Unusual charge stabilization of NADP+ in 17beta-hydroxysteroid dehydrogenase.

Type 1 17beta-hydroxysteroid dehydrogenase (17beta-HSD1), a member of the short chain dehydrogenase reductase (SDR) family, is responsible for the synthesis of 17beta-estradiol, the biologically active estrogen involved in the genesis and development of human breast cancers. Here, we report the crystal structures of the H221L 17beta-HSD1 mutant complexed to NADP+ and estradiol and the H221L mutant/NAD+ and a H221Q mutant/estradiol complexes. These structures provide a complete picture of the NADP+-enzyme interactions involving the flexible 191-199 loop (well ordered in the H221L mutant) and suggest that the hydrophobic residues Phe192-Met193 could facilitate hydride transfer. 17beta-HSD1 appears to be unique among the members of the SDR protein family in that one of the two basic residues involved in the charge compensation of the 2'-phosphate does not belong to the Rossmann-fold motif. The remarkable stabilization of the NADP+ 2'-phosphate by the enzyme also clearly establishes its preference for this cofactor relative to NAD+. Analysis of the catalytic properties of, and estradiol binding to, the two mutants suggests that the His221-steroid O3 hydrogen bond plays an important role in substrate specificity.

Type 1 17␤-hydroxysteroid dehydrogenase (17␤-HSD1), a member of the short chain dehydrogenase reductase (SDR) family, is responsible for the synthesis of 17␤-estradiol, the biologically active estrogen involved in the genesis and development of human breast cancers. Here, we report the crystal structures of the H221L 17␤-HSD1 mutant complexed to NADP ؉ and estradiol and the H221L mutant/NAD ؉ and a H221Q mutant/estradiol complexes. These structures provide a complete picture of the NADP ؉ -enzyme interactions involving the flexible 191-199 loop (well ordered in the H221L mutant) and suggest that the hydrophobic residues Phe 192 -Met 193 could facilitate hydride transfer. 17␤-HSD1 appears to be unique among the members of the SDR protein family in that one of the two basic residues involved in the charge compensation of the 2-phosphate does not belong to the Rossmann-fold motif. The remarkable stabilization of the NADP ؉ 2-phosphate by the enzyme also clearly establishes its preference for this cofactor relative to NAD ؉ . Analysis of the catalytic properties of, and estradiol binding to, the two mutants suggests that the His 221 -steroid O 3 hydrogen bond plays an important role in substrate specificity.
The 17␤-estradiol (E 2 ) is known to promote the genesis and development of human breast cancers (1,2). Its presence in tumor cells comes from in situ synthesis (3), and its concentration is significantly increased in malignant breast tissues (4 -6). Type 1 17␤-hydroxysteroid dehydrogenase (17␤-HSD1) 1 catalyzes the reversible transformation of estrone (E 1 ) into the biologically active estradiol (E 2 ) (7). Thus, preventing the formation of E 2 by a specific inhibition of 17␤-HSD1 activity should be of paramount importance for cancer therapy.
Based on amino acid sequence alignments, 17␤-HSD1 was thought to belong to the group of NAD(H)-preferring enzymes (22). However, biochemical studies (7) and the structure of the 17␤-HSD1⅐E 2 ⅐NADP ϩ ternary complex (25) have shown that 17␤-HSD1 is able to bind both NAD(H) and NADP(H). 17␤-HSD1 appears to be unique among the SDR family because it lacks both the aspartic acid residue at position 36 (Leu 36 in 17␤-HSD1), characteristic of NAD(H) preferring enzymes, and the basic residue located in the consensus sequence of the dinucleotide binding motif Gly-Xaa-Xaa-Xaa-Gly-Xaa-Gly (which is replaced by Ser 12 in 17␤-HSD1). This motif forms an ionic interaction with the ribose 2Ј-phosphate and is characteristic of NADP(H)-preferring enzymes.
His 221 , first identified by affinity labeling studies (27)(28)(29), was thought to be involved in the specific binding of the steroid. Indeed, the construction of an H221A mutant led to an enzyme displaying a higher K m and lower V max relative to the wild type (16). Furthermore, the crystal structure of the enzyme complexed with E 2 (25,26) revealed that His 221 is directly involved in the specific binding of the steroid.
Here, we report the construction of H221L and H221Q mutants, their characterization by enzymatic assays, their crystallization, and the determination of the structures of the binary complexes H221Q⅐E 2 and H221L⅐NAD ϩ and the ternary complex H221L⅐NADP ϩ ⅐E 2 at 2.7, 3.0, and 2.7 Å resolution, respectively. We show for the first time a well ordered conformation for the 191-199 loop and speculate about its role in cofactor binding and hydride transfer. Moreover, the specificity of the enzyme for estrogens is reassessed. lactalbumin hydrolysate. The Sf9 cells were grown as monolayers at 28°C in TNM-FH medium supplemented with 10% fetal bovine serum. Cells were infected with virus at a multiplicity of infection of 0.1-1 plaque-forming unit to produce virus stocks or at multiplicity of infection Ͼ10 for maximal protein expression. Cells were harvested 60 h after infection.
Site-directed Mutagenesis-The two mutants H221Q and H221L were constructed with one round of polymerase chain reaction made on the pVL/17␤-HSD transfer vector previously constructed for 17␤-HSD1 overexpression in baculovirus (30). These mutations use two primers. The first one is located on the cDNA, upstream of a PstI unique site (5Ј-GTAGTAGGGACTGTGCGG-3Ј). The second one introduces the mutation. For H221Q and H221L, it overlaps the codon to be mutated and a NruI unique site (5Ј-GCCGCCTCGCGAAAGACTTGCTTGCTT-TGGGCGAGG-3Ј and 5Ј-GCCGCCTCGCGAAAGACTTGCTTGCTGA-GGGCGAGG-3Ј, respectively). Amplified fragments and pVL/17␤-HSD1 were digested with NruI and PstI. The mutated fragments were then cloned instead of the nonmutated one. Mutations were checked by dideoxynucleotide sequencing, and mutated transfer vectors were cotransfected in Sf9 cells with the wild-type AcNPV virus following the protocol described by Invitrogen. Recombinant viruses were purified by plating.
Protein Purification and Enzyme Assay-The purification of mutant proteins was performed as described previously for the wild-type protein (30,25). The only modification concerned the H221L mutant which was purified at pH 7.0 instead of pH 7.5. The activities of the wild-type and the mutated 17␤-HSD1s were measured as described by Langer and Engel (31) with few modifications. The enzymes were assayed by spectrophotometric measurement of the concentration changes of NADPH at 340 nm. Reactions were run in 50 mM NaHCO 3 /Na 2 CO 3 buffer, pH 9.2, for the oxidation and 50 mM KH 2 PO 4 /K 2 HPO 4 buffer, pH 5.8, for the reduction, both containing 0.5-16 M steroid (estradiol or estrone, respectively). In both cases, 10 l of enzyme preparation (from 55 to 90 g/ml) were added to 500 l of buffer. Reactions were initiated by the addition of 5 l of a 10 mM NADP ϩ or NADPH solution, and the absorbance variations were measured against a blank containing all components except steroid and coenzyme. Controls containing all components except steroid were also run. Reactions were followed from 30 s to 3 min after addition of NADP ϩ on a Shimadzu UV-160A spectrophotometer. For NADP ϩ reduction measurements, the same protocol was used except that the estradiol concentration was fixed to 25 M while NADP ϩ concentrations varied from 0.05 up to 5 mM. Velocities were calculated from the slopes of the zero order portion of the kinetics obtained and were corrected for the control absorbance. The enzymatic activity was calculated as described previously (8) and defined in units per milliliter (taking a micromole per min as a unit). Lineweaver-Burk plots were used to determine the K m and the V max values. Protein concentrations of enzyme preparations were measured using a micro BCA protein reagent kit from Pierce. The catalytic activity parameters are presented in Tables IV and V. Crystallization-All recombinant mutant enzymes were concentrated to 5 mg/ml, in a buffer containing 40 mM Tris, pH 7.5, 1 mM EDTA, 0.2 mM DTT, 20% glycerol, and 2 mM ␤-octylglucoside and crystallized at room temperature. The H221Q mutant was crystallized in conditions similar to those of the wild-type enzyme (100 mM HEPES, pH 7.0, 100 mM MgCl 2 , 0.5 mM E 2 , 2-4% propanediol, 30% polyethylene glycol 4000) (25,32); crystals appeared after 6 days. The H221L mutant was crystallized in 100 mM sodium phosphate buffer, pH 6.0 -6.3, 1 mM NAD(P) ϩ , 100 mM NaCl and 2.2-4.4 mM decyl-␤-D-maltoside or 9 -18 mM octyl-␤-D-thioglucopyranoside and 2-2.4 M ammonium sulfate as precipitant. Space groups and unit cell dimensions of these crystals are summarized in Table I.
The H221L crystals used were incubated (from a few minutes up to one night) in a buffer containing 100 mM HEPES, pH 6.5, 1 mM NADP ϩ , 100 mM NaCl, 30% PEG 4000 with and without 0.5 mM E 2 to obtain both the ternary and the binary complex. Crystals were then flash-cooled under a nitrogen gas stream at Ϫ150°C, transferred into liquid propane at Ϫ160°C, and conserved in solid propane at Ϫ196°C until data collection was carried out at a synchrotron radiation source.
Data Collection and Processing-A data set of the H221Q mutant was collected at room temperature on the in-house Siemens/Xentronics area detector X1000 system, mounted on a Rigaku RU200 rotatinganode-x-ray generator. Two data sets of the H221L mutant were collected at low temperature (Ϫ150°C) on the D2AM French-Collaborative Research Group synchrotron beam line at the European Synchrotron Radiation Facility (ESRF, Grenoble), using an XRII-CCD detector (33). Data were processed with XENGEN (34) for the former and with a modified version of XDS (35, 36) for the others. Data collection statistics are summarized in Table I.
Structure Determination-For the H221Q mutant, a rigid body refinement with the model of the wild-type 17␤-HSD1 was sufficient to place it correctly in the unit cell. The H221L mutant structure was solved by molecular replacement methods with the wild-type 17␤-HSD1 as a search model using AMoRe (37). Four molecules per asymmetric unit were found. The mutants structures were refined with X-PLOR (Version 3.1) (38) and REFMAC (CCP4 Suite of Programs) (39), and model corrections were made with O (40). The refinement statistics are presented in Table II. The coordinates have been deposited with the Protein Data Bank (codes: 1FDU for the H221L⅐E 2 ⅐NADP ϩ complex, 1FDV, for the H221L⅐NAD ϩ complex, and 1FDW for the H221Q⅐E 2 complex).

RESULTS
Two different crystalline complexes were obtained with the H221L mutant: the first one by co-crystallizing the protein with NADP ϩ and diffusing E 2 in the crystal, and the second one by crystallizing the protein in the presence of NAD ϩ . Both complexes crystallize in the monoclinic P2 1 space group (Table I) with four molecules in the asymmetric unit corresponding to two biologically active dimers (Fig. 1), here named mA/mB and mC/mD, respectively.
In the H221L⅐NADP ϩ ⅐E 2 ternary complex, the electron density is very well defined all along the polypeptide chain, including the 191-199 loop that was disordered in the wild-type  (25): a ϭ 122. 8  Unusual Charge Stabilization of NADP ϩ in 17␤-HSD1 structure (Fig. 2). The observed conformation for this loop is completely different from that proposed by Ghosh et al. (24) but the 190 -192 segment conformation is similar to that of residues 184 -186 in mouse lung carbonyl reductase (MLCR) (22) (Fig. 3).
The electron density map is also very well defined for the steroid and cofactor (Fig. 4). As a result, NADP ϩ and estradiol have been modeled with full occupancies in the four subunits.
In the H221L⅐NAD ϩ binary complex, the electron density for NAD ϩ and the 191-200 loop is well defined for monomers mA and mC but discontinuous for monomers mD and mB. Consequently, the 191-199 loop was not included in the models of these two monomers. Since this site is known to have Michaelian kinetics, we do not understand the observed difference in NAD ϩ occupancy. It may be due to subtle packing effects difficult to characterize at this resolution.
Both NADP ϩ and NAD ϩ bind in the same extended confor-mation already observed for the wild-type enzyme: the cofactor points toward the active site of the enzyme with the nicotinamide ring in the syn conformation and the adenine in the anti conformation (25). The structure of the H221Q⅐E 2 binary complex displays two conformations for Gln 221 . In one of these, the amide group of the Gln 221 forms a hydrogen bond with the steroid O 17 atom; whereas in the other, Gln 221 is oriented toward the solvent. The double conformation of the Gln 221 may be a consequence of the partial occupation of the steroid-binding site, as it was already suggested in the case of the wild-type enzyme (25). No electron density was found for the cofactor.
Having a complete model of the cofactor binding site allows for a full description of the NAD(P) ϩ /protein interactions. Some of these interactions are common to NADP ϩ and NAD ϩ , and most of them were present in the wild-type ternary complex  (48) and Render (49,50).) (25). However, due to the disorder in the neighborhood of the NADP ϩ binding site in the latter structure, two major interactions were not observed: the extensive hydrophobic contact between the nicotinamide moiety and the Phe 192 side chain, and the charge compensation of the dinucleotide 2Ј-phosphate through salt bridges with Arg 37 and Lys 195 side chains (Figs. 4 and 5). In addition, the 2Ј-phosphate is further stabilized by a hydrogen bond with Ser 11 O ␥ . In the NAD ϩ ⅐H221L complex, the binding pocket of the ribose 2Ј-phosphate of monomers mA and mC is occupied by a sulfate ion, presumably coming from the crystallization solution. This ion is bound to the protein⅐NAD ϩ complex through a hydrogen bond network involving 2Ј-OH adenine ribose, Lys 195 , Arg 37 , Ser 11 , and Thr 41 side chains. This phenomenon has already been observed in glutathione reductase where an inorganic phosphate ion substitutes for the missing 2Ј-phosphate group when NAD ϩ is bound (41). In both NAD ϩ and NADP ϩ complexes, the NH 3 ϩ group of Lys 195 interacts with the O1A of the pyrophosphate.
The 191-199 loop, located between the ␤F sheet and the ␣G helix, seems to be predominantly stabilized by its interactions with the dinucleotide, particularly by the salt bridge between Lys 195 and the 2Ј-phosphate. In the H221L⅐NAD ϩ complex, where a sulfate ion replaces the 2Ј-phosphate, the electron density corresponding to the 191-199 loop is less well defined even though the NAD ϩ site appears to be fully occupied. This implies that although the sulfate ion establishes a series of interactions with both the protein and the cofactor, these are less efficient in stabilizing the 191-199 loop than those formed by the covalently bound 2Ј-phosphate. Once stabilized, the 191-199 loop appears to protect the coenzyme from solvent as the NADP ϩ accessible surface (42) is reduced from 122 Å 2 when the loop is removed to 38 Å 2 when it is well ordered. The E 2 molecule and the residues forming the steroid binding site are well superposed in the two wild-type structures (Table III). On the other hand, the rms differences values resulting from the superposition of each mutant onto the wildtype models show a significant deviation for estradiol, relative to the residues involved in the hydrophobic site (Fig. 6). In this respect, the loss of a hydrogen bond between residue 221 and estradiol may be responsible for the increased steroid mobility observed in the H221L mutant. There is also a significant difference in substrate position between two non-equivalent monomers such as mC and mD. The slight substrate reorientation observed in the H221Q mutant is likely to be due to a shift of the hydrogen bond between E 2 -O 3 and Gln 221 -N ⑀1 .
The catalytic efficiency is strongly affected by mutations at the His 221 position (Table IV). Reduction is decreased 10-fold for the H221L mutant and 4-fold for the H221Q mutant (18-fold and 6-fold, respectively, for the oxidation). This may be partially explained by a 4.4-fold increase in the reduction reaction K m for H221L and a respective 2.3-fold increase for the H221Q mutant (4-fold and 3.4-fold, respectively, for the oxidation reaction). As the histidine-to-glutamine mutation preserves the hydrogen bond with C 3 -OH, this interaction seems to be essential for the catalytic activity. As expected from the structure, the H221L mutation has a limited effect on the cofactor binding site architecture, the K m value for NADP ϩ reduction being close to that of the wild-type enzyme (Table V). DISCUSSION The Role of the 191-199 Loop-The H221L mutant provides the first image of a well ordered 191-199 loop in a 17␤-HSD1 structure. The conformation of this loop in the H221L⅐E 2 ⅐ NADP ϩ complex is very likely to be identical to that of the wild-type enzyme when fully complexed to the cofactor. This is supported, on the one hand, by biochemical results that indicate that the affinity of the H221L mutant enzyme for NADP ϩ is similar to that of the wild-type protein (Table V) and, on the other, by the fact that the 191-199 loop is located at the protein surface, and it is not involved in potentially constraining crystal packing interactions in either structure. The stabilization of the 191-199 loop, that is disordered in the absence of cofactor, is mediated by the interactions of Lys 195 and Phe 192 with NADP ϩ . Furthermore, loop residues Phe 192 and Met 193 shield NADP ϩ from solvent and contribute to the hydrophobic character of the nicotinamide binding pocket.
Other dehydrogenases also have a similarly flexible loop near their dinucleotide binding site. One example is the mobile loop comprising residues 16 -20 in Escherichia coli dihydrofolate reductase which is involved in hydride transfer (43). This loop, which becomes well ordered only in the enzyme⅐ NADP ϩ ⅐folate complex, has been found to shield the nicotinamide moiety from solvent and to participate on the transitionstate stabilization (44). Replacement of the Met 16 -Ala 19 stretch by glycine results in a 550-fold decrease in the hydride transfer rate. Another case is lactate dehydrogenase. As in 17␤-HSD1, a loop comprising residues 97-123 is stabilized by interactions with NADH and shields the cofactor from solvent (45). In the SDR family, E. coli 7␣-HSD (23) exhibits a large conformational change of the 195-210 loop upon substrate binding that also results in shielding of the catalytic site. In the 3␣,20␤-HSD structure (19), a small conformational change of the 184 -189 loop located close to the active site is also observed. In the MLCR⅐NADP ϩ complex (22), the Met 186 interaction with the nicotinamide moiety is similar to the one observed between Phe 192 and the NADP ϩ in 17␤-HSD1 (Fig. 3). However, it is not known whether the 184 -190 loop has the same conformation in the MLCR apoenzyme.
All the SDR loops described above are located between the ␤F sheet and the ␣G helix and are flanked by two proline  residues (Pro 187 and Pro 200 in 17␤-HSD1). As it was previously suggested by Tanaka et al. (23), these two prolines may prevent conformational changes from propagating to the rest of the protein. An amino acid sequence alignment of SDR enzymes reveals that a proline residue located at position 185 (17␤-HSD1 numbering) is highly conserved, except for 17␤-HSD1, where this residue is found at position 187 (Fig. 7). Furthermore, a hydrophobic residue, equivalent to Phe 192 of 17␤-HSD1, is found in 38 out of the 46 amino acid sequences analyzed (Fig. 7). These loops, that are stabilized by both substrate or cofactor, may be characteristic of enzymes belonging to the SDR family. In turn, their protection of the nicotinamide moiety and contribution of hydrophobic residues to the active site structure suggest they are involved in catalytic hydride transfer. The Coenzyme Specificity-As indicated by the different xray complex structures, and by previous biochemical results (7), 17␤-HSD1 is able to bind both NADP ϩ and NAD ϩ cofactors. The 17␤-HSD1⅐NADP ϩ complex has been shown to be stabilized through hydrogen bonds established between the dinucleotide 2Ј-phosphate moiety and the main chain NHs of residues Cys 10 , Ser 11 , and Arg 37 (25). In most known structures of NADP ϩ -preferring enzymes, the two negative charges of the 2Ј-phosphate group are compensated by one or two positively charged residues. Due to disorder of the 191-199 loop, these interactions were not observed in our NADP ϩ /wild-type 17␤-HSD1 complex structure. Now, the H221L mutant structure with its well defined 191-199 loop, allows us to definitely classify 17␤-HSD1 among the NADP ϩ -preferring enzymes. Three residues, Ser 11 , Arg 37 , and Lys 195 , interact with the dinucleotide 2Ј-phosphate group of the cofactor (Fig. 5). Arg 37 forms a salt bridge that compensates one of the two negative charges of the phosphate. Incidentally, this residue, as Arg 39 , is also present in the MLCR⅐NADP ϩ complex (22). In both structures, it is located in the turn between the ␤-sheet B and the ␣-helix C of the ␤A-␣B-␤B-␣C-␤C-␣D-␤D motif that forms the Rossmann fold ( Figs. 1 and 7). This arginine is conserved in 26 of the 46 proteins known to be NADP ϩpreferring enzymes (46) (Fig. 7).
A second basic residue, located in the 4th position of the Gly-Xaa-Xaa-Xaa-Gly-Xaa-Gly consensus sequence of the dinucleotide binding motif, often further compensates the 2Ј-phosphate charge. Indeed, sequence alignment studies reveal that this residue is conserved in 54% of the NADP ϩ -preferring SDR proteins (Fig. 7). In 17␤-HSD1, there is no basic residue in the 4th position. Instead, the 2Ј-phosphate group interacts with Ser 11 , which is located in the third position of the consensus sequence. In fact, among the 21 NADP ϩ -preferring enzymes of the SDR family that do not have a basic residue at this position, 15 are found to have either serine or threonine instead (Fig. 7). In 17␤-HSD1, further charge compensation of the 2Ј-phosphate group is afforded by Lys 195 . This residue is not conserved in the SDR family of amino acid sequences, and superposition of the known SDR structures shows that it is located in a variable loop. Moreover, in most of the NADP ϩ ⅐enzyme complex structures having a Rossmann-fold binding motif (47), the 2Ј-phosphate group interacts with one to three residues located in either the ␤A␣B or the ␤B␣C turn (Fig. 1). Thus 17␤-HSD1 is the first SDR structure for which such an atypical charge compensation is observed. This may be relevant to the understanding of the evolution of the NADP(H) binding motif.
The Role of His 221 in Substrate Specificity-In the wild-type structure, His 221 was found to hydrogen bond to the C 3 -OH of the estradiol moiety. This hydrogen bond is preserved in the H221Q mutant. In the H221L mutant, however, the mutated  leucine residue can only participate in the formation of the hydrophobic pocket of the steroid binding site. Consequently, it can be concluded that this hydrogen bond is not essential to substrate binding. However, comparison of all the available x-ray structures shows that, as a function of the immediate environment, estradiol can occupy slightly different positions in its hydrophobic pocket and that the largest differences concern its C and D rings (Fig. 6). The lack of hydrogen bond formation in the H221L mutant reduces the catalytic efficiency by 9 to 18-fold. This can be compared with a 4-to 5-fold reduction in the H221Q mutant (Table IV). Taken together, these results suggest that the hydrogen bond between His 221 and the C 3 -OH group is important for enzyme specificity because it helps in establishing the catalytically relevant steroid/ protein interactions of O 17 , Tyr 155 O , and Ser 142 O ␥ . These observations may explain the specificity of 17␤-HSD1 for aromatic A ring-containing substrates. With substrates like testosterone or 4-androstenedione, the orientation of the C 3 -O 3 bond relative to the C 17 -O 17 bond is different. Accordingly, we suggest that for these substrates, the interaction of His 221 with the O 3 atom would place the O 17 in a position which is unfavorable for catalysis.