Structure–Activity Relationships of Hexahydrocyclopenta[c]quinoline Derivatives as Allosteric Inhibitors of CDK2 and EGFR

Article abstract of DOI:10.1002/cmdc.201800687

Following the discovery of a type III allosteric modulator of cyclin-dependent kinase 2 (CDK2) characterized by a hexahydrocyclopenta[c]quinolone scaffold, three different series of its derivatives were synthesized and biologically evaluated. Docking of t

Full text of DOI:10.1002/cmdc.201800687

DOI: 10.1002/cmdc.201800687
Full Papers
Structure–Activity Relationships of
Hexahydrocyclopenta[c]quinoline Derivatives as Allosteric
Inhibitors of CDK2 and EGFR
Luca Carlino+,[a] Michael S. Christodoulou+,[a, b] Valentina Restelli,[c] Fabiana Caporuscio,[a]
Francesca Foschi,[a, b] Marta S. Semrau,[d] Elisa Costanzi,[e] Annachiara Tinivella,[a] Luca Pinzi,[a]
Leonardo Lo Presti,[b, f] Roberto Battistutta,[e] Paola Storici,[d] Massimo Broggini,[c]
Daniele Passarella,[b] and Giulio Rastelli*[a]
Following the discovery of a type III allosteric modulator of
cyclin-dependent kinase 2 (CDK2) characterized by a hexahy-
drocyclopenta[c]quinolone scaffold, three different series of its
derivatives were synthesized and biologically evaluated. Dock-
ing of the synthesized compounds into the allosteric pocket of
CDK2 allowed the elucidation of structure–activity relationships
(SARs). Moreover, the compounds were tested on the wild-
type epidermal growth factor receptor (EGFR) kinase domain
(KD) and its clinically relevant T790M/L858R mutant form.
Herein we describe the first SAR investigation of allosteric li-
gands that bind to the type III inhibitor pocket of CDK2 and
EGFR-KD. Although the activity of the synthesized inhibitors
needs to be improved, the obtained results provide clear-cut
indications about pharmacophore requirements and selectivity
determinants. Remarkably, this study led to the identification
of a selective T790M/L858R EGFR allosteric inhibitor that is in-
active toward both wild-type EGFR and CDK2. Finally, docking
into the T790M/L858R EGFR-KD led us to hypothesize that the
compounds bind to the double-mutant EGFR-KD by adopting
a binding mode different from that in CDK2, thus rationalizing
the observed selectivity profile.
Introduction
Due to their pivotal role in the modulation of cell pathways,
protein kinases (PKs) are the targets of choice for a number of
human diseases.[1] The standard approach of targeting the ATP
binding site has recently come up against selectivity and drug
resistance issues, which can be considerably decreased by tar-
geting allosteric pockets.[2] In fact, allosteric modulators of pro-
tein kinases bind to less conserved binding sites thus allowing
to bypass drug resistance issues clinically observed with ATP-
competitive drugs.[3] Moreover, they do not have to compete
with the high intracellular levels of ATP, as is the case with
type I, I1/2, and II inhibitors.
We have recently discovered that the hexahydrocyclopen-
ta[c]quinoline derivative 3 f (Figure 1) is an allosteric inhibitor
of cyclin-dependent kinase 2 (CDK2).[4] Further investigations
by X-ray crystallography, stereoselective synthesis of its pure
enantiomers, modeling studies, biological evaluations and mu-
tagenesis experiments provided the first thorough chemical
and biological characterization of an allosteric modulator of
this protein, and confirmed that 3 f binds to CDK2 via a type III
allosteric mechanism.[5]
[a] Dr. L. Carlino,+ Dr. M. S. Christodoulou,+ Dr. F. Caporuscio, Dr. F. Foschi,
Dr. A. Tinivella, Dr. L. Pinzi, Prof. G. Rastelli
Dipartimento di Scienze della Vita, Universitꢀ degli Studi di Modena e
Reggio Emilia, Via Campi 103, 41125 Modena (Italy)
E-mail: giulio.rastelli@unimore.it
[b] Dr. M. S. Christodoulou,+ Dr. F. Foschi, Dr. L. Lo Presti, Prof. D. Passarella
Dipartimento di Chimica, Universitꢀ degli Studi di Milano, Via Golgi 19,
20133 Milano (Italy)
[c] Dr. V. Restelli, Dr. M. Broggini
IRCSS Istituto di Ricerche Farmacologiche Mario Negri, Via La Masa 19,
20156 Milano (Italy)
In this study, we explored the
[d] M. S. Semrau, Dr. P. Storici
structure–activity
relationships
Structural Biology Lab, Elettra Sincrotrone Trieste S.C.p.A., SS 14 km 163.5,
AREA Science Park, 34149 Trieste (Italy)
(SARs) of this class of compounds
by synthesizing and testing three
series of differently substituted de-
rivatives (series 3a–f, i, series
3g,h, k–q, 4a,b, 5a,b, 7, and
series 3j,r–y). Moreover, as the
CDK2 mechanism of activation
was proposed to mimic that of ep-
idermal growth factor receptor
(EGFR)[6–8] and the two proteins
[e] Dr. E. Costanzi, Dr. R. Battistutta
Dipartimento di Scienze Chimiche, Universitꢀ degli Studi di Padova, Via
Marzolo 1, 35131 Padova (Italy)
[f] Dr. L. Lo Presti
Istituto di Scienze e Tecnologie Molecolari, Consiglio Nazionale delle Ricer-
che, Via Golgi 19, 20133 Milano (Italy)
[+] These authors contributed equally to this work.
Supporting information and the ORCID identification number(s) for the
author(s) of this article can be found under:
https://doi.org/10.1002/cmdc.201800687.
Figure 1. Chemical structure
of compound 3 f.
ChemMedChem 2018, 13, 1 – 9
1
ꢀ 2018 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim
&
These are not the final page numbers! ÞÞ

Full Papers
show similar inactive conformations,[9] the compounds were
also tested on the kinase domains of wild-type EGFR and the
T790M/L858R mutant. The latter double mutant is of particular
clinical relevance, because it constitutively activates the full-
length EGF receptor and is a key driver of drug resistance to
EGFR tyrosine kinase inhibitors gefitinib, erlotinib, and afati-
nib.[10,11] Induced-fit docking simulations of the synthesized
compounds to the CDK2 and the wild-type and double-mutant
EGFR allosteric pockets allowed us to rationalize and discuss
the obtained SARs.
Results and Discussion
Chemistry
Scheme 2. Reagents and conditions: a) MeNH2, HATU, DIPEA, THF, RT, over-
night, 85–90%; b) MeI (1.2 equiv), K2CO3, DMF, 808C, 4 h, 70%; c) MeI
(2.4 equiv), K2CO3, DMF, 808C, 4 h, 75%; d) NH2OTHP, HATU, Et3N, THF, RT,
overnight, 58%; e) HCl (3m), MeOH, RT, 3 h, 15%.
The synthesis of compounds 3a–y (Scheme 1, Tables 1–3) is
based on a multicomponent Povarov reaction to give com-
pounds 2a–y, followed by an addition reaction with ortho-phe-
nylsulfenyl chloride derivatives. In the Povarov reaction, TFA (a,
path i) was used as a catalyst in the presence of either elec-
tron-deficient or inactivated aldehydes 1 (synthesis of com-
pounds 2a–u).[5] On the other hand the use of AlCl3 (a, path ii)
is essential to promote both the imine formation and the cy-
cloaddition in the presence of electron-rich aldehydes (synthe-
sis of compounds 2v–y). For the synthesis of racemic com-
pounds 3a–y, 2-nitrobenzenesulfenyl chloride was purchased
from Alfa Aesar, whereas all the other ortho-phenylsulfenyl
chlorines were prepared according to published procedures.[12]
Racemic compounds 4a and 4b were prepared by a conden-
sation reaction of compounds 3 f and 3i, respectively, whereas
racemic compounds 5a and 5b were obtained from com-
pound 3m by using 1 and 2.5 equiv of MeI, respectively
(Scheme 2). Finally, the hydroxamic acid 7 was furnished by de-
protection of compound 6 in the presence of HCl (Scheme 2).
NMR studies confirmed the relative configuration of the ob-
tained molecules. Further X-ray diffraction experiments of a
single crystal of 3j confirmed the structure (see Experimental
Section and Supporting information).
tein–ligand complexes. The predicted binding mode was vali-
dated through mutagenesis experiments and direct binding
experiments on the wild-type and mutant CDK2 with micro-
scale thermophoresis.[5] As previously reported, 3 f was pro-
posed to bind to the allosteric pocket of CDK2 with the car-
boxylate moiety forming a salt bridge with the polar head of
the conserved Lys33 residue, and the nitro group making hy-
drogen bonds with the backbone NHs of the Asp145 and
Phe146 residues of the DFG motif (Figure 2).[5]
Therefore, the first series of compounds was designed to in-
vestigate the importance of the carboxylic acid and the nitro
group (Table 1). Biological evaluations clearly showed that the
removal of the carboxylic acid (3c and 3d), the nitro group
(3e), or both (3b and 3a), completely abolished binding to
CDK2. These results are consistent with the previously reported
mutagenesis studies that showed that mutation of Lys33 to an
alanine significantly impaired binding of 3 f to CDK2, and pro-
vided further validation of the proposed binding mode.[5]
Substitution with an additional chlorine atom in the para
position of the 2-phenyl ring attached to the tetrahydroquino-
line moiety as shown by compound 3i (IC50 =25.8Æ11.9 mm;
Figure S1N) and compound 3d did not affect the CDK2 inhibi-
tory activity. In conclusion, results obtained from this first
series of compounds indicated that the carboxylic acid and the
nitro group are mandatory for the CDK2 inhibitory activity.
Having ascertained that the carboxylic acid and the nitro
groups cannot be removed, the second series of compounds
Structure–activity relationships in CDK2
The progenitor molecule 3 f (IC50 =10.2Æ2.5 mm; Figure S1N)
was thoroughly characterized by X-ray crystallography and the
structural model of its interaction with CDK2 was obtained by
molecular docking to selected CDK2 crystal structures and ex-
tensive molecular dynamics simulations of the resulting pro-
Scheme 1. Reagents and conditions: a) path i, synthesis of 2a–u: various amines, cyclopentadiene, TFA, MeCN, 08C, 2 h, 50–90%; path ii, synthesis of 2v–y:
various amines, cyclopentadiene, AlCl3, MeCN, 08C, 8 h, 58–76%; b) various ortho-phenylsulfenyl chlorines, MeCN, RT, 30 min, 50–90%.
&
ChemMedChem 2018, 13, 1 – 9
www.chemmedchem.org
2
ꢀ 2018 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim
ÝÝ These are not the final page numbers!

Full Papers
Figure 2. Binding mode of compound 3 f within CDK2 obtained via induced fit docking of the compound to the 2EXM crystal structure (see Christodoulou
et al.[5] for details). The protein is represented by a pink surface and pink cartoons, whereas the residues forming the binding cavity are labeled with the
three-letter code and shown as pink sticks. Compound 3 f is represented by burgundy spheres and burgundy balls and sticks within the box.
Table 1. Structures and biological activity of the investigated compounds synthesized to explore the importance of the carboxylic acid and nitro groups.
Biological activities are reported as percent inhibition at 10 and 50 mm.
Compd
R1
R2
R3
R4
R5
CDK2
wild-type EGFR
L858R/T790M EGFR
10 mm 50 mm
10 mm
50 mm
10 mm
50 mm
3a
3b
3c
3d
3e
3 f
3i
H
H
H
H
H
H
Cl
Cl
Cl
Cl
Cl
Cl
H
H
H
H
H
H
H
H
H
H
Cl
H
H
Cl
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
12
29
6
26
4
0
0
0
0
0
0
0
0
NO2
NO2
H
NO2
NO2
H
0
COOH
COOH
COOH
28
81
55
52
56
65
66
38
28
10
18
was designed to investigate whether the two groups could be
replaced by different chemical moieties (Table 2). The acetic
acid derivative 3o partially retained CDK2 inhibitory activity. In-
duced-fit docking of this compound to the CDK2 allosteric site
showed that the acetic acid moiety of 3o forms a salt bridge
with the Lys33 residue and, in addition, it hydrogen bonds to
the backbone NHs of Thr14 and Tyr15 (Figure S1A). On the
contrary, the esterification of the carboxylic acid (3k and 3l) or
replacement of the carboxylic group with an acetyl group (3p)
provided inactive compounds. These modifications strongly af-
fected the mandatory electrostatic interaction with the Lys33
side chain without forming any additional interaction, and an
ethyl or a methyl moiety are accommodated in the proximity
of the polar region of the cavity, thus resulting in loss of activi-
ty. The modification of the carboxylic acid group with a car-
boxamide (4a and 4b), an acetamido group (3q), or a hydrox-
amate (7) provided compounds able to retain CDK2 inhibitory
activity. In the CDK2/4a and CDK2/4b complexes, the carbonyl
and the amino groups of the amide hydrogen bond to the
Lys33 side chain and the Lys34 backbone carbonyl, respectively
(Figure S1B). In the acetamido analogue 3q, the amino and
the carbonyl groups make hydrogen bonds with the Asp145
ChemMedChem 2018, 13, 1 – 9
www.chemmedchem.org
3
ꢀ 2018 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim
&
These are not the final page numbers! ÞÞ

Full Papers
Table 2. Structures and biological activity of the investigated compounds synthesized to explore the effects of replacement of the carboxylic acid and the
nitro group with various chemical moieties. Biological activities are reported as percent inhibition at 10 and 50 mm.
Compd
R1
R2
R5
CDK2
wild-type EGFR
L858R/T790M EGFR
10 mm 50 mm
10 mm
50 mm
10 mm
50 mm
3g
3h
3k
3l
COOH
COOH
COOEt
COOEt
SO2NH2
SO2CH3
CH2COOH
COCH3
NHCOCH3
CONHCH3
CONHCH3
SO2NHCH3
SO2N(CH3)2
CONHOH
OCH3
CF3
H
H
H
Cl
H
H
H
Cl
Cl
H
Cl
H
H
H
0
0
0
0
46
0
0
0
0
0
63
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
17
0
0
42
0
25
0
17
0
8
0
0
23
0
0
0
0
28
0
0
0
24
74
0
0
9
0
74
0
0
0
NO2
NO2
NO2
NO2
NO2
NO2
NO2
NO2
NO2
NO2
NO2
NO2
3m
3n
3o
3p
3q
4a
4b
5a
5b
7
33
0
81
0
16
55
57
23
18
30
13
56
62
41
21
82
0
0
0
0
0
0
0
0
0
0
side chain and with the Tyr15 and Gly16 backbone carbonyls,
respectively (Figure S1C). The hydroxamate moiety of 7 forms
hydrogen bonds with the Asp145 side chain (Figure S1D).
Moreover, the substitution of the carboxylic acid with sulfona-
mide moieties provided compounds whose activities are in the
order: SO2NH2 >SO2NHMe>SO2NMe2 (compounds 3m, 5a,
5b, respectively). The unsubstituted sulfonamide 3m (IC50 =
59.7Æ21.1 mm; Figure S1N) hydrogen bonds to the polar head
of Lys33 via both the SO2 and the amino groups (Figure S1E).
On the contrary, the methyl-substituted sulfonamides 5a and
5b only interact via the SO2 group, due to the steric repulsion
of the methyl or the dimethyl sulfonamide substituents (Fig-
ure S1F, S1G).
(3s) was detrimental for activity. This result is consistent with
the hydrophobic nature of this pocket (Figure 2). The replace-
ment of the phenyl ring with a naphthalene (3r) moiety was
also tolerated, whereas the introduction of a polar hydroxy
group in meta or para positions (3v, 3w, 3x, 3y) resulted in
less active or inactive compounds, in agreement with the hy-
drophobic character of this sub-pocket.
Structure–activity relationships in EGFR
All compounds were inactive on wild-type EGFR-KD (Tables 1–
3). Only two compounds (3w and 3y, Table 3) showed moder-
ate wild-type EGFR-KD inhibitory activity. The superimposition
of the structure of the CDK2–3 f complex[5] with the crystal
structure of wild-type EGFR-KD in the inactive conformation
(PDB ID: 3W32)[13] showed that compound 3 f, if adopting a
CDK2-like binding mode, would form severe steric clashes with
several residues of EGFR. These residues include Leu858,
Val786, Ile759, Met766, Leu861, and Leu747. Indeed, standard
induced-fit docking calculations of 3 f to the allosteric pocket
of wild-type EGFR-KD in the inactive conformation[13] failed to
produce allosteric binding modes. Induced-fit docking with an
extended sampling protocol (see the Experimental Section)
generated allosteric binding modes that were quite different
from those observed in CDK2. In particular, none of the gener-
ated binding modes showed the carboxylate and the nitro
moieties, respectively engaged in interactions with the Lys745
(corresponding to Lys33 in CDK2) side chain and with the NH
backbone atoms of Asp855 (Asp145) and Phe856 (Phe146), as
observed in CDK2. Altogether, these findings provided an ex-
planation for the observed lack of activity of the investigated
compounds on wild-type EGFR.
For the sulfonylmethane derivative 3n a similar behavior is
also predicted (Figure S1H). Finally, the modification of the
nitro substituent of 3 f with a methoxy (3g) or a trifluorocar-
bon (3h) group completely abolished the inhibitory activity.
This finding, combined with the observed lack of activity of
3a, 3b, and 3e, in which the nitro group was removed, sug-
gested that the interaction of the nitro group with the Asp145
and Phe146 NH groups of the DFG motif (Figure 2) is strictly
required for CDK2 inhibitory activity.
The third series of compounds was synthesized to analyze
the effect of different substituents on the 2-phenyl ring at-
tached to the tetrahydroquinoline moiety (Table 3). In the
CDK2–3 f complex, the ortho-chloro phenyl ring of 3 f is tightly
packed against the short helix at the beginning of the activa-
tion loop, which is typical of PK inactive conformations, and
the hydrophobic residues of the b4-b5 hairpin (Figure 2). In
this series, substitution of the ortho-chloro atom with a methyl
(3j), a trifluorocarbon (3u), or a bromine (3t) was tolerated,
whereas substitution with a less hydrophobic fluorine atom
&
ChemMedChem 2018, 13, 1 – 9
www.chemmedchem.org
4
ꢀ 2018 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim
ÝÝ These are not the final page numbers!

Full Papers
Table 3. Structures and biological activity of the investigated compounds synthesized to explore the role of phenyl ring substitutions. Biological activities
are reported as percent inhibition at 10 and 50 mm.
Compd
R3
R4
H
R5
R6
CDK2
wild-type EGFR
L858R/T790M EGFR
10 mm 50 mm
10 mm
50 mm
10 mm
50 mm
3j
3r
3s
3t
3u
3v
3w
3x
3y
CH3
H
H
H
H
H
H
H
OH
H
H
H
H
H
H
OH
F
38
36
0
55
49
22
0
40
36
0
63
65
43
0
0
0
0
0
17 64
phenyl
27
23
0
41
28
39
41
0
0
0
0
52
70
72
79
17
30
30
33
F
Br
CF3
F
H
H
H
H
H
H
OH
H
0
0
0
19
0
19
20
23
31
32
28
F
F
13
29
51
62
OH
H
In contrast, the compounds could be readily docked to the
crystal structure of the T790M/L858R double mutant EGFR-KD
(PDB ID: 3W2R),[14] in which the unfolding of the short helix at
the beginning of the activation loop due to the L858R muta-
tion makes the allosteric pocket significantly more accessible
to the ligands.[14–16] Induced-fit docking of the carboxylic acid
derivatives to the T790M/L858R EGFR-KD crystal structure
showed that the compounds bind to the allosteric pocket by
accommodating the 2-phenyl-substituted ring into the deeper
hydrophobic pocket lined by Leu788, Met766, Met790, and
Phe856, with the amino group of the hexahydrocyclopenta[c]-
quinoline hydrogen bonding with the Phe856 backbone car-
bonyl of the DFG motif, and the carboxylate group forming a
salt bridge with the polar heads of Arg858 and Arg748 (3s,
Figure 3). In particular, the interaction of the carboxylate with
the mutated Arg858 residue may further explain the observed
selectivity of the compounds for the double mutant EGFR-KD
with respect to the wild-type EGFR-KD. Indeed, all the com-
pounds in which the carboxylate is removed or modified were
inactive when tested on the T790M/L858R protein, except for
the acetic acid derivative 3o that was still able to form a salt
bridge with Arg858 (Tables 1 and 2). Finally, the different sub-
stituents of the 2-phenyl ring, which is accommodated into
the deeper hydrophobic pocket lined by the side-chains of
Figure 3. Induced-fit docking binding mode of compound 3s within the T790M/L858R EGFR-KD crystal structure (PDB ID: 3W2R). The protein is represented
by a cyan surface and cyan cartoons, whereas the residues forming the binding cavity are labeled with the three-letter code and shown as cyan sticks. Com-
pound 3s is represented by pink spheres and pink balls and sticks within the box.
ChemMedChem 2018, 13, 1 – 9
www.chemmedchem.org
5
ꢀ 2018 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim
&
These are not the final page numbers! ÞÞ

Full Papers
Phe856, Leu788, Met766, Met790, Phe723, and Lys745, were
found to play a role in modulating the selectivity toward the
double mutant EGFR. In particular, the fluorine substituted
compound 3s was a selective allosteric inhibitor of the double
mutant T790M/L858R EGFR-KD with respect to either wild-type
EGFR-KD and CDK2.
whereas the majority of ChEMBL compounds belongs to the
type I, I1/2, and II classes of PK inhibitors.
Conclusions
This study was devised to explore the SAR of a promising class
of type III allosteric inhibitors of CDK2 through target-guided
synthetic modifications. Moreover, for the first time, the report-
ed CDK2 allosteric inhibitors were tested on the wild-type
EGFR kinase domain and its clinically relevant T790M/L858R
mutant, leading to the identification of active compounds.
Docking studies and molecular dynamics simulations were per-
formed to elucidate the SARs. Through this strategy, the impor-
tance of the carboxylic acid and the nitro groups for the CDK2
and T790M/L858R EGFR inhibitory activity were established.
The carboxylic acid is strictly required for binding to the
T790M/L858R EGFR, whereas it is potentially replaceable in
CDK2 by different chemical moieties able to make hydrogen
bonds with the side-chain of Lys33. We also demonstrated that
2-phenyl ring substituents that fit into an additional hydropho-
bic pocket of the allosteric site are important to fine-tune the
activity and selectivity of the investigated compounds. Our
data, although the synthesized compounds need to be opti-
mized for potency, explored the relatively uncharted field of
type III allosteric inhibitors and demonstrated that switching
the inhibitory activity from CDK2 to double mutant EGFR is to
be considered a feasible strategy. Our results pointed to differ-
ent binding modes in CDK2 and double mutant EGFR-KD, thus
reflecting the peculiarities of the respective allosteric sites and
potentially enabling the development of selective allosteric li-
gands. Remarkably, compound 3s was shown to be an alloste-
ric inhibitor of the T790M/L858R EGFR that spares both wild-
type EGFR and CDK2, and will be further investigated for the
design of more potent and selective T790M/L858R EGFR inhibi-
tors.
Molecular dynamics simulations of compound 3s within the
T790M/L858R EGFR-KD (see the Experimental Section) showed
that the proposed binding mode was stable and the key inter-
actions within the allosteric pocket were maintained through-
out the simulation (100 ns; two replicas with different starting
velocities, Figure S1I).
Dose–response curves (Figure 4) showed that 3s was able to
inhibit the T790M/L858R EGFR-KD with an EC50 of 44Æ4 mm,
whereas inhibition was not observed with wild-type EGFR-KD.
To demonstrate its allosteric behavior on the T790M/L858R
EGFR-KD, compound 3s was also tested in the presence of a
higher concentration of ATP (1 mm). The obtained dose–re-
sponse curves clearly demonstrated that compound 3s main-
tained its inhibitory profile, thus confirming its allosteric mech-
anism of action (Figure S1O).
Figure 4. Dose–response curves of compound 3s tested on wild-type and
T790M/L858R EGFR.
Experimental Section
Chemical diversity with respect to known CDK2 and EGFR
inhibitors reported in ChEMBL
Chemistry: All reactions were carried out in oven-dried glassware
and dry solvents under nitrogen atmosphere. Unless otherwise
stated, all solvents were purchased from Sigma Aldrich and used
without further purification. Substrates and reagents were pur-
chased from Sigma Aldrich and used as received. 2-Nitrobenzene-
sulfenyl chloride was purchased from Alfa Aesar and used as re-
ceived. Thin-layer chromatography (TLC) was performed on Merck
precoated 60 F254 plates. Reactions were monitored by TLC on
silica gel, with detection by UV light (254 nm) or by a solution of
ninhydrin with heating. Flash chromatography was performed
using silica gel (240–400 mesh, Merck). All tested compounds pos-
sessed a purity of >98% confirmed via elemental analyses (CHN)
with a PerkinElmer 2400 instrument. 1H NMR spectra were recorded
on a Bruker DRX-400 instrument and are reported relative to resid-
ual CDCl3 and DMSO. 13C NMR spectra were recorded on the same
instrument (100 MHz) and are reported relative to residual CDCl3
and DMSO. Chemical shifts (d) for proton and carbon resonances
are quoted in parts per million (ppm) relative to tetramethylsilane
(TMS), which was used as an internal standard. COSY, HSQC, HMBC,
and NOESY experiments were used in the structural assignment.
MS spectra were recorded using electrospray ionization (ESI) tech-
The chemical novelty of the presented hexahydrocyclopen-
ta[c]quinolone inhibitors was evaluated by means of 2D simi-
larity analyses with respect to the CDK2 and EGFR ligands col-
lected in the ChEMBL database. Chemical structures of the
compounds with activity annotations on human CDK2 and
EGFR were extracted from the ChEMBL database and similarity
analyses were performed by using the MACCS and circular fin-
gerprints (ECFP6 fingerprint), as described in the experimental
section. The 2D fingerprint-based retrospective analyses re-
vealed that the disclosed compounds are, on average, highly
dissimilar to the already reported CDK2 and EGFR inhibitors,
the averaged similarities being far below the commonly ac-
cepted similarity thresholds (Table S1).[17,18] Therefore, results of
the cheminformatic analyses demonstrated that the ligands
herein reported are structurally different from those in the ana-
lyzed ChEMBL subsets. This result is not surprising considering
that the new class of inhibitors bind to an allosteric pocket,
&
ChemMedChem 2018, 13, 1 – 9
www.chemmedchem.org
6
ꢀ 2018 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim
ÝÝ These are not the final page numbers!

Full Papers
nique on a Waters Micromass Q-Tof micro mass spectrometer. De-
tailed methods for the preparation of the compounds reported in
Tables 1–3 can be found in the Supporting Information.
column (Merck KGaA, Darmstadt, Germany) and eluted with a NaCl
gradient from 10 to 1000 mm in 25 mm HEPES pH 8, 10% glycerol,
5 mm DTT in 30 CV. The protein fractions were gel filtered on Su-
perdex 75 (10/300 GL) column (GE Healthcare Life Sciences,
Milano, Italy) in 25 mm HEPES pH 8, 250 mm NaCl, 10% glycerol
and 2 mm TCEP. Stock protein samples were flash frozen at the
final concentration of 0.5–1 mgmLÀ1 and stored in aliquots at
À808C till further use. Typical yield of purified EGFR-KD constructs
was 0.6–1.8 mgmLÀ1 from 1ꢂ109 Sf9 cells.
Single-crystal X-ray analysis of racemic compound 3j: X-ray-qual-
ity crystals were obtained by slow crystallization from MeOH. The
data collection was performed at room temperature on a three-
circle Bruker AXS Smart diffractometer equipped with an APEX-II
CCD detector. A 99.3% complete full sphere of data was collected
using graphite-monochromated radiation (l=0.71073 ꢁ) with a
nominal power of 50 kVꢂ30 mA. Data were corrected for absorp-
tion and beam anisotropy effects, and the structural model was
obtained up to a maximum resolution of 0.77 ꢁ through the full-
matrix least-squares procedure implemented in SHELXL-2014/4.[19]
The interested reader can find full information on the crystal struc-
ture of 3j in the Supporting Information. CCDC 1856132 contains
the supplementary crystallographic data for this paper. These data
can be obtained free of charge from The Cambridge Crystallo-
graphic Data Centre.
Biological assays: CDK2 inhibitory activities were determined
using the previously described ANS displacement assay.[4,20] The
assay was performed in 96-well black plates in a total volume of
50 mL of 50 mm HEPES pH 7.5 with DTT 1 mm and ANS (final con-
centration 62.5 mm). Various concentrations of the compounds
were added and the fluorescence (excitation 405 nm, emission
460 nm) measured using an Infinite M200 Microplate Reader
(Tecan) to determine the intrinsic fluorescence of the compounds.
Finally, 1 mg of recombinant CDK2 was added and a second fluo-
rescence measurement was taken at the same wavelengths. EGFR
inhibitory activities were determined using the ADP-Glo kinase
assay. The ADP-Glo kinase assay (Promega) is a luminescent kinase
assay that measures the ADP formed from the EGFR kinase reac-
tion. With this method, the luminescent signal generated is propor-
tional to the ADP concentration produced and is correlated with
the kinase activity. The kinase reaction was performed in 96-well
white plates as already described,[21] in a total volume of 25 mL and
incubated for 1 h at room temperature. Then, an equal volume of
ADP-Glo reaction was added followed by, after 40 min incubation,
the addition of 50 mL of kinase detection reagent. The lumines-
cence was read using the GloMax plate reader (Promega). Control
experiments with the reference inhibitors are reported in the Sup-
porting Information (Figure S1M).
CDK2 production and purification: E. coli BL21(DE3) cells were
transformed with a vector enclosing the human CDK2 coding se-
quence with a GST tag at the N-terminus. Cells were grown for 2–
3 h at 378C, then the temperature was decreased to 168C prior to
induction. Expression was induced with 0.1 mm IPTG at OD600 =0.9
for 20–24 h at 168C. Cells were harvested by centrifugation for
20 min at 4000ꢂg at 48C. Bacteria were suspended in 50 mm
Tris·HCl pH 8.0, 150 mm NaCl, and 1 mm DTT (buffer A) supple-
mented with the Complete Mini Protease Inhibitor Cocktail (Roche)
and were lysed using a French pressure cell press. The lysate was
centrifuged for 30 min at 12000 rpm (15500ꢂg, on a Beckman
Coulter TA-14–50 rotor) at 48C and the supernatant was applied
onto a GST-affinity column (GE Healthcare) equilibrated with the
buffer A. GST-CDK2 was eluted using the buffer A supplemented
with 10 mm reduced glutathione. The final protein was diluted at
1 mgmLÀ1, aliquoted, flash-frozen in liquid nitrogen, and stored at
À808C.
Docking calculations: The protein structures were prepared with
the Schrçdinger Suite 2014-3 Protein Preparation Wizard (PPW)
tool. PPW automatically adjusted the ionization and tautomeriza-
tion state of the proteins at a neutral pH (PROPKA was used to pre-
dict the pKa of residues at pH 7), set the orientation of any misor-
iented groups (Asn, Gln, and His residues), and optimized the hy-
drogen bond network. Finally, the structures were refined to re-
Wild-type and T790M/L858R EGFR production and purification:
DNA encoding for the kinase domain residues 696–1022 of the
human EGFR wild-type was inserted in pFB-LIC-Bse and pFB-6HZB
transfer vectors (kindly provided by Opher Gileadi, SGC-Oxford).
The T790M/L858R mutation was introduced in the wild-type con-
structs by the Q5 site directed mutagenesis based method (New
England Biolabs) and sequence was verified. Recombinant baculo-
virus was generated by using the Bac-to-Bacꢃ Baculovirus Expres-
sion System (Invitrogen by Life Technologies Corporation, CA, USA)
by transfecting Sf9 cells seeded in six-well plates (1.5ꢂ106 cells per
well) with FuGENE HT (Promega Italia Srl, Milano, Italy). The high
titer viral stock was generated by two rounds of amplifications and
used for protein expression in Sf9 cells at 278C. The cells were har-
vested by centrifugation 72 h post-infection. For protein purifica-
tion, cells were homogenized (Emulsiflex, Avestin) in lysis buffer
(50 mm Tris pH 8, 500 mm NaCl, 10% glycerol, 10 mm imidazole,
2 mm TCEP) supplemented with 10 mgmLÀ1 DNaseI, 1 mm MgCl2
and protease inhibitor cocktail (complete EDTA-Free, Roche Diag-
nostics GmbH, Manheim, Germany). The lysate was cleared by cen-
trifugation at 30000ꢂg for 1 h at 48C and the supernatant was in-
cubated with NiNTA beads (Qiagen, Milano, Italy) for 1 h at 48C.
The beads were washed with 30 CV of the lysis buffer and the
bound proteins were eluted with lysis buffer containing 300 mm
imidazole. Eluted protein fractions were dialysed overnight at 48C
against 25 mm Tris pH 8, 0.2m NaCl, 10% glycerol, 5 mm DTT in
the presence of the Tobacco Etch Virus protease to remove tags.
The cleaved proteins were purified on a Fractogelꢃ EMD TMAE (M)
lieve steric clashes with
a restrained minimization with the
OPLS2005 force field till a final RMSD of 0.30 ꢁ with respect to the
input protein coordinates. Docking calculations to CDK2 were per-
formed by extracting the protein from the CDK2–3 f complex ob-
tained in our previous publication.[5] Docking calculations to the
wild-type EGFR and the T790M/L850R EGFR kinase domains were
performed by using the 3W32[13] and 3W2R crystal structures,[14] re-
spectively, prepared as described above. All docking calculations
were performed with the Induced Fit Docking (IFD) protocol avail-
able in the Schrçdinger Suite 2014-3, by using default settings. In
addition, the wild-type EGFR-KD–3 f complex was also investigated
by using the extended sampling protocol available in the Schrç-
dinger Suite 2014-3, which automatically prepares the receptor by
choosing the residues to trim and atom-specific van der Waals scal-
ing factors on the basis of solvent-accessible surface areas, B-fac-
tors, the presence of salt bridges, and rotamer searches. In this
case, a maximum number of 80 poses were saved for each ligand
and submitted to the subsequent Prime side-chain orientation pre-
diction of residues within a shell of 5 ꢁ around the ligand. After
the Prime minimization of the selected residues and the ligand for
each pose, a Glide SP re-docking of each protein–ligand complex
structure was performed. Finally, the binding energy (IFDScore) for
ChemMedChem 2018, 13, 1 – 9
www.chemmedchem.org
7
ꢀ 2018 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim
&
These are not the final page numbers! ÞÞ

Full Papers
each output pose was calculated.[22] In all cases, grids were cen-
tered to the allosteric pocket of the investigated kinases.
Keywords: allosteric inhibitors · cyclin-dependent kinase 2 ·
docking · drug design · epidermal growth factor receptor ·
structure–activity relationships
Molecular dynamics: The complex between the T790M/L858R
EGFR-KD and 3s obtained by IFD was saved in the PDB format and
prepared with the tleap module of the Amber14 suite, using the
ff14SB force field.[23,24] The protein–ligand complex was enclosed in
a truncated octahedral TIP3P water box with a radius of 10 ꢁ. Na+
counter-ions were added to neutralize the complex. Long-range
electrostatic interactions were treated using the Particle-Mesh
Ewald (PME) method,[25] while a short-range cutoff of 8 ꢁ was ap-
plied for all non-bonded interactions. Before data acquisition the
system was equilibrated in a stepwise fashion to eliminate bad
atomic contacts. The solvent was minimized, while keeping the
protein and ligand fixed, by using 800 steps of the steepest de-
scent (SD) method and 800 steps of the conjugate gradient (CG)
method. A second minimization was performed by keeping the
protein fixed with 800 steps of the SD method and 800 steps of
the CG method. A third minimization was performed by keeping
the ligand fixed with 800 steps of the SD method and 800 steps of
the CG method. Finally, an unrestrained minimization was per-
formed on the whole system by using 800 steps of the SD method
and 800 steps of the CG method. The system was subsequently
heated to 300 K in two steps of 150 ps each, in the NVT ensemble.
Then, equilibration was performed in six separate runs of 400 ps
each, in the NPT ensemble. Constraints on the protein and ligand
atoms were gradually decreased throughout the equilibration
steps. All constraints were removed in the production run (100 ns).
The time step was set to 2 fs, and bonds involving hydrogen
atoms were constrained using the SHAKE algorithm.[26] Two replicas
with different starting velocities were performed. Trajectories were
analyzed by calculating the RMSD of protein alpha carbons and
ligand heavy atoms.
[1] K. A. Stenberg, P. T. Riikonen, M. Vihinen, Nucleic Acids Res. 1999, 27,
362–364.
[2] Z. Fang, C. Grꢄtter, D. Rauh, ACS Chem. Biol. 2013, 8, 58–70.
[3] P. Wu, M. H. Clausen, T. E. Nielsen, Pharmacol. Ther. 2015, 156, 59–68.
[4] G. Rastelli, A. Anighoro, M. Chripkova, L. Carrassa, M. Broggini, Cell Cycle
2014, 13, 2296–2305.
[5] M. S. Christodoulou, F. Caporuscio, V. Restelli, L. Carlino, G. Cannazza, E.
Costanzi, C. Citti, L. Lo Presti, P. Pisani, R. Battistutta, M. Broggini, D. Pas-
sarella, G. Rastelli, ChemMedChem 2017, 12, 33–41.
[6] X. Zhang, J. Gureasko, K. Shen, P. A. Cole, J. Kuriyan, Cell 2006, 125,
1137–1149.
[7] N. Jura, X. Zhang, N. F. Endres, M. A. Seelinger, T. Schindler, J. Kuriyan,
Mol. Cell 2011, 42, 9–22.
[8] L. Palmieri, G. Rastelli, Drug Discovery Today 2013, 18, 407–414.
[9] N. Sturm, A. Tinivella, G. Rastelli, J. Chem. Info. Model. 2018, 58, 1094–
1103.
[10] S. V. Sharma, D. W. Bell, J. Settleman, D. A. Haber, Nat. Rev. Cancer 2007,
7, 169–181.
[11] C. Ricordel, L. Friboulet, F. Facchinetti, J. C. Soria, Ann. Oncol. 2018, 29,
i28–37.
[12] L. Zai-Shun, W. Wei-Min, L. Wei, N. Cong-Wei, L. Yong-Hong, L. Zheng-
Ming, W. Jian-Guo, Bioorg. Med. Chem. Lett. 2013, 23, 3723–3727.
[13] Y. Kawakita, M. Seto, T. Ohashi, T. Tamura, T. Yusa, H. Miki, H. Iwata, H.
Kamiguchi, T. Tanaka, S. Sogabe, Y. Ohta, T. Ishikawa, Bioorg. Med. Chem.
2013, 21, 2250–2261.
[14] S. Sogabe, Y. Kawakita, S. Igaki, H. Iwata, H. Miki, D. R. Cary, T. Takagi, S.
Takagi, Y. Ohta, T. Ishikawa, ACS Med. Chem. Lett. 2013, 4, 201–205.
[15] C. H. Yun, T. J. Boggon, Y. Li, M. S. Woo, H. Greulich, M. Meyerson, M. J.
Eck, Cancer Cell 2007, 11, 217–227.
[16] K. S. Gajiwala, J. Feng, R. Ferre, K. Ryan, O. Brodsky, S. Weinrich, J. C.
Kath, A. Stewart, Structure 2013, 21, 209–219.
[17] S. W. Muchmore, D. A. Debe, J. T. Metz, S. P. Brown, Y. C. Martin, P. J.
Hajduk, J. Chem. Inf. Model. 2008, 48, 941–948.
[18] S. Jasial, Y. Hu, M. Vogt, J. Bajorath, F1000 Research 2016, 5, 591.
[19] G. M. Sheldrick, Acta Crystallogr. Sect. C 2015, 71, 3–8.
[20] M. P. Martin, R. Alam, S. Betzi, D. J. Ingles, J.-Y. Zhu, E. Schçnbrunn,
ChemBioChem 2012, 13, 2128–2136.
[21] F. Caporuscio, A. Tinivella, V. Restelli, M. S. Semrau, L. Pinzi, P. Storici, M.
Broggini, G. Rastelli, Future Med. Chem. 2018, 10, 1545–1553.
[22] M. D. Parenti, G. Rastelli, Biotechnol. Adv. 2012, 30, 244–250.
[23] J. A. Maier, C. Martinez, K. Kasavajhala, L. Wickstrom, K. E. Hauser, C. Sim-
merling, J. Chem. Theory Comput. 2015, 11, 3696–3713.
[24] D. A. Case, V. Babin, J. T. Berryman, R. M. Betz, Q. Cai, D. S. Cerutti, T. E.
Cheatham III, T. A. Darden, R. E. Duke, H. Gohlke, A. W. Goetz, S. Gusarov,
N. Homeyer, P. Janowski, J. Kaus, I. Kolossvꢅry, A. Kovalenko, T. S. Lee, S.
LeGrand, T. Luchko, R. Luo, B. Madej, K. M. Merz, F. Paesani, D. R. Roe, A.
Roitberg, C. Sagui, R. Salomon-Ferrer, G. Seabra, C. L. Simmerling, W.
Smith, J. Swails, R. C. Walker, J. Wang, R. M. Wolf, X. Wu, P. A. Kollman,
AMBER14, 2014, University of California, San Francisco.
[25] U. Essmann, L. Perera, M. L. Berkowitz, T. Darden, H. Lee, L. G. Pedersen,
J. Chem. Phys. 1995, 103, 8577–8593.
2D similarity analyses: 2D similarity analyses were carried out on
the ChEMBL database (release 23, accessed on May 29, 2018) to
evaluate the chemical novelty of the identified CDK2 and EGFR al-
losteric inhibitors.[27] In particular, similarity calculations were per-
formed on already known CDK2 and EGFR ChEMBL compounds
with activity annotations characterized by an assay confidence
score equal or greater than 8. Two different types of molecular fin-
gerprints implemented in the OpenEye OEGraphSim Python Tool-
kits (release 2017.Jun.1), i.e., MACCS and ECFP6, were used to
encode the structural details of the molecules.[28–30] Ligand similari-
ties were evaluated in terms of the Tanimoto similarity score by
considering CDK2 or EGFR active compounds as reference queries.
A total of 56 ranks (two molecular fingerprints ꢂ 28 active com-
pounds) were obtained for each target (CDK2 and EGFR) in the 2D
retrospective analyses on ChEMBL. Results and statistics of the sim-
ilarity calculations are summarized in Table S1 of the Supporting In-
formation.
[26] J. P. Ryckaert, G. Ciccotti, H. J. C. Berendsen, J. Comput. Phys. 1977, 23,
327–341.
[27] A. Gaulton, L. J. Bellis, A. P. Bento, J. Chambers, M. Davies, A. Hersey, Y.
Light, S. McGlinchey, D. Michalovich, B. Al-Lazikani, J. P. Overington, Nu-
cleic Acids Res. 2012, 40, D1100–D1107.
Acknowledgements
[28] OpenEye Python Toolkits 2017.Jun.1, OpenEye Scientific Software, Santa
Fe, NM (USA), http://www.eyesopen.com.
[29] J. L. Durant, B. A. Leland, D. R. Henry, J. G. Nourse, J. Chem. Inf. Comput.
Sci. 2002, 42, 1273–1280.
This work was supported by a grant from the Associazione Ital-
iana per la Ricerca sul Cancro (AIRC IG 15993 and IG 16792).
[30] D. Rogers, M. Hahn, J. Chem. Inf. Model. 2010, 50, 742–754.
Conflict of interest
Manuscript received: October 20, 2018
Accepted manuscript online: && &&, 0000
Version of record online: && &&, 0000
The authors declare no conflict of interest.
&
ChemMedChem 2018, 13, 1 – 9
www.chemmedchem.org
8
ꢀ 2018 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim
ÝÝ These are not the final page numbers!

FULL PAPERS
Selectivity determinants revealed: The
structure–activity relationships of a
promising class of type III allosteric
cyclin-dependent kinase 2 (CDK2) inhibi-
tors were explored. Additional biological
evaluations allowed the discovery of a
selective allosteric inhibitor of the
L858R/T790M epidermal growth factor
receptor (EGFR) double mutant. Model-
ing studies shed light on the pharmaco-
phore requirements for switching the
inhibitory activity from CDK2 to the
double-mutant EGFR.
L. Carlino, M. S. Christodoulou, V. Restelli,
F. Caporuscio, F. Foschi, M. S. Semrau,
E. Costanzi, A. Tinivella, L. Pinzi,
L. Lo Presti, R. Battistutta, P. Storici,
M. Broggini, D. Passarella, G. Rastelli*
&& – &&
Structure–Activity Relationships of
Hexahydrocyclopenta[c]quinoline
Derivatives as Allosteric Inhibitors of
CDK2 and EGFR
ChemMedChem 2018, 13, 1 – 9
www.chemmedchem.org
9
ꢀ 2018 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim
&
These are not the final page numbers! ÞÞ