Computational design and discovery of nanomolar inhibitors of IκB kinase β

Article abstract of DOI:10.1021/ja510636t

IκB kinase β (IKKβ) is a useful target for the discovery of new medicines for cancer and inflammatory diseases. In this study, we aimed to identify new classes of potent IKKβ inhibitors based on structure-based virtual screening, de novo design, and chemical synthesis. To increase the probability of finding actual inhibitors, we improved the scoring function for the estimation of the IKKβ-inhibitor binding affinity by introducing proper solvation free energy and conformational destabilization energy terms for putative inhibitors. Using this modified scoring function, we have been able to identify 15 submicromolar-level IKKβ inhibitors that possess the phenyl-(4-phenyl-pyrimidin-2-yl)-amine moiety as the molecular core. Decomposition analysis of the calculated binding free energies showed that a high biochemical potency could be achieved by lowering the desolvation cost and the conformational destabilization for the inhibitor required for binding to IKKβ as well as by strengthening the interactions in the ATP-binding site. The formation of two hydrogen bonds with backbone amide groups of Cys99 in the hinge region was found to be necessary for tight binding of the inhibitors in the ATP-binding site. From molecular dynamics simulations of IKKβ-inhibitor complexes, we also found that complete dynamic stability of the bidentate hydrogen bond with Cys99 was required for low nanomolar-level inhibitory activity. This implies that the scoring function for virtual screening and de novo design would be further optimized by introducing an additional energy term to measure the dynamic stability of the key interactions in enzyme-inhibitor complexes.

Full text of DOI:10.1021/ja510636t

Article
pubs.acs.org/JACS
Computational Design and Discovery of Nanomolar Inhibitors of IκB
Kinase β
Hwangseo Park,*,† Yongje Shin,‡ Hyeonjeong Choe,‡ and Sungwoo Hong*,‡
†Department of Bioscience and Biotechnology, Sejong University, Seoul 143-747, Korea
‡Center for Catalytic Hydrocarbon Functionalization, Institute for Basic Science (IBS) and Department of Chemistry, Korea
Advanced Institute of Science and Technology (KAIST), Daejeon, 305-701, Korea
S
* Supporting Information
ABSTRACT: IκB kinase β (IKKβ) is a useful target for the discovery of new
medicines for cancer and inflammatory diseases. In this study, we aimed to
identify new classes of potent IKKβ inhibitors based on structure-based virtual
screening, de novo design, and chemical synthesis. To increase the probability
of finding actual inhibitors, we improved the scoring function for the
estimation of the IKKβ-inhibitor binding affinity by introducing proper
solvation free energy and conformational destabilization energy terms for
putative inhibitors. Using this modified scoring function, we have been able to
identify 15 submicromolar-level IKKβ inhibitors that possess the phenyl-(4-
phenyl-pyrimidin-2-yl)-amine moiety as the molecular core. Decomposition
analysis of the calculated binding free energies showed that a high biochemical
potency could be achieved by lowering the desolvation cost and the
conformational destabilization for the inhibitor required for binding to IKKβ
as well as by strengthening the interactions in the ATP-binding site. The
formation of two hydrogen bonds with backbone amide groups of Cys99 in the hinge region was found to be necessary for tight
binding of the inhibitors in the ATP-binding site. From molecular dynamics simulations of IKKβ-inhibitor complexes, we also
found that complete dynamic stability of the bidentate hydrogen bond with Cys99 was required for low nanomolar-level
inhibitory activity. This implies that the scoring function for virtual screening and de novo design would be further optimized by
introducing an additional energy term to measure the dynamic stability of the key interactions in enzyme−inhibitor complexes.
The multisubunit IKK complex consists of two catalytic
subunits (IKKα and IKKβ) and an essential regulatory subunit
NEMO (IKKγ).4 It is generally agreed that IKKβ plays a more
critical role than IKKα in activating the NF-κB pathway because
the former has even higher kinase activity than the latter.5,6
Therefore, the inhibition of IKKβ has received considerable
attention as an effective strategy for the treatment of
inflammation-related diseases such as asthma, rheumatoid
arthritis, and cancer.7−9
A number of small molecules have been reported through
intensive studies and drug discovery programs for the
development of potent and selective IKKβ inhibitors.10−20 In
addition to various ATP competitive inhibitors, the allosteric
inhibitors have also been shown to selectively impair the
activity of IKKβ.21 Although most IKKβ inhibitors were
identified via the high-throughput screening of a chemical
library and the structural modifications of known inhibitors,
some rational drug design approaches have also been actively
pursued based on a variety of computational methods.22−25
However, the lack of 3D structure of IKKβ has limited the
applicability of the computational methods due to the
INTRODUCTION
■
Nuclear factor kappa-B (NF-κB) protein is a ubiquitously
expressed transcription factor and known to be a key regulator
of immune responses, cell proliferation, cell death, and
inflammation.1,2 In normal cells, NF-κB is maintained in an
inactive state in the cytoplasm through binding to inhibitor of
kappa-B (IκB) anchoring protein. However, when cells are
stimulated by a variety of factors such as interleukin-1, tumor
necrosis factor, lipopolysaccharide, hormone, and other signals,
IκB is phosphorylated by the IκB kinase (IKK) complex, which
results in IκB degradation and NF-κB activation. Ultimately,
NF-κB proteins released from IκB translocate freely into the
nucleus and induce the transcription of genes responsible for
inflammation and cell survival. For this reason, NF-κB has been
regarded as a crucial mediator associated with a number of
human cancers and inflammatory diseases. Several approaches
to suppress NF-κB activation are now being investigated. For
example, a number of therapeutic agents including IKK
inhibitors, proteasome inhibitors, and steroidal and non-
steroidal anti-inflammatory drugs are known to mediate the
blockade of the NF-κB signaling pathway.3 The inhibition of
IKK can thus be a promising strategy for developing
therapeutics for diseases caused by aberrant NF-κB activation.
Received: October 16, 2014
© XXXX American Chemical Society
A
DOI: 10.1021/ja510636t
J. Am. Chem. Soc. XXXX, XXX, XXX−XXX

Journal of the American Chemical Society
Article
the ligand binding affinity by accounting for the effects of ligand
solvation on the protein−ligand association. The modified scoring
function to calculate the protein−ligand binding free energy in
solution (ΔGabq) can be expressed as follows.
considerable uncertainty in estimating the biochemical
potencies of putative inhibitors. Recently, X-ray crystal
structures of IKKβ have been reported both in the resting
form and in complex with a potent inhibitor.26,27 Such
structural information about the interactions between IKKβ
and small-molecule ligands can greatly aid in the design of more
effective IKKβ inhibitors.
The present study was undertaken to identify new potent
IKKβ inhibitors by consecutively applying the virtual screening
of a large chemical library to find a lead compound and de novo
design to optimize the inhibitory activity of the lead by
chemical modifications. The newly identified IKKβ inhibitors
have potential to reveal anticancer activities because the
impairment of IKKβ activity leads to the perturbation of the
inflammatory process required for cancer development. Besides
the discovery of potent IKKβ inhibitors, therefore, we also
investigated the presence of their anticancer activities with
respect to the pancreatic cancer cells.
Even when using a 3D structure of the target protein, virtual
screening and de novo design have not always been successful
due to the imperfections in the scoring function for the
estimation of the binding free energy between the target
protein and a putative ligand.28,29 One of the reasons for this
inaccuracy lies in the neglect of ligand solvation effects, which
leads to the overestimation of the binding affinity of a ligand
with many polar groups.30 Therefore, to enhance the efficiency
of virtual screening and de novo design, we modified the existing
scoring function by adding a proper molecular solvation free
energy function. The addition of this new energy term reflects
the desolvation cost for a ligand to be bound in the ATP-
binding site of IKKβ.
⎛
⎞
Aij
B
ij
ri6j ⎠
ΔGbaq = WvdW ∑ ∑ ⎜
−
⎜
⎟
⎟
12
r
ij
⎝
i=1 j=1
⎛
⎝ ij
⎞
qiqj
Cij
r12
D
ij
⎜
⎜
⎟
⎟
+ Whbond ∑ ∑ E(t)
−
+ W ∑ ∑
elec
10
ij ⎠
ε(r )r
r
ij ij
i=1 j=1
i=1 j=1
2
2
+ W Ntor + W ∑ Si(Omax − ∑ Vje−r /2σ
)
ij
tor
sol
i
i=1
j≠i
(1)
The weighting factors for van der Waals interactions, hydrogen
bonds, electrostatic interactions, and the entropic penalty, and the
molecular solvation free energy term, were represented by WvdW
,
Whbond, Welec, Wtor, and Wsol, respectively. The interatomic distance is
rij, and Aij, Bij, Cij, and Dij, are Leonard-Jones potential parameters. The
hydrogen-bond energy term includes an additional weighting factor,
E(t), to reflect the angle-dependent directionality. In calculating the
electrostatic interaction energy term, we used the sigmoidal function
ε(rij) as the distance-dependent dielectric constant of IKKβ.34 In the
torsional term, Ntor denotes the number of rotatable bonds in the
ligand. The final term in eq 1 represents the desolvation cost of the
ligand for binding to IKKβ and includes three atomic parameters, Si,
Oimax, and Vi, which denote the atomic solvation free energy per unit
volume, maximum atomic occupancy, and fragmental volume for atom
i, respectively.35 This desolvation term was calculated using the atomic
parameters reported by Park because they proved to be successful in
estimating the hydration energies of various organic molecules.36
Using this modified scoring function, we carried out docking
simulations in the ATP-binding pocket of IKKβ to score and rank
the potential IKKβ inhibitors according to the calculated binding
affinities.
Furthermore, most popular docking and de novo design
programs concentrate on finding the conformation of a ligand
that can be bound most tightly in the binding pocket of the
receptor. Although there might be a large energy difference
between this bioactive conformation of a ligand and its
minimum-energy structure,31 the conformational destabiliza-
tion of the ligand is neglected or roughly approximated from
molecular mechanics in most scoring functions. It is therefore
apparent that the inaccuracy of the scoring function can be
attributed, at least in part, to the improper description of the
ligand conformation energy. Therefore, to further enhance the
possibility of finding the potent inhibitors via de novo design, we
calculated the ligand conformational energy term in the scoring
function with ab initio quantum chemical calculations. We thus
anticipated that the present virtual screening and de novo design
procedures would enrich the chemical library with molecules
that have a good inhibitory activity against IKKβ. Finally, we
also addressed the dynamical features required for highly potent
IKKβ inhibitors based on molecular dynamics (MD)
simulations of the IKKβ−inhibitor complexes. The results
obtained in these comprehensive computational and exper-
imental studies are expected to provide useful information for
the discovery of new potent IKKβ inhibitors.
De novo Design. The structure-guided de novo design of IKKβ
inhibitors was performed in three steps. First, we made some structural
modifications to the initial IKKβ inhibitor found from the preceding
virtual screening to obtain the best scaffold from which even more
potent inhibitors could be derivatized. The LigBuilder program37 was
used in this structural transformation to obtain phenyl-(4-phenyl-
pyrimidin-2-yl)-amine (PPA) as a promising scaffold for designing new
potent inhibitors. In the second step, a variety of PPA derivatives were
generated based on the calculated binding mode of PPA in the ATP-
binding site of IKKβ. To obtain various PPA derivatives as candidates
for a potent inhibitor, a genetic algorithm was employed to generate
the derivatives by changing the chemical moieties at the substitution
positions. In this step, we used the empirical scoring function
suggested by Wang et al.38 to score and rank the generated PPA
derivatives. Because the central pyrimidin-2-yl amine group of PPA
appeared to be bound tightly to the ATP-binding site, we limited the
substitution positions to the two terminal phenyl rings. To reduce the
computational time, only the generated derivatives that could satisfy
the bioavailability rules as a drug candidate were selected for further
analysis. Based on the filtration criteria, we obtained 3617 PPA
derivatives that were estimated to have higher inhibitory activity
against IKKβ than PPA itself.
In the final de novo design step, the generated PPA derivatives were
further screened with the new scoring function constructed by
introducing a conformational destabilization energy term of ligand into
the modified scoring function shown in eq 1. This new scoring
function has the following form.
MATERIALS AND METHODS
■
Virtual Screening of IKKβ Inhibitors with Docking Simu-
lations. We used the automated AutoDock program32,33 to calculate
the binding mode and binding free energy of each molecule in the
docking library. The procedure for constructing this docking library is
detailed in Supporting Information. To improve the accuracy of virtual
screening, the original AutoDock scoring function was modified to
include a solvation free energy term for the ligand. The introduction of
this additional energy term seemed to guarantee better prediction of
ΔGbconf = ΔGbaq + ΔEconf
(2)
Here, ΔEconf represents the energy required for the conformational
change of a ligand from its ground state to the best binding mode in
the ATP-binding site of IKKβ. This new energy term was augmented
to the scoring function to enhance the accuracy of predicting the
binding free energy of a ligand and could be given by the electronic
B
DOI: 10.1021/ja510636t
J. Am. Chem. Soc. XXXX, XXX, XXX−XXX

Journal of the American Chemical Society
Article
energy difference between the fully optimized structure of the ligand
and its best binding conformation with respect to IKKβ. We calculated
this conformational destabilization energy with ab initio calculations at
B3LYP/3-21G* level of theory. The PPA derivatives selected from the
precedent de novo design were then rescored according to the binding
affinities for IKKβ calculated from additional docking simulations with
the improved scoring function (ΔGcbonf) in eq 2. Finally, the 100 top-
ranked derivatives were inspected for the possibility of chemical
synthesis.
RESULTS AND DISCUSSION
Virtual Screening and Enzyme Assays. Of the 260,000
■
compounds in the docking library screened for tight binding in
the ATP-binding site of IKKβ, the 100 top-ranked compounds
were selected as virtual hits. All these compounds were tested
for the presence of inhibitory activity against IKKβ at a
concentration of 50 μM by a high-throughput binding assay.
Five compounds revealed significant biochemical potency with
percent of control (POC) values lower than 30, and these were
selected for the determination of the IC50 values. The synthetic
intermediates to obtain the candidate compounds for IKKβ
inhibitors were also screened for biochemical potency with
respect to IKKβ. However, no significant activity was observed
at the concentration of 50 μM. The chemical structures and
inhibitory activities of the newly found IKKβ inhibitors are
presented in Figure 1. As the common structural features,
Synthesis of PPA Derivatives. As illustrated in Scheme 1, the
designed derivatives could be conveniently prepared by a two-step
Scheme 1. Preparation of Aminopyrimidine or Triazine
a
Derivatives
a
Reagents and conditions: (a) PdCl2(PPh3)2, arylboronic acid, K3PO4,
1,4,-dioxane/H2O, 50 or 70 °C, 4−12 h; (b) Pd(PPh3)4, 2 N Na2CO3,
MeCN, 90 °C, 3 h; (c) aniline, cat. HCl, EtOH, 150 °C, 30 min,
microwave; (d) aniline, DMF, 80 °C, 30 min, then 1 N HCl, 80 °C, 4
h; (e) p-aminobenzenesulfonamide, DMF, 0 °C, 3 h; (f) Pd(dppf)Cl2·
CH2Cl2, arylboronic acid, K3PO4, 1,4-dioxane/H2O, 140 °C, 1 h,
microwave.
Figure 1. Chemical structures and IC50 values of five IKKβ inhibitors
identified through the virtual screening of a chemical library with
molecular docking.
sequence: (1) aryl groups were installed at the 4-position of the
pyrimidine core via palladium-catalyzed cross-coupling reactions with
corresponding arylboronic acids; next, (2) the chloro compounds 1
were substituted with appropriate anilines by an SNAr reaction to allow
for the direct facile conversion to the product 2. In the case of the
triazine scaffold, a 4-sulfonamide aniline group was introduced at the
triazine ring prior to Suzuki coupling to afford the target compounds.
Enzyme Assays. The inhibitory activities of all compounds with
respect to IKKβ were measured by Reaction Biology Corp. (Malvern,
PA, USA) by means of radiometric kinase assays ([γ-33P]-ATP). The
enzymatic activity of IKKβ was monitored using 20 μM of IKKtide
substrate dissolved in the freshly prepared Reaction Buffer (20 mM
HEPES (pH 7.5), 10 mM MgCl2, 1 mM EGTA, 0.02% BRIJ-35, 0.02
mg/mL BSA, 0.1 mM Na3VO4, 2 mM DTT, 1% DMSO). Each
putative IKKβ inhibitor was dissolved in 100% DMSO at the specific
concentration and diluted in serial manner with epMotion 5070 in
DMSO. Four nM of IKKβ/IKBKB (Invitrogen) was added into the
reaction buffer including 20 μM of IKKtide substrate. After delivering
the candidate inhibitor dissolved in DMSO into the kinase reaction
mixture by Acoustic technology (Echo550; nanoliter range), the
reaction mixture was incubated for 20 min at room temperature. To
initiate the enzymatic reaction, 33P-ATP with specific activity of 10
μCi/μL was delivered into the reaction mixture to reach the final ATP
concentration of 1 μM. Radioactivity was then monitored by the filter-
binding method after the incubation of reaction mixture for 2 h at
room temperature. At given concentrations of the inhibitor,
biochemical potency was measured by the percent remaining kinase
activity with respect to vehicle (dimethyl sulfoxide) reaction. Curve fits
and IC50 values were then obtained using the PRISM program
(GraphPad Software). The ATP-competitive inhibitor staurosporine
(STSP) was employed as the positive control in this study because of
its high biochemical potency against various kinases including IKKβ.39
compounds 5−9 possess several hydrogen-bonding groups and
nonpolar aromatic rings. Hence, they seem to be capable of
establishing multiple hydrogen bonds and hydrophobic
contacts in the ATP-binding site of IKKβ.
To address the inhibitory mechanisms of the newly identified
IKKβ inhibitors, we compared their binding modes calculated
with the modified AutoDock scoring function (eq 1). The
lowest-energy conformations of 5−9 are overlaid in Figure 2a
in the ATP-binding site of IKKβ. All five inhibitors appear to be
stabilized in the unique binding pocket comprising the Gly loop
(residues 20−30), the hinge region (residues 95−100) of the
ATP-binding site, and the activation loop (residues 166−194)
at the interface of the N- (residues 1−109) and C-terminal
(residues 110−307) domains. It is a common feature in the
calculated binding modes for all five inhibitors that they involve
multiple hydrogen bonds and van der Waals contacts with the
residues in the hinge region and those at the top of the C lobe,
respectively. The binding modes of 5−9 are also similar in that
some nonpolar moieties are directed to the Gly loop, which
would act as a receptor with respect to the phosphate group in
ATP binding. Therefore, the formation of hydrogen bonds with
the hinge region and hydrophobic interactions with the Gly
loop seem to contribute to the inhibition of IKKβ by the
effective blocking of ATP binding in a cooperative fashion. As a
check on the possibility of the allosteric inhibition of IKKβ by
5−9, we conducted additional docking simulations with
extended 3D grid maps to include the whole kinase domain.
C
DOI: 10.1021/ja510636t
J. Am. Chem. Soc. XXXX, XXX, XXX−XXX

Journal of the American Chemical Society
Article
Energetic Features of the Identified Inhibitors. To
investigate the effect of adding a ligand solvation term to the
scoring function on the accuracy of the calculated binding
affinities for IKKβ, we carried out a decomposition analysis of
the calculated binding free energies of 5−9. Because the
binding free energy for the protein−ligand complex in solution
(ΔGabq) can be given by subtracting the ligand solvation free
energy (ΔGsol) from that in the gas phase (ΔGgbas), we
computed the two energy terms separately to estimate their
relative contributions to ΔGaq. Table 1 lists ΔGbgas and ΔGabq
Table 1. Binding Free Energies in the Gas Phase (ΔGgbas) and
in Solution (ΔGabq), Solvation Free Energy (ΔGsol), and
Conformational Destabilization Energy (ΔEconf) for the
a
IKKβ Inhibitors 5−9
b
compd
ΔGgbas
ΔGsol
ΔGabq
ΔEconf
ΔGcbonf
ΔGebxp
5
6
7
8
9
−24.3
−26.8
−23.9
−24.3
−22.5
−12.7
−14.5
−13.3
−13.8
−11.2
−11.6
−12.3
−10.6
−10.5
−11.3
3.0
5.6
4.7
4.1
−8.6
−6.7
−5.9
−6.4
−4.7
−7.0
−6.4
−5.5
−5.4
−5.3
6.6
b
a
Each energy value is given in kcal/mol. The experimental ΔGb
(ΔGebxp) values were obtained from the corresponding Ki values
calculated using the Cheng−Prusoff equation [Ki = IC50/(1 + [ATP]/
Km)].
Figure 2. (a) Comparison of the docking poses for 5−9 in the ATP-
binding site of IKKβ. The carbon atoms of 5−9 are indicated in green,
cyan, black, gray, and pink, respectively. (b) Detailed binding mode of
5 in the ATP-binding site of IKKβ. The carbon atoms of 5 and IKKβ
are shown in green and cyan, respectively. Hydrogen bonds are
indicated with dotted lines.
values for the five inhibitors in comparison to the
corresponding ΔGsol and the experimental ΔGb (ΔGebxp) values.
The calculated binding free energies of 5−9 are in better
agreement with the experimental results due to the change of
the original scoring function (ΔGgbas) to a more physically
reasonable one (ΔGabq) that includes the ligand desolvation
term (−ΔGsol). For example, the ΔGgbas value of 5 appears to be
higher than that of 6 by 2.5 kcal/mol and similar to those of 7
and 8, in conflict with its biochemical potency against IKKβ
being highest. On the other hand, 5 is found to have higher
ΔGsol value than 6−8, which implies a lower desolvation cost
for binding in the ATP-binding site of IKKβ. Thus, 5 can only
be identified as a potent IKKβ inhibitor using a scoring function
that takes into account the desolvation cost for a ligand to form
the required protein−ligand complex as well as the strength of
interactions in the ATP-binding site. This result exemplifies the
need to include the ligand solvation free energy term in the
scoring function for successful virtual screening.
Despite the significant improvement, 6 is still predicted to be
a more potent inhibitor than 5 according to the ΔGabq values in
contrast to the experimental results. Furthermore, the ΔGbaq
value of 9 appears to be similar to that of 5, which is
inconsistent with the more than 10-fold lower potency of the
former (Figure 1). These discrepancies imply the necessity for
further modification of the scoring function.
Recognizing the possibility that the conformational energy
change of an inhibitor upon binding to the target protein
(ΔEconf) would have a significant influence on the binding
affinity of an enzyme−inhibitor complex,40,41 we examined the
effect of including the quantum mechanically calculated
conformational destabilization energy on the accuracy of the
scoring function for the virtual screening of IKKβ inhibitors.
Listed in the fifth column of Table 1 are the ΔEconf values of 5−
9 associated with the conformational change from their fully
optimized geometries to the docked poses in the ATP-binding
site of IKKβ. These ΔEconf values, calculated at B3LYP/3-21G*
However, no peripheral binding pocket was found in which 5−
9 could be stabilized with a negative binding free energy. It is
thus expected that the inhibitory activity of 5−9 stems from
specific binding in the ATP binding site of IKKβ.
Because 5 is a small molecule with a molecular weight of
294.3 and has inhibitory activity at the micromolar level,
detailed analysis of its binding mode would be helpful for
designing new inhibitors with increased potency. Figure 2b
illustrates the lowest-energy binding mode of 5. We note that
the [1,3,5]triazine-2,4-diamine group of 5 donates two
hydrogen bonds to the backbone aminocarbonyl oxygens of
Glu97 and Cys99 and acts as a hydrogen-bond receptor with
respect to the backbone amidic nitrogen of Cys99. These three
hydrogen bonds seem to play a critical role in the inhibition of
IKKβ by 5, as the establishment of multiple hydrogen bonds
with the backbone amide groups in the hinge region was found
to be necessary for the binding of the inhibitor in the ATP-
binding site of IKKβ.27 In the calculated IKKβ-5 complex, two
additional hydrogen bonds are established at the top of the C
lobe between the terminal aniline moiety of 5 and the
aminocarbonyl oxygens of Glu149 (backbone) and Asn150
(side chain). These two hydrogen bonds also seem to serve as a
significant binding force to accommodate 5 in the ATP-binding
site. Further stabilization of 5 in the binding pocket appears to
be achieved through the hydrophobic interactions of its
aromatic rings with the nonpolar side chains of Leu21, Val29,
Val74, Tyr98, Val152, and Ile165. Judging from the features of
the calculated binding mode of 5, the low micromolar
inhibitory activity can be attributed to the simultaneous
establishments of van der Waals contacts and multiple
hydrogen bonds in the ATP-binding site of IKKβ.
D
DOI: 10.1021/ja510636t
J. Am. Chem. Soc. XXXX, XXX, XXX−XXX

Journal of the American Chemical Society
Article
Table 2. Calculated ΔGgbas, ΔGsol, and ΔGabq Values of the
Newly Designed Inhibitor Scaffolds (11−13) in Comparison
level of theory, are small in magnitude as compared to the
ΔGsol and ΔGgbas values. This is not surprising because the
molecular electronic energy is essentially calculated based on
variational methods. Nonetheless, the introduction of the
ΔEconf term into the scoring function improves the accuracy in
estimating the ligand binding affinity to a significant extent.
When ΔGebxp values were compared to those of ΔGabq plus
ΔEconf (ΔGbconf), for example, it follows immediately that the
binding affinities of 6 and 9 with respect to IKKβ can fall below
that of 5 due to the higher conformational destabilization upon
binding in the ATP-binding site. Related to the accuracy
enhancement by adding the ΔEconf term, the root-mean-square
(RMS) error with respect to the ΔGebxp values of 5−9 decreases
(from 5.35 to 0.95 kcal/mol) when the scoring function
changes from ΔGabq to ΔGbconf. Overall, the squared Pearson
correlation coefficient (R2) between the experimental and
computational binding free energies of 5−9 increases from 0.31
to 0.76 due to the modifications of the scoring function by the
addition of ΔGsol and ΔEconf terms. This indicates that the
scoring function becomes even more powerful when the effects
of ligand solvation and conformational destabilization are taken
into account simultaneously.
a
with Those of the Input Scaffold (10)
scaffold
10
11
12
13
ΔGgbas
ΔGsol
ΔGabq
−15.9
−6.8
−9.1
−17.1
−5.5
−11.6
−16.7
−6.4
−10.3
−16.9
−7.2
−9.7
a
Each energy value is given in kcal/mol.
aromatic rings. 11−13 seem to be the better inhibitor scaffolds
than 10, because the former are predicted to bind in the ATP-
binding site of IKKβ more tightly than the latter. Using 11−13
as the input structures, we carried out the second structure-
based de novo design to find new potent IKKβ inhibitors. To
increase the accuracy, we used ΔGcbonf in eq 2 as the scoring
function to estimate the biochemical potencies of the generated
derivatives of 11−13 considering the desolvation cost and the
conformational destabilization associated with binding to IKKβ.
To address the possibility of tight binding of the newly
designed IKKβ inhibitor scaffolds in the ATP-binding site, we
compared their binding modes. Figure 4 shows the binding
De novo Design for the Change of the Inhibitor
Scaffold. Although 5−9 are newly identified IKKβ inhibitors
and expected to have satisfactory physicochemical properties
for drug candidates, they are poor lead compounds for drug
discovery due to the low biochemical potencies. Therefore, we
decided to make some structural modifications to the identified
inhibitors to find more potent IKKβ inhibitors with at least
submicromolar-level inhibitory activity. Because 5−9 were
actually selected from a chemical library of commercially
available compounds that had been synthesized for a different
purpose rather than as IKKβ inhibitors, it would be reasonable
to change their structural skeletons to design the new potent
IKKβ inhibitors. For this reason, we started with the structure-
based de novo design of new IKKβ inhibitors from the design of
the new inhibitor scaffolds by modifying the 2-(4-phenoxy-
phenyl)-[1,3,5]triazine (10) moiety contained in 5.
Two major criteria were applied to generate the new
inhibitor scaffolds with de novo design. First, we selected only
the structures similar to 10 with a Tanimoto coefficient higher
than 7.0 because 10 proved to be well-accommodated in the
ATP binding site (Figure 2). The output structures were
screened further for having the lower ΔGabq values than that of
10 to select a good inhibitor scaffold with an increased binding
affinity for IKKβ.
Figure 4. Calculated binding modes of 11−13 in the ATP-binding site
of IKKβ. The carbon atoms of IKKβ, 11, 12, and 13 are indicated in
cyan, green, gray, and pink, respectively. Hydrogen bonds are indicated
with dotted lines.
modes of 11−13 calculated with the modified AutoDock
scoring function. All three putative inhibitor scaffolds appear to
be in close contact with the hinge region through the two
hydrogen bonds with the backbone groups of Cys99 in the
bidentate form. This result is consistent with the X-ray
crystallographic study on the IKKβ-inhibitor complex, in
which double hydrogen bonds with the backbone amide
groups of the hinge region were found to be necessary for the
effective inhibition of IKKβ.27 It is also a common feature in the
calculated binding modes of 11−13 that three aromatic rings
are stabilized through the hydrophobic interactions with the
nonpolar residues in the Gly loop (Leu21 and Val29) and those
at the top of C lobe (Val152 and Ile165). Because the para
positions of the two terminal phenyl rings of 11−13 are farthest
from the protein atoms, many potent IKKβ inhibitors seem to
be identified by introducing various chemical moieties at the
two positions.
As a consequence of this multistep de novo design, we were
able to find three structures (11−13) that were anticipated to
be promising scaffolds from which new potent IKKβ inhibitors
could be derivatized. The structures and calculated binding free
energies of 11−13 are shown in Figure 3 and Table 2,
respectively, in comparison with those of 10. As can be
expected from the use of a similarity criterion in the de novo
design, 11−13 are structurally similar with three six-membered
Structures and Potencies of the Designed and
Synthesized IKKβ Inhibitors. The PPA moiety might not
be a new scaffold in the field of kinase inhibitors in the strict
sense because it contains the amino pyrimidine group that has
been considered as a useful structural element of kinase
Figure 3. Structures of the inhibitor scaffolds (11−13) identified with
de novo design using 10 as the input structure.
E
DOI: 10.1021/ja510636t
J. Am. Chem. Soc. XXXX, XXX, XXX−XXX

Journal of the American Chemical Society
Article
inhibitors. However, many kinase inhibitors derivatized from
the amino pyrimidine core were shown to have too poor
physicochemical properties for cellular activity and bioavail-
ability.42−44 In this study, therefore, we aimed to find the new
PPA-based IKKβ inhibitors with novel side chains through the
simultaneous optimization of biochemical potency and drug-
like properties in the structure-based de novo design. As a
consequence of this design effort, we were able to synthesize
the PPA derivatives as novel IKKβ inhibitors that were
expected to have better physicochemical properties and
biochemical potencies than the known ones with the amino
pyrimidine moiety.
Synthesis of compounds bearing modifications on the phenol
group was carried out by Mitsunobu or SN2 reactions to
provide the linked analogues 15. In addition, phenol
intermediate 14 was reacted with substituted 1-fluoro-2-
nitrobenzene to undergo nucleophilic aromatic substitution to
produce the desired nitro derivatives 16, which were further
reduced under Fe/NH4Cl conditions to afford amino
derivatives 17.
0079543A1), no structural and biological data have been
reported so far. Hence, 18−42 are the first reported
compounds as IKKβ inhibitor.
In Table 3, it should be noted that the majority of designed
inhibitors include the sulfonamide moiety on the terminal
phenyl ring connected to the central heterocyclic structure with
the −NH− linker (X position). Because the terminal phenyl
ring is most likely to reside between the hinge region and Gly
loop (Figures 2 and 4) that play a critical role in ATP binding,
the introduced sulfonamide group seems to serve as an effective
surrogate for the phosphate group of ATP. The introduction of
a sulfonamide group at the X position allows the high inhibitory
activity against IKKβ to be maintained for a variety of
substituents at the para position of the terminal phenyl ring
at the opposite side of the PPA core (Y position), which is
evidenced by the submicromolar inhibitory activities of 18, 24−
27, 33−39, 41, and 42. The necessity of a terminal sulfonamide
moiety for high biochemical potency becomes more apparent
when the IC50 value of 27 is compared to those of 28−32,
which indicates that change or removal of the sulfonamide
group leads to a 30−150-fold decrease in the inhibitory activity.
Thus, a derivative of 11 with a sulfonamide moiety at the X
position can be suggested as a good lead compound for drug
discovery with IKKβ as the target protein.
Scheme 2. Synthesis of Elaborated Aminopyrimidine
a
Derivatives
Whereas the replacement of chloride at Y position in 18 with
intact and fluorinated phenoxy groups in 21−24 lowers the
biochemical potency by 10−100 times, some highly potent
inhibitors with IC50 values of 60−80 nM can be obtained upon
the substitution of 2-aminoethoxy group (25 and 26) at Y
position. Judging from the IC50 values of 35−39 lower than
those of 33 and 34 by 1 order of magnitude, the nitrophenoxy
moiety seems to be the better substituent than the amino-
phenoxy moiety in terms of conferring high inhibitory activity
against IKKβ. A potent IKKβ inhibitor (41) with a similar IC50
value to those of 35−39 can also be obtained by simply
substituting the −NH2 group at the Y position. It is also
noteworthy that the biochemical potency decreases to the
micromolar level in going from 35 to 40 due to the methyl
substitution at the terminal sulfonamide group, which confirms
its crucial role as a structural element of potent IKKβ inhibitors.
Among the variety of substituents at the Y position studied,
the sulfonamide moiety in 42 is found to be most efficient for
the inhibition of IKKβ. As a consequence of the substitutions of
two sulfonamide moieties at the X and Y positions of the PPA
core, the highly potent IKKβ inhibitor 42 is identified, with an
IC50 value of 4 nM. Because the terminal phenyl ring that
includes the Y position is expected to reside at the top of the C
lobe (Figure 4), the sulfonamide group at the Y position seems
to be stabilized through hydrogen bonds with the polar groups
at the interface between the Gly loop and the C lobe.
a
Reagents and conditions: (a) 2-(dimethylamino)-ethyl chloride·HCl,
KI, K2CO3, DMF, 90 °C, 2 h, then 1.25 M HCl in MeOH, rt, 30 min;
(b) N-Boc-ethanolamine, PPh3, DIAD, THF, rt, 48 h, then 1.25 M
HCl in MeOH, 50 °C, 1 h; (c) substituted 1-fluoro-2-nitrobenzene,
K2CO3, DMSO, 130 °C, 2 h; (d) Fe, NH4Cl, EtOH/H2O, 85 °C, 30
min.
Table 3 lists the structures and IC50 values of various
derivatives of 11−13 that were selected for synthesis among the
virtual hits of de novo design. Of the 25 derivatives synthesized
and tested, 15 compounds reveal submicromolar-level inhib-
itory activities against IKKβ. This high hit rate exemplifies the
accuracy of the modified scoring function (ΔGcbonf in eq 2) with
which the new inhibitors can be designed to maximize the
binding affinity for IKKβ and simultaneously minimize the
desolvation cost and the conformational destabilization
associated with binding in the ATP-binding site of IKKβ. As
a check on the novelty of the synthesized IKKβ inhibitors, we
examined the availability of structural and functional data for all
24 synthesized compounds in the two popular chemical
databases, SciFinder and PubChem. Compounds 19−42 were
found to be completely new substances. Although 18 proved to
be one of synthetic examples of the patent (US2006/
Consistent with the ΔGabq value of 11 being lower than those
of 12 and 13, which is mainly due to the decrease in the
desolvation cost (Table 2), the inhibitory activity of 18 is found
to be even higher than those of 19 and 20 by more than 3
orders of magnitude. This result is surprising due to the high
degree of similarity among the molecular structures of 18−20.
Such an unexpectedly large difference in the IC50 values actually
indicates the involvement of an additional factor beyond the
desolvation effect in the variation of inhibitory activities with
respect to the structural perturbation in the central aromatic
heterocycle.
To investigate the biological activities of the IKKβ inhibitors
found in this study, we carried out the cell-based assays for the
F
DOI: 10.1021/ja510636t
J. Am. Chem. Soc. XXXX, XXX, XXX−XXX

Journal of the American Chemical Society
Article
a
Table 3. Structures and Inhibitory Activities of the Derivatives of 11−13 against IKKβ
a
Asterisk indicates the atom bonded to substitution position.
26, 35, 41, and 42 using the pancreatic cancer cell lines (BxPC-
3). In this study, cell viability was measured at varying
concentrations of the inhibitors with 3-(4,5-dimethylthiazol-2-
yl)-2,5-diphenyltetrazolium bromide method. All four com-
pounds appear to have significant inhibitory activity against the
proliferation of the pancreatic cancer cell lines in a dose-
dependent manner with IC50 values ranging from 3 to 10 μM
(Table 4). These apparent antiproliferative activities of 26, 35,
study to become a good lead compound for the discovery of
new cancer medicines.
As a check on the promiscuous inhibition of various kinases
by the IKKβ inhibitors identified in this study, we investigated
the selectivity profiles of the two representative compounds (26
and 35) via high-throughput target screening assay (KINO-
MEscan, Ambit Biosciences). The inhibitory activities of 26 and
35 were measured by POC values at the concentration of 10
μM over a panel of 50 cancer-related kinases. The results for
the measurements of these off-target activities are presented in
Table S1. Consistent with the nanomolar inhibitory activities
for IKKβ, both compounds appear to bind tightly at the ATP-
binding site of IKKβ with zero POC value. Aside from the two
IKK isoforms, only glycogen synthase kinase 3 beta (GSK3β)
and vascular endothelial growth factor receptor 2 (VEGFR2)
are found to be adequate for effective binding of 26 with the
associated POC values lower than 20. Similarly, 35 reveals a
significant off-target binding affinity only for stem cell factor
receptor (c-KIT), polo-like kinase 3 (PLK3), and MAP/
microtubule affinity-regulating kinase 3 (MARK3) with the
POC values of 9.5, 20, and 21, respectively. Such a preference
of binding for a few kinases supports the possibility that the
impairment of kinase activity of IKKβ by the inhibitors found in
Table 4. IC50 Values with Respect to the Inhibition of BxPC-
3 Cell Proliferation by IKKβ Inhibitors
26
35
41
42
IC50 (μM)
6.2
9.6
3.7
8.5
41, and 42 with respect to BxPC-3 cancer cell line imply that
they can serve as a starting point for the discovery of new
anticancer medicines. However, we note that the biochemical
potencies decrease from nanomolar level in enzyme inhibition
assays to micromolar level in cell-based assays. As widely
discussed in the literature,45−47 the relatively low anticellular
activity of highly potent enzyme inhibitors can be attributed in
a large part to the poor cell permeability. Therefore, further
structural modifications to improve the cellular permeability
seem to be necessary for the IKKβ inhibitors found in this
G
DOI: 10.1021/ja510636t
J. Am. Chem. Soc. XXXX, XXX, XXX−XXX

Journal of the American Chemical Society
Article
this study would stem from the specific binding in the ATP-
binding site rather than compound promiscuity.
We next focus our attention on the energetic features
associated with the stabilization of the newly identified IKKβ
inhibitors in the ATP-binding site. Table 5 lists the calculated
solvent due to the screening effects of the neighboring phenyl
and −NH− groups. In addition to such a reduced desolvation
cost, the 2-aminopyrimidine core in 18 can also achieve the cis
conformation with respect to the C−NH bond without a
significant loss of conformational stability because there is little
steric repulsion between the central pyrimidine and the
neighboring phenyl ring, which facilitates the establishment of
a stable bidentate hydrogen bond with the backbone groups of
Cys99. This is actually not the case for 4-aminopyrimidine core
in 19 because the exchange of the nitrogen and −CH− group
at the Z and W positions (Table 3) would force the two
aromatic rings out of planarity to avoid a poor van der Waals
contact between the hydrogen at the Z position and the
neighboring phenyl ring. The weakening of the biochemical
potency with the change of the scaffold from 2- to 4-
aminopyrimidine was also observed in SAR studies for the
inhibitors of Bcr-Abl kinase.48 In case of 20, the low
biochemical potency can be attributed in large part to the
increase in the desolvation cost for binding to IKKβ due to the
change of the central aromatic ring from pyrimidine to triazine,
which is implied in the 2.2 kcal/mol decrease of the ΔGsol value
in going from 18 to 20.
The importance of the contribution from the conformational
destabilization for a ligand to the binding free energy can
become more apparent by comparing the ΔGabq and ΔEconf
values of potent inhibitors, such as 41 and 42, to those of
inhibitors with moderate strength including 21, 28, and 40.
Although the average ΔGabq value of the former is 1.6 kcal/mol
higher than that of the latter, a decrease of more than 2.7 kcal/
mol in the conformational destabilization for binding in the
ATP-binding site makes the ΔGcbonf values of the former lower
than those of the latter, which is consistent with the
experimental results.
The high inhibitory activities of 35−39 are surprising
because they are predicted to have a relatively high desolvation
cost and to undergo significant conformational destabilization
upon binding to IKKβ (Table 5). However, as can be inferred
from the highly negative ΔGgbas values, both potency-lowering
effects can be successfully compensated by strengthening the
interactions with the residues in the ATP-binding site. This
indicates that one can also augment the potency of an IKKβ
inhibitor with structural modifications in such a way as to make
the interactions with IKKβ strong enough to surmount both
the increased desolvation cost and the conformational
destabilization upon binding in the ATP-binding site. Thus,
the decomposition analyses of ΔGcbonf provide energetic insight
into the biochemical action of IKKβ inhibitors.
Because all inhibitors possessing sulfonamide and nitro-
phenoxy groups at the X and Y positions (35−39), respectively,
were shown to have high biochemical potency due to the low
ΔGgbas values, we investigated their binding modes with docking
simulations. As shown in Figure 5, both the sulfonamide and
nitrophenoxy moieties of 35 and 38 appear to be in close
contact with protein groups. More specifically, their nitro-
phenoxy group is well-accommodated in a small binding pocket
comprising the side chains of Asn28, Lys44, Lys147, Glu149,
Asn150, Ile165, and Asp166. Because most of these amino acid
residues are hydrophilic, the terminal nitro group of 35−39
seems to be stabilized through the attractive electrostatic
interactions. More remarkably, we see that the sulfonamide
moiety of 35 and 38 establishes three hydrogen bonds around
the ATP-binding site with the side-chain of Lys106 and the
backbone amide groups of Leu21 and Asp103. This is why the
Table 5. Calculated ΔGgbas, ΔGsol, ΔGbaq, ΔEconf, and ΔGcbonf
values (in kcal/mol) for the IKKβ Inhibitors in Comparison
to the Corresponding ΔGebxp Values Calculated Using the
Cheng−Prusoff Equation
compd
ΔGgbas
ΔGsol
ΔGabq
ΔEconf
ΔGcbonf
ΔGebxp
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
STSP
−22.0
−22.4
−21.5
−25.3
−26.0
−25.5
−25.2
−25.0
−25.1
−27.0
−26.7
−25.5
−23.8
−23.2
−24.8
−27.6
−26.3
−27.1
−27.9
−28.7
−28.2
−28.4
−26.7
−24.1
−27.8
−26.6
−11.2
−12.5
−13.4
−13.6
−13.8
−14.1
−14.6
−12.9
−13.3
−15.4
−13.6
−15.0
−12.5
−11.2
−11.8
−16.2
−15.3
−14.7
−14.4
−15.4
−15.1
−14.9
−13.8
−13.3
−16.6
−14.8
−10.8
−9.9
−8.1
1.7
3.7
2.3
4.6
5.4
4.5
4.2
4.0
4.7
3.2
5.3
5.6
5.9
5.6
6.3
4.1
3.4
4.6
4.3
4.8
3.7
3.9
6.1
1.9
1.8
2.2
−9.1
−6.2
−5.8
−7.1
−6.8
−6.9
−6.4
−8.1
−7.1
−8.4
−7.8
−4.9
−5.4
−6.4
−6.7
−7.3
−7.6
−7.8
−9.2
−8.5
−9.4
−9.6
−6.8
−8.9
−9.4
−9.6
−10.5
−6.1
−6.2
−7.9
−8.2
−8.1
−9.0
−9.7
−9.8
−9.6
−7.6
−6.6
−6.9
−6.7
−6.7
−8.8
−8.2
−10.2
−10.0
−10.5
−10.3
−10.2
−8.1
−10.1
−11.4
−10.3
−11.7
−12.2
−11.4
−10.6
−12.1
−11.8
−11.6
−13.1
−10.5
−11.3
−12.0
−13.0
−11.4
−11.0
−12.4
−13.5
−13.3
−13.1
−13.5
−12.9
−10.8
−11.2
−11.8
ΔGgbas, ΔGabq, ΔGsol, ΔEconf, and ΔGbconf values for 18−42 in
comparison to the corresponding ΔGebxp values. The standard
deviations of ΔGgbas, ΔGsol, and ΔEconf values among 25
inhibitors amount to 2.0, 1.4, and 1.3 kcal/mol, respectively.
This similarity indicates that these three energy components
contribute to ΔGcbonf to a comparable extent and confirms that
both the desolvation cost for a ligand and its conformational
destabilization upon binding to IKKβ should be considered
important energy terms in the scoring function for virtual
screening and de novo design of potent IKKβ inhibitors.
Therefore, to increase the biochemical potency of an IKKβ
inhibitor with structural modifications, the resultant strengthen-
ing of enzyme−inhibitor interactions should be sufficient to
surmount the combined potency-lowering effects that stem
from the increased stabilization in solution and the increased
conformational destabilization at the ATP-binding site.
Comparison of the ΔGgbas, ΔGbaq, ΔGbconf, and ΔGebxp values of
18 with those of 19 and 20 clearly reveals that the much lower
inhibitory activities of the latter can be attributed to the
combined effect of the increases in the desolvation cost and
conformational instability. Within the framework of the
solvation model inherent in the fifth term in eq 1, 18 should
be less stable in solution than 19 because it is difficult to
completely expose one of the aromatic nitrogens to the bulk
H
DOI: 10.1021/ja510636t
J. Am. Chem. Soc. XXXX, XXX, XXX−XXX

Journal of the American Chemical Society
Article
conformational destabilization energy terms for the ligand. The
RMS error of the binding free energies of 18−42 estimated
with ΔGcbonf with respect to the experimental values is only 1.43
kcal/mol, compared to 17.0 kcal/mol in the absence of the two
additional terms. This improvement is consistent with the
results for 5−9 (Table 1). These substantial enhancements in
both correlation and precision confirm the necessity for
including the desolvation and the conformational destabiliza-
tion energy of ligand in calculating the protein−ligand binding
affinity.
Despite the substantial accuracy improvement due to the two
additional energy terms, the scoring function ΔGcbonf remains
inadequate for fully elucidating the relative potencies of 18−42.
For example, the predicted binding free energies of 18, 24, 26,
and 35 become worse with the addition of the conformational
Figure 5. Calculated binding modes of 35 and 38 in the ATP-binding
site of IKKβ. The carbon atoms of IKKβ, 35, and 38 are indicated in
cyan, green, and gray, respectively.
destabilization energy term to ΔGbaq to obtain ΔGbconf
.
Furthermore, when the ΔGcbonf and ΔGebxp values are compared,
it follows immediately that the binding affinities of 24, 26, 35,
37, and 42 are underestimated by more than 2 kcal/mol.
Dynamic Stabilities of the Hydrogen Bonds between
the Pyrimidin-2-lyamine Group and Cys99. Because the
bidentate hydrogen bond between the backbone groups of
Cys99 and the central pyrimidin-2-lyamine moiety appears to
be necessary for the tight binding of potent IKKβ inhibitors,
their relative biochemical potencies may be elucidated on the
basis of the stabilities of the two hydrogen bonds. To examine
the effects of dynamic properties of the bidentate hydrogen
bond on the inhibitory activity, we carried out 10.2 ns MD
simulations of the IKKβ-18, IKKβ-38, and IKKβ-42 complexes
in aqueous solution. These three enzyme−inhibitor complexes
were selected in the comparative MD studies to find the
characteristic dynamic properties of the bidentate hydrogen
bond for 42 with low nanomolar inhibitory activity that could
be discriminated from those for the inhibitors with IC50 values
ranging from 10 to 50 nM such as 18 and 38 (Table 3). This
comparative study seems to be necessary to obtain insight into
designing nanomolar IKKβ inhibitors because the ΔGbconf
scoring function is incapable of distinguishing the inhibitory
activity of 42 from those of 18 and 38 (Table 5).
most potent IKKβ inhibitors listed in Table 3 possess a
sulfonamide moiety at the X position. The multiple hydrogen
bonds involving the sulfonamide moiety are most likely to be
critically important for the inhibition of IKKβ because they
involve the residues that belong to the hinge region (Asp103)
and Gly loop (Leu21), which are the two key functional motifs
for kinase action.
To investigate the effects of including the desolvation and
conformational energy terms in the scoring function on the
accuracy of the de novo design, we plotted the correlation
diagrams between the experimental binding free energies and
those calculated with the ΔGgbas and ΔGbconf functions. As
observed in Figure 6a, we obtained an R2 value of 0.316 for the
comparison of ΔGbexp and ΔGgbas data for 18−42, which
indicates a very weak correlation between the experimental and
calculated results. In contrast, the R2 value increased to 0.759
after changing the variables from ΔGgbas and ΔGbconf (Figure 6b).
This implies that the scoring function can become even more
physically reasonable with the addition of the desolvation and
Figure 7 shows the time evolutions of the interatomic
distances to measure the stabilities of the bidentate hydrogen
bonds between Cys99 and the central pyrimidin-2-ylamine
moiety of the inhibitor in IKKβ-18, IKKβ-38, and IKKβ-42
complexes. We note that the N−H···N hydrogen bond between
amidic nitrogen of Cys99 and the central pyrimidine ring
becomes dynamically more stable with the change of the
inhibitor from 18 and 38 to 42 (Figure 7a). For example, it
appears to be retained for 99.6% of simulation time in IKKβ-42
complex when the distance limit for a hydrogen bond is
assumed to be 2.5 Å,49 as compared to 88.4% and 82.6% of
residence time in IKKβ-18 and IKKβ-38 complexes,
respectively. Similarly, the O···H−N hydrogen bond between
the aminocarbonyl oxygen of Cys99 and the pyrimidin-2-
ylamine moiety of the inhibitors is observed in 99.8% of the
trajectory snapshots of IKKβ-42 complex, whereas it is
preserved for 96.3% and 97.9% of simulation time in IKKβ-
18 and IKKβ-38 complexes (Figure 7b), respectively. It is thus
apparent that the bidentate hydrogen bond between Cys99 and
the inhibitor in IKKβ-42 complex should be dynamically more
stable than those in the IKKβ-18 and IKKβ-38 complexes.
Therefore, the higher biochemical potency of 42 than 18 and
38 can be attributed, at least in part, to the strengthening of the
bidentate hydrogen bonds with Cys99 in the ATP-binding site.
Figure 6. Correlation diagrams for the experimental binding free
energies (ΔGebxp) of 18−42 and STSP versus those calculated with (a)
ΔGgbas and (b) ΔGcbonf including the desolvation and conformational
destabilization energy terms for a ligand.
I
DOI: 10.1021/ja510636t
J. Am. Chem. Soc. XXXX, XXX, XXX−XXX

Journal of the American Chemical Society
Article
chains of Lys44 and Asp166, respectively, at the interface
between the N and C lobes. These hydrogen bonds seem to be
strong because they involve a charged group. The formation of
five additional hydrogen bonds by the two sulfonamide groups
is expected to have an effect of supporting the bidentate
hydrogen bonds between Cys99 and 42 by preventing the
dissociation of the inhibitor from the ATP-binding site. The
high inhibitory activity of 42 may thus be elucidated in terms of
the stability of the bidentate hydrogen bond with Cys99.
The highly negative ΔGbgas value of 42 (Table 5) actually
stems from the formation of a total of seven stable hydrogen
bonds in the ATP-binding site of IKKβ. However, these
multiple hydrogen bonds may be insufficient to compensate for
the high desolvation cost for binding to IKKβ (Table 5) due to
the presence of two highly polar sulfonamide groups. 42 can
thus be a nanomolar-level inhibitor by successfully overcoming
the high desolvation cost with the two additional potency-
enhancing factors including the low conformational destabiliza-
tion upon binding in the ATP-binding site and the increased
dynamic stability of the bidentate hydrogen bonds with Cys99.
CONCLUSIONS
■
We have identified new potent IKKβ inhibitors by means of the
structure-based virtual screening of a large chemical library and
subsequent de novo design using the PPA as the molecular core.
To improve the efficiency of this computer-aided drug design,
we improved the scoring function by introducing proper
desolvation energy and conformational destabilization energy
terms for ligands. As a consequence of these modifications, the
discovery of new IKKβ inhibitors was successful to the extent
that more than the half of the designed PPA derivatives had
submicromolar inhibitory activities. In particular, the derivative
with sulfonamide moieties on both terminal phenyl rings (42)
was anticipated to serve as a good lead compound for the
discovery of new small-molecule medicines because of its low
nanomolar-level inhibitory activity. We found that the
biochemical potency could be optimized by lowering the
desolvation cost and the conformational destabilization for the
inhibitor required for binding to IKKβ as well as by reinforcing
the interactions in the ATP-binding site. Consistent with the
previous experimental findings, the establishment of the
hydrogen bonds with the backbone amide groups of Cys99
in the hinge region proved to be indispensable for high
inhibitory activity against IKKβ. The MD simulation studies of
IKKβ−inhibitor complexes showed that the dynamic stabilities
of the bidentate hydrogen bonds between Cys99 and the
inhibitor would serve as the yardstick for discriminating the
submicromolar and low nanomolar IKKβ inhibitors. Therefore,
to identify the highly potent IKKβ inhibitors with virtual
screening and de novo design, the scoring function should
include the energy term to measure the dynamic stability of the
bidentate hydrogen bonds with Cys99 in addition to those to
describe the strength of interactions in the ATP-binding site,
the ligand desolvation cost, and the conformational destabiliza-
tion of the inhibitor upon binding to IKKβ.
Figure 7. Time evolutions of the interatomic distances associated with
(a) N−H···N and (b) O···H−N hydrogen bonds established between
the backbone amide groups of Cys99 and the central pyrimidin-2-
ylamine moiety of 18, 38, and 42.
This result exemplifies the need to consider the detailed
dynamical aspects of enzyme−inhibitor interactions to precisely
estimate the inhibitory activity of the highly potent inhibitors.
Because 42 possesses two sulfonamide moieties at both X
and Y positions in the PPA core, they are expected to
contribute significantly to the high dynamic stability of the
bidentate hydrogen bonds in the IKKβ-42 complex. Figure 8
Figure 8. A representative MD trajectory snapshot of IKKβ-42
complex. The carbon atoms of IKKβ and 42 are indicated in cyan and
green, respectively. Hydrogen bonds are indicated with dotted lines.
Solvent molecules are omitted for clarity.
displays a representative MD trajectory snapshot for the IKKβ-
42 complex. The interactions of 42 in the ATP-binding site of
IKKβ are similar to those of 35 and 38 (Figure 5) in that the
sulfonamide group at the X position forms three hydrogen
bonds with the side chain of Lys106 and the backbone groups
of Leu21 and Asp103 at the interface between the hinge region
and the Gly loop. The second sulfonamide moiety at the Y
position also establishes multiple hydrogen bonds with the side
ASSOCIATED CONTENT
* Supporting Information
Computational details, experimental procedures, selectivity
■
S
profiles of 26 and 35 with respect to kinase inhibitions,
characterization data, and copies of H and 13C NMR spectra
1
for all compounds. This material is available free of charge via
the Internet at http://pubs.acs.org.
J
DOI: 10.1021/ja510636t
J. Am. Chem. Soc. XXXX, XXX, XXX−XXX

Journal of the American Chemical Society
Article
(20) Frelin, C.; Imbert, V.; Griessinger, E.; Loubat, A.; Dreano, M.;
Peyron, J. F. Oncogene 2003, 22, 8187.
AUTHOR INFORMATION
■
Corresponding Authors
hspark@sejong.ac.kr
hongorg@kaist.ac.kr
(21) Burke, J. R.; Pattoli, M. A.; Gregor, K. R.; Brassil, P. J.;
MacMaster, J. F.; Mclntyre, K. W.; Yang, X.; Iotzova, V. S.; Clarke, W.;
Strnad, J.; Qiu, Y.; Zusi, F. C. J. Biol. Chem. 2003, 278, 1450.
(22) Hu, H.; Snyder, J. P. J. Chem. Inf. Model. 2012, 52, 3190.
(23) Lauria, A.; Ippolito, M.; Fazzari, M.; Tutone, M.; Blasi, F. D.;
Mingoia, F.; Almerico, A. M. J. Mol. Graph. Model. 2010, 29, 72.
(24) Noha, S. M.; Atanasov, A. G.; Schuster, D.; Markt, P.;
Fakhrudin, N.; Heiss, E. H.; Schrammel, O.; Rollinger, J. M.; Stuppner,
H.; Dirsch, V. M.; Wolber, G. Bioorg. Med. Chem. Lett. 2011, 21, 577.
(25) Nagarajan, S.; Choo, H.; Cho, Y. S.; Oh, K. S.; Lee, B. H.; Shin,
K. J.; Pae, A. N. Bioorg. Med. Chem. 2010, 18, 3951.
(26) Xu, G.; Lo, Y.-C.; Li, Q.; Napolitano, G.; Wu, X.; Jiang, X.;
Dreano, M.; Karin, M.; Wu, H. Nature 2011, 472, 325.
(27) Liu, S.; Misquitta, Y. R.; Olland, A.; Johnson, M. A.; Kelleher, K.
S.; Kriz, R.; Lin, L. L.; Stahl, M.; Mosyak, L. J. Biol. Chem. 2013, 288,
22758.
Notes
The authors declare no competing financial interest.
ACKNOWLEDGMENTS
■
This research was supported by National Research Foundation
of Korea (NRF) funded by the Ministry of Education, Science
and Technology (NRF-2011-0022858) and the Institute for
Basic Science (IBS-R010-G1).
REFERENCES
■
(1) Nakanishi, C.; Toi, M. Nat. Rev. Cancer 2005, 5, 297.
(2) Perkins, N. D. Nat. Rev. Mol. Cell Biol. 2007, 8, 49.
(3) Karin, M.; Yamamoto, Y.; Wang, Q. M. Nat. Rev. Drug Discovery
2004, 3, 17.
(4) Yamamoto, Y.; Gaynor, R. B. Trends Biochem. Sci. 2004, 29, 72.
(5) Li, J.; Peet, G. W.; Pullen, S. S.; Schembri-King, J.; Warren, T. C.;
Marcu, K. B.; Kehry, M. R.; Barton, R.; Jakes, S. J. Biol. Chem. 1998,
273, 30736.
(6) Hu, Y.; Baud, V.; Delhase, M.; Zhang, P.; Deerinck, T.; Ellisman,
M.; Johnson, R.; Karin, M. Science 1999, 284, 316.
(7) Bamborough, P.; Callahan, J. F.; Christopher, J. A.; Kerns, J. K.;
Liddle, J.; Miller, D. D.; Morse, M. A.; Rumsey, W. L.; Williamson, R.
Curr. Top. Med. Chem. 2009, 9, 623.
(28) Huang, S. Y.; Grinter, S. Z.; Zou, X. Phys. Chem. Chem. Phys.
2010, 12, 12899.
(29) Warren, G. L.; Andrews, C. W.; Capelli, A. M.; Clarke, B.;
LaLonde, J.; Lambert, M. H.; Lindvall, M.; Nevins, N.; Semus, S. F.;
Senger, S.; Tedesco, G.; Wall, I. D.; Woolven, J. M.; Peishoff, C. E.;
Head, M. S. J. Med. Chem. 2006, 49, 5912.
(30) Shoichet, B. K.; Leach, A. R.; Kuntz, I. D. Proteins 1999, 34, 4.
(31) Viji, S. N.; Prasad, P. A.; Gautham, N. J. Chem. Inf. Model. 2009,
49, 2687.
(32) Morris, G. M.; Goodsell, D. S.; Halliday, R. S.; Huey, R.; Hart,
W. E.; Belew, R. K.; Olson, A. J. J. Comput. Chem. 1998, 19, 1639.
(33) Park, H.; Lee, J.; Lee, S. Proteins 2006, 65, 549.
(8) Suzuki, J.; Ogawa, M.; Muto, S.; Itai, A.; Isobe, M.; Hirata, Y.;
Nagai, R. Expert Opin. Investig. Drugs 2011, 20, 395.
(9) Colotta, F.; Allavena, P.; Sica, A.; Garlanda, C.; Mantovan, A.
Carcinogenesis 2009, 30, 1073.
(34) Mehler, E. L.; Solmajer, T. Protein Eng. 1991, 4, 903.
(35) Stouten, P. F. W.; Frommel, C.; Nakamura, H.; Sander, C. Mol.
̈
Simul. 1993, 10, 97.
(36) Park, H. J. Comput.-Aided Mol. Des. 2014, 28, 175.
(37) Wang, R.; Gao, Y.; Lai, L. J. Mol. Model. 2000, 6, 498.
(38) Wang, R.; Liu, L.; Lai, L. J. Mol. Model. 1998, 6, 379.
(39) Liu, T.; Zhan, W.; Wang, Y.; Zhang, L.; Yang, B.; Dong, X.; Hu,
Y. Eur. J. Med. Chem. 2014, 73, 167.
(40) Viji, S. N.; Prasad, P. A.; Gautham, N. J. Chem. Inf. Model. 2009,
49, 2687.
(10) Liddle, J.; Bamborough, P.; Barker, M. D.; Campos, S.; Chung,
C. W.; Cousins, R. P.; Faulder, P.; Heathcote, M. L.; Hobbs, H.;
Holmes, D. S.; Ioannou, C.; Ramirez-Molina, C.; Morse, M. A.;
Osborn, R.; Payne, J. J.; Pritchard, J. M.; Rumsey, W. L.; Tape, D. T.;
Vicentini, G.; Whitworth, C.; Williamson, R. A. Bioorg. Med. Chem.
Lett. 2012, 22, 5222.
(11) Christopher, J. A.; Bamborough, P.; Alder, C.; Campbell, A.;
Cutler, G. J.; Down, K.; Hamadi, A. M.; Jolly, A. M.; Kerns, J. K.;
Lucas, F. S.; Mellor, G. W.; Miller, D. D.; Morse, M. A.; Pancholi, K.
D.; Rumsey, W.; Solanke, Y. E.; Williamson, R. J. Med. Chem. 2009, 52,
3098.
(12) Qiu, X.; Du, Y.; Lou, B.; Zuo, Y.; Shao, W.; Huo, Y.; Huang, J.;
Yu, Y.; Zhou, B.; Du, J.; Fu, H.; Bu, X. J. Med. Chem. 2010, 53, 8260.
(13) Lorenzo, P.; Alvarez, R.; Ortiz, M. A.; Alvarez, S.; Piedrafita, F.
J.; de Lera, A. R. J. Med. Chem. 2008, 51, 5431.
(14) Crombie, A. L.; Sum, F. W.; Powell, D. W.; Hopper, D. W.;
Torres, N.; Berger, D. M.; Zhang, Y.; Gavriil, M.; Sadler, T. M.; Arndt,
K. Bioorg. Med. Chem. Lett. 2010, 20, 3821.
(15) Bingham, A. H.; Davenport, R. J.; Fosbeary, R.; Gowers, L.;
Knight, R. L.; Lowe, C.; Owen, D. A.; Parry, D. M.; Pitt, W. R. Bioorg.
Med. Chem. Lett. 2008, 18, 3622.
(16) Waelchli, R.; Bollbuck, B.; Bruns, C.; Buhl, T.; Eder, J.; Feifel,
R.; Hersperger, R.; Janser, P.; Revesz, L.; Zerwes, H. G.; Schlapbach, A.
Bioorg. Med. Chem. Lett. 2006, 16, 108.
(17) Palanki, M. S. S.; Erdman, P. E.; Ren, M.; Suto, M.; Bennett, B.
L.; Manning, A.; Ransone, L.; Spooner, C.; Desai, S.; Ow, A.; Totsuka,
R.; Tsao, P.; Toriumi, W. Bioorg. Med. Chem. Lett. 2003, 13, 4077.
(18) Kishore, N.; Sommers, C.; Mathialagan, S.; Guzova, J.; Yao, M.;
Hauser, S.; Huynh, K.; Bonar, S.; Mielke, C.; Albee, L.; Weier, R.;
Graneto, M.; Hanau, C.; Perry, T.; Tripp, C. S. J. Biol. Chem. 2003,
278, 32861.
(19) Ziegelbauerm, K.; Gantner, F.; Lukacs, N. W.; Berlin, A.;
Fuchikami, K.; Niki, T.; Sakai, K.; Inbe, H.; Takeshita, K.; Ishimori,
M.; Komura, H.; Murata, T.; Lowinger, T.; Bacon, K. B. Br. J.
Pharmacol. 2005, 145, 178.
(41) Kontoyianni, M.; McClellan, L. M.; Sokol, G. S. J. Med. Chem.
2004, 47, 558.
(42) Palmer, W. S.; Alam, M.; Arzeno, H. B.; Chang, K.-C.; Dunn, J.
P.; Goldstein, D. M.; Gong, L.; Goyal, B.; Hermann, J. C.; Hogg, J. H.;
Hsieh, G.; Jahangir, A.; Janson, C.; Jin, S.; Kammlott, R. U.;
Kuglstatter, A.; Lukacs, C.; Michoud, C.; Niu, L.; Reuter, D. C.;
Shao, A.; Silva, T.; Trejo-Martin, T. A.; Stein, K.; Tan, Y.-C.;
Tivitmahaisoon, P.; Tran, P.; Wagner, P.; Weller, P.; Wu, S.-Y. Bioorg.
Med. Chem. Lett. 2013, 23, 1486.
(43) Gong, L.; Han, X.; Silva, T.; Tan, Y.-C.; Goyal, B.;
Tivitmahaisoon, P.; Trejo, A.; Palmer, W.; Hogg, H.; Jahagir, A.;
Alam, M.; Wagner, P.; Stein, K.; Filonova, L.; Loe, B.; Makra, F.;
Rotstein, D.; Rapatova, L.; Dunn, J.; Zuo, F.; Porto, J. D.; Wong, B.;
Jin, S.; Chang, A.; Tran, P.; Hsieh, G.; Niu, L.; Shao, A.; Reuter, D.;
Hermann, J.; Kuglstatter, A.; Goldstein, D. Bioorg. Med. Chem. Lett.
2013, 23, 3565.
(44) Deininger, M.; Buchdunger, E.; Druker, B. J. Blood 2005, 105,
2640.
(45) Kerns, E. H.; Di, L. Drug Discovery Today 2003, 8, 316.
(46) Innocenti, P.; Cheung, K.-M. J.; Solanki, S.; Mas-Droux, C.;
Rowan, F.; Yeoh, S.; Boxall, K.; Westlake, M.; Pickard, L.; Hardy, T.;
Baxter, J. E.; Aherne, G. W.; Bayliss, R.; Fry, A. M.; Hoelder, S. J. Med.
Chem. 2012, 55, 3228.
(47) Ginman, T.; Viklund, J.; Malmstrom, J.; Blid, J.; Emond, R.;
̈
Forsblom, R.; Johansson, A.; Kers, A.; Lake, F.; Sehgelmeble, F.;
Sterky, K. J.; Bergh, M.; Lindgren, A.; Johansson, P.; Jeppsson, F.;
Falting, J.; Gravenfors, Y.; Rahm, F. J. Med. Chem. 2013, 56, 4181.
̈
K
DOI: 10.1021/ja510636t
J. Am. Chem. Soc. XXXX, XXX, XXX−XXX

Journal of the American Chemical Society
Article
́
(48) Deng, X.; Okram, B.; Ding, Q.; Zhang, J.; Choi, Y.; Adrian, F. J.;
Wojciechowski, A.; Zhang, G.; Che, J.; Bursulaya, B.; Cowan-Jacob, S.
W.; Rummel, G.; Sim, T.; Gray, N. S. J. Med. Chem. 2010, 53, 6934.
(49) Jeffrey, G. A. An Introduction to Hydrogen Bonding; Oxford
University Press: Oxford, 1997.
L
DOI: 10.1021/ja510636t
J. Am. Chem. Soc. XXXX, XXX, XXX−XXX