Development of nonsteroidal anti-inflammatory drug analogs and steroid carboxylates selective for human aldo-keto reductase isoforms: Potential antineoplastic agents that work independently of cyclooxygenase isozymes

Article abstract of DOI:10.1124/mol.104.006569

Human aldo-keto reductases (AKRs) regulate nuclear receptors by controlling ligand availability. Enzymes implicated in regulating ligand occupancy and trans-activation of the nuclear receptors belong to the AKR1C family (AKR1C1-AKR1C3). Nuclear receptors regulated by AKR1C members include the steroid hormone receptors (androgen, estrogen, and progesterone receptors) and the orphan peroxisome proliferator-activated receptor (PPARγ). In human myeloid leukemia (HL-60) cells, ligand access to PPARγ is regulated by AKR1C3, which diverts PGD₂ metabolism away from J-series prostanoids (Desmond et al., 2003). Inhibition of AKR1C3 by indomethacin, a nonsteroidal anti-inflammatory drug (NSAID), caused PPARγ-mediated terminal differentiation of the HL-60 cells. To discriminate between antineoplastic effects of NSAIDs that are mediated by either AKR1C or cyclooxygenase (COX) isozymes, selective inhibitors are required. We report a structural series of N-phenylanthranilic acid derivatives and steroid carboxylates that selectively inhibit recombinant AKR1C isoforms but do not inhibit recombinant COX-1 or COX-2. The inhibition constants, IC₅₀, K₁ values, and inhibition patterns were determined for the NSAID analogs and steroid carboxylates against AKR1C and COX isozymes. Lead compounds, 4-chloro-N-phenylanthranilic acid and 4-benzoyl-benzoic acid for the N-phenylanthranilic acid analogs and most steroid carboxylates, exhibited IC₅₀ values that had greater than 500-fold selectivity for AKR1C isozymes compared with COX-1 and COX-2. Crystallographic and molecular modeling studies showed that the carboxylic acid of the inhibitor ligand was tethered by the catalytic Tyr55-OH₂⁺ and explained why A-ring substituted N-phenylanthranilates inhibited only AKR1C enzymes. These compounds can be used to dissect the role of the AKR1C isozymes in neoplastic diseases and may have cancer chemopreventive roles independent of COX inhibition.

Full text of DOI:10.1124/mol.104.006569

0026-895X/05/6701-60–68$20.00
MOLECULAR PHARMACOLOGY
Copyright © 2005 The American Society for Pharmacology and Experimental Therapeutics
Mol Pharmacol 67:60–68, 2005
Vol. 67, No. 1
6569/1189059
Printed in U.S.A.
Development of Nonsteroidal Anti-Inflammatory Drug Analogs
and Steroid Carboxylates Selective for Human Aldo-Keto
Reductase Isoforms: Potential Antineoplastic Agents That
Work Independently of Cyclooxygenase Isozymes
David R. Bauman, Stephen I. Rudnick, Lawrence M. Szewczuk, Yi Jin, Sridhar Gopishetty,
and Trevor M. Penning
Department of Pharmacology, University of Pennsylvania School of Medicine, Philadelphia, Pennsylvania
Received August 26, 2004; accepted October 5, 2004
ABSTRACT
Human aldo-keto reductases (AKRs) regulate nuclear receptors lates that selectively inhibit recombinant AKR1C isoforms
by controlling ligand availability. Enzymes implicated in regu- but do not inhibit recombinant COX-1 or COX-2. The inhibi-
lating ligand occupancy and trans-activation of the nuclear tion constants, IC50, KI values, and inhibition patterns were
receptors belong to the AKR1C family (AKR1C1-AKR1C3). Nu- determined for the NSAID analogs and steroid carboxylates
clear receptors regulated by AKR1C members include the ste- against AKR1C and COX isozymes. Lead compounds,
roid hormone receptors (androgen, estrogen, and progesterone 4-chloro-N-phenylanthranilic acid and 4-benzoyl-benzoic
receptors) and the orphan peroxisome proliferator-activated acid for the N-phenylanthranilic acid analogs and most ste-
receptor (PPAR␥). In human myeloid leukemia (HL-60) cells, roid carboxylates, exhibited IC50 values that had greater than
ligand access to PPAR␥ is regulated by AKR1C3, which diverts 500-fold selectivity for AKR1C isozymes compared with
PGD2 metabolism away from J-series prostanoids (Desmond et COX-1 and COX-2. Crystallographic and molecular modeling
al., 2003). Inhibition of AKR1C3 by indomethacin, a nonsteroi- studies showed that the carboxylic acid of the inhibitor li-
ϩ
dal anti-inflammatory drug (NSAID), caused PPAR␥-mediated gand was tethered by the catalytic Tyr55-OH2 and ex-
terminal differentiation of the HL-60 cells. To discriminate plained why A-ring substituted N-phenylanthranilates inhib-
between antineoplastic effects of NSAIDs that are mediated ited only AKR1C enzymes. These compounds can be used to
by either AKR1C or cyclooxygenase (COX) isozymes, selec- dissect the role of the AKR1C isozymes in neoplastic dis-
tive inhibitors are required. We report a structural series of eases and may have cancer chemopreventive roles indepen-
N-phenylanthranilic acid derivatives and steroid carboxy- dent of COX inhibition.
The human AKR1C isozymes are hydroxysteroid dehydro- ligand occupancy and trans-activation of the nuclear steroid
genases (HSDs) and are involved in the prereceptor regula-
tion of steroid hormone action (Penning, 1997; Dufort et al.,
1999; Dufort et al., 2001; Rizner et al., 2003). AKR1C
isozymes regulate the concentration of active and inactive
androgens, estrogens, and progestins in target tissues by
catalyzing the reduction of ketosteroids at the C3, C17, or
C20 positions (Dufort et al., 1999; Penning et al., 2000; Du-
fort et al., 2001). Therefore, AKR1C isozymes regulate the
hormone receptors, which include the androgen, estrogen
(ER), and progesterone receptors. Therefore, they maybe im-
portant in regulating steroid hormone concentrations in tar-
get tissues by modulating their intracrine formation (Labrie
et al., 2001). Selective inhibitors of AKR1C isoforms provide
an approach to obtain tissue specific effects of steroid hor-
mones and are termed selective intracrine modulators
(SIMs).
The human AKR1C isozymes involved in regulating the
local concentration of steroid hormones include AKR1C1
(20␣-HSD), AKR1C2 (Type 3 3␣-HSD), and AKR1C3 (Type 5
17␤-HSD, Type 2 3␣-HSD) and share greater than 86%
amino acid sequence identity (Penning et al., 2000). Despite
This work was supported by National Institutes of Health grants R01-
CA90744 (to T.M.P.) and training grant 1R25-CA101871-D1 (to D.R.B.).
Article, publication date, and citation information can be found at
http://molpharm.aspetjournals.org.
doi:10.1124/mol.104.006569.
ABBREVIATIONS: AKR, aldo-keto reductase; HSD, hydroxysteroid dehydrogenase; ER, estrogen receptor; SIM, selective intracrine modulator;
NSAID, nonsteroidal anti-inflammatory drug; PG, prostaglandin; PPAR␥, peroxisome proliferator-activated receptor ␥; COX, cyclooxygenase; MS,
mass spectroscopy; ES, electrospray; TMPD, N,N,NЈ,NЈ-tetramethyl-1,4-phenylenediamine; DMSO, dimethyl sulfoxide; SAR, structure activity
relationship.
60

NSAID Analogs That Target Human AKR1Cs and Not COX
61
their high sequence identity, they display different substrate regulating myeloid leukemia cell differentiation and indicated
preferences, inhibition profiles, and tissue-specific expres- that NSAIDs can have antineoplastic properties via a non–
sion patterns (Matsuura et al., 1997; Dufort et al., 1999; cyclooxygenase (COX)-2 pathway. This result not only validated
Penning et al., 2000; Dufort et al., 2001). AKR1C isozymes the concept of SIMs, whereby targeting a specific AKR1C
catalyze the reduction of 3-, 17- and 20-ketosteroids in dif- isozyme has beneficial therapeutic effects, but also indicated a
ferent ratios so that AKR1C1 is considered a 20␣-HSD that need to develop NSAID analogs that discriminate between
will inactivate progesterone, AKR1C2 is considered a periph- AKR1C isozymes and their traditional targets, COX-1 and
eral 3␣-HSD that will inactivate 5␣-dihydrotestosterone, and COX-2. In 1983, Penning and Talalay showed that the com-
AKR1C3 is considered a peripheral 17␤-HSD that will form monly prescribed NSAIDs indomethacin and mefenamic acid
testosterone and 17␤-estradiol from their less active precur- were potent inhibitors of AKR1C9 at pharmacologically rele-
sors (Fig. 1) (Dufort et al., 1999; Penning et al., 2000; Dufort vant concentrations. AKR1C9 is a model for the human AKR1C
2001; Rizner et al., 2003). Several of these isoforms will also isozymes because it has 69% sequence identity at the amino
transform prostaglandins; for example, AKR1C3 is also acid level with the human AKR1C isozymes. This suggests that
known as prostaglandin F2␣ synthase (Matsuura et al., the human enzymes may be important targets for NSAIDs, and
1998). Therefore, AKR1C isozymes may be further involved their inhibition may contribute to the antineoplastic effects of
in the prereceptor regulation of a large group of nuclear these drugs.
receptors (e.g., steroid receptors and nuclear orphan recep-
tors).
Nonselective NSAIDs do not discriminate between COX-1
and COX-2 (Laneuville et al., 1994). Therefore, selective
Desmond et al. (2003) recently showed that inhibition of COX-2 inhibitors were developed to reduce the side effects of
AKR1C3 by the nonsteroidal anti-inflammatory drug (NSAID) classic NSAIDs. However, mounting evidence suggests non-
indomethacin prevented the proliferation of human myeloid selective and selective NSAID inhibitors have other targets
leukemia HL-60 cells. AKR1C3 prevented the conversion of (Wick et al., 2002; Keller and Giardiello, 2003; Leng et al.,
PGD2 to 15-⌬12,14-PGJ2 [a ligand for the nuclear orphan perox- 2003; Reid et al., 2003; Sanchez-Alcazar et al., 2003). These
isome proliferator-activated receptor ␥ (PPAR␥)]. By converting include the multidrug resistance proteins, transcription fac-
PGD2 to PGF2␣, AKR1C3 deprived PPAR␥ of its ligand and tors, mitogen-activated protein kinase kinases, cell cycle reg-
prevented terminal differentiation of the myeloid leukemia cells ulatory proteins, and the human AKR1C isozymes, which are
(Fig. 2). This phenotypic change suggests a role for AKR1C3 in the focus of this article (Penning and Talalay, 1983; Askonas
et al., 1991; Tegeder et al., 2001; Wick et al., 2002; Desmond
et al., 2003; Keller and Giardiello, 2003; Leng et al., 2003;
Reid et al., 2003; Sanchez-Alcazar et al., 2003). To test the
hypothesis that AKR1C isozymes are involved in COX-inde-
pendent neoplasias, compounds are needed that are selective
for AKR1C isozymes but not for COX-1 or COX-2.
Fig. 2. Potential therapeutic benefit of inhibiting AKR1C3 using an
NSAID analog that does not inhibit COX-1 leading to the activation of the
Fig. 1. Regulation of ligand concentrations for nuclear receptors by
AKR1C isozymes. All AKR1C isozymes are shown working in the reduc-
tion direction only.
PPAR␥ receptor and ultimately cellular differentiation. PGG2, prosta-
glandin G2; PGD2, prostaglandin D2; PGF2␣, 9␣,11␤-prostaglandin F2␣
;
15⌬-PGJ2, 15-⌬12,14- prostaglandin J2; RXR, retinoic acid receptor.

62
Bauman et al.
We report a set of AKR1C inhibitors based on N-phenylan- of 90% ethanol containing 0.6 g of HCl was stirred at room temper-
ature for 8 h. After partial concentration, the residue was extracted
with dichloromethane. The organic phase was washed with water
until neutral and then with saturated brine, dried over sodium
sulfate, and evaporated under reduced pressure. The overall yield
was 190 mg (95%) and the structure was confirmed by 1H NMR and
MS 1H NMR (500 mHz) indicated a loss of the carboxylic acid proton
at ␦ ppm 12.0 (1H, -COOH, s) and a gain in the ethyl group at ␦ ppm
4.1 and 3.6 (2H, -CH2, q and 3H, -CH3, t). The MS (ESϩ) gave a
parent ion MHϩ m/z ϭ 427.32 (ϩNa), predicted m/z ϭ 403.31.
thranilic acids and cholanic acids that can discriminate be-
tween AKR1C and COX isozymes. Lead compounds that
specifically inhibit AKR1C isozymes but not COX were iden-
tified. Because AKR1C isozymes are involved in prereceptor
regulation of nuclear receptors, these new compounds may
lead to cancer chemopreventive agents that exert their ef-
fects independently of COX. The selective COX-2 inhibitor
celecoxib (Celebrex) was also found to inhibit the AKR1C
isozymes. Therefore, the AKR1C selective inhibitors de-
scribed will be able to discriminate between the effects of
celecoxib on COX-2 versus those effects on AKR1C isozymes.
Purification of AKR1C Isoforms and COX Isoforms
Recombinant homogenous AKR1C isozymes were purified accord-
ing to published procedures (Burczynski et al., 1998; Penning et al.,
2000). Purification to homogeneity of COX-1 from ram seminal ves-
icles and COX-2 from baculovirus-infected SF-21 cells was performed
as described previously (George et al., 1996, and Smith et al., 2000,
respectively).
Materials and Methods
Materials
2-Chloro-5-methyl-benzoic acid was purchased from Pfaltz and
Bauer Chemicals (Waterbury, CT), and 2-chloro-4-nitro-benzoic acid
was purchased form Acros Organics (Pittsburgh, PA). Glycocholic
acid, taurodeoxycholic acid, 4-carboxy-2Ј,4Ј-dinitrodiphenylamine,
4-chloro-N-phenylanthranilic acid, 4-chloro-N-(4Ј-toly)-anthranilic
acid, 3-amino-4-phenylamino-benzoic acid, and 4-benzoyl-benzoic
acid were purchased from Sigma-Aldrich (St. Louis, MO). Mefenamic
acid and ursodeoxycholic acid were purchased from ICN Biomedical
Inc. (Aurora, OH). 5␤-Cholanic acid-3-one, 5␤-cholanic acid-3␣,7␣-
diol, and 5␤-cholanic acid-3␣-ol (lithocholic acid) were purchased
from Steraloids (Wilton, NH). Celecoxib was purchased from
ChemPacific USA (Baltimore, MD). Arachidonic acid was purchased
from Cayman Chemicals (Ann Arbor, MI).
Steady-State Kinetic Assays
A continuous assay was used to monitor AKR1C activity. This
assay monitored the oxidation of 1-acenaphthenol by measuring the
increase in absorbance of NADH at 340 nm ⑀ (molar extinction
coefficient) ϭ 6270 MϪ1 cmϪ1. Reactions were performed in 1.0-ml
systems containing 100 mM potassium phosphate, pH 7.0, 1% BSA,
2.3 mM NADϩ, increasing concentrations of 1-acenaphthenol rang-
ing from 5 to 500 ␮M, and 4% methanol. Using this assay, the
specific activities were determined to be 2.1 ␮mol/min/mg for
AKR1C1, 2.5 ␮mol/min/mg for AKR1C2, and 2.8 ␮mol/min/mg for
AKR1C3 at 1 mM 1-acenaphthenol (Burczynski et al., 1998; Penning
et al., 2000). A continuous assay was also used to monitor COX-
activity. This assay monitored the oxidation of N,N,NЈ,NЈ-tetram-
ethyl-1,4-phenylenediamine (TMPD) when it is used as coreductant
to reduce PGG2 to PGH2 starting with arachidonic acid (Benedetto et
al., 1987). The assay monitors the formation of N,N,NЈ,NЈ-tetram-
ethyl-1,4-phenylenediimine (⑀ ϭ 12,000 MϪ1 cmϪ1) at 610 nm. Reac-
tions were performed in 1.0-ml systems containing 100 mM Tris-
HCl, pH 8.0, 80 ␮M TMPD, 2 ␮M ferriprotoporphyrin IX, 60 ␮M
arachidonic acid, and 5% DMSO. Using this assay, the specific ac-
tivities were determined to be 25 and 16.8 ␮mol of TMPD oxidized/
min/mg of enzyme at 150 ␮M arachidonic acid for COX-1 and COX-2,
respectively.
Synthesis of Nonsteroidal Inhibitors: N-Phenylanthranilic
Acid Derivatives
The synthesis of the N-phenylanthranilic acid derivatives used the
Ullmann-Goldberg reaction that couples benzoic acid and aniline
derivatives as described previously (Ozaki et al., 1985). The general
method was as follows: 0.04 mol of the halogenated benzoic acid
derivative, 0.08 mol of aniline, 0.05 mol of potassium carbonate,
0.003 mol of copper(II) powder, and 15% (w/w) pyridine were refluxed
in 25 ml of amyl alcohol for 18 h. The amyl alcohol was removed by
rotary evaporation, and a mixture of potassium carbonate and water
was added to the resultant slurry. The mixture was acidified with
HCl (1:1) to pH 4 to 5. The resultant solid was filtered, washed with
cold H2O, extracted in CHCl3, and recrystallized from EtOH. The
structure was confirmed by 1H-NMR (500 mHz) in CDCl3 and by
mass spectroscopy (MS). All chemical shifts are reported as parts per
million relevant to a tetramethylsilane internal standard. The high-
resolution MS data were obtained on a Micromass AutoSpec instru-
ment (Waters, Milford, MA) carried out using electrospray in either
the positive or negative ionization mode.
Initial velocities were obtained by linear regression to reaction
progress curves. Plots of velocity versus substrate concentration
were hyperbolic and could be fit to the Michaelis-Menten equation
(v ϭ (Vmax ϫ S)/(KM ϩ S) to yield values (Ϯ S.D.) for Vmax, kcat, and
KM for a given substrate.
Reversible Inhibition Studies
IC50 values for AKR1C enzymes were measured by varying the
inhibitor concentrations while holding the 1-acenaphthenol concen-
tration equal to the KM for AKR1C1 and AKR1C2. Because of the
solubility constraints of 1-acenaphthenol, the KM for AKR1C3 could
not be accurately determined, but estimates approached 2 mM, and
the resulting IC50 experiments were performed at a concentration of
1-acenaphthenol equal to 100 ␮M (KM of AKR1C2). Inhibitors were
dissolved in DMSO, and the final concentration of cosolvent was 4%
(2% DMSO and 2% methanol) for AKR1C isozymes. IC50 values were
measured for COX isozymes by varying inhibitor concentrations
while holding the arachidonic acid concentration equal to 60 ␮M
(Benedetto et al., 1987). Inhibitors were dissolved in DMSO, and the
final concentration of cosolvent was 5% for the COX measurements.
The presence of the DMSO in the assay had no effect on initial
velocities. Inhibition data were fit using Grafit 5.0 [y ϭ (range)/
[1 ϩ (I/IC50)S] ϩ background] yielding the IC50 value. Reversible
5-Methyl-N-phenylanthranilic Acid (C14NO2H13). This was
synthesized using 2-chloro-5-methyl-benzoic acid and aniline with a
46% yield, purified, and recrystallized as described above. 1H NMR
(500 mHz) indicated ␦ ppm 2.3 (3H, -CH3, s), ␦ ppm 4.3 (1H, NH, s),
and ␦ ppm 7.5 to 8.2 (8H, aromatic). MS (ESϩ) gave a parent ion
MHϩ m/z ϭ 286.14 (ϩNa, ϩCl), predicted m/z ϭ 227.26.
4-Nitro-N-phenylanthranilic Acid (C13N2O4H10). This was
synthesized using 2-chloro-4-nitro-benzoic acid and aniline with an
11% yield, purified, and recrystallized as described above. 1H NMR
(500 mHz) indicated ␦ ppm 4.3 (1H, NH, s), ␦ ppm 7.5 to 8.2 (8H,
aromatic), and ␦ ppm 9.5 (1H, -COOH, s). The MS (ESϪ) gave a
parent ion MϪ-H m/z ϭ 257.14, predicted m/z ϭ 258.06.
Synthesis of Steroidal Inhibitors: Formation of Ethyl
Esters
The lithocholic ester was synthesized as described previously inhibition patterns for the AKR1C and COX isozymes were deter-
(Barton et al., 1989). A solution of 200 mg of lithocholic acid in 15 ml mined by varying the concentration of inhibitor (0.125–3 ϫ IC50
)

NSAID Analogs That Target Human AKR1Cs and Not COX
63
over fixed substrate concentrations (0.2–5 ϫ KM), and the reac- the Ullmann-Goldberg coupling reaction in which any halogenated
tions were initiated with enzyme. Initial velocity data were glo-
bally fit to the competitive, noncompetitive, and uncompetitive
inhibition models and the results presented represent the best fit
as judged by standard deviation using Grafit 5.0 to determine the
inhibition constant KI. Using the association that exists between
the IC50 value, KM, and the substrate concentration, the KI values
were calculated for many inhibitors using the Cheng-Prusoff
(1973) relationship, [KI ϭ (IC50)/(1 ϩ S/KM)], where the lead
compound in that class displayed competitive inhibition kinetics.
benzoic acid is coupled to any aniline derivative, providing a route to
a combinatorial library (Fig. 3A).
Another structural lead to AKR1C inhibitors exists in the C24-
cholanic acids (bile acids). AKR1C2 is also known as the human bile
acid binding protein and binds bile acids with nanomolar affinity,
whereas AKR1C1 and AKR1C3 bind bile acids with micromolar
affinity (Matsuura et al., 1997, 1998). This selectivity is impressive
considering that these enzymes share greater than 86% amino acid
identity. More impressive is that AKR1C1 and AKR1C2 are greater
than 97% similar at the amino acid level and differ by only one amino
acid in the active site, yet bile acids are more than 100-fold selective
for AKR1C2 than AKR1C1. This result indicates that selective in-
hibitors maybe attainable for each AKR1C isozyme.
KI values are more representative of the inhibition potency and
will be used to compare inhibition potency between AKR1C isozymes
because KI values were obtainable for most of the compounds tested
(Fig. 4). However, in a comparison of inhibition potency between the
AKR1C isozymes and COX isozymes, IC50 values were used because
KI values were unobtainable for most of the compounds tested for the
COX isozymes.
Time-Dependent Inhibition of Celecoxib for COX and
AKR1C Isozymes
The inhibition potency of celecoxib for COX and AKR1C isozymes
was determined by two methods. First, celecoxib was added directly
to the enzyme assays as described above under conditions in which it
would compete for substrate. Second, the time-dependent inhibition
of COX and AKR1C isozymes by celecoxib (0.0005–50 ␮M for COX-2,
0.005–200 ␮M for COX-1, and 0.005–100 ␮M for AKR1C isozymes)
was investigated. In this method, the AKR1C enzymes (1.5 ␮g for
AKR1C1, 1.5 ␮g for AKR1C2, and 11.6 ␮g for AKR1C3) and COX
enzymes (10.7 ␮g for COX-1 and 7.6 ␮g for COX-2) were preincu-
bated with inhibitor for 10 min and the reaction was initiated with
substrate (Gierse et al., 1996). Preincubation of COX-1 and COX-2
with celecoxib for 10 min is sufficient for this slow tight-binding
inhibitor to cause maximal loss of enzyme activity (Copeland et al.,
1994; Gierse et al., 1996).
Results
Inhibitors That Will Discriminate between AKR1C
Isozymes and COX Isozymes
There are several major chemical classes of classic NSAIDs that
target COX and AKR1C isozymes. Of these, the two most potent are
the indole acetic acids (e.g., indomethacin) and the N-phenylanthra-
nilic acids (e.g., mefenamic acid and meclofenamic acid). In this
study, we concentrated on the N-phenylanthranilic acids for several
reasons. First, their known structure-activity relationships (SAR)
points to subtle changes that could be exploited to eliminate COX
inhibition (Fig. 3B) but retain AKR1C inhibition (Scherrer, 1974;
Lombardino, 1985). These structural alterations include changing
the heteroatom, substitution on the amine, substitution on the
A-ring, and varying the ortho relationship that exists between
the amine and the carboxylic acid in the A-ring. Second, N-pheny-
lanthranilic acids can be synthesized via a single-pot reaction using
Fig. 3. A, the Ullmann-Goldberg coupling reaction. B, SAR trends in
N-phenylanthranilic acids that eliminate COX-1 inhibition.
Fig. 4. Inhibitors screened.

64
Bauman et al.
Inhibition Patterns for the AKR1C Isozymes and the COX
Isozymes
the Chang-Prusoff (1973) relationship, KI values were calculated for
the remainder of the compounds in each class as described under
Materials and Methods (Tables 1 and 2).
Inhibition patterns for the AKR1C isozymes and COX-1 were
determined for the N-phenylanthranilic acid parent compound mefe-
namic acid (1), 5-methyl-N-phenylanthranilic acid (2), and 5␤-
cholanic-3-one (9). Representative competitive inhibition patterns
for AKR1C2 with compounds 1, 2, and 9 were obtained (Fig. 5). The
representative inhibition patterns indicate a competitive mode of
inhibition for each chemical class, whereby the substrate and inhib-
itor compete for the same binary complex (AKR1C2-NADϩ). Using
N-Phenylanthranilic Acids—Enzyme Selectivity, Potency,
and Trends
The parent compound mefenamic acid (1) inhibited both AKR1C
and COX isozymes (Fig. 6). Inhibition by 1 indicated that it is more
selective for the AKR1C isozymes than the COX isozymes, but it was
able to inhibit all enzymes tested. The rank order of potency against
these enzymes was: AKR1C3 (IC50 ϭ 0.39 ␮M), AKR1C2 (IC50 ϭ 0.96
␮M), AKR1C1 (IC50 ϭ 4.0 ␮M), COX-1 (IC50 ϭ 33.3 ␮M), and finally
COX-2 (IC50 ϭ 225 ␮M) (Table 1). Representative inhibition patterns
with the NSAID analogs showed that the pattern of inhibition was
competitive. Thus the IC50 values obtained with each of the AKR1C
isozymes were directly comparable. In addition, because of the rela-
tionship that exists between IC50 values and KM under competitive
inhibition conditions, KI values could be computed for each of the
remaining inhibitors. By contrast, IC50 values for COX-1 and COX-2
inhibition were obtained at 60 ␮M arachidonic acid as described
previously (Gierse et al., 1996). Our COX data are consistent with
previous findings, which showed that the N-phenylanthranilic acids
are more potent against COX-1 (Gierse et al., 1996). Although the
IC50 values for the COX isozymes were not determined at KM, the
N-phenylanthranilic acid analogs can be directly compared with the
values obtained for mefenamic acid, the parent compound (1).
N-Phenylanthranilic acid derivatives were examined for their se-
lectivity to inhibit AKR1C isozymes versus COX isozymes (Table 1).
The N-phenylanthranilic acid analogs were found to be potent inhib-
itors for the AKR1C isozymes but did not inhibit COX isozymes, as
expected. The loss of COX inhibition by simple modifications to the
A-ring was predicted by existing SAR; however, the maintenance of
AKR1C inhibition was a unique finding. For example, halogenation
of the A-ring led to 4, which remained a potent inhibitor for the
AKR1C isozymes. However, 4 did not inhibit the COX isozymes (Fig.
7). Direct comparison of 1 and 4 indicated that halogenation of the
A-ring led to a decrease in inhibition potency by more than 30-fold for
COX-1 and 5-fold for COX-2. The rank order for enzyme inhibition
with 4 was AKR1C2 (IC50 ϭ 1.15 ␮M) Ͼ AKR1C3 (IC50 ϭ 3.01 ␮M) Ͼ
AKR1C1 (IC50 ϭ 8.2 ␮M) Ͼ COX-1 and COX-2 (IC50 Ͼ 1000 ␮M). The
selectivity for the N-phenylanthranilic acid analogs ranged from
30-fold (AKR1C1, IC50 ϭ 8.3 ␮M; COX-1, IC50 ϭ 240 ␮M) for com-
pound 7, which contained substitution on both rings, to greater than
1200-fold (AKR1C3, IC50 ϭ 0.40 ␮M; COX-1, IC50 ϾϾ 500 ␮M) for
compound 3, in which the ortho relationship between the carboxylate
and amine was altered to para. No time dependence was noticed for
the inhibition of AKR1C and COX isozymes by the N-phenylanthra-
nilic acid derivatives (data not shown).
Analysis of the N-phenylanthranilic acid derivatives revealed po-
tential trends in loss of COX isozyme inhibition. Replacement of the
heteroatom almost completely abolished COX inhibition but did not
affect AKR1C inhibition. For example, compound 8 did not inhibit
COX-1 or COX-2 activity (COX-1, IC50 ϾϾ 1000 ␮M; COX-2, IC50 ϾϾ
1000 ␮M) compared with the parent compound (1), which inhibited
both COX isozymes but remained a potent inhibitor for the AKR1C
isozymes (AKR1C3, IC50 ϭ 2.04 ␮M; AKR1C1, IC50 ϭ 33.2 ␮M; and
AKR1C2, IC50 ϭ 36.9 ␮M). Movement of the carboxylic acid from the
ortho to the para position on the A-ring decreased inhibition dramat-
ically for COX-1 and COX-2 but did not affect AKR1C inhibition. For
example, compound 6 gave IC50 values of Ͼ 1000 ␮M for COX-1 and
COX-2 and remained a potent inhibitor for the AKR1C isozymes
(AKR1C3, IC50 ϭ 10.7 ␮M; AKR1C2, IC50 ϭ 26.2 ␮M; and AKR1C1,
IC50 ϭ 49.4 ␮M), whereas compound 1 gave IC50 values of 33.3 and
225 ␮M for COX-1 and COX-2, respectively. However, substituents
on the B-ring resulted in a slight increase in COX inhibition potency.
For example, addition of a para-methyl group to the B-ring increased
the inhibition potency for COX-1 more than 4-fold by 7 (IC50 ϭ 240
Fig. 5. Representative inhibition patterns for AKR1C2 using 1-acenaph-
thenol as substrate. A, competitive inhibition pattern for AKR1C2 by
mefenamic acid. B, competitive inhibition pattern for AKR1C2 by 5-methyl-
N-phenylanthranilic acid. C, competitive inhibition pattern for AKR1C2 by
5␤-cholanic acid-3-one. V, initial velocity; S, substrate (1-acenaphthenol). KI
values were determined as described under Materials and Methods, and
each point represents one experiment performed in triplicate.

NSAID Analogs That Target Human AKR1Cs and Not COX
65
␮M) compared with 4 (IC50 Ͼ 1000 ␮M). By contrast, substitution on AKR1C isozymes tested, whereas the COX isozymes were not inhibited.
the A-ring also substantially reduced potency against the COX The selectivity for the AKR1C isozymes versus COX isozymes ranged
isozymes but retained AKR1C inhibition. For example, 2 was 13-fold from 5-fold for 13 to greater than 30,000 for 12.
and 5-fold less potent against COX-1 and COX-2, respectively, com-
pared with 1.
Comparison of Inhibitors for AKR1C Isoform Selectivity
Steroid Carboxylates—Isoform Selectivity, Potency, and
Trends. C24-cholanic acids are known inhibitors of AKR1C2, but
N-Phenylanthranilic Acid—Isoform Selectivity, Potency,
their inhibition potencies toward other AKR1C and COX isozymes and Trends. The N-phenylanthranilic acid analogs were analyzed
are unknown. Screening the steroid carboxylates indicated that for AKR1C isozyme selectivity. In general, the selectivity among the
these compounds were both more potent and selective for the AKR1C individual AKR1C isozymes was not as dramatic as the selectivity
isozymes than the COX isozymes (Table 2). The AKR1C isozymes between AKR1C and COX isozymes. However, compounds 4, 7, and
were potently inhibited by the bile acids, whereas the COX isozymes
8 were more than 6-fold more selective for different AKR1C
were not inhibited. For example, 5␤-cholanic acid-3-one (9) potently isozymes. Further analysis indicated potential trends that could be
inhibited the AKR1C isozymes, whereas the COX isozymes were exploited to increase isozyme selectivity. Movement of the carboxylic
unaffected at all concentrations tested (Fig. 8). The rank order of acid from the ortho to the para position on the A-ring increased the
enzyme inhibition with compound 9 was AKR1C2 (IC50 ϭ 0.04 selectivity for AKR1C3 compared with AKR1C1 or AKR1C2. For
␮M) Ͼ AKR1C3 (IC50 ϭ 1.02 ␮M) Ͼ AKR1C1 (IC50 ϭ 14.0 ␮M) ϾϾ example, 3 and 8 were more selective for AKR1C3 (KI ϭ 0.38 ␮M and
COX-1 and COX-2 (IC50 ϾϾ 250 ␮M). As expected, AKR1C2 was much KI ϭ 1.9 ␮M, respectively) than AKR1C1 (KI ϭ 2.66 ␮M and KI ϭ
more potently inhibited by the bile acids compared with the other 12.5 ␮M, respectively) or AKR1C2 (KI ϭ 1.32 ␮M and KI ϭ 13.3 ␮M,
TABLE 1
Inhibition of AKR1C and COX isozymes by N-phenylanthranilic acid analogs and celecoxib
KI values for AKR1C3 were computed assuming a KM of 2 mM for 1-acenaphthenol. Selectivity ratio refers to comparison of KI values.
IC50
Selectivity Ratio
Compound
AKR1C1
vs
AKR1C2
AKR1C1
vs
AKR1C3
AKR1C2
vs
AKR1C3
AKR1C1
AKR1C2
AKR1C3
COX-1
COX-2
␮M
1
2
4.00 Ϯ 0.5
KI ϭ 0.81
3.2 Ϯ 0.13
KI ϭ 0.88
7.1 Ϯ 0.80
KI ϭ 2.66
8.2 Ϯ 0.90
KI ϭ 3.10
9.2 Ϯ 0.50
KI ϭ 3.50
49.4 Ϯ 8.90
KI ϭ 18.5
8.30 Ϯ 0.50
KI ϭ 3.10
33.2 Ϯ 4.40
KI ϭ 12.5
39.6 Ϯ 3.0
KI ϭ 14.9
0.96 Ϯ 0.11
KI ϭ 0.22
1.23 Ϯ 0.06
KI ϭ 0.53
3.69 Ϯ 0.20
KI ϭ 1.32
1.15 Ϯ 0.06
KI ϭ 0.41
2.09 Ϯ 0.40
KI ϭ 0.75
26.2 Ϯ 3.24
KI ϭ 9.40
0.56 Ϯ 0.17
KI ϭ 0.20
36.9 Ϯ 2.45
KI ϭ 13.3
23.9 Ϯ 1.6
KI ϭ 8.6
0.39 Ϯ 0.09
KI ϭ 0.30
10.8 Ϯ 0.43
KI ϭ 7.27
0.40 Ϯ 0.06
KI ϭ 0.38
3.01 Ϯ 0.22
KI ϭ 2.90
1.98 Ϯ 0.28
KI ϭ 1.90
10.7 Ϯ 1.50
KI ϭ 10.2
1.47 Ϯ 0.37
KI ϭ 1.40
2.04 Ϯ 0.19
KI ϭ 1.9
33.3
KI ϭ 3.90
434
225
ϾϾ1000
N.D.
3.7
1.7
2.0
7.5
4.6
2.0
15
2.7
8.3
6.9
1.1
1.8
1.8
2.2
6.4
2.9
1.4
14
3
ϾϾ500
Ͼ1000
ϾϾ250
3.5
7.0
2.5
1.1
7.0
6.8
1.7
4
ϾϾ1000
N.D.
5
6
ف1000
240
ϾϾ1000
ϾϾ1000
ϾϾ1000
0.04 Ϯ 0.00
7
8
ϾϾ1000
23.7 Ϯ 2.5
1.1
1.7
16
5.45 Ϯ 0.4
KI ϭ 5.2
N.D., not determined.
TABLE 2
Inhibition of AKR1C and COX isozymes by steroid carboxylates
KI values for AKR1C3 were computed using a KM of 2 mM for 1-acenaphthenol. Selectivity ratio refers to comparison of KI values.
IC50
Selectivity Ratio
Compound
AKR1C1
vs
AKR1C2
AKR1C1
vs
AKR1C3
AKR1C2
vs
AKR1C3
AKR1C2
AKR1C2
AKR1C3
COX-1
COX-2
␮M
9
10
11
12
13
14
15
14.0 Ϯ 2.80
KI ϭ 1.30
22.2 Ϯ 2.50
KI ϭ 8.30
9.20 Ϯ 0.60
KI ϭ 3.45
13.1 Ϯ 1.10
KI ϭ 4.90
177 Ϯ 27.0
KI ϭ 66.0
61.2 Ϯ 5.90
KI ϭ 23.0
ϾϾ100
0.04 Ϯ 0.00
KI ϭ 0.021
0.05 Ϯ 0.00
KI ϭ 0.017
0.09 Ϯ 0.00
KI ϭ 0.032
0.03 Ϯ 0.00
KI ϭ 0.011
2.00 Ϯ 0.30
KI ϭ 0.72
0.56 Ϯ 0.06
KI ϭ 0.20
2.50 Ϯ 0.25
KI ϭ 0.90
1.02 Ϯ 0.10
KI ϭ 0.37
10.5 Ϯ 2.70
KI ϭ 10.0
1.57 Ϯ 0.18
KI ϭ 1.50
1.38 Ϯ 0.13
KI ϭ 1.30
216 Ϯ 40.0
KI ϭ 206
ϾϾ250
ϾϾ1000
ϾϾ500
ϾϾ250
ϾϾ1000
ϾϾ500
N.D.
63
483
110
450
92
3.5
1.2
2.3
3.7
3.1
4.3
2.5
18
578
48
ϾϾ1000
ϾϾ1000
ϾϾ1000
ϾϾ250
120
287
493
106
ϾϾ1000
ϾϾ1000
N.D.
104 Ϯ 19.0
KI ϭ 9.0
Ͼ 100
114
42
N.D., not determined.

66
Bauman et al.
respectively). It is interesting that substitution on the A-ring with AKR1C2 than compounds 6 and 8, and it is more than 26-fold and
the carboxylic acid at the para position increased the potency for 5-fold more potent for AKR1C3 than compounds 6 and 8, respec-
AKR1C3 more than 5-fold (compare 6 and 8). Replacement of the tively. Introduction of a chloride at the fourth position of the A-ring
heteroatom seems to favor selectivity for AKR1C3 over AKR1C1 and
increases the selectivity for AKR1C2, given that 4 and 7 are more
AKR1C2 (compare 8 and 6). This increase in selectivity for AKR1C3 than 7-fold more selective for AKR1C2 (KI ϭ 0.41 ␮M, KI ϭ 0.2 ␮M)
might be a result of the position of the carboxylic acid on the A-ring, than for AKR1C1 (KI ϭ 3.1 ␮M, KI ϭ 3.1 ␮M), and AKR1C3 (KI ϭ 2.9
replacement of the heteroatom, or a combination of both. Multiple
␮M, KI ϭ 1.4 ␮M). Furthermore, introduction of a methyl group at
substitutions on the B-ring seem to favor AKR1C isozyme potency, in the fifth position of the A-ring produced a compound (2) that was
that 1 and 3 are very potent inhibitors for all AKR1C isozymes 14-fold more selective for AKR1C2 than AKR1C3, but this compound
compared with COX isozymes. Furthermore, when compounds con- did not discriminate between AKR1C2 and AKR1C1 (compare 1 and
taining the carboxylic acid in the para position are compared (3, 6, 2). It is clear that small changes on the A-ring or B-ring of the
and 8), it is apparent that multiple substitutions on the B-ring N-phenylanthranilic acids can introduce selectivity for the different
enhanced the inhibition potency for the AKR1C isozymes. For exam- AKR1C isozymes.
ple, compound 3 is 7-fold and 4-fold more potent for AKR1C1 than
When comparing the inhibition of the N-phenylanthranilic acid
compounds 6 and 8. Compound 3 is 7-fold and 9-fold more potent for analogs, the electronegativity of the carboxylic acid is determined by
the different substituents on the A-ring, thus altering the pKa for the
carboxylic acid. For example, the addition of a chloride group at the
fourth position on the A-ring decreased the inhibition potency by
more than 2-fold for all the AKR1C isozymes, comparing 1 and 4
(AKR1C1 KI ϭ 0.81 ␮M for 1 compared with KI ϭ 3.1 ␮M for 4,
AKR1C2 KI ϭ 0.22 ␮M for 1 compared with KI ϭ 0.41 ␮M for 4, and
AKR1C3 KI ϭ 0.3 ␮M for 1 compared with KI ϭ 2.9 ␮M for 4). The
effect of the halogen on the A-ring decreases the pKa of the carboxylic
acid, thereby increasing the strength of the carboxylic acid and
consequently affecting the overall inhibition potency. The relative
pKa may be an important factor that influences the overall inhibition
potency by increasing or decreasing the propensity of the acid to be
deprotonated, and is further influenced by the presence of the amine.
Steroid Carboxylates—Isoform Selectivity, Potency, and
Trends. The steroid carboxylates tested were potent inhibitors for
Fig. 6. Inhibition of AKR1C isozymes and COX isozymes by mefenamic
acid. IC50 values were determined as described under Materials and
Methods and performed in triplicate. E, AKR1C1; F, AKR1C2; Ⅺ,
AKR1C3; f, COX-1; ‚, COX-2.
all the AKR1C isozymes; however, these compounds were more se-
lective for AKR1C2 than AKR1C1 or AKR1C3. The selectivity for
AKR1C2 versus AKR1C1 and AKR1C3 ranges from greater than
48-fold for 11 to greater than 570-fold for 10 when KI values are
compared. Increasing the side-chain length and changing the acid at
position C24 resulted in the decrease of inhibition potency for all the
AKR1C isozymes. For example, by comparing the KI values for 10
with 13 and 14, a decrease in potency is seen for the inhibition of
AKR1C1 (10 is 7-fold more potent than 13 and 2-fold more potent
than 14), AKR1C2 (10 is 41-fold more potent than 13 and 11-fold
more potent than 14), and AKR1C3 (10 is 20-fold more potent than
13 and 9-fold more potent than 14). The SAR with steroid carboxy-
lates indicates that replacement of the carboxylic acid (13) with a
more electronegative acid, such as sulfonic acid (14), increases the
potency by more than 2-fold for all the AKR1C isozymes (AKR1C1 KI
ϭ 23 ␮M for 14 compared with KI ϭ 66 ␮M for 13, AKR1C2 KI ϭ 0.2
␮M for 14 compared with KI ϭ 0.72 ␮M for 13, and AKR1C3 KI ϭ 99
␮M for 14 compared with KI ϭ 206 ␮M for 13). The importance of the
acid was substantiated when the carboxylic acid for compound 11
was esterified with an ethyl group (15). This small change resulted in
a severe decrease in inhibition potency for all the AKR1C isozymes
(compare 11 and 15). The inhibition potency of 15 for AKR1C1
decreased by more than 9-fold (IC50 ϭ 9.2 ␮M for 11 and IC50 ϾϾ 100
␮M for 15), the inhibition potency for AKR1C2 decreased by more
than 28-fold (IC50 ϭ 0.09 ␮M for 11 and IC50 ϭ 2.5 ␮M for 15), and
inhibition potency for AKR1C3 decreased by more than 64-fold (IC50
ϭ 1.57 ␮M for 11 and IC50 Ͼ 100 ␮M for 15).
Fig. 7. Inhibition of AKR1C isozymes and COX isozymes by 4-chloro-N-
phenylanthranilic acid. IC50 values were determined as described under
Materials and Methods and performed in triplicate. E, AKR1C1; F,
AKR1C2; Ⅺ, AKR1C3; f, COX-1; ‚, COX-2.
Inhibition of COX and AKR1C Isozymes by the COX-2
Selective Inhibitor Celecoxib
Recent data suggest that the COX-2 selective inhibitor celecoxib
may have other targets at physiological concentrations. Therefore,
celecoxib (16) was tested against the AKR1C isozymes and was found
to be a potent inhibitor for all the AKR1C isozymes tested (Table 1)
yielding the following rank order AKR1C3 (KI ϭ 5.2 ␮M) Ͼ AKR1C2
(KI ϭ 8.6 ␮M) Ͼ AKR1C1 (KI ϭ 14.9 ␮M). It was interesting that
when celecoxib was tested as a reversible inhibitor, there was no
Fig. 8. Inhibition of AKR1C isozymes and COX isozymes by 5␤-cholanic
acid-3-one. IC50 values were determined as described under Materials
and Methods and performed in triplicate. E, AKR1C1; F, AKR1C2; Ⅺ,
AKR1C3; f, COX-1; ‚, COX-2.

NSAID Analogs That Target Human AKR1Cs and Not COX
67
difference in the inhibition potency between COX-1 (IC50 ϭ 23.5 ␮M) differently (Fig. 9). The A-ring of the anthranilic acid (con-
and COX-2 (IC50 ϭ 25.0 ␮M) (data not shown). However, when
celecoxib was tested as a time-dependent inhibitor, inhibition on
COX-2 was more than 600-fold more selective (COX-2 IC50 ϭ 0.04
␮M versus COX-1 IC50 ϭ 23.7 ␮M), which has been reported previ-
ously (Gierse et al., 1996) (Table 1). No time dependence was noticed
for the inhibition of the AKR1C isozymes (data not shown).
taining the carboxylate) can tolerate bulky substitutions be-
cause the substituent projects into the larger steroid-binding
pocket. Therefore, AKR1C isoform selectivity may not be
achieved by only modifying the A-ring.
Comparison of the flufenamic acid and ursodeoxycholate
complex structures also indicated that the B-ring of the
N-phenylanthranilic acid is bound in a smaller pocket formed
from amino acids 118, 120, 167, 306, and 311 on the C-loop
Discussion
Classic NSAIDs and selective COX-2 inhibitors also po- adjacent to the larger steroid binding pocket. This smaller
tently inhibit AKR1C isozymes implicated in regulating li- binding pocket may be exploited to develop AKR1C isoform
gand concentrations for nuclear receptors. We show that selective inhibitors, given that variability exists in the size of
AKR1C isozymes are inhibited by celecoxib at concentrations this pocket. Structural data indicate that this additional
similar to its peak plasma concentrations (Chow et al., 2004). binding pocket is larger in AKR1C3 than in AKR1C2. Re-
To discern the cancer chemopreventive properties of these placement of amino acids Phe319, Phe118, and Leu308 in
drugs, we report NSAID analogs and steroid carboxylates AKR1C2 for Tyr319, Ser118, and Ser308 in AKR1C3 and the
that inhibit the AKR1C enzymes but do not inhibit COX positioning of the main chain at Ser308 cause an increase in
isozymes. Small changes in the structures of N-phenylan- the pocket size for AKR1C3.
thranilic acid derivatives resulted in a loss of COX-inhibition,
Inspection of COX crystal structures with bound NSAIDs
but AKR1C inhibition was retained. Using these compounds, explains the observed selectivity of the NSAID analogs for
the roles of AKR1C and COX-isozymes in malignancies af- AKR1C isozymes (Selinsky et al., 2001). In these structures,
fected by trans-activation of nuclear receptors can now be the NSAID pocket is composed of a long, narrow hydrophobic
discerned.
channel with a constricted binding region defined by amino
N-Phenylanthranilic Acid Derivatives. Using the SAR acids His90, Arg120, Tyr355, and Glu524, where the A-ring
criteria for N-phenylanthranilic acid inhibition of COX, we of a N-phenylanthranilic acid binds. Therefore, COX cannot
developed analogs that were selective for the AKR1C isozymes, tolerate substitutions on the A-ring as easily as AKR1C
whereas their effects on COX isozymes were minimal. Changes isozymes.
to the N-phenylanthranilic acid scaffold (A-ring substitution
AKR1C Isozymes and Neoplastic Diseases. NSAIDs
and heteroatom substitution) dramatically decreased COX are effective in the treatment of breast, prostate, lung, and
isozyme inhibition in some instances by more than 30-fold colon cancers (Marnett, 1995; Harris et al., 2000; Wick et al.,
(compounds with IC50 Ն 1000 ␮M), but potent inhibition of 2002; Sanchez-Alcazar et al., 2003). It is believed that these
the AKR1C isozymes was retained.
beneficial effects are mediated through inhibition of COX-2.
The selectivity between the AKR1C isozymes was less dra- Human AKR1C isozymes are also potently inhibited by both
matic than that observed between the AKR1C and COX classic and COX-2 selective NSAIDs and may be an alternate
isozymes. However, movement of the carboxylic acid from the target for their antineoplastic effects.
ortho to the para position on the A-ring favored AKR1C3
inhibition, and/or additional substitution of the B-ring fa- differentiation of HL-60 cells by preventing the PPAR␥ re-
vored isoform selectivity.
ceptor from being activated by its ligand, 15-⌬12,14-PGJ2.
Desmond et al. (2003) showed that AKR1C3 prevented the
Steroid Carboxylate Derivatives. The steroid carboxy- This indicated that the AKR1C isozymes are able to regulate
lates, including the bile acids, were known selective inhibi- ligand access to orphan nuclear receptors and may be in-
tors for the AKR1C isoforms with preference for AKR1C2. volved in the progression of neoplastic diseases. Inhibition of
Bile acids show 100-fold selectivity for AKR1C2 over AKR1C3 could lead to the activation of PPAR␥ receptor,
AKR1C1 yet differ by a single amino acid at the active site. which induces differentiation, is antiproliferative, and re-
The structural basis for this difference in inhibition potency sults in apoptosis in many cell types and cancers (Wick et al.,
is explained by the AKR1C2-NADPϩ-Ursodeoxycholate crys-
tal structure, which indicated that replacement of Val54 in
AKR1C2 to a leucine in AKR1C1 disrupts binding of the bile
acids (Jin et al., 2001). The significant finding is that these
steroid carboxylates do not inhibit COX isozymes. Further
selectivity in these compounds for AKR1Cs may be achieved
by restricting the rotation around the steroid side chain.
Comparison of the Ligand Binding Sites of AKR1C
Isozymes and COX. Crystal structures exist for the AKR1C
isozymes with steroid carboxylates (AKR1C2-NADPϩ-urso-
deoxycholate, Protein Data Bank entry 1IHI) and NSAIDs
(AKR1C3-NADPϩ-flufenamic acid, Protein Data Bank entry
1S2C) (Jin et al., 2001, and Lovering et al., 2004, respec-
tively). Both ternary complex structures have the carboxylic
Fig. 9. Positions of flufenamic acid (FLU) and ursodeoxycholate (URS) in
AKR1C2 (blue) and AKR1C3 (gold). Ursodeoxycholate and flufenamic
acid are in ball-and-stick representation with their carboxylic acid oxygen
(red) anchored by Tyr55. These two structures were superimposed using
acid of the inhibitor bound to the catalytic tyrosine and can
provide starting points for inhibitor design. When the two
active sites in these ternary complexes are overlaid, it is
apparent that flufenamic acid and ursodeoxycholate bind the C␣ chain as a reference with an RMSD of 1.0 Å.

68
Bauman et al.
prostaglandin G/H synthase from baculovirus-infected insect cells. Protein Expr
Purif 7:19–26.
2002; Desmond et al., 2003). Therefore, the beneficial effects
of NSAIDs as cancer chemopreventives could be through the
inhibition of the AKR1C isozymes rather than by COX.
COX-2–selective inhibitors, such as celecoxib, are also effec-
tive in decreasing the proliferation of breast cancer (Marnett,
1995; Harris et al., 2000; Wick et al., 2002; Davies, 2003;
Sanchez-Alcazar et al., 2003; DuBois, 2004). Nonselective and
selective NSAIDs are believed to act beneficially in the breast
by inhibiting COX-2 and decreasing PGE2 levels (Davies, 2003;
DuBois, 2004). PGE2 induces the transcription of cyp19-aro-
matase, which converts testosterone to the potent estrogen 17␤-
estradiol, thereby increasing active estrogens in the breast
(Zhao et al., 1996; Davies, 2003; DuBois, 2004). Therefore, in-
hibition of COX-2 leads to a decrease in the production of 17␤-
estradiol. An alternate mechanism may involve the inhibition of
AKR1C3, which is localized in the secretory epithelial cells in
both the acini and terminal ducts of the mammary gland (Pel-
letier et al., 2001). AKR1C3 not only reduces estrone (a weak
estrogen) to 17␤-estradiol (a potent estrogen) but also reduces
androstenedione to testosterone, which can be aromatized to
17␤-estradiol. AKR1C3 inhibition by NSAIDs would deprive
the ER of its ligand and provide a mechanism for their antineo-
plastic effects on hormone-dependent breast cancer.
Gierse JK, McDonald JJ, Hauser SD, Rangwala SH, Koboldt CM, and Seibert K (1996)
A single amino acid difference between cyclooxygenase-1 (COX-1) and -2 (COX-2)
reversus the selectivity of COX-2 specific inhibitors. J Biol Chem 271:15810–15814.
Harris RE, Alshafie GA, Abou-Issa H, and Seibert K (2000) Chemoprevention of
breast cancer in rats by celecoxib, a cyclooxygenase 2 inhibitor. Cancer Res 60:
2101–2103.
Jin Y, Stayrook SE, Albert RH, Palackal NT, Penning TM, and Lewis M (2001) Crystal
structure of human type 3 3alpha-hydroxysteroid dehydrogenase/bile acid binding protein
complexed with NADPϩ and ursodeoxycholate. Biochemistry 40:10161–10168.
Keller JJ and Giardiello FM (2003) Chemoprevention strategies using NSAIDs and
COX-2 inhibitors. Cancer Biol Ther 2(Suppl 1):S140–S149.
Labrie F, Luu-The V, Labrie C, and Simard J (2001) DHEA and its transformation
into androgens and estrogens in peripheral target tissues: intracrinology. Front
Neuroendocrinol 22:185–212.
Laneuville O, Breuer DK, Dewitt DL, Hla T, Funk CD, and Smith WL (1994)
Differential inhibition of human prostaglandin endoperoxide H synthases-1 and -2
by nonsteroidal anti-inflammatory drugs. J Pharmacol Exp Ther 271:927–934.
Leng J, Han C, Demetris AJ, Michalopoulos GK, and Wu T (2003) Cyclooxygenase-2
promotes hepatocellular carcinoma cell growth through Akt activation: evidence
for Akt inhibition in celecoxib-induced apoptosis. Hepatology 38:756–768.
Lombardino JG (1985) Nonsteroidal Antiinflammatory Drugs pp 251–431, Wiley
Interscience, New York.
Lovering AL, Ride JP, Bunce CM, Desmond JC, Cummings SM, and White SA (2004)
Crystal structures of prostaglandin D(2) 11-ketoreductase (AKR1C3) in complex
with the nonsteroidal anti-inflammatory drugs flufenamic acid and indomethacin.
Cancer Res 64:1802–1810.
Marnett LJ (1995) Aspirin and related nonsteroidal anti-inflammatory drugs as
chemopreventive agents against colon cancer. Prev Med 24:103–106.
Matsuura K, Deyashiki Y, Sato K, Ishida N, Miwa G, and Hara A (1997) Identifica-
tion of amino acid residues responsible for differences in substrate specificity and
inhibitor sensitivity between two human liver dihydrodiol dehydrogenase isoen-
zymes by site-directed mutagenesis. Biochem J 323:61–64.
Matsuura K, Shiraishi H, Hara A, Sato K, Deyashiki Y, Ninomiya M, and Sakai S
(1998) Identification of a principle mRNA species for human 3alpha-hydroxy-
steroid dehydrogenase isoform (AKR1C3) that exhibits high prostaglandin D2
11-ketoreductase activity. J Biochem 124:940–946.
Ozaki K, Yamada Y, Oine T, Ishizuka T, and Iwasawa Y (1985) Studies on 4(1H)-
quinazolinones. 5. Synthesis and antiinflammatory activity of 4(1H)-quinazolin-
one derivatives. J Med Chem 28:568–576.
Pelletier G, Luu-The V, El-Alfy M, Li S, and Labrie F (2001) Immunoelectron
microscopic localization of 3beta-hydroxysteroid dehydrogenase and type 5 17beta-
hydroxysteroid dehydrogenase in the human prostate and mammary gland. J Mol
Endocrinol 26:11–19.
Acknowledgments
The COX-2 protein used in this study was a kind gift of Michelle V.
Williams (University of Pennsylvania School of Medicine, Philadelphia,
PA). A preliminary account of this work was presented at the 85th
Annual Meeting of the Endocrine Society in Philadelphia, 2003, and
again at the 95th Annual Meeting of the American Association for
Cancer Research in Orlando in 2004. A provisional patent application
has been submitted to the United States Patent and Trademark Office
for the development of selective intracrine modulators.
Penning TM (1997) Molecular endocrinology of hydroxysteroid dehydrogenases. En-
docr Rev 18:281–305.
Penning TM, Burczynski ME, Jez JM, Hung CF, Lin HK, Ma H, Moore M, Palackal
N
and Ratnam K (2000) Human 3alpha-dehydrogenase isoforms (AKR1C1-
References
AKR1C4) of the aldo-keto reductase superfamily: functional plasticity and tissue
distribution reveals roles in the inactivation and formation of male and female sex
hormones. Biochem J 351:67–77.
Askonas LJ, Ricigliano JW, and Penning TM (1991) The kinetic mechanism cata-
lyzed by homogeneous rat liver 3 alpha-hydroxysteroid dehydrogenase. Evidence
for binary and ternary dead-end complexes containing non-steroidal anti-
inflammatory drugs. Biochem J 278:835–841.
Barton DHR, Wozniak J, and Zard SZ (1989) A short and efficient degradation of the
bile acid side chain. Some novel reactions of sulphines and ␣-ketoesters. Tetrahe-
dron 45:3741–3754.
Benedetto C, Mc Donald-Gibson RG, Nigram S, and Slater TF (1987) Prostaglandins
and Related Substances: A Practical Approach, pp 209–227, IRL Press, Oxford
Burczynski ME, Harvey RG, and Penning TM (1998) Expression and characteriza-
tion of four recombinant human dihydrodiol dehydrogenase isoforms: oxidation of
trans-7, 8-dihydroxy-7,8-dihydrobenzo[a]pyrene to the activated o-quinone metab-
olite benzo[a]pyrene-7,8-dione. Biochemistry 37:6781–6790.
Cheng Y and Prusoff WH (1973) Relationship between the inhibition constant (K1)
and the concentration of inhibitor which causes 50 per cent inhibition (I50) of an
enzymatic reaction. Biochem Pharmacol 22:3099–3108.
Chow HH, Anavy N, Salazar D, Frank DH, and Alberts DS (2004) Determination of
celecoxib in human plasma using solid-phase extraction and high-performance
liquid chromatography. J Pharm Biomed Anal 34:167–174.
Copeland RA, Williams JM, Giannaras J, Nurnberg S, Covington M, Pinto D, Pick S,
and Trzaskos JM (1994) Mechanism of selective inhibition of the inducible isoform
of prostaglandin G/H synthase. Proc Natl Acad Sci USA 91:11202–11206.
Davies GL (2003) Cyclooxygenase-2 and chemoprevention of breast cancer. J Steroid
Biochem Mol Biol 86:495–499.
Desmond JC, Mountford JC, Drayson MT, Walker EA, Hewison M, Ride JP, Luong
QT, Hayden RE, Vanin EF, and Bunce CM (2003) The aldo-keto reductase
AKR1C3 is a novel suppressor of cell differentiation that provides a plausible
target for the non-cyclooxygenase-dependent antineoplastic actions of nonsteroidal
anti-inflammatory drugs. Cancer Res 63:505–512.
Penning TM and Talalay P (1983) Inhibition of a major NAD(P)-linked oxidoreduc-
tase from rat liver cytosol by steroidal and nonsteroidal anti-inflammatory agents
and by prostaglandins. Proc Natl Acad Sci USA 80:4504–4508.
Reid G, Wielinga P, Zelcer N, van der Heijden I, Kuil A, de Haas M, Wijnholds J, and
Borst P (2003) The human multidrug resistance protein MRP4 functions as a
prostaglandin efflux transporter and is inhibited by nonsteroidal antiinflamma-
tory drugs. Proc Natl Acad Sci USA 100:9244–9249.
Rizner TL, Lin HK, Peehl DM, Steckelbroeck S, Bauman DR, and Penning TM (2003)
Human type 3 3alpha-hydroxysteroid dehydrogenase (Aldo-keto reductase 1C2)
and androgen metabolism in prostate cells. Endocrinology 144:2922–2932.
Sanchez-Alcazar JA, Bradbury DA, Pang L, and Knox AJ (2003) Cyclooxygenase
(COX) inhibitors induce apoptosis in non-small cell lung cancer through cycloox-
ygenase independent pathways. Lung Cancer 40:33–44.
Selinsky BS, Gupta K, Sharkey CT, and Loll PJ (2001) Structural analysis of NSAID
binding by prostaglandin H2 synthase: time-dependent and time-independent
inhibitors elicit identical enzyme conformations. Biochemistry 40:5172–5180.
Scherrer RA (1974) Antiinflammatory Agents; Chemistry and Pharmacology pp
45–89, Academic Press, New York.
Smith T, Leipprandt J, and DeWitt D (2000) Purification and characterization of the
human recombinant histidine-tagged prostaglandin endoperoxide H synthases-1
and -2. Arch Biochem Biophys 375:195–200.
Tegeder I, Pfeilschifter J, and Geisslinger G (2001) Cyclooxygenase-independent
actions of cyclooxygenase inhibitors. FASEB J 15:2057–2072.
Wick M, Hurteau G, Dessev C, Chan D, Geraci MW, Winn RA, Heasley LE and
Nemenoff RA (2002) Peroxisome proliferator-activated receptor-␥ is a target of
nonsteroidal anti-inflammatory drugs mediating cyclooxygenase-independent in-
hibition of lung cancer cell growth. Mol Pharmacol 62:1207–1214.
Zhao Y, Agarwal VR, Mendelson CR, and Simpson ER (1996) Estrogen biosynthesis
proximal to a breast tumor is stimulated by PGE2 via cyclic AMP, leading to activa-
tion of promoter II of the CYP19 (aromatase) gene. Endocrinology 137:5739–5742.
DuBois RN (2004) Aspirin and breast cancer prevention: the estrogen connection.
J Am Med Assoc 29:2488–2489.
Dufort I, Labrie F, and Luu-The
V (2001) Human types 1 and 3 3 alpha-
hydroxysteroid dehydrogenases: differential lability and tissue distribution. J Clin
Endocrinol Metab 86:841–846.
Dufort I, Rheault P, Huang XF, Soucy P, and Luu-The V (1999) Characteristics of a highly
labile human type 5 17beta-hydroxysteroid dehydrogenase. Endocrinology 140:568–574.
George HJ, Van Dyke DE, Straney RA, Trzaskos JM, and Copeland RA (1996)
Expression purification and characterization of recombinant human inducible
Address correspondence to: Dr. Trevor M. Penning, Department of Phar-
macology, University of Pennsylvania School of Medicine, 130C John Morgan
Building, 3620 Hamilton Walk, Philadelphia, PA 19104-6084. E-mail:
penning@pharm.med.upenn.edu