Fragment based discovery of arginine isosteres through REPLACE: Towards non-ATP competitive CDK inhibitors

Article abstract of DOI:10.1016/j.bmc.2013.10.039

In order to develop non-ATP competitive CDK2/cyclin A inhibitors, the REPLACE strategy has been applied to generate fragment alternatives for the N-terminal tetrapeptide of the cyclin binding motif (HAKRRLIF) involved in substrate recruitment prior to phosphotransfer. The docking approach used for the prediction of small molecule mimics for peptide determinants was validated through reproduction of experimental binding modes of known inhibitors and provides useful information for evaluating binding to protein-protein interaction sites. Further to this, potential arginine isosteres predicted using the validated LigandFit docking method were ligated to the truncated C-terminal peptide, RLIF using solid phase synthesis and evaluated in a competitive binding assay. After testing, identified fragments were shown to represent not only appropriate mimics for a critical arginine residue but also to interact effectively with a minor hydrophobic pocket present in the binding groove. Further evaluation of binding modes was undertaken to optimize the potency of these compounds. Through further application of the REPLACE strategy in this study, peptide-small molecule hybrid CDK2 inhibitors were identified that are more drug-like and suitable for further optimization as anti-tumor therapeutics.

Full text of DOI:10.1016/j.bmc.2013.10.039

Bioorganic & Medicinal Chemistry 22 (2014) 616–622
Contents lists available at ScienceDirect
Bioorganic & Medicinal Chemistry
journal homepage: www.elsevier.com/locate/bmc
Fragment based discovery of arginine isosteres through REPLACE:
Towards non-ATP competitive CDK inhibitors
Padmavathy Nandha Premnath , Shu Liu , Tracy Perkins, Jennifer Abbott, Erin Anderson,
⇑
Campbell McInnes
Drug Discovery and Biomedical Sciences, South Carolina College of Pharmacy, University of South Carolina, Columbia, SC 29208, United States
a r t i c l e i n f o
a b s t r a c t
Article history:
In order to develop non-ATP competitive CDK2/cyclin A inhibitors, the REPLACE strategy has been applied
to generate fragment alternatives for the N-terminal tetrapeptide of the cyclin binding motif (HAKRRLIF)
involved in substrate recruitment prior to phosphotransfer. The docking approach used for the prediction
of small molecule mimics for peptide determinants was validated through reproduction of experimental
binding modes of known inhibitors and provides useful information for evaluating binding to protein–
protein interaction sites. Further to this, potential arginine isosteres predicted using the validated Ligand-
Fit docking method were ligated to the truncated C-terminal peptide, RLIF using solid phase synthesis
and evaluated in a competitive binding assay. After testing, identified fragments were shown to represent
not only appropriate mimics for a critical arginine residue but also to interact effectively with a minor
hydrophobic pocket present in the binding groove. Further evaluation of binding modes was undertaken
to optimize the potency of these compounds. Through further application of the REPLACE strategy in this
study, peptide-small molecule hybrid CDK2 inhibitors were identified that are more drug-like and suit-
able for further optimization as anti-tumor therapeutics.
Received 22 August 2013
Revised 16 October 2013
Accepted 24 October 2013
Available online 7 November 2013
Keywords:
Cyclin dependent kinase
Drug discovery
Inhibitor
Anti-tumor therapeutic
REPLACE
Ó 2013 Elsevier Ltd. All rights reserved.
1. Introduction
to at least a 10 fold loss in binding.8,9 In this present study, the RE-
PLACE (Replacement with Partial Ligand Alternatives through
CDKs associate with cyclins to regulate the cell cycle check-
points and control cell proliferation.1 CDK2/cyclin A (CDK2A) con-
trols DNA replication through phosphorylation of the transcription
factor E2F-1, the activity of which is often deregulated in tumor
cells. Inhibition of CDK2A has been shown to selectively induce
apoptosis of cancer cells through the E2F-1 pathway and therefore
is an attractive target for controlling abnormal cell proliferation.2,3
Currently, available CDK inhibitors primarily target the highly con-
served ATP binding site and generally inhibit both cell cycle and
transcriptional CDKs potentially leading to toxicities in normal
cells.3,4 In our present study we utilize an alternative approach to
selectively inhibit cell cycle CDKs by targeting protein–protein
interactions distinct from the ATP binding pocket. CDK complexes
recruit substrates and endogenous inhibitory proteins through the
cyclin binding groove (CBG) only in the cell cycle CDK context
(CDK2/Cyclin A, E; CDK4/cyclin D).5–7 The CBG is recognized by a
conserved cyclin binding motif (CBM), has been truncated and
optimized to potent octapeptides including HAKRRLIF,8 and further
minimized to small peptides retaining low micromolar binding
affinity.8,9 Arg4 of the 8mer is particularly important for activity
since modification to even the uncharged isostere, citrulline leads
Computational Enrichment) strategy has been applied to identify
fragment based alternatives for the N-terminus of CBG-peptides
and suitable mimetics for the critical arginine in order to convert
the octamer to a less peptidic inhibitor.10,11 Validation of the
LigandFit docking method12 was carried out as a prelude to compu-
tationally evaluating fragment alternatives. Predicted N-terminal
capping groups were then incorporated as Fragment Ligated Inhib-
itory Peptides (FLIPs) through solid phase synthesis and after
in vitro evaluation, furoic, phenyl acetic and picolinic acid derived
groups were shown to inhibit binding to CDK2/cyclin A while
improving the drug likeness. These compounds represent the basis
for further optimization of cell cycle CDK inhibitors as preclinical
candidates for cancer therapy.
2. Materials and methods
2.1. Computational chemistry
The parameters of the LigandFit (Discovery Studio 3.0, Accelrys)
docking method were validated using ligands from cyclin A/CDK2
crystal structures. The crystallographic ligands 1-(3,5-dichloro-
phenyl)-5-methyl-1H-1,2,4-triazole-3-carbaldehyde
(3,5-DCPT)
(PDB ID: 2UUE) and 1-(4-chlorophenyl)-5-methyl-1H-1,2,4-tria-
zole-3-carbaldehyde (4-CPT) (PDB ID: 2V22) were used as positive
⇑
Corresponding author. Tel.: +1 (803) 576 5684.
E-mail address: mcinnes@sccp.sc.edu (C. McInnes).
These authors made equal contributions to this manuscript.
0968-0896/$ - see front matter Ó 2013 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.bmc.2013.10.039

P. N. Premnath et al. / Bioorg. Med. Chem. 22 (2014) 616–622
617
controls and 5-chloro-2-phenyl-1,8a-dihydroimidazo[1,2-a]pyri-
dine-3-carbaldehyde was evaluated as a negative control. The
three ligands were docked successively into the cyclin grooves of
two structures (2V22, 2UUE) and 20 poses were generated for each.
This was repeated by variation of the LigandFit parameters includ-
ing the forcefield used for the energy grid (Dreiding, CFF and PLP1),
use of minimization sphere (on or off) and different scoring func-
tions (Ligscore1_Dreiding, Ligscore2_Dreiding, PLP1, PLP2, PMF,
DOCKSCORE) to determine which generated a calculated binding
energy most predictive of the experimental binding mode. For each
parameter and scoring function, the number of correct poses of the
positive controls in the top 25 ranked binding modes (out of 60
possible, 20 for each of the three ligands) was determined. A library
of 20 potential fragment alternatives was manually built using
ChemDraw for Excel (Perkin Elmer) and subsequently imported
into DiscoveryStudio 3.0 (Accelrys). For docking of unknown com-
pounds, 10 poses were generated since this was sufficient to gen-
erate correct poses for the control ligands.
(q, J = 6.75 Hz, 2H), 3.50 (d, J = 7.26 Hz, 2H), 1.31 (t, J = 6.87 Hz,
3H), 1.16 (t, J = 7.02 Hz, 3H).
EI 208.
2.2.1.2. Propyl 2-(3-propoxyphenyl)acetate (1b).
Reddish oil
(yield-28.2%) 1H NMR (Chloroform-D, 300 MHz) d (ppm) 7.25–7.19
(m, 1H), 6.83 (t, J = 8.07 Hz, 3H), 4.05 (t, J = 6.78 Hz, 2H), 3.91
(t, J = 6.39 Hz, 2H), 3.59 (d, J = 6.96 Hz, 2H), 1.80 (q, J = 7.35 Hz,
2H), 1.64 (q, J = 6.00 Hz, 2H), 1.03 (t, J = 7.32 Hz, 3H), 0.91
(t, J = 7.33 Hz, 3H).
EI 224.
2.2.1.3. Isobutyl 2-(3-isobutoxyphenyl)acetate (1c).
Reddish
oil (yield-24.2%) 1H NMR (Chloroform-D, 300 MHz) d (ppm) 7.22 (t,
J = 8.34 Hz, 1H), 6.85–6.79 (m, 3H), 3.89 (d, J = 6.48 Hz, 2H), 3.71 (d,
J = 6.93 Hz, 2H), 3.60 (s, 2H), 2.12–2.03 (m, 1H), 1.97–1.88 (m, 1H),
1.03 (d, J = 6.18 Hz, 6H), 0.92 (d, J = 6.60 Hz, 6H).
EI 264.
2.2.1.4. Step 2: 2-(3-Ethoxyphenyl) acetic acid (2a)13
.
The
2.2. Chemistry
ethyl 2-(3-ethoxyphenyl) acetate (1.41 g, 5 mmol) obtained from
step 1 was treated with sodium hydroxide (2.8 g, 70 mmol) and a
solution of 15 mL ethanol and 15 mL water. The reaction mixture
was refluxed for 2 h and the reaction was monitored by TLC (eth-
ylacetate:hexanes = 35:65). After the completion of reaction, the
reaction mixture was cooled, the alcohol was evaporated and
diluted with water. The reaction mixture was acidified with 1 N
hydrochloric acid and stirred to precipitate the product as white
solid (0.73 g, yield-81.2%).
All the starting materials, solvents and reagents were used as
obtained without further purification. Analytical thin layer chro-
matography was performed on silica gel (GF-254 plates). 1H and
13C NMR spectra were recorded with a Varian Mercury 300 and
400 Spectrometer, respectively. Mass spectra were measured with
a Micromass QTOF (Tandem quadruple-time of flight mass spec-
trometer), electrospray ionization (ESI) and VG 70S (Double-focus-
ing magnetic sector mass spectrometer, EI). Analytical purities of
evaluated compounds were >95% unless stated otherwise. The
following analytical method (unless stated otherwise) was used
on a Waters Alliance 2695 HPLC with a 2996 diode-array detector
1H NMR (chloroform-D, 500 MHz) d (ppm) 7.28–7.24 (m, 1H),
6.88–6.83 (m, 3H), 4.05 (q, J = 6.80 Hz, 2H), 3.63 (s, 2H), 1.42 (t,
J = 7.25 Hz, 3H).
13C NMR (chloroform-D, 125 MHz) d (ppm) 177.15, 159.14,
134.61, 129.60, 121.55, 115.65, 113.41, 63.41, 41.01, 14.80.
EI 180.
and equipped with a C18 (2) 100 A, 250 ꢀ 4.6 mm, 5
lm column
(Phenomenox Luna). A gradient from 100% water (0.1% trifluoro-
acetic acid) to 60% acetonitrile (0.1% trifluoroacetic acid) was run
over 30 min and held for 4 min. The chromatograms were
extracted at 226 and 254 nm.
2.2.1.5. 2-(3-Propoxyphenyl)acetic acid (2b).
White solid
(yield-85.2%) 1H NMR (chloroform-D, 300 MHz) d (ppm) 7.21(d,
J = 7.44 Hz, 2H), 6.83 (t, J = 7.86 Hz, 2H), 3.90 (t, J = 6.03 Hz, 2H),
3.61 (s, 2H), 1.79 (q, J = 6.42 Hz, 2H), 1.02 (t, J = 7.05 Hz, 3H).
EI 194.
2.2.1. Synthesis of capping groups
Furoic acid, picolinic acid and 2-(3,4-dihydroxyphenyl)acetic
acid N-terminal capping groups were all obtained commercially
(Chembridge, Matrix Scientific and UKROrgSyth). 3-Alkoxy phenyl
acetic acid derivatives were synthesized using the following
procedure.
2.2.1.6. 2-(3-Isobutoxyphenyl)acetic acid (2c).
White solid
(yield-82.8%) 1H NMR (chloroform-D, 300 MHz) d (ppm) 7.22 (d,
J = 8.19 Hz, 1H), 6.83 (t, J = 8.70 Hz, 3H), 3.70 (d, J = 7.26 Hz, 2H),
3.62 (s, 2H), 2.11–2.00 (m, 1H), 1.01 (d, J = 6.69 Hz, 6H).
EI 208.
O
O
O
R
Br
NaOH
EtOH-H2O
Reflux 100°C
R
R
R
K2CO3
HO
OH
O
O
O
OH
DMF
Reflux 150°C
18hrs
2a-c
1a-c
R=Et, Pr, Ibu
2hrs
R=Et, Pr, Ibu
2.2.1.1. Step 1: Synthesis of ethyl 2-(3-ethoxyphenyl) acetate
(1a)13
3-hydroxyphenylacetic acid (0.76 g, 5 mmol), ethyl
2.2.2. Peptide and FLIP synthesis
.
Unless otherwise indicated, capping groups were obtained
commercially from ChemBridge and UORSYS and were used as
provided. Peptides were assembled by using standard solid-phase
synthesis methods.10 A sample procedure is given as follows:
5 equiv of the C-terminal amino acid were coupled to Rink resin
at the first place using DIEA (0.082 mL) and HBTU (189.6 mg) in
5 mL of DMF for 1 h. The Fmoc of the C-terminal amino acid was
removed using 20% piperidine in 5 mL of DMF for 10 min before
addition of 5 equiv of the next amino acid using DIEA (0.082 mL)
and HBTU (189.6 mg) in 5 mL of DMF. Wash cycles (5 ꢀ 10 mL
of DMF + 5 ꢀ 10 mL of DCM) were applied to each step in
bromide (0.25 g, 23 mmol) and potassium carbonate (0.13 g,
16.5 mmol) were refluxed in 5 mL of DMF for 12–18 h. The reaction
was monitored by TLC (ethylacetate:hexanes = 35:65), after the
reaction was complete, the reaction mixture was cooled to room
temperature; water was added to dissolve potassium carbonate.
The aqueous layer was extracted with ethyl acetate and the com-
bined organic layers were washed with saturated sodium bicar-
bonate, brine, dried over sodium sulfate and evaporated to yield
a reddish oil (0.31 g, yield-30.0%).
1H NMR (Chloroform-D, 300 MHz) d (ppm) 7.13 (t, J = 7.32 Hz,
1H), 6.75 (t, J = 8.73 Hz, 3H), 4.06 (q, J = 7.80 Hz, 2H), 3.93

618
P. N. Premnath et al. / Bioorg. Med. Chem. 22 (2014) 616–622
between coupling and deprotection of Fmoc. Upon completion of
assembly, side chain protecting groups were removed, and
FLIPs were cleaved from Rink resin using 90:5:5 mixtures of
TFA/H2O/TIPS. Crude FLIPs were purified using reverse-phase flash
chromatography and semi preparative reverse-phase HPLC meth-
ods. Pure compounds were lyophilized and characterized using
mass spectrometry and analytical HPLC (see Supplementary
Table 1).
in vitro activity. In addition, docking methods generally are specific
for a particular binding site and therefore require judicious selec-
tion. Furthermore, protein–protein interactions are known to be
difficult to study through computational analysis and therefore
validation exercises would provide insights into the best force-
fields and scoring functions for these. To achieve this, the N-termi-
nal capping groups (Ncaps) from the crystal structures of two FLIPs
bound to CDK2A (1-(3,5-dichlorophenyl)-5-methyl-1H-1,2,4-tria-
zole-3-carboxamide-Arg-Leu-Ile-4-fluoroPhe (3,5-DCPT) (PDB
ID: 2UUE) and 1-(4-chlorophenyl)-5-methyl-1H-1,2,4-triazole-3-
carboxamide-Arg-Leu-Ile-4-fluoroPhe) (4-CPT) (PDB ID: 2V22) were
used as positive controls while an inactive compound 5-chloro-2-
phenyl-1,8a-dihydroimidazo[1,2-a]pyridine-3-carbaldehyde was
evaluated as negative control.10
2.3. Fluorescent polarization binding assay
2.3.1. CDK4D1 FP assay
This assay was performed using black 384-well plates using a
previously described procedure14 with the following modifications.
To each well were added: 5
recombinant human kinase complex from Invitrogen),
compound solution, 30 nM fluoresceinyl-Ahx-Pro-Val-
l
l CDK4D1 (0.3
l
g/well purified
The Ncap positive controls (3,5-DCPT and 4-CPT) when bound
to the CBG make important hydrogen bonding interactions with
residues Gln254 and Trp217.10 To determine if the docking method
used (LigandFit, DiscoveryStudio) is predictive in terms of
reproducing experimental binding modes, the docked poses of
the 3,5-DCPT and 4-CPT with 2UUE were examined for similar ori-
entation of the aromatic and triazole rings and also retention of the
two hydrogen bonding interactions. Reproducibility of the docking
method was assessed through examining the number of correct
poses (i.e., those superimposable with the crystal structure) or
close poses (poses that show a RMSD value 62.0 Å) with the crystal
structure) of the positive control ligands generated and the
number of negative control poses in top the 25 ranked conforma-
tions (20 poses generated for each ligand, 60 in total). This was
repeated for both cyclin subunits (B and D) observed in the asym-
metric unit of the crystal structure. The result for subunit B showed
10 and 14 correct/close poses of 3,5-DCPT and 4-CPT whereas
those for subunit D showed only 11 close poses of 4-CPT and no
correct/close poses for 3,5-DCPT. There were no poses of the
negative control in the top 25 poses for either subunit. Due to
the better data obtained for subunit B, this was considered for fur-
ther parameter optimization. For each energy grid available in
LigandFit, the number of top ranked correct/close poses was
calculated for each scoring function (Ligscore1_Dreiding, Lig-
score2_Dreiding, PLP1, PLP2, Jain, PMF and DOCKSCORE). The
Dreiding, CFF and PLP1 energy grids generated 2, 4 and 6 correct
poses (in the top 25 ranked) respectively for 3,5-DCPT for the best
scoring function. There were 18 and 15 poses of the negative con-
trol ligands generated using Dreiding and CFF grids, respectively,
however none were observed for the PLP1 method in top 25 ranked
compounds and therefore indicating this as the most successful for
this validation (Table 1). Further analysis of the docked poses re-
vealed that the most effective scoring function was determined
to be PLP1 since it resulted in 9 and 6 correct or close poses for
3,5-DCPT and 4-CPT, respectively. The optimized docking protocol
was therefore found to include use of the PLP1 energy grid, and the
PLP1 scoring function with 10 poses generated for each structure.
These results demonstrated full validation of the docking approach
and strongly suggest that results obtained with unknown ligands
will be predictive.
5 ll
5 ll
Lys-Arg-Arg-Leu-(3ClPhe)-Gly tracer peptide. Compounds and ki-
nase complexes were diluted using assay buffer (25 mM HEPES
pH 7, 10 nM NaCl, 0.01% Nonidet P-40, 1 mM dithiothretiol). The
plate was centrifuged for 1 min at 500 rpm and then incubated
with shaking for 45 min at room temperature. Fluorescence polar-
ization was measured on a DTX880 multimode detector (Beckman
Coulter, Brea, CA) fitted with 485/535 nm excitation/emission
filters and a dichroic mirror suitable for fluorescein. Relative mp
was calculated for each concentration tested using the equation
showing below. IC50 values were determined by logarithmic
regression by correlating mean polarizations (mp) and testing
concentrations.
Relative mean polarizationðRMPÞ
mPðcompoundÞ ꢁ mPðDMSO; protein; tracerÞ
¼
mPðDMSO; proteinÞ ꢁ mPðDMSO; protein; tracerÞ
2.3.2. CDK2A2 FP assay
This assay was performed using black 384-well plates. To each
well were added: 5
ll CDK2A2 (0.3
lg/well purified recombinant
human kinase complex), 5
l
l compound solution, 5
ll 30 nM fluo-
resceinyl-Ahx-Pro-Val-Lys-Arg-Arg-Leu-Phe-Gly tracer peptide.
Compounds and kinase complexes were diluted using assay buffer
(25 mM HEPES pH 7, 10 nM NaCl, 0.01% Nonidet P-40, 1 mM dithi-
othretiol). The plate was centrifuged for 1 min at 500 rpm and then
incubated with shaking for 45 min at room temperature. Fluores-
cence polarization was read on a DTX880 multimode detector
(Beckman Coulter, Brea, CA) fitted with 485/535 nm excitation/
emission filters and a dichroic mirror suitable for fluorescein.
Relative mp was calculated for each concentration tested using
the equation showing below. IC50 values were determined by
logarithmic regression by correlating mean polarization (mp) and
testing concentrations.
Relative mean polarizationðRMPÞ
A library of compounds based on the native ligand (1-(3,5-
dichlorophenyl)-5-methyl-1H-1,2,4-triazole-3-carbonyl and the
parent peptide (HAKRRLIF) was designed to include pharmaco-
phoric features observed in this series. A library of 20 compounds
was built on core scaffolds including phenylacetic acids, five-mem-
bered heterocyclic analogs, and also picolinic acid derivatives. These
fragments were designed to interact with Trp217 and Gln254
through H-bonds, the secondary hydrophobic pocket (alkyl/aryl
group for van der Waals interaction), and the acidic region (hetero-
cyclic and/or amine groups) through ion-pairing interactions with
Glu224/Glu220. Phenyl acetic acid derivatives were designed with
alkoxy groups at position 3 and/or 4 to make van der Waals
mPðcompoundÞ ꢁ mPðDMSO; protein; tracerÞ
¼
mPðDMSO; proteinÞ ꢁ mPðDMSO; protein; tracerÞ
3. Results
3.1. Validation of REPLACE docking method
Validation of the method used for computational aspects of RE-
PLACE was undertaken to demonstrate that docking results ob-
tained for unknown compounds would be predictive of their

P. N. Premnath et al. / Bioorg. Med. Chem. 22 (2014) 616–622
619
Table 1
Optimization of docking parameters for LigandFit method
Positive control
O
Negative control
N
H
N
O
Cl
N
N
N
N
N
O
Cl
Cl
Cl
35DCPT
4CPT
Energy grid
Dreiding
CFF
PLP1
No. of correct poses 3,5-DCPT
No. of correct poses 4-DCPT
No. of negative control poses in top 25
2
—
4
—
6
8
0
–PLP1(4), –PLP2(4), Jain (4), PMF(4),
DOCK SCORE(6)
–PMF (7), DOCK SCORE
(6)
Best scoring functions
LigScore2
PLP1, PLP2
PLP1, PLP2
3,5-DCPT (rank of top 25 correct/closer poses for the best
scoring function)
4,5 (For all the scoring functions)
PLP1(9–12, 25), PLP2(13– PLP1(7–10, 14, 11–13, 25)
16, 25)
PLP2(11–18)
4-DCPT (rank of top 25 correct/closer poses for the best
scoring function)
—
PLP1 (1–8), PLP2(17–24)
PLP1(1–6), PLP2(20–25)
interactions with the secondary pocket. Picolinic acid derivatives
were substituted at the 3 position in order to interact with the ala-
nine pocket and this series also included basic substitutions at the 4
position (guanidinomethyl, diethyl aminomethyl, piperazinylmeth-
yl) for ion-pairing interactions. In addition, five-membered hetero-
cycles included 1,2,4-triazole, pyrazole, furan, pyrrole and
thiophene carboxylic acid were substituted with another heterocy-
clic ring structure (including imidazole, pyrazole, thiophene). Other
modifications of the furan carboxylic acid included addition of
diethylaminomethyl and piperazinylmethyl groups selected as
arginine isosteres due to their potential for ion-pairing interactions.
The binding site for the potential capping groups was created by
deletion of the N-terminus of the crystallographic ligand from the
CBG. After generation of the site points, fragment alternatives were
docked using the optimized LigandFit parameters. Docked frag-
ments were analyzed and shortlisted based on the required interac-
tions, PLP1 scoring, and correct positioning of the carbonyl for H-
bonding. The interactions of 5856 (PLP1 score = ꢁ35.03), 5581
(PLP1, -52.1) bound to the cyclin A groove are shown in Figure 1
and 5589 (PLP1, -42.61) in the cyclin D groove are shown in Figure 2.
Figure 2. Modeled structures of 5856 (cyan ligand) and 5589 (yellow ligand) in
complex with cyclin D1 (grey Connolly surface; 2W96). Cyclin residues are labeled
as three letter amino acid codes whereas peptide residues as one letter codes.
Favorable interactions between acidic residues of the cyclin (Glu 66 and 70) and the
piperazinylmethyl group of the inhibitor are represented by solid lines.
3.2. Structure–activity relationship of furoic and thiazole
carboxylic acid derivatives as N-Cap core structures
We have previously described Fragment Ligated Inhibitory
Peptides (FLIPs) with comparable potency to the native p21 penta-
peptide (RRLIF) in binding to CDK2/cyclin A (CDK2A) and CDK4/cy-
clin D1 (CDK4D1) complexes.15 CDK4/cyclin D1 is another cell
cycle regulatory kinase that controls the G1/S transition and is a
validated cancer target. The capping groups described were phe-
nylheterocyclic core structures as alternatives for the N-terminal
residues and resulted in inhibitors with more drug-like properties
including more optimal C log P values and polar surface area.15 Rel-
ative to the CDK4D1 activity of this pentapeptide, a significant po-
tency increase was obtained for phenyltriazole FLIPS. Binding of
the N-caps to the secondary hydrophobic pocket occurs through
van der Waals interactions with the phenyl ring bypassing the argi-
nine binding site.6 Potential for ion pairing interactions made by
the N-terminal residues of the peptide (HAKR), have therefore
not been fully exploited in the design of partial ligand alternatives.
Appropriate substitution of a heterocyclic scaffold would allow
complementarity with both the secondary hydrophobic pocket of
the groove and the acidic residues that surround this subsite. This
would build on the previous Ncaps where interaction of the phe-
Figure 1. Docked pose of the Ncap for 5581 (blue carbons) with cyclin A (grey
Connolly surface; 2V22) illustrating how the fragment alternatives are generated.
Cyclin residues are labeled as three letter amino acid codes whereas peptide
residues as one letter codes. The peptidic component retained during the docking is
shown with green carbons. 5581 interacts both with the secondary hydrophobic
pocket (through the right hand aminoethyl group) and through ion pairing
interactions with Glu220 and Glu224.

620
P. N. Premnath et al. / Bioorg. Med. Chem. 22 (2014) 616–622
nyltriazole rings (and pyrazole) occurs primarily through van der
Waals contacts with the alanine pocket. To achieve these addi-
tional interactions, functional groups including N,N-diethylamino-
methyl, pyrazolylmethyl and piperazinylmethyl groups were
selected from docking results and incorporated in the furoic acid
context in order to simultaneously interact with both subsites of
the cyclin. To facilitate synthetic accessibility, these were initially
pursued in the context of furoic acid and thiazole-4-carboxylic acid
core structures as a scaffold to present requisite functionality both
to the secondary hydrophobic pocket and to the acidic region of the
CBG.
A series of furoic acid and thiazole-4-carboxylic acid analogs
were therefore incorporated into FLIPs and evaluated as capping
group replacements for the N-terminus of the octapeptide. The
capping groups were ligated to the RLIF tetrapeptide and their
activity compared to the pentapeptide with the native arginine
residue (RRLIF) through testing in a fluorescence polarization
competitive binding assay (Table 2). Although this assay does not
greater than 100
analog with a 3-thienyl substitution (5582) was at least three
times greater than that of the 2-thienyl analog (5584, 56.2 M vs
>180 M) in the CDK2A2 assay.
lM. The affinity of the thiazole-4-carboxylic acid
l
l
3.3. Structure–activity relationship of picolinic acid and
phenylacetic carboxylic acid derivatives as N-Cap core
structures
In addition to the core structures described above, two alterna-
tive scaffolds were investigated. Fragment alternatives containing
picolinic acid and phenylacetic acid substructures were employed
in order to mimic the HAKR binding site. As observed in the phe-
nylheterocyclic series, a critical hydrogen bond is formed between
the N2 of the 1,2,4 triazole and indole NH of Trp217 of cyclin A.10
Structural comparison of triazole and picolinic acid core structures
indicate that the pyridine ring of the picolinic acid would present
an H-bond acceptor in a similar position to the N2 of the triazole
and thus represents a viable alternative scaffold.
provide
a readout of functional kinase activity, previous
comparisons have shown that there is a close correlation between
competitive binding and inhibition of CDK functional activity
through the cyclin groove.8,14 Of the FLIP analogs synthesized
and tested, the N-ethyl-N-methylethanamine analog (5581)
proved to be the most potent compound in this series not only
against CDK2A2 (43.5 lM) but also against CDK4D1 (41.9 lM).
The 1-ethyl-1H-imidazole analog (5585) and 1-ethyl-4-methyl-
1H-pyrazole analog (5761) only showed modest activities for
Picolinic acid derivatives were designed in order to incorporate
substituents interacting with either the secondary hydrophobic
pocket or the arginine binding site. R1 substituents including,
methyl (5845), methoxy (524) and ethoxy (523) substitutions (6
position) and at R2, a piperazine substitution (5856) was incorpo-
rated (Table 3). Analogs were synthesized in the RLNpfluorophe-
nylalanine (RLNpfF) or RLIF contexts with the former having a
slightly greater affinity as a pentapeptide.8 After testing in the
competitive binding assays, the unsubstituted analog (525) was
CDK2A with IC50 values falling to 100
des-methyl pyrazole analogs (5761, 5586) were shown to be weak
binders to CDK2A2 with IC50 values between 100 and 180 M. The
lM. Both the methyl and
found to have an IC50 of 29.2
(5845) and 6-methoxy (524) substitutions resulted in decreased
potency with IC50s of >100 and 36.4 M, respectively. The 6-ethoxy
containing FLIP (523) was determined to have an IC50 of 39.7
lM against CDK2A. Both the 6-methyl
l
(4-methylpiperazin-1-yl)methyl analog (5589) was shown to have
similar respectable activity against CDK2A2 and CDK4D1
l
lM
(IC50 = 50.3 and 43.2
within 2–3-fold that of RRLIF (16.1
imidazole methyl analog (5585) for CDK4D1 was approximately
150 M; slightly weaker than its CDK2A2 activity. The pyrazole
l
M, respectively), with its potency being
suggesting that affinity was not gained through substitution at
R1. In the CDK2A context, the 5-piperazine derivative (5856) did
not increase the binding affinity relative to 525 as the IC50 was
lM). The IC50 value for the
l
determined to be approximately 100 lM. This analog did however
Ncap analogs (5586 and 5761) were determined to have IC50 values
exhibit increased potency relative to 525 in the CDK4D1 (two-fold,
Table 2
Structure activity of 2-furoic acid and thiazole-4 carboxylic acid FLIPS
O
S
O
N
O
NH RLIF
NH RLIF
R
R
1
2
SCCP ID
5581
R
Core
1
CDK2/cyclin A IC50 (mM)
43.5 1.2
CDK4/cyclin D1 IC50 (mM)
41.9 3.1
N
N
—
N
5585
5586
1
1
—
—
>100
>100
158.6 5.2
>100
N
N
N
N
5589
5761
1
1
—
—
50.3 1.3
>100
43.2 5.0
>180
N
N
5582
5584
—
—
2
2
56.2 1.2
>180
>180
>180
S
S

P. N. Premnath et al. / Bioorg. Med. Chem. 22 (2014) 616–622
621
Table 3
Structure activity of picolinic acid FLIPs
O
O
R2
R2
N
NH RLIF
N
NH RLNpfF
R1
R1
1
2
SCCP ID
R1
R2
Core
CDK2/cyclin A IC50 (mM)
CDK4/cyclin D1 IC50 (mM)
525
5845
524
523
H
H
H
H
H
1
—
1
—
2
—
—
29.2 1.3
>100
36.4 1.2
39.7 1.2
48.4 1.2
85.7 1.9
36.9 1.4
>100
Me
MeO
EtO
1
HN
N
5856
H
—
2
>100
21.9 1.1
IC50 = 21.9
l
M). The 6-methyl (5845) and 6-ethoxy (523) analogs
binding modes of new fragment alternatives for the N-terminal
tetrapeptide of HAKRRLIF. Using this computational approach,
fragment ligated inhibitory peptides containing various N-cap
structures (5-furoic acid, thiazole-4-carboxylic acid, 5 and 6 substi-
tuted picolinic acids and 3 and 4 substituted phenylacetic acids)
were designed with the purpose of reproducing interactions with
CBG observed in peptidic inhibitors. After solid phase synthesis,
the inhibitory activity of these compounds against both CDK2A
and CDK4D1 was determined using an FP competitive binding as-
say to generate SAR information useful in the development of po-
tential cell cycle specific inhibitors. As a whole, scoring functions
used in high-throughput docking are qualitative and are used to
differentiate binders from nonbinders. It is therefore not surprising
that no correlation was observed between the PLP1 scoring values
calculated in these REPLACE iterations and the observed IC50
values.
were weaker inhibitors than 525 and 5856 for CDK4D1 however
the 6-methoxy derivative (524) had a slight potency increase to
36.9 lM compared to both of these FLIPs.
In a further series of N-terminal capping groups, the presence of
a bridging methylene between the linking carboxylate group and
the aryl ring was explored through the synthesis of phenylacetic
acid Ncaps so as to optimize interactions with the secondary lipo-
philic pocket. The SAR of four derivatives was determined in both
CDK binding assays (Table 4) and revealed that a 3,4-diethoxy ana-
log (530) was found to be the most potent against both CDK2A2
and CDK4D1 (5.2 lM vs 3.0 lM). Single R1 substitutions were ex-
plored and included ethoxy (5854), n-propyloxy (5853), and iso-
butoxy (5855). None of these analogs in the FLIP context
possessed appreciable binding affinity in either assay.
The most potent furoic acid based N-caps identified for both
CDK2A and CDK4D1 contained either a diethylaminomethyl or a
piperazinylmethyl group. Both of these derivatives would be posi-
tively charged at physiological pH, therefore indicating the impor-
tance of establishing ionic interactions with E220 (E66 in cyclin
D1) and thus confirming that they are reasonable mimics for the
native peptide arginine. The structural basis for the similar potency
observed with 5581 (Fig. 1) and 5589 (Fig. 2) for CDK2A and
CDK4D1 can be explained analyzing the cyclin contact residues
in each context. Both cyclin A and D have an acidic region
(Glu220, Glu224 in cyclin A, Glu66 in cyclin D) that interacts with
the arginine residue. The charged basic substituents of the two
capping groups therefore interact effectively with this acidic region
through ion pairing contacts. In addition, the ethyl groups of 5581
interact with the minor hydrophobic pocket of the cyclin groove
providing additional affinity for the CDK complexes studies
(Fig. 1). Although the 3-thienyl (5582) and 2-thienyl (5584) deriv-
atives both contain a similar heterocyclic moiety, 5582 displays in-
creased activity probably due to the secondary hydrophobic pocket
in the cyclin A context which is able to accommodate the larger
sulfur atom in the relevant position of thiophene ring.
4. Discussion
Prior to docking unknown fragments into the cyclin groove, the
LigandFit method was validated with known partial ligand alterna-
tives. For docking methods in general, the binding affinity of a li-
gand is estimated by a scoring function which evaluates the
receptor–ligand interactions. The PLP energy grid and scoring func-
tions were found to be the most effective for reproducing the
experimental binding mode of the 35DCPT and 4CPT N-terminal
capping groups (Table 1). Of the scoring functions used, only
PLP1 and PLP2 explicitly estimate contributions from hydrogen
bonding.12 Previous SAR has demonstrated that H-bonding to
Trp217 and Gln254 of cyclin A are critical interactions of a poten-
tial ligand,15 therefore suggesting why the PLP functions are the
most effective. This result indicates that the LigandFit docking pro-
tocol with the PLP energy grid and scoring function would be a
beneficial strategy in application to protein–protein interactions
which typically involve numerous interactions including multiple
H-bonds. Further to this validation, LigandFit was used to evaluate
Table 4
Comparison of the activity of the furoic acid derived FLIPs with
the phenylfuroic acid derivatives previously described suggests
that introduction of basic substituents results in significantly more
potent cyclin groove inhibitors. The significant activity of 5581 and
5589 indicates the increased contribution of the dimethylamino-
methyl group relative to the phenyl substituent. In addition, due
to the increased H-bonding ability of the triazole system relative
to the furan, phenyltriazole substituted FLIPs are about 100-fold
more active. Future design efforts should combine substitutions
(i.e., from 5581, 5589) onto a 1,2,4 triazole and therefore should
generate more potent CDK inhibitors.
Structure activity of Phenylacetic acid FLIPs
H
N
R1
R2
RLNpfF
O
SCCP
ID
R1
R2
CDK2/cyclin A IC50
(mM)
CDK4/cyclin D1 IC50
(mM)
5854
5853
5855
530
H
H
H
EtO
>100
>100
>100
>50
>100
CH3CH2CH2O
(CH3)2CHCH2O >100
5.2 1.1
EtO EtO
3
1.4

622
P. N. Premnath et al. / Bioorg. Med. Chem. 22 (2014) 616–622
The observed activity for 525 and the increase in potency ob-
conversion to nonATP competitive CDK inhibitors for development
as anti-tumor therapeutics.
tained for 5856 suggests that the picolinic acid scaffold is a viable
one for further development and especially in the CDK4D1 context
where addition of the piperazine to the aryl ring leads to increased
activity. Binding is more favorable in the CDK4D1 scenario since
the charged nitrogen is in closer proximity to E66 in this context
and is able to closely mimic the interactions of Arg4 (Fig. 2). The
increased activity of the 6-methoxy derivative (524) and the
decreased potency of the 6-ethoxy (523) relative to the unsubsti-
tuted picolinic acid version suggests that there is an optimal size
of the substituent for complementarity with the secondary
hydrophobic pocket and that steric clashes with Trp217 limit the
placement of larger groups on the aromatic ring. Furthermore the
results obtained for the phenylacetic acid Ncap (most potent Ncap
identified in this study) indicate the high potential of this scaffold
for further development. Modeling of the diethoxy analog (530)
concludes that only one of the substituents interacts with the
minor pocket of the cyclin groove while the other one limits the
conformational space available for the Ncap and therefore provides
an entropic advantage for binding. Placement of similar basic func-
tional groups to those in the furoic acid context instead of the
4-ethoxy substituent should provide increased affinity relative to
530.
Acknowledgments
We thank Dr’s. Michael Walla and William Cotham in the
Department of Chemistry and Biochemistry at the University of
South Carolina for assistance with Mass Spectrometry, Helga Co-
hen and Dr. Perry Pellechia for NMR spectrometry. This work
was funded by the National Institutes of Health through the re-
search project Grant, 5R01CA131368.
Supplementary data
Supplementary data associated with this article can be found, in
the online version, at http://dx.doi.org/10.1016/j.bmc.2013.10.039.
References and notes
1. Sherr, C. J. Science 1996, 274, 1672.
2. Chen, Y. N.; Sharma, S. K.; Ramsey, T. M.; Jiang, L.; Martin, M. S.; Baker, K.;
Adams, P. D.; Bair, K. W.; Kaelin, W. G., Jr. Proc. Natl. Acad. Sci. U.S.A. 1999, 96,
4325.
3. Shapiro, G. I. J. Clin. Oncol. 2006, 24, 1770.
4. McInnes, C. Drug Discovery Today 2008, 13, 875.
As a whole, these results demonstrate that N-cap core struc-
tures containing furoic acid, thiazole acid, picolinic acid and phen-
ylacetic acid represent viable mimetics for the critical arginine
residue and other determinants of potent cyclin groove binding.
A computational docking approach was validated as a useful ap-
proach for predicting protein–protein interaction inhibitors and
provided partial ligand alternatives that interact with both the
minor hydrophobic pocket and the acidic region of the CBG. Syn-
thesis of FLIP hybrid compounds based on the predicted structures
validated the hypothesis that this approach would provide in-
creased affinity relative to previous inhibitors that only interact
with the minor pocket alone. Overall, a greater relative potency in-
crease for FLIP inhibitors was observed for CDK4D1 than for CDK2A
compared to the native pentapeptide (RRLIF) for a number of
compounds and with potential improvement of their drug-like
properties. These compounds therefore more completely mimic
the contacts of the optimized peptide and are useful for further
5. Zheleva, D. I.; McInnes, C.; Gavine, A. L.; Zhelev, N. Z.; Fischer, P. M.; Lane, D. P. J.
Pept. Res. 2002, 60, 257.
6. McInnes, C.; Andrews, M. J.; Zheleva, D. I.; Lane, D. P.; Fischer, P. M. Curr. Med.
Chem. Anticancer Agents 2003, 3, 57.
7. Kirkland, L. O.; McInnes, C. Biochem. Pharmacol. 2009, 77, 1561.
8. Kontopidis, G.; Andrews, M. J.; McInnes, C.; Cowan, A.; Powers, H.; Innes, L.;
Plater, A.; Griffiths, G.; Paterson, D.; Zheleva, D. I.; Lane, D. P.; Green, S.;
Walkinshaw, M. D.; Fischer, P. M. Structure 2003, 11, 1537.
9. Kontopidis, G.; Andrews, M. J.; McInnes, C.; Plater, A.; Innes, L.; Renachowski,
S.; Cowan, A.; Fischer, P. M. ChemMedChem 2009, 4, 1120.
10. Andrews, M. J.; Kontopidis, G.; McInnes, C.; Plater, A.; Innes, L.; Cowan, A.;
Jewsbury, P.; Fischer, P. M. Chembiochem 1909, 2006, 7.
11. McInnes, C. In Annual Reports in Medicinal Chemistry In ; Bernstein, P., Desai, M.,
Eds.; Elsevier, 2012; Vol. 47, p 459.
12. Venkatachalam, C. M.; Jiang, X.; Oldfield, T.; Waldman, M. J. Mol. Graphics
Modell. 2003, 21, 289.
13. Guy, S.; Yacine, P. J.; Jean, M. Tetrahedron 2003, 59, 3315.
14. Andrews, M. J.; McInnes, C.; Kontopidis, G.; Innes, L.; Cowan, A.; Plater, A.;
Fischer, P. M. Org. Biomol. Chem. 2004, 2, 2735.
15. Liu, S.; Premnath, P. N.; Bolger, J. K.; Perkins, T. L.; Kirkland, L. O.; Kontopidis,
G.; McInnes, C. J. Med. Chem. 2013, 56, 1573.