Structural Determinants of the Selectivity of 3-Benzyluracil-1-acetic Acids toward Human Enzymes Aldose Reductase and AKR1B10

Article abstract of DOI:10.1002/cmdc.201500393

The human enzymes aldose reductase (AR) and AKR1B10 have been thoroughly explored in terms of their roles in diabetes, inflammatory disorders, and cancer. In this study we identified two new lead compounds, 2-(3-(4-chloro-3-nitrobenzyl)-2,4-dioxo-3,4-dihy

Full text of DOI:10.1002/cmdc.201500393

DOI: 10.1002/cmdc.201500393
Full Papers
Structural Determinants of the Selectivity of 3-
Benzyluracil-1-acetic Acids toward Human Enzymes Aldose
Reductase and AKR1B10
Francesc X. Ruiz,*[a] Alexandra Cousido-Siah,[a] Sergio PortØ,[b] Marta Domínguez,[c]
Isidro Crespo,[b] Chris Rechlin,[d] AndrØ Mitschler,[a] ngel R. de Lera,[c] María Jesffls Martín,[e]
Jesffls ngel de la Fuente,[e] Gerhard Klebe,[d] Xavier ParØs,[b] Jaume FarrØs,[b] and
Alberto Podjarny*[a]
The human enzymes aldose reductase (AR) and AKR1B10 have
been thoroughly explored in terms of their roles in diabetes,
inflammatory disorders, and cancer. In this study we identified
two new lead compounds, 2-(3-(4-chloro-3-nitrobenzyl)-2,4-
dioxo-3,4-dihydropyrimidin-1(2H)-yl)acetic acid (JF0048, 3) and
2-(2,4-dioxo-3-(2,3,4,5-tetrabromo-6-methoxybenzyl)-3,4-dihy-
dropyrimidin-1(2H)-yl)acetic acid (JF0049, 4), which selectively
target these enzymes. Although 3 and 4 share the 3-benzylura-
cil-1-acetic acid scaffold, they have different substituents in
their aryl moieties. Inhibition studies along with thermodynam-
ic and structural characterizations of both enzymes revealed
that the chloronitrobenzyl moiety of compound 3 can open
the AR specificity pocket but not that of the AKR1B10 cognate.
In contrast, the larger atoms at the ortho and/or meta positions
of compound 4 prevent the AR specificity pocket from open-
ing due to steric hindrance and provide a tighter fit to the
AKR1B10 inhibitor binding pocket, probably enhanced by the
displacement of a disordered water molecule trapped in a hy-
drophobic subpocket, creating an enthalpic signature. Further-
more, this selectivity also occurs in the cell, which enables the
development of a more efficient drug design strategy: com-
pound 3 prevents sorbitol accumulation in human retinal
ARPE-19 cells, whereas 4 stops proliferation in human lung
cancer NCI-H460 cells.
Introduction
Human aldose reductase (AR, ALR2, or AKR1B1) and AKR1B10
are enzymes of biomedical interest because of their involve-
ment in diabetes (AR) and in cancer (AR and AKR1B10). They
belong to the aldo–keto reductase superfamily (AKR) and are
NADPH-dependent enzymes, folding into a highly conserved
(a/b)8 barrel. They have different substrate specificity and in-
hibitor selectivity due to residue differences in their three ex-
ternal and variable loops.[1]
[a] Dr. F. X. Ruiz,+ ++ A. Cousido-Siah,+ Dr. A. Mitschler, Dr. A. Podjarny
Department of Integrative Biology
AR presents ubiquitous tissue expression and catalyzes the
reduction of glucose and a wide range of aldehydes, remarka-
bly, glutathione conjugates of lipid peroxidation products. In-
creased AR activity results in NADPH depletion, sorbitol accu-
mulation, decreased cellular levels of reduced glutathione and
oxidative stress,[2] contributing to inflammation-related diseas-
es (e.g., atherosclerosis, sepsis, asthma, colorectal carcinoma
[AR inhibition prevented colon cancer metastasis in mice]).[3]
AKR1B10 shares 71% amino acid sequence identity with AR,
but its tissue expression is restricted mainly to colon and small
intestine. It is overexpressed in lung and liver cancers, among
others, making it a potential cancer diagnostic and/or prognos-
tic marker.[4] AKR1B10 catalyzes the reduction of chemothera-
peutic drugs, lipid peroxidation of free aldehydes and all-trans-
retinaldehyde, but it does not catalyze the reduction of glu-
cose.[5] AKR1B10 may impact the carcinogenesis process
through its involvement in several pathways, for example, reti-
noic acid signaling,[4b,6] lipid aldehyde detoxification, and lipid
synthesis.[4c,7]
Institut de GØnØtique et de Biologie MolØculaire et Cellulaire
CNRS, INSERM, UdS, rue Laurent Fries, 67404 Illkirch CEDEX (France)
E-mail: fxavier.ruiz@gmail.com
podjarny@igbmc.fr
[b] Dr. S. PortØ, I. Crespo, Prof. Dr. X. ParØs, Prof. Dr. J. FarrØs
Department of Biochemistry and Molecular Biology
Universitat Autònoma de Barcelona, 08193 Bellaterra, Barcelona (Spain)
[c] Dr. M. Domínguez, Prof. Dr. . R. de Lera
Departmento de Química Orgµnica and
Centro de Investigaciones BiomØdicas (CINBIO)
Universidade de Vigo, 363100 Vigo (Spain)
[d] C. Rechlin, Prof. Dr. G. Klebe
Institute of Pharmaceutical Chemistry
University of Marburg, Marbacher Weg 6, 35032 Marburg (Germany)
[e] Dr. M. J. Martín, Dr. J. . de la Fuente
Biomar Microbial Technologies S.A.
Parque Tecnológico de León, 24009 León (Spain)
[+] These authors contributed equally to this work.
++] Present address: Center for Advanced Biotechnology and Medicine, Depart-
ment of Chemistry and Chemical Biology, Rutgers University, Piscataway, NJ
08854-5627 (USA)
[
Despite the fact that a large number of AR inhibitors (ARIs)
have been designed and even assayed in clinical trials, epalre-
Supporting information for this article is available on the WWW under
http://dx.doi.org/10.1002/cmdc.201500393.
ChemMedChem 2015, 10, 1989 – 2003
1989
ꢀ 2015 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim

Full Papers
stat (16) (all chemical structures apart from those involved in
synthesis are shown in Scheme S1 of the Supporting Informa-
tion) is the only ARI approved for therapy, although restricted
to Japan, China, and India. The failure of ARIs as therapeutic
agents has been mainly due to poor pharmacokinetic proper-
ties, lack of clinical efficacy, and/or unacceptable side effect-
s.[2a,b,8] Most ARIs can be roughly grouped in those containing:
1) a carboxylic acid moiety, or 2) a cyclic imide group. These
moieties bind to the anion-binding pocket, formed by Tyr48,
His110, Trp111, and the positively charged nicotinamide
moiety of the cofactor. Most ARIs possess a second moiety
with one or more aromatic groups, which bind into a hydro-
phobic subpocket of AR, bordered by Trp111, Phe122 and
Leu300. This subpocket is open, due to a conformational
change and is known as the specificity pocket.[9]
Figure 1. Structures of the polybrominated diphenyl ether compound from
the marine sponge Disydea herbacea (2,3,4,5-tetrabromo-6-(3,5-dibromo-2-
hydroxyphenoxy)phenol) (1), JF0064 (2), JF0048 (3), and JF0049 (4).
Initially, the only enzyme analyzed for off-target effects of
ARIs was aldehyde reductase (or AKR1A1).[5a,10] Several studies
have recently addressed the selectivity of ARIs compared with
the more closely related AKR1B10. Gallego and co-workers[6a]
solved the first three-dimensional structure of AKR1B10 in
complex with NADP+ and tolrestat (18); subsequently, other
researchers have reported potent and selective AKR1B10 inhib-
itors using in vitro assays.[11] In the last two years, additional
AKR1B10 three-dimensional structures have been solved. Crys-
tal structures of the AKR1B10 holoenzyme and of the ternary
complexes with several ARIs and with AKR1B10 inhibitors were
reported by the Hu group.[12] They observed that the orienta-
tion of the Trp112 side chain is critical for the selectivity of
selectivity of 3 for AR was justified by its ability to open the
enzyme specificity pocket, while in AKR1B10, this ligand might
adopt a different conformation, resulting in weaker binding.
On the other hand, the selectivity of 4 for AKR1B10 is probably
due to the inability of its bulkier aryl moiety to occupy the AR
specificity pocket and its improved fitting into the larger
AKR1B10 active site, including a subpocket defined by loop A.
This study paves the way for future efforts in structure-guided
drug discovery, directed to both AR and the relatively unex-
plored AKR1B10.
AKR1B10
inhibitors.
Meanwhile,
we
solved
1) the
AKR1B10 V301L–fidarestat (17) complex,[13] which served as the
basis to explain the selectivity of 17 for AR; 2) the methylated
AKR1B10K125R/V301L–2 complex,[1] which characterized as Results and Discussion
a novel lead a tetrafluorophenol moiety that targets both AR
Chemistry
and AKR1B10; and 3) the methylated AKR1B10K125R/V301L–
sulindac (26) complex, which showed that 26 and its sulfone
might be a drug lead for selective AR and AKR1B10 inhibi-
tors.[14]
The synthesis of 3-benzyluracil-1-acetic acid derivatives 3 and
4 followed the methodology described for the latter com-
pound,[16] involving the alkylation of ethyl 2-(2,4-dioxo-3,4-dihy-
dropyrimidin-1(2H)-yl)acetate 5[17] with the corresponding ben-
zylbromides 11 and 8, respectively (Scheme 1). Benzylic bro-
mides 8 and 11 were obtained in 96 and 47% yields, respec-
tively, by treatment of 7 (prepared by methylation of commer-
cially available 3,4,5,6-tetrabromocresol 6 with K2CO3 and MeI
in acetonitrile) and 10 using bromine under photochemical ir-
radiation conditions. A second alkylation of the monoalkylated
uracil 5 with 8 and 11 using NaH in DMF provided the N,N-dia-
lkylated uracil derivatives 9 and 12, respectively, in 85% and
75% yield. Finally, saponification of the esters afforded the de-
sired carboxylic acids 3 and 4.
Previously, Biomar Microbial Technologies screened a library
of about 2000 marine natural products and found a polybromi-
nated diphenyl ether compound (1, Figure 1) from a marine
sponge (Disydea herbacea) that showed an IC50 value of 6.4 mm
against AR. This hit led to the synthesis of different series of
synthetic polyhalogenated compounds (such as 2), some of
which targeted both AR and AKR1B10.[1,15]
In the search for novel ligands selective for AKR1B10, we fo-
cused our attention on compound JF0048 (3), which was previ-
ously shown to be selective for rat lens AR versus rat kidney al-
dehyde reductase.[16] We surmised that the modification of the
3-benzyluracil-1-acetic acid scaffold through the incorporation
of additional substituents at the (poly)halogenated aryl moiety,
reminiscent of the sponge natural products (i.e., compound 1,
Figure 1), might shift the selectivity of the designed compound
to target AKR1B10 instead.
Inhibition assays and thermodynamic signature of AR/
AKR1B10 complexes with 3 and 4
Kotani et al.[16] previously assayed 3 against rat lens AR and rat
kidney aldehyde reductase, which were purified from the cor-
responding tissues. They found that 3 was one of the best hits
for rat AR, with more than 4000-fold selectivity versus rat alde-
hyde reductase. We confirmed that 3 exhibited strong inhibi-
In the present work, this assumption was indeed confirmed,
as compound JF0049 (4) (Figure 1) was found to be selective
for AKR1B10. In addition, structural insight was gained from
several X-ray structures of the corresponding complexes. The
ChemMedChem 2015, 10, 1989 – 2003
www.chemmedchem.org
1990
ꢀ 2015 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim

Full Papers
Scheme 1. Reagents and conditions: a) MeI, K2CO3, MeCN, reflux, 4 h, 95%; b) Br2, CCl4, tungsten lamp 250 W, 908C, 4 h (8, 96%; 11, 48%); c) NaH, DMF, 0!
258C, 3 h (9, 85%; 12, 75%); d) 10% NaOH, MeOH, 808C, 3 h (4, 90%; 3, 80%).
tion against human AR (IC50 =170 nm),[9,18] and was 127-fold
and >4000-fold more potent toward human AR than toward
AKR1B10 and AKR1A1, respectively. On the other hand, 4 inhib-
ited AKR1B10 with an IC50 value of 450 nm, being eightfold
and >200-fold more potent than for AR and AKR1A1, respec-
tively (Table 1).
agreement with the inhibition assays for the binding of 3 and
4 to AR. For the binding of 4 to AKR1B10, the Kd value was
slightly higher than the corresponding IC50 value.
Structure–activity relationships for 3 and 4 with AR and
AKR1B10
We also performed isothermal titration calorimetry (ITC) to
study the binding of 3 and 4 to both enzymes. As shown in
Table 2 (and Figure S1), the Gibbs free binding energy (DG8)
and the dissociation constants (Kd) obtained by ITC were in
With the aim of understanding the rationale behind the inhibi-
tory properties of 3 and 4 with both AR and AKR1B10, we
sought to determine the X-ray structures of the ternary com-
plexes. We were able to obtain all except that of the
AKR1B10–3 complex. The structures folded into the expected
(a/b)8-TIM barrel, with the cofactor position conserved. Table 3
shows the data collection and refinement statistics.
Table 1. IC50 values of halogenated lead compounds.
Compd
IC50 [mm][a]
1B10/AR AR/1B10
AR
AKR1B10
AKR1A1
3
4
0.17Æ0.02 21.6Æ1.7
3.6Æ0.4
>100
127
0.1
0.008
8
(5.6)[b]
Compound 3 opens the AR specificity pocket
0.45Æ0.04 >100
(27.3)[b]
The X-ray structure of the AR-NADP+–3 complex was obtained
at an atomic resolution of 1.0 Š, with the FoÀFc omit map
showing clearly the inhibitor electron density (Figure 2A). The
carboxylic acid of 3 was assumed to be deprotonated, as ob-
[a] Values are the meanÆSEM of n=3 independent experiments. [b] Pa-
rentheses indicate percent inhibition of AKR1A1 at 100 mm.
Table 2. Thermodynamic data of the binding events of 3 and 4 with AR and of 4 with AKR1B10.[a]
[b]
[c]
[b]
[c]
Complex
Kd [nm]
DG8 [kJmolÀ1
]
DH8obs [kJmolÀ1
]
DH8bind [kJmolÀ1
]
ÀTDS8obs [kJmolÀ1
]
ÀTDS8bind [kJmolÀ1
]
AR–NADP+–3
AR–NADP+–4
AKR1B10–NADP+–3[d]
AKR1B10–NADP+–4
91.1Æ39.8
8894.2Æ1451.6
ND
2313.3Æ516.7
À40.4Æ1.3
À28.8Æ0.4
ND
À32.2Æ0.5
À46.7Æ0.5
À10.7Æ1.9
ND
À38.7Æ0.7
À59.7
ND
ND
À41.6
6.1Æ0.6
À18.1Æ1.5
ND
6.0Æ0.6
19.3
ND
ND
9.3
[a] Values are the meanÆSEM of n=3 independent experiments. [b] Thermodynamic data for the measurement in HEPES (DH8obs and ÀTDS8obs). [c] If pos-
sible, the values were corrected for the protonation effect (DH8bind and ÀTDS8bind). [d] The binding of 3 to AKR1B10 could not be measured due to the low
affinity of 3 for AKR1B10. ND: not determined.
ChemMedChem 2015, 10, 1989 – 2003
www.chemmedchem.org
1991
ꢀ 2015 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim

Full Papers
Table 3. Data collection and refinement statistics.
AR–NADP+–3
AR–NADP+–4
AKME2MU–NADP+–4 methylated AKR1B10 holoenzyme AKME2MU holoenzyme
inhibitor conc. [mm]
PDB ID
wavelength [Š]
5
5
4XZI
1
50–2.45
(2.54–2.45)
I222
30
4XZL
1.54178
50-1.70
(1.76–1.70)
P31
–
–
4XZH
4XZM
0.9202
50–1.75
(1.81–1.75)
P31
79.4, 79.4, 49.8
908, 908, 1208
11 2502
35411
3.2 (3.0)
99.0 (100.0)
16.64 (2.97)
25.75
0.065 (0.375)
0.203
0.2499
2861
2625
56
4XZN
0.7085
50–1.00
(1.04–1.00)
P1
46.5, 46.7, 68.3
75.08, 78.98, 74.48 908, 908, 908
390678
246041
1.6 (1.3)
85.9 (73.3)
22.31 (2.44)
8.56
0.028 (0.232)
0.1488
0.161
6175
0.91907
50–1.70
(1.76–1.70)
P31
79.9, 79.9, 50.6
908, 908, 1208
129558
37876
3.4 (3.5)
93.9 (94.3)
14.51 (3.36)
25.79
0.068 (0.322)
0.1955
0.2283
2908
2659
56
resolution [Š]
space group
unit cell [Š]
74.1, 84.8, 105.3 79.1, 79.1, 50.2
908, 908, 1208
172274
70350
2.4 (1.4)
91.0 (66.0)
18.79 (2.77)
23.81
0.048 (0.223)[a]
0.2085
0.2385
2896
total reflections
unique reflections
multiplicity
completeness [%]
mean I/s(I)
Wilson B-factor [Š2]
50932
22204
2.3 (2.0)
94.4 (88.4)
10.67 (2.38)
43.42
0.076 (0.321)[a]
0.2188
0.2895
2716
[b]
Rsym
[c]
Rfactor
[d]
Rfree
number of atoms
macromolecules
ligands
5322
46
2579
98
2667
81
water
807
38
143
178
180
protein residues
RMS bonds [Š]
RMS angles [8]
Ramachandran favored [%]
Ramachandran outliers [%]
clash score
average B-factor [Š2]
macromolecules
solvent
634
0.008
1.32
97
0
12.98
11.48
7.88
316
0.012
1.5
93
0
21.34
26.2
26.5
316
0.009
1.24
98
0
7.53
26.5
26.3
316
0.01
1.41
97
317
0.01
1.48
97
0
0
13.84
29.3
29.1
32.9
13.05
21.1
20.8
26.4
23.55
20.2
30.6
Statistics for the highest-resolution shell are shown in parentheses. [a] Anomalous pairs were treated separately. [b] Rsym =SjIÀ‘I’j/SI, for which I=ob-
served intensity and ‘I’=statistically weighted average intensity of multiple observations of symmetry-related reflections. [c] Rfactor =Sj jFo jÀjFc j j/SjFo j,
for which Fo =observed structure factor amplitudes and Fc =calculated structure factor amplitudes. [d] Rfree: same definition as that for Rfactor for a cross-vali-
dation set of ~5% of the reflections.
served previously for IDD594,[19] with the oxygen atoms acting
Compound 4 is too bulky to fit into the AR specificity pocket
as hydrogen bond acceptors to the Tyr48 OH (2.7 Š), the
His110 Ne2 (2.7 Š), and the Trp111 Ne1 (3.0 Š), and the creation
of an electrostatic contact with the charged nicotinamide N of
NADP+ (4.1 Š). The uracil-1-acetic acid moiety creates hydro-
phobic contacts with Trp20, Trp111, Trp219, and Cys298. Its
aryl moiety is sandwiched between the indole moieties of
Trp20 and Trp111, the latter through a face-to-face oriented
p–p stacking interaction (with approximately a 3.4 Š distance
between the two planes), and the side chains of Phe122,
Leu300, and Cys303 lining the specificity pocket. One of the
oxygen atoms of the nitro group forms a hydrogen bond with
the Leu300 main chain nitrogen atom (2.9 Š), and the chlorine
atom makes a halogen bond with the Thr113 OH moiety
(2.9 Š) (Figure 2A).
The X-ray structure of the AR–NADP+–4 complex was obtained
at 2.45 Š resolution. It must be noted that crystals were ob-
tained by co-crystallization with 5 mm of 4 in a rarely applied
reservoir condition for AR crystallization, replacing the habitual
ammonium citrate, pH 5.0, buffer with Tris-HCl, pH 8.0. This is
the second AR structure obtained by co-crystallization at this
pH, after the ternary complex of the AR holoenzyme with the
inhibitor IDD552 (23) (Protein Data Bank (PDB) ID 1T41).[22] In
addition, the crystal grew in a very unusual space group, I222,
only observed in one AR X-ray crystal structure previously,
a holoenzyme complex (PDB ID 1ABN).[23] The FoÀFc omit map
unequivocally shows the inhibitor electron density in two dif-
ferent conformations (A and B), where the neighboring residue
(Phe122) adopts two different conformations in accordance
with the inhibitor positioning (Figure 2B and 2C). Both confor-
mations of 4 are stabilized by crystallographic contacts with
the complex in the neighboring asymmetric unit. Conformatio-
n A of 4 (60% occupancy) is stabilized by the neighbor confor-
mation A molecule, with a Br–Br contact between the two bro-
mine atoms in para, while conformation B of 4 is stabilized
through halogen bonds with the main chain of loop A residues
of the complex in the neighboring asymmetric unit (Fig-
Additionally, two non-classical hydrogen bonds are formed
between the nitro oxygen atoms and the Tyr309 Cd1 and Ce1
moieties (3.5 Š for both, Figure 2A), as observed for IDD393
(24) and other nitro-containing ARIs.[20] Apparently, the sum of
a halogen bond (as in the case of IDD388 [25] and 22), with
the interactions provided the by nitro substituent at the meta
position, do not trigger stronger binding. This is not surprising,
taking into account that molecular interactions behave in
a highly non-additive fashion.[21]
ChemMedChem 2015, 10, 1989 – 2003
www.chemmedchem.org
1992
ꢀ 2015 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim

Full Papers
Figure 2. X-ray structures of the AR–NADP+–3 and AR–NADP+–4 complexes. A) Atomic representation of the inhibitor binding site for the AR-NADP+–3 com-
plex, with the inhibitor electron density shown as a sA-weighted FoÀFc omit map contoured at the 6s level in grey mesh, and the protein residues, the cofac-
tor and 3 represented as sticks (white, orange, and cyan, respectively); B) Atomic representation of the inhibitor binding site for the AR–NADP+–4 complex
conformation ensemble, with the inhibitor electron density shown as a sA-weighted FoÀFc omit map contoured at the 3s level in grey mesh (with both con-
formations showing the inhibitor bound (top) and separated (down)), and the protein residues in white sticks (except for Phe122, which follows the coloring
of 4), the cofactor in orange sticks, and 4 in green sticks (blue-green for conformation A, olive green for conformation B); C) Atomic representation of the in-
hibitor binding site for the AR–NADP+–4 complex, with each conformation shown in a separate drawing (for color coding, see legend for panel B); D) Atomic
representation of the inhibitor binding site for the AR–NADP+–4 complex conformation ensemble, including the intermolecular crystallographic contacts
with the symmetry-related molecules, in pink lines or sticks for the protein and in violet sticks for the symmetry-related 4, following the same color code as
for the other panels. Hydrogen bond distances are shown as green dashed lines.
The large aryl moiety of 4 binds in the more external
hydrophobic subpocket of AKR1B10
ure 2D). Therefore, the observed interaction of 4 with the AR
holoenzyme confirms that the conformation adopted by 3 in
the cognate complex is not possible for the former compound.
In this regard, we superimposed 4 in the position of 3 in its AR
complex, with the methoxy group oriented toward Ala299 or
Phe122 (Figure 3). In each case, one of the bromine atoms at
the meta position (ortho to the methoxy group) would clash
against the CH2 group of Trp79 (2.4 Š distance). For the ortho
substituents, if the methoxy would be facing Phe122 (Fig-
ure 3A), there would be a clash between the two groups (2.1
and 2.2 Š distance from Phe122 Ce2 and CZ, respectively). If the
methoxy group was in the other orientation (Figure 3B), there
would be steric clashes with the nitrogen and carbon atoms of
the backbone of Ala299 (2.5 Š distance).
In the case of AKR1B10, although new structures of AKR1B10–
NADP+–inhibitor have been solved in the last two years,[12a,24]
we have only succeeded in obtaining the ternary complex
with 4 in the methylated AKR1B10K125R/V301L mutant (from
now on, denoted AKME2MU), as in the case of inhibitors 2 and
26.[1,14] The AKME2MU holoenzyme in complex with 4 was ob-
tained at 1.70 Š resolution, with the FoÀFc omit map showing
the inhibitor, and the anomalous difference map confirming
the position of the four bromine atoms of compound 4 (Fig-
ure 4A). The carboxylic acid oxygen atoms display hydrogen
bonds with the Tyr49 OH (2.9 Š), His111 Ne2 (2.8 Š), and
Trp112 Ne1 moieties of the conformer A (3.0 Š) (Figure 4A). The
electron-deficient uracil ring is also involved in further interac-
tions: 1) with the Trp21 aromatic moiety through a parallel-dis-
placed stacking (3.5 Š distance); and 2) the carbonyl oxygen
ChemMedChem 2015, 10, 1989 – 2003
www.chemmedchem.org
1993
ꢀ 2015 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim

Full Papers
Modeling the binding of 3 to
AKME2MU predicts a geometric
interaction similar to that of 4
The AKME2MU–NADP+–3 com-
plex was solved, but no electron
density corresponding to the in-
hibitor could be distinguished.
We
superimposed
the
AKME2MU–NADP+–3 (PDB, MTZ,
and validation files are given in
the Supporting Information), and
the
AKME2MU
holoenzyme
maps (PDB ID 4XZN, Table 3
shows the data collection and
refinement statistics), and it
could be inferred that the
former corresponds to the
enzyme with the open specificity
pocket (loop C open and Trp112
in flipped conformation) (Fig-
Figure 3. Steric hindrance of 4 within the AR–NADP+–3 active-site conformation. Atomic representation of the in-
hibitor binding site for the AR–NADP+–3 complex with A) 4 superimposed with the methoxy group pointing to
loop A, or B) 4 superimposed with the methoxy group pointing to loop C. The protein residues, the cofactor, 3,
and 4 are shown as sticks (white, orange, cyan, and green, respectively). The short distances, demonstrating 4
steric hindrance in this AR active site conformation, are displayed as red dashed lines and red labels.
atoms are hydrogen bonded to Phe123 Ce1 and to Trp220 CH2,
the two being non-classical hydrogen bonds[20,25] (Figure 4A,
atomic distances of 3.2 Š for both interactions, not shown).
The methylene bridge and the tetrabromomethoxy aryl
moiety are surrounded by a hydrophobic subpocket formed
by Trp80, Trp112, Phe116, Phe123, Ala131, Leu301 (Val301 in
the wild-type enzyme), and the aliphatic parts of Arg125
(Lys125 in the wild-type form) and Gln303 (Figure 4B). The
most significant interactions are the edge-to-face stacking that
the aryl moiety performs with the side chain of Phe123 (3.8 Š
distance, data not shown) and the interactions of two of the
bromine atoms of 4 with the aromatic ring systems of Trp80
and Phe116, respectively (3.5 Š in both cases, data not shown).
Interestingly, Trp112 was observed in two different conforma-
tions (Figure 4C): conformation A (flipped), with 61% occupan-
cy and positioned as in the AKR1B10–18 complex (PDB ID
1ZUA), and conformation B (native), with 39% occupancy and
positioned as in the AKR1B10 holoenzyme structure (PDB ID
4GQG). Trp112 conformation B is stabilized by a specific hydro-
gen bonding network centered on Gln114 conformation B
(58% occupancy) and including Ser304 (Figure 4C). The latter
conformation was also observed for several specific AKR1B10
inhibitors.[12] In fact, the occupancy of 4 was found to be 64%,
which might suggest that two conformations of the protein
coexist in the crystal; conformation A may be that of the
AKME2MU holoenzyme in complex with the inhibitor (with
Trp112 flipped), while conformation B may correspond to that
of the unliganded AKME2MU holoenzyme (with Trp112 in the
native position). In addition, there is an interstitial water mole-
cule (HOH116) interacting with two of the Br atoms of 4 (3.3
and 3.2 Š, respectively) and with Ne1 (2.9 Š) of conformation A
of Trp112 (Figure 4C), which is also present in conformation B,
as it displays 100% occupancy and has no clashes.
ure S2A). Then, the obtained AKME2MU–NADP+–3 coordi-
nates, with no observable electron density for 3, were used as
a macromolecule in AutoDock 4.2[26] in order to generate an
AKME2MU–NADP+–3 model. The ligand coordinates were gen-
erated by superimposing
3 to the observed 4 in the
AKME2MU–NADP+–4 X-ray structure. The pose chosen after
the docking attempt (DLG file in the Supporting Information)
was the one with the ligand ranked highest in DG, with the
closest RMSD to the reference and with conserved hydrogen
bonds between the carboxylic acid oxygen atoms and Tyr49,
His111, and Trp112 side chains, which is prototypical for car-
boxylic AR inhibitors (ARIs).[9]
In the resulting AKME2MU–NADP+–3 model, the inhibitor is
bound in a similar position to that of the cognate AKME2MU–
NADP+–4 (Figure 5B). Indeed, the hydrogen bond pattern with
the carboxylic acid groups is conserved as aforementioned,
but the stacking interaction with Trp21 is lost in the former.
Furthermore, the chloronitrophenyl moiety is involved in an
edge-to-face stacking interaction with the Phe123 side chain
(3.6 Š distance, data not shown), and the nitro group oxygen
atoms are hydrogen bonded to Gln303 Ne2 on one side (2.6 Š)
and to Arg125 NH1 (3.1 Š) in the other. A total of 135 of the
150 docking poses obtained have the nitro group in this posi-
tion (Figure 5A). This conformation could not occur in AR, as
Gln303 in AKR1B10 is Ser302 in AR, and Lys125 in AKR1B10
(Arg125 in the model; however, Lys125 could also be hydro-
gen bonded) is Leu124 in AR.
ITC experiments reveal a strong enthalpic contribution for the
binding of 3 to AR and of 4 to AKR1B10
ITC experiments were performed to gain insight into the ther-
modynamic profiles of the binding events of the inhibitors to
AR and AKR1B10. The binding of the carboxylate-type inhibi-
tors to AR is accompanied by an uptake of 0.7–0.9 protons per
mole of formed complex with Tyr48.[27] To correct for buffer de-
ChemMedChem 2015, 10, 1989 – 2003
www.chemmedchem.org
1994
ꢀ 2015 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim

Full Papers
Figure 4. X-ray structure of the AKME2MU–NADP+–4 complex. A) Atomic
representation of the inhibitor binding site for the AKME2MU–NADP+–4
complex, with the inhibitor electron density shown as a sA-weighted FoÀFc
omit map contoured at the 2s level as a cyan mesh, and as an anomalous
difference map, measured on the bromine edge, contoured at the 2s level
as a purple mesh. The protein residues, the cofactor, and 4 are shown as
white, orange, and gold-yellow sticks, respectively (mutated residues are la-
beled in italics); B) Atomic representation of the inhibitor binding site for
the AKME2MU–NADP+–4 complex, including the surface representation, col-
ored according to the local electrostatic potential (blue, positive charge;
white, neutral; red, negative charge), calculated using PyMOL, the inhibitor
binding site residues and NADP+ as white and orange lines, respectively,
and 4 as gold-yellow sticks; C) Detail of the inhibitor binding site for the
AKME2MU–NADP+–4 complex, following the same color code as in panel A,
and including the interstitial water molecule HOH116 as a red sphere, con-
formation B of Trp112 and Gln114 in green sticks, and the sA-weighted
FoÀFc omit map, contoured at the 2s level as a cyan mesh of the aforemen-
tioned water molecule, Trp112, Gln114, and Ser304. Hydrogen bond distan-
ces are shown in blurred green dashed lines.
tected (Figure S4), which is in good agreement with previous
results for similar ARIs. As already mentioned, the formed con-
tacts of the nitro group in the specificity pocket are linked to
a strong exothermic enthalpic signal.[20] Presumably, these in-
teractions are responsible for the strong enthalpic binding pro-
file of 3 to AR. The binding of 4 to AR could only be measured
in HEPES, as the enthalpic signal was too low to be evaluated
in the other two buffer systems (Figure S1). Nevertheless,
a buffer dependency for the binding of 4 to AR could be de-
tected. In HEPES, the binding of 4 to AR is much less enthalpy-
driven than for 3 (À10.7 vs. À46.7 kJmolÀ1). The binding of
other carboxylate-type inhibitors to AR, as measured by us (un-
published results), showed that the enthalpic contribution nec-
essary to be corrected for the protonation effect is between
À11.5 to À14.0 kJmolÀ1 in HEPES. Both types of ligands—
those accommodating the specificity pocket and those leaving
the pocket closed—show this amount of buffer dependence in
HEPES. It is assumed that the protein binding of 4 is accompa-
nied by similar proton uptake. Even if a similar impact on the
enthalpic signal is anticipated for 3, owing to proton uptake,
the enthalpic term of 4 will be much smaller than that of 3.
Thus, the different binding modes of 3 (addresses the specifici-
ty pocket) and 4 (leaves the specificity pocket unoccupied)
with AR have a significant influence on the thermodynamic
profile.
As 4 binds to AR in two alternative orientations, the inter-
pretation of its thermodynamic profile is even more complicat-
ed. The following aspects were observed: in conformation A of
4, a water molecule is fixed, due to hydrogen bonds from the
oxygen atom of the benzyluracil scaffold of 4 (3.2 Š), the nitro-
gen atom of the Leu300 backbone (2.6 Š), and the nitrogen of
the Ala299 backbone (2.9 Š) (Figure 2C). In conformation B,
a hydrogen bond between the methoxy substituent of 4 and
Trp20 (3.1 Š) can be found. The bromine atoms of 4 form con-
tacts to the polar groups of the AR. In conformation A, such an
interaction is visible with the oxygen of Ser302 and, in confor-
mation B, with the nitrogen of Gln49. However, all of these in-
teractions do not sum to an enthalpic contribution similar to
that of the 4-chloro-3-nitro-phenyl moiety of 3. For AKR1B10,
pendencies, we performed measurements in HEPES, Tricine,
and Tris buffer and plotted the measured enthalpy (H8obs
against the heat of ionization (H8ion) of the corresponding
buffer system.[28] The corrected thermodynamic profile of 3 to
)
AR showed
a
strong favorable enthalpic contribution
(À59.7 kJmolÀ1), while the entropic term was unfavorable (+
19.3 kJmolÀ1). An entrapment of 0.6 protons per mole was de-
ChemMedChem 2015, 10, 1989 – 2003
www.chemmedchem.org
1995
ꢀ 2015 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim

Full Papers
4 to AKR1B10, the displacement
of a water molecule might also
contribute to the enthalpic sig-
nature, as will be discussed in
the following section.
Desolvation costs between hol-
oenzymes may contribute to dif-
ferences in binding affinities
The modeled binding of 3 to
AKME2MU has a similar geome-
try to that of 4, while their inhib-
itory potency is quite different.
Thus we decided to perform
docking and thermodynamic
analyses of the binding events
to determine the underlying rea-
sons for this discrepancy. First,
we wished to assess the effect of
the V301L and K125R mutations,
which were required to obtain
Figure 5. Model of the AKME2MU–NADP+–3 complex. A) Atomic representation of the inhibitor binding site for
the AKME2MU–NADP+–3 complex, obtained by a combination of X-ray crystallography and molecular docking,
with the protein residues, NADP+, and 3 shown as sticks (grey, orange, and blue, respectively); B) Superimposition
of the AKME2MU–NADP+–3 complex model and the X-ray structure of the AKME2MU–NADP+–4 complex, with
the same color code as the legend in panel A for 3 and as Figure 4 for 4. Hydrogen bond distances are shown as
blurred green dashed lines, and the mutated residues are labeled in italics.
only the binding of 4 could be measured, as the affinity of 3
toward the AKR1B10 holoenzyme was too low to be character-
ized by a direct ITC experiment.
the crystal structure of AKR1B10 complexed to 4.[1] The K125R
mutation does not involve major changes regarding the active
site conformation (Figure S3). As for the V301L mutation, the
superimposition of all deposited AKR1B10 wild-type (Val301)
and V301L mutant structures (Leu301, equivalent to Leu300 in
AR) show that, while the side chains of Leu301 mainly point to
the active site, Val301 side chains point to the solvent (Fig-
ure S3). The mutation also changes the orientation of the
neighboring residues Leu302 and Gln303, because the larger
Leu301 is fixed by a hydrophobic interaction with the side
chain of Trp112 in the flipped conformation that does not
occur in the AKR1B10 wild-type complexes with the smaller
Val301, which moves toward the solvent. This particular effect
of residue 301 was previously noticed for the AKR1B10 wild-
type holoenzyme–NADP+–2 model obtained by molecular dy-
namics (MD) simulations.[1] Clearly, inhibitor binding is also able
to induce this contact by means of the opening of the specific-
ity pocket (AKR1B10–18, PDB ID 1ZUA).
Surprisingly, the measurement of 4 with AKR1B10 in the
three different buffers revealed only minor buffer dependence
compared with AR, as only a negligible addition of 0.1 protons
per mole formed complex is detected (Figure S4). As described,
Trp112 changes from conformation B to A during binding of 4
relative to the holoenzyme. This is accompanied by a rupture
of the internal hydrogen bond network, including Gln114 and
Ser304. A more profound study of the electrostatic properties
of the binding pocket of AKR1B10, especially with regard to
the structural changes upon ligand binding, is required to
quantify the pKa values of the local residues and their putative
changes upon ligand binding to better understand this deviat-
ing behavior of AKR1B10 relative to AR. However, this example
underlines the importance of studying buffer dependencies in
ITC.
The thermodynamic profile of AKR1B10 binding of 4, which
was corrected for the protonation effect, showed a strong en-
thalpic dominance (À41.6 kJmolÀ1), while it was accompanied
by an unfavorable entropic term (+9.3 kJmolÀ1). The enthalpic
gain of the newly formed interactions between 4 and the pro-
tein clearly overcompensates for the enthalpic penalty linked
to the rupture of the internal hydrogen bond network. As dis-
cussed above, we estimate the thermodynamic binding profile
of 4 to AR to be much less enthalpy-dominated than for bind-
ing of 4 to AKR1B10. Accordingly, binding of 4 to AKR1B10 re-
sults in a more favorable enthalpic profile.
As an alternative to the more time-consuming MD analyses,
we inspected the available AKR1B10 wild-type structures com-
plexed with inhibitors in the PDB and we found three main
conformations: 1) AKR1B10–18 (PDB ID 1ZUA), with a more ex-
ternally oriented pocket; 2) AKR1B10–caffeic acid phenethyles-
ter (21) (PDB ID 4GQ0), very similar to the AKR1B10 holoen-
zyme pocket (PDB ID 4GQG) and to complexes with flufenamic
acid (20) and 16, (PDB IDs 4I5X and 4JIH, respectively); and
3) AKR1B10-zopolrestat (19) (PDB ID 4JII), with a more buried
specificity pocket. Next, we docked compounds 3 and 4 into
each of the three different holoenzyme conformations as re-
ceptors, but we used FlexX[29] instead of AutoDock 4.2,[26] as
the latter has proven to be influenced by the starting confor-
mation.[30]
In summary, the comparison of binding of 3 and 4 to AR
and the binding of 4 to AR and AKR1B10 indicate that the
opening and accommodation of the specificity pocket lead to
an enthalpic advantage. For the binding of 3 to AR, the inter-
actions formed inside the specificity pocket seem to be re-
sponsible for the strong exothermic signal. For the binding of
In the case of 4, it can be observed that the inhibitor can
only access the protein in the conformation found in 1ZUA
(see docking in mol2 files in the Supporting Information). The
ChemMedChem 2015, 10, 1989 – 2003
www.chemmedchem.org
1996
ꢀ 2015 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim

Full Papers
superimposition of this pose with the crystal structure of the
AKME2MU–NADP+–4 complex shows that both are very similar
(Figure S2B), thus validating the latter structure as a reliable
starting point for structure-based drug design (SBDD). In the
case of 3, only the conformation found in 4GQ0 was unable to
accommodate the inhibitor (see docking in mol2 files in the
Supporting Information). The 4JII conformer displays 3 bound
in a very similar geometry to the cognate AR–NADP+–3 com-
plex (Figure S2C). However, while the AR complex provides the
chlorine atom of 3 with a halogen bond acceptor (Thr113),
AKR1B10 would not, as the larger Gln114 might rearrange its
side chain conformation away from the chlorine to avoid
a mutual clash (see PDB ID 4JII for further clarity). The 1ZUA
conformer binds 3 in a more similar fashion to that of the
AKME2MU–NADP+–3 model, but the inhibitor leans significant-
ly closer to loop C than in the model (like reference ligand 18),
losing the stacking to Phe123. Thus, several conformations
could exist for the AKR1B10 holoenzyme–3 complex. However,
given the flexible nature of the loop C–loop A region in
AKR1B10, in the same fashion as in AR, the energetic cost
would be likely moderate and could be compensated,[31] prob-
ably discarding it as a cause of the different binding affinities
of 3 and 4 toward AKR1B10.
The binding thermodynamics are another aspect worthy of
consideration. Apart from the experimentally determined ener-
getic terms (Table 2), it is important to consider the ligand and
the protein prior to complex formation, the formed protein–
ligand complex, and all changes that occur with water mole-
cules and the various components solvated in the water envi-
ronment. It is known that desolvation costs sometimes can be
responsible for the strongly decreased potency of some li-
gands, even though they seem to fit well into a binding site.[32]
In our case, we applied the HYDE scoring function[33] with the
two AKR1B10–3/4 complexes (obtained using the PDB files of
the model and the X-ray structure, respectively, mutated in
silico with Coot[34] to the wild-type AKR1B10). HYDE considers
the contribution of hydrogen bonds and dehydration energies
in protein–ligand complexes and estimates the free binding
energy.[35] HYDE data indicate that 3 would suffer from a large
desolvation penalty because of the oxygen atom of the nitro
group facing loop C, which could explain its low affinity (see
Table S1 in the Supporting Information): the oxygen atom of
the nitro group would provoke its hydrogen bonding partner
Gln303 side chain amide nitrogen atom to remain poorly des-
olvated (Figure 5A). Using HYDE, the calculated Gibbs free
energy values for binding to AKR1B10 gave Kd values of 2.9 mm
(similar to the experimental Kd value of 2.3 mm, Table 2) and
132 mm for 4 and 3, respectively, a 45-fold difference. This is
roughly the same difference between the experimental IC50
values (48-fold difference); thus, the computed HYDE score
serves as a qualitative value to explain the difference in poten-
cy.
assumed to be entropy-driven, but it has been shown in sever-
al cases that displacement of disordered water molecules from
a hydrophobic pocket can also create an enthalpic signature.[36]
Indeed, the region located at the base of loop A in AKR1B10
presents a loosely packed subpocket that is able to trap
a water molecule, as discovered in the AKR1B10V301L–NADP+
–17 complex (and also observed in other ternary complexes,
PDB IDs: 1ZUA, 4JII, and 4WEV), deviating from AR, which pres-
ents a well-packed hydrophobic subpocket.[13] The analysis of
the AKR1B10 holoenzymes crystallized by us and others (PDB
IDs: 4GQG, 4XZM, and 4XZN) also showed this buried water
molecule (Figure 6). The absence of other water molecules
nearby suggests that this subpocket is insufficiently hydrated;
that is, in spite of available space, no other water molecules
are crystallographically observed, possibly meaning that they
will be highly mobile and disordered. In the AKME2MU–NADP+
–4 complex, the inhibitor is able to displace this water mole-
cule upon binding (Figure 6C), along with the other disordered
water molecules. Their release should not contribute signifi-
cantly to a large entropy gain; instead, new hydrogen bonds
can be formed with other water molecules in the bulk phase
which may add a significant enthalpic benefit, reflected in ITC
experiments (Table 2). In the case of the AKME2MU–NADP+–3
complex model, this water molecule is even closer (less than
2 Š, Figure 6C). Therefore, the low affinity of 3 for AKR1B10
could be due to the incapability of displacing this observed
water molecule, or the new interactions formed (i.e., the hy-
drogen bond between the oxygen of the nitro group with the
carboxamide nitrogen of the Gln303 side chain) may cancel
out the enthalpic gain of displacing this water molecule.
Inhibition of the biological activity of AR and AKR1B10 by 3
and 4 in cellular models
To test the intracellular inhibitory potency of 3 and 4, we used
two different approaches. First, we compared their ability to
prevent sorbitol accumulation within human retinal ARPE-19
cells cultured in a medium supplemented with 50 mm glucose,
reproducing the typical hyperglycemic conditions of diabetes,
in which AR is overexpressed.[15b,37] The IC50 values for sorbitol
accumulation (Table 4) were in accordance with the recombi-
nant protein inhibition assays, as 3 prevented half of sorbitol
accumulation at 0.6 mm, while 4 only succeeded at 16 mm.
Table 4. Inhibitory activity on sorbitol accumulation in human ARPE-19
retinal cells.
Compound
IC50 [mm][a]
1[b]
2
3
4
18[b]
3
0.4
0.6
16
We expected the hydrophobic methoxytetrabromobenzyl
moiety of 4 to entail binding to AKR1B10 with a predominantly
entropic signature. However, to our surprise, the thermody-
namic profile of 4, corrected for the small protonation effect,
was mainly enthalpy-driven (Table 2). The hydrophobic effect is
0.02
[a] Compound concentration required to inhibit sorbitol accumulation by
50%; all values are the mean of at least three experiments. [b] Deter-
mined previously by de la Fuente et al.[15a]
ChemMedChem 2015, 10, 1989 – 2003
www.chemmedchem.org
1997
ꢀ 2015 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim

Full Papers
Figure 6. Impact of a crystallographically observed buried water molecule in
a subpocket of the AKR1B10 holoenzyme active site on the binding of 3 and
4. A) Superimposition of the holoenzyme X-ray structures of wild-type
AKR1B10 (PDB ID 4GQG), methylated wild-type AKR1B10 (PDB ID 4XZM),
and AKME2MU (PDB ID 4XZN), with the cofactor in orange sticks and the
protein residues and the buried water molecule in pink, green, and cyan
sticks and spheres, respectively (mutated residues are labeled in italics);
B) Superimposition of the buried water molecules of the aforementioned
AKR1B10 holoenzymes with the protein represented as a light grey surface;
C) Superimposition of the holoenzyme structures with the buried water mol-
ecules and the AKME2MU–NADP+–3/AKME2MU–NADP+–4 complexes, with
the protein represented as a cartoon, following the same color code for all
structures. The mutated residues are labeled in italics.
Figure 7. Effect of compounds 3 and 4 on cell viability. A) Western blot anal-
ysis of several human cell lines against AKR1B10 and b-actin (loading con-
trol). Lanes 1–7: total cell extracts from NCI-H460 (non-small cell lung carci-
noma, NSCLC), HepG2 (hepatocellular carcinoma), HCT-116 (colorectal carci-
noma), BT-20 (mammary gland carcinoma), A549 (NSCLC), BEAS-2B (non-
tumor bronchial epithelium), and HEK293T (embryonic kidney), respectively.
Purified recombinant AKR1B10 (200 ng) was used as a positive control
(lane 8). B),C) Bar graphs showing the percentage of NCI-H460 (B) and BEAS-
2B (C) cell viability at increasing concentrations (0–200 mm) of inhibitor: 3
(light gray bars), 4 (dark gray bars), 16 (black bars). Bars indicate the mean
Æstandard error of at least four determinations.
Secondly, to assess the degree of inhibition of 3 and 4
against AKR1B10 biological activity, we determined the effect
of each compound on NCI-H460 and BEAS-2B cell viability
(Figure 7). Western blot analysis showed that several tumor cell
lines express AKR1B10, while non-tumor BEAS-2B cells did not
(Figure 7A). NCI-H460 and BEAS-2B cells were chosen for inhib-
itor studies, as both are derived from human lung tissue. In
NCI-H460 cells, a good correlation with the in vitro enzymatic
inhibition tests was observed: cell viability was decreased to
less than 50% with 100 mm of 4, while concentrations higher
than 200 mm of 3 did not have any effect (Figure 7B). To link
the effect of compounds on cell viability to the in vitro
ChemMedChem 2015, 10, 1989 – 2003
www.chemmedchem.org
1998
ꢀ 2015 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim

Full Papers
AKR1B10 inhibition, we tested 3 and 4 in the BEAS-2B cell line,
a non-tumor human lung cell line. Our results show that 3 and
4 did not affect cell viability up to 100 mm (Figure 7C).
potential application for targeting the aforementioned pathol-
ogies.
Experimental Section
Conclusions
Chemistry
A large number of ARIs have been developed for the treat-
ment of diabetic complications, with huge amounts of preclini-
cal and clinical data collected through academic and industrial
programs. In the last ten years, AKR1B10 has been shown to
be overexpressed in a plethora of different cancers and has
emerged as a protein related to tumorigenesis. Additionally,
AR has been established as an important mediator in oxidative
stress and inflammation-related processes, including colon
cancer.[38] Both AR and AKR1B10 siRNA knockdown or drug in-
hibition decreased tumor growth of several cell line xenografts
(from different tissues) transplanted into immune-deficient
mice.[39] Therefore, AR and AKR1B10 inhibitors may represent
a novel class of antitumor agents, and the clinical data assem-
bled in diabetes clinics may ease the transition of these inhibi-
tors to cancer chemotherapy, that is, drug repositioning (re-
viewed in ref. [40]).
General. All chemicals (purchased from Sigma–Aldrich, Acros,
Fluka, and Merck) were used as received unless otherwise noted.
All non-aqueous reactions were performed in flame-dried glass-
ware under an argon atmosphere. Acetone and CCl4 were dried
prior to use at reflux over CaH2. DMF, MeCN, and MeOH were dried
using a Pure Solv Innovative Technology apparatus. Column chro-
matography was performed on Merck silica gel 60 (0.040–
0.063 mm). Thin-layer chromatography (TLC) analysis was per-
formed on aluminum TLC sheets (Merck, silica gel 60, F245) and
glass TLC plates (Merck, RP-18F254s). The compounds were visual-
ized by UV245nm or spraying with phosphomolybdic acid solution,
followed by heating. 1H and 13C NMR spectra were recorded on
a Bruker AV 400 spectrometer (400 MHz and 100 MHz, respectively).
NMR spectra were recorded in CDCl3, CD2Cl2, CD3OD, or (CD3)2SO.
Chemical shifts (d) are given in parts per million (ppm) and were
referenced to residual solvent peaks as an internal standard. Cou-
pling constants (J) are given in Hertz (Hz). The proton spectra are
reported as follows: d (multiplicity, coupling constant J, number of
protons). DEPT135 was used to aid in the assignment of signals in
the 13C NMR spectra. Additional COSY and HSQC spectra were re-
Here, starting from the structure of uracilacetic acid 3, we
developed and synthesized compound 4, inspired by the halo-
genated biaryls and diaryl ethers present in natural products.
The inhibition assays, ITC experiments, X-ray crystallographic
determinations, and in silico structural analysis of holoenzyme–
3/4 complexes enabled determination of AR and AKR1B10 se-
lectivity and potency. The increased volume of the aryl moiety
of 4 shifts the selectivity to AKR1B10, and the smaller aryl
group of 3 is able to open the AR buried specificity pocket,
while the bulky substituents at the ortho and meta positions of
4 prevent binding in this pocket. Conversely, the large aryl
moiety of 4 enables binding into a more external subpocket in
AKR1B10. The modeled binding of 3 to AKR1B10 is similar to
that of 4, but the nitro group of 3 might prevent proper bind-
ing to AKR1B10. We hypothesize that binding of 3 would
imply a large desolvation penalty to the neighboring Gln303
residue because of the repulsive interaction with one of the
nitro oxygen atoms, thus explaining its low affinity.
1
corded in particular cases to enable interpretation of H NMR data.
Electron impact ionization (EI) mass spectra were recorded on
a VG-autospec M instrument. For electrospray ionization (ESI), an
APEX IIIFT-ICR MS (Bruker Daltonics, Billerica, MA), equipped with
a 7T actively shielded magnet was used, and ions were generated
using an Apollo API ESI source, with voltage between 1800 and
2200 V (to optimize ionization efficiency) applied to the needle
and a counter voltage of 450 V applied to the capillary. The ESI
spectra samples were prepared by adding a spray solution of
70:29.9:0.1 (v/v/v) MeOH/H2O/formic acid to a solution of the
sample at a v/v ratio of 1 to 5% to give the best signal-to-noise
ratio. IR spectra were recorded on a Jasco FT/IR-4200 spectropho-
tometer and are reported in cmÀ1
.
1,2,3,4-Tetrabromo-5-methoxy-6-methylbenzene (7): Iodome-
thane (0.88 mL, 14.16 mmol) was added to a solution of 2,3,4,5-tet-
rabromo-6-methylphenol 6 (5.0 g, 11.8 mmol) and K2CO3 (4.89 g,
35.4 mmol) in MeCN (47.2 mL), and the mixture was stirred at
reflux for 4 h. The reaction mixture was cooled to room tempera-
ture and extracted with Et2O (3”100 mL). The combined organic
layers were dried (Na2SO4), filtered, and concentrated. The residue
was purified by column chromatography (silica gel, hexanes) to
afford 4.93 g (95%) of 7 as a white solid: mp: 138–1408C (EtOH);
1H NMR (400 MHz, CDCl3): d=3.78 (s, 3H), 2.47 ppm (s, 3H);
13C NMR (100 MHz, CDCl3): d=155.6 (s), 134.4 (s), 128.1 (s), 126.2
(s), 124.5 (s), 121.3 (s), 60.6 (q), 19.6 ppm (q); HRMS (TOF-EI): calcd
for C8H679Br281Br2O ([M]+), 437.7111; found, 437.7121; IR: n˜ =2935
(w, CÀH), 2845 (w, CÀH), 1439 (m), 1361 (s), 1338 (s), 1151 cmÀ1
(m).
The ITC data showed that the binding of both 3 to AR and 4
to AKR1B10 display large enthalpic signatures. In the first case,
the presence of the nitro group in 3 and its interactions with
the open AR specificity pocket might explain the high binding
enthalpy to AR (as previously observed for 24). In the second
case, compound 4 might additionally displace a disordered
water molecule trapped in the AKR1B10 holoenzyme in the
aforementioned subpocket at the base of loop A, creating the
overall observed large enthalpic signature. Additionally, both
enzymes exhibit different proton uptake upon inhibitor bind-
ing, unveiling the difference in electrostatic properties be-
tween the active sites of the two enzymes.
1,2,3,4-Tetrabromo-5-(bromomethyl)-6-methoxybenzene (8): Bro-
mine (0.471 mL, 9.13 mmol) was slowly added over 5 min to a solu-
tion of compound 7 (2.0 g, 4.57 mmol) in CCl4 (11.14 mL),. The mix-
ture was heated at 908C and irradiated with a tungsten lamp
(250 W) for 4 h. The mixture was washed with an aqueous solution
of Na2S2O3 (3”30 mL). The organic layer was dried (Na2SO4), fil-
tered, and concentrated. The residue was purified by column chro-
Remarkably, the selectivity of 3 for AR and of 4 for AKR1B10
observed in recombinant protein inhibition assays correlated
well with the biological activity tests under cellular conditions.
Thus, the striking differences found here could be exploited to
develop selective and potent AR and AKR1B10 inhibitors, with
ChemMedChem 2015, 10, 1989 – 2003
www.chemmedchem.org
1999
ꢀ 2015 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim

Full Papers
matography (silica gel, hexanes) to afford 2.27 g (96%) of 8 as
pale-yellow solid: mp: 126–1298C (hexane/EtOAc); 1H NMR
(silica gel, from 100:0 to 99:1 CH2Cl2/MeOH) and crystallization
(mp: 117–1188C, hexane/EtOAc). The spectroscopic data matched
those described in the literature.[8b] 1H NMR (400 MHz, CD2Cl2): d=
7.92 (d, J=1.9 Hz, 1H), 7.62 (dd, J=8.3, 2.0 Hz, 1H), 7.49 (d, J=
8.3 Hz, 1H), 7.13 (d, J=7.9 Hz, 1H), 5.79 (d, J=7.9 Hz, 1H), 5.10 (s,
2H), 4.44 (s, 2H), 4.21 (q, J=7.1 Hz, 2H), 1.25 ppm (t, J=7.1 Hz,
3H); 13C NMR (100 MHz, CD2Cl2): d=167.3 (s), 162.4 (s), 151.4 (s),
147.8 (s), 143.0 (d), 137.4 (s), 133.8 (d), 131.7 (d), 125.9 (d), 125.8 (s),
101.9 (d), 62.2 (t), 50.0 (t), 43.0 (t), 13.8 ppm (q).
a
(400 MHz, CDCl3): d=4.76 (s, 2H), 4.01 ppm (s, 3H); 13C NMR
(100 MHz, CDCl3): d=155.0 (s), 133.9 (s), 130.1 (s), 128.0 (s), 125.6
(s), 121.8 (s), 61.7 (t), 29.6 ppm (q); HRMS (TOF-EI): calcd for
C8H579Br381Br2O ([M]+), 515.6172; found, 515.6233; IR: n˜ =2935 (w,
CÀH), 2846 (w, CÀH), 1531 (m), 1359 (s), 1343 (s), 1219 cmÀ1 (m);
UV (MeOH): lmax =226 nm.
Ethyl 2-(2,4-dioxo-3-(2,3,4,5-tetrabromo-6-methoxybenzyl)-3,4-
dihydropyrimidin-1(2H)-yl)acetate (9): A solution of ethyl 2-(2,4-
2-(3-(4-Chloro-3-nitrobenzyl)-2,4-dioxo-3,4-dihydropyrimidin-
1(2H)-yl)acetic acid 3. Following the procedure detailed above,
from ethyl 2-(3-(4-chloro-3-nitrobenzyl)-2,4-dioxo-3,4-dihydropyri-
midin-1(2H)-yl)acetate 12 (0.1 g, 0.27 mmol), 2-(3-(4-chloro-3-nitro-
dioxo-3,4-dihydropyrimidin-1(2H)-yl)acetate
(5)[17]
(0.058 g,
0.29 mmol) was added to a cold (08C) suspension of NaH (0.014 g,
0.36 mmol) in DMF (0.75 mL), After stirring for 1.5 h at 08C, a solu-
tion of 8 (0.15 g, 0.29 mmol) in DMF (1.75 mL) was added, and the
mixture was stirred for 2 h at 08C and then for 1 h at room tem-
perature. The mixture was poured into H2O and extracted with
EtOAc (3”25 mL). The combined organic layers were washed with
brine (3x15 mL), dried (Na2SO4), filtered, and concentrated. The resi-
due was purified by column chromatography (silica gel, CH2Cl2/
MeOH, 98:2) to afford 0.156 g (85%) of 9 as a mixture of rotamers
in a 5:1 ratio as a pale-yellow solid: mp: 148–1508C (hexane/
EtOAc); (major) 1H NMR (400 MHz, CDCl3): d=7.09 (d, J=7.9 Hz,
1H), 5.81 (d, J=7.9 Hz, 1H), 5.42 (s, 2H), 4.45 (s, 2H), 4.25 (q, J=
7.4 Hz, 3H), 3.86 (s, 3H), 1.29 ppm (t, J=7.4 Hz, 3H); (minor)
1H NMR (400 MHz, CDCl3): d=7.43 (d, J=7.2 Hz, 1H), 5.94 (d, J=
7.2 Hz, 1H), 5.64 (s, 2H), 4.63 (s, 2H), 4.29 (q, J=7.2 Hz, 2H), 3.88
(s, 3H), 1.34 ppm (t, J=7.2 Hz, 3H); HRMS (ESI+): calcd for
C16H1579Br4N2O5 ([M+H]+), 630.7709; found, 630.7705; IR: n˜ =2980
(w, CÀH), 1745 (m, C=O), 1709 (m, C=O), 1656 (s, C=O), 1450 (m),
1352 (m), 1200 (m), 999 cmÀ1 (m).
benzyl)-2,4-dioxo-3,4-dihydropyrimidin-1(2H)-yl)acetic
acid
3
(0.073 g, 80%) it was obtained, after purification by crystallization
from EtOH (mp: 183–1848C). The spectroscopic data matched
those reported in the literature:[8b] 1H NMR (400 MHz, (CD3)2CO):
d=7.97 (d, J=1.7 Hz, 1H), 7.66 (d, J=8.0 Hz, 1H), 7.7–7.6 (m, 2H),
5.79 (d, J=7.8 Hz, 1H), 5.14 (s, 2H), 4.62 ppm (s, 2H); 13C NMR
(100 MHz, (CD3)2CO): d=169.4 (s), 163.3 (s), 152.4 (s), 148.8 (s),
145.3 (d), 139.3 (s), 134.3 (d), 132.4 (d), 126.1 (d), 125.3 (s), 101.5
(d), 50.3 (t), 43.4 ppm (t).
Protein purification, crystallography, and biological assays
Site-directed mutagenesis, enzyme expression, and purification.
cDNAs of AR and AKR1A1 were subcloned into the pET15b plas-
mid, while cDNA of AKR1B10 was subcloned into the pET16b plas-
mid (pET30-Xa/LIC was used for the K125R/V301Lmutant). AR, wild-
type and mutant AKR1B10, and AKR1A1 were recombinantly ex-
pressed in the E. coli BL21(DE3) strain (Novagen) and were purified
using the procedures described previously for AR,[41] AKR1B10,[6a]
and AKR1A1.[42] Purity was confirmed by SDS-PAGE, and protein
2-(2,4-Dioxo-3-(2,3,4,5-tetrabromo-6-methoxybenzyl)-3,4-dihy-
dropyrimidin-1(2H)-yl)acetic acid (4): A 10% aqueous solution of
NaOH (0.12 mL) was added to a solution of 9 (0.075 g, 0.12 mmol)
in MeOH (3.54 mL), and the mixture was stirred at 808C for 3 h.
The mixture was cooled to room temperature, acidified with 10%
HCl, and extracted with EtOAc (3”10 mL). The combined organic
layers were dried (Na2SO4), filtered, and concentrated. The residue
was purified by recrystallization from EtOH to afford 0.065 g (90%)
of 4 as a white solid: mp: 227–2308C (EtOH); 1H NMR (400 MHz,
CD3OD): d=7.33 (d, J=7.9 Hz, 1H), 5.74 (d, J=7.8 Hz, 1H), 5.37 (s,
2H), 4.45 (s, 2H), 3.80 ppm (s, 3H); 13C NMR (100 MHz, CDCl3): d=
169.8 (s), 164.0 (s), 156.8 (s), 151.9 (s), 144.3 (d), 132.6 (s), 129.1 (s),
127.7 (s), 125.6 (s), 121.9 (s), 101.7 (d), 61.9 (q), 50.1 (t), 43.9 ppm
(t); HRMS (TOF-ESI+): calcd for C14H1179Br4N2O5 ([M+H]+), 602.7396;
found, 602.7424; IR: n˜ =3500–3000 (br, OH), 2922 (w, CÀH), 1709
(s, C=O), 1660 (s, C=O), 1457 (m), 1361 cmÀ1 (m); UV (MeOH):
lmax =269, 220 nm.
concentration was determined using
(Thermo Scientific).
a NanoDrop ND-1000
Inhibitor screening. The IC50 activity assays were carried out accord-
ing to the quantification of NADPH consumption, which takes
place when the enzyme catalyzes the conversion of d,l-glyceralde-
hyde into glycerol. The assays were performed at 258C in 100 mm
sodium phosphate buffer, pH 7.0, with a protein amount equiva-
lent to the Vmax and 0.2 mm NADPH. The final reaction volume was
500 mL. All compounds assayed were dissolved in dimethyl sulfox-
ide (DMSO), with the corresponding solution added to 2% (v/v)
final DMSO concentration. They were incubated for 5 min at 258C
prior to addition of the substrate. The reaction was initiated by ad-
dition of 1, 60, and 5 mm d,l-glyceraldehyde (for AR, AKR1B10, and
AKR1A1, respectively), and the decrease in optical density at
340 nm was monitored for 3 min at 258C in a UV–vis spectropho-
tometer (UV-1700 PharmaSpec, Shimadzu). The IC50 value was de-
termined to be the compound concentration that inhibits enzy-
matic activity by 50% and was calculated using the Grafit program
(version 5.0; Erithacus Software). Values are given as the mean Æ
standard error of three experiments.
4-(Bromomethyl)-1-chloro-2-nitrobenzene (11): According to the
procedure described above, compound 10 (1.5 g, 8.74 mmol) in
CCl4 (21.3 mL) was treated with Br2 (0.896 mL, 17.48 mmol), heated
at 908C, and irradiated with a tungsten lamp (250 W) for 6 h. Purifi-
cation of the residue by distillation afforded 0.772 g (51%) of start-
ing material and 1.05 g (48%) of a 11 as a yellow oil. Spectroscopic
data matched those described in the literature.[8b]
Isothermal titration calorimetry (ITC). ITC experiments were carried
out with an ITC200 Micro Titration Calorimeter (MicroCal), as de-
scribed previously.[43] The experiments were performed in three dif-
ferent buffer systems (10 mm HEPES, Tricine, and Tris, pH 8.0).
Measurements were performed at 298.15 K with 3% (v/v) final
DMSO concentration. An excess of NADP+ cofactor was added.
The cell was filled with the solution containing the respective pro-
tein, NADP+, and DMSO. The syringe was filled with the respective
inhibitor, NADP+, and DMSO. Final inhibitor concentrations were
Ethyl 2-(3-(4-chloro-3-nitrobenzyl)-2,4-dioxo-3,4-dihydropyrimi-
din-1(2H)-yl)acetate (12): Following the procedure detailed above,
from ethyl 2-(2,4-dioxo-3,4-dihydropyrimidin-1(2H)-yl)acetate 5[17]
(0.336 g, 1.69 mmol) and 4-(bromomethyl)-1-chloro-2-nitrobenzene
11 (0.42 g, 1.66 mmol) in DMF (10 mL),ethyl 2-(3-(4-chloro-3-nitro-
benzyl)-2,4-dioxo-3,4-dihydropyrimidin-1(2H)-yl)acetate 12 (0.459 g,
75%) was obtained, after purification by column chromatography
ChemMedChem 2015, 10, 1989 – 2003
www.chemmedchem.org
2000
ꢀ 2015 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim

Full Papers
between 300 and 1800 mm; final protein concentrations ranged be-
tween 20 and 160 mm. Inhibitor solutions were titrated into a stirred
cell containing the protein solution after a stable baseline was ach-
ieved. The protocol consisted of an initial injection of 0.5 mL to pre-
vent any artefacts arising from the filling of the syringe (not used
in data fitting), followed by injections of 1.5 mL, each spaced 240 s
apart, until complete saturation of the enzyme binding site was
achieved. Raw data were integrated and fitted with NITPIC[44] and
SEDPHAT,[45] plots of the thermograms and fitted curves were pre-
pared using GUSSI.[46] Fitting a single-site-binding isotherm into the
data led to the enthalpy of binding (DH8) and the dissociation con-
stant (Kd) values. Measurements were performed in triplicate,
except for the measurement of 4 with AR, which was done in du-
plicate because of the high amount of protein needed for the titra-
tion. The enthalpies were corrected for protonation effects, as de-
scribed in the literature.[27b,28] The heat of ionization (DH8ion) of the
buffer substances was taken from literature[47] and plotted against
well. After a 16 h incubation at 378C with 5% CO2, the accumulat-
ed sorbitol inside the cells was extracted by lysis with 8% perchlor-
ic acid and then neutralized with KOH. Quantification of sorbitol
was carried out using a colorimetric method (d-sorbitol/xylitol,
Boehringer Mannheim). The IC50 value for each test compound was
determined as the compound concentration that inhibited sorbitol
accumulation by 50%. Values are given as the mean of three ex-
periments.
Cell viability assay. NCI-H460 and BEAS-2B cells were obtained from
the ATCC and were incubated in RPMI 1640 and keratinocyte
serum-free media (Life Technologies), respectively. All experiments
were carried out in 96-well plates. Cellular densities of 2000 and
10000 cellscmÀ2 were reached for NCI-H460 and BEAS-2B, respec-
tively. Seeded cells were incubated in a humidified atmosphere of
5% CO2/95% air at 378C for 24 h, prior to inhibitor treatment.
Then, compounds were added to each well to reach the desired
concentration with a final concentration of 0.2% DMSO (v/v), and
cells were further incubated for 72 h. Cell viability was determined
using the XTT Cell Proliferation Kit II (Roche), following the manu-
facturer’s instructions. After adding XTT, cells were incubated for
90 min before reading the absorbance at 485 and 620 nm in a Wal-
lace 1420 VICTOR3 V multiplate reader (PerkinElmer Life Sciences).
The absorbance at 620 nm was subtracted from the absorbance at
485 nm, and the resulting value was corrected for the control wells
without cells. The percentage of viable cells was calculated by di-
viding this value by the result of wells treated with 0.2% DMSO.
The results are shown as the mean Æstandard deviation of at least
four experimental values.
the measured enthalpy (DH8obs
) in the corresponding buffer
system. A line of best fit of the form DHobs =DHbind +n”DHion was
matched to the data points. The positive slope of the line, as ob-
tained here, indicates how many protons are picked up. The nega-
tive slope indicates how many protons are released during the
binding event. The interception point with the y-axis reflects the
enthalpy corrected for the protonation effect. Mean values and
standard deviations are shown in Table 2, with the measured
values for DH8obs shown in Tables S1C and S1D in the SI. Example
curves for the different measurements are shown in Figure S1 in
the SI.
Western blot analysis. All cell lines were obtained from the Ameri-
can Type Culture Collection (ATCC) following the recommended
protocols. Cells were collected from a confluent culture in a 25 cm2
flask, using the standard method of trypsinization. After neutraliz-
ing the trypsin, the cell pellet was washed two times in phos-
phate-buffered saline (PBS) and was frozen at À208C to facilitate
cell lysis. Then, cells were thawed and resuspended with RIPA
Buffer (Tris-HCl, pH 8.0, 1% IGEPAL CA-630, 0.5% sodium deoxy-
cholate, 0.1% SDS, 10 mm sodium orthovanadate, 1 mm PMSF).
20 mg of cell extract and 200 ng purified recombinant AKR1B10
were loaded onto an SDS-PAGE, and separation was performed at
150 V for 90 min. Then, proteins were transferred for 1 h at 100 V
onto a polyvinylidenedifluoride (PVDF) membrane (Millipore). Once
the transfer was completed, the membrane non-specific interaction
sites were blocked by 5% skim milk solution in 0.1% Tween 20-Tris
buffered saline (TTBS) for 90 min. The membrane was incubated
for 90 min with rabbit primary polyclonal antibody against
AKR1B10 (Bio-Rad) (diluted 1:2000 in 2.5% skim milk solution,
0.1% TTBS). Finally, the membrane was incubated for 90 min with
goat peroxidase-conjugated secondary polyclonal antibody against
rabbit antibody constant fraction (Bio-Rad), (diluted 1:5000 in 2.5%
skim milk solution, 0.1% TTBS). The membrane was stained by
a chemiluminescent method with luminol and hydrogen peroxide.
A digital camera (Bio-Rad) was used to measure band intensities.
b-actin was used for normalization of protein loading.
Crystallization and structure determination
For the AR–NADP+–3 complex, crystals were obtained by co-crys-
tallization under previously published conditions:[41,48] hanging
drop method at 298 K, reservoir with 50 mm ammonium citrate,
pH 5.0, 20% polyethylene glycol (PEG) 6000, with 5 mm of 3. Nei-
ther co-crystallization nor soaking was successful for the AR–
NADP+–4 complex using these conditions. Later, co-crystals were
obtained using the following conditions: 10 mm Tris-HCl, pH 8.0,
20% PEG 6000, with 5 mm of 4. Regarding the AKR1B10 com-
plexes, all crystals were obtained with the AKME2MU system (re-
ductive lysine methylation plus K125R/V301L mutations), except
for the methylated AKR1B10 holoenzyme. Co-crystals of the
AKME2MU–NADP+–4 (30 mm inhibitor) complex and of both
mutant holoenzymes were obtained by the hanging drop vapor
diffusion method at 298 K. The protein solution was mixed with an
equal volume of precipitating solution consisting of 100 mm
sodium cacodylate, pH 9.0, and 30% PEG 6000. The detailed proto-
col is given elsewhere.[1] X-ray data were collected on the home
source for the AKME2MU–NADP+–4 complex, at the Swiss Light
Source on the X06SA beamline for the AR–NADP+–3 complex, and
on X06DA for the methylated AKR1B10 holoenzyme and for AR–
NADP+–4 (for data collection statistics, see Table 3). All data sets
were processed with the program HKL2000.[49] The structures were
solved by molecular replacement with Phaser[50] (PDB model used
in each case is indicated in Table 3), and finalized sets of atomic co-
ordinates were obtained after iterative rounds of model modifica-
tion with the program Coot[34] and refinement with REFMAC5[51]
and PHENIX.[52] Ligand coordinates and restraints were generated
as previously reported[1] or using eLBOW.[53] The coordinate data
for the structures were deposited, with PDB IDs as indicated in
Table 3. Related figures were prepared with PyMOL (v 1.3; Schrç-
dinger).
Intracellular sorbitol accumulation in ARPE-19human retinal cells. In-
tracellular concentrations of sorbitol in ARPE-19 human retinal cells
obtained from the ATCC were measured following the recom-
mended protocol.[15b] In brief, 1”107 cells were cultured in 2.5 mL
of minimum essential medium (JRH Biosciences) supplemented
with 0.5% fetal calf serum. Glucose (50 mm) was added to the cul-
ture medium to reproduce the intracellular hyperglycemic condi-
tions of diabetes mellitus. The compounds to be assayed were dis-
solved in DMSO, and the corresponding solution was added to the
ChemMedChem 2015, 10, 1989 – 2003
www.chemmedchem.org
2001
ꢀ 2015 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim

Full Papers
Oates, Curr. Opin. Invest. Drugs 2010, 11, 402–417; c) W. H. Tang, K. A.
Martin, J. Hwa, Front. Pharmacol. 2012, 3, 87.
[3] a) K. V. Ramana, S. K. Srivastava, Int. J. Biochem. Cell Biol. 2010, 42, 17–
20; b) R. Tammali, A. B. Reddy, A. Saxena, P. G. Rychahou, B. M. Evers, S.
Qiu, S. Awasthi, K. V. Ramana, S. K. Srivastava, Carcinogenesis 2011, 32,
1259–1267.
Docking simulations
Molecular docking to obtain the AKME2MU–NADP+–3 model was
performed with the program AutoDock 4.2[26] on a Linux worksta-
tion. Coordinates of 3 were taken from the AR cognate structure
(PDB ID 4XZH). For the protein molecule, crystallographic ligands
and water molecules were removed from the PDB file correspond-
ing to the co-crystal of AKME2MU holoenzyme–JF0048–NADP+–3,
in which the inhibitor could not be observed. Polar hydrogens
were added by using the hydrogen module in AutoDock Tools
(ADT), then Gasteiger united atom partial charges were assigned,
except for NADP+, where the nicotinamide charge was kept posi-
tive.[54] The remaining settings were the same as previously report-
ed.[55] All structures generated for the same compound were sub-
jected to cluster analysis with a tolerance of 2.0 Š for an all-atom
RMSD from a lower energy structure. Related figures were pre-
pared with PyMOL (v 1.3; Schrçdinger). Further docking studies
were performed with the docking program FlexX,[29] included in
the LeadIT suite (BioSolveIT). Docking poses were selected follow-
ing the same criteria as explained for AutoDock. The crystal struc-
tures with PDB IDs 1ZUA, 4JII, and 4GQ0 were used to obtain the
corresponding protein conformers. The ligand and all crystallo-
graphic water molecules were removed. The compounds were sub-
sequently docked.
[4] a) S. Fukumoto, N. Yamauchi, H. Moriguchi, Y. Hippo, A. Watanabe, J.
Shibahara, H. Taniguchi, S. Ishikawa, H. Ito, S. Yamamoto, H. Iwanari, M.
Hironaka, Y. Ishikawa, T. Niki, Y. Sohara, T. Kodama, M. Nishimura, M. Fu-
kayama, H. Dosaka-Akita, H. Aburatani, Clin. Cancer Res. 2005, 11, 1776–
1785; b) F. X. Ruiz, O. Gallego, A. Ard›vol, A. Moro, M. Domínguez, S. l-
varez, R. lvarez, A. R. de Lera, C. Rovira, I. Fita, X. ParØs, J. FarrØs,
Chem.-Biol. Interact. 2009, 178, 171–177; c) C. Wang, R. Yan, D. Luo, K.
Watabe, D. F. Liao, D. Cao, J. Biol. Chem. 2009, 284, 26742–26748; d) E.
Zeindl-Eberhart, S. Haraida, S. Liebmann, P. R. Jungblut, S. Lamer, D.
Mayer, G. Jager, S. Chung, H. M. Rabes, Hepatology 2004, 39, 540–549.
[5] a) O. A. Barski, S. M. Tipparaju, A. Bhatnagar, Drug Metab. Rev. 2008, 40,
553–624; b) O. Gallego, O. V. Belyaeva, S. PortØ, F. X. Ruiz, A. V. Stetsen-
ko, E. V. Shabrova, N. V. Kostereva, J. FarrØs, X. ParØs, N. Y. Kedishvili, Bio-
chem. J. 2006, 399, 101–109.
[6] a) O. Gallego, F. X. Ruiz, A. Ard›vol, M. Domínguez, R. lvarez, A. R.
de Lera, C. Rovira, J. FarrØs, I. Fita, X. ParØs, Proc. Natl. Acad. Sci. USA
2007, 104, 20764–20769; b) T. M. Penning, Clin. Cancer Res. 2005, 11,
1687–1690; c) A. M. Quinn, R. G. Harvey, T. M. Penning, Chem. Res. Toxi-
col. 2008, 21, 2207–2215; d) F. X. Ruiz, S. PortØ, X. ParØs, J. FarrØs, Front.
Pharmacol. 2012, 3, 58.
[7] J. Ma, R. Yan, X. Zu, J. M. Cheng, K. Rao, D. F. Liao, D. Cao, J. Biol. Chem.
2008, 283, 3418–3423.
[8] a) P. J. Oates, Curr. Drug Targets 2008, 9, 14–36; b) M. V. Reddy, P. Venka-
tapuram, M. R. Mallireddigari, V. R. Pallela, S. C. Cosenza, K. A. Robell, B.
Akula, B. S. Hoffman, E. P. Reddy, J. Med. Chem. 2011, 54, 6254–6276.
[9] O. El-Kabbani, A. Podjarny, Cell. Mol. Life Sci. 2007, 64, 1970–1978.
[10] H. Steuber, A. Heine, A. Podjarny, G. Klebe, J Mol Biol 2008, 379, 991–
1016.
Acknowledgements
The authors thank the Institut de GØnØtique et de Biologie Mo-
lØculaire et Cellulaire (IGBMC) Structural Genomics Platform staff
(in particular, Dr. Catherine Birck, Dr. Alastair McEwen, and Pierre
Poussin-Courmontagne). The crystallographic experiments were
performed on the X06SA and X06DA beamlines at the Swiss Light
Source synchrotron, Paul Scherrer Institut, Villigen, Switzerland.
The authors thank Drs. Vincent Olieric and Takashi Tomizaki for
their help on the beamlines and Dr. Marc Ruff for the X-ray data
collection of the AR–NADP+–3 complex. This work was funded
by the National Center for Scientific Research (CNRS), the Institut
National de la SantØ et de la Recherche MØdicale (INSERM), the
UniversitØ de Strasbourg, the RØgion Alsace, the Hôpital Civil de
Strasbourg, Instruct (part of the European Strategy Forum of Re-
search Infrastructures; ESFRI), the French Infrastructure for Inte-
grated Structural Biology (FRISBI) (ANR-10-INSB-05-01), the Span-
ish Ministerio de Economía y Competitividad (BFU2011-24176),
Generalitat de Catalunya (2009 SGR 795), the Fondation pour la
Recherche MØdicale (Code FRM: SPF20121226275) and the Bayer
HealthCare ‘Grants4Targets’ program (entitled ‘AKR1B10 as a mo-
lecular target for cancer control’, 2009–2010). This work was sup-
ported by European Research Council (ERC) grant 268145–Drug-
ProfilBind, kindly provided by the EU to G.K.
[11] a) B. Diez-Dacal, J. Gayarre, S. Gharbi, J. F. Timms, C. Coderch, F. Gago, D.
PØrez-Sala, Cancer Res. 2011, 71, 4161–4171; b) S. Endo, T. Matsunaga,
K. Kuwata, H. T. Zhao, O. El-Kabbani, Y. Kitade, A. Hara, Bioorg. Med.
Chem. 2010, 18, 2485–2490; c) S. Endo, T. Matsunaga, H. Mamiya, C.
Ohta, M. Soda, Y. Kitade, K. Tajima, H. T. Zhao, O. El-Kabbani, A. Hara,
Arch. Biochem. Biophys. 2009, 487, 1–9; d) S. Endo, T. Matsunaga, M.
Soda, K. Tajima, H. T. Zhao, O. El-Kabbani, A. Hara, Biol. Pharm. Bull.
2010, 33, 886–890; e) T. Matsunaga, S. Endo, M. Soda, H. T. Zhao, O. El-
Kabbani, K. Tajima, A. Hara, Biochem. Biophys. Res. Commun. 2009, 389,
128–132; f) M. Soda, D. Hu, S. Endo, M. Takemura, J. Li, R. Wada, S.
Ifuku, H. T. Zhao, O. El-Kabbani, S. Ohta, K. Yamamura, N. Toyooka, A.
Hara, T. Matsunaga, Eur. J. Med. Chem. 2012, 48, 321–329; g) M. Take-
mura, S. Endo, T. Matsunaga, M. Soda, H. T. Zhao, O. El-Kabbani, K.
Tajima, M. Iinuma, A. Hara, J. Nat. Prod. 2011, 74, 1201–1206.
[12] a) L. Zhang, H. Zhang, Y. Zhao, Z. Li, S. Chen, J. Zhai, Y. Chen, W. Xie, Z.
Wang, Q. Li, X. Zheng, X. Hu, FEBS Lett. 2013, 587, 3681–3686; b) L.
Zhang, H. Zhang, X. Zheng, Y. Zhao, S. Chen, Y. Chen, R. Zhang, Q. Li, X.
Hu, ChemMedChem 2014, 9, 706–709.
[13] F. X. Ruiz, A. Cousido-Siah, A. Mitschler, J. FarrØs, X. ParØs, A. Podjarny,
Chem.-Biol. Interact. 2013, 202, 178–185.
[14] A. Cousido-Siah, F. X. Ruiz, I. Crespo, S. PortØ, A. Mitschler, X. ParØs, A.
Podjarny, J. FarrØs, Chem.-Biol. Interact. 2015, 234, 290–296.
[15] a) J. A. de La Fuente, S. Manzanaro, Nat. Prod. Rep. 2003, 20, 243–251;
b) J. A. de La Fuente, S. Manzanaro, M. J. Martín, T. G. de Quesada, I. Rey-
mundo, S. M. Luengo, F. Gago, J. Med. Chem. 2003, 46, 5208–5221; c) S.
Manzanaro, J. Salva, J. A. de La Fuente, J. Nat. Prod. 2006, 69, 1485–
1487.
Keywords: aldose reductase · AKR1B10 · drug design · steric
hindrance · buried water molecule
[16] T. Kotani, Y. Nagaki, A. Ishii, Y. Konishi, H. Yago, S. Suehiro, N. Okukado,
K. Okamoto, J. Med. Chem. 1997, 40, 684–694.
[17] O. M. Ali, H. H. Amer, A. A. H. Abdel-Rahman, Synthesis 2007, 2823–
2828.
[1] A. Cousido-Siah, F. X. Ruiz, A. Mitschler, S. PortØ, . R. de Lera, M. J.
Martín, S. Manzanaro, J. A. de La Fuente, F. Terwesten, M. Betz, G. Klebe,
J. FarrØs, X. ParØs, A. Podjarny, Acta Crystallogr. Sect. D 2014, 70, 889–
903.
[2] a) Z. Changjin in Diabetes Mellitus—Insights and Perspectives (Ed.: O. O.
Oguntibeju), InTech—Open Access Company, 2013, pp. 17–46; b) P. J.
[18] K. M. Bohren, C. E. Grimshaw, Biochemistry 2000, 39, 9967–9974.
[19] a) M. P. Blakeley, F. Ruiz, R. Cachau, I. Hazemann, F. Meilleur, A. Mitschler,
S. Ginell, P. Afonine, O. N. Ventura, A. Cousido-Siah, M. Haertlein, A. Joa-
chimiak, D. Myles, A. Podjarny, Proc. Natl. Acad. Sci. USA 2008, 105,
1844–1848; b) E. I. Howard, R. Sanishvili, R. E. Cachau, A. Mitschler, B.
Chevrier, P. Barth, V. Lamour, M. Van Zandt, E. Sibley, C. Bon, D. Moras,
ChemMedChem 2015, 10, 1989 – 2003
www.chemmedchem.org
2002
ꢀ 2015 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim

Full Papers
T. R. Schneider, A. Joachimiak, A. Podjarny, Proteins Struct. Funct. Bioinf.
2004, 55, 792–804.
E862–E871; b) R. Satow, M. Shitashige, Y. Kanai, F. Takeshita, H. Ojima, T.
Jigami, K. Honda, T. Kosuge, T. Ochiya, S. Hirohashi, T. Yamada, Clin
Cancer Res. 2010, 16, 2518–2528; c) R. Tammali, K. V. Ramana, S. S. Sin-
ghal, S. Awasthi, S. K. Srivastava, Cancer Res. 2006, 66, 9705–9713; d) W.
Zhang, H. Li, Y. Yang, J. Liao, G.-Y. Yang, Cancer Lett. 2014, 355, 273–
280.
[20] H. Steuber, A. Heine, G. Klebe, J. Mol. Biol. 2007, 368, 618–638.
[21] C. Bissantz, B. Kuhn, M. Stahl, J. Med. Chem. 2010, 53, 5061–5084.
[22] F. Ruiz, I. Hazemann, A. Mitschler, A. Joachimiak, T. Schneider, M. Kar-
plus, A. Podjarny, Acta Crystallogr. Sect. D 2004, 60, 1347–1354.
[23] D. W. Borhani, T. M. Harter, J. M. Petrash, J. Biol. Chem. 1992, 267,
24841–24847.
[40] J.-U. Peters, J. Med. Chem. 2013, 56, 8955–8971.
[41] V. Lamour, P. Barth, H. Rogniaux, A. Poterszman, E. Howard, A. Mitschler,
A. Van Dorsselaer, A. Podjarny, D. Moras, Acta Crystallogr. Sect. D 1999,
55, 721–723.
[24] L. Zhang, H. Zhang, X. Zheng, Y. Zhao, S. Chen, Y. Chen, R. Zhang, Q. Li,
X. Hu, ChemMedChem 2014, 9, 706–709.
[25] Z. S. Derewenda, L. Lee, U. Derewenda, J. Mol. Biol. 1995, 252, 248–262.
[26] G. M. Morris, R. Huey, W. Lindstrom, M. F. Sanner, R. K. Belew, D. S. Good-
sell, A. J. Olson, J. Comput. Chem. 2009, 30, 2785–2791.
[27] a) M. Stefek, M. Soltesova Prnova, M. Majekova, C. Rechlin, A. Heine, G.
Klebe, J. Med. Chem. 2015, 58, 2649–2657; b) H. Steuber, P. Czodrowski,
C. A. Sotriffer, G. Klebe, J. Mol. Biol. 2007, 373, 1305–1320.
[28] F. Dullweber, M. T. Stubbs, D. Musil, J. Stürzebecher, G. Klebe, J. Mol.
Biol. 2001, 313, 593–614.
[42] O. A. Barski, K. H. Gabbay, C. E. Grimshaw, K. M. Bohren, Biochemistry
1995, 34, 11264–11275.
[43] a) C. Koch, A. Heine, G. Klebe, J. Mol. Biol. 2011, 406, 700–712; b) T. Pet-
rova, H. Steuber, I. Hazemann, A. Cousido-Siah, A. Mitschler, R. Chung,
M. Oka, G. Klebe, O. El-Kabbani, A. Joachimiak, A. Podjarny, J. Med.
Chem. 2005, 48, 5659–5665.
[44] S. Keller, C. Vargas, H. Zhao, G. Piszczek, C. A. Brautigam, P. Schuck, Anal.
Chem. 2012, 84, 5066–5073.
[29] M. Rarey, B. Kramer, T. Lengauer, G. Klebe, J. Mol. Biol. 1996, 261, 470–
489.
[45] H. Zhao, G. Piszczek, P. Schuck, Methods 2015, 76, 137–148.
[46] C. A. Brautigam, The University of Texas Southwestern Medical Center,
Dallas, TX, 2012.
´
[30] D. Plewczynski, M. Łazniewski, R. Augustyniak, K. Ginalski, J. Comput.
Chem. 2011, 32, 742–755.
[31] C. A. Sotriffer, O. Kramer, G. Klebe, Proteins Struct. Funct. Bioinf. 2004, 56,
52–66.
[32] a) G. Klebe, Nat. Rev. Drug Discovery 2015, 14, 95–110; b) R. Talhout, A.
Villa, A. E. Mark, J. B. F. N. Engberts, J. Am. Chem. Soc. 2003, 125, 10570–
10579.
[33] N. Schneider, G. Lange, S. Hindle, R. Klein, M. Rarey, J. Comput.-Aided
Mol. Des. 2013, 27, 15–29.
[34] P. Emsley, B. Lohkamp, W. G. Scott, K. Cowtan, Acta Crystallogr. Sect. D
2010, 66, 486–501.
[35] a) M. T. Barho, S. Oppermann, F. C. Schrader, I. Degenhardt, K. Elsässer,
C. Wegscheid-Gerlach, C. Culmsee, M. Schlitzer, ChemMedChem 2014, 9,
2260–2273; b) N. T. Tzvetkov, S. Hinz, P. Küppers, M. Gastreich, C. E.
Müller, J. Med. Chem. 2014, 57, 6679–6703.
[36] a) E. Barratt, R. J. Bingham, D. J. Warner, C. A. Laughton, S. E. V. Phillips,
S. W. Homans, J. Am. Chem. Soc. 2005, 127, 11827–11834; b) A. Biela,
N. N. Nasief, M. Betz, A. Heine, D. Hangauer, G. Klebe, Angew. Chem. Int.
Ed. 2013, 52, 1822–1828; Angew. Chem. 2013, 125, 1868–1876; c) L.
Englert, A. Biela, M. Zayed, A. Heine, D. Hangauer, G. Klebe, Biochim.
Biophys. Acta Gen. Subj. 2010, 1800, 1192–1202; d) S. W. Homans, Drug
Discovery Today 2007, 12, 534–539; e) T. Young, R. Abel, B. Kim, B. J.
Berne, R. A. Friesner, Proc. Natl. Acad. Sci. USA 2007, 104, 808–813.
[37] K.-C. Chang, J. Ponder, D. V. LaBarbera, J. M. Petrash, Invest. Ophthalmol.
Visual Sci. 2014, 55, 2853–2861.
[47] H. Fukada, K. Takahashi, Proteins Struct. Funct. Bioinf. 1998, 33, 159–166.
[48] A. Cousido-Siah, T. Petrova, I. Hazemann, A. Mitschler, F. X. Ruiz, E.
Howard, S. Ginell, C. Atmanene, A. Van Dorsselaer, S. Sanglier-Cianferani,
A. Joachimiak, A. Podjarny, Proteins Struct. Funct. Bioinf. 2012, 80, 2552–
2561.
[49] Z. Otwinowski, W. Minor, Macromol. Crystallogr. Part A 1997, 276, 307–
326.
[50] A. J. McCoy, R. W. Grosse-Kunstleve, P. D. Adams, M. D. Winn, L. C. Storo-
ni, R. J. Read, J. Appl. Crystallogr. 2007, 40, 658–674.
[51] G. N. Murshudov, P. Skubak, A. A. Lebedev, N. S. Pannu, R. A. Steiner,
R. A. Nicholls, M. D. Winn, F. Long, A. A. Vagin, Acta Crystallogr. Sect. D
2011, 67, 355–367.
[52] P. D. Adams, P. V. Afonine, G. Bunkoczi, V. B. Chen, I. W. Davis, N. Echols,
J. J. Headd, L. W. Hung, G. J. Kapral, R. W. Grosse-Kunstleve, A. J. McCoy,
N. W. Moriarty, R. Oeffner, R. J. Read, D. C. Richardson, J. S. Richardson,
T. C. Terwilliger, P. H. Zwart, Acta Crystallogr. Sect. D 2010, 66, 213–221.
[53] N. W. Moriarty, R. W. Grosse-Kunstleve, P. D. Adams, Acta Crystallogr.
Sect. D 2009, 65, 1074–1080.
[54] J. De Ruyck, M. FamerØe, J. Wouters, E. A. Perp›te, J. Preat, D. Jacque-
min, Chem. Phys. Lett. 2007, 450, 119–122.
[55] S. PortØ, F. X. Ruiz, J. GimØnez, I. Molist, S. Alvarez, M. Domínguez, R. Al-
varez, A. R. de Lera, X. ParØs, J. FarrØs, Chem.-Biol. Interact. 2013, 202,
186–194.
[38] R. Maccari, R. Ottanà, J. Med. Chem. 2015, 58, 2047–2067.
[39] a) J. Ma, D. X. Luo, C. Huang, Y. Shen, Y. Bu, S. Markwell, J. Gao, J. Liu, X.
Zu, Z. Cao, Z. Gao, F. Lu, D. F. Liao, D. Cao, Int. J. Cancer 2012, 131,
Received: September 1, 2015
Published online on November 9, 2015
ChemMedChem 2015, 10, 1989 – 2003
www.chemmedchem.org
2003
ꢀ 2015 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim