Structural basis for the high all-trans-retinaldehyde reductase activity of the tumor marker AKR1B10

Oriol Gallego*†, F. Xavier Ruiz*, Albert Arde` vol‡, Marta Domı´nguez§, Rosana Alvarez§, Angel R. de Lera§, Carme Rovira‡¶, Jaume Farre´ s*, Ignacio Fitaʈ, and Xavier Pare´ s*,**

*Department of Biochemistry and Molecular Biology, Universitat Auto`noma de Barcelona, E-08193 Bellaterra, Barcelona, Spain; ʈInstitut de Biologia Molecular de Barcelona, Consejo Superior de Investigaciones Cientı´ficas,Institut de Recerca Biome`dica, and ‡Centre de Recerca en Quı´micaTeo`rica, Parc Cientı´ficde Barcelona, Josep Samitier 1–5, E-08028 Barcelona, Spain; ¶Institucio´Catalana de Recerca i Estudis Avanc¸ats, Passeig Lluı´s Companys 23, E-08010 Barcelona, Spain; and §Departamento de Quı´micaOrga´nica, Universidade de Vigo, E-36200 Vigo, Spain

Edited by Wayne A. Hendrickson, Columbia University, New York, NY, and approved November 6, 2007 (received for review June 16, 2007) AKR1B10 is a human aldo-keto reductase (AKR) found to be elevated medium-chain dehydrogenases/reductases and retinol dehydroge- in several cancer types and in precancerous lesions. In vitro, AKR1B10 nases from the short-chain dehydrogenases/reductases (SDR) (8). exhibits a much higher retinaldehyde reductase activity than any Thus, highly distinct structures have converged to the same other human AKR, including AKR1B1 (). We here function. The multiplicity of members of the three super- demonstrate that AKR1B10 also acts as a retinaldehyde reductase in families that can contribute to retinol–retinaldehyde interconver- vivo. This activity may be relevant in controlling the first step of sion stresses the importance of this critical step in retinoic acid retinoic acid synthesis. Up-regulation of AKR1B10, resulting in retinoic synthesis (7). An alteration of this step, e.g., by changing enzymatic acid depletion, may lead to cellular proliferation. Both in vitro and in levels, may result in disturbances in control and cell prolifer- vivo activities of AKR1B10 were inhibited by tolrestat, an AKR1B1 ation. Thus, the up-regulation of retinaldehyde reductase activity of inhibitor developed for diabetes treatment. The crystal structure of AKR1B10 found in cancer could be linked to the depletion of the ternary complex AKR1B10–NADP؉–tolrestat was determined at retinoic acid levels and subsequent loss of cell differentiation and 1.25-Å resolution. Molecular dynamics models of AKR1B10 and cancer development (9). AKR1B1 with retinaldehyde isomers and site-directed mutagenesis In addition to AKR1B10, other human AKRs—AKR1B1 (al- show that subtle differences at the entrance of the retinoid-binding dose reductase) (3, 7), AKR1C3, and AKR1C4 (F.X.R. and X.P., site, especially at position 125, are determinant for the all-trans- unpublished data)—are also active with retinaldehyde, but with retinaldehyde specificity of AKR1B10. Substitutions in the retinalde- very low kcat values. AKR1B10 is an exception; it exhibits low Km hyde cyclohexene ring also influence the specificity. These structural and high kcat values, resulting in a catalytic efficiency comparable to features should facilitate the design of specific inhibitors, with that of the best retinaldehyde reductases from the SDR superfamily potential use in cancer and diabetes treatments. (7). In the present work we have further studied the functional properties of AKR1B10, in vivo and in vitro, and have obtained its aldo-keto reductases ͉ aldose reductase ͉ protein structure ͉ crystal structure to discover the specific features that provide its retinoic acid ͉ tolrestat high activity with retinoids. We have compared the functional and structural properties of AKR1B10 with those of AKR1B1 as a KR1B10 (human small intestine aldose reductase or aldose model of an AKR with low retinaldehyde reductase activity. Thus, Areductase-like) is a recently identified NADPϩ-dependent we have demonstrated the inhibition of AKR1B10 with tolrestat, a aldo-keto reductase (AKR) (1, 2). Similar to other members of the powerful inhibitor of AKR1B1, designed to treat secondary com- enzymatic family, AKR1B10 is a monomer of Mr 36,000 and can plications of diabetes (10). The crystal structures show that reduce a variety of aldehydes and ketones, such as glyceraldehyde, AKR1B10 and AKR1B1 similarly bind tolrestat, but differences are methylglyoxal, diacetyl, and aromatic aldehydes (1, 3). However, its observed in the corresponding models of complexes with retinal- physiological function remains still unclear. Two recent findings dehyde. This could be related to the distinct kinetics of the two have made the study of this enzyme especially relevant: its detection with retinoids, which is supported by results from site- in several human cancer types and its high activity with directed mutagenesis. retinaldehyde. AKR1B10 was reported to be increased in several studies on Results human hepatocellular carcinoma (1, 4). Recently it was proved that AKR1B10 Acts as a Retinaldehyde Reductase in a Cellular Environment. the expression of the enzyme was induced in patients with non- After the recent demonstration of a relevant retinaldehyde reduc- small cell lung carcinoma (NSCLC), which is mainly linked to tase activity in vitro (7), we here investigated the involvement of tobacco consumption (5). Accordingly, AKR1B10 has been pro- AKR1B10 in retinoid metabolism in vivo. COS-1 cells were trans- posed as a new diagnostic marker for smoking-related NSCLC. Interestingly, expression of AKR1B10 was found to be induced already in squamous metaplasia, a precancerous lesion in squamous Author contributions: O.G., C.R., J.F., I.F., and X.P. designed research; O.G., F.X.R., A.A., and I.F. performed research; M.D., R.A., A.R.d.L., and C.R. contributed new reagents/analytic cell carcinoma, suggesting that the enzyme could be involved in the tools; O.G., F.X.R., A.A., C.R., J.F., I.F., and X.P. analyzed data; and O.G., A.R.d.L., C.R., J.F., onset of carcinogenesis and stressing its potential use as a thera- I.F., and X.P. wrote the paper. peutic target (5). Consistent with this report is the several-fold The authors declare no conflict of interest. induction of AKR1B10 in oral cancer cells by cigarette smoke This article is a PNAS Direct Submission. condensate (6). Data deposition: The atomic coordinates have been deposited in the , We have demonstrated that AKR1B10 and other AKRs catalyze www.pdb.org (PDB ID code 1ZUA). retinol–retinaldehyde conversion (3, 7), the essential first step in the †Present address: European Molecular Biology Laboratory, Meyerhofstrasse 1, D-69117 retinoic acid synthesis pathway. Retinoic acid participates in a large Heidelberg, Germany. number of biological processes, ranging from fetal development to **To whom correspondence should be addressed. E-mail: [email protected]. cell proliferation and differentiation, controlling the expression of This article contains supporting information online at www.pnas.org/cgi/content/full/ multiple (8). Two other enzyme types are well known to be 0705659105/DC1. involved in this initial redox step, alcohol dehydrogenases from the © 2007 by The National Academy of Sciences of the USA

20764–20769 ͉ PNAS ͉ December 26, 2007 ͉ vol. 104 ͉ no. 52 www.pnas.org͞cgi͞doi͞10.1073͞pnas.0705659105 Downloaded by guest on September 28, 2021 in the absence of detergent. With the new methodology, we determined an IC50 of Ϸ10 nM (Fig. 1B), a value similar to that published for the tolrestat inhibition of the glyceraldehyde reduc- tase activity of AKR1B1 (12). Tolrestat also affected the retinaldehyde reductase activity of COS-1 cells expressing AKR1B10 (Fig. 1C). When cells were incubated with 1 ␮M tolrestat, the activity contributed by the enzyme was reduced by Ϸ50%, whereas it was completely sup- pressed by 10 ␮M tolrestat. Thus, tolrestat was revealed as a potent inhibitor of AKR1B10 both in vitro and in vivo.

.Crystal Structure of AKR1B10 Complexed with NADP؉ and Tolrestat The crystal structure of AKR1B10, complexed with the NADPϩ and the inhibitor tolrestat, was solved and refined at 1.25-Å resolution (Fig. 2 A and B and SI Table 3). The structure of AKR1B10 showed the (␣/␤)8 barrel topology, characteristic of the AKR superfamily. The quality and the resolution of the final electron density map allowed the accurate determination of the conformation of all residues from Met-1 to Tyr-316 (henceforth residue numbering will correspond to that of AKR1B10 with the Swiss-Prot entry O60218) and all non-hydrogen atoms of the cofactor and the inhibitor (SI Fig. 6). The highest mobility, as derived from the thermal B factors, corresponds to a region from Lys-125 to Ala-131 included in the loop A, which contributes to the upper lid of the active-site pocket (Fig. 2B). The NADPϩ cofactor binds at the carboxyl edge of the ␤-strands of the barrel in an extended conformation that is perpendicular to the axis defined by the strands and with the adenine and nicotinamide moieties located at the periphery and at the center of the barrel, respectively. In turn, tolrestat binds mostly parallel to that axis with its carboxyl group at distances of 2.63 and 3.27 Å from the essential Tyr-49 hydroxyl group and from the cofactor C4 atom, Fig. 1. Retinaldehyde reductase activity of AKR1B10 and effect of tolrestat in vitro and in vivo.(A) Retinoid metabolism in COS-1 cells transiently express- respectively. This geometry is expected to be very close to that ing AKR1B10. Cellular retinoid content was measured by HPLC after incubat- of a substrate bound in a catalytically productive manner. ing cells for 30 min with 10 ␮M retinaldehyde or 10 ␮M retinol. (B) Determi- The analysis of interactions between AKR1B10 and the nation of tolrestat IC50 for retinaldehyde reductase activity of AKR1B10 using inhibitor shows that residues Tyr-49, His-111, and Trp-112 0.5 ␮M retinaldehyde as a substrate. (C) Tolrestat inhibition of cellular establish hydrogen bonds with tolrestat atoms O2 and O3 of the AKR1B10 activity. COS-1 cells transfected with pCMV-HA-AKR1B10 were in- carboxyl group and, together with the cofactor-positive charge, cubated with 10 ␮M all-trans-retinaldehyde and different concentrations of define an anion-binding pocket (Fig. 2C). Trp-21, Val-48, Trp- tolrestat. Data are expressed as the percentage of conversion of the retinoid 80, Trp-112, Phe-116, Phe-123, Trp-220, Cys-299, Val-301, Gln- taken up by cells (reduced retinaldehyde or oxidized retinol). Conversion for 303, and the nicotinamide moiety of the cofactor define a COS-1 cells transfected with empty vector (pCMV-HA) is shown as a control. Results are expressed as the mean Ϯ SEM of at least three determinations. strongly hydrophobic pocket, the so-called specificity pocket in AKR1B1, where tolrestat is found. For comparison, a similar analysis was performed for AKR1B1 (13) [Protein Data Bank fected with a suitable plasmid containing AKR1B10 cDNA [sup- (PDB) entry 2FZD] in Fig. 2D. For residues interacting with porting information (SI) Fig. 5]. all-trans-Retinaldehyde and all- tolrestat, only positions 301 and 303 differ between AKR1B10 trans-retinol were measured in COS-1 cells transfected with empty and AKR1B1 (Fig. 2D), which might be consistent with the vector or with AKR1B10 vector, after incubation with all-trans- inhibitor having similar effects on AKR1B10 and AKR1B1. retinaldehyde (Fig. 1A). Production of retinyl esters or retinoic acid was undetectable under the conditions used. Control cells had a Computer Modeling of Retinaldehyde Binding. Models of the ternary ϩ all trans marked ability to reduce retinaldehyde to retinol because they complexes of AKR1B10 and AKR1B1 with NADP and - - BIOCHEMISTRY converted Ϸ25% of intracellular retinaldehyde. However, cells retinaldehyde were built based on the corresponding crystal struc- tures with tolrestat (Fig. 3). In the two enzymes, the retinaldehyde transiently expressing AKR1B10 showed a 2-fold-higher rate of molecule replaced the inhibitor, fitting nicely into the binding reduction, reaching Ϸ55% of conversion. In contrast, incubation pocket. The resulting complexes were submitted to molecular with retinol did not produce a significant increase in intracellular dynamics simulation (SI Text). The average structures, during a retinaldehyde content. Thus, although AKR1B10 can use both time window of 250 ps after equilibration of the complex, were retinol and retinaldehyde in vitro, it appears to function only as a taken for analysis. In both cases, rotation of the cyclohexene ring to retinaldehyde reductase in vivo. accommodate in the binding pocket was observed just after Ϸ350 ps. During the rest of the simulation the two C1-methyl Tolrestat Inhibition of Retinaldehyde Reductase Activity of AKR1B10 groups pointed toward the interior of the enzyme. In the anion- in Vitro and in Vivo. In a previous work we tested the effect of binding pocket, the carbonyl group was placed at catalytic dis- tolrestat on the retinaldehyde reductase activity of AKR1B1 and tances (Ϸ2.8 Å) from the hydroxyl group of Tyr-49 and the cofactor AKR1B10, using Tween 80 as a retinoid solvent (3). Similar Ki C4 atom (SI Fig. 7 A and B). Besides this, the larger size of the values were obtained for the two enzymes, but they were 10-fold retinaldehyde molecule with respect to tolrestat resulted in new higher than those reported for AKR1B1 using glyceraldehyde as a hydrophobic interactions that were mainly located in the more substrate (11). Here we reevaluated the effect of the inhibitor on the external part of the . Residues implicated in retinalde- retinaldehyde reductase activity of AKR1B10 measured by HPLC hyde binding are strictly conserved in the inner part of the pocket

Gallego et al. PNAS ͉ December 26, 2007 ͉ vol. 104 ͉ no. 52 ͉ 20765 Downloaded by guest on September 28, 2021 ϩ Fig. 2. Crystal structure of AKR1B10 complexed with NADP and tolrestat. (A) View from the top of the (␣/␤)8 barrel. The catalytic site is located in the center of the barrel. (B) View of the (␣/␤)8 barrel after rotating 90°. Cofactor approaches the catalytic site from one side of the barrel, and tolrestat enters the barrel from the upper face. (C) LIGPLOT (32) describing interactions of the tolrestat molecule in the AKR1B10–NADPϩ–tolrestat complex. (D) LIGPLOT of tolrestat in the AKR1B1–NADPϩ–tolrestat complex.

(conformed by Trp-21, Tyr-49, Trp-80, His-111, Phe-116, Phe-123, differences between AKR1B10 and AKR1B1 with retinoids, Trp-220, and Cys-299), resulting in equivalent interactions for the AKR1B10 mutants K125L, V301L, and K125L/V301L were two enzymes. In contrast, all residues (except Pro-124) differ prepared. In these substitutions the AKR1B10 residues were between AKR1B1 and AKR1B10 in the more external part of changed to the corresponding ones in AKR1B1. the channel: Leu125Lys and Val131Ala (loop A), Leu301Val, The kinetics of AKR1B1, AKR1B10, and the AKR1B10 mutants Ser303Gln, and Cys304Ser (loop C). In particular, whereas Leu-125 were analyzed toward D,L-glyceraldehyde and the all-trans- and results in tight packing in AKR1B1 (Fig. 3 C and D), the side chain 9-cis-retinaldehyde isomers (Table 1). Comparison of AKR1B1 and of Lys-125 is substantially displaced with respect to the conforma- AKR1B10 revealed distinct kinetic features. AKR1B1 exhibited a tion seen in the crystal structure of AKR1B10 with tolrestat (Fig. 100-fold-lower Km value with glyceraldehyde, but a similar kcat.In 3 A and B). contrast, Km values for all-trans-retinaldehyde were similar for the Models for the complexes of AKR1B10 and AKR1B1 with two enzymes, but kcat was 100-fold higher in AKR1B10. Remark- 9-cis-retinaldehyde were similarly obtained (Fig. 4). In AKR1B10, ably, the two AKRs showed identical kinetic results with 9-cis- all residues adopted essentially the same conformation found in the retinaldehyde, with low Km and kcat values. all-trans isomer complex, with the exception of loop A residues (SI The AKR1B10 mutants, built to mimic the retinoid-binding site of AKR1B1, changed the kinetic constants in the direction pre- Fig. 8). The rmsd between the backbone atoms of the all-trans and dicted by our models. Thus, with glyceraldehyde, a small substrate 9-cis complexes is much higher for loop A (2.4 Å) than for the rest that occupies a binding position far from the substituted residues, of the protein (0.6 Å). The most noticeable change in the binding- the kinetic constants were identical to those of wild-type AKR1B10. pocket residues with respect to the all-trans complex corresponds to Interestingly, activities with all-trans-retinaldehyde revealed a 13.5- Lys-125, which in the 9-cis complex retained the conformation fold decrease in the k value for the K125L mutant and a 3.5-fold observed for tolrestat binding. The different conformation of cat decrease for the V301L mutant. Although the Km value did not Lys-125 side chain in the complexes with all-trans and 9-cis sub- change for the V301L enzyme, it showed a 4-fold-lower value for strates is apparent from the representation of the evolution of the ␣ ␤ ␥ the K125L mutant. Both mutations failed to modify the kinetic N-C -C -C dihedral angle during the simulation (SI Figs. 9 and properties of AKR1B10 toward the 9-cis isomer. The double 10). Thus, when comparing the four complexes obtained with the ␧ mutation did not further decrease the kcat value for all-trans- two enzymes and the two retinaldehyde isomers, the N atom of retinaldehyde, but yielded a lower k /K than the single mutations. ϩ cat m Lys-125 in the AKR1B10–NADP –all-trans-retinaldehyde com- In addition, the mutants exhibited IC50 values with tolrestat plex has to move 4 Å away to allow for the complex formation. The identical to that for wild-type AKR1B10 (data not shown), dem- location of residue 125 at the entrance of the binding pocket may onstrating that the mutated residues, important for retinoid kinet- have variable effects on substrate binding and/or product release. ics, are not relevant for tolrestat binding.

Site-Directed Mutagenesis. To investigate the participation of Kinetics with 4-Hydroxyretinaldehyde. The structural models of the residues predicted by the modeling in the remarkable kinetic complexes AKR–retinaldehyde indicate that position 125 and

20766 ͉ www.pnas.org͞cgi͞doi͞10.1073͞pnas.0705659105 Gallego et al. Downloaded by guest on September 28, 2021 Ϫ 1 , 170 160 190 140 m Ϯ Ϯ Ϯ Ϯ ⅐ min / K ND Ϫ 1 cat k mM Ϫ 1 0.1 1,500 0.1 1,300 0.10.01 1,300 1,800 Ϯ Ϯ Ϯ Ϯ ND , min cat k 9- cis -Retinaldehyde M 0.1 0.7 0.1 0.9 0.10.03 0.9 0.60

␮ Ϯ Ϯ Ϯ Ϯ , ND m K Ϫ 1 , m 7,600* 0.7 2,000 0.7 30* 0.4 1,400 0.30 320 ⅐ min / K Ϯ Ϯ Ϯ Ϯ Ϯ Ϫ 1 cat k mM

Fig. 3. Models of all-trans-retinaldehyde docked into the AKR1B10 and AKR1B1 structures. (A) Tolrestat-binding pocket in the AKR1B10–NADPϩ– Ϫ 1 0.01* 320 1* 45,000 0.06 13,100 0.3 12,700 tolrestat crystal. (B) all-trans-retinaldehyde-binding pocket of AKR1B10 pre- 0.01 3,500 ϩ Ϯ Ϯ Ϯ Ϯ dicted by our model. (C) Tolrestat-binding pocket in the AKR1B1–NADP – Ϯ , min

tolrestat crystal (PDB entry 2FZD). (D) all-trans-retinaldehyde-binding pocket of cat AKR1B1 predicted by our model. The molecular surface is colored according to k

the local electrostatic potential as calculated with the program PYMOL (www- all - trans -Retinaldehyde .pymol.org). Residues around the substrate define a highly hydrophobic and well adjusted pocket, protecting the retinaldehyde molecule from the polar solvent. M 0.1* 0.35 0.02 2 0.1* 27 0.1 7.7 0.3 3.5

␮ Ϯ Ϯ Ϯ Ϯ Ϯ , m 1 K 1.1 0.6 0.6

others from the external part of the substrate-binding pocket 0.15 interact with the cyclohexene ring. If these interactions contribute to the distinct specificity of the AKRs with retinoids, a structural change in the ring is likely to affect the kinetics. To test this Ϫ 1

hypothesis, (S)-all-trans-4-hydroxyretinaldehyde was synthesized , 80 m 1 1 0.5 1 ⅐ min Ϯ and the kinetics were performed. Unfortunately, the BSA/HPLC / K Ϯ Ϯ Ϯ Ϯ Ϫ 1 cat method was not feasible, essentially because BSA did not efficiently 7 k 660

solubilize the hydroxyl-retinoid and because of a low recovery in the mM extraction with organic solvents. Therefore, the Tween 80/ spectrophotometer methodology was used, although we had shown that this method does not yield reliable Km values because of Ϫ 1 1 16 0.1 0.9 4.9 17 Ϯ Ϯ Ϯ Ϯ Ϯ BIOCHEMISTRY , min -Glyceraldehyde L , cat D k 0.01 31 135 0.8 35.6 0.7 41 0.7 35.4 Ϯ Ϯ Ϯ Ϯ Ϯ ,mM m 6 K 5.2 6.0 0.05

Fig. 4. Retinaldehyde isomers docked into the AKR1B10 and AKR1B1 struc- tures. (A) Superimposition of critical residues for 9-cis-retinaldehyde (blue) and

all-trans-retinaldehyde (orange) binding predicted by the AKR1B10 model. (B) ND, not determined. AKR1B1 AKR1B10 AKR1B10 K125L *Data taken from ref. 7. Table 1. Kinetic analysis for human AKRs and mutant enzymes using the BSA/HPLC method AKR1B10 V301L Superimposition for 9-cis- and all-trans-retinaldehyde in the AKR1B1 model. AKR1B10 K125L/V301L 7.3

Gallego et al. PNAS ͉ December 26, 2007 ͉ vol. 104 ͉ no. 52 ͉ 20767 Downloaded by guest on September 28, 2021 Table 2. Catalytic constants with all-trans-4-OH-retinaldehyde in tion of AKR1B10 by AKR1B1 inhibitors may alter retinoid me- Tween 80 tabolism and contribute to their nonspecific actions. Hence, the

Ϫ1 design and selection of AKR1B1 inhibitors should also consider kcat, min their effect on AKR1B10. The recognition of the structural basis of all-trans- (S)-all-trans-4- the AKR1B10 specificity with retinoids, and the interest of design- Retinaldehyde Hydroxyretinaldehyde ing specific inhibitors of potential anticancer activity, make relevant the crystallization and determination of the x-ray structure of the AKR1B1 0.37 Ϯ 0.02 5.80 Ϯ 0.16 AKR1B10–NADPϩ–tolrestat complex presented here. Tolrestat is AKR1B10 17.8 Ϯ 1.0 17.3 Ϯ 1.5 bound in the substrate-binding pocket, establishing essentially the AKR1B10 K125L 4.40 Ϯ 0.02 18.5 Ϯ 2.6 same interactions as in AKR1B1 (13). AKR1B1 and AKR1B10 Activities were performed in 0.1 M sodium phosphate (pH 7.5)/0.2 mM share 70% sequence identity, and, accordingly, their three- NADPH/0.02% Tween 80 at 25°C (7). dimensional structures are very similar, with an rmsd value between the C␣ atoms of 0.95 Å. Therefore, it was expected that a compar- ative examination of their structures might explain the 100-fold- apparent competitive inhibition by Tween 80. Nevertheless, the kcat higher catalytic efficiency of AKR1B10 with all-trans-retinalde- results were quite consistent with those obtained with the BSA/ hyde. Attempts to obtain crystals of AKR1B1 and AKR1B10 with HPLC method (7). Comparison of the kcat values for the all-trans- NADPϩ and retinoids were unsuccessful. Very few structures of and (S)-all-trans-4-OH-isomers (Table 2) strongly supports the enzymes with retinoid analogs have been reported so far (19), and participation of the cyclohexene ring in the specificity for retinoids. not a single structure has been experimentally Thus, the OH substitution dramatically increased the kcat value of obtained. Therefore, models of the two enzymes with all-trans- and the low-activity enzymes AKR1B1 and AKR1B10 K125L whereas 9-cis-retinaldehyde were built based on the crystal structures with its value did not change for the highly active AKR1B10. NADPϩ and tolrestat. all-trans-Retinaldehyde is a larger molecule than tolrestat, which results in its cyclohexene ring interacting with Discussion the external part of the cleft, mainly the loops A and C. Comparison Previous reports have shown a high catalytic constant (kcat) for the of the models for AKR1B1 and AKR1B10 indicates that the residue retinaldehyde reductase activity of AKR1B10 (3, 7), which is at position 125 (Lys in AKR1B10 and Leu in AKR1B1) shows the unique in human AKRs, making the enzyme comparable to largest differences. In AKR1B10, binding of the all-trans- retinaldehyde reductases from the SDR superfamily in terms of retinaldehyde molecule requires that Lys-125 move toward the catalytic efficiency (kcat/Km). Here we have confirmed this obser- solvent. Such a displacement is not needed in any of the other vation and have demonstrated the remarkable specificity of models analyzed: AKR1B10 with 9-cis-retinaldehyde and AKR1B1 AKR1B10 toward all-trans-retinaldehyde. Thus, activity with 9-cis- with all-trans-or9-cis-retinaldehyde. Therefore, Lys-125 appears to retinaldehyde, also a physiological compound, is 35-fold lower be a key residue for AKR1B10 to attain the highest catalytic (Table 1), similar to the activity of other human AKRs, such as efficiency with all-trans-retinaldehyde. AKR1B1, with any retinaldehyde isomer. Therefore, AKR1B10 is To explore this hypothesis, the AKR1B10 K125L mutant was strictly an all-trans-retinaldehyde reductase. studied. As expected, the activity with glyceraldehyde was not The multiple reports on the overexpression of AKR1B10 in affected by a mutation located in a distant position from the binding different human cancer types (1, 4–6, 14) have increased the search pocket of the small substrate. In contrast, with all-trans- for physiological or xenobiotic substrates for the enzyme that could retinaldehyde, the kcat value decreased by 13.5-fold with respect to reveal a link between a physiological function and its participation that of the wild-type enzyme. Consistently, all-trans-retinaldehyde in cancer development. Thus, the reducing activity of AKR1B10 has modeling into AKR1B10 K125L did not require any rearrangement been found with the tobacco carcinogen 4-(methylnitrosamino)-1- in the mutated residue, similar to what had been observed for the (3-pyridyl)-1-butanone, the antiemetic, type 3 serotonin (5-HT3) other enzyme–substrate pairs having poor activity. Kinetics with receptor antagonist dolasteron, and the antitumor drugs daunoru- 9-cis-retinaldehyde was not affected by the mutation. These results bicin and oracin, suggesting that AKR1B10 may play a role in the fully support a relationship between residue-125 conformational chemoresistance of tumors toward carbonyl group-bearing drugs change and AKR1B10 specificity. The AKR1B10 V301L mutant (15). In addition, AKR1B10 is able to activate polycyclic aromatic also exhibited a small decrease in the kcat value for all-trans- hydrocarbons (16). However, these substrates exhibit 1,000-fold- retinaldehyde, indicating that other residues of the cyclohexene- lower catalytic efficiencies than all-trans-retinaldehyde, stressing binding site may also contribute to AKR1B10 specificity. The effect the extreme specificity of the enzyme toward this retinoid and on kcat was not additive in the AKR1B10 K125L/V301L double reinforcing the concept that a pathological effect of AKR1B10 mutant, but its kcat/Km value further decreased with respect to the overexpression in cancer may be through an alteration of the single mutants. retinoic acid synthesis pathway. The involvement of the substrate-binding region that interacts AKR1B10, expressed in COS-1 cells, efficiently reduced retinal- with the cyclohexene ring in the specificity for retinoids was further dehyde, providing evidence that the enzyme functions as a retinal- supported by using (S)-all-trans-4-hydroxyretinaldehyde as sub- dehyde reductase also in vivo. The in vitro and in vivo activity of strate. The cyclohexene hydroxyl group, well oriented toward the AKR1B10 was inhibited by tolrestat, a well known AKR1B1 125 position as observed in the models (SI Fig. 11), has a strong inhibitor designed to prevent diabetic complications by blocking influence on the kinetics (Table 2). When interacting with a sorbitol accumulation. Tolrestat decreased retinaldehyde reductase hydrophilic residue, such as Lys-125 in the highly active AKR1B10, activity of AKR1B10 with the same IC50 as that reported for the the kcat value is not altered. However, if a hydrophobic residue is inhibition of glyceraldehyde reduction by AKR1B1 (IC50 Ϸ 10 nM) found, such as in the less active AKR1B1 or AKR1B10 K125L, then (12). Moreover, tolrestat inhibited AKR1B10 expressed in COS-1 the kcat value increases. Apparently, to reach high activity a hydro- cells with an effect similar to that found on AKR1B4 (the rat philic group, either in the substrate cyclohexene moiety or in the ortholog of AKR1B1, IC50 Ϸ 1 ␮M) (17). Tolrestat is therefore an enzyme position 125, is required. The two situations, either sepa- efficient AKR1B10 inhibitor, and thus the overexpression of rately or simultaneously present, provide an effect on the substrate AKR1B10 in cancer could be counteracted by tolrestat or similar binding or, most probably, on the product release, which is AKR1B1 inhibitors, suggesting potential applications in cancer translated into a higher kcat value. control. On the other hand, many AKR1B1 inhibitors have side The strong differences in kcat values suggest that the limiting step effects precluding their pharmacological use (10, 18). The inhibi- in the reaction mechanism with retinaldehyde differs between the

20768 ͉ www.pnas.org͞cgi͞doi͞10.1073͞pnas.0705659105 Gallego et al. Downloaded by guest on September 28, 2021 enzymes studied. The limiting step for the low-activity species 2695 HPLC. Elution was monitored at 350 nm with a Waters 2996 photodiode seems slower than cofactor dissociation, the common limiting step array detector (7). for AKRs with the best substrates (20, 21). The present results are Cell Culture and Transfection. Cells were grown in DMEM supplemented with consistent with the dissociation of the retinol product being limiting 10% (vol/vol) FBS (GIBCO BRL). Transfection was done with Lipofectamine Plus in these cases, but this should be further investigated by using (Invitrogen) as detailed in SI Text. transient kinetics analysis. In conclusion, the structural features that make AKR1B10 highly Crystallization and Structure Determination. Crystals were grown at 20°C by efficient for all-trans-retinaldehyde are localized in the external part vapor diffusion using the hanging-drop method. One microliter of protein solu- of the substrate-binding site, including positions 125 and 301, where tion (18 mg/ml) was mixed with 1 ␮l of precipitant solution containing polyeth- ylene glycol 6000 and 100 mM sodium cacodylate (pH 9.0). The hanging drop was the retinoid cyclohexene ring binds. Substitutions in the cyclohex- equilibrated with 0.8 ml of precipitant solution. Crystals belonged to the hexag- ene ring also influence the specificity. These structural character- onal space group P61 (a ϭ b ϭ 89.1 Å, c ϭ 78.4 Å, ␣ ϭ ␤ ϭ 90.0°, ␥ ϭ 120.0°), with istics, in both the enzyme and the ligand, should be taken into one molecule in the asymmetric unit. Data were collected at 100 K on an ADSC account for the rational design of new and more specific inhibitors Q4R CCD detector in the beamline ID14-4 at the European Synchrotron Radiation against AKR1B10 for potential cancer control and against Facility. Diffraction data were integrated and scaled by using DENZO and SCALE- PACK (24). The structure was solved by molecular replacement with MOLREP (25), AKR1B1 for diabetes treatment. using as a searching model the PDB coordinates with entry 1FRB (26). Subsequent refinement was carried out by using REFMAC (27) and manual model building in Materials and Methods O (28) (SI Table 3). Building of the AKR1B10–NADPϩ–retinaldehyde complexes Site-Directed Mutagenesis. K125L, V301L, and K125L/V301L mutants were ob- was performed with the graphic program O and regularized with the idealization tained by standard procedures performed as described in SI Text. option in REFMAC. AUTODOCK 3.05 software (29) was used to build the initial structure of the AKR1B1–NADPϩ–retinaldehyde complexes, using the crystal Expression and Purification of AKR1B10. Wild-type AKR1B10 and mutants were structure of AKR1B1 (13) (PDB entry 2FZD). purified as described (7). Details of purification are presented in SI Text. Molecular Dynamics Simulations. Molecular dynamics simulations were per- formed to allow retinaldehyde substrates to accommodate in the binding cavity . Standard activities were measured before each kinetic experi- as described in SI Text. Analysis of the trajectories was carried out by using ment by using D,L-glyceraldehyde as a substrate (3). D,L-Glyceraldehyde kinetics standard tools of AMBER (30) and with VMD (31). were performed as described (3) but in the presence of 90 mM potassium phosphate (pH 7.4)/40 mM KCl (reaction buffer) at 37°C. Activity with retinoids in Coordinates. Atomic coordinates and structure factors of the AKR1B10–NADPϩ– the presence of BSA was performed in reaction buffer at 37°C in siliconized glass tolrestat complex have been deposited in the PDB (ID code 1ZUA). tubes (22), and the products were analyzed by HPLC. Activity in the presence of Tween 80 was determined as described (7). The kinetic constants were expressed ACKNOWLEDGMENTS. We thank Professor T. Geoffrey Flynn (Department of as the mean Ϯ SEM of at least three independent determinations. (S)-all-trans- Biochemistry, Queen’s University, Kingston, Ontario, Canada) for his support. 4-Hydroxyretinaldehyde was synthesized as reported (23). We acknowledge the computer support, technical expertise, and assistance provided by the Barcelona Supercomputing Center—Centro Nacional de Su- percomputacio´n. This work was supported by grants from Direccio´n General HPLC Analysis. After extraction, retinoids were separated by chromatography on de Investigacio´n (BMC2003-09606, BFU2005-02621, BFU2005-08686-C02-01, a Spherisorb S3W column (4.6 ϫ 100 mm; Waters) in hexane:methyl-tert-butyl and FIS2005-00655) and Generalitat de Catalunya (2005SGR-00036 and ether (96:4, vol/vol) mobile phase, at a flow rate of 2 ml/min using Waters Alliance 2005SGR-00112).

1. Cao D, Fan ST, Chung SS (1998) J Biol Chem 273:11429–11435. 16. Quinn AM, Penning TM (2005) Chem Res Toxicol 18:1986. 2. Hyndman DJ, Flynn TG (1998) Biochim Biophys Acta 1399:198–202. 17. Ramana KV, Chandra D, Srivastava S, Bhatnagar A, Aggarwal BB, Srivastava SK (2002) 3. Crosas B, Hyndman DJ, Gallego O, Martras S, Pare´s X, Flynn TG, Farre´s J (2003) Biochem J Biol Chem 277:32063–32070. J 373:973–979. 18. Boulton AJ, Vinik AI, Arezzo JC, Bril V, Feldman EL, Freeman R, Malik RA, Maser RE, 4. Scuric Z, Stain SC, Anderson WF, Hwang JJ (1998) Hepatology 27:943–950. Sosenko JM, Ziegler D, American Diabetes Association (2005) Diabetes Care 28:956– 5. Fukumoto S, Yamauchi N, Moriguchi H, Hippo Y, Watanabe A, Shibahara J, Taniguchi 962. H, Ishikawa S, Ito H, Yamamoto S, et al. (2005) Clin Cancer Res 11:1776–1785. 19. Pakhomova S, Buck J, Newcomer ME (2005) Protein Sci 14:176–182. 6. Nagaraj NS, Beckers S, Mensah JK, Waigel S, Vigneswaran N, Zacharias W (2006) Toxicol 20. Kubiseski TJ, Hyndman DJ, Morjana NA, Flynn TG (1992) J Biol Chem 267:6510–6517. Lett 165:182–194. 21. Grimshaw CE, Bohren KM, Lai CJ, Gabbay KH (1995) Biochemistry 34:14356–14365. 7. Gallego O, Belyaeva OV, Porte´S, Ruiz FX, Stetsenko AV, Shabrova EV, Kostereva NV, 22. Gough WH, VanOoteghem S, Sint T, Kedishvili NY (1998) J Biol Chem 273:19778–19785. Farre´s J, Pare´s X, Kedishvili NY (2006) Biochem J 399:101–109. 23. Domı´nguezM, Alvarez R, Borra`s E, Farre´s J, Pare´s X, de Lera AR (2006) Org Biomol Chem 8. Blaner WS, Olson JA (1994) in The Retinoids: Biology, Chemistry and Medicine, eds 4:155–164. Sporn MB, Roberts AB, Goodman DS (Raven, New York), pp 229–255. 24. Otwinowski Z, Minor W (1997) Methods Enzymol 276:307–326. 9. Penning TM (2005) Clin Cancer Res 11:1687–1690. 25. Vagin A, Teplyakov A (1997) J Appl Crystallogr 30:1022–1025. 10. Foppiano M, Lombardo G (1997) Lancet 349:399–400. 26. Wilson DK, Nakano T, Petrash JM, Quiocho FA (1995) Biochemistry 34:14323–14330. 11. Vander Jagt DL, Hunsaker LA, Robinson B, Stangebye LA, Deck LM (1990) J Biol Chem 265:10912–10918. 27. Collaborative Computational Project, Number 4 (1994) Acta Crystallogr D 50:760–763. 12. Ehrig T, Bohren KM, Prendergast FG, Gabbay KH (1994) Biochemistry 33:7157–7165. 28. Jones TA, Zou JY, Cowan SW, Kjeldgaard M (1991) Acta Crystallogr A 47:110–119. BIOCHEMISTRY 13. Steuber H, Zentgraf M, Gerlach C, Sotriffer CA, Heine A, Klebe G (2006) J Mol Biol 29. Morris GM, Goodsell DS, Halliday RS, Huey R, Hart WE, Belew RK, Olson AJ (1998) 363:174–187. J Comput Chem 19:1639–1662. 14. Zeindl-Eberhart E, Haraida S, Liebmann S, Jungblut PR, Lamer S, Mayer D, Jager G, 30. Pearlman DA, Case DA, Caldwell JW, Ross WS, Cheatham TE, Debolt S, Ferguson D, Chung S, Rabes HM (2004) Hepatology 39:540–549. Seibel G, Kollman PA (1995) Comput Phys Commun 91:1–41. 15. Martin HJ, Breyer-Pfaff U, Wsol V, Venz S, Block S, Maser E (2006) Drug Metab Dispos 31. Humphrey W, Dalke A, Schulten K (1996) J Mol Graphics 14:33–38. 34:464–470. 32. Wallace AC, Laskowski RA, Thornton JM (1995) Protein Eng 8:127–134.

Gallego et al. PNAS ͉ December 26, 2007 ͉ vol. 104 ͉ no. 52 ͉ 20769 Downloaded by guest on September 28, 2021