Electron density guided fragment-based lead discovery of ketohexokinase inhibitors

Article abstract of DOI:10.1021/jm100677s

A fragment-based drug design paradigm has been successfully applied in the discovery of lead series of ketohexokinase inhibitors. The paradigm consists of three iterations of design, synthesis, and X-ray crystallographic screening to progress low molecula

Full text of DOI:10.1021/jm100677s

J. Med. Chem. 2010, 53, 7979–7991 7979
DOI: 10.1021/jm100677s
Electron Density Guided Fragment-Based Lead Discovery of Ketohexokinase Inhibitors†
Alan C. Gibbs, Marta C. Abad, Xuqing Zhang, Brett A. Tounge, Francis A. Lewandowski, Geoffrey T. Struble,
Weimei Sun, Zhihua Sui, and Lawrence C. Kuo*
Johnson & Johnson Pharmaceutical Research and Development, Welsh and McKean Roads, Spring House,
Pennsylvania 19477, United States
Received June 7, 2010
A fragment-based drug design paradigm has been successfully applied in the discovery of lead series
of ketohexokinase inhibitors. The paradigm consists of three iterations of design, synthesis, and X-ray
crystallographic screening to progress low molecular weight fragments to leadlike compounds.
Applying electron density of fragments within the protein binding site as defined by X-ray crystal-
lography, one can generate target specific leads without the use of affinity data. Our approach contrasts
with most fragment-based drug design methodology where solution activity is a main design guide.
Herein we describe the discovery of submicromolar ketohexokinase inhibitors with promising druglike
properties.
Introduction
will have high binding efficiencies as optimized larger
molecules.14,15 Compounds with a high binding efficiency,
definedashigh free energy of binding per non-hydrogen atom,
serve as optimal starting points for library design and com-
pound elaboration. A successful FBDD program also requires
the integrated collaboration between scientific disciplines
including structural biology, medicinal chemistry, and com-
putational chemistry.
Fragment-based drug design (FBDDa) has become an
important tool for drug discovery over the past decade.1-3
Typical FBDD campaigns employ screening of compounds
with low molecular weight and low structural complexity
against a therapeutic target of interest.4 Fragments that bind
or interact with the target are identified by way of a robust
biophysical technique, including X-ray crystallography,5-8
nuclear magnetic resonance spectroscopy,9 mass spectrometry,10
isothermal titration calorimetry,11 and surface plasmon
resonance.12,13 These techniques, although disparate in atomic-
level binding information, are ideal for FBDD screening
because they are able to characterize fragments with weak
affinity (millimolar to micromolar range) in a semiautomated
to fully automated medium-throughput format. The screen is
designed to identify fragments that interact with the target.
These fragments are then evolved to leads using synthetic
methods. Fragment elaboration, which is used throughout the
FBDD campaign, routinely employs the growing, linking, or
fusing of fragments alone or in combination.
Numerous FBDD success stories have been published, and
more than a handful of FBDD-derived compounds have
entered phase I clinical studies. A few of those have advanced
to phase II/III clinical trials.16 The successes originate from
many different pharmaceutical companies covering a diverse
range of target classes, including protein kinases, nuclear
hormone receptors, and protein chaperones.17 The FBDD
methodologies leveraged apply varied detection methods,
screening technologies, chemistry strategies, and protocols.
Often more than one screening technology is used in concert
with computational techniques such as virtual screening to
further complement the discovery effort. Although successful
FBDD campaigns differ in methodology and target class,
they are often similar in that a major emphasis is placed on
bioaffinity measurements to guide compound design.
The reported FBDD methodology herein is unique in that
the rounds of elaboration from primary to secondary to
tertiary library of compounds are conducted without any
guidance from affinity data. Instead, they are guided solely
with X-ray protein crystallographic data. Fragment electron
density is utilized to reveal if, where, and how a molecular
fragment binds. Comparison and superpositioning of two or
more occurrences of fragment electron density, within a series
of experiments, reveal the similarities and differences in bind-
ing interactions among the fragments and offer a “binding
surface” map. This map is then used to direct the design of
subsequent fragment libraries by either sprouting from a
fragment of choice or fusion of non-hydrogen atoms from mul-
tiple fragments that occupy different areas in the binding map.
Underlying FBDD are the following basic assumptions.
First, the fragment as part of an elaborated molecule will bind
in a similar orientation in the active site as an isolated
molecule. This assumption, which holds in the majority of
cases, forms the basis for rational structure based design
of potent compounds. Second, the low molecular weight
fragments (typically <250 Da) used in fragment screening
†Coordinates and structure factors for ketohexokinase complexes
with compounds 1 and 2 have been deposited in the Protein Data Bank
with accession code 3NBV, and those with compounds 3, 4, 5, and 6 have
been deposited as 3NBW, 3NC2, 3NCA, and 3NC9, respectively.
*To whom correspondence should be addressed. Phone: 215-628-
7850. Fax: 215-540-4632. E-mail: LKUO@its.jnj.com.
a Abbreviations: AMP-PNP, adenylyl imidodiphosphate; ATP, ade-
nosine triphosphate; CMC, Comprehensive Medicinal Chemistry
database; FBDD, fragment-based drug design; HATU, O-(7-azabenzo-
triazol-1-yl)-N,N,N0,N0-tetramethyluronium hexafluorophosphate;
rmsd, root-mean-square deviation.
r
2010 American Chemical Society
Published on Web 10/29/2010
pubs.acs.org/jmc

7980 Journal of Medicinal Chemistry, 2010, Vol. 53, No. 22
Gibbs et al.
We believe that by not ranking compounds based on their
solution activity, we enhance the design process with a
“normalization” of weak and strong binders and encourage
focus of the design to only the interactions between the
fragment and target.
The FBDD protocol followed here is initiated by X-ray
crystallography screening of a primary library of fragments
followed by iterative design and synthesis of secondary and
tertiary libraries. The primary library is a general purpose
library; it is composed of approximately 900 fragments with
no target class biases and no target-privileged functional
groups or pharmacophores. Following screening of the pri-
mary library, a secondary library is designed based on the
hits discovered. Similarly, a final tertiary library is designed
following X-rayscreening of the secondarylibrary. Incontrast
to the primary library, the secondary and tertiary libraries are
composed of 300-500 compounds each. Upon completion of
tertiary library synthesis, bioaffinity data are collected. Ter-
tiary library compounds meeting predefined thresholds in
various bioactivity, pharmacokinetic, and selectivity assays
will advance to “lead” status, and subsequent lead optimiza-
tion will commence.
Figure 1. Two-dimensional representations of small molecules that
bind ketohexokinase. X-ray crystallography derived structures of
these molecules, together with ketohexokinse, have been deter-
mined during this study.
selectivity against a panel of protein kinases as well as very
favorable pharmacokinetic properties.
Results and Discussion
The value of FBDD as a lead discovery engine derives from
its ability to act as an alternative source of leads for ongoing
projects or as a primary source of leads for projects where
traditional approaches have been unsuccessful. Using protein
crystallography as the screening tool may only be realized if
the protein target of interest possesses the following qualities:
a robust supply of single crystals; crystals that are able to
withstand fragment soaking at high concentration; a crystal
form that allows fragments to soak into the active site of the
protein; no occluded active sites when crystallized; better than
Ketohexokinase Structure. Ketohexokinase catalyzes the
phosphorylation of a wide range of furanose substrates,
most notably fructofuranose.18 The ketohexokinase mono-
mer is 298 amino acid residues in length and is a member of
the ribokinase superfamily of kinases.22 Enzymes within
the ribokinase family show similar tertiary structures de-
spite a low amino acid sequence homology. They catalyze
the phosphorylation of a variety of substrates. Examples
include D-fructose,23 adenosine,24 pyridoxal,25 4-amino-5-
hydroxymethly-2-methylpyrimidine,26 and 2-keto-3-deoxy-
gluconate.27
An alternatively spliced gene gives rise to a hepatic active
and a peripherally distributed, inactive isoform of KHK.28
This study focuses specifically on the active ketohexokinase
C isoform. The ketohexokinase protein construct used in this
study contains 294 residues (amino acids 5-298). Three-
dimensional structures of both isoforms have been deter-
mined via X-ray crystallography29 with ketohexokinase A and
ketohexokinase C in the apo form and ketohexokinase A
in complex with AMP-PNP (an ATP surrogate). The two
isoforms have high sequence homology, with differences
confined to a region of 44 residues primarily localized in the
β-sheet domain. Figure 1 illustrates two-dimensional repre-
sentations of AMP-PNP (compound 1) and pertinent small
molecules whose ketohexokinase complex structures have
been determined in this study.
Ketohexokinase tertiary structure is organized into two
domains, a large R/β/R fold and a small β-sheet region.
The biologically relevant quaternary structure of ketohexo-
kinase is a homodimer.23 The dimer interface between the
two monomers occurs through the β-sheet region. The two
β-sheet regions interact with each other to form a twisted
β-barrel, also known as the β-clasp30 as shown in Figure 2A.
The β-clasp (residues 13-40 and 98-114) acts as a lid for the
active site and is thought to allow substrate, product, and
cofactor to enter and exit. One active site per monomer is
located between the R/β/R fold and β-sheet domains. The
active site contains residues responsible for positioning ATP,
furanose, and a Kþ ion for chemistry. Figure 2B depicts
an enlarged view of the ketohexokinase active site with
bound AMP-PNP and D-fructose (compounds 1 and 2).
˚
2.8 A resolution X-ray diffraction. Our ketohexokinase crys-
tals meeting all requisite crystallography qualities renders
ketohexokinase an ideal candidate for FBDD.
Ketohexokinase catalyzes, with adenosine triphosphate
(ATP) and potassium ion (Kþ), the conversionof the furanose
form of D-fructose to fructose 1-phosphate.18 Ketohexoki-
nase, aldolase B, and triokinase are the enzymes catalyzing the
bulk of dietary fructose metabolism. A significant correlation
has recently been observed between increased dietary sugar
intake and the prevalence of non-insulin-dependent diabetes
mellitus.19 Non-insulin-dependent diabetes mellitus is often
a consequence of the metabolic syndrome, a collection of
disorders that include to varying degrees obesity, hypertri-
glyceridemia, insulin resistance, and hypertension.20 It is
because of this connection that regulation of dietary sugar
metabolism may have therapeutic benefit for people suffer-
ing non-insulin-dependent diabetes mellitus, obesity, or
hypertension. Ketohexokinase is directly implicated in reg-
ulation of dietary sugar metabolism; the product of its
catalysis fructose-1-phosphate is shunted directly into the
glycolytic pathway to provide a significant supply of carbon
for the biosynthesis of fatty acids and very low density
lipoproteins.21 As dietary sugar intake increases, fatty acid
and very low density lipoprotein syntheses are amplified. It is
hypothesized that inhibition of ketohexokinase activity
would suppress carbon supply for fatty acid and very low
density lipoprotein syntheses. Small molecule inhibitors of
ketohexokinase may be a practical means to achieve keto-
hexokinase modulation.
In this paper we describe the use of fragment electron
density guided FBDD to design small molecule inhibitors of
ketohexokinase. The eventual lead, compound 6, exhibits

Article
Journal of Medicinal Chemistry, 2010, Vol. 53, No. 22 7981
The AMP-PNP bound structure of ketohexokinase A has
previously been published,29 the AMP-PNP bound structure
of ketohexokinase C has not.
Inspection of the dimer backbone CR trace reveals two
distinct monomer conformations. One monomer possesses
an open active site, and the other conformer possesses a
closed active site. As shown in Figure 2C, the CR traces of the
two monomers are very similar throughout the R/β/R
domains but deviate extensively in the β-sheet domain,
˚
showing a root-mean-square deviation (rmsd) of 4.3 A for
the non-hydrogen atoms. A CR backbone superimposition of
ketohexokinase with two other known dimers (adenosine
kinase31 and ribokinase32) of the superfamily also reveals a
low rmsd for the CR backbone within the R/β/R fold domain
and a high backbone non-hydrogen atom rmsd for the
β-sheet domain.
Our ketohexokinase crystals in space group P212121 pos-
sess both open and closed conformations of ketohexokinase
independent of ligand binding. On the other hand, ligand
binding can be conformation dependent. Specifically, we
have observed that some fragments bind only the closed-
conformation monomer, some bind only the open-confor-
mation monomer, and some bind both. Herein, we designate
the closed conformation as monomer-A and the open con-
formation as monomer-B.
It is monomer-A that is catalytically competent. The
closed conformation of ketohexokinase positions all catal-
ytic residues, substrate, cofactor, and Kþ appropriately to
allow phosphorylation to occur. Residues aligned for catal-
ysis include the conjugate base, Asp258, and the β-clasp
residue Arg108. Asp258 extracts the 1-OH proton from
D-fructose, facilitating a 1-O- nucleophilic substitution at
the γ-phosphate of ATP which is anchored in position by
Arg108.30 Although its catalytic involvement is unclear,
Asp27 in monomer-B (Asp27B) may be important for other
processes. Asp27B extends from monomer-B into the active
˚
site of monomer-A to within approximately 6 A of bound
AMP-PNP, a feature not previously observed in any crystal
structure from the ribokinase superfamily. This feature is
exploited during fragment library design throughout our
FBDD campaign.
Primary Library. In putting together the primary library,
we applied a variety of filters to include the number of non-
hydrogen atoms being g6 and e15, number of hydrogen
bond acceptors being e3, number of hydrogen bond donors
being e3, number of rings being e2, and no unspecified
chiral centers. Compounds with moieties containing reac-
tive, toxic, or nondruglike groups are removed. In total 45
specific functional groups and substructures were excluded
(see Supporting Information for a complete list).
Next, the fragments were compared to the Comprehensive
Medicinal Chemistry database (CMC),33 a collection of over
8000 known pharmaceutical compounds, to determine if a
fragment was a substructure in a CMC compound. If so, the
compound was given a CMC subsimilarity (CMC_Subsim)
score of 1. If not, it was assigned a score of 0. An overall
FBDD score (FBDD_Score) was created by applying the
equation
˚
Figure 2. Structural features of ketohexokinase at 2.3 A resolution.
(A) Ketohexokinase dimer topology. The interlocking β-clasp is
visible at the dimer interface. Monomer-A (gray) is in the closed
conformation, and monomer-B (white) is in the open conformation.
AMP-PNP is partially visible in monomer-A. (B) Catalytic site of
ketohexokinase C with AMP-PNP (compound 1) and D-fructose
(compound 2). AMP-PNP forms hydrogen bonds through N-3 of
adenine to the conserved water (red sphere) and through HO-30 of
ribose to Gly229. D-Fructose is positioned below the terminal amido-
phosphateofAMP-PNP(γ-phosphateofATP). ExtensionofAsp27B
into the active site is clearly visible in the bottom of the figure.
Although not visible in the density, Mg2þ or Mn2þ may also facilitate
ATP and AMP-PNP binding to ketohexokinase. D-Fructose binds in
a sterically constrained pocket and is held in place by a bidentate
interaction between HO-30 and HO-40 with Asp15, a hydrogen bond
between HO-10 and Asn45, and a hydrogen bond between HO-60 and
Glu29B. (C) Ribbon representation of the backbone superimposition
of monomer-A (gray) and monomer-B (white). The R/β/R domains
are very similar in conformation. In contrast, the β-sheet domain
(comprising the β-clasp) conformations are very different and the
difference between the open and closed states is dramatic. The arrow
FBDD Score ¼ CMC Subsim
- ½ðnumber of non-hydrogen atoms - 6Þ=9ꢀ
- ðpercent heteroatomÞ
˚
indicates a distance of 11 A between the open and closed β-sheets. All
three-dimensional structure figures were created by MOE (Chemical
Computing Group, http://www.chemcomp.com).
The result was a 900-member primary library of fragments.

7982 Journal of Medicinal Chemistry, 2010, Vol. 53, No. 22
Table 1. Crystallographic Data Collection and Refinement Statisticsa
Gibbs et al.
compd, PDB accession no.
4, 3NC2
parameter
1 and 2, 3NBV
P212121
3, 3NBW
P212121
5, 3NCA
P212121
6, 3NC9
P212121
space group
unit cell
P212121
˚
a, A
82.8
86.0
82.7
85.6
83.4
85.1
82.9
85.5
82.7
85.6
˚
b, A
˚
c, A
˚
resolution, A
136.9
2.3
136.6
2.3
137.2
2.5
137.5
2.6
136.8
2.4
completeness, %
merge,b %
98.2 (98.3)
7.1 (32.6)
18.9 (3.6)
20.7
94.4 (78.8)
5.3 (28.9)
16.1 (3.1)
22.4
98.8 (99.1)
4.5 (30.0)
16.0 (3.4)
23.3
99.6 (99.5)
6.1 (37.0)
21.1 (3.9)
19.2
99.2 (99.9)
5.9 (39.0)
10.6 (2.7)
23.6
R
ÆIæ/ÆσIæ
Rfactor,c %
R
free,d %
rmsd bond, A
23.6
0.008
1.12
27.4
0.008
1.17
27.1
0.008
1.10
23.9
0.009
1.17
28.6
0.009
1.16
˚
rmsd angle, deg
a Values in parentheses refer to the highest resolution shell. b Rmerge
and ÆIæ is the average intensity for this reflection, with summation over all data. Rfactor
withheld.
P
P
=
I (|II - ÆIæ|/ÆIæ), where II is an individual intensity measurement
hkl
P
P
c
=
||Fo| - |Fc||/ |Fo|. d 10% of the total reflections
Fragment Screening. Fragments or compounds from both
the primary and secondary libraries were screened in groups
of five by selecting five of the nearest-neighbor compounds in
a structural-similarity space, based on substructure or two-
dimensional topological descriptors. Although clustering by
structural similarity would be counterintuitive if one were to
deconvolute bound structures, grouping similar fragments
reinforced identification of binding of a common chemo-
type, thus minimizing scaffold ambiguity when electron
density was found in the active site. This was particularly
advantageous in cases where electron density in the binding
site lacked perfect definition. The combination of struc-
turally similar fragments in a cocktail provided a “binary”
answer of “yes” or “no” on binding. In our FBDD protocol,
X-ray crystallographic screening was complete after the
secondary library screen. Tertiary library compounds were
not clustered into groups of five; selected compounds were
Figure 3. X-ray crystallographic determination of all primary li-
brary “hits” in the active site of ketohexokinase. Sixty fragments
are shown. The fragments are accommodated in a wide range of
binding orientations. The resolution of the bound structures ranges
soaked individually for nonscreening mode analysis.
ATP-Site Feature Map. Ketohexokinase crystals were
soaked overnight with the primary library fragment cocktails
and subjected to X-ray diffraction analysis. Protein structure
determination was performed on all data sets. The resolved
structures found to contain extra electron density within the
protein active site due to fragment binding were further
refined. During structure refinement, one of the five frag-
ments in the cocktail was modeled into the observed extra
electron density in the active site of the protein. The model
was then optimized with electron density weighting giving
the most reasonable orientations and binding modes as
measured by typical crystallographic statistics, as shown in
Table 1. Occasionally the electron density of the fragment
was ambiguous allowing more than one orientation to fit
or a second fragment from the cocktail to fit. In these cases,
all possible structures were refined. In total approximately 60
data sets were refined, giving an unusually high hit rate of
30%, where a hit was defined as discernible active-site
fragment density. The majority of fragment hits were found
to bind monomer-A. For this reason, we used only the
structure of this monomer in the design of compounds for
the secondary and tertiary libraries. Unless stated otherwise,
all discussions below refer to monomer-A.
˚
from 2.3 to 3.0 A.
˚
Figure 4. Ketohexokinase ATP-site feature map at 2.3 A resolu-
tion. Regions of interest, largely defined by primary library frag-
ment binding, are highlighted red; see text for details. A primary
library pyrazole (compound 3) is shown hydrogen-bonding to the
conserved water, red sphere. The water molecule, in the reference
frame, is at 12 o’clock.
Figure 3 illustrates the structural diversity and scope
of binding within the active site of ketohexokinase of hits
from the primary library fragments. The hit fragments are

Article
Journal of Medicinal Chemistry, 2010, Vol. 53, No. 22 7983
accommodated in a moderately sized volume roughly char-
3
indicate that Asp27B is the only residue involved in ATP-site
fragment binding. The negatively charged carboxylate of
Asp27B attracts fragments containing charged amines and
basic groups, revealing another potentially favorable spot
for ligand interaction. This interaction is successfully
exploited in the design of a number of tertiary library
compounds (described below).
Between the 8 and 10 o’clock positions is the egress of the
active site to the bulk solvent, possibly the entrance for ATP
and D-fructose as well as exit for ADP and fructose 1-phos-
phate. At the 10 o’clock position, three sequential proline
residues (Pro246, Pro247, and Pro248) define the perimeter
of a small hydrophobic binding pocket. Fragments and
functional groups located in this region are composed of
single ring aromatics or small (cyclo)alkyl moieties.
Secondary and Tertiary Libraries. The aim of the second-
ary library screen is to explore sprouting or merging of
fragment hits found from the primary library. The aim of
the tertiary library screen is to focus on a small number of
chemotypes and to apply a greater emphasis on the synthesis
of compounds with leadlike properties. The secondary li-
brary contains greater compound diversity for a larger
number of chemotypes, whereas the tertiary library contains
a concentration of leadlike compounds centered on a fewer
number of chemotypes. How strict one selects primary
library hits for the design of a secondary library is largely
dictated by the hit rate. In the case of ketohexokinase, the
primary library hit rate is high allowing a more stringent
selection process to include novelty, diversity, ease of synthe-
sis, and quality of fragment electron density.
˚
˚
˚
˚
acterized by a 14 A ꢁ 11 A ꢁ 3 A box, or ∼460 A , to yield an
“ATP-site feature map”.
Figure 4 details the ATP-site feature map as defined by hits
from the primary library screen. At the 12 o’clock position of
the figure, an explicit density from a conserved water mole-
cule, donating a hydrogen bond to the main-chain Cys282
carbonyl, is often well resolved in all our ketohexokinase
structures. The conserved water molecule is observed in
many related kinases of the ribokinase superfamily although
the residue it interacts with is not conserved. In the majority
of cases, we find that the conserved water also donates a
hydrogen bond to the bound ligand. This protein-water-
ligand interaction plays a key role in mediating binding
of ATP, AMP-PNP (through N-3 of adenine; see also
Figure 2B), and the majority of fragments to ketohexoki-
nase. Interestingly, among all fragment-bound ketohexoki-
nase structures solved in this study, only one fragment
displaced the conserved water.
At the 2 o’clock position, the side chains of Cys282
and Phe260 define a small pocket that accommodates the
20 and 30 ribose hydroxyl groups of the endogenous ligand
ATP (see also Figure 2B). Fragments that bind near this
pocket reveal the inherent specificity of the pocket for small
alkyl (with apparent preference for S-methyl) and aromatic
groups of non-ATP molecules. Often observed are π-π
interactions between the aromatic side chain of Phe260 and
(hetero)aromatic fragments. T-shaped (edge-to-face) inter-
actions are usually observed, although face-to-face interac-
tions are also represented.
At the 4 o’clock position is the “phosphate channel”,
where the endogenous ATP phosphate groups reside. This
is a polar region with few primary library fragment hits.
A common observation among our structures is the presence
of a sulfate ion, a constituent of the crystallization buffer,
occupying a similar position as the γ-phosphate of ATP and
AMP-PNP. Residues Gly255, Gly257, and Arg108 coordi-
nate the sulfate ion. Coincidentally, a sulfate ion is also
observed in the apo crystal structure of monomer-A of
ketohexokinase.29 Because of its polar nature, this region
provides sought-after protein hotspots for lead design.
At the 5 o’clock position is a small, extremely hydrophilic
pocket where the furanose form of D-fructose binds. In the
absence of sugar, the pocket recruits four water molecules.
The oxygen atoms of these water molecules overlay with the
four D-fructose hydroxyl oxygen atoms when superimposed,
with a low rmsd. Several fragments are found to bind in this
pocket, but they are not part of the design of compounds in
this study.
At the 6 o’clock position, or the central region of the active
site, is where the adenine heterocycle of AMP-PNP, or ATP,
binds. This region is surrounded by nonaromatic, hydro-
phobic residues from both the R/β/R and β-sheet domains
and binds almost exclusively flat (hetero)aromatic frag-
ments. The pocket is very narrow in the closed conformation
of the enzyme, limiting accommodation and accessibility
primarily to flat fragments, but displays little preference as
to the composition of allowed aromatic rings. Five-member,
six-member, fused bicyclic, fused tricyclic-aromatic, and
heteroaromatic rings have all been found to bind.
Increasing bias is given to structural novelty as the chemo-
type progresses throughout secondary and tertiary library
design, synthesis, and screening. The designed libraries must
also be synthetically accessible, contain features that help
elucidate the nature of specific protein-chemotype interac-
tions, and possess leadlike physicochemical properties. Often
various synthetic routes are required in order to accomplish
these goals.
Our FBDD design method for ketohexokinase focuses on
˚
fragment electron density found within a radius of 5 A from
the location of the two bound ligands defined in the X-ray
structure of ketohexokinase crystallized with AMP-PNP and
D-fructose (shown in Figure 2B). The electron density fea-
tures tracked during the course of this study are size, com-
plementarity (goodness of fit), and location (see Figure 5).
Screening the Secondary Library. The secondary library
was designed around hits found from the primary library to
produce 6 scaffolds: II_A (pyrazoles), II_B (arylamides),
II_C (benzotriazoles), II_D (pyridines), II_E (sulfones),
and II_F (triazolones). They are shown in Figure 6 together
with the substituents proposed for elaboration. Substituents
are chosen on the basis of the electron density of the hits and
the ATP-site feature map derived thereof. Approximately 350
compounds were synthesized. These compounds were screened
in a manner identical to that performed for the primary library
fragments. Hits were only discovered for compounds belong-
ing to the sublibraries II_A, II_B, II_C, and II_D.
Screening the Tertiary Library. In contrast to the second-
ary library where fragment elaboration consisted primarily
of growing, fragment fusion was heavily used in the prepara-
tion of fragments in the tertiary library. This is illustrated
in Figure 7 for the tertiary sublibrary designated as III_A.
Electron density relationships between pyrazole-containing
fragments and various six-member aromatic rings found
At the 7 o’clock position, the carboxylate side chain of
β-clasp residue Asp27B enters the catalytic site. Although
it is possible that other residues from the monomer partner
are involved in binding and/or catalysis, our observations

7984 Journal of Medicinal Chemistry, 2010, Vol. 53, No. 22
Gibbs et al.
and were the basis for the incorporation of 6:5 fused hetero-
cycles into tertiary sublibrary III_A. Additional support for
the incorporation of 6:5 analogues came from pyrazole hits
containing 4-carboxamide and 5-amine functional groups
that formed pseudo-six-member rings via intramolecular
hydrogen bonding, as revealed by molecular mechanics
and virtual docking exercises. The 6:5 fused systems of III_A
(indazoles, imidazopyridines, pyrazolopyrimidines, and
pyrazolopyrimidinones) provided the following compound
properties: (1) As basic amines were commonly located
around the 7 o’clock region of the active site (see Figure 4)
making favorable interactions with Asp27B, extension of the
scaffold relative to the parent pyrazole provided alternative,
well positioned “handles” for attachment of amines. (2) The
6:5 analogues maintained hydrogen bonding with the con-
served water via an appropriately positioned atom within the
five-member heterocycle (equivalent to N-2 of the pyrazole).
(3) Both aryl and small alkyl substituents on either side of the
five-member ring, as dictated by the ATP-site feature map,
could be maintained by the 6:5 systems.
Figure 9 outlines the design scheme for the tertiary sub-
library III_B. This series of fragments required a carbox-
amide for accepting a hydrogen bond from the conserved
water and tolerated little diversity in the amide substituent
relative to the parent cyclopropylethyl. Similar to sublibrary
III_A, the core aromatic scaffold was extended on the basis
of electron density comparison and virtual docking experi-
ments, in attempts to make the available Asp27B interaction
with a pendent basic amine.
In contrast to sublibraries III_A and III_B, compounds
belonging to the III_C sublibrary primarily consisted of
modification to the core scaffold shown in Figure 10 with
maintenance of a 6:5 heterocycle. This core scaffold did not
interact directly with the conserved water. Interactions with
the water molecule were observed for parent analogues in
secondary library II_C that contained hydrogen bonding
substituents extending off the scaffold, e.g., pendent imida-
zole rings. Analogues were designed into this tertiary library
to follow this trend.
Fragments of secondary sublibrary II_D were positioned
in the active site by assuming interaction of the pyridine
nitrogen with the conserved active site water. This nitrogen
in the context of a six-member ring was held constant mostly
in the design of the final tertiary sublibrary, III_D. Figure 11
illustrates the tertiary sublibrary III_D design scheme. In
some cases, the six-member ring was extended to isoquino-
line or semisaturated heterocycles. It was hoped that exten-
sion of the pyridine scaffold would provide a reasonable
handle for not only basic amine attachment to interact with
Asp27B but also phosphate-channel attractive substituents
(see the 4 o’clock position in Figure 4). Unfortunately,
compounds in any series with polar functional groups that
were predicted to produce favorable interactions in the
phosphate channel region did not show any discernible
electron density in the bound structure of ketohexokinase.
Compound 6. All tertiary library compounds were tested for
solution activity in an enzyme assay. This step was taken to
transition the FBDD effort into a lead optimization effort. We
found many tertiary library scaffolds offered at least one repre-
sentative displaying an IC50 of <5 μM in the ketohexokinase
enzyme activity assay. These scaffolds are highlighted by the
boxed structures in Figures 7, 9, and 10. The most promising
ketohexokinase inhibitors came from sublibrary III_A, the
indazoles, which led to the discovery of compound 6.
Figure 5. Electron density characteristics used for defining fragment
constitution and relationships. A small sulfate ion is illustrated in panel
˚
Aat2.3Aresolution. The density (at 2.5 A resolution) of compound 4in
˚
panel B allows more than one orientation of the symmetric fragment to
˚
fit. The density (at 2.3 A resolution) of compound 3 in panel C allows an
unambiguous fit. Compound 2 in panel D resides in the fructose binding
0
˚
site (2.3 A resolution). For comparison the O-1 of compound 2 is
˚
9 A from the nitrile N of compound 3 (C) when superimposed.
for hits from the primary and secondary (II_A) libraries
revealed non-hydrogen atom overlaps (see Figure 8)

Article
Journal of Medicinal Chemistry, 2010, Vol. 53, No. 22 7985
Figure 6. Secondary library design scheme. The six primary library fragments selected for secondary library design are shown at left.
Corresponding constitution of the six sublibraries are illustrated at right: II_A (pyrazoles); II_B (aryl amides); II_C (benzotriazoles); II_D
(pyridines); II_E (sulfones); II_F (triazolones).
The structural evolution of lead compound 6 from a
representative pyrazole hit (compound 3) is depicted in
Figure 12. Our data indicate that large modifications are
tolerated around certain regions of the parent pyrazole
scaffold, whereas in other regions little change is tolerated.
Although the search for optimal substituents was not ex-
haustive, minimal tolerance to modification of the S-methyl
(at 1 o’clock as shown in Figure 4) and N-phenyl (at 11
o’clock as shown in Figure 4) is observed. In contrast, large
structural modifications to the pyrazole ring extending the
scaffold into the 6 and 7 o’clock regions of the active site
(as shown in Figure 4), via a 6:5 fusion, are well tolerated and
very advantageous for inhibitor design. Although many
tertiary library indazoles displayed submicromolar affinity,
all tertiary library pyrazoles or precursors of the indazoles
showed IC50 values greater than 5 μM in the enzyme assay.

7986 Journal of Medicinal Chemistry, 2010, Vol. 53, No. 22
Gibbs et al.
Figure 7. Tertiary sublibrary III_A design scheme. Various single ring and fused ring heterocycles make up the sublibrary. Boxed structures
indicate scaffolds with at least one tertiary library representative testing below 5 μM in the ketohexokinase enzyme assay.
Compound 6 and its synthesis are shown in Scheme 1.
Cyclization of 2-amino-4-nitrobenzoic acid 7 with N,N0-
carbonyldiimidazole yielded the 7-nitro-1H-benzo[d ][1,3]-
oxazine-2,4-dione intermediate, which was then reacted with
phenylhydrazine to generate compound 8 (∼34% yield in
two steps). Compound 8 was treated with NaNO2 and HCl
to give the indazolone intermediate which was then reacted
with P2S5 to yield the corresponding indazolethione 9
(∼11% yield in two steps). Methylation of compound 9, with
MeI in the presence of Cs2CO3 followed by reduction of the
nitro group with hydrogenation at normal atmospheric
pressure via the catalysis of Pd on carbon, gave compound
10 (∼77% yield in two steps). Compound 10 was coupled
with piperidine-1,4-dicarboxylic acid mono-tert-butyl ester
in the presence of O-(7-azabenzotriazol-1-yl)-N,N,N0,N0-
tetramethyluronium hexafluorophosphate (HATU). The
adduct was then deprotected using HCl to yield the final
product, compound 6, in two steps with a 40% yield.
Compound 6 displays an IC50 value of 330 nM in the
ketohexokinase enzyme assay. In inhibition assays with
recombinant CYP450 enzymes, compound 6 shows no sig-
nificant activity against any of the five major human CYP450
enzymes (CYP1A2, CYP3A4, CYP2C9, CYP2C19, CYP2D6,
>10 μM). The compound is stable in human and rat liver
microsomes (96% remaining at 10 min in HLM and 95%
in RLM). Compound 6 possesses high aqueous solubility at
both pH 7.4 and pH 2.0 (45.2 μM in PBS, pH 7.4; 35.3 μM
in PBS at pH 2.0; 40.7 μM in water; 40 μM in ACN). The
compound is found to be permeable through a monolayer
of Caco-2 cells in vitro, with efflux ratio of 5.74 (Papp(AfB)=
1.65 ꢁ 10-6 cm/s; Papp (BfA) = 9.47 ꢁ 10-6 cm/s). Com-
pound 6 is highly orally bioavailable, exhibiting an oral bio-
availability of 75.4% in rats when dosed at 10 mg/kg using 20%
hydroxypropyl-β-cyclodextrin (HPBCD) as the vehicle (10
mg/kg po, Cmax = 529 ng/mL, AUClast = 9031 h ng/mL).
3
Compound 6 shows moderate in vivo systemic clearance (11.6
(mL/min)/kg) and a high volume of distribution (8.53 L/kg).
The potential for cardiovascular side effects with com-
pound 6 was investigated. The compound shows an IC50 of
8.13 μM in hERG channel astemizole binding assays. In
single-cell patch clamp studies, inhibitions of the membrane
Kþ current Ikr are observed at the three concentrations tested
(18% at 0.1 μM, 27% at 0.3 μM, and 57% at 3 μM, compared
to10%, 16%, and19%withsolvent asthe control). Selectivity
of compound 6 against protein kinases (Invitrogen), G-pro-
tein-coupled receptors (Cerep), and ion channels (Cerep) were
also evaluated. The Invitrogen protein kinase panel contains
200 kinases; compound 6 inhibits protein kinase A (99%
inhibition) at 10 μM compound and 10 μM ATP. It does
not inhibit any of the other protein kinases greater than 37%
at 10 μM. As indicated by the Cerep panel screening,
compound 6 inhibits three receptors greater than 70% at
10 μM; they are 5-HT1A (83% inhibition), 5-HT2B (91%
inhibition), and 5-HT7 (95% inhibition). No ion channels are
inhibited by compound 6. As these apparent inhibition values
are based on a single concentration of compound 6, true
inhibitions would need to be determined by dose response.
Selectivity within the ribokinase super family was not tested.

Article
Journal of Medicinal Chemistry, 2010, Vol. 53, No. 22 7987
LCMS-2110EV instrument with analytical LC using Shiseido
Capcell Pak C18 MGS3 column (3.0 mm ꢁ 50 mm, 3.5 μm),
0.05% TFA in water as mobile phase A, and 0.05% TFA in
acetonitrile as mobile phase B at 1 mL/min flow, gradient from
10% to 100% B in 1.7 min, 100% B for 1.5 min, 100% to 10% B
in 0.13 min, monitored by UV absorption at 215 nm using PDA
and ELSD. The purity of all compounds was found to be
>95%. Thin-layer chromatography was performed on Merck
PLC prescored plates 60F254. Unless otherwise noted, reagents
were obtained from commercial sources and were used without
further purification.
Preparation of 6: N-(3-(Methylthio)-1-phenyl-1H-indazol-6-yl)
piperidine-4-carboxamide Hydrochloride Salt. A solution of
2-amino-4-nitrobenzoic acid (54.6 g, 300.00 mmol, 1.00 equiv)
in tetrahydrofuran (500 mL) and di(1H-imidazol-1-yl)methanone
(58.32 g, 360.00 mmol) was added to a 1000 mL four-necked
round-bottom flask, then purged and maintained with an inert
atmosphere of nitrogen. The resulting solution was stirred for 2
h at 10 ꢀC. The solids were collected by filtration. This resulted in
56 g (90%) of 7-nitro-1H-benzo[d ][1,3]oxazine-2,4-dione as a
yellow solid.
A solution of 7-nitro-1H-benzo[d ][1,3]oxazine-2,4-dione(56g,
269.23 mmol, 1.00 equiv) in ethanol (500 mL) and 1-phenylhy-
drazine (31.98 g, 296.11 mmol, 1.10 equiv) was added to a 1000 mL
three-necked round-bottom flask. The resulting solution was
heated to reflux for 5 min. The resulting mixture was concen-
trated under vacuum. The residue was applied onto a silica gel
column with dichloromethane/petroleum ether (1:2 to 1:0) and
ethyl acetate/petroleum ether (1:1). This resulted in 28 g (38%)
of 2-amino-4-nitro-N0-phenylbenzohydrazide as a yellow solid.
MS (m/z): 273 [MHþ].
A solution of 2-amino-4-nitro-N0-phenylbenzohydrazide (23 g,
84.56 mmol, 1.00 equiv) in ethanol (250 mL), 1 N hydrogen
chloride (250 mL), and tetrahydrofuran (100 mL) was added to
a 1000 mL four-necked round-bottom flask. This was followed
by the addition of a solution of NaNO2 (17.5 g, 253.62 mmol,
3.00 equiv) in water (45 mL) dropwise with stirring at 72 ꢀC in
40 min. The resulting solution was heated to reflux for 1 h. The
reaction mixture was cooled to 25 ꢀC. The solids were collected
by filtration. This resulted in 11.5 g (53%) of 6-nitro-1-phenyl-
1,2-dihydroindazol-3-one as a brown solid. MS (m/z): 256 [MHþ].
6-Nitro-1-phenyl-1,2-dihydroindazol-3-one (15 g, 58.82 mmol,
1.00 equiv), xylene (150 mL), and P2S5 (52.2 g, 235.14 mmol)
were added to a 250 mL three-necked round-bottom flask. The
resulting solution was heated to reflux for 40 min in an oil bath.
The reaction mixture was cooled to 25 ꢀC. The resulting solution
was diluted with 50 mL of ethyl acetate. The solids were filtered
out. This resulted in 3.1 g (19%) of 6-nitro-1-phenyl-1,2-dihy-
droindazole-3-thione as a yellow solid. MS (m/z): 272 [MHþ].
A solution of 6-nitro-1-phenyl-1,2-dihydroindazole-3-thione
(3.1 g, 11.44 mmol, 1.00 equiv) in CH3CN (20 mL) and Cs2CO3
(21 g, 64.42 mmol, 5.63 equiv) was added to a 100 mL three-
necked round-bottom flask. This was followed by the addition
of a solution of iodomethane (11.3 g, 79.58 mmol, 6.96 equiv) in
CH3CN (15 mL) dropwise with stirring at 10 ꢀC in 30 min. The
resulting solution was stirred for 30 min at 10 ꢀC. The solids were
filtered out. The resulting mixture was concentrated under
vacuum. The residue was applied onto a silica gel column with
ethyl acetate/petroleum ether (1:100). This resulted in 2.8 g
(86%) of 3-(methylthio)-6-nitro-1-phenyl-1H-indazole as a yel-
low solid. MS (m/z): 285 [MHþ].
Figure 8. Example of electron density comparison used to fuse two
fragment substructures. Panel A illustrates the density overlap
between two experiments as shown by yellow and green density
mesh (yellow density = compound 3, green density = compound 5).
Panel B shows the non-hydrogen atom overlap between the frag-
ments, the genesis of 6:5 fused rings from compound 3 as shown in
sublibrary III_A.
Conclusion
The methodology described here provides a unique route
to the discovery of lead compounds by FBDD-based ap-
proaches. The discovery of the lead series of ketohexokinase
inhibitors shown here is driven with fragments from a primary
library, followed by two rounds of compound elaboration,
and directed by X-ray electron density maps without guidance
from affinity data. The outcomes are compounds that are
structurally unique, are leadlike in pharmacokinetic proper-
ties, are minimally restrained by restricted intellectual prop-
erty space, and possess significant binding affinity toward the
target of interest. All of these properties are embodied in
compound 6.
Experimental Section
Chemistry. 1H nuclear magnetic resonance spectra were
determined with a Bruker Biospin International AG-400 spec-
trometer at 400 MHz. Chemical shifts (δ) are reported in parts
per million (ppm) relative to residual chloroform (7.26 ppm),
TMS (0 ppm), or CD3OD (4.87 ppm) as an internal reference
with coupling constants (J ) reported in hertz (Hz). The peak
shapes are denoted as follows: s, singlet; d, doublet; t, triplet;
q, quartet; m, multiplet; br, broad. Electrospray mass spectra
were recorded in positive or negative mode on a Micromass
Platform spectrometer. The purity of all compounds described
here was determined by analytical LC/MS using a Shimadzu
A 100 mL round-bottom flask was purged, flushed, and main-
tained with a hydrogen atmosphere. A solution of 3-(methylthio)-
6-nitro-1-phenyl-1H-indazole (3.5 g, 12.28 mmol, 1.00 equiv) in
ethyl acetate (30 mL) was added to the flask followed by 5%
palladium on carbon (2.2 g), then bubbled with H2 (gas). The
resulting solution was stirred for 48 min at room temperature. The
solids were filtered out. The resulting mixture was concentrated
under vacuum. This resulted in 2.8 g (89%) of 3-(methylthio)-1-
phenyl-1H-indazol-6-amine as brown oil. MS (m/z): 256 [MHþ].

7988 Journal of Medicinal Chemistry, 2010, Vol. 53, No. 22
Gibbs et al.
Figure 9. Tertiary (III_B) sublibrary design scheme. The carboxamide, important for hydrogen bonding to the conserved water, is maintained
in all compounds. Boxed structures indicate scaffolds with at least one tertiary library representative testing below 5 μM in the ketohexokinase
enzyme assay.
A solution of 4-(3-methylsulfanyl-1-phenyl-1H-indazol-6-yl-
carbamoyl)piperidine-1-carboxylic acid tert-butyl ester (112 mg,
0.24 mmol, 1.00 equiv) in dichloromethane (4 mL) was added to
a 50 mL three-necked round-bottom flask. The solution was
bubbled with hydrogen chloride (gas) and stirred for 30 min
at 0 ꢀC. The solids were collected by filtration. This resulted
in 46 mg (52%) of N-(3-(methylthio)-1-phenyl-1H-indazol-6-
yl)piperidine-4-carboxamide hydrochloride as a yellow solid.
MS (m/z): 367 [M - HCl þ H]þ. 1H NMR (400 MHz, MeOD,
ppm): δ 10.082 (1H, s), 8.303 (1H, s), 7.628-7.609 (2H, d),
7.568-7.546 (1H, d), 7.498-7.461 (2H, t), 7.316-7.279 (1H, t),
7.151-7.130 (1H, d), 3.421-3.390 (2H, d), 3.044-3.983 (2H, t),
2.703-2.648 (1H, d), 2.597 (3H, s), 2.054-2.020 (2H, d),
1.947-1.857 (2H, m).
Enzyme Assay. An enzymatic assay was developed to measure
ketohexokinase-catalyzed conversion of D-fructose to fructose
1-phosphate, using a LCMS instrument to quantify the product.
This assay was adapted to a high throughput format with use of
the BioTrove RapidFire.
Inhibition of ketohexokinase was analyzed with determina-
tion of IC50 values. Twelve-point titration curves were generated
with concentration of a compound ranging from 511 to 0.5 μM
or from 50 μM to 50 nM, depending on the tightness of binding.
Figure 10. Tertiary (III_C) sublibrary design scheme. No fragment
fusion was involved in the design of sublibrary III_C. Boxed
All kinetic reactions were conducted under steady-state condi-
tions with use of 200 μM D-fructose, 100 μM ATP, and 2 nM
structures indicate scaffolds with at least one tertiary library repre-
ketohexokinase at 25 ꢀC. The reaction time was 60 min, and
sentative testing below 5 μM in the ketohexokinase enzyme assay.
initial velocity was measured. The assay was carried out in 384-
well plate format and referenced to an internal standard com-
pound included in every plate as a control.
A 50 mL three-necked round-bottom flask was purged and
maintained with an inert atmosphere of nitrogen. A solution of
3-(methylthio)-1-phenyl-1H-indazol-6-amine (500 mg, 1.96 mmol,
1.00 equiv) in N,N-dimethylformamide (5 mL), triethylamine
(0.5 mL), piperidine-1,4-dicarboxylic acid mono-tert-butyl ester
(536 mg, 2.34 mmol, 1.19 equiv), and HATU (2.235 g, 5.88 mmol,
3.00 equiv) was added to the flask. The resulting solution was
stirred for 2 h at 15 ꢀC. The solids were filtered out. The resulting
mixture was concentrated under vacuum. The residue was ap-
plied onto a silica gel column with ethyl acetate/petroleum ether
(1:5 to 1:3). This step resulted in 705 mg (77%) of 4-(3-methyl-
sulfanyl-1-phenyl-1H-indazol-6-ylcarbamoyl)piperidine-1-carbo-
xylic acid tert-butyl ester as a white solid.
Rat Pharmacokinetics. Male SPF Sprague-Dawley rats, each
weighing ∼250 g, were used in pharmacokinetics (PK) studies.
Compound 6 as HCl salt was dissolved in 20% HPBCD for total
dissolution, and the solution was adjusted to pH 4.5 with
NaOH. Before dosing, the formulations were stored at room
temperature, protected from light, and analyzed quantitatively;
the stability of the formulations was checked at the day of
dosing. Oral dose (10 mg/kg) administration was by gastric
intubation, and blood samples were collected at 0.5, 1, 2, 4, 8,
and 24 h after administration. Intravenous dose administration
(2.0 mg/kg) was in the jugular vein, and blood samples were col-
lected at 7 and 20 min and at 1, 2, 4, 8, and 24 h after administration.

Article
Journal of Medicinal Chemistry, 2010, Vol. 53, No. 22 7989
In Vitro Kinetic Solubility Assay.34,35 In a typical kinetic
solubility assay, solid material was separated from the liquid
phase after an incubation period by filtration or centrifugation.
The supernatant was analyzed for compound in solution by a
UV plate reader and/or HPLC analysis. Aliquots of DMSO
stock solution (8 μL of 5 mM DMSO solution) were diluted into
buffer solutions (0.01 N HCl; buffer at pH 4.0; buffer at pH 7.4)
to a final DMSO concentration of 2% (representing a maximum
possible solubility of 100 μM) and shaken for 4 h at room
temperature. Insoluble material was removed by centrifugation
at 3700 rpm for 10 min, and the concentration of compound in
the supernatant was measured by UPLC/UV using a three-point
reference calibration prepared by diluting the original DMSO
stock solution.
Caco-2 Permeability Assay.36,37 These studies employed
Caco-2 cells after a 21 day cell culture in 24-well Transwell
plates. Test and reference compounds (propranolol and
vinblastine) were prepared in Hanks’ balanced salt solution
containing 25 mM HEPES (pH 7.4) and added to either the
apical or basolateral chambers of the Transwell plate assembly
at 10 μM. Lucifer Yellow (LY) was added to the donor buffer in
all wells to assess integrity of the cell layers by monitoring LY
permeation. As LY could not freely permeate lipophilic barriers,
a high degree of LY transport indicates poor integrity of the cell
layer. After 1 h of incubation at 37 ꢀC, aliquots were taken from
both apical (A) and basal (B) chambers and added to acetonitrile
containing analytical internal standard (carbamazepine) in a 96-
well plate. Concentrations of compound in the samples were
measured by high performance liquid chromatography/mass
Figure 11. Tertiary (III_D) sublibrary design scheme. Although
numerous fused ring analogues were designed for sublibrary III_D,
the internal pyridine ring was held constant in order to maintain the
hydrogen bond to conserved water.
spectrometry (LC-MS/MS). Apparent permeability (Papp
)
values are calculated as given in references 36 and 37. The
efflux ratios, as an indication of active efflux from the apical
cell surface, were calculated using the ratio of Papp(BfA)/
P
app(AfB).
P450 Inhibition Assays.38,39 Compounds were incubated in
human liver microsomes over a concentration range appropri-
ate for characterizing the IC50 for inhibition of the probe
substrate by the test inhibitor. Positive control incubations in
which a reference inhibitor was used were also included for each
individual assay. The protein concentration, incubation time,
and probe substrate concentration for each isoform/substrate
pairing could vary. After a defined incubation time, the reaction
was stopped and centrifuged and the supernatant prepared for
analysis typically by LC/MS or fluorescence measurements. An
appropriate internal standard could also be added. The mea-
sured concentration of the metabolite formed by the probe
substrate was plotted against inhibitor concentration and an
IC50 calculated by curve fitting.
Cloning and Expression.40 Human ketohexokinase (EC
2.7.1.3) construct 6H.Tb.ketohexokinase 5-298 was cloned
into pET28s and transformed into BL21(DE3) cells. Cells
were grown in LB supplemented with 0.5% glucose and
kanamycin (50 μg/mL) until the OD (A600) reached 0.8. The
cultures were chilled to 15 ꢀC on ice and induced with 1 mM
IPTG and grown for an additional 16 h at 15 ꢀC.
Figure 12. Evolution of hit compound 6. Compared here are the
˚
primary library pyrazole (compound 3) at 2.3 A resolution and the
˚
tertiary library indazole lead compound (compound 6) at 2.4 A
resolution.
Purification41 and Crystallization. All manipulations were
performed at 4 ꢀC. Bacterial cell pellet was thawed, suspended
in lysis buffer [25 mM Tris-HCl, pH 8.0, 500 mM NaCl, 10 mM
imidazole, 10 mM β-mercaptoethanol, 1ꢁ complete protease
inhibitor cocktail (Roche)], and lysed by sonication. Crude cell
lysate was cleared by centrifugation at 1000g for 20 min. Cell
lysate was further clarified by centrifugation at 100000g for
60 min. Supernatant containing 6His-tagged ketohexokinase
was applied on 10 mL Ni-NTA agarose column (Qiagen)
equilibrated in lysis buffer. The column was washed with 10
column volumes of lysis buffer and developed by 10 column
volumes of lysis buffer supplemented with 250 mM imidazole.
Protein-containing fractions were pooled and concentrated to
8 mg/mL using a centrifugal filtering device (Millipore, 10 kDa
molecular-weight cutoffs). Protein homogeneity and identity
Complete plasma profiles were taken from four different animals.
A limited PK analysis was performed using WinNonlintm
Professional (version 3.3, Pharsight, Mountain View, CA).
Analysis was carried out on the plasma concentration time
profiles obtained from each animal (n=4). Oral PK parameters
observed were maximum plasma concentration (Cmax), time
to reach the maximum plasma concentration (Tmax), plasma
half-life (t1/2), and exposure of the compound calculated by
the area under the curve (AUClast and AUCinf). PK param-
eters after iv administration were volume of distribution,
total plasma clearance, and area under the curve (AUClast
and AUCinf).

7990 Journal of Medicinal Chemistry, 2010, Vol. 53, No. 22
Scheme 1. Synthesis of Compound 6a
Gibbs et al.
a (a) N,N0-Carbonyldiimidazole, 90%; (b) PhNHNH2, 38%; (C) NaNO2, HCl, 53%; (d) P2S5, 19%; (e) Cs2CO3, MeI, 86%; (f) H2, Pd/C, 89%; (g)
piperidine-1,4-dicarboxylic acid mono-tert-butyl ester, HATU, 77%; (h) HCl, 52%.
were confirmed by SDS-PAGE (purity >98%) and mass
spectrometry (34 078 Da). For crystallography, purified keto-
hexokinase was dialyzed overnight against 25 mM Tris, pH 8.0,
250 mM NaCl using Pierce Slide-A-Lyzer dialysis cassette with
10 kDa molecular-weight cutoff. Ketohexokinase crystallized at
295 K from a solution containing 17% PEG 8K, 100 mM Na
citrate, pH 4.5, and 0.2 M ammonium sulfate.29
synthesis were allowed to flexibly dock into the receptors using
standard precision (SP) or extra precision (XP) Glide. Glide-
Score, the default internal docking scoring function in Glide,
evaluated the interaction energies of fragments with the
Coulombic and van der Waals grids. The best scoring poses
were minimized using the OPLS-AA force field51 on precom-
puted van der Waals and electrostatic grids.
Crystal Soaking, Data Collection, and Structure Determina-
tion. Fragment plating was performed by first solubilizing each
fragment cocktail in methanol to 20 mM/fragment. Once solu-
bilized 6.5 μL of each fragment cocktail was pipetted into the
well of an Imp@ct plate (Hampton Research) and allowed to
evaporate at room temperature. Fragment cocktails contained a
maximum of five fragments in each well. The fragment cock-
tails in each well were resuspended in 6.5 μL of 24% PEG 8K,
100 mM Na citrate, pH 4.5, and 0.2 M ammonium sulfate.
Crystals were soaked in the fragment cocktails overnight. The
next day the crystals were transferred to a cryoprotectant
solution containing 24% PEG 8K, 100 mM Na citrate, pH
4.5, 0.2 M ammonium sulfate, and 15% glycerol, mounted, and
quickly frozen by immersion in liquid nitrogen.
Acknowledgment. The authors thank Keli Dzordzorme,
Shariff Bayoumy, Robyn William, Alexander Barnakov,
Ludmila Barnakova, and Cynthia Milligan for ketohexoki-
nase cloning, expression, purification, and crystallization. The
authors also thank the ADME, Analytical Research, and
Lead Generation Biology teams for bioanalysis support.
The use of the IMCA-CAT beamlines 17-ID and 17-BM at
the Advanced Photon Source is supported by the companies
of the Industrial Macromolecular Crystallography Associa-
tion through a contract with Hauptman-Woodward Medical
Research Institute. Use of the Advanced Photon Source
is supported by the U.S. Department of Energy, Office of
Science, Office of Basic Energy Sciences, under Contract No.
W-31-109-Eng-38.
X-ray diffraction data for the various fragment structures
˚
were collected with resolutions between 2.3 and 3.0 A at the
IMCA-CAT beamlines ID-17 and BM-17 at the Argonne Na-
tional Laboratory. Diffraction data were indexed, integrated,
and scaled using d*trek.42 The crystals belong to the P212121
space group, with two ketohexokinase molecules in the asym-
metric unit. The unit cell parameters for selected structure
representatives are listed in Table 1. The structure was deter-
mined by molecular replacement using Auto-MR from the
PHENIX suite.43 One ketohexokinase molecule from the tetra-
meric structure of ketohexokinase-A (PDB code 2HLZ)
was used as the search model.44 Model building was performed
using Coot.45 Ligand building was performed using eLBOW.43
Refinement and map calculations were carried out using
PHENIX.43 Refinement statistics for structure representatives
are listed in Table 1. The atomic coordinates and structure
factors for the ketohexokinase complexes with compounds 1-6
have been deposited in the Protein Data Bank under accession
codes 3NBV, 3NBW, 3NC2, 3NCA, and 3NC9, respectively.46
Computational Fragment Docking. Various in-house ketohexo-
kinase structures were used as protein coordinates in Glide
docking studies. All fragment 3D coordinates were generated
with an in-house implementation of a stochastic proximity
embedding (SPE) algorithm,47 followed by minimization using
the MMFF94s force field.48 Explicit hydrogens were added to
the proteins using the “all-atom with no lone pair treatment”
followed by restrained minimization, as implemented in PPREP
utility of Glide.49,50 Various Coulombic and van der Waals grids
were generated with and without hydrogen bond constraints to
the conserved active site water. Fragments considered for
Supporting Information Available: List of structures and
substructures used to filter compounds to produce candidates
for the primary library. This material is available free of charge
via the Internet at http://pubs.acs.org.
References
(1) Hajduk, P. J.; Greer, J. A decade of fragment-based drug design:
strategic advances and lessons learned. Nat. Rev. Drug Discovery
2007, 6, 211–219.
(2) Warr, W. A. Fragment-based drug discovery. J. Comput.-Aided
Mol. Des. 2009, 23, 453–458.
(3) Congreve, M.; Chessari, G.; Tisi, D.; Woodhead, A. J. Recent
developments in fragment-based drug discovery. J. Med. Chem.
2008, 51, 3661–3680.
(4) Hajduk, P. J. Puzzling through fragment-based drug design. Nat.
Chem. Biol. 2006, 2, 658–659.
(5) Murray, C. W.; Callaghan, O.; Chessari, G.; Cleasby, A.; Congreve,
M.; Frederickson, M.; Hartshorn, M. J.; McMenamin, R.; Patel, S.;
Wallis, N. Application of fragment screening by X-ray crystallography
to β-secretase. J. Med. Chem. 2007, 50, 1116–1123.
(6) Congreve, M.; Aharony, D.; Albert, J.; Callaghan, O.; Campbell,
J.; Carr, R. A. E.; Chessari, G.; Cowan, S.; Edwards, P. D.;
Frederickson, M.; McMenamin, R.; Murray, C. W.; Patel, S.;
Wallis, N. Application of fragment screening by X-ray crystal-
lography to the discovery of aminopyridines as inhibitors of
β-secretase. J. Med. Chem. 2007, 50, 1124–1132.
(7) Murray, C. W.; Callaghan, O.; Chessari, G.; Cleasby, A.; Con-
greve, M.; Frederickson, M.; Hartshorn, M. J.; McMenamin, R.;
Patel, S.; Wallis, N. Application of fragment screening by X-ray

Article
crystallography to beta-secretase. J. Med. Chem. 2007, 50, 1116–
Journal of Medicinal Chemistry, 2010, Vol. 53, No. 22 7991
(30) Sigrell, J. A.; Cameron, A. D.; Jones, T. A.; Mowbray, S. L.
Structure of Escherichia coli ribokinase in complex with ribose
1123.
˚
and dinucleotide determined to 1.8 A resolution: insights into a new
(8) Nienaber, V. L.; Richardson, P. L.; Klighofer, V.; Bouska, J. J.;
Giranda, V. L.; Greer, J. Discovering novel ligands for macro-
molecules using X-ray crystallographic screening. Nat. Biotechnol.
2000, 18, 1105–1108.
family of kinase structures. Structure 1998, 6, 183–193.
(31) Mathews, I. I.; Erion, M. D.; Ealick, S. E. Structure of human
˚
adenosine kinase at 1.5 A resolution. Biochemistry 1998, 37, 15607–
(9) Jhoti, H.; Cleasby, A.; Verdonk, M.; Williams, G. Fragment-based
screening using X-ray crystallography and NMR spectroscopy.
Curr. Opin. Chem. Biol. 2007, 11, 485–493.
(10) Poulsen, S.-A.; Kruppa, G. H. In situ fragment-based medicinal
chemistry: screening by mass spectrometry. Fragment-Based Drug
Discovery 2008, 159–198.
(11) Ciulli, A.; Williams, G.; Smith, A. G.; Blundell, T. L.; Abell, C.
Probing hot spots at protein-ligand binding sites: a fragment-
based approach using biophysical methods. J. Med. Chem. 2006,
49, 4992–5000.
15620.
(32) Sigrell, J. A.; Cameron, A. D.; Jones, T. A.; Mowbray, S. L.
Purification, characterization, and crystallization of Escherichia
coli ribokinase. Protein Sci. 1997, 6, 2474–2476.
(33) Symyx. Comprehensive Medicinal Chemistry Database, CMC.
http://www.symyx.com/products/databases/bioactivity/cmc/
index.jsp.
(34) Alsenz, J.; Kansy, M. High throughput solubility measurement in
drug discovery and development. Adv. Drug Delivery Rev. 2007, 59,
546–567.
(12) Neumann, T.; Junker, H.; Schmidt, K.; Sekul, R. SPR-based
fragment screening: advantages and applications. Curr. Top. Med.
Chem. 2007, 7, 1630–1642.
(13) Miura, T. Fragment screening using surface plasmon resonance
optical biosensor technology. Yakugaku Zasshi 2010, 130, 341–348.
(14) Kuntz, I. D.; Chen, K.; Sharp, K. A.; Kollman, P. A. The maximal
affinity of ligands. Proc. Natl. Acad. Sci. U.S.A. 1999, 96, 9997.
(15) Bembenek, S. D.; Tounge, B. A.; Reynolds, C. H. Ligand efficiency
and fragment-based drug discovery. Drug Discovery Today 2009,
14, 278–283.
(16) Chessari, G.; Woodhead, A. J. From fragment to clinical candi-
date;a historical perspective. Drug Discovery Today 2009, 14,
668–675.
(17) de Kloe, G. E.; Bailey, D.; Leurs, R.; de Esch, I. J. Transforming
fragments into candidates: small becomes big in medicinal chem-
istry. Drug Discovery Today 2009, 14, 630–646.
(18) Raushel, F. M.; Cleland, W. W. Bovine liver fructokinase: puri-
fication and kinetic properties. Biochemistry 1977, 16, 2169–2175.
(19) Gross, L. S.; Li, L.; Ford, E. S.; Liu, S. Increased consumption of
refined carbohydrates and the epidemic of type 2 diabetes in the
United States: an ecologic assessment. Am. J. Clin. Nutr. 2004, 79,
774–779.
(20) Basciano, H.; Federico, L.; Adeli, K. Fructose, insulin resistance,
and metabolic dyslipidemia. Nutr. Metab. 2005, 2, 5.
(21) Elliott, S. S.; Keim, N. L.; Stern, J. S.; Teff, K.; Havel, P. J.
Fructose, weight gain, and the insulin resistance syndrome. Am. J.
Clin. Nutr. 2002, 76, 911–922.
(35) Kerns, E. D.; Di, L.; Carter, G. T. In vitro solubility assays in drug
discovery. Curr. Drug Metab. 2008, 9, 879-885.
(36) Irvine, J.; Takahashi, L.; Lockhart, K.; Cheong, J.; Tolan, J.;
Selick, H.; Grove, J. MDCK cells: a tool for membrane perme-
ability screening. J. Pharm. Sci. 1999, 88, 28–33.
(37) Artursson, P. Epithelial transport of drugs in cell culture. I: A
model for studying the passive diffusion of drugs over intestinal
absorbtive (Caco-2) cells. J. Pharm. Sci. 1990, 79, 476–482.
(38) Walsky, R.; Obach, R. Validated Assays for Human Cytochrome
P450 Activities. Drug Metab. Dispos. 2004, 32, 647–660.
(39) Tucker, G.; Houston, J.; Huang, S. Optimizing drug development:
strategies to assess metabolism/transporter interaction potential;
towards a consensus. Br. J. Clin. Pharmacol. 2001, 52, 107–117.
(40) Bayoumy, S. S.; Dzordzorme, K. C. Unpublished results.
(41) Barnakov, A. N.; Barnakova, L. A. Unpublished results.
(42) Pflugrath, J. W. The finer things in X-ray diffraction data collec-
tion. Acta Crystallogr., Sect. D: Biol. Crystallogr. 1999, 55, 1718–
1725.
ꢀ
(43) Adams, P. D.; Afonine, P. V.; Bunkoczi, G.; Chen, V. B.; Davis,
I. W.; Echols, N.; Headd, J. J.; Hung, L.; Kapral, G. J.; Grosse-
Kunstleve, R. W.; McCoy, A. J.; Moriarty, N. W.; Oeffner, R.;
Read, R. J.; Richardson, D. C.; Richardson, J. S.; Terwilliger,
T. C.; Zwart, P. H. PHENIX: a comprehensive Python-based
system for macromolecular structure solution. Acta Crystallogr.,
Sect. D: Biol. Crystallogr. 2010, 66, 213–221.
(44) Rabeh, W.; Tempel, W.; Nedyalkova, L.; Landry, R.; Arrowsmith,
C.; Edwards, A.; Sundstrom, M.; Weigelt, J.; Bochkarev, A.;
Park, H. Deposited in the Protein Data Bank Research Collabora-
tory for Structural Bioinformatics, Rutgers University, New Brunswick,
NJ (http://www.rcsb.org), PDB entry 2HLZ. Unpublished
results.
(22) Bork, P.; Sander, C.; Valencia, A. Convergent evolution of similar
enzymatic function on different protein folds: the hexokinase,
ribokinase, and galactokinase families of sugar kinases. Protein
Sci. 1993, 2, 31–40.
(23) Renze, B.; James, H. M.; Allan, M.; Conyers, R. A. The purifica-
tion and properties of human liver ketohexokinase. Biochem. J.
1985, 230, 53–60.
(45) Emsley, P.; Cowtan, K. Coot: model-building tools for molecular
graphics. Acta Crystallogr., Sect. D: Biol. Crystallogr 2004, 60,
2126–2132.
(46) Rutgers University, New Brunswick, NJ. Protein Data Bank,
Research Collaboratory for Structural Bioinformatics. http://
www.rcsb.org.
(47) Xu, H.; Izrailev, S.; Agrafiotis, D. K. Conformational sampling
by self-organization. J. Chem. Inf. Comput. Sci. 2003, 43, 1186–
1191.
(48) Halgren, T. A. Merck molecular force field. I. Basis, form, scope,
parameterization, and performance of MMFF94. J. Comput.
Chem. 1996, 17, 490–519.
(49) Friesner, R. A.; Banks, J. L.; Murphy, R. B.; Halgren, T. A.; Klicic,
J. J.; Mainz, D. T.; Repasky, M. P.; Knoll, E. H.; Shelley, M.; Perry,
J. K.; Shaw, D. E.; Francis, P.; Shenkin, P. S. Glide: a new
approach for rapid, accurate docking and scoring. 1. Method
and assessment of docking accuracy. J. Med. Chem. 2004, 47,
1739–1749.
(50) Halgren, T. A.; Murphy, R. B.; Friesner, R. A.; Beard, H. S.; Frye,
L. L.; Pollard, W. T.; Banks, J. L. Glide: a new approach for rapid,
accurate docking and scoring. 2. Enrichment factors in database
screening. J. Med. Chem. 2004, 47, 1750–1759.
(51) Jorgensen, W. L.; Maxwell, D. S.; Tirado-Rives, J. Development
and testing of the OPLS all-atom force field on conformational
energetics and properties of organic liquids. J. Am. Chem. Soc.
1996, 118, 11225–11236.
(24) Muchmore, S. W.; Smith, R. A.; Stewart, A. O.; Cowart, M. D.;
Gomtsyan, A.; Matulenko, M. A.; Yu, H.; Severin, J. M.; Bhagwat,
S. S.; Lee, C.; Kowaluk, E. A.; Jarvis, M. F.; Jakob, C. L. Crystal
structures of human adenosine kinase inhibitor complexes reveal
two distinct binding modes. J. Med. Chem. 2006, 49, 6726–6731.
(25) Tang, L.; Li, M.; Cao, P.; Wang, F.; Chang, W.; Bach, S.;
Reinhardt, J.; Ferandin, Y.; Galons, H.; Wan, Y.; Gray, N.;
Meijer, L.; Jiang, T.; Liang, D. Crystal structure of pyridoxal
kinase in complex with roscovitine and derivatives. J. Biol. Chem.
2005, 280, 31220–31229.
(26) Cheng, G.; Bennett, E. M.; Begley, T. P.; Ealick, S. E. Crystal
structure of 4-amino-5-hydroxymethyl-2-methylpyrimidine phos-
˚
phate kinase from Salmonella typhimurium at 2.3 A resolution.
Structure 2002, 10, 225–235.
(27) Ohshima, N.; Inagaki, E.; Yasuike, K.; Takio, K.; Tahirov, T. H.
Structure of Thermus thermophilus 2-keto-3-deoxygluconate ki-
nase: evidence for recognition of an open chain substrate. J. Mol.
Biol. 2004, 340, 477–489.
(28) Hayward, B. E.; Bonthron, D. T. Structure and alternative splicing
of the ketohexokinase gene. Eur. J. Biochem. 1998, 257, 85–91.
(29) Trinh, C. H.; Asipu, A.; Bonthron, D. T.; Phillips, S. E. V. Struc-
tures of alternatively spliced isoforms of human ketohexokinase.
Acta Crystallogr., Sect. D: Biol. Crystallogr. 2009, 65, 201–211.