Porcine Carbonyl Reductase

Porcine testicular carbonyl reductase (PTCR) belongs to the short chain dehydrogenases/reductases (SDR) superfamily and catalyzes the NADPH-dependent reduction of ketones on steroids and prostaglandins. The enzyme shares nearly 85% sequence identity with the NADPH-dependent human 15-hydroxyprostaglandin dehydrogenase/carbonyl reductase. The tertiary structure of the enzyme at 2.3 Å reveals a fold characteristic of the SDR superfamily that uses a Tyr-Lys-Ser triad as catalytic residues, but exhibits neither the functional homotetramer nor the homodimer that distinguish all SDRs. It is the first known monomeric structure in the SDR superfamily. In PTCR, which is also active as a monomer, a 41-residue insertion immediately before the catalytic Tyr describes an all-helix subdomain that packs against interfacial helices, eliminating the four-helix bundle interface conserved in the superfamily. An additional anti-parallel strand in the PTCR structure also blocks the other strand-mediated interface. These novel structural features provide the basis for the scaffolding of one catalytic site within a single molecule of the enzyme.

PTCR is highly homologous to NADP(H)-dependent human and rat CRs, except for its 13 additional amino acid residues at the C terminus (2). The sequence identity of PTCR with the human CR, also known as 15-hydroxyprostaglandin dehydrogenase/9-ketoprostaglandin reductase, is about 85% (2). Alignment of amino acid sequences of PTCR and human CR with those of other well known members of SDR superfamily, such as bacterial 3␣,20␤-HSD (7,8), human 17␤-HSD type 1 (9), human 11␤-HSD (10), and Drosophila alcohol dehydrogenase (11), suggests that these CRs have a 41-residue insertion at a strategic location, before the conserved Tyr-X-X-X-Lys motif. Nearly all SDRs are known to utilize a Tyr-Lys-Ser triad as catalytic residues (1). The Tyr hydroxyl has been proposed to be the proton donor in an electrophilic attack on the substrate carbonyl in a reduction reaction (8).
The other distinguishing feature of all SDRs is that the functional units are either homotetramers or homodimers (8,9,(12)(13)(14)(15). Surprisingly, the active units of both PTCR and human CR were reported to be monomeric (5,16). Although more than half of the 15 or so crystal structures of SDRs to date exhibit the same tetrameric quaternary structure, only the four-helix bundle or the Q axis interface that involves the largest surface area of association (8) is conserved among SDRs. The other surface of close association of monomers in tetrameric SDRs is the so-called P axis interface involving an anti-parallel pair of strands and a pair of helices (8) that has been observed only once in a dimeric SDR (17). The interfacial four-helix bundle that borders two catalytic sites and buries hydrophobic surfaces of helices is predicted to be important for the integrity of the active site clefts (18,19). Here we present the first crystal structure of a monomeric SDR and explain how the functional determinants of oligomeric SDRs could be retained within a monomer of PTCR having the basic SDR fold.

EXPERIMENTAL PROCEDURES
Purification, Crystallization, and Data Collection-The enzyme was purified as described (5). Crystallization and data collection details have also been published (20). Briefly, the enzyme was crystallized from a solution of 36 -37% saturated ammonium sulfate in 10 mM MES buffer, pH 6.0. Crystals of diffraction size were grown by macroseeding. The space group is P4 3 2 1 2 and the unit cell parameters are a ϭ b ϭ * This work was supported in part by Grant DK26546 from the National Institutes of Health. The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked "advertisement" in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.
The 58.53 Å and c ϭ 165.64 Å with one molecule (288 amino acid residues) in the asymmetric unit. Data collection was carried out on a Rigaku R axis II image plate detector at ambient temperature and processed with the software package DENZO (21). Data from three crystals between resolution range 55.2 and 2.30 Å were merged to yield a data set of 10792 unique reflections (F Ͼ 0) with an average of about three observations per reflection. The overall R merge was 0.066 on the intensities. The data were 80% complete overall and 52% complete in the resolution shell between 2.50 and 2.30 Å.
Structure Solution-The structure was solved by the single isomorphous replacement and anomalous scattering method, combined with phases from a molecular replacement solution. Two platinum compounds, potassium chloroplatinate (PtCl4) and di--iodobis(ethylenediamine)diplatinum(II) nitrate (PIP), were used to prepare heavy atom derivative crystals, both yielding the same major platinum-binding site. A total of 4351 reflections to 3.00 Å was phased using the PIP derivative, and 3564 reflections to 3.20 Å were phased using the PtCl4 derivative. The latter data set also contained anomalous signal for 1660 reflections to 3.40 Å. The centric R value and the phasing power for the PIP and PtCl4 derivatives were 0.56 and 1.64 and 0.63 and 1.41, respectively. The phasing power for the anomalous data was 1.92. The software package PHASES (22) was used for calculating and combining phases.
The molecular replacement solution was obtained using a search model, which was a superposition of equally weighted structures of 3␣,20␤-HSD (8), 17␤-HSD1 (9), mouse lung carbonyl reductase (13), and 7␣-HSD (14). A large number of rotation search solutions were computed using compact (without loop regions) and extended (all loops through the catalytic triad present) search models. The optimal solution was obtained with the CNS program suite (23) and the extended search model with data between 15.00 and 3.50 Å, against an E 2 E 2 synthesis. The translation function search was conducted in both P4 3 2 1 2 and P4 1 2 1 2 space groups. Phases from the most likely solutions were computed and used to phase a difference Fourier synthesis with the PIP derivative and native data to 3.00 Å. The correct translation function solution in the space group P4 3 2 1 2 produced the largest peak at the platinum-binding site, having a height about three times the background peaks. The heavy atom hand and the choice of origin were fixed by this peak position. Phases calculated from the model were combined with single isomorphous replacement and anomalous scattering phases using the PHASES package. In all, 4698 reflections were phased to 3.00 Å, with a combined figure of merit of 0.809, a correlation coefficient of 0.792, and an R inverse value of 0.266.
Model Building and Refinement-Model building was carried out on a Silicon Graphics R10000 Indigo 2 work station using a figure of merit weighted F o map, phased with the combined experimental phases, and the software CHAIN (24). Initially, 235 of 288 amino acids were fit in the electron density map, which also unequivocally showed the NADP density. The model was subjected to 26 cycles of refinement and rebuilding processes, using XPLOR (25) and CNS (23) routines. The resolution was gradually extended to 2.30 Å. The C-terminal tail section was built during the latter part of the refinement. Various omit maps with calculated phases were computed to resolve regions of ambiguity.
Except for a few residues in highly exposed regions of the molecule, such as Glu-261, Lys-219, and Lys-238, side chains were in good agreement with their electron densities. A sulfate ion and 58 well ordered water molecules were introduced in the last three cycles of refinement and rebuilding.
A summary of refinement results is provided in Table I. The analysis of the molecular geometry was performed with PROCHECK (26), CNS (23), and CCP4 (27) packages. The illustrations were made with SETOR (28). The atomic coordinates have been deposited with the Protein Data Bank (code 1HU4).

RESULTS AND DISCUSSION
The Tertiary Structure of PTCR, an Overview of SDRs-All 288 amino acid residues of one polypeptide chain of PTCR (excluding the N-terminal methionine) were located from electron density maps. Amino acid residues are numbered 1-288 starting with the Ser at position 2 in the cDNA sequence (2). A ribbon diagram of the tertiary structure of PTCR is shown in Fig. 1a. A schematic description of the structure is provided in Fig. 1b. The basic SDR fold includes a seven-stranded parallel ␤-sheet (␤A to ␤G) flanked by three parallel helices on each side (␣B, ␣C, ␣G and ␣D, ␣E, ␣F). The segment ␤A to ␤F is a doubly wound ␣/␤ motif, with alternating ␤-strands and ␣-helices.
Whereas the ␤A to ␤F segment constitutes the classic "Rossmann fold" associated with the binding of the coenzyme NADPH, the ␤D to ␤G segment, in addition to being partially within the Rossmann fold, governs quaternary association and substrate binding. The two longest helices ␣E and ␣F form the interfacial four-helix bundle, typical of dimeric and tetrameric forms.
The view in Fig. 1a is along the central ␤-sheet of the PTCR molecule, overlooking the active site delineated by the coenzyme NADP and catalytic residues Tyr-193, Lys-197, and Ser-139. As shown in Fig. 1, a and b, the basic SDR fold described above is well preserved in PTCR, except for two important differences. First, the 41-residue insertion before Tyr-193 describes four helices ␣FЈ-1 (residues 140 -148), ␣FЈ-2 (residues 151-158), ␣FЈ-3 (residues 164 -179), and ␣FЈ-4 (residues 182-186). The first turn of ␣FЈ-1 coincides with the helical turn immediately following the catalytic serine (Ser-139 in PTCR) present in nearly all SDRs. This all-helix subdomain folds away from the active site cleft, toward the dimer interface of other SDRs. Second, after a tight turn the polypeptide following ␤G describes an additional strand ␤H (residues 272-275) that is anti-parallel to ␤G. The last 13 C-terminal residues form an extended tail-like structure that approaches the active site of a 2-fold symmetry-related molecule. The terminal carboxyl group forms a salt bridge with the Arg-37 side chain of the second molecule. A similar eighth anti-parallel strand is present only in one other member of the SDR family, dihydropteridine reductase (12). Fig. 1c is a superposition of monomers of crystal structures of all four steroid dehydrogenases belonging to the SDR superfamily (bacterial 3␣,20␤-HSD, human 17␤-HSD1, Escherichia coli 7␣-HSD (14), and PTCR). The viewing direction in this figure is roughly the same as in Fig. 1a. 3␣,20␤-HSD and 7␣-HSD have homotetrameric structures, whereas 17␤-HSD1 is homodimeric (Fig. 2). The structural diversity among monomers of all members of the SDR family is well represented by this group. The overlap of the Rossmann fold and the catalytic triad among these four enzymes, excluding the 41-residue insertion in PTCR, is striking. Furthermore, two long helices ␣E and ␣F from 3␣,20␤-HSD, 7␣-HSD, and 17␤-HSD1 that make up the four-helix bundle dimer interface superimpose remarkably well with the ones from PTCR, which are not involved in dimer formation.
The other important observation from this superposition is the diversity of the three-dimensional structures of the regions primarily contributed by the C-terminal halves of polypeptide chains. This region, which constitutes the outer walls of the substrate-binding cleft of the active site, provides specific interactions that are critical to the selectivity of substrates and to the mechanism of molecular recognition by the enzyme (19,29). Although this region is well developed for 17␤-HSD1 having additional structural elements from an insertion between ␤F and ␣G, and from an extended C terminus that surrounds the catalytic cleft, fewer residues from similar regions of 3␣,20␤-HSD and 7␣-HSD are present in the area. In 3␣,20␤-HSD, a segment of the polypeptide chain from the loop between ␤F and ␣G (Met-184 to Pro-206) and C-terminal residues Trp-243 to Gln-255 border the active site cleft (8). Similarly, in 7␣-HSD, residues Ala-191 to Pro-214 between ␤F and ␣G and C-terminal residues Gly-250 to Asn-255 line the active site (14). Thus, the outer wall structures in 3␣,20␤-HSD and 7␣-HSD differ significantly from that of 17␤-HSD1. These differences in the architecture of active sites probably explain why 17␤-HSD1 has high specificity for estrone as a substrate, whereas the other two enzymes are not known to be substrate-specific.
Nonetheless, all three enzymes possess the so-called substrate entry-loop between ␤F and ␣G, which tightens on the substrate upon its entry into the active site providing additional substrate-specific interactions (8,14,19,29). The entrance to the active site in PTCR is more open than any other HSD. This is a direct consequence of a shortened three-residue linkage between ␤F and ␣G, connecting Gly-236 and Pro-240 (corresponding to residues Thr-190 and Val 210, respectively, in 3␣,20␤-HSD). Consequently, the substrate-entry loop is almost non-existent in PTCR. Moreover, only residues Glu-281 to Ala-288 from the C-terminal tail are in the vicinity of the active site. The number of residues that border the substrate-binding cleft is thus only 15 in PTCR, as opposed to 37 in 3␣,20␤-HSD, 30 in 7␣-HSD, and 60 in 17␤-HSD1. The openness of the substrate-binding cleft is consistent with the observed promiscuity in substrate specificity of PTCR.
The Lys-238 side chain in human placental CR is autocatalytically modified to carboxyethyllysine (30). This side chain in PTCR is in a highly exposed region, and its electron density is not discernible beyond the C␦ atom owing probably to dynamic disorders.
The Quaternary Structure of SDRs and How a PTCR Monomer Mimics It-The Q axis dimer interface involving the fourhelix bundle and the P axis interface involving anti-parallel association of ␤G strands and ␣G helices in the prototypical tetrameric 3␣,20␤-HSD are shown in Fig. 2, a and b. The four-helix bundle, conserved in dimeric SDRs such as 17␤-HSD1 shown in Fig. 2c, conceals most of the hydrophobic surfaces that are at dimer interfaces (8). Efforts to disrupt the Q axis interface by mutations resulted in inactive and insoluble monomers (18). Furthermore, the four-helix bundle is considered critical to function, since disruption of the helical bundle interferes with the integrity of the two active sites it borders, as shown in Fig. 2, a and c. The P axis interface does not involve significant hydrophobic interactions (8) and is less critical to the structural integrity of the active site.
The all-helix subdomain of the 41-amino acid insertion region obstructs access to ␣E and ␣F. Helices ␣FЈ-1 and ␣FЈ-2 and the Trp-276 side chain from the C-terminal end of the new strand ␤H surround ␣F such that four of its seven turns are shielded from solvent exposure. The four and a half turn helix ␣FЈ-3, the longest in the subdomain, partially shields ␣F and packs over five of the seven turns of ␣E, as shown in Fig. 2d. Together, ␣FЈ-2, ␣FЈ-3, and ␣FЈ-4 occupy the position of sym-metry-related helices ␣E and ␣F in 3␣,20␤-HSD, 17␤-HSD1, and 7␣-HSD, blocking the interface and preventing dimer formation. However, as shown in a close-up view of the area in Fig. 2d, this arrangement does not completely shield the interface leaving ␣E and ␣F partially exposed. Selective substitutions of residues on these helices in PTCR render the exposed surface more hydrophilic than in oligomeric SDRs, as shown in Fig. 2d (substitutions from 3␣,20␤-HSD to PTCR are as follows: Phe-116 to Arg-118, Thr-121 to Glu-123, Val-168 to Arg-209, Gly-171 to Arg-212, and Thr-172 to Glu-213). All four helices of the insertion domain are amphipathic; their hydrophobic surfaces face the outer hydrophobic surfaces of ␣E and ␣F, and their hydrophilic surfaces are exposed to the solvent, as shown for ␣FЈ-3 in Fig. 2d. Interestingly, the inner hydrophobic surface of ␣FЈ-3 contains Met-171 which packs against Met-110, in close proximity to the Met-115 side chain, both from ␣E. The S␦ atom of Met-171 is at a distance of 3.76 Å from S␦ of Met-110, which is 6.04 Å from S␦ of Met-115. This partially exposed "methionine trap" is responsible for the high affinity binding of platinum ions (mono-and di-platinate) to the crystalline protein that resulted in the only useful isomorphous derivatives for solving the phase problem (see under "Experimental Procedures").
Why Is PTCR NADPH-specific?-Although no coenzyme was added in the crystallization medium, a molecule of NADP(H) was bound to the protein molecule. The coenzyme is in the extended conformation as observed for all other NAD(P)(H) bound to SDRs (8,(12)(13)(14)(15)29). The distance between C-2 of the nicotinamide and C-6 of the adenine ring, a parameter used to compare the degree of extension of coenzymes, is 14.65 Å, well within the range for SDRs. The nicotinamide ring is in syn conformation, with the 4-pro-S hydride facing the catalytic site, typical of all SDRs. The coenzyme molecule and its electron density are shown in Fig. 3a, along with side chains in the vicinity. Puckering of the nicotinamide ring was not discernible at the working resolution, and we used parameters of an oxidized coenzyme to restrain its geometry during refinement. Hydrogen bonding and charge interactions between NADPH and neighboring amino acid residues are drawn schematically in Fig. 3b.
The high specificity of the enzyme for reduced ␤-NADPH has been established. At pH 7, PTCR has the highest affinity for ␤-NADPH as evidenced by the lowest K m of 7.2 M and the highest V max /K m ratio of 16.2 nmol min Ϫ1 mg Ϫ1 M Ϫ1 of all coenzymes (6). In the crystal structure, we find that the negatively charged 2Ј-phosphate moiety is surrounded by three positively charged side chains (Arg-37, Arg-41, and Lys-14), one hydrogen bond forming polar side chain (Asn-13), and a water molecule. All of these make specific contacts with the 2Ј-phosphate group, either through hydrogen bond formation or electrostatic charge neutralization, as shown in Fig. 3. The Asn-13 side chain carbonyl also accepts a proton from the 3Ј-hydroxyl of the adenine ribose. In addition to the presence of charged Arg and Lys residues near the 2Ј-phosphate of NADPH noted before (13,29), specific interactions of the Asn-13 side chain with phosphate and ribose groups may play an important role in coenzyme selectivity. The location of Asp-62 near the amide group of the adenine ring is analogous to the Asp-60 in 3␣,20␤-HSD (8) and the Asp-68 in 7␣-HSD (14). The carboxamide group of the nicotinamide ring interacts with the main chain atoms of Val-230, whereas the Thr-232 side chain makes a hydrogen bond with the bridging pyrophosphate. Interestingly, a sulfhydryl group from Met-234 is also present in the general vicinity, as has been observed in other SDR structures (Met-189 in 3␣,20␤-HSD and Met-193 in 17␤-HSD1). Two water molecules are hydrogen-bonded to the phosphates.
Catalytic side chains Tyr-193 and Lys-197 and the main chain carbonyl group of Asn-89 have contacts with the 2Ј-and 3Ј-hydroxyls of the nicotinamide ribose moiety, as shown in Fig. 3. Not shown in the figure is a short contact (3.1 Å) between the main chain carbonyl oxygen of Gly-228 of PTCR, FIG. 3. a, a stereo view of the bound NADP molecule, within its final (2F o Ϫ F c ) electron density map. The map is contoured at 1. Side chains in the surrounding neighborhood, including the ones that have direct contact with the coenzyme, are shown. The color codes for atoms are the same as in Fig. 1. Three-letter codes are used to identify amino acids. W denotes a water molecule. All the direct contacts with NADP are shown schematically in b, and distances are given in Å. Residues that are in the immediate surroundings but do not have any direct contact are also shown.
conserved in all steroid dehydrogenases and many other SDRs, and C-4 of the nicotinamide ring. This interaction and the orientation of carbonyl with respect to the nicotinamide ring have been proposed to have an important role in driving the hydride transfer from the coenzyme to the substrate (19).
Crystal Packing-The most interesting intermolecular interaction among crystallographically related monomers of PTCR is the formation of a crystallographic dimer by the packing of two loops between ␤D and ␣E (Ile-92 to Pro-101) about a 2-fold rotation axis. Because of the openness of the active site cleft, the loops penetrate each other's active site and the Asp-97 side chain from one loop makes two strong hydrogen bonds to the catalytic residues Tyr-193 and Ser-139 (2.57 and 2.58 Å, respectively) of the other molecule, as shown in Fig. 4a. Interaction of the hydroxyl group of the catalytic Tyr with a carboxyl acid group was previously observed in the structure of the carbenoxolone complex of 3␣,20␤-HSD (31), in which the hemisuccinate side chain of the inhibitor formed a strong hydrogen bond with the Tyr-152 hydroxyl. The main chain amide of Asp-97 makes a hydrogen bond (2.96 Å) with Glu-141 O⑀2 from a symmetry-related molecule. The 2-fold rotation axis passes through the center of two Gln-95 side chains, whose O⑀1 atoms are 3.48 Å apart and make hydrogen bonds (2.84 Å) with the main chain amides of Leu-96s of symmetry-related molecules. The Leu-96 side chain stacks against the Ala-93 and Tyr-193 side chains, and Asn-98 side chain stacks against the Trp-229 side chain.
The conformation of this loop is considerably different in PTCR than in other SDRs (Fig. 4b). The similarity of the conformations of the loops in 3␣,20␤-HSD, 17␤-HSD1, and 7␣-HSD is characterized by the presence of a distorted 3 10 helical turn (residues 95-99 in 3␣,20␤-HSD, 98 -102 in 17␤-HSD1, and 103-106 in 7␣-HSD), which is absent in PTCR. The loop adopts an extended conformation in PTCR. The distances between C␣-99 and the equivalent C␣-95 in 3␣,20␤-HSD, C␣-98 in 17␤-HSD1, and C␣-103 in 7␣-HSD are 12.02, 9.33, and 8.59 Å, respectively. It is not clear whether this difference in conformation of the loop results from the interaction of the Asp-97 side chain with catalytic residues or if this altered conformation leads to dimerization.
The other interesting aspect of this packing interaction is the formation of a salt bridge between the free C-terminal of one molecule and the Arg-37 side chain of the other, as shown in Fig. 4c. This is the same Arg side chain that is involved in a specific interaction with the 2Ј-phosphate group of the coenzyme. Furthermore, the extended loop described above (residues 92-101) approaches the tail region of the symmetry-related monomer, producing the following intermolecular stacking interactions among side chains: Ile-92 with Ala-288, Phe-94 with Val-286, and Ala-288, Pro-99 with Pro-284, and Pro-101 with Tyr-283.
Thus the two most extended and isolated regions of the PTCR molecule are stabilized in the crystal by intermolecular interactions with the same regions of a symmetry-related molecule. Such contacts between two molecules can only be present in the crystal, as the functional unit of PTCR was shown to be a monomer in solution from the gel filtration data (5).
Concluding Remarks-Previously determined structures of oligomeric SDRs demonstrated how multiple copies of polypeptides could be assembled to build functional enzymes with multiple independent active sites. When allosterism or co-operativity is not required for function, multimeric assembly may not be an efficient utilization of biological resources. The PTCR structure demonstrates how alteration of the protein by substitution and insertion can alleviate the problem of interdependence among protein products, leading to efficient scaffolding of one fully functional active site within an independent molecule. It is possible that monomeric SDRs like PTCR have appeared later in the evolutionary pathway than their oligomeric counterparts. FIG. 4. a, interactions between two PTCR monomers related by a crystallographic 2-fold rotation axis. The Asp-97 side chain of one monomer forms hydrogen bonds with catalytic Tyr-193 and Ser-139 side chains of the other (distances are 2.57 and 2.58 Å, respectively). NADP and other side chains have been removed for clarity. The backbones of two loops between ␤D and ␣E are shown. The view is down the 2-fold axis. b, a close-up of the superposition of a section of this loop for four SDRs, illustrating the difference in its conformation between PTCR and other SDRs. The backbone atoms of PTCR are color-coded, whereas those of 3␣,20␤-HSD, 17␤-HSD1, and 7␣-HSD are shown in green, blue, and magenta, respectively. c, interaction of the free carboxyl end of one monomer with the Arg-37 side chain of the other. The Arg-37 side chain also have specific contacts with 2Ј-phosphate of the bound NADP molecule.