Structure-guided design of AMP mimics that inhibit fructose-1,6- bisphosphatase with high affinity and specificity

Article abstract of DOI:10.1021/ja074869u

AMP binding sites are commonly used by nature for allosteric regulation of enzymes controlling the production and metabolism of carbohydrates and lipids. Since many of these enzymes represent potential drug targets for metabolic diseases, efforts were ini

Full text of DOI:10.1021/ja074869u

Published on Web 11/28/2007
Structure-Guided Design of AMP Mimics That Inhibit
Fructose-1,6-bisphosphatase with High Affinity and Specificity
Mark D. Erion,*,† Qun Dang,† M. Rami Reddy,† Srinivas Rao Kasibhatla,†,‡
Jingwei Huang,§ William N. Lipscomb,§ and Paul D. van Poelje†
Contribution from the Departments of Medicinal Chemistry, Biochemistry, and Molecular
Modeling, Metabasis Therapeutics, Inc., 11119 North Torrey Pines Road, La Jolla, California
92037, and Department of Chemistry and Chemical Biology, HarVard UniVersity, Cambridge,
Massachusetts 02138
Received July 2, 2007; E-mail: erion@mbasis.com
Abstract: AMP binding sites are commonly used by nature for allosteric regulation of enzymes controlling
the production and metabolism of carbohydrates and lipids. Since many of these enzymes represent potential
drug targets for metabolic diseases, efforts were initiated to discover AMP mimics that bind to AMP-binding
sites with high affinity and high enzyme specificity. Herein we report the structure-guided design of potent
fructose 1,6-bisphosphatase (FBPase) inhibitors that interact with the AMP binding site on FBPase despite
their structural dissimilarity to AMP. Molecular modeling, free-energy perturbation calculations, X-ray
crystallography, and enzyme kinetic data guided our redesign of AMP, which began by replacing the 5′-
phosphate with a phosphonic acid attached to C8 of the adenine base via a 3-atom spacer. Additional
binding affinity was gained by replacing the ribose with an alkyl group that formed van der Waals interactions
with a hydrophobic region within the AMP binding site and by replacing the purine nitrogens N1 and N3
with carbons to minimize desolvation energy expenditures. The resulting benzimidazole phosphonic acid,
16, inhibited human FBPase (IC50 ) 90 nM) 11-fold more potently than AMP and exhibited high specificity
for the AMP binding site on FBPase. 16 also inhibited FBPase in primary rat hepatocytes and correspondingly
resulted in concentration-dependent inhibition of the gluconeogenesis pathway. Accordingly, these results
suggest that the AMP site of FBPase may represent a potential drug target for reducing the excessive
glucose produced by the gluconeogenesis pathway in patients with type 2 diabetes.
Introduction
acts intracellularly to preserve ATP levels through simultaneous
inhibition of pathways that consume ATP (e.g., gluconeogenesis
Adenosine 5′-monophosphate (AMP) and the related nucleo-
side adenosine are increasingly recognized as important physi-
ological sensors of cellular energy status that modulate con-
sumption and production of energy during times of cellular
stress.1 Both are produced from intracellular ATP stores during
net ATP breakdown to minimize further decreases in ATP.
Significant decreases in cellular ATP levels arising from
increased ATP catabolism (e.g., during exercise) or decreased
ATP production (e.g., as a consequence of oxygen deprivation
or mitochondrial toxicity) lead to disruption of essential cellular
processes and cell death. Adenosine preserves ATP levels
through activation of extracellular adenosine receptors, which
leads to decreased ATP consuming activities (e.g., neuron firing,
muscle contraction) and increased ATP production through
enhanced blood flow and oxygen supply.2 In contrast, AMP
and cholesterol biosynthesis) and activation of pathways that
produce ATP (e.g., glycolysis, glycogenolysis, and free fatty
acid â-oxidation).3
AMP increases intracellular ATP levels in large part by
regulating flux through pathways governing carbohydrate and
lipid biosynthesis and metabolism (Figure 1). AMP acts by
directly or indirectly modulating the activity of the key rate-
controlling enzymes in these pathways. Catalytic activity is
typically affected through the binding of AMP to allosteric sites,
which may therefore represent potential drug targets for
metabolic diseases such as diabetes, hyperlipidemia, and hy-
percholesterolemia. For example, the AMP binding enzymes,
fructose-1,6-bisphosphatase4 (FBPase) and glycogen phospho-
rylase5 (GP), control the flux through the gluconeogenesis and
glycogenolysis pathways, respectively, and therefore represent
potential drug targets for reducing the excessive glucose
† Metabasis Therapeutics.
‡ Present address: Biogen Idec, 5200 Research Place, San Diego, CA
92122.
(3) Hardie, D. G. ReV Endocr. Metab. Disord. 2004, 5, 119-125.
(4) (a) Benkovic, S. J.; deMaine, M. M. AdV. Enzymol. Relat. Areas Mol. Biol.
1982, 53, 45-82. (b) Ke, H. M.; Liang, J. Y.; Zhang, Y. P.; Lipscomb, W.
N. Biochemistry 1991, 30, 4412-4420. (c) Ke, H. M.; Zhang, Y. P.;
Lipscomb, W. N. Proc. Natl. Acad. Sci. U.S.A. 1990, 87, 5243-5247. (d)
Shyur, L.-F.; Aleshin, A. E.; Honzatko, R. B.; Fromm, H. J. J. Biol. Chem.
1996, 271, 33301-33307.
§ Harvard University.
(1) (a) Hardie, D. G. Nature 1994, 370, 599-600. (b) Hardie, D. G.; Hawley,
S. A. Bioessays 2001, 23, 1112-1119. (c) Kahn, B. B.; Alquier, T.; Carling,
D.; Hardie, D. G. Cell Metab. 2005, 1, 15-25. (d) Hardie, D. G.; Hawley,
S. A.; Scott, J. W. J. Physiol. 2006, 574, 7-15.
(2) (a) Schrader, J. Circulation 1990, 81, 389-391. (b) Erion, M. D. Annu.
Rep. Med. Chem. 1993, 28, 295-304.
(5) Barford, D.; Hu, S. H.; Johnson, L. N. J. Mol. Biol. 1991, 218, 233-260.
9
15480
J. AM. CHEM. SOC. 2007, 129, 15480-15490
10.1021/ja074869u CCC: $37.00 © 2007 American Chemical Society

Structure-Guided Design of AMP Mimics
A R T I C L E S
Figure 1. Pathways and enzymes that either affect intracellular AMP levels or are regulated by AMP. AMP is synthesized by either the de noVo purine
synthesis pathway [a], ATP degradation, cAMP hydrolysis, or Ado phosphorylation. AMP is metabolized either by phosphorylation to ATP, deamination to
IMP, or dephosphorylation to Ado. AMP inhibits pathways that consume energy such the purine [a], glucose (gluconeogenesis) [b], cholesterol [c], and fatty
acid [d] biosynthesis pathways. AMP activates pathways that produce energy such as glycolysis [e], glycogenolysis [f], and fatty acid oxidation [g]. AMP
affects these pathways by inhibiting (-) or activating (+) key rate-controlling enzymes either directly through allosteric binding sites or indirectly by first
activating AMP activated protein kinase (AMPK), which in turn affects the activity of the indicated enzymes through protein phosphorylation.
Abbreviations: ACC, acetyl CoA carboxylase; Ado, adenosine; AK, adenosine kinase; AMPDA, AMP deaminase; AMPK, AMP-activated protein kinase;
AS, adenylosuccinate synthetase; CPT1, carnitine palmitoyltransferase-1; FBP, fructose-1,6-bisphosphatase; G-6-P, glucose-6-phosphate; GPAT, glutamine
phosphoribosylpyrophosphate amidotransferase; GP, glycogen phosphorylase; HMR, HMG CoA reductase; 5′NT, 5′-nucleotidase; PDE, phosphodiesterase;
PFK, phosphofructokinase; Pyr, pyruvate; TGs, triglycerides; VLDL, very low-density lipoprotein.
production that accompanies type 2 diabetes.6 Another AMP-
binding enzyme generating considerable interest as a potential
drug target is AMP-activated protein kinase7 (AMPK), which
affects cholesterol, fatty acid, and triglyceride biosynthesis as
well as fatty acid oxidation by phosphorylating the rate-limiting
enzymes in these pathways (e.g., HMG CoA reductase and
acetyl-CoA carboxylase).
Despite numerous discovery programs over the past 20 years
targeting various AMP-binding enzymes, few compounds have
emerged with suitable potency, specificity, and biological
activity. Screening large, diverse compound libraries for AMP
deaminase (AMPDA) inhibitors, FBPase inhibitors,8 GP inhibi-
tors,9 and more recently AMPK activators10 failed to identify
compounds with high affinity for the AMP site. Failure was
attributed to the local environment within these sites, which
unlike most drug binding sites is hydrophilic and highly
dependent on hydrogen bond interactions. Since hydrogen bond
strength is sensitive to both distance and bond angle, compounds
with the desired affinity not only had to possess the comple-
mentary hydrogen bond donor and acceptor groups but also had
to have a structure that properly aligned each within the binding
site. These strict structural requirements were met most readily
by close structural analogs of AMP, which as a compound class
proved to be unsuitable as AMP mimics because of their
inability to achieve the desired cellular activity due to poor
plasma stability and cell penetration properties. Efforts to find
nucleoside analogs that generate nucleoside monophosphates
(NMPs) inside cells capable of functioning as potent and specific
AMP mimics were also unsuccessful. Failure stemmed from
the inability of most nucleoside analogs to generate high
intracellular levels of the NMP either because of poor initial
phosphorylation11 or because the NMP was rapidly phospho-
rylated to the corresponding nucleoside triphosphate. 5-Ami-
noimidazole-4-carboxamide 1-â-D-ribofuranoside (AICA ribo-
side), a nucleoside analog we serendipitously discovered in the
late 1980s to lower glucose levels,12 was one of the rare
exceptions. Unfortunately, the monophosphate (ZMP, Chart 1)
that accumulated inside hepatocytes proved to be a relatively
poor AMP mimic that interacted nonspecifically with numerous
AMP-binding enzymes, including FBPase,13 GP,14 AMPK,15 and
phosphofructokinase (PFK).16
(6) (a) DeFronzo, R. A. Diabetes 1988, 37, 667-687. (b) Magnusson, I.;
Rothman, D. L.; Katz, L. D.; Shulman, R. G.; Shulman, G. I. J. Clin. InVest.
1992, 90, 1323-1327. (c) Pilkis, S. J.; Granner, D. K. Annu. ReV. Physiol.
1992, 54, 885-909.
(7) Hardie, D. G. Annu. ReV. Pharmacol. Toxicol. 2007, 47, 185-210.
(8) (a) Choe, J. Y.; Nelson, S. W.; Arienti, K. L.; Axe, F. U.; Collins, T. L.;
Jones, T. K.; Kimmich, R. D.; Newman, M. J.; Norvell, K.; Ripka, W. C.;
Romano, S. J.; Short, K. M.; Slee, D. H.; Fromm, H. J.; Honzatko, R. B.
J. Biol. Chem. 2003, 278, 51176-51183. (b) Wright, S. W.; et al. J. Med.
Chem. 2002, 45, 3865-3877. (c) Wright, S. W.; Carlo, A. A.; Danley, D.
E.; Hageman, D. L.; Karam, G. A.; Mansour, M. N.; McClure, L. D.; Pandit,
J.; Schulte, G. K.; Treadway, J. L.; Wang, I. K.; Bauer, P. H. Bioorg. Med.
Chem. Lett. 2003, 13, 2055-2058. (d) Wright, S. W.; Hageman, D. L.;
McClure, L. D.; Carlo, A. A.; Treadway, J. L.; Mathiowetz, A. M.; Withka,
J. M.; Bauer, P. H. Bioorg. Med. Chem. Lett. 2001, 11, 17-21. (e) Lai, C.;
Gum, R. J.; Daly, M.; Fry, E. H.; Hutchins, C.; Abad-Zapatero, C.; von
Geldern, T. W. Bioorg. Med. Chem. Lett. 2006, 16, 1807-1810. (f) von
Geldern, T. W.; Lai, C.; Gum, R. J.; Daly, M.; Sun, C.; Fry, E. H.; Abad-
Zapatero, C. Bioorg. Med. Chem. Lett. 2006, 16, 1811-1815.
(9) (a) Kristiansen, M.; Andersen, B.; Iversen, L. F.; Westergaard, N. J. Med.
Chem. 2004, 47, 3537-3545. (b) Hoover, D. J.; Lefkowitz-Snow, S.;
Burgess-Henry, J. L.; Martin, W. H.; Armento, S. J.; Stock, I. A.;
McPherson, R. K.; Genereux, P. E.; Gibbs, E. M.; Treadway, J. L. J. Med.
Chem. 1998, 41, 2934-2938.
(10) Cool, B.; Zinker, B.; Chiou, W.; Kifle, L.; Cao, N.; Perham, M.; Dickinson,
R.; Adler, A.; Gagne, G.; Iyengar, R.; Zhao, G.; Marsh, K.; Kym, P.; Jung,
P.; Camp, H. S.; Frevert, E. Cell Metab. 2006, 3, 403-416.
(11) Yamanaka, G.; Wilson, T.; Innaimo, S.; Bisacchi, G. S.; Egli, P.; Rinehart,
J. K.; Zahler, R.; Colonno, R. J. Antimicrob. Agents Chemother. 1999, 43,
190-193.
(12) Gruber, H. E. US Patent 5,658,889, August 19, 1997.
9
J. AM. CHEM. SOC. VOL. 129, NO. 50, 2007 15481

A R T I C L E S
Chart 1
Erion et al.
X-ray structures of porcine FBPase complexed with AMP5b,c
(1FBP) and human FBPase complexed with AMP21 (1FTA) and
ZMP22 suggested that the decrease in potency was likely due
to the loss of seven hydrogen bonds formed between the
phosphate and residues in the positively charged phosphate
binding site (Figure 2). Accordingly, our first design challenge
was to identify a phosphate surrogate that retained most of the
associated binding affinity but, unlike organophosphates, ex-
hibited good stability in biological fluids.
Given the location and orientation of the hydrogen bond
donors within the phosphate binding site, most of the design
strategies incorporated tetrahedral, negatively charged phosphate
surrogates rather than non-tetrahedral groups such as carboxylic
acids.23 The phosphonic acid group was particularly attractive
as a phosphate surrogate, since it represents a close structural
analog and is resistant to phosphatases. Somewhat surprisingly,
simple phosphonate analogs of AMP (1-2, Chart 1) exhibited
a >2000-fold decrease in inhibitory potency (data not shown).
Free-energy calculations and inhibition studies using the Y113F
mutant21a suggested that the loss in inhibitor potency for 1 was
at least partially due to the loss in the hydrogen bond between
Y113 and the 5′ oxygen.24 Decomposing the relative binding
free-energy difference into van der Waals and electrostatic
energies showed that most of the lost affinity was due to
decreases in electrostatic energy, which was attributed primarily
to less favorable interaction energies with Y113, K112, and
L30.24d
These results prompted a further examination of the FBPase-
AMP complex, which led to the realization that the phosphate
binding site might be accessible from the purine base. The
binding conformation of AMP and associated torsion angles [ø
(glycosidic bond, C4-N9-C1′-O1′) ) -153°; γ (O5′-C5′-C4′-
C3′) ) 64°; â (P-O5′-C5′-C4′) ) -140°] orient the adenine
base such that the C8-H bond points directly toward the
phosphate binding site (Figure 2). Accordingly, our initial goal
was to identify molecular fragments that could link the purine
base with a phosphonic acid in a manner that preserved the
key interactions within both the purine and phosphate binding
sites.
With little hope that NMPs could be found with sufficient
potency and specificity to serve as drug candidates, we turned
our attention to non-nucleoside AMP mimics and used a
structure-based drug design strategy to help guide our discovery
efforts. Reported herein is the design and discovery of the lead
FBPase inhibitor series that demonstrates the potential of AMP
binding sites as drug targets. Following this manuscript is an
article17 that describes our lead optimization efforts that led to
the discovery of MB05032 and the MB05032 prodrug MB06322
(CS-917): a compound that significantly reduces glucose levels
in animal models of type 2 diabetes18,19 as well as in diabetic
patients.20
Results
Design Strategy. AMP binds to FBPase at an allosteric
binding site and either induces a protein conformational change
from the R-state (high catalytic activity) to the T-state (low
catalytic activity) or acts to stabilize the T-state.4 Thus, the
challenge we faced at the outset of our program was to find an
AMP mimic that not only bound to FBPase with high affinity
but also favored the R- to T-protein conformational change. In
addition, results from earlier efforts evaluating AMP analog
specificity suggested that complete discrimination between the
AMP binding site of FBPase and those of other AMP-binding
enzymes would likely require the AMP mimic to be as
structurally dissimilar to AMP as possible. Last, the AMP mimic
also needed to be both metabolically stable and cell penetrable
to reach intracellular levels required for effective inhibition of
FBPase and the gluconeogenesis pathway. With these general
criteria in mind, our design efforts began by focusing on the
5′-phosphate and its interactions with the AMP binding site.
Inhibition studies showed that removal of the phosphate resulted
in a >106-fold decrease in inhibitory potency (data not shown).
Spacer Optimization. Using a computer model developed
from the X-ray structure of the human liver FBPase-ZMP
complex,22a molecular fragments (termed spacer groups) of 1-5
atoms in length were attached to C8 and the phosphorus atom,
with the ribose either removed or allowed to freely move. Low
energy conformations were energy minimized in solvent after
(13) (a) Vincent, M. F.; Marangos, P. J.; Gruber, H. E.; Van den Berghe, G.
Diabetes 1991, 40, 1259-1266. (b) Vincent, M. F.; Erion, M. D.; Gruber,
H. E.; Van den Berghe, G. Diabetologia 1996, 39, 1148-1155.
(14) Longnus, S. L.; Wambolt, R. B.; Parsons, H. L.; Brownsey, R. W.; Allard,
M. F. Am. J. Physiol. Regul. Integr. Comp. Physiol. 2003, 284, R936-
R944.
(21) (a) Gidh-Jain, M.; Zhang, Y.; van Poelje, P. D.; Liang, J. Y.; Huang, S.;
Kim, J.; Elliott, J. T.; Erion, M. D.; Pilkis, S. J.; Raafat el-Maghrabi, M.;
Lipscomb, W. N. J. Biol. Chem. 1994, 269, 27732-27738. (b) Iversen, L.
F.; Brzozowski, M.; Hastrup, S.; Hubbard, R.; Kastrup, J. S.; Larsen, I.
K.; Naerum, L.; Norskov-Lauridsen, L.; Rasmussen, P. B.; Thim, L.;
Wiberg, F. C.; Lundgren, K. Protein Sci. 1997, 6, 971-982.
(22) (a) Lipscomb, W. N. Unpublished results. The ZMP binding conformation
is closely related to the AMP binding conformation found in the human
FBPase (ref 21a) and pig kidney FBPase (1FBP, ref 4b) structures. (b) ref
21b.
(23) Carboxylic acids were used previously in the design of a series of AMP
deaminase inhibitors. High affinity was achieved despite suboptimal
interactions in the phosphate binding site by using transition-state mimicry.
Erion, M. D.; Srinivas, R. K.; Bookser, B. C.; van Poelje, P. D.; Reddy,
M. R.; Gruber, H. E.; Appleman, J. R. J. Am. Chem. Soc. 1999, 121, 308-
319.
(15) van den Berghe, G. PCT Appl. US92/06828. (b) Sullivan, J. E.; Brockle-
hurst, K. J.; Marley, A. E.; Carey, F.; Carling, D.; Beri, R. K. FEBS Lett.
1994, 353, 33-36. (c) Merrill, G. F.; Kurth, E. J.; Hardie, D. G.; Winder,
W. W. Am. J. Physiol. 1997, 273, E1107-E1112.
(16) Vincent, M. F.; Bontemps, F.; Van den Berghe, G. Biochem. J. 1992, 281,
267-272.
(17) Dang, Q.; Kasibhatla, S. R.; Reddy, M. R.; Jiang, T.; Reddy, K. R.; van
Poelje, P. D.; Huang, J.; Lipscomb, W. N.; Erion, M. D. J. Am. Chem.
Soc. 2007, 129, 15491-15502.
(18) Erion, M. D.; van Poelje, P. D.; Dang, Q.; Kasibhatla, S. R.; Potter, S. C.;
Reddy, M. R.; Reddy, K. R.; Jiang, T.; Lipscomb, W. N. Proc. Natl. Acad.
Sci. U.S.A. 2005, 102, 7970-7975.
(24) (a) Erion, M. D.; van Poelje, P. D.; Reddy, M. R. J. Am. Chem. Soc. 2000,
122, 6114-6115. (b) Reddy, M. R.; Erion, M. D. J. Am. Chem. Soc. 2001,
123, 6246-6252. (c) Reddy, M. R.; Erion, M. D. In Free Energy
Calculations in Rational Drug Design; Reddy, M. R., Erion, M. D., Eds.;
Kluwer/Plenum Press: New York, 2001; pp 285-297. (d) Reddy, M. R.;
Erion, M. D. J. Am. Chem. Soc. 2007, 129, 9296-9297.
(19) van Poelje, P. D.; Potter, S. C.; Chandramouli, V. C.; Landau, B. R.; Dang,
Q.; Erion, M. D. Diabetes 2006, 55, 1747-1754.
(20) Triscari, J.; Walker, J.; Feins, K.; Tao, B.; Bruce, S. R. Diabetes 2006, 55
(Suppl 1), 444P.
9
15482 J. AM. CHEM. SOC. VOL. 129, NO. 50, 2007

Structure-Guided Design of AMP Mimics
A R T I C L E S
Figure 2. Stereodiagram of the AMP binding conformation and binding site interactions present in the computer model of the human FBPase-AMP
complex that was generated from the X-ray structure of the FBPase-ZMP complex (see Methods). Dashed lines designate hydrogen bond interactions
between AMP and the denoted FBPase binding site residues.
Table 1. Interaction Energies of 8-Substituted Adenosine Analogs
Table 2. Structure Activity Relationships
cmpd
spacer
X
X
′
R
IC50 (µM)
spacer (C8
f
P)
C8
−
Pa,b (Å)
H2N−
Pa,c (Å)
IEd (kcal/mol)
AMP
ZMP
3
4
5
6
7
8
9
10
11
12
13
14
15
16
1 ( 0.05
12 ( 1.4
>1000a
118 ( 9
>1000b
500 ( 10
∼1000
N(H)CH2
N(H)CH2CH2
N(H)CH2CH2CH2
CH2N(H)CH2
CH2CH2
CH2CH2CH2
CH2CH2CH2CH2
CH2OCH2
2,5-furanyl
C(dO)CH2
C(dN)HCH2
CHdCHCH2 (cis)
CHdCHCH2 (trans)
4.08
5.16
6.14
5.20
4.24
5.31
6.30
4.91
5.18
5.10
4.98
5.05
5.17
6.20
6.96
8.02
6.98
6.42
7.09
8.16
7.05
7.75
6.96
6.85
6.88
7.01
11.8
-35.7
33.5
-9.9
3.4
-43.6
28.8
-38.9
-69.2
10.5
-2.7
-5.6
-19.7
N(H)CH2
N
N
N
N
N
N
N
N
N
N
N
N
CH
CF
N
N
N
N
N
N
N
N
N
N
N
N
ribose
ribose
ribose
H
N(H)CH2CH2
N(H)(CH2)3
N(H)CH2CH2
N(H)CH2
N(H)CH2CH2
CH2CH2CH2
CH2OCH2
2,5-furanyl
2,5-furanyl
2,5-furanyl
2,5-furanyl
2,5-furanyl
2,5-furanyl
chexylethyl
chexylethyl
chexylethyl
chexylethyl
phenethyl
chexylethyl
iBu
neoPentyl
iBu
iBu
97 ( 20
26 ( 1
15 ( 3
5 ( 0
1.2 ( 0.9
2.2 ( 0.2
1.1 ( 0.2
1.6 ( 0.1
0.09 ( 0.1
CH
C-Et
a Through space distance between designated atoms in energy minimized
structure. b Distance is 4.24 Å for AMP complexed to human FBPase.
c Distance is 7.27 Å for AMP complexed to human FBPase. d Interaction
energy determined as described in Methods section.
a Inhibition is 27% at 1 mM. b Inhibition is 33% at 1 mM
spacer with an ether spacer (e.g., 9 vs 10). Subsequent efforts
focused on restricting the conformational freedom25 of the ether
led to the discovery of the 2,5-furanyl spacer (e.g., 12), which
provided an additional ∼10-fold increase in inhibitory potency.
Ribose Replacement. Linking the phosphonate to the adenine
through C8 eliminated the need for the ribose as the link.
Analysis of the X-ray structures of FBPase-AMP complexes
placed the ribosyl moiety of AMP near the solvent interface
and in position to form 2 hydrogen bonds (O5′ and O3′) with
the protein. Free-energy perturbation (FEP) calculations showed
that the hydrogen bonds between the 3′ hydroxyl and R140/
docking in the AMP binding site using a four-stage protocol.
The minimized structures were then used in calculations of
interaction energies and ligand strain energies (Table 1). The
energy minimized complexes were also evaluated graphically
for the ability of these low-energy spacer conformations to
replicate the hydrogen bond networks formed by AMP within
the phosphate and purine binding sites. The results, while
qualitative, suggested that the molecular fragments that most
optimally bridged the space between C8 and P contained 3
contiguous atoms and a length of 4.9-5.3 Å (Table 1).
Synthesis and analysis of the inhibition kinetics of 8-ami-
noadenosine analogs with an alkyl phosphonic acid attached to
the 8-amino group (2- to 4-atom spacers) provided the first
experimental data supporting the concept and the preference
for a 3-atom spacer (Table 2, compounds 3-5). Subsequent
analogs confirmed these findings (Table 2, compounds 7-8)
and led to efforts to improve the inhibition potency, which was
at least 15-fold weaker than AMP but still more potent than
phosphonate 1. Analogs covering over 22 different spacers were
synthesized and evaluated (unpublished data). A modest increase
in potency (∼2-fold) was achieved by replacing the all carbon
Y113 contributed little to the overall binding affinity.24
A
Connolly surface map26 color coded by atomic polarity showed
that a portion of the volume accessible to substituents attached
to N9 is relatively hydrophilic and exposed to solvent (Figure
3). In addition, the map revealed a hydrophobic surface located
along the edge of the adenine base between N9 and N3/C2
composed of side chain atoms from three residues, namely
(25) Chang, C. A.; Chen, W.; Gilson, M. K. Proc. Natl. Acad. Sci. U.S.A. 2007,
104, 1534-1539.
(26) Connolly, M. L. Science 1983, 221, 709-713.
9
J. AM. CHEM. SOC. VOL. 129, NO. 50, 2007 15483

A R T I C L E S
Erion et al.
Scheme 2 a
Figure 3. Connolly surface generated using the structure of the human
FBPase-ZMP complex. ZMP was replaced with 16 in the binding
conformation present in the FBPase-16 model. The surface was generated
using a van der Waals radius of 1.4 Å. The surface is color coded based on
electrostatic potential with surface points colored red indicating positive
charge, green indicating neutral, and blue indicating negative charge.
a Conditions: (a) iPr-CHO, NaBH(OAc)3, ClCH2CH2Cl, 98%; (b) FeCl3,
5-diethylphosphono-2-furaldehyde, DMF, 90 °C, 52%; (c) vinyl-SnBu3,
Pd(PPh3)2Cl2, DMF, 92%; (d) H2, Pd-C, EtOH, 99%; (e) TMSBr, CH2Cl2,
86%.
phosphonate ester in 90% yield. Conversion of the 6-chloro
group to the corresponding amine followed by ester hydrolysis
with TMSBr gave 14 in 54% overall yield. The benzimidazole
16 was prepared from 6-bromo-4-fluoro-3-nitro-1,2-benzene-
diamine (18) (Scheme 2), which was synthesized via a 2,1,3-
benzoselenadiazole intermediate according to a reported pro-
cedure with some modifications.28 Selective reductive amination
of the more reactive 1-amino group with isobutyraldehyde in
near quantitative yield followed by an iron (III) chloride-
promoted cyclization reaction with 5-diethylphosphono-2-fural-
dehyde gave the benzimidazole 19 in 52% yield. The ethyl group
was introduced via a Stille coupling with tributylvinyl tin
followed by hydrogenation. Hydrolysis of the diethyl ester with
TMSBr gave 16 in 78% overall yield from bromide 19.
Free-Energy Calculations. A series of post hoc FEP
calculations29 provided insight into the molecular interactions
contributing to the inhibitory potencies of the lead compounds
(Table 3). Calculations comparing close structural analogs were
performed using a single topology and a total of 479 ps of
molecular dynamics (MD) simulation, whereas calculations
comparing analogs with more substantial structural changes were
performed using the thread method and a total of 632 ps of
MD simulation (see Methods). The results from the FEP
calculations were consistent with the experimental findings and
indicated that two-atom and four-atom spacers were associated
with reduced binding affinities relative to the three-atom spacers,
which typically bridged the space between C8 and P without
making unfavorable contacts and with retention of nearly the
full set of hydrogen bonds formed between AMP and the purine
and phosphate binding sites. In the modeled structure of
FBPase-16, the phosphonate forms six hydrogen bonds,
including a hydrogen bond with Y113. The furanyl spacer is
tilted slightly out of plane relative to the benzimidazole with
its oxygen atom, forming a weak electrostatic contact with the
NH of L30 (4.5 Å). The amino group retains the two hydrogen
Scheme 1. Preparation of 14a
a Reagents and conditions: (a) LDA, THF, -78 °C; ClP(O)(OEt)2, 90%;
(b) ammonia, DMSO-THF, 25 °C, 56%; (c) TMSBr, CH2Cl2, 97%.
M177, L30, and V160. Consistent with our predictions derived
from these structural findings, removal of the ribose led to less
than a 5-fold loss in potency (compound 6 vs 4, Table 2),
whereas replacement of the ribose with a cyclohexylethyl group
afforded comparable potency (compound 8 vs 4). As noted
above, incorporation of the 2,5-furanyl spacer in this series
provides a major boost in potency (compound 12 vs 8),
independent of the alkyl group (compounds 11, 13, and 14).
Purine Base Optimization. FEP calculations and experi-
mental data suggested that the hydrogen bonds identified in the
X-ray structure of the FBPase-AMP complex between FBPase
6
and both the N7 and the NH2 groups contribute significantly
to the overall binding affinity.24 In contrast, neither N1 nor N3
formed a favorable interaction, and both were associated with
significant desolvation costs (0.7 and 1.1 kcal/mol, respectively)
that could be avoided by replacement with CH.24 Synthesis of
the corresponding benzimidazole analog (15) showed no loss
in inhibitory potency but little benefit. Fortunately, synthesis
of benzimidazole analogs with various substituents attached to
these carbons led to further gains in inhibitory potency and to
the lead compound 16, a compound with 24-fold greater potency
than the corresponding purine analog 13 and a potency 11-fold
greater than AMP (Table 2).
Synthesis of Lead Compounds. The purine 14 was ef-
ficiently synthesized from 6-chloro-8-(2-furanyl)-9-neopen-
tylpurine (17) (Scheme 1), which was prepared as previously
described.27 Lithiation of 17 with LDA followed by addition
of diethyl chlorophosphate gave the corresponding diethyl
(28) Pesin, V. G.; D’yachenko, S. A.; Khaletskii, A. M. J. Gen. Chem., USSR
(Eng. Transt.), 1964, 34, 3807-3812.
(29) (a) Zwanzig, R. J. J. Chem. Phys. 1954, 22, 1420-1426. (b) Tembe, B.
L.; McCammon, J. A. Comput. Chem. 1984, 8, 281-283. (c) Pearlman,
D. A. In Free Energy Calculations in Rational Drug Design; Reddy, M.
R., Erion, M. D., Eds.; Kluwer/Plenum Press: New York, 2001; pp 9-35.
(27) Dang, Q.; Brown, B. S.; Erion, M. D. Tetrahedron Lett. 2000, 41, 6559-
6562.
9
15484 J. AM. CHEM. SOC. VOL. 129, NO. 50, 2007

Structure-Guided Design of AMP Mimics
A R T I C L E S
Table 3. Relative Free Energies for AMP Mimics
relative free energies (kcal/mol)
transformation (S1
f
S2)
∆∆Gsol
∆∆Gbind
∆∆G
bind (exp)
AMP f ZMPa,d,f
AMP f 1c,e,f
4 f 6d,g
Adenine f AICARb
O f CH2
Ribose f H
H f cyclo-hexylethyl
-1.5 ( 0.7
1.1 ( 0.5
3.1 ( 0.8
1.3 ( 0.8
1.6 ( 0.7
1.1 ( 0.6
2.3 ( 0.6
1.9 ( 0.6
-1.5 ( 0.5
-0.7 ( 0.7
1.4 ( 0.5
0.9 ( 0.6
1.7 ( 0.9
4.6 ( 0.6
1.6
5.5
0.8
-1.4
-0.2
<-1.5
-2.2
-0.8
-0.4
-1.5
-0.2
-1.9
c
1.2 ( 0.9
6 f 8d,g
-1.7 ( 0.9
-0.5 ( 0.8
-1.8 ( 0.7
-2.8 ( 0.7
-1.1 ( 0.7
-0.8 ( 0.6
-2.0 ( 0.8
-0.6 ( 0.6
-2.2 ( 0.7
3 f 5d,g
N(H)CH2 f N(H)CH2CH2CH2
N(H)CH2 f N(H)CH2CH2
N(H)CH2 f CH2CH2CH2
N(H)CH2CH2 f CH2CH2CH2
CH2CH2CH2 f CH2OCH2
CH2OCH2 f 2,5-furanyl
X,X' ) N f CH
3 f 4d,g
7 f 9e,h
8 f 9e,h
9 f 10e,h
10 f 12d,h
13 f 15e,h
15 f 16e,h
X ) CH f C-F; X' ) CH f C-Et
a Calculation reported in ref 24c. b AICAR ) 5-aminoimidazole-4-carboxamide 1-â-D-ribofuranoside. c Calculation reported in ref 24b. d Double topology.
e Single topology. f Model generated from FBPase-ZMP X-ray structure. g Model generated from FBPase-4 X-ray structure. h Model generated from FBPase-
11 X-ray structure.
6
bonds formed between the NH2 group of AMP and residues
V17 and T31 (Figure 2). Unlike AMP, however, the imidazole
nitrogen of 16 forms only an electrostatic interaction with T31
and not a hydrogen bond, largely because the benzimidazole
undergoes a slight in-plane rotation relative to the purine base
(see X-ray crystallography results).
the use of the thread method30 and long MD simulation times
(632 ps). As shown in Table 3, the ribose moiety is associated
with significant desolvation costs that are neutralized by a few
relatively weak interactions with binding site residues. In
contrast, the cyclohexylethyl is readily desolvated while making
enthalpic gains through van der Waals interactions with the side
chain atoms of several hydrophobic residues. Overall, the results
were consistent with the inhibition potencies reported in Table
2 and the design strategy focused on enhancing inhibitory
potency through replacement of the ribose with hydrophobic
groups.
Relative improvements in inhibitory potency corresponding
to the spacer were analyzed using FEP calculations (Table 3).
Transformation of 8 to 9 (-NHCH2CH2- f -CH2CH2CH2-)
and 10 to 9 (-CH2OCH2- f -CH2CH2CH2-) were associated
with a 1.9 and 1.5 kcal/mol reduction in desolvation costs
(∆∆Gsol ) ∆Gaq - ∆Ggas), respectively. The unfavorable
relative binding free energy associated with the latter transfor-
mation reflects a decrease in enthalpic energy arising from the
loss of the electrostatic interaction with the NH of L30.
Transformation of ether 10 to the 2,5-furanyl analog 12 showed
that the furanyl analog is associated with a greater desolvation
penalty but an overall more favorable relative free energy of
binding (∆∆Gbind ) ∆Gcom - ∆Gaq of -2.0 kcal/mol). A
decrease in the entropic penalty arising from decreased con-
formational freedom presumably accounts for a portion of the
improved binding affinity. The remaining portion is enthalpic
and likely represents stronger interactions within the phosphate
and purine binding sites due to the difference in the C8 to P
distance (Table 1).
FEP calculations were also used to assess purine base
modifications, including the effect of various substituents.
Previous calculations showed that neither N1 nor N3 contributed
significantly to the overall binding affinity of AMP and that
both nitrogens were associated with modest desolvation penal-
ties.24 Consistent with these findings and the experimental results
(Table 2), the relative binding free-energy difference between
the purine analog 13 and benzimidazole analog 15 is -0.6 kcal/
mol. FEP calculations also accurately reflected the favorable
binding interactions gained by replacing the ribose with cyclo-
hexylethyl (4 f 6 f 8). These transformations entailed
relatively large structural perturbations and therefore required
X-ray Structure of FBPase-Inhibitor Complexes. The
three-dimensional structure of the human FBPase-4 complex
was solved to 2.8 Å resolution (W. N. Lipscomb and Y. Zhang,
unpublished data) during the lead optimization phase of the
discovery program. The structure confirmed the computer
modeling predictions that a 3-atom spacer enabled retention of
the key phosphate and purine binding site interactions. The
structure also revealed that the ribosyl group was rotated away
from the site it occupies in the AMP complex and toward the
solvent interface. This change in glycosidic bond torsion is
attributed to the relative absence of strong binding site interac-
tions with the ribose coupled with the preference of C8-
substituted purine nucleosides to exist in the syn-conformation.31
More recently and years after the discovery of 16, the structure
of the FBPase-16 complex was solved to 2.0 Å resolution
(Table 4, Figure 4). Importantly, the binding site interactions
determined by X-ray crystallography (Figure 5) were identical
to those predicted years in advance by computer modeling
(Table 5). Moreover, a comparison of the X-ray structure and
the computer model showed that the binding site interaction
(30) (a) Singh, U. C.; Benkovic, S. J. Proc. Natl. Acad. Sci. U.S.A. 1988, 85,
9519-9523. (b) Reddy, M. R.; Viswanadhan, V. N.; Weinstein, J. N. Proc.
Natl. Acad. Sci. U.S.A. 1991, 88, 10287-10291. (c) Reddy, M. R.; Bacquet,
R. J.; Zichi, D.; Mathews, D. A.; Welsh, K. M.; Jones, T. R.; Freer, S. J.
Am. Chem. Soc. 1992, 114, 10117-10122.
(31) Saenger, W. In Principles of Nucleic Acid Structure; Cantor, C. R., Ed.;
Springer-Verlag: New York, 1984; pp 51-104.
9
J. AM. CHEM. SOC. VOL. 129, NO. 50, 2007 15485

A R T I C L E S
Erion et al.
Table 4. Data Processing and Refinement Statisticsa
Table 5. Binding Site Interaction Distances
hydrogen bond
residue
FBPase
modela
−
16
FBPase
modelc
−
AMP
FBPase−16
ligand
X-rayb
X-rayd
space group
P21
cell dimensions
a ) 67.8 Å; b ) 83.9 Å;
c ) 280.7 Å; â ) 89.9
NH2
NH2
N
O (furanyl)
5′O (ribose)
3′O (ribose)
3′O (ribose)
OP
OP
OP
OP
OP
OP
OP
OdC (V17)
HO (T31)
HO (T31)
HN (L30)
HO (Y113)
HO (Y113)
HNê (R140)
HN (E29)
HN (G28)
HO (T27)
HN (T27)
HNꢀ (K112)
HO (Y113)
HN (L30)
2.61
3.16
4.11
4.32
NA
3.33
3.19
4.88
4.56
NA
2.76
3.07
3.33
NA
2.98
3.23
3.54
NA
data collection and processing
wavelength (Å)
resolution (Å)
0.9124
2.0 (2.07-2.00)
492 209 (16 358)
200 729
92.0 (75.3)
2.5 (2.0)
10.4 (48.3)
10.1 (2.0)
8
2.94
2.94
2.90
3.10
2.80
2.92
3.31
2.72
2.94
2.99
4.27
7.94
2.98
2.53
2.95
2.78
2.89
2.73
2.88
2.91
2.70
3.19
4.24
7.27
reflections
NA
NA
NA
NA
unique reflections
completeness (%)
multiplicity
2.54
2.52
2.87
3.19
2.51
3.29
NA
2.70
3.17
2.58
2.76
2.88
2.56
NA
Rsymb (%)
mean I/σ
number of molecules in
asymmetric unit
refinement statistics
c
Rworking
Rfree
21.3 (32.5)
21.9 (31.3)
22.6
0.0061
1.32
base to P distance
NH2 to P distance
5.28
8.20
5.19
7.40
d
average B-factor (Å2)
R.M.S. bond (Å)
a Distances in angstroms for computer model of human FBPase-16
complex; NA ) not applicable. b Distances in angstroms for X-ray structure
of human FBPase-16 complex (2.0 Å resolution); NA ) not applicable.
c Distances in angstroms for computer model of human FBPase-AMP
complex, which was generated using human FBPase-ZMP structure; NA
) not applicable. d Distances in angstroms for X-ray structure of human
FBPase-AMP complex complex (2.3 Å resolution, ref 21a); NA ) not
applicable.
R.M.S. angle (deg)
number of protein atoms
number of water molecules
19 480
1407
a Values in the parentheses apply to the outer resolution shell. Rsym
)
b
∑
∑i |Ii (hkl) - I (hkl) |/∑hkl ∑i Ii (hkl), where Ii (hkl) is the intensity
hkl
of an individual reflection, and I (hkl) is the mean intensity obtained from
the multiple observations of symmetry-related reflections taken from one
c
crystal. Rworking ) ∑hkl (|Fo(hkl) - Fc (hkl) |)/∑hkl(|Fo (hkl)|), calculated
to the imidazole nitrogen (Table 5). The larger distance between
T31 and the imidazole nitrogen of 16 (4.88 Å) appears related
to the tilting of the benzimidazole base relative to the adenine
of AMP, which presumably is necessary to accommodate the
furanyl spacer and the inherently longer C8 to P distance. The
furanyl spacer of 16 is engaged in a weak electrostatic
interaction with the amide nitrogen of L30, which may at least
partially compensate for the loss of the hydrogen bond to the
imidazole nitrogen. As shown in Figure 6, the ribose of AMP
and the alkyl sidechains of 16 reside in different regions within
the AMP binding site and form favorable interactions with
different binding site residues. Specifically, the ribose is
positioned near the protein-solvent interface with O3′ in some
instances forming a hydrogen bond with the guanidine of R140.
In contrast, the alkyl substituents of 16 reside near a hydrophobic
surface within the AMP binding site and form van der Waals
contacts with the sidechains of M177 (S with 7-ethyl, 3.6 Å),
L30 (Cδ1 with 1-i-butyl, 4.0 Å), and V160 (Cγ1 with 1-i-butyl,
3.4 Å).
d
from reflections in the working set. Rfree) ∑hkl (|Fo(hkl) - Fc (hkl) |)/
∑
hkl(|Fo (hkl)|), calculated from reflections in the test set, which is a random
5% data set selected from the whole data set.
Inhibition Kinetics. Compounds in Table 2 inhibited FBPase
in a concentration-dependent manner. The inhibition kinetic
profiles of 16 and AMP were similar with both exhibiting the
same maximal FBPase inhibition (∼99%) achieved largely over
a narrow concentration range (∼1 log).18 Moreover, like AMP,
16 exhibited noncompetitive inhibition kinetics (data not shown).
FBPase Specificity. The AMP binding site specificity of 16
was assessed by evaluating the effect of 16 on five AMP-binding
enzymes, namely human erythrocytic AMP deaminase, rabbit
muscle glycogen phosphorylase, human AMP-activated protein
kinase, rabbit muscle adenylate kinase, and rabbit liver phos-
phofructokinase. No effects were observed at concentrations
1000-fold above the human liver FBPase IC50 for 16. In addition,
no significant effects were observed in a screen against >70
enzymes, ion channels, and receptors (Panlabs, Bothel, WA) at
10 µM 16.
Figure 4. The 2F0-Fc electron density map is contoured at 1 σ and
superimposed on the refined model of the human FBPase-16 complex at
2.0 Å resolution.
distances and binding conformation of 16 were similar (Table
5). A least-squares superposition of the CR traces revealed only
minor differences in the backbone and binding site side chain
atoms (0.48 Å and 0.93 Å average rms deviations, respectively).
Superimposition of the CR carbons of the FBPase-16
complex onto the FBPase-AMP complex revealed only minor
differences in backbone atoms and binding site side chain atoms
(0.55 and 1.02 Å average rms deviations, respectively) (Figure
6). A comparison of the structures showed that 16 forms the
same interactions with the phosphate and purine binding site
residues as AMP with the exception of the T31 hydrogen bond
Inhibition of Glucose Production. Freshly isolated rat
hepatocytes were used to assess the ability of 16 to penetrate
9
15486 J. AM. CHEM. SOC. VOL. 129, NO. 50, 2007

Structure-Guided Design of AMP Mimics
A R T I C L E S
Figure 5. Stereodiagram derived from the X-ray structure of the human FBPase-16 complex showing the interactions between the AMP binding site of
FBPase and 16. Putative hydrogen bonds are shown as dotted lines when the distance between the acceptor and donor is less than 3.3 Å.
Figure 7. Glucose production from 10 mM lactate/1 mM pyruvate (red
b, IC50 ) 0.38 ( 0.02 µM) and intracellular fructose-1,6-bisphosphate
(FBP) levels (blue 1) in freshly isolated rat hepatocytes incubated for 1 h
with the indicated concentration of 16.
ribose, and purine base. Efforts to reduce the desolvation costs
and entropic penalty associated with the spacer while gaining
affinity through favorable electrostatic interactions led to the
discovery of the 2,5-furanyl spacer. The ribose moiety, which
became expendable with the spacer strategy, was replaced with
less polar alkyl groups capable of forming favorable van der
Waals contacts with a hydrophobic surface within the AMP
binding site. Last, the purine base was replaced with a
substituted benzimidazole. Ultimately, a combination of these
structural changes led to the benzimidazole phosphonic acid 16,
an AMP mimic that inhibits FBPase 11-fold more potently than
AMP.
Figure 6. Superimposition of the CR traces of FBPase from the human
FBPase-16 (carbons colored yellow) and human FBPase-AMP (carbons
colored green) complexes. The average rms deviation was 0.55 Å for CR
and 1.02 Å for the sidechains.
hepatocytes and inhibit glucose production by an FBPase-
dependent mechanism. 16 inhibited glucose production from
lactate/pyruvate (gluconeogenesis) in a concentration-dependent
manner with an IC50 of 2.8 ( 0.5 µM (n ) 2 studies, duplicate
samples) and a maximal inhibition of approximately 100%
relative to vehicle-treated hepatocytes (Figure 7). Inhibition was
associated with an approximate 10-fold increase in intracellular
fructose-1,6-bisphosphate levels.
Evidence that 16 not only bound to the AMP site of FBPase
but also functioned as an AMP mimic was apparent from both
X-ray structural data as well as inhibition kinetics. The structural
data demonstrated that like AMP, 16 was complexed to the
T-state of FBPase. Consistent with the X-ray data, 16 inhibited
FBPase in solution to the same extent as AMP (99%) and like
AMP was associated with a steep concentration-dependent
inhibition curve that exhibited near-complete inhibition of
FBPase at concentrations only 10-fold higher than the IC10
concentration.
Discussion
Our discovery of a series of non-nucleoside AMP mimics
with high inhibitory potencies and specificities for FBPase
benefited from structural information obtained by X-ray crystal-
lography and molecular modeling studies. Central to our success
was the initial recognition that the phosphate binding site was
accessible from the purine base using 3-atom spacers to link
C8 to the PO32-. The inhibitory potency of the initial lead
inhibitor, 4 (IC50 ) 118 µM), was subsequently improved more
than 1000-fold through structural modifications in the spacer,
The high FBPase specificity achieved by 16 presumably arose
from the large structural differences between 16 and AMP
combined with the structural differences between the FBPase
AMP binding site and the binding sites for other AMP-binding
enzymes. A review of X-ray structures of nucleotide binding
9
J. AM. CHEM. SOC. VOL. 129, NO. 50, 2007 15487

A R T I C L E S
Erion et al.
sites32 and more specifically of AMP binding sites suggested
that the origin of this high specificity may arise from the
combined effects of several structural changes. Linking C8 of
the adenine base to PO32- by a 3-atom spacer may be the largest
single contributor to the increased specificity, since its success
depended on the preference of FBPase to bind AMP in the anti,
+sc, -ac conformation31 and correspondingly on the angle and
distance between C8 and the phosphorus atom. AMP binding
sites favoring different AMP binding conformations correspond-
ingly favor different distances and/or angles33b between the base
bridge the space between the purine base and PO32- group in a
manner that simultaneously preserved the favorable interactions
between these groups and their respective biding sites. Modeling
studies and FEP calculations also provided the incentive to
replace the purine base with a benzimidazole and the ribose
with alkyl groups to minimize desolvation costs and gain van
der Waals interactions. Importantly, the binding site interactions
predicted by computer modeling and used as an integral part
of our design process nearly 10 years ago were replicated
experimentally when the structure of the FBPase-16 complex
was solved recently by X-ray crystallography.
As described in the subsequent article,17 this work was
followed by a substantial lead optimization effort that led to
the discovery of MB05032 and an orally bioavailable phospho-
nic diamide prodrug of MB05032, MB06322 (designated as CS-
917). CS-917 was shown to lower glucose in animal models of
mild and severe diabetes.18,19 Moreover, CS-917 resulted in
pronounced glucose lowering in type 2 diabetics20 and may
therefore represent an important new therapy for this growing
and costly disease that afflicts nearly 170 million people
worldwide.
2-
and PO3 and therefore are unlikely to recognize compounds
2-
that link the base and PO3 by the rigid 2,5-furanyl spacer.
Specificity is also attributed to differences in AMP binding site
interactions with the purine and ribose moieties. Unlike many
other nucleotide binding sites, FBPase binds AMP by forming
6
hydrogen bonds with N7 and the NH2 group but not N1 or
N3. In addition, the replacement of the ribose with an alkyl or
an alkylaryl group eliminates binding to AMP binding sites that
require interactions with the ribose for high affinity or are unable
to accommodate hydrophobic groups near N9.
The final step prior to designating 16 as the lead AMP mimic
was demonstration of cellular uptake and inhibition of intrac-
ellular FBPase activity. Phosphonic acids usually fail to enter
cells due to their high negative charge at physiological pH.
Fortunately, the cells that produce glucose via the gluconeo-
genesis pathway, i.e., liver and kidney cells, contain cell surface
transporters known as organic anion transporters that often aid
the uptake and elimination of negatively charged compounds.33
Results from studies with primary rat hepatocytes demonstrated
that 16 is rapidly transported into hepatocytes. Cell penetration
was subsequently confirmed by studies showing that 16 inhibited
the conversion of lactate/pyruvate to glucose in a concentration
dependent manner and that this inhibition was via FBPase based
on the corresponding rise in intracellular levels of fructose-1,6-
bisphosphate.
The discovery of 16 highlights the value of structure-based
drug design, especially in light of the apparent difficulty in
finding suitable AMP mimics by screening large compound
libraries. The X-ray structure of the human FBPase-AMP
complex provided the initial understanding of the AMP binding
site interactions, whereas FEP calculations24 added important
insights into the contribution of each individual interaction to
the overall binding affinity. Molecular modeling helped guide
medicinal chemistry efforts to retain the key binding site
interactions formed by AMP while attempting to exploit other
features within the binding site that could enhance either binding
site affinity or specificity. The early decision to focus on
phosphonic acids stemmed from the recognition that the
phosphate binding site interactions were very important to the
overall binding affinity and that the phosphate binding site was
most compatible with a negatively charged tetrahedral group.
Discovery of the initial lead, 4, was guided by modeling studies
demonstrating that the phosphate binding site was accessible
from the C8 carbon on the adenine base. Subsequent modeling
studies provided additional information that aided in the spacer
optimization and the selection of 3-atom spacers that could
Conclusions
AMP binding sites represent potential drug targets that remain
largely unexplored due to long-standing difficulties in finding
potent, selective, and cell penetrable compounds. FBPase is one
such AMP-binding enzyme that has received considerable
attention based on its potential as a drug target for type 2
diabetes, but without success. To address this drug discovery
challenge, we used several X-ray structures of FBPase-inhibitor
complexes and numerous computer modeling studies to guide
the design of a series of AMP mimics that bear little structural
resemblance to AMP. The lead compound, 16, was shown to
inhibit FBPase 11-fold more potently than AMP and to exhibit
high specificity for FBPase. Moreover, 16 was shown to undergo
rapid uptake by rat hepatocytes, which resulted in potent
inhibition of FBPase and glucose production via the gluconeo-
genesis pathway. These results suggest that AMP mimics with
suitable drug-like properties are achievable and therefore that
AMP-binding sites represent potential drug targets worthy of
further exploration.
Experimental Procedures
General Methods. All moisture-sensitive reactions were performed
under a nitrogen atmosphere using oven-dried glassware and anhydrous
solvents purchased and used without further manipulation. TLC was
performed on Merck silica gel 60 F254 TLC glass plates and visualized
with UV light or a suitable stain. Flash chromatography was performed
on 230-400 mesh EM Science silica gel 60. Melting points were
recorded on a Thomas-Hoover capillary melting point apparatus and
are uncorrected. 1H, 13C, and 31P NMR spectra were recorded on Varian
Gemini or Mercury spectrometers operating respectively at 200 or 300
MHz for proton, 50 or 75 MHz for carbon, and 121 MHz for
phosphorus spectra. Analytical HPLC was performed on a 4.6 × 250
mm YMC ODS-AQ 5 µm column eluting with a 0.1% aqueous AcOH/
MeOH gradient at a flow rate of 1 mL/min and the detector set at 280
nm. Mass spectrometry data were determined at Mass Consortium Corp.
(San Diego, CA) or on a Perkin-Elmer Sciex API2000 LC-MS system.
Elemental microanalyses were performed by NuMega Resonance Labs,
Inc. (San Diego, CA) or by Robertson Microlit Laboratories (Madison,
NJ).
(32) (a) Moodie, S. L.; Mitchell, J. B.; Thornton, J. M. J. Mol. Biol. 1996, 263,
486-500. (b) Moodie, S. L.; Thornton, J. M. Nucleic Acids Res. 1993, 21,
1369-1380.
(33) van Montfoort, J. E.; Hagenbuch, B.; Groothuis, G. M.; Koepsell, H.; Meier,
P. J.; Meijer, D. K. Curr. Drug Metabol. 2003, 4, 185-211.
[5-(7-Bromo-5-fluoro-1-isobutyl-4-nitro-1H-benzimidazol-2-yl)-
furan-2-yl]-phosphonic Acid Diethyl Ester (19). A solution of
9
15488 J. AM. CHEM. SOC. VOL. 129, NO. 50, 2007

Structure-Guided Design of AMP Mimics
A R T I C L E S
6-bromo-4-fluoro-3-nitrobenzene-1,2-diamine (18, 94.0 g, 375 mmol)
in 1,2-dichloroethane (950 mL) at 0 °C was treated with isobutyral-
dehyde (27.1 g, 375 mmol) followed by sodium triacetoxyborohydride
(159 g, 750 mmol) and then dropwise with glacial acetic acid (84.0
mL, 1460 mmol) over 15 min while maintaining the temperature
between 0 and 5 °C.34 After stirring at 0 °C for an additional hour, the
resulting mixture was quenched with a saturated NaHCO3 solution and
extracted with EtOAc (3 × 750 mL). The combined organic extracts
were washed with water (2 × 1 L) and brine (750 mL), dried (MgSO4),
filtered, and concentrated under reduced pressure. The resulting residue
was coevaporated with toluene (750 mL) to afford a brown oil (113 g,
98%). 1H NMR (DMSO-d6): δ 0.93 (d, 6H, J ) 6.6 Hz), 1.8 (m, 1H),
2.53 (m, 2H), 3.87 (br s, 1H, D2O exchangeable), 6.78 (br s, 2H, D2O
exchangeable), 6.88 (d, 1H, J ) 11.6 Hz). The above material (90.8 g,
296 mmol) was dissolved in DMF (500 mL) and treated with
5-diethylphosphono-2-furaldehyde (75.6 g, 325 mmol). Air was bubbled
through the solution while iron (III) chloride (52.8 g, 325 mmol) was
added portionwise.35 Air bubbling was continued while the resulting
dark mixture was stirred and heated at 90 °C. After 2 h, the mixture
was cooled to room temperature and the solvent was evaporated under
reduced pressure. The dark oil was coevaporated with toluene (3 ×
200 mL), and the residue was dissolved in EtOAc (3.0 L). The EtOAc
solution was washed with water (3 × 600 mL) and brine (600 mL)
and then dried (MgSO4), filtered, and concentrated under reduced
pressure to provide a dark tar. The crude material was purified by
chromatography on silica gel (2.0 kg, 6 in. diameter column) using
40-60% EtOAc: hexanes as eluting solvents to give 19 as a brown
solid (80.0 g, 52%). 1H NMR (DMSO-d6): δ 0.8 (d, 6H, J ) 6.6 Hz),
1.26 (t, 3H, J ) 7.0 Hz), 2.1 (m, 1H), 4.12 (m, 4H), 4.72 (d, 2H, J )
7.8 Hz), 7.48 (m, 2H), 7.93 (d, 1H, J ) 11.0 Hz).
concentrated under reduced pressure to give an orange syrup (56 g,
100%), which was used in the next step without further purification.
1H NMR (DMSO-d6): δ 0.73 (d, 6H, J ) 6.6 Hz), 1.05-1.35 (m,
6H), 1.85 (m, 1H), 2.82 (q, 2H, J ) 7.0 Hz), 4.1 (m, 4H), 4.38 (d, 2H,
J ) 7.8 Hz), 5.18 (br s, 2H, exchangeable with D2O), 6.8 (d, 1H, J )
11.5 Hz), 7.2 (m, 1H), 7.4 (m, 1H).
A solution of the above material in dichloromethane (500 mL) was
cooled to 0 °C and slowly treated with bromotrimethylsilane (112 mL,
848 mmol). After warming to room temperature and stirring for 16 h,
the solvent was evaporated under reduced pressure to afford a residue
that was subsequently coevaporated with acetone (200 mL). Water (800
mL) and acetone (150 mL) were added to the resulting thick, orange
tar, and the suspension was stirred vigorously. After 2 h, the suspension
was filtered, and the solid was collected, washed with water (3 × 150
mL), and dried under vacuum to give crude 16 as a fine, yellow powder
1
(48.1 g, 103%). H NMR (DMSO-d6): δ 0.69 (d, 6H, J ) 6.6 Hz),
1.21 (t, 3H, J ) 7.2 Hz), 1.8 (m, 1H), 2.82 (q, 2H, J ) 7.2 Hz), 4.38
(d, 2H, J ) 7.8 Hz), 6.83 (d, 1H, J ) 12.8 Hz), 7.0 (m, 1H), 7.1 (m,
1H).
Crude 16 (114.7 g, obtained by combining several batches) was
treated with aqueous sodium hydroxide (1 M, 570 mL, 570 mmol).
After stirring at room temperature for 2 h, the cloudy, dark-orange
solution was extracted with EtOAc (2 × 350 mL). The dark-orange
aqueous phase was then diluted with methanol (350 mL) and treated
with activated carbon (Norit SA-3, 4.0 g). The mixture was filtered
twice through Celite and once through filter paper. The filtrate (1.4 L
total volume) was treated with concentrated hydrochloric acid (48 mL,
570 mmol) dropwise with vigorous stirring. A persistent yellow
precipitate formed during the second half of the addition. After the
addition was complete (45 min), the mixture was stirred at room
temperature for 2 h and then at 0 °C for 1 h. The solid was collected
by filtration and washed with MeOH-water (60:40, 100 mL) and then
water (3 × 200 mL). The solid was dried under vacuum to give 16 as
[5-(5-Fluoro-1-isobutyl-4-nitro-7-vinyl-1H-benzimidazol-2-yl)fu-
ran-2-yl]-phosphonic Acid Diethyl Ester (20). A solution of 19 (80.0
g, 154 mmol) in DMF (650 mL) was treated with dichloro bis-
(triphenylphosphine)palladium(II) (4.5 g, 6.4 mmol). Nitrogen was
bubbled through the stirred mixture while tributylvinyl tin (51.5 mL,
176 mmol) was slowly added. After the addition was complete (5 min),
the mixture was warmed to 70 °C and stirred for 1 h with continued
nitrogen purge. The mixture was cooled to room temperature and
concentrated under reduced pressure. The resulting dark oil was
dissolved in EtOAc (2.1 L) and treated with a mixture of sodium
fluoride (113 g, 2690 mmol) in water (475 mL). After stirring for 16
h, the mixture was diluted with water (350 mL) and filtered through a
pad of Celite in a Buchner funnel (11 cm). The layers were separated,
and the aqueous phase was extracted with EtOAc (800 mL). The
combined organic extracts was washed with water (600 mL) and brine
(600 mL), dried (MgSO4, 125 g), filtered, and concentrated under
reduced pressure to give a dark oil (88.6 g). The crude material was
purified by chromatography on silica gel (1.75 kg, 6 in. diameter
column) using 3:1 EtOAc-hexanes (16 L) as the eluting solvents to
1
a fine, yellow-orange powder (93.4 g, 86%). mp >220 °C. H NMR
(DMSO-d6): δ 0.68 (d, 6H, J ) 6.6 Hz), 1.21 (t, 3H, J ) 7.2 Hz),
1.7-2.0 (m, 1H), 2.83 (q, 2H, J ) 7.2 Hz), 4.37 (d, 2H, J ) 7.4 Hz),
6.83 (d, 1H, J ) 12.8 Hz), 7.02 (m, 1H), 7.12 (m, 1H). 13C NMR
(DMSO-d6): δ 15.79, 19.25, 23.59, 30.47, 52.08, 112.66, 112.89, 113.1,
114.86, 115.00, 119.64, 120.10, 124.43, 124.73, 129.77, 133.49, 133.67,
143.15, 143.32, 147.65, 147.81, 148.02, 149.16, 153.66. [MH]+ calcd
for C17H21N3O4PF: 382. Found: 382. Anal. calcd for C17H21N3O4PF:
C, 53.54; H, 5.55; N, 11.02; F, 4.98. Found: C, 53.27; H, 5.47; N,
10.92; F, 4.92.
Computational Details. All molecular dynamics, molecular me-
chanics, and free-energy perturbation calculations were carried out with
the AMBER program using an all atom force field36 and the SPC/E
model potential37 to describe water interactions. Procedures for generat-
ing the computer model of the FBPase complexes and for developing
all force field parameters for nonstandard residues are as described
previously and detailed in the Supporting Information.
1
give 20 as a pale-orange powder (66.7 g, 92%). mp 77-78.5 °C. H
NMR (DMSO-d6): δ 0.73 (d, 6H, J ) 6.6 Hz), 1.25 (t, 3H, J ) 7.0
Hz), 1.9 (m, 1H), 4.1 (m, 4H), 4.58 (d, 2H, J ) 7.8 Hz), 5.7 (d, 1H,
J ) 9.4 Hz), 6.0 (d, 1H, J ) 11.5 Hz), 7.3-7.6 (series of m, 4H).
Spacer groups were evaluated computationally starting with the three-
dimensional structure of the human FBPase-ZMP complex22a and later
complexes of FBPase with 4 and 11. The ZMP complex was used to
generate a computer model of the FBPase-AMP complex.24 AMP was
replaced with analogs denoted in Table 1 by positioning the phosphonate
in the phosphate binding site in a manner that retained the full
complement of hydrogen bonds. Low-energy spacer conformations were
selected that positioned the adenine base and the ribose in the general
vicinity of where they bound in the AMP complex. The complex was
[5-(4-Amino-7-ethyl-5-fluoro-1-isobutyl-1H-benzoimidazol-2-yl)-
furan-2-yl]-phosphonic Acid (16). A solution of 20 (57.1 g, 122 mmol)
in ethanol (425 mL) was treated with palladium on carbon (4.5 g) under
a continuous flow of nitrogen. The resulting mixture was then stirred
under one atmosphere of hydrogen for 20 h at room temperature. The
reaction mixture was flushed with nitrogen and filtered through a pad
of Celite, and the filtrate was concentrated under reduced pressure to
provide a thick, orange syrup. The material was dissolved in toluene
(250 mL) and filtered through a pad of Celite. The filtrate was
(36) (a) Weiner, S. J.; Kollman, P. A.; Case, D. A.; Singh, U. C.; Ghio, C.;
Alagoha, G.; Profeta, S., Jr.; Weiner, P. K. J. Am. Chem. Soc. 1984, 106,
765-784. (b) Singh, U. C.; Weiner, P. K.; Caldwell, J. K.; Kollman, P. A.
AMBER Version 3.0; University of California at San Francisco: San
Francisco, CA, 1986.
(37) (a) Berendsen, H. J. C.; Grigera, J. R.; Straatsma, T. P. J. Phys. Chem.
1987, 91, 6269-6271. (b) Reddy, M. R.; Berkowitz, M. J. Chem. Phys.
1988, 88, 7104-7110.
(34) The order of reagent addition and addition times are critical for successful
conversion. The addition of acetic acid before the reducing agent results
in exclusive formation of the dihydrobenzimidazole.
(35) An exotherm was observed during the iron (III) chloride addition.
9
J. AM. CHEM. SOC. VOL. 129, NO. 50, 2007 15489

A R T I C L E S
Erion et al.
then energy minimized using a four-stage protocol and the BORN
module of the AMBER program. Energy minimization at each stage
entailed 500 steps of steepest descent followed by 2000 steps of
conjugate gradient. In stage one, the inhibitor and protein were fixed
and only the water molecules were minimized to relieve any bad
contacts that arose as a result of solvating the system. In stage two, all
heavy atoms were fixed and only the hydrogens were allowed to relax
to obtain optimal hydrogen bond geometries and interactions. In stage
three, atoms associated with the protein and atoms that coincide with
AMP atoms were fixed whereas the remaining atoms and water
molecules were left unrestrained. Last, in stage four, all residues within
25 Å from the center of the AMP mimic were allowed to relax.
A four-stage protocol was also used in the energy minimization of
the solvated inhibitor. Energy minimization was carried out using
periodic boundary conditions in all directions. Each stage entailed 500
steps of steepest descent and 2000 steps of conjugate gradient. Similar
to above, stage one minimized the water molecules while keeping the
inhibitor fixed, stage two relaxed the hydrogen atoms, stage three fixed
only the ligand atoms in common with the molecule (AMP, 4 or 11)
in the X-ray structure, and stage four allowed all atoms to relax.
The minimized structures for all the inhibitors in the complex and
solvated states were used for calculating the following energy differ-
ences:
Glucose production from freshly isolated rat hepatocytes was
measured as previously described18 after a 1 h incubation with 16.
Intracellular fructose-1,6-bisphosphate levels were determined enzy-
matically at the same timepoint using aldolase- and triosephosphate
isomerase-containing reactions that coupled the production of glycer-
aldehyde 3-phosphate to the oxidation of NADPH.
Source of FBPase and Crystal Growth. The pET3a expression
vector containing the human liver FBPase gene was a gift from M.
Raafat El-Maghrabi (State University of New York, Stony Brook). The
enzyme was expressed in Escherichia coli containing this vector and
purified to homogeneity as previously described.21a Crystals of the
FBPase-16 complex were grown at room temperature by the hanging
drop method: 2 µL FBPase (24 mg/mL) in buffer (2.7 mM 16, 5 mM
sodium malonate, pH 7.2, 1 mM dithiothreitol, 0.5 mM phenylmeth-
ylsulfonyl fluoride, 0.1 mM EDTA, and 2 mM fructose-1,6-bisphos-
phate) was mixed with 2 µL of 6% (vol/vol) polyethylene glycol 8000
(PEG-8K)/0.1 M Tris-HCl (pH 8.2), and equilibrated against 1 mL 6%
PEG-8K/0.1 M Tris-HCl (pH 8.2). The following day, the resulting
crystals were soaked in 20% glycerol/80% crystallographic buffer (6%
PEG-8K/0.1 M Tris-HCl, pH 8.2) (vol/vol) for about 5 s, plunged into
liquid nitrogen (77 K), and then stored until data collection.
Data Collection and Analysis. The diffraction data were collected
at 100 K on beamline F-1 of the Cornell High-energy Synchrotron
Source (CHESS). The crystal was in the space group P21. There are
two tetramers in the asymmetric unit. A search model was obtained
from the PDB code 1FTA.21a Molecular replacement was then carried
out by using AMORE.39 The model of 16 was generated by CS
Chem3D Pro (http://www.camsoft.com), modified according to the
electron density in the simulated annealing omit map, and the
conformation of the model was optimized by energy minimization in
CS Chem3D Pro. The structure of human liver FBPase complexed with
16 was refined by using the CNS_SOLVE package.40 The model was
then visualized and refined against 2Fo-Fc and Fo-Fc maps by using
the program O.41 The stereochemical properties of the intermediate and
the final structures were examined by PROCHECK.42 In the final
structure, 89.8% residues were in the most favored regions, 9.9%
residues were in additional allowed regions, and 0.3% residues were
in generously allowed regions.
∆Ebind (intra) ) Ecom(intra) - Esol(intra)
∆Ebind (inter) ) Ecom(inter) - Esol (inter)
∆Ebind (total) ) ∆Ebind (intra) + ∆Ebind (inter)
(1)
(2)
(3)
where, ∆Ebind(intra) and ∆Ebind(inter) are relative intra- and intermo-
lecular binding interaction energies of an inhibitor, respectively, and
where Ecom(intra), Ecom(inter), Esol(intra), and Esol(inter) are intra- and
intermolecular interaction energies of an inhibitor in the complex and
solvated states, respectively. ∆Ebind (total) is the total binding energy
of an inhibitor.
Free-energy calculations were conducted as previously described.24,38
Prior to the mutation, the system was minimized using the four-stage
protocol followed by an equilibration period using 20 ps of MD
simulation and a 2 fs time step. Mutations suitable for the single
topology method were conducted using 51 windows with each window
comprising 3 ps of equilibration and 6 ps of data collection or a total
of 479 ps of MD simulation. The double topology method30 was used
for mutations involving larger structural changes. In this case, calcula-
tions entailed two stages with each stage using 51 windows and each
window comprising 2 ps of equilibration and 4 ps of data collection or
a total of 632 ps of MD simulation. Each mutation followed the
doublewide sampling procedure, and the results reported are based on
the averages from the backward and forward simulations of the
mutation. Errors were estimated for each window by dividing the
window statistics into four groups (in both forward and backward
directions) and computing the standard deviation for the indicated free-
energy change. The root-mean-square of these window errors are
reported as a measure of the statistical uncertainty in the result for each
mutation.
Biological Assays. FBPase activity and the activity of the AMP-
binding enzymes human erythrocytic AMP deaminase, rabbit muscle
glycogen phosphorylase, human AMP-activated protein kinase, rabbit
muscle adenylate kinase, and rabbit liver phosphofructose kinase were
measured as previously described18 and as detailed in the Supporting
Information. FBPase IC50 values were calculated by 4-parameter
logistics regression analysis of plots of FBPase activity versus inhibitor
concentration using Statview software (SAS Institute, Cary, NC).
Acknowledgment. We thank Dr. Edward Robinson, Dr.
Gerard Scarlato, Brian Brown, Dr. Raja Reddy, Yan Liu, Wei
Xiao, Jay DaRe, Joe Kopcho, and Robert Rydzewski for their
assistance in compound synthesis; Tim Colby and Emily
Topczewski for characterization of the compounds in the
biological assays; and Dr. Sridhar Prasad for graphical repre-
sentations of the binding interactions.
Supporting Information Available: Analytical data and
procedures for the preparation of compounds 3-15 and
intermediates generated during the synthesis of 16. Enzyme
assays and rat hepatocyte assays. Procedure for developing force
field parameters for nonstandard residues, as well as procedures
for energy minimizations in solvent and the complex and for
the development of the computer models of FBPase complexed
with AMP and 16. Complete ref 8b. This material is available
free of charge via the Internet at http://pubs.acs.org.
JA074869U
(39) Navaza, J. Acta Cryst. 1994, A50, 157-163.
(40) Brunger, A. T.; Adams, P. D.; Clore, G. M.; DeLano, W. L.; Gros, P.;
Grosse-Kunstleve, R. W.; Jiang, J.-S.; Kuszewski, J.; Nilges, N.; Pannu,
N. S.; Read, R. J.; Rice, L. M.; Simonson, T.; Warren, G. L. Acta Cryst.
1998, D54, 905-921.
(41) Jones, T. A.; Zou, J.-Y.; Cowan, S. W.; Kjeldgaard, M. Acta Cryst. 1991,
A47, 110-119.
(38) Reddy, M. R.; Erion, M. D.; Agarwal, A. In ReViews in Computational
Chemistry; Lipkowitz, K. B., Boyd, D. B., Eds.; Wiley-VCH: New York,
2000; Vol. 16, pp 217-304.
(42) Laskowski, R. A.; MacArthur, M. W.; Moss, D. S.; Thornton, J. M. J.
App. Cryst. 1993, 26, 283-291.
9
15490 J. AM. CHEM. SOC. VOL. 129, NO. 50, 2007