Explicit treatment of active-site waters enhances quantum mechanical/implicit solvent scoring: Inhibition of CDK2 by new pyrazolo[1,5-a]pyrimidines

Article abstract of DOI:10.1016/j.ejmech.2016.12.023

We present comprehensive testing of solvent representation in quantum mechanics (QM)-based scoring of protein-ligand affinities. To this aim, we prepared 21 new inhibitors of cyclin-dependent kinase 2 (CDK2) with the pyrazolo[1,5-a]pyrimidine core, whose

Full text of DOI:10.1016/j.ejmech.2016.12.023

Accepted Manuscript
Explicit treatment of active-site waters enhances quantum mechanical/implicit solvent
scoring: Inhibition of CDK2 by new pyrazolo[1,5-a]pyrimidines
Michaela Hylsová, Benoit Carbain, Jindřich Fanfrlík, Lenka Musilová, Susanta Haldar,
Cemal Köprülüoğlu, Haresh Ajani, Pathik S. Brahmkshatriya, Radek Jorda, Vladimír
Kryštof, Pavel Hobza, Aude Echalier, Kamil Paruch, Martin Lepšík
PII:
S0223-5234(16)31029-7
10.1016/j.ejmech.2016.12.023
EJMECH 9114
DOI:
Reference:
To appear in: European Journal of Medicinal Chemistry
Received Date: 15 September 2016
Revised Date: 7 December 2016
Accepted Date: 9 December 2016
Please cite this article as: M. Hylsová, B. Carbain, J. Fanfrlík, L. Musilová, S. Haldar, C. Köprülüoğlu,
H. Ajani, P.S. Brahmkshatriya, R. Jorda, V. Kryštof, P. Hobza, A. Echalier, K. Paruch, M. Lepšík,
Explicit treatment of active-site waters enhances quantum mechanical/implicit solvent scoring: Inhibition
of CDK2 by new pyrazolo[1,5-a]pyrimidines, European Journal of Medicinal Chemistry (2017), doi:
10.1016/j.ejmech.2016.12.023.
This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to
our customers we are providing this early version of the manuscript. The manuscript will undergo
copyediting, typesetting, and review of the resulting proof before it is published in its final form. 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.

ACCEPTED MANUSCRIPT

ACCEPTED MANUSCRIPT
Explicit Treatment of Active-site Waters Enhances Quantum
Mechanical/Implicit Solvent Scoring: Inhibition of CDK2 by New
Pyrazolo[1,5-a]pyrimidines
a
a,f
b
a
‡
Michaela Hylsová, ‡ Benoit Carbain, ‡ Jindřich Fanfrlík, Lenka Musilová, Susanta
Haldar,b,c Cemal Köprülüoğlu,b,c Haresh Ajani,b,c Pathik S. Brahmkshatriya,b Radek Jorda,d
Vladimír Kryštof,d Pavel Hobza,b,c Aude Echalier,e Kamil Paruch,a,f* Martin Lepšíkb*
a
Department of Chemistry, CZ Openscreen, Masaryk University, Kamenice 5, 625 00 Brno, Czech
Republic.
b
Institute of Organic Chemistry and Biochemistry, Academy of Sciences of the Czech Republic,
v.v.i., Flemingovo nam. 2, 166 10 Prague 6, Czech Republic.
c
Regional Center of Advanced Technologies and Materials, Department of Physical Chemistry,
Palacký University, Olomouc, 771 46 Olomouc, Czech Republic.
d
Laboratory of Growth Regulators, Faculty of Science, Palacký University & Institute of
Experimental Botany, Šlechtitelů 27, 783 71 Olomouc, Czech Republic.
e Centre de Biochimie Structurale. CNRS UMR 5048 - UM - INSERM U 1054. 29 rue de Navacelles.
34090 Montpellier. France, Present address: Departments of Molecular and Cell Biology and of
Cancer Studies. Henry Wellcome Building. University of Leicester. Lancaster Road. Leicester. LE1
9HN. United Kingdom.
f
International Clinical Research Center, St. Anne’s University Hospital Brno, Pekařská 53, 656 91
Brno, Czech Republic.
KEYWORDS
cyclin-dependent kinase 2; ATP-competitive type I inhibitors; pyrazolo[1,5-a]pyrimidine; quantum
mechanical scoring; protein-ligand binding; molecular dynamics; water thermodynamics; X-ray
crystal structure
1

ACCEPTED MANUSCRIPT
ABSTRACT
We present comprehensive testing of solvent representation in quantum mechanics (QM)-
based scoring of protein-ligand affinities. To this aim, we prepared 21 new inhibitors of cyclin-
dependent kinase 2 (CDK2) with the pyrazolo[1,5-a]pyrimidine core, whose activities spanned three
orders of magnitude. The crystal structure of a potent inhibitor bound to the active CDK2/cyclin A
complex revealed that the biphenyl substituent at position 5 of the pyrazolo[1,5-a]pyrimidine scaffold
was located in a previously unexplored pocket and that six water molecules resided in the active site.
Using molecular dynamics, protein-ligand interactions and active-site water H-bond networks as well
as thermodynamics were probed. Thereafter, all the inhibitors were scored by the QM approach
utilizing the COSMO implicit solvent model. Such a standard treatment failed to produce a correlation
with the experiment (R2=0.49). However, the addition of the active-site waters resulted in significant
improvement (R2 = 0.68). The activities of the compounds could thus be interpreted by taking into
account their specific noncovalent interactions with CDK2 and the active-site waters. In summary,
using a combination of several experimental and theoretical approaches we demonstrate that the
inclusion of explicit solvent effects enhance QM/COSMO scoring to produce a reliable structure–
activity relationship with physical insights. More generally, this approach is envisioned to contribute
to increased accuracy of the computational design of novel inhibitors.
2

ACCEPTED MANUSCRIPT
1. Introduction
Reliable prediction of protein–ligand (P–L) binding affinity is the major task of
computer-aided drug design. In the structure-based drug design, the 3D coordinates of the P
–
L complex (obtained mostly by X-ray crystallography, NMR or docking) are used for the
evaluation of binding free energy using scoring or other approaches. The application of
quantum mechanics (QM) in P
–
L scoring1 has expanded the methodological repertoire and
L interactions,2,3 metals (either in proteins1,4 or
allowed a quantitative treatment of P
–
ligands),5 exotic ligands,6 and noncovalent interactions of quantum origin, e.g. σ-hole
bonding,7,8 or covalent binding of ligands.9
The treatment of solvation in QM scoring needs to be fast and is thus usually done
using continuum approaches of varying degree of accuracy.1,10-12 Despite the success of QM-
based models (such as COSMO),13 some P
–L complexes need explicit and dynamic treatment
of solvation for a correct description.14,15 This approach can be further extended to evaluate
enthalpic and entropic contributions of specific water molecules to ligand binding.16-19
Theoretical approaches crucially rely on data obtained by various experimental methods, such
as X-ray crystallography, isothermal microcalorimetry or kinetic assays.17,20
Cyclin-dependent kinase 2 (CDK2) is the best studied member of the CDK family of
Ser/Thr protein kinases, which are essential components of cell-cycle regulation.21 Since the
activities of CDKs are frequently deregulated in various types of cancers, these kinases have
been targeted by numerous experimental anticancer therapeutics.22 The recent approval of the
CDK4/CDK6 inhibitor palbociclib for the treatment of ER-positive and HER2-negative breast
cancer is a culmination of these efforts.23 CDKs phosphorylate their substrates using ATP as a
phosphate donor. The ATP molecule binds to the CDK active site located in a cleft (“hinge
region”) which is located between the N-terminal β-sheet and C-terminal α-helical domains.
The vast majority of known CDK inhibitors are of type I, i.e. directly competing with ATP for
the binding site.24 Detailed understanding of the binding of small-molecule inhibitors to
CDK2 has been obtained through hundreds of X-ray structures of co-crystal complexes.24
However, only about one-fifth of them include the active state of CDK2 (as a ternary complex
with a regulatory cyclin subunit), whose geometry is more suitable for the prediction of an
interaction with small-molecule inhibitors.25 In addition, the water content in the active site of
CDK2 complexed with cyclin is different due to differently placed segments or amino-acid
side chains (such as Lys33, Asp145 or Glu51).25
3

ACCEPTED MANUSCRIPT
One class of potent and selective CDK2 inhibitors (which includes dinaciclib,
currently profiled in clinical trials26) utilizes the pyrazolo[1,5-a]pyrimidine (PP) core (Fig.
1).27-30 The exploration of the structure–activity relationship (SAR) at positions 3 and 5 of the
core has revealed that: i) the substitution of hydrogen (as R3) by bromine or ethyl can lead to a
10- to 100-fold increase in the binding affinity, but analogs containing larger substituents are
significantly less active,28,30 and ii) compounds possessing proper unsaturated and saturated
five- or six-membered rings at position 5 exhibit IC50 values below 20 nM.28 The significant
increase in activity upon replacing R3=H by the bulkier Br or Et moieties has been ascribed to
the optimal filling of the hydrophobic cavity in the gatekeeper region around Phe80.28 This
explanation, based on inactive binary CDK2/inhibitor co-crystal structures, also includes the
expulsion of an extra water molecule (denoted here as W*), which can be accommodated in
sub-series with R3=H but not with R3=Br/Et.28 However, individual contributions of these
factors to the overall binding affinity have not been reported to date.
Figure 1
The project reported herein is a non-trivial extension of our ongoing efforts in the area
of protein kinase inhibitor design.30-34 Specifically, we report the synthesis and biological
activity of a series of heretofore unknown 21 PP-based CDK2 inhibitors bearing
“standard“ H, Br and Et substituents at position 3 and previously unexplored biphenyl
substituents at position 5. The crystal structure of the active CDK2/cyclin A ternary complex
with one of the most potent inhibitors from the series serves as a starting point for molecular
dynamics (MD) calculations, water thermodynamics and QM/COSMO-based scoring taking
into account explicit-solvent effects. Overall, the project is a comprehensive experimental–
theoretical study of challenging CDK2/inhibitor complexes and suggests some potential
avenues for near-future computer-aided drug design.
2. Results and Discussion
2.1. Compound Design and Synthesis
We prepared a series of new PP-based compounds 1a–u (Table 1) with H, Br or Et at position
3, the 3-aminomethylpyridine moiety at position 7, and biphenyl substituents at position 5. To
our knowledge, no PP-based kinase inhibitors with biphenyls at position 5 have been
described to date. We put several polar substituents on the distal phenyl ring, hypothesizing
4

ACCEPTED MANUSCRIPT
that binding to the enzyme could be enhanced via noncovalent interactions to this previously
unexplored part of the active site.
Adopting the methodology we had developed previously,28,35 we prepared key intermediates
6a–6c and 7a–7b, which were used for the subsequent installation of the substituted biphenyl
moiety via Suzuki coupling (Scheme 1). The syntheses of the biphenyls containing single
substituents were quite straightforward and provided the target compounds 1a–1q indicated in
Table 1.
Table 1
Scheme 1
While the syntheses described above proceeded uneventfully, the preparation of brominated
analogs 1r and 1s proved non-trivial, as the bromination of hydroxylated biphenyl derivatives
was non-regioselective and proceeded on both the phenol and the pyrazole rings. Eventually,
a late-stage selective bromination of derivatives containing 2- and 3-hydroxy-[1,1'-biphenyl]-
4-carboxamide moieties was achieved by deactivating the phenol moiety via its protection as
mesylate or pivaloate. The final cleavage of the mesylate or pivaloate under basic conditions,
followed by the removal of the Boc group, afforded the target compounds 1r and 1s (Scheme
2).
Scheme 2
The synthesis of the analogs with nitrile-containing side chains was also relatively
challenging. Direct Heck couplings of acrylonitrile with commercially available 2-bromo-4-
nitrobenzoic acid or its amide were unsuccessful. However, the reaction did proceed on ester
9, providing the acrylic nitrile derivative, which after hydrogenation afforded compound 11.
Next, the reduction of the nitro group followed by the conversion of the amine into bromide
provided compound 13, which then provided (via the Sandmeyer reaction and conversion of
the ester into an amide) the desired, heretofore unknown, intermediate 15. The subsequent
Suzuki reaction, bromination and final deprotection proceeded smoothly and afforded the
target compounds 1t and 1u (Scheme 2).
5

ACCEPTED MANUSCRIPT
2.2. The Structure–Activity Relationship
The experimental IC50 values spanned three orders of magnitude, from the weakest-binding
compound 1a, bearing unsubstituted biphenyl at position 5 and hydrogen at position 3 (12.6
µM), to the most active compound 1s with a doubly substituted distal phenyl ring (R5 = p-
CONH2-o-OH-Ph) and R3 = Br (0.015 µM) (Table 1). In general, the compounds could be
clearly divided into two groups based on the identity of the R3 substituent: compounds with
R3 = H were weaker binders with IC50 ≥ 2 µM, whereas compounds with R3 = Br or Et were
significantly more potent (IC50 ≤ 0.5 µM, Table 1). This structure–activity relationship (SAR),
compatible with the previous observations for PP-based CDK2 inhibitors with smaller
(phenylic) R5 substituents,28 was previously explained by the filling of the hydrophobic cavity
near the gatekeeper residue Phe80.27,28 Since this interaction profoundly influences the
compounds’ activity, we address it here using advanced computations (see Section 2.6).
The SAR at position 5 was independent from that at position 3. It revealed these trends:
i.
Unsubstituted biphenyls (i.e. compounds 1a and 1b) are less potent than analogous
PP-based inhibitors with phenylic substituents at position 5.28
ii.
iii.
However, installation of polar substituents on the biphenyl moiety can
significantly improve the potency.
Polar substituents have to be at a proper distance from the distal phenyl ring (cf.
compounds 1d and 1i); additional suitably oriented polar motifs can further
improve the inhibitory activity (cf. compounds 1l and 1s).
The interaction of the polar motifs on the biphenyl part with the protein is therefore one of the
key determinants of the compounds’ activity; consequently, we addressed it by QM scoring
(see Section 2.7).
2.3. The Crystal Structure of the Active Ternary CDK2/Cyclin A/1l Complex
In order to gain additional insight into the interaction of our inhibitors with the enzyme, we
determined the crystal structure at 2.4 Å resolution of the active Thr160-phosphorylated
(pThr160) CDK2/cyclin A ternary complex with one of the most potent compounds of the
series, compound 1l (PDB code: 5LMK). The data collection and refinement statistics have
been compiled in the Supporting Information (Table S1). The inhibitor is bound in the active
6

ACCEPTED MANUSCRIPT
site using the standard binding mode observed for type I inhibitors. The asymmetric unit
contains two CDK2/cyclin A heterodimers. Overall, the two CDK2 molecules present in the
asymmetric unit (chains A and C) assume very similar structures (the backbone RMSD of
0.47 Å excluding the flexible regions not supported by electron density: the missing loops 38–
41 in the A chain and 220–251 in the C chain and the C-terminus in both chains, residues
294–298). The inhibitor bound in their ATP binding sites interacts via the same binding
mode. The water-molecule patterns in their respective ATP cavities are also very similar. The
description of the inhibitor bound to the CDK2 ATP binding site that follows is based on the
analysis of the CDK2 chain A. Any deviation from this binding mode and the water molecule
pattern in the two CDK2 chains is indicated explicitly.
Specifically, the pyrazolo[1,5-a]pyrimidine core binds to CDK2 via three hinge region H-
bonds (two classic 7-amino N-H...O:Leu83 and N1...HN:Leu83 bonds and one weak C2-
H...O:Glu81 bond) (Fig. 1 – dashed; Fig. 2 – dotted). The R3 substituents of the PP core face
the gatekeeper Phe80 and the pyridyl moiety occupies the specificity surface (Figure 1) of the
ATP-binding domain of CDK2. The proximal phenyl ring of the R5 biphenyl moiety (C19-
C24, Figure 2) is connected to the PP core residue in the CDK2 ribose-binding pocket roughly
in-plane with a C24-C19-C3-N4 dihedral of 33°. The distal phenyl ring (C25-C30, Figure 2)
extends in the phosphate-binding pocket toward Lys129 on the surface of the enzyme (Figures
1 and 2). It is again roughly in-plane with the PP core (a C30-C25-C3-N4 dihedral of -29°),
but tilted in the opposite direction as compared to the proximal phenyl ring. The distal-to-
proximal phenyl dihedral C26-C25-C22-C21 is thus twisted at -63°/-87° (the A/C chain of
CDK2, respectively). The terminal carboxamide group of the inhibitors is slightly rotated with
respect to the distal phenyl – the O32-C31-C28-C29 dihedral angle is 42/28° (the A/C chain
of CDK2, respectively, Figure 2). It is noteworthy that the electron density corresponding to
the carboxamide moiety of the inhibitor bound to the CDK2 chain C is significantly weaker
than in the chain A, probably due to a different crystal environment. In the surroundings of
the inhibitor, there is a chain of five water molecules (W77, W206, W194, W147 and W130)
occupying the region from the gatekeeper Phe80 to the phosphate-ribose pocket (Figures 1
and 2). Additionally, a water molecule with low occupancy (W224; ca 0.30) has been found
between the distal phenyl ring and the CDK2 side chain of Asp145 (Figure 2). The
corresponding water molecule bound to the CDK2 chain C (W171) has been modeled with
full occupancy.
Figure 2
7

ACCEPTED MANUSCRIPT
2.4. The Thermodynamics of Explicit Active-site Waters in the CDK2/1l Complex
The CDK2/1l crystal structure described above has been treated by the WaterMap program to
assess the dynamics and thermodynamics of the water molecules. The clustering has
confirmed the sites for all six active-site waters (W77, W206, W194, W147, W224 and
W130) with full occupancy (Figure 2, Table S2, Figure S1). This computational
overestimation of the structural stability of water sites (cf. the crystallographic occupancy of
W224 of 0.30, see above) stems mostly from the rigid protein/rigid ligand approximation
embodied in the WaterMap. Thermodynamically, the binding of all six water molecules in
their protein sites is unfavorable with respect to their ∆G in the bulk solution (Table 2). This
is mainly due to the entropic cost of trapping them in the protein, which is not overcome by
sufficiently strong H-bond interactions (enthalpy gain) with the protein/ligand/water in the
cavity. In the case of W206 and W130, even the ∆H term is unfavorable (Table 2) because of
the closeness of the hydrophobic Phe80 and the proximal phenyl of 1l, respectively. As
discussed below, the thermodynamic parameters of water are crucial in QM scoring.
Table 2
2.5. Molecular Dynamics of the CDK2/1l Complex
To understand the full dynamics of the CDK2/1l complex, we performed standard molecular
dynamics (MD) on the crystal structure surrounded by explicit water molecules. A correct
description of the bromine atom of the inhibitors (within the limits of the MD approach) was
enabled by the use of an explicit sigma hole (ESH) (see the Methods) to account for possible
X…π interactions with Phe80. Two variants of the terminal para-carboxamide moiety were
used, because the potential hydrogen bond with Lys129 represents only a weak
crystallographic restraint (the 1l:N33…NZ:Lys129 distance of 3.7/3.8 Å in the A/C chain of
CDK2, respectively). The original crystallographic variant is denoted as amide A
conformation (i.e. an O32-C31-C28-C27 dihedral of 42°) and the alternative orientation, B
conformation, related by a roughly 180° flip (due to a slight nonplanarity on C31), has a
dihedral of -139°. Because the biphenyl moiety also rotates in MD, we have opted for the
O32-C31-C3-N4 dihedral (i.e. the orientation of the terminal carboxamide with respect to the
PP core) to monitor the orientational preferences of the terminal carboxamide group. The
crystallographic values of the O32-C31-C3-N4 dihedral are -158/159°, respectively, for the
8

ACCEPTED MANUSCRIPT
A/C chain of CDK2 for the original A conformation and 15/-25°, respectively, for the B
conformation (Figure S2, blue and gray bars, respectively).
Terminal Carboxamide Rotation. First, we compared the root-mean-square deviations
(RMSD) of the protein backbone and of non-hydrogen inhibitor atoms in the course of 80 ns.
Upon comparing the A and B conformations, the latter was found to be stabler both for the
protein (the average RMSD of 1.4 Å and a drift of 0.2 Å vs. 1.1 Å and a drift of 0.1 Å,
respectively) (Figure S3A) and the inhibitor (3 plateaus with the RMSD up to 1.5 Å vs. 1
plateau of the RMSD at 0.9 Å). Since the greater stability of B was also confirmed by QM
calculations (as described below), only this conformer was taken into account. The inhibitor
RMSD oscillated with the amplitude of ca 0.4 Å in the first 40 ns and dropped down to ca 0.1
Å in the second half of the trajectory, being especially stable in the last 20 ns (Figure S3B).
This was concomitant with the formation of the H-bond between the inhibitor’s carboxamide
in the conformation B and Lys129 (1l:O32…NZ:Lys129) after 40 ns and its maintenance after
60 ns (Figure S3A). The plotting of the populations of the O32-C31-C3-N4 dihedral
throughout the whole 80 ns trajectory (Figure S2, violet curve) has revealed that both
conformations A and B are present and correspond to the crystallographic values (Figure S2,
blue and gray bars). Figures S2, S4B (i.e. plots of isolated populations in different parts of the
trajectory) demonstrate that the conformation A prevails in the first 40 ns while the
conformation B is present predominantly after 40 ns. Taken together, the analyses of 80-ns
MD show that both conformations A and B of the terminal carboxamide can be present in the
structure of the CDK2/1l complex.
Active-site Water Molecules. Water densities in the CDK2 active site averaged over the 80-ns
MD trajectory clearly show four spherical regions corresponding to the crystallographic water
molecules W77, W206, W194 and W147 (Figure S5). These water molecules are present
throughout most of the simulation time, forming a hydrogen-bonded chain and linking the
inhibitor N4 nitrogen (Figure 2) with the terminal side-chain amine of Lys33. In line with the
crystallographic structure, W147 H-bonds with W194 (the corresponding crystallographic
distance is 3.0 Å). The MD density for W130 is also present at the crystallographic position,
but it is weaker when compared to the four above-mentioned water molecules. This
corresponds to the lack of anchoring interactions and only a weakly directional O-H…π
interaction with the proximal phenyl of 1l (Figure S5). The site for W224 is split into two
densities, one closer to W130 and one farther toward the bulk solvent. Each of the MD
9

ACCEPTED MANUSCRIPT
densities is occupied by a different water molecule, thus completing the H-bond chain
extending from W130 toward the bulk solvent (Figure S5). This transient water chain could
not be captured by X-ray crystallography due to its dynamic nature (the crystallographic
W130…W224 inter-oxygen distance was 4.5 Å, cf. Figure 2). This was also evidenced by the
low occupancy of W224 in the A chain of the crystallographic structure.
2.6. The QM/COSMO Treatment of the Effects of R3 Substituents
As described in Section 1.2., the studied inhibitors can be clearly divided into two groups
based on the identity of the R3 substituent: compounds with R3=H are weaker binders with
IC50 ≥ 2 µM while compounds with bromine or ethyl moiety as R3 have IC50 ≤ 0.5 µM. The
previous qualitative structure-based explanation was based on filling the hydrophobic cavity
near the gatekeeper residue Phe80.27,28
Direct R3-Gatekeeper Interactions. Here, this question is approached quantitatively based on
QM/COSMO interaction ‘free’ energies between three inhibitors possessing all three
substituent types at R3 (i.e. compounds 1j, 1k and 1l; Table 1) and the Phe80 side chain. We
have found that compounds 1k and 1l have stronger interactions with the side chain of Phe80
(-1.7 and -1.8 kcal/mol, respectively) than 1j (-0.8 kcal/mol), which is in agreement with the
previous explanation.27,28 More specifically, the effect can be attributed to dispersion-driven
noncovalent interactions (X…π and CH…π, respectively) of Br and Et substituents at
position 3 with the phenyl ring of Phe80 and the lack thereof when R3=H. Based on this
indication of the affinity difference between the R3 subseries, we proceeded to the modeling
and scoring of the whole 21-compound series (Table 1), using the advanced QM/SQM
scoring function.2,3
2.7. The QM/COSMO Scoring of the Entire Inhibitor Series
Terminal Carboxamide Conformation. Prior to productive QM/SQM/COSMO scoring, we
carefully selected the conformation of the terminal carboxamide. The preliminary scoring of
the A and B conformations of inhibitors 1j–1u (Table 1) revealed that the scores for the
conformation B were systematically stabler by approximately 3 kcal/mol. Structurally, this
was caused by the formation of the 1l:O32...NZ:Lys129 H-bond (the O...N distance of 2.9 Å),
which is similar to that found in MD after 60 ns (cf. Figure S4A). We therefore used the
conformation B for the QM scoring of all inhibitors possessing the terminal carboxamide (and
N,N-dimethyl carboxamide) motif.
10

ACCEPTED MANUSCRIPT
Implicit Solvent QM Scoring. All complexes of CDK2 and the PP-based inhibitors, including
the conformational variants of the distal phenyl substituents, were optimized using the
QM/SQM/COSMO procedure. The resulting structures will be briefly described here. The
terminal carboxamide formed the H-bond mentioned above. The amide methylation (1m, 1n)
did not change the geometry of the H-bond. Similarly, the hydroxymethyl group of 1g, 1h and
1i formed the O32...H-NZ:Lys129 H-bond with the O...N distance of 2.9 Å. Wherever the
carboxamide group was replaced by smaller amino or hydroxyl groups (1c, 1d, 1e or 1f), the
H-bond with Lys129 was weakened (the O...N distance increased to about 3.7 Å). The
introduction of a second substituent on the distal phenyl did not change the orientation of the
biphenyl in the geometry of the complex.
For the entire inhibitor series, we used our standard QM-based scoring,8 i.e. no explicit water
molecules were present and the solvent was modeled by the implicit COSMO model.36 As
shown in Figure 3A, there was a poor correlation of the QM score with the experimentally
derived ∆Gbind (expressed as RT*ln(IC50/2). The coefficient of determination, R2, was 0.49
and the predictive index, PI, was 0.76. The PI represents a measure of the rank-order
prediction: 1 stands for an always correct prediction, 0 for a random prediction and -1 for an
always incorrect prediction.37 Therefore, the differences in affinity between the R3=H and
R3=Br/Et subseries cannot be explained solely by the small interaction with Phe80 (as
calculated in Section 1.6), which corresponds to ca1 kcal/mol.
Figure 3
Mixed Explicit/Implicit Solvent QM Scoring. We and others have recently realized that
structural waters have important consequences for docking and QM scoring.13,16,19 Therefore,
we considered the six explicit water molecules (W77, W206, W194, W147, W224 and W130)
revealed by X-ray crystallography and confirmed by WaterMap simulations in the CDK2/1l
complex (see above) in QM optimization and scoring.
In the R3=H subseries, the possibility of an additional water molecule (W*) filling the space
around the position 3 of the PP scaffold was suggested by WaterMap calculations on
CDK2/1j (Table S2, Figure S1); therefore, we included it optionally. In specific cases, such as
1o, the possible expulsion of W224 was also taken into account. For mixed explicit/implicit
solvent QM scoring, the entropies of bridging water molecules, as obtained from WaterMap
calculations (Table S2), were added to the QM score calculations (see the Methods).
11

ACCEPTED MANUSCRIPT
Overall, the QM scoring with mixed explicit/implicit solvent treatment accounted correctly
for the affinity difference between the R3=H and R3=Br/Et subseries and also made it possible
to correctly describe the interactions of the substituents in ortho- and meta- positions of the
biphenyl moiety. This resulted in a fair coefficient of determination (R2 of 0.63) and the
predictive index (PI of 0.80) – see Figure 3B. Upon including W* in the R3=H subseries, the
correlation of the whole inhibitor set slightly increased to R2 of 0.68 and PI to 0.85. It is thus
evident that the proper addressing of both the structural and thermodynamic effects of all the
conserved active-site water molecules is indispensable in the QM scoring for reliable a
description of the affinity trends. Next, we investigated the molecular determinants of affinity.
2.8. The Molecular Determinants of Potency
All the QM scoring terms for the entire inhibitor series are summarized in Table S3. In line
with our previous studies,2,3 the computed scores were dominated by the ∆Eint and ∆∆Gsolv
terms. The ∆G'conf(L) and -T∆Ssolv terms had similar values for all the studied inhibitors. The
least active compound 1a had the least negative ∆Eint term. Although the desolvation penalty
(∆∆Gsolv) of 1a was also the smallest within the whole series, it did not compensate for the
weak interactions of the unsubstituted phenyl ring and the missing substituent at the R3
position. On the other hand, the most active compounds 1r and 1s possessed the para-
carboxamide group and the hydroxyl group in meta- or ortho- positions, respectively, on the
distal phenyl ring of the biphenyl, and a Br atom at the R3 position. Interestingly, the hydroxyl
group was crucial for the formation of two important H-bonds: with W224 and Lys129 in the
former case and with W224 and W130 in the latter (Figure 4). Such interactions were
responsible for the highest QM/SQM score of 1s, due to the most negative ∆Eint.
Figure 4
3. Conclusions
Herein we report the design, synthesis, activity measurements, and crystallographic and
computational analyses of a series of 21 new pyrazolo[1,5a]pyrimidine-based CDK2
inhibitors. The compounds contain R5-substituted biphenyl moieties that allowed for the
exploration of the phosphate-ribose pocket of the CDK2 active site. The measured activities
of all the new compounds against CDK2 span three orders of magnitude. For the molecular
12

ACCEPTED MANUSCRIPT
understanding of the potency determinants, we have determined the X-ray structure of the
active ternary complex of CDK2/cyclin A/1l.
We have shown that active-site explicit waters and their thermodynamics need to be included
in the computational description of biomolecule-inhibitor systems in order to obtain a good
correlation with the experimental binding affinities. Modeling, QM scoring, MD simulations
and water thermodynamics calculations have revealed that explicit-solvent effects need to be
included in the scoring procedure for this system to obtain meaningful results. Using a non-
trivial combination of the above-mentioned techniques, we have been able to dissect the role
of interactions of R3 substituents with the Phe80 gatekeeper residue, the effect of active-site
waters, and the roles of the distal phenyl substitutions.
This study has important specific consequences for the design of CDK inhibitors.We envision
that our methodology (when applied more generally) will contribute to the efforts focused on
increasing the prediction accuracy of computer-aided drug design.
4. Methods
4.1. Organic Synthesis
The conditions and reagents used in the synthesis of all new compounds are described in
Schemes 1 and 2. Detailed experimental procedures and the characterization of the
compounds are given in the Supporting Information.
4.2. Biochemical Measurements
The tested compounds were dissolved in DMSO and diluted with water (the concentration of
DMSO in the reaction never exceeded 0.2%). The CDK2/Cyclin E complex was produced in
Sf9 insect cells via baculoviral infection and purified on a Ni2+NTA column (Qiagen). Kinase
(approx. 10 ng) was assayed using a mixture of the following: 1 mg/mL of histone H1, 15 µM
of ATP, 0.05 of µCi [γ-33P]ATP, the tested compound, and reaction buffer, in a final volume
of 10 µL. The reaction buffer consisted of: 60 mM of HEPES-NaOH, pH 7.5, 3 mM of
MgCl2, 3 mM of MnCl2, 3 µM of Na-orthovanadate, 1.2 mM of DTT, and 2.5 µg / 50 µl of
PEG20.000. The reactions were stopped by adding 5 µL of 3% aqueous H3PO4. Aliquots were
spotted onto P-81 phosphocellulose (Whatman), washed 3 times with 0.5% aqueous H3PO4,
and finally air-dried. Kinase inhibition was quantified using a FLA-7000 digital image
13

ACCEPTED MANUSCRIPT
analyzer (Fujifilm). The concentration of each tested compound required the decrease of the
CDK activity by 50%. The IC50 values were determined from the dose-response curve.
4.3. X-ray Crystallography
The expression of the recombinant human pThr160 CDK2 and Cyclin A proteins in
Escherichia coli and the purification of the binary complex were carried out as described in.38
The co-crystallization of the protein complex at 12 mg.mL-1 in the presence of the inhibitor
was done using published crystallization conditions, with (NH4)2SO4 and KCl as the
precipitant agents,39 after the incubation of the protein complex with 2 mM of inhibitor 1l
dissolved in DMSO for 20 min on ice. Crystals grew in a few weeks and were subjected to
brief soaking in 7 M of sodium formate as a cryoprotectant step before being flash-frozen in
liquid nitrogen. The diffraction data were collected on the European Synchrotron Radiation
Facility (ESRF) ID29 beamline. Data processing was carried using programs of the CCP4
suite.40 The structure of pThr160 CDK2/Cyclin A in complex with inhibitor 1l was solved by
molecular replacement using the program Phaser41 and the high resolution structure of
pThr160 CDK2/Cyclin A in complex with roscovitine (PDB code 3DDQ)42 as the search
model. A clear solution emerged from the molecular-replacement procedure with two ternary
complexes in the asymmetric unit. After rigid body refinement, the initial difference electron
density clearly highlighted the presence of the inhibitor bound in the ATP-binding site in both
copies of CDK2 found in the asymmetric unit. A standard procedure of iterative building in
Coot43 and refinement in Refmac44 was used. The final complex was saved in the Protein Data
Bank (PDB) under accession code 5LMK.
4.4. Computational Methodology
For the three types of calculations, i.e. molecular dynamics (MD), WaterMap and quantum
chemical (QM) scoring, specific preparations of the crystal structure of the CDK2/Cyclin A/1l
ternary complex were done as described below. The CDK2 chain A was consistently used,
whereas the cyclin A subunit and the magnesium ion were removed.
4.4.1. Molecular Dynamics Simulations.
System setup. The missing loop between the residues 38–41 was modeled using the
MODELLER program45 via the UCSF Chimera package.46 Hydrogens were added by the
LEaP program of the AMBER14 suite.47 Subsequently, the loop with its flanking residues
(residues 37–44) was minimized using the Sander program of AMBER14 package47 using
14

ACCEPTED MANUSCRIPT
500 cycles of the steepest descent, followed by 300 cycles of the conjugate gradient method.
The CDK2/1l complex was prepared by a procedure described previously (see below).8
Explicit TIP3P water molecules were added to fill the dodecahedron box repeating in space
using periodic boundary conditions (PBC). The system was neutralized by adding two Cl-
ions. For the protein, the AMBER ff03.r1 force field48 was used. For the ligands, the general
AMBER force field (GAFF)49 with HF/6-31G* RESP charges50 and fragment-based
approach51 were used. Explicit sigma-hole (ESH) using the “all-fit” (AF) approach52 was
utilized for the ligands with R3=Br to describe potential X…π interactions.
MD simulations. All the MD simulations and most of the analyses were run in GROMACS
4.5.453 on CPUs. The starting relaxation steps included minimizations and heating MD steps,
followed by the MD production run. First, only the explicit waters were minimized and then
heated slowly in MD to 300 K in the canonical NVT ensemble, i.e. with a fixed number of
particles (N), volume (V) and temperature (T). Next, the whole system was minimized except
for the backbone of the protein. Then, the whole system was heated to 300 K. The
equilibration simulation in an isothermal–isobaric (NPT) ensemble, i.e. with a fixed number
of particles (N), pressure (P), and temperature (T) consisted of a 5-ns run with a velocity-
rescale thermostat and a Berendsen barostat, respectively. Finally, the production run of 80 ns
was carried out with a Nose–Hoover thermostat (the reference temperature of 300 K, the
coupling coefficient τT=0.5 ps) and a Parrinello–Rahman barostat (the reference pressure of 1
bar and the coupling coefficient of τP 0.5 ps).
The Newton’s equations of motion were integrated using a 1.0-fs time step with the
leap-frog algorithm. The 10 Å non-bonded cut-off was used for long-range electrostatics by
the particle-mesh Ewald (PME) method as well as for the Lennard–Jones (LJ) interactions.
All the bonds were constrained with the LINCS algorithm.
The visualization of the MD trajectories was done in VMD 1.9.1.54 The water density
was calculated by the VolMap tool55 in VMD 1.9.1.
4.4.2. WaterMap Calculations.
The CDK2/1l and CDK2/1j complexes were prepared with the Protein Preparation Wizard in
Maestro56 using the default options. The crystallographic water molecules were retained in
CDK2/1l and transferred to CDK2/1j. The standard settings for the WaterMap57,58
calculations were used except for the following: the protein residues outside the 15 Å shell
around the inhibitor were removed and the MD of explicit waters was run for 10 ns.
15

ACCEPTED MANUSCRIPT
4.4.3. Quantum Mechanical (QM) Calculations.
System setup. Residues 38 and 41 were capped by N-methyl amide and acetamide,
respectively, at their C- and N-termini. All the crystallographic waters were initially deleted
following the standard procedure.3 The mixed explicit–implicit solvent procedure (see the
Results) included six structural water molecules (W77, W206, W194, W147, W130 and
W224) in their crystallographic positions (Figure 2). For the ligands with R3=H, one
additional water molecule (W*) was optionally added to the position defined by WaterMap.
The compound 1o was additionally scored without W224.
The REDUCE program of AMBER1447 was used to determine potential flips of Asn, Gln and
His residues. Histidines were modeled as neutral or monoprotonated, based on the visual
inspection of the surroundings: His121 was protonated on Nδ and other His residues on Nε.
All the Lys, Arg, Asp, Glu residues as well as the N- and C-termini of CDK2 were modeled in
their charged state. Hydrogen atoms were added by the LEaP module of AMBER1447 and
relaxed by a short high-temperature molecular dynamics run (for 2 ps at 2400 K and then
cooled down to 0 K for 8 ps).
Modeling. All 21 inhibitors in complex with CDK2 were modeled from the crystallographic
CDK2/1l complex in PyMol59 by deleting/building atoms. The compounds with substituents
in the ortho- and meta- positions on the distal biphenyl ring were considered in two
orientations, related by a 180° flip of the distal phenyl, to account for both possible binding
orientations with respect to the protein. Similarly, the terminal amide in the para- position of
the biphenyl was considered in both A and B conformations. Hydrogen atoms were added to
the ligands with the UCSF Chimera program.46 All the added atoms were relaxed by
simulated annealing from 600 to 0 K. The cooling runs were 2 ps long using the Berendsen
thermostat, 1-fs time step, and the generalized Born solvent model (Bondi radii and igb=7
sander option).60 Such a preparation protocol has been reported to be optimal for QM
scoring.30
QM/SQM optimization. We used an improved multi-layer setup8 by adopting the
QM/SQM/COSMO methodology. The rationale of this approach was that COSMO had been
shown to be more reliable than GB for the solvation of neutral ligands.61 The procedure is
identical to that reported in Ref.8 with the following exceptions: we considered only residues
within 10 Å of the inhibitor for optimization while all the residues farther than 8 Å from the
16

ACCEPTED MANUSCRIPT
inhibitor were frozen during the optimization. All the complexes were optimized by the
QM/SQM/COSMO method and the FIRE optimization algorithm prior to scoring.
QM scoring. For QM scoring (single-point calculations), the whole CDK2/inhibitor
complexes were used. The QM part comprised the inhibitors, all explicit water molecules and
residues Ala33, Val64, Phe80, Asp127, Lys129, Pro130, Gln131 and Asn132. The small QM
region was treated at the DFT-D3 (the TPSS functional with the TZVPP basis set) level of
theory using the Turbomole 6.5 program.62 The rest of the system (SQM region) was
calculated by PM6-D3H4X3 using MOPAC2009.63
The binding free energy was approximated by the total score expressed by Eq. 1.2 with the
new consistent notation described in Ref.3:
Score = ∆Eint + ∆∆Gsolv + ∆G'confw(L) – T∆Ssolv
[Eq. 1].
The individual terms describe the gas-phase interaction energy (∆Eint), the interaction
solvation/desolvation free energy (∆∆Gsolv), the change of the conformational “free” energy
of the ligand (∆G'confw(L)), and optionally the entropy of the explicit water molecules
(T∆Ssolv). It should be mentioned, however, that another portion of the entropy contribution is
already included in the ∆∆Gsolv term via the parameterization of the implicit solvent models.
∆Eint was calculated using the QM/SQM method described above. The solvation free energy
was determined using two implicit solvent models: COSMO36 at the PM6 level and SMD64 at
the HF/6-31G** level (the latter only for the ligands to increase the accuracy).61 For the
evaluation of the change of the conformational “free” energy of the ligand (∆G'confw(L)), the
gas-phase DFT-D3 (TPSS/TZVPP single-point energy at the B-LYP/SVP optimized
geometry) energy is combined with the SMD solvation free energy.
QM fragmentation. The interaction energies between the inhibitor and protein fragments (side
chains and peptide bonds) were determined at the DFT-D3 (TPSS/TZVPP)62 level combined
with the COSMO implicit solvent model.36
17

ACCEPTED MANUSCRIPT
ASSOCIATED CONTENT
The X-ray data collection and the refinement statistics, analyses of MD trajectories, water site
location and thermodynamics from WaterMap, energy terms of the QM score and NMR, IR
and MS spectra of all the compounds are presented in the Supporting Information.
AUTHOR INFORMATION
Author Contributions
‡ these authors contributed equally to this work
Corresponding Authors
*
E-mail: paruch@chemi.muni.cz
*E-mail: lepsik@uochb.cas.cz
ACKNOWLEDGEMENT
We are grateful to Dr. F. Hoh for diffraction data collection. We thank Dr. Michaela
Nekardová for helpful discussions. This work has been supported by the project “Employment
of
Best
Young
Scientists
for
International
Cooperation
Empowerment”
(CZ.1.07/2.3.00/30.0037) co-financed from the European Social Fund and the state budget of
the Czech Republic, and by the European Regional Development Fund under grant FNUSA-
ICRC (No. CZ.1.05/1.1.00/02.0123). This work has been supported by the Ministry of
Education, Youth and Sports of the Czech Republic (project No. L01305) and also from the
Large Infrastructures for Research, Experimental Development and Innovations project
“IT4Innovations National Supercomputing Center – LM2015070”. RJ and VK thank the
Czech Science Foundation (15-15264S). JF, PH and ML thank the Gilead Sciences & IOCB
Research Centre. JF, SH, CK, HA, PH and ML thank the research project RVO 61388963,
awarded by the Academy of Sciences of the Czech Republic and the Czech Science
Foundation (grant number P208/12/G016). KP thanks the project CZ-OPENSCREEN:
National Infrastructure for Chemical Biology (LM2015063).
18

ACCEPTED MANUSCRIPT
REFERENCES
(1) K. Raha, K.M. Merz A quantum mechanics-based scoring function: study of zinc ion-
mediated ligand binding, J. Am. Chem. Soc. 126 (2004), 1020–1021.
(2) J. Fanfrlik, A.K. Bronowska, J. Rezac, O. Prenosil, J. Konvalinka, P. Hobza, A reliable
docking/scoring scheme based on the semiempirical quantum mechanical PM6-DH2 method
accurately covering dispersion and H-bonding: HIV-1 protease with 22 ligands, J. Phys.
Chem. B. 114 (2010), 12666–12678.
(3) M. Lepšík, J. Řezáč, M. Kolář, A. Pecina, P. Hobza, J. Fanfrlík, The semiempirical
quantum mechanical scoring function for in silico drug design, ChemPlusChem. 78 (2013),
921–931.
(4) A. Pecina, M. Lepšík, J. Řezáč, J. Brynda, P. Mader, P. Řezáčová, P, Hobza J. Fanfrlík,
QM/MM calculations reveal the different nature of the interaction of two carborane-based
sulfamide inhibitors of human carbonic anhydrase II, J. Phys. Chem. B 117 (2013), 16096–
16104.
(5) A. Ciancetta, S. Genheden, U. Ryde, A QM/MM study of the binding of RAPTA ligands
to cathepsin B, J. Comput. Aided. Mol. Des. 25 (2011), 729–742.
(6) J. Fanfrlík, J. Brynda, J. Rezác, P. Hobza, M. Lepsík, Interpretation of protein/ligand
crystal structure using QM/MM calculations: case of HIV-1 protease/metallacarborane
complex, J. Phys. Chem. B 112 (2008), 15094–15102.
(7) J. Fanfrlík, M. Kolář, M. Kamlar, D. Hurný, F.X. Ruiz, A. Cousido-Siah, M. Mitschler, J.
Rezáč, E. Munusamy, M. Lepšík, P. Matějíček, J. Veselý, A. Podjarny, P. Hobza, Modulation
of aldose reductase inhibition by halogen bond tuning, ACS Chem. Biol. 8 (2013), 2484–
2492.
(8) J. Fanfrlík, F.X. Ruiz, A. Kadlčíková, J. Řezáč, A. Cousido-Siah, A. Mitschler, S. Haldar,
M. Lepšík, M.H. Kolář, P. Majer, A.D. Podjarny, P. Hobza, The Effect of Halogen-to-
19

ACCEPTED MANUSCRIPT
Hydrogen Bond Substitution on Human Aldose Reductase Inhibition, ACS Chem. Biol. 10
(2015), 1637–1642.
(9) J. Fanfrlík, P.S. Brahmkshatriya, J. Řezáč, A. Jílková, M. Horn, M. Mareš, P. Hobza, M.
Lepšík, Quantum mechanics-based scoring rationalizes the irreversible inactivation of
parasitic Schistosoma mansoni cysteine peptidase by vinyl sulfone inhibitors, J. Phys. Chem.
B 117 (2013), 14973–14982.
(10) M. Kolář, J. Fanfrlík, M. Lepšík, F. Forti, J.F. Luque, P. Hobza, Assessing the Accuracy
and Performance of Implicit Solvent Models for Drug Molecules: Conformational Ensemble
Approaches, J. Phys. Chem. B 117 (2013), 5950-5962.
(11) P. Soderhjeim, J. Kongsted, U. Ryde, Ligand Affinities Estimated by Quantum Chemical
Calculations, J. Chem. Theory Comput. 6 (2010), 1726–1737.
(12) S. Decherchi, M. Masetti, I. Vyalov, W. Rocchia, Implicit solvent methods for free
energy estimation, Eur. J. Med. Chem. 91 (2015), 27–42.
(13) A. Pecina, R. Meier, J. Fanfrlík, M. Lepšík, J. Řezáč, P. Hobza, C. Baldauf, The
SQM/COSMO filter: reliable native pose identification based on the quantum-mechanical
description of protein–ligand interactions and implicit COSMO solvation, Chem. Commun.
52 (2016), 3312–3315.
(14) S. Genheden, P. Mikulskis, L. Hu, J. Kongsted, P. Söderhjelm, U. Ryde, Accurate
predictions of nonpolar solvation free energies require explicit consideration of binding-site
hydration, J. Am. Chem. Soc. 133 (2011), 13081–13092.
(15) P. Setny, R. Baron, P.M. Kekenes-Huskey, A. McCammon, J. Dzubiella, Solvent
fluctuations in hydrophobic cavity–ligand binding kinetics, Proc. Natl. Acad. Sci. USA 110
(2013), 1197–1202.
(16) B. Vorlová, D. Nachtigallová, J. Jirásková-Vaníčková, H. Ajani, P. Jansa, J. Rezáč, J.
Fanfrlík, M. Otyepka, P. Hobza, J. Konvalinka, M. Lepšík, Malonate-based inhibitors of
20

ACCEPTED MANUSCRIPT
mammalian serine racemase: kinetic characterization and structure-based computational
study. Eur. J. Med. Chem. 89 (2015), 189–197.
(17) B. Breiten, M. R. Lockett, W. Sherman, S. Fujita, M. Al-Sayah, H. Lange, C.M. Bowers,
A. Heroux, G. Krilov, G.M. Whitesides, Water networks contribute to enthalpy/entropy
compensation in protein–ligand binding, J. Am. Chem. Soc. 135 (2013), 15579–15584.
(18) D. Robinson, T. Bertrand, J.C. Carry, F. Halley, A. Karlsson, M. Mathieu, H. Minoux
M.A. Perrin, B. Robert, L. Schio W. Sherman, Differential Water Thermodynamics
Determine PI3K-Beta/Delta Selectivity for Solvent-Exposed Ligand Modifications, J. Chem.
Inf. Model. 56 (2016), 886–894.
(19) A. Cousido-Siah, F.X. Ruiz, J. Fanfrlik, J. Giménez-Dejoz, A. Mitschler, M. Kamlar, J.
Vesely, H. Ajani, X. Pares, J. Farrés, P. Hobza, A.D. Podjarny, IDD388 polyhalogenated
derivatives as probes for an improved structure-based selectivity of AKR1B10 inhibitors,
ACS Chem. Biol. 11 (2016), 2693–2705.
(20) A. Biela, M. Betz, H. Heine, G. Klebe, Water makes the difference: rearrangement of
water solvation layer triggers non-additivity of functional group contributions in protein–
ligand binding, Chem. Med. Chem. 7 (2012), 1423–1434.
(21) M. Malumbres, Cyclin-dependent kinases. Genome Biol. 15 (2014), 122.
(22) U. Asghar, A.K. Witkiewicz, N.C. Turner, E.S. Knudsen, The history and future of
targeting cyclin-dependent kinases in cancer therapy, Nat. Rev. Drug Discov. 14 (2015), 130–
46.
(23) J.A. Beaver, L. Amiri-Kordestani, R., Charlab, W. Chen, T. Palmby, A. Tilley, J.F.
Zirkelbach, J. Yu, Q. Liu, L. Zhao, J. Crich, X.H. Chen, M. Hughes, E. Bloomquist, S. Tang,
R. Sridhara, P.G. Kluetz, G. Kim, A. Ibrahim, R. Pazdur, P.C. Cortazar, FDA Approval:
Palbociclib for the Treatment of Postmenopausal Patients with Estrogen Receptor-Positive,
HER2-Negative Metastatic Breast Cancer, Cancer Res. 21 (2015), 4760–4766.
21

ACCEPTED MANUSCRIPT
(24) A. Huwe, R. Mazitschek, A. Giannis, Small molecules as inhibitors of cyclin-dependent
kinases. Angew. Chem. Int. Ed. Engl. 42 (2003), 2122–2138.
(25) G. Kontopidis, C. McInnes, S.R. Pandalaneni, I. McNae, D. Gibson, M. Mezna, M.
Thomas, G. Wood, S. Wang, M.D. Walkinshaw, P.M. Fischer, Differential binding of
inhibitors to active and inactive CDK2 provides insights for drug design, Chem Biol. 13
(2006), 201–211.
(26) K. Paruch, M.P. Dwyer, C. Alvarez, C. Brown, T.Y. Chan, R.J. Doll, K. Keertikar, C.
Knutson, B. McKittrick, J. Rivera, R. Rossman, G. Tucker, T. Fischmann, A., Hruza, V.
Madison, A.A. Nomeir, Y. Wang, P. Kirschmeier, E. Lees, D. Parry, N. Sgambellone, W.
Seghezzi, L. Schultz, F. Shanahan, D. Wiswell, X. Xu, Q. Zhou, R.A. James, V.M. Paradkar,
H. Park, L.R. Rokosz, T.M. Stauffer, T.J. Guzi, Discovery of Dinaciclib (SCH 727965): A
Potent and Selective Inhibitor of Cyclin-Dependent Kinases, ACS Med. Chem. Lett. 1 (2010),
204–208.
(27) P.M. Dwyer, K. Paruch, C. Alvarez, R.J. Doll, K. Keertikar, J. Duca, T.O. Fischmann, A.
Hruza, V. Madison, E. Lees, D. Parry, W. Seghezzi, N. Sgambellone, F. Shanahan, D.
Wiswell, T.J. Guzi, Versatile templates for the development of novel kinase inhibitors:
Discovery of novel CDK inhibitors, Bioorg. Med. Chem. Lett. 17 (2007), 6216–6219.
(28) K. Paruch, P.M. Dwyer, C. Alvarez, C. Brown, T.Y. Chan, R.J. Doll, K. Keertikar, C.
Knutson, B. McKittrick, J. Rivera, R. Rossman, G. Tucker, T.O. Fischmann, A. Hruza, V.
Madison, A.A., Nomeir, Y. Wang, E. Lees, D. Parry, N. Sgambellone, W. Seghezzi, L.
Schultz, F. Shanahan, D. Wiswell, X. Xu, Q. Zhou, R.A. James, V.M. Paradkar, H. Park, L.R.
Rokosz, T.M. Stauffer, T.J. Guzi, Pyrazolo[1,5-a]pyrimidines as orally available inhibitors of
cyclin-dependent kinase 2. Bioorg. Med. Chem. Lett. 17 (2007), 6220–6223.
(29) R. Jorda, K. Paruch, V. Krystof, Cyclin-dependent kinase inhibitors inspired by
roscovitine: purine bioisosteres, Curr Pharm Des. 18 (2012), 2974–2980.
22

ACCEPTED MANUSCRIPT
(30) P.S. Brahmkshatriya, P. Dobeš, J. Fanfrlik, J. Rezác, K. Paruch, A. Bronowska, M.
Lepšík, P. Hobza, Quantum mechanical scoring: structural and energetic insights into cyclin-
dependent kinase 2 inhibition by pyrazolo[1,5-a]pyrimidines, Curr. Comput. Aided Drug Des.
9 (2013), 118–129.
(31) T. Gucký, R. Jorda, M. Zatloukal, V. Bazgier, K. Berka, E. Řezníčková, T. Béres, M.
Strnad, V. Kryštof, A novel series of highly potent 2,6,9-trisubstituted purine cyclin-
dependent kinase inhibitors. J. Med. Chem. 56 (2013), 6234–6247.
(32) R. Jorda, L. Havlícek, I.W. McNae, M.D. Walkinshaw, J. Voller, A. Sturc, J.
Navrátilová, M., Kuzma, M. Mistrík, J. Bártek, M. Strnad, V. Krystof, Pyrazolo[4,3-
d]pyrimidine bioisostere of roscovitine: evaluation of a novel selective inhibitor of cyclin-
dependent kinases with antiproliferative activity, J. Med. Chem. 54 (2011), 2980–2993.
(33) P. Dobeš, J. Fanfrlík, J. Rezáč, M. Otyepka, P. Hobza, Transferable scoring function
based on semiempirical quantum mechanical PM6-DH2 method: CDK2 with 15 structurally
diverse inhibitors, J. Comput. Aided Mol. Des. 25 (2011), 223–235.
(34) P. Dobes, J. Rezác, J. Fanfrlík, M. Otyepka, P. Hobza, Semiempirical quantum
mechanical method PM6-DH2X describes the geometry and energetics of CK2-inhibitor
complexes involving halogen bonds well, while the empirical potential fails, J Phys Chem B.
115 (2011), 8581–8589.
(35) T.J. Guzi, K. Paruch, M.P. Dwyer, M. Labroli, F. Shanahan, N. Davis, L. Taricani, D.
Wiswell, W. Seghezzi, E. Penaflor, B. Bhagwat, W. Wang, D. Gu, Y. Hsieh, S. Lee, M. Liu,
D. Parry, Targeting the replication checkpoint using SCH 900776, a potent and functionally
selective CHK1 inhibitor identified via high content screening, Mol. Cancer Ther. 10 (2011),
591–602.
23

ACCEPTED MANUSCRIPT
(36) A. Klamt, G. Schüürmann. COSMO: A New Approach to Dielectric Screening in
Solvents with Explicit Expressions for the Screening Energy and its Gradient. J. Chem. Soc.
Perkin Trans. 2 (1993), 799–805.
(37) D.A. Pearlman, P.S. Charifson, Are free energy calculations useful in practice? A
comparison with rapid scoring functions for the p38 MAP kinase protein system, J. Med.
Chem. 44 (2001), 3417-3423.
(38) N.R. Brown, M.E. Noble, J.A. Endicott, L.N. Johnson, The structural basis for specificity
of substrate and recruitment peptides for cyclin-dependent kinases, Nat. Cell Biol. 1 (1999),
438–443.
(39) T.G. Davies, J. Bentley, C.E. Arris, F.T. Boyle, N.J. Curtin, J.A. Endicott, A.E. Gibson,
B.T. Golding, R.J. Griffin, I.R. Hardcastle, Structure-based design of a potent purine-based
cyclin-dependent kinase inhibitor, Nat. Struct. Mol. Biol. 9 (2002), 745-749.
(40) Collaborative Computational Project, Number 4, The CCP4 suite: programs for protein
crystallography. Acta Cryst. D 50 (1994), 760–763.
(41) A.J. McCoy, R.W. Grosse-Kunstleve, P.D. Adams, M.D. Winn, L.C. Storoni, L. C.,
R.J.X. Read, Phaser crystallographic software, J. Appl. Cryst. 40 (2007), 658–674.
(42) K. Bettayeb, N. Oumata, A. Echalier, Y. Ferandin, J.A. Endicott, H. Galons, L. Meijer,
CR8, a potent and selective, roscovitine-derived inhibitor of cyclin-dependent kinases,
Oncogene 27 (2008), 5797–807.
(43) P. Emsley, B. Lohkamp, W.G. Scott, K. Cowtan, Features and development of Coot,
Acta Cryst. D. 66 (2010), 486–501.
(44) G.N. Murshudov, A.A. Vagin, E.J. Dodson, Refinement of Macromolecular Structures
by the Maximum-Likelihood Method, Acta Cryst. D 53 (1997), 240–255.
(45) A. Fiser, R.K. Do, A. Sali, Modeling of loops in protein structures, Protein Science 9
(2000), 1753–1773.
24

ACCEPTED MANUSCRIPT
(46) E.F. Pettersen, T.D. Goddard, C.C. Huang, G.S. Couch, D.M. Greenblatt, E.C. Meng,
T.E. Ferrin, UCSF Chimera--a visualization system for exploratory research and analysis. J.
Comput. Chem. 25 (2004), 1605–1612.
(47) D.A. Case, R.M. Betz, W. Botello-Smith, D.S. Cerutti, T.E. Cheatham, III, T.A. Darden,
R.E. Duke, T.J. Giese, H., Gohlke, A.W. Goetz, N. Homeyer, S. Izadi, P. Janowski, J. Kaus,
A. Kovalenko, T.S. Lee, S. LeGrand, P. Li, C. Lin, T. Luchko, R. Luo, B. Madej, D.
Mermelstein, K.M. Merz, G. Monard, H. Nguyen, H.T. Nguyen, I. Omelyan, A. Onufriev,
D.R. Roe, A. Roitberg, C. Sagui, C.L. Simmerling, J. Swails, R.C. Walker, J. Wang, R.M.
Wolf, X. Wu, L. Xiao, P.A. Kollman, AMBER 2014 (2014), University of California, San
Francisco.
(48) Y. Duan, C. Wu, S. Chowdhury, M.C. Lee, G. Xiong, W. Zhang, R. Yang, P. Cieplak, R.
Luo, T. Lee, A point-charge force field for molecular mechanics simulations of proteins based
on condensed-phase quantum mechanical calculations, J. Comput. Chem. 24 (2003), 1999–
2012.
(49) J. Wang, R.M. Wolf, J.W. Caldwell, P.A. Kollman, D.A. Case, Development and Testing
of a General Amber Force Field, J. Comput. Chem. 25 (2004), 1157–1174.
(50) C.I. Bayly, P. Cieplak, W. Cornell P.A. Kollman, A Well-Behaved Electrostatic Potential
Based Method Using Charge Restraints for Determining Atom-Centered Charges: The RESP
Model, J. Phys. Chem. 97 (1993), 10269–10280.
(51) M. Lepsík, Z. Kríz, Z. Havlas, Efficiency of a second-generation HIV-1 protease
inhibitor studied by molecular dynamics and absolute binding free energy calculations,
Proteins 57 (2004), 279–293.
(52) M. Kolar, P. Hobza, On Extension of the Current Biomolecular Empirical Force Field for
the Description of Halogen Bonds, J. Chem. Theory Comput. 8 (2012), 1325–1333.
25

ACCEPTED MANUSCRIPT
(53) D. van der Spoel, E. Lindahl, B. Hess, A.R. van Buuren, E., Apol, P.J. Meulenhoff, D.P.
Tieleman, A.L.T.M. Sijbers, K.A. Feenstra, R. van Drunen H.J.C. Berendsen, Gromacs User
Manual, Version 4.5.4 (2010), www.gromacs.org.
(54) W. Humphrey, A. Dalke, K. Schulten, VMD – Visual Molecular Dynamics, J. Mol.
Graph., 14 (1996), 33–38.
(55) J. Cohen, A. Arkhipov, R. Braun K. Schulten, Imaging the migration pathways for O2,
CO, NO, and Xe inside myoglobin. Biophysical Journal 91 (2006), 1844–1857.
(56) Schrödinger Release 2014-4: Maestro, Version 10.0, Schrödinger, LLC, New York, NY,
2014.
(57) Schrödinger Release 2014-4: WaterMap, Version 2.1, Schrödinger, LLC, New York,
NY, 2014.
(58) R. Abel, T. Young, R. Farid B.J. Berne R.A. Friesner, Role of the active-site solvent in
the thermodynamics of factor Xa ligand binding, J. Am. Chem. Soc. 130 (2008), 2817–2831.
(59) The PyMOL Molecular Graphics System, Version 1.7.6.3., Schrödinger, LLC.
(60) J. Mongan, C. Simmerling, A.J. McCammon, D. Case, A. Onufriev, Generalized Born
with a simple, robust molecular volume correction, J. Chem. Theory Comput. 3 (2007), 156–
169.
(61) M. Kolář, J. Fanfrlík, M. Lepšík, F. Forti, F.J. Luque, P. Hobza, Assessing the accuracy
and performance of implicit solvent models for drug molecules: conformational ensemble
approaches, J Phys Chem B. 117 (2013), 5950–5962.
(62) R. Ahlrichs, M. Bar, M. Haser, H. Horn, C. Kolmel, Electronic Structure Calculations on
Workstation Computers: The Program System Turbomole, Chem. Phys. Lett. 162 (1989),
165–169.
(63) J.J.P. Stewart, MOPAC2012 (2012), Stewart Computational Chemistry, Colorado
Springs, CO; http://OpenMOPAC.net.
26

ACCEPTED MANUSCRIPT
(64) A.V. Marenich, C.J. Cramer, D.G. Truhlar, Universal solvation model based on solute
electron density and on a continuum model of the solvent defined by the bulk dielectric
constant and atomic surface tensions, J. Phys. Chem. B 113 (2009), 6378–6396.
27

ACCEPTED MANUSCRIPT
FIGURE CAPTIONS
Figure 1. Schematical representation of interaction of pyrazolo[1,5-a]pyrimidines with the
CDK2 active site with indicated key amino acids. Three hinge region hydrogen bonds are
dashed.
Scheme 1. Synthesis of intermediates 6 and 7 leading to targets 1a to 1q.
Reagents and conditions: i. LDA, THF, -78 °C, then 4-iodobenzoyl chloride or 4-
bromobenzoyl chloride, THF, -78 °C to r.t., (2a: 81%), (2b: 92%); ii. (a) 3-aminopyrazole or
(b) 3-amino-4-ethylpyrazole, AcOH, reflux, (3a : 61%), (3b: 71%), (3c: 65%); iii. POCl3,
pyr., r.t., (4a: 85%), (4b: 42%), (4c: 95%); iv. 3-picolylamine, MeCN, reflux, (5a: 93%), (5b:
79%), (5c: 97%); v. Boc2O, DMAP, CH2Cl2, r.t., (6a: 92%), (6b: 90%), (6c: 94%); vi.
B2Pin2, K3PO4, PdCl2dppf, DME, H2O, reflux, (7a: 72% from 6a, 85% from 6c), (7b: crude
used as such in next step); vii. K3PO4, PdCl2dppf, DME, H2O, reflux; viii. AcCl, CH2Cl2,
NEt3, 0 °C to r.t., (8d: 82%); ix. NaBH4, MeOH, 0 °C to r.t., (8i: 88%), (8l: 88%); x. NBS,
MeCN, r.t.; xi. 3M aq. HCl, EtOH, 60 °C; xii. TFA, CH2Cl2, r.t.
Scheme 2. Synthesis of compounds 1p, 1r, 1s, 1t and 1u.
Reagents and conditions: i. SOCl2, MeOH, reflux, 95%; ii. acrylonitrile, Pd(P(t-Bu)3)2,
DIPEA, dioxane, 100 °C, 49%; iii. H2, Pd(OAc)2, PPh3, toluene, 80 °C, 70%; iv. SnCl2,
EtOAc, H2O, reflux, 55%; v. isoamylnitrite, CuBr2, CH3CN, 60 °C, 77%; vi. NaOH, THF, r.t.,
quant.; vii. SOCl2, DMF,THF , r.t., then NH3 in dioxane, THF, 0 °C, 72%; viii. Compound
7a, PdCl2dppf, K3PO4, DME, H2O, reflux, (20a: 77%); ix. SOCl2, THF, reflux then NH3 in
MeOH, THF, 0 °C, 66%; x. PivCl, NEt3, CH2Cl2, THF, 75%; xi. Compound 7a, PdCl2dppf,
K3PO4, DME, H2O, reflux, (20b: 40%); xii. Na2CO3, SOCl2, NH3 in MeOH, 80 °C, 68%; xiii.
BCl3, CH2Cl2, r.t., 90%; xiv. Compound 7a, K3PO4, PdCl2dppf, DME, H2O, reflux, (20c:
80%); xv. MsCl, NEt3, DMF, 0 °C, (20d: 87%); xvi. NBS, CH2Cl2, 0 °C, (21a: 91%), (21b:
86%), (21c: 83 %); xvii. NaOH, THF, reflux, (22: 90%); xviii. K2CO3, MeOH, r.t., (23: 86%);
xix. HCl, EtOH, 60 °C, (1r: 85%), (1s: quant.), (1u: 90%) ; xx. HCl, EtOH, 60 °C, (1p:
90%), (1t: 78%).
Figure 2. Inhibitor binding mode, atom numbering and structural water molecules in the
crystallographic complex of CDK2/1l. Three hinge-region hydrogen bonds between the
protein and the inhibitor are shown as dotted lines. Color coding of sticks: grey – carbon, blue
– nitrogen, red – oxygen, brown – bromine. Spheres represent water sites as obtained from
28