bonds with Glu66 side chain and Asp173 NH-backbone residue in
the DMG area. On the other hand, the N of the fused pyridine of the tricyclic scaffold
established another H-bond with the flexible Lys52 side chain. Moreover, a π-π interaction
between Phe97 and the fused aromatic ring of its tricyclic scaffold were observed.
Additional pi-sigma, pi-anion and other hydrophobic interactions between CDK8 and
inhibitor 2 are established (a full detailed diagram is provided in Figure S2).

Figure 4. Close view of co-crystal structure of compound 2 and CDK8 highlighting H-
bond interactions.
The co-crystal structure showed electron density in the P-loop area which allowed the
resolution of its backbone as well as the orientation of the flexible Tyr32 residue.
Unfortunately, the high flexibility of the P-loop did not led to a clear electron density for
some side-chains, such as Val27, Arg29 and Thr31 which were introduced as glycines for
crystal structure solution purposes. In spite of this, the main-chain was clear and well-
defined along the whole loop.
Importantly, the accommodation of inhibitor 2 in the pocket was achieved by tilting its
structure in comparison with Sorafenib to avoid the potential steric clashes suggested in
Figure 1 with residues Val35 and Lys52 and by the induction of a clear displacement of the
whole P-loop substructure to open the binding pocket cavity. The flexible Tyr32 residue
was positioned towards the inner area of the ATP catalytic site roofing the inhibitor.

The displacement of the P-loop induced by CDK8 inhibitor 2 is remarkable in
comparison with the aligned structure 4F6W in which the orientation of Tyr32 is also
similar. This effect can be visualized in Figure 5.
Figure 5. Close view of co-crystal structure of compound 2 and CDK8 (blue) and 4F6W
and its inhibitor (orange). Key residues are highlighted to observe alignment of the proteins
and the induced displacement of the P-loop by compound 2 denoted by Tyr32 position in
each structure.
2
.4. Selectivity
A potential effect of filling the cavity flanked by residues Tyr32-Val35-Ala50-Lys52 of
CDK8 and the observed P-loop displacement induced by the designed type II class of
tricyclic inhibitors could be an improvement of its on-target selectivity in comparison with

Sorafenib. It is plausible that the additional fused ring systems of these compounds could
impair their binding to a wider number of kinases than the more adaptable “monocyclic
Sorafenib”. In order to assess this hypothesis, compound 2 was screened at 1.0
µM
concentration (~150 fold higher than its IC against CDK8) in a panel of 456 kinases using
5
0
TM
a high-throughput competition binding assay (KINOMEscan ). Compound 2 exhibited a
signal below 35% of the produced by the untreated controls in this assay (a value widely
accepted for a significant binding in this platform) for only 13 kinases: CDK8 and its
paralog CDK19 (also named CDK11), CDK7, CLK2, CLK4, DDR1, ERK8, HIPK4,
IRAK1, IKKα, JNK2, LOK and PLK3. The complete data set is provided in Table S1 in
supplementary material. The calculated selectivity score S(35) of compound 2 is 0.03,
indicating a high level of selectivity for a kinase inhibitor [20].
TM
To easily visualize these results, the TREEspot interaction map is given (Figure 6A),
which represents each kinase activity versus control after incubation with compound 2.
Furthermore, inhibitor 2 was profiled internally against CDK19 and the off-targets
CDK7, CLK4 and DDR1 in dose response experiments using orthogonal assays for
validation purposes. As expected, compound 2 demonstrated a potent activity against
CDK19 (IC50 = 29 nM), due to the high homology degree between both paralogs.
Interestingly, compound 2 displayed negligible activity against CDK7 and CLK4 (IC50 > 10
µ
M) and moderated DDR1 inhibition (IC = 0.180
µM).
5
0
The promising kinase selectivity results of inhibitor 2 prompted us to compare them with
those of Sorafenib. A similar broad kinase selectivity profiling for Sorafenib using
TM
KINOMEscan platform was published using the inhibitor at 10 µM concentration [21].
However, for a direct comparison with compound 2, we screened Sorafenib at the same

concentration (1 µM) in the same platform against a panel of 468 kinases (Figure 6B, Table
S2). Sorafenib exhibited a selectivity score S(35) of 0.1 and 41 kinases (including CDK8
and CDK19) showed a signal below 35% of the produced by the untreated controls in this
assay. Two of these kinases were not included in the original panel of 456 kinases of
compound 2. After a comparative analysis, it was reasonable to conclude that inhibitor 2
was significantly more selective for CDK8/19 than Sorafenib, the origin of its design.
Therefore, the presence of the tricyclic pyrido[2,3-b][1,5]benzoxazepin-5(6H)-one scaffold
had a beneficial impact for its selectivity, although contributions of the hinge binding
moieties in 2 (4-pyridyl) and Sorafenib (2-picolinamide) cannot be ruled out.

Figure 6. Selectivity profile of A) compound 2 and B) Sorafenib. Kinases found to bind
percent of control <35) are marked with red circles, where larger circles indicate higher-
(
affinity binding. CDK8 and CDK19 are highlighted in blue color. Image generated using
TM
TREEspot Software Tool and reprinted with permission from KINOMEscan®, a division
of DiscoveRx Corporation, © DISCOVERX CORPORATION 2010.
2
.5. Cellular activity
Ser727
Next, we examined the modulation of phospho-STAT1
in cells as an in vivo
biomarker for CDK8 inhibition [22], by our inhibitors (2, 8 and 9) and Sorafenib. We only
observed modulation of STAT1 phosphorylation in SW620 cells following 8 h treatment
with compound 2, showing moderate inhibition (IC = 1.3
µ
M; maximum inhibition of ~
M ) whereas compound 8, 9 and Sorafenib showed negligible activity (IC50
M) (Figure S3).
In spite of the good biochemical activity, these compounds and Sorafenib presented
5
0
6
1
0% at 10
µ
>
0
µ
disappointing CDK8 cellular activity. Our results were in line with those reported
previously for other type II CDK8 inhibitors [11,13] where a similar and strong
disconnection between biochemical and cellular CDK8 inhibition was observed. The
explanation of this behavior was based on the notion that the inactive form of CDK8
targeted by this type of inhibitors was poorly available in cells. A very recent article [23]
describes the molecular mechanism by which CDK8 is blocked in its DMG in (active)
conformation when is part of the mediator complex due to its association with MED12. The
preference for a DMG-in conformation of the active site disfavors type II kinase inhibitors

from binding and inhibiting CDK8 in ternary CDK8/Cyclin C/MED12 complexes, which
are present in cells [23].
We alternatively determined the cellular inhibition of the off-target DDR1 kinase by
compound 2. Interestingly, 2 was able to block collagen induced DDR1
autophosphorylation in transfected HEK293T cells with DDR1 (IC p-DDR1 value of 170
5
0
nM, Figure S4). This result triggered the possibility of positioning the pyrido[2,3-
b][1,5]benzoxazepin-5(6H)-one derivatives as starting points to develop novel cell active
DDR1 inhibitors. As a matter of information, DDR1 kinase is involved in several diseases
such as cancer (tumor progression and metastasis), lung and liver fibrosis,
glomerulonephritis and atheroesclerosis among others [24].
3
. Conclusions
Tricyclic pyrido[2,3-b][1,5]benzoxazepin-5(6H)-one derivatives were designed,
synthesized and evaluated for their activity towards CDK8. This led to the identification of
the most potent inhibitor of the series, compound 2 with IC50 of 8.25 nM, 4-fold more
potent than Sorafenib, the compound used for its design. Moreover, compound 2 showed a
remarkable selectivity over a panel of 456 kinases. Co-crystal studies showed similar
binding mode than Sorafenib, with the induction of displacement of P-loop to
accommodate compound 2 in the ATP cavity. It has been demonstrated that rational
modifications of the scaffold of the multi-kinase inhibitor Sorafenib leading to tricyclic
pyrido[2,3-b][1,5]benzoxazepin-5(6H)-one derivatives, exemplified by compound 2, is an
attractive strategy to improve its CDK8 kinase activity and selectivity. Unfortunately, and
in line with other reported type II CDK8 inhibitors, the novel compounds were unable to
strongly modulate the activity of CDK8 in cells. Last but not least, the promising

biochemical and cellular activity against DDR1 of representative compound 2 would allow
its use as starting point to develop new DDR1 inhibitors.
4
. Experimental section
4
.1. General Procedures
All reagents and solvents were purchased from commercial sources and were used
1
13
without further purification. H NMR spectra were recorded at 300 or 700MHz and C
NMR at 75MHz using Bruker instruments operating at indicated frequencies. Chemical
1
13
6
shifts are given in ppm relative to DMSO-d residual peak ( H, 2.50 ppm; C, 39.51 ppm).
The following abbreviations are used for the description: br = broad, s = singlet, d =
doublet, dd = doublet of doublets, t = triplet, q = quartet, m = multiplet. Purification by
silica flash chromatography was performed using automated Biotage systems with pre-
packed cartridges. Chemical purities for final compounds were > 95% measured by LC/MS
analysis at UV (254 nm), Gemini-NX C18 (100 x 2.0 mm; 5µm), Solvent A: water with
0
.1% formic acid; Solvent B: acetonitrile with 0.1% formic acid. Gradient: 5% of B to
1
00% of B within 8 min at 50 ºC on Agilent 1200/6120 Quadrupole system. Purifications
by preparative HPLC were performed on Agilent 1100; HT Zorbax SB-C18 (100 x 2.12
mm; 5µm) with appropriate gradient (solvent A: water with 0.1% formic acid; Solvent B:
acetonitrile with 0.1% formic acid). Other abbreviations used along the document: rt =
room temperature, DCM = dichloromethane, DMAc = dimethylacetamide, DMF = N,N-
dimethylformamide, THF = tetrahydrofuran, Ar = Argon, eq = equivalent, Rt = retention
time, TLC = thin layer chromatography.
4
.1.1. Synthesis of compounds following Scheme 1.

4
.1.1.1. 2-Chloro-N-(2-hydroxy-4-nitro-phenyl)-nicotinamide (
-Hydroxynicotinic acid (4 g, 17.97 mmol) was suspended in dichloromethane (115 mL)
and DMF (1.15 mL) and then oxalyl chloride (2M in CH Cl solution, 3 eq, 43 mL) was
3)
2
2
2
added slowly under Ar flow (gas evolved). The mixture was stirred at rt till completion of
reaction by LC-MS, 4 h. The reaction mixture was concentrated under vacuum and further
dried in vacuum pump. The crude product, 2-chloronicotinoyl chloride, was used in next
step assuming quantitative yield. A suspension of 2-chloronicotinoyl chloride (3.16 g, 17.95
mmol) in THF (72 mL) was added to a mixture of 2-amino-5-nitrophenol (1.15 eq, 3.18 g)
and DIPEA (1.7 eq, 5.3 mL) in THF (72 mL) at 0 ºC. The reaction mixture was stirred at 0
ºC for 2 h and then allowed to get rt. The solids in suspension were filtered out and washed
with ethyl acetate. The filtrate was washed with water and brine. The organic layer was
dried (Na SO ) and concentrated to render a solid that was triturated with CH Cl affording
2
4
2
2
1
3
1
6
4
(brownish solid, 3.76 g, 70%); H NMR (700 MHz, DMSO) δ 8.68 (dd, J = 4.7, 2.0 Hz,
H), 8.64 (m, 1H), 8.10 (d, J = 2.5 Hz, 1H), 7.97 (dd, J = 9.1, 2.5 Hz, 1H), 7.67 (m, 1H),
+
.85 (m, 3H). LCMS (ESI): m/z 212 [M+ H] ; Rt: 4.05 min.
.1.1.2. 7-Nitro-10H-5-oxa-4,10-diaza-dibenzo[a,d]cyclohepten-11-one (4)
Intermediate 3 (3.74 g, 12.74 mmol) was dissolved in DMF (85 mL) and NaOH (1.5 eq,
.76 g, powdered pellets) was added. The mixture in pressure tube was heated at 120 ºC for
0 min. The reaction mixture was poured into ice-water. The solid was filtered, washed
2
9
with water and triturated with methanol-Et O to render compound 4 (off-white solid, 2.1 g,
2
pure required product). From the mother liquor, another batch of product precipitated which
was filtered and washed with water affording 400 mg of pure compound 4 (total amount 2.5
1
g, 76%); H NMR (700 MHz, DMSO)
δ 11.30 (s, 1H), 8.57 (dd, J = 4.7, 2.0 Hz, 1H), 8.31

(dd, J = 7.5, 1.9 Hz, 1H), 8.15 (dd, J = 8.8, 2.5 Hz, 1H), 8.11 (d, J = 2.5 Hz, 1H), 7.53 (dd, J
+
=
7.5, 4.8 Hz, 1H), 7.40 (d, J = 8.8 Hz, 1H); LCMS (ESI): m/z 258 [M+ H] , Rt: 3.48 min.
4
.1.1.3. 7-Amino-10H-5-oxa-4,10-diaza-dibenzo[a,d]cyclohepten-11-one (
Intermediate 4 (2.05 g, 7.97 mmol) was suspended in ethanol (80 mL) and Pd/C 10% wt
400 mg) was slowly added. Then, hydrogen balloon was connected to the reaction mixture
5)
(
and it was stirred at rt. After 3 h more Pd/C was added (100 mg) and the mixture was stirred
overnight. The reaction mixture was diluted with ethanol and refluxed for 2 h (in order to
dissolve maximum amount of product) and then it was filtered through Watsman® paper.
The filtrate was concentrated to afford pure compound 5 (pale brown solid, 1.29 g, 71%);
1
H NMR (300 MHz, DMSO)
δ 10.30 (s, 1H), 8.46 (dd, J = 4.8, 2.0 Hz, 1H), 8.21 (dd, J =
7
.6, 2.1 Hz, 1H), 7.42 (dd, J = 7.6, 4.8 Hz, 1H), 6.82 (d, J = 8.5 Hz, 1H), 6.51 (d, J = 2.4
+
Hz, 1H), 6.39 (dd, J = 8.5, 2.4 Hz, 1H), 5.24 (s, 2H); LCMS (ESI): m/z 228 [M+ H] , Rt:
0
.8 min. When the reaction was performed at scale below 1 g, the next conditions also
worked well: Intermediate 4 (1 eq, 400 mg, 1.55 mmol) was suspended in EtOH-H O 1:1
15 mL) and then Fe (5 eq, 434 mg) and acetic acid (1.5 eq, 0.14 mL) were added. The
2
(
mixture was heated at 80 ºC for 1 h. The reaction mixture was filtered through Celite® and
rinsed thoroughly with methanol. The filtrate was concentrated and triturated with methanol
rendering the compound 5 (255 mg, 72%). Removal of iron salts from the low soluble
required product was difficult and messy at scale bigger than 1 g.
4
.1.1.4. 1-(4-Chloro-3-trifluoromethyl-phenyl)-3-(11-oxo-10,11-dihydro-5-oxa-4,10-diaza-
dibenzo[a,d]cyclohepten-7-yl)-urea (6a
)
Intermediate 5 (550 mg, 2.42 mmol) was suspended in 1,4-dioxane (20 mL) and 4-
chloro-3-(trifluoromethyl)phenyl isocyanate (536 mg, 2.42 mmol) was added. The mixture

under Ar atmosphere was heated at 55 ºC for 2 h. The mixture was concentrated on Celite®
and loaded into flash silica column. Elution with gradient c-Hex-EtOAc 25-100% of EtOAc
1
afforded compound 6a (white solid, 315 mg, 29%); H NMR (300 MHz, DMSO)
δ 10.65
(s, 1H), 9.22 (s, 1H), 9.02 (s, 1H), 8.51 (m, 1H), 8.27 (m, 1H), 8.10 (s, 1H), 7.63 (s, 3H),
7
.46 (dd, J = 7.5, 4.9 Hz, 1H), 7.21 (d, J = 8.8 Hz, 1H), 7.10 (d, J = 8.7 Hz, 1H); LCMS
+
(
ESI): m/z 449 [M+ H] , Rt: 4.48 min.
4
.1.1.5.
phenyl-urea (6b
Intermediate 5 (32 mg, 0.141 mmol) was suspended in dichloromethane (1.5 mL) and
phenyl isocyanate (1.1 eq, 17 L) was added. The mixture was stirred at rt for 28 h and
then more isocyanate (1.5 eq, 23 L) was added stirring for additional 96 h. The mixture
1-(11-Oxo-10,11-dihydro-5-oxa-4,10-diaza-dibenzo[a,d]cyclohepten-7-yl)-3-
)
µ
µ
was concentrated and purified by HPLC to afford compound 6b (15 mg, white solid, 30%);
1
H NMR (700 MHz, DMSO)
δ 10.62 (s, 1H), 8.94 (s, 1H), 8.81 (s, 1H), 8.51 (dd, J = 4.7,
2
.0 Hz, 1H), 8.26 (dd, J = 7.5, 2.0 Hz, 1H), 7.66 (d, J = 2.2 Hz, 1H), 7.46 (m, 3H), 7.28 (t, J
=
7.9 Hz, 2H), 7.16 (dd, J = 8.6, 2.3 Hz, 1H), 7.09 (d, J = 8.6 Hz, 1H), 6.97 (t, J = 7.3 Hz,
+
1
H); LCMS (ESI): m/z 347.2 [M+ H] , Rt: 4.27min.
4
.1.1.6. 1-(4-Chloro-3-trifluoromethyl-phenyl)-3-(11-oxo-10-pyridin-4-yl-10,11-dihydro-5-
oxa-4,10-diaza-dibenzo [a,d]cyclohepten-7-yl)-urea (
2)
Intermediate 6a (75 mg, 0.167 mmol), 4-iodopyridine (1.5 eq, 51 mg) and K CO (2.0 eq,
2
3
4
6 mg) were suspended in 1,4-dioxane (2 mL). Argon was bubbled into the suspension for
several minutes and then, CuI (0.1 eq, 2mg) and N,N’-dimethylethylenediamine (0.2 eq,
L, one drop of glass pipette) were added. The reaction mixture, in pressure tube, was
degassed and filled with Ar before heating at 120 ºC. After 6 h, more reactants were added
4
µ

(
4-iodopyridine, 20 mg; K2CO , 20 mg, CuI, 2 mg, dimethylethylenediamine, 1 drop) and
3
heating was continued up to 15 h. The reaction mixture was concentrated into Celite® and
loaded into flash silica column. Elution with gradient EtOAc and MeOH from 0-5% of
MeOH afforded compound 2 still impure, which was purified by preparative HPLC to
1
render pure compound 2 (white solid, 8 mg, 9%); H NMR (300 MHz, DMSO)
δ 9.55 (s,
1
1
H), 9.47 (s, 1H), 8.72 (s, 2H), 8.54 (dd, J = 4.8, 2.0 Hz, 1H), 8.34 (dd, J = 7.6, 1.9 Hz,
H), 8.10 (d, J = 2.2 Hz, 1H), 7.80 (d, J = 2.3 Hz, 1H), 7.64 (m, 2H), 7.52 (m, 3H), 7.16
1
3
(
dd, J = 8.9, 2.4 Hz, 1H), 6.77 (d, J = 8.8 Hz, 1H). C NMR (75 MHz, DMSO)
δ
163.42,
1
1
63.10, 152.36, 152.27, 152.11, 151.04, 148.61, 143.17, 139.06, 138.76, 132.04, 127.30,
26.24, 123.34, 123.25, 123.03, 122.66, 120.64, 116.58, 111.36. LCMS (ESI): m/z 526.1
+
[
M+ H] , Rt: 5.03 min.
4
.1.1.7. 1-[10-(2-Amino-pyridin-4-yl)-11-oxo-10,11-dihydro-5-oxa-4,10-diaza-dibenzo[a,d]
cyclohepten -7-yl]-3-(4-chloro-3-trifluorome- thyl-phenyl)-urea (
8)
Intermediate 6a (50 mg, 0.111 mmol), 2-amino-4-bromopyridine (1.5 eq, 0.167 mmol, 29
mg), K CO (2.5 eq, 38mg) and CuI (0.2 eq, 4 mg) were suspended in dioxane (1 mL).
2
3
Argon was bubbled and then N,N’-dimethylethylenediamine (0.5 eq, 6 µL) was added. The
reaction mixture, in pressure tube, was degassed and filled with Ar before heating at 120
ºC. After 6 h more reactants were added bromopyridine (15 mg), K CO (20 mg) and CuI
2
3
(2 mg) stirring the reaction up to 24 h. The reaction mixture was concentrated into Celite®
and loaded into flash silica column. Elution with gradient EtOAc and MeOH from 0-10%
of MeOH afforded compound 8 still impure, which was purified by preparative HPLC to
1
render pure compound 8 (white solid, 5 mg, 9%); H NMR (300 MHz, DMSO)
δ 10.39 (s,
1
H), 10.27 (s, 1H), 8.51 (dd, J = 4.8, 2.0 Hz, 1H), 8.29 (dd, J = 7.6, 2.0 Hz, 1H), 8.13 (d, J