The Crystal Structures of Dihydropyrimidinases Reaffirm the Close Relationship between Cyclic Amidohydrolases and Explain Their Substrate Specificity*

In eukaryotes, dihydropyrimidinase catalyzes the second step of the reductive pyrimidine degradation, the reversible hydrolytic ring opening of dihydropyrimidines. Here we describe the three-dimensional structures of dihydropyrimidinase from two eukaryotes, the yeast Saccharomyces kluyveri and the slime mold Dictyostelium discoideum, determined and refined to 2.4 and 2.05 Å, respectively. Both enzymes have a (β/α)8-barrel structural core embedding the catalytic di-zinc center, which is accompanied by a smaller β-sandwich domain. Despite loop-forming insertions in the sequence of the yeast enzyme, the overall structures and architectures of the active sites of the dihydropyrimidinases are strikingly similar to each other, as well as to those of hydantoinases, dihydroorotases, and other members of the amidohydrolase superfamily of enzymes. However, formation of the physiologically relevant tetramer shows subtle but nonetheless significant differences. The extension of one of the sheets of the β-sandwich domain across a subunit-subunit interface in yeast dihydropyrimidinase underlines its closer evolutionary relationship to hydantoinases, whereas the slime mold enzyme shows higher similarity to the noncatalytic collapsin-response mediator proteins involved in neuron development. Catalysis is expected to follow a dihydroorotase-like mechanism but in the opposite direction and with a different substrate. Complexes with dihydrouracil and N-carbamyl-β-alanine obtained for the yeast dihydropyrimidinase reveal the mode of substrate and product binding and allow conclusions about what determines substrate specificity, stereoselectivity, and the reaction direction among cyclic amidohydrolases.

which represents the second step in the three-step reductive degradation pathway of uracil, thymine, and several anti-cancer drugs (1,2). The ring cleavage reaction is reversible and is achieved by hydrolysis of the amide bond between nitrogen 3 and carbon 4 of the dihydropyrimidine ring (Scheme 1).
As an integral part of the pyrimidine catabolic pathway, DHPase activity is required for the regulation of the pyrimidine pool size available for nucleic acid synthesis and for supplying the cell with ␤-alanine. The reductive degradation of uracil represents the major pathway providing the non-proteinogenic amino acid in plants and filamentous fungi and its sole source in mammalian tissues (3). It is thought to have neurotransmitter function because of its chemical similarity to ␥-amino-n-butyric acid and glycine (4), and it is accumulated in biologically active dipeptides such as carnosin and anserine.
The medical condition of DHPase deficiency (dihydropyrimidinuria) is a rare event and has thus far been described for five patients who showed a variable clinical phenotype consisting of seizures, mental retardation, growth retardation, and dysmorphic features (5). DHPase deficiency has been associated with a high risk of 5-fluorouracil toxicity (6,7). 5-Fluorouracil is one of the most commonly used chemotherapeutic agents for the treatment of carcinoma of the colorectum, breast, and head/neck. More than 80% of the administered dose of 5-fluorouracil is rapidly catabolized by the pyrimidine-degrading enzymes (8), which thus represent major determinants of drug efficacy, pharmacokinetics, and toxicity. DHPase has also been implicated in the metabolism of dexrazoxane, an antioxidant agent used to reduce the cardiotoxic side effects of the anticancer drug doxorubicin (9).
DHPases from a range of eukaryotic species (10 -16) have been purified from natural sources or as recombinant proteins. All are homotetrameric Zn 2ϩ -metalloenzymes, with subunit molecular masses ranging from 56 kDa for DHPase from slime mold to 65 kDa for the fruit fly enzyme (16). Besides the reversible ring opening of its physiological substrates dihydrouracil and dihydrothymine, DHPase also catalyzes the hydrolysis of a variety of 5-monosubstituted hydantoins and succinimides (17), which led to the hypothesis that DHPase is identical to the enzyme hydantoinase (Hyd) and hence to the synonymous use of both names in the EC nomenclature. However, the DHPases from Saccharomyces kluyveri and Dictyostelium discoideum do not hydrolyze hydantoins (16), and not all Hyds, i.e. the bacterial counterparts of DHPase, hydrolyze dihydropyrimidines and are therefore not likely to be involved in the reductive pyrimidine catabolism (18,19). Hyds are utilized for the biotechnological production of optically pure D-amino acids (20), but their natural substrates are often unknown. It has been suggested to classify the enzymes involved in the reductive pathway of pyrimidine degradation as DHPases, whereas the name hydantoinase should be reserved for all enzymes that hydrolyze (5-monosubstituted) hydantoin and have other physiological roles (19).
DHPases exhibit 24 -43% amino acid sequence identity to Hyds, whereas pairwise sequence identities between DHPases range from 27% for evolutionary distant species (human/yeast) to 94% for close relatives (rat/mouse). Significant homology to other amidohydrolases such as dihydroorotase (DHOase), which catalyzes the mirror reaction of DHPase in the pyrimidine biosynthesis, urease, and allantoinase suggests evolutionary relationship between these enzymes (15,21,22). In addition, similarity has been detected within genes/ proteins involved in the development of the nervous system, such as collapsin response-mediator protein (CRMP) (23,24). In recent years structural information has been obtained for most of these DHPasehomologous amidohydrolases, e.g. for Hyd (25)(26)(27)(28)(29), DHOase (30), and urease (31), as well as for the noncatalytic CRMP1 (32). They share a common subunit structure with a (␤/␣) 8 -barrel core. Often it is accompanied by a ␤-sandwich domain, which can significantly vary in size and is missing completely, for example, in phosphotriesterase (33). With the exception of CRMP1, all these proteins contain two metal ions and a post-translationally modified lysine residue for their coordination in the active site.
In this study, we report the first crystal structures of eukaryotic DHPases, from the yeast S. kluyveri ( Sk DHPase) and the slime mold D. discoideum ( Dd DHPase). Although their overall fold is closely related, the assembly of the physiologically relevant tetramer differs with regard to the formation of the dimer-dimer interface. The Sk DHPase sequence contains several short insertions that are not found in other DHPases or Hyds. Complexes of this enzyme with the substrate 5,6-dihydrouracil (DHU) and the product N-carbamyl-␤-alanine (NC␤A) reveal amino acid residues that play a crucial role in substrate binding.

EXPERIMENTAL PROCEDURES
Enzymatic Assay-DHPase activity was determined spectrophotometrically by measuring the change in the absorbance at 225 nm caused by the hydrolysis of the DHU ring (15). A UV-3000 spectrophotometer (Hitachi) thermostated at 30°C was used. The molar absorption coefficient for DHU at 225 nm is 1287 M Ϫ1 cm Ϫ1 (12). For routine enzyme assays, the reaction mixture contained 0.1 M potassium phosphate, pH 8.0, and 0.25 mM DHU. One unit of DHPase catalyzes the hydrolysis of 1 mol of DHU/min at 30°C. Protein concentrations were determined by the method of Bradford (34) using bovine serum albumin as a standard.
Expression, Purification, and Crystallization-Sk DHPase was expressed, purified, and crystallized as described (15,35). The crystals resulting in data set Sk DHPase-nat were obtained by equilibration against a reservoir solution of 100 mM Bistris, pH 6.5, 20% (w/v) PEG 3350, 100 mM ammonium sulfate, and 1 M ZnCl 2 . The protein solution was prepared by diluting an Sk DHPase stock (20 mg/ml) with 50 mM Tris, pH 7.5, 100 mM NaCl to a final protein concentration of 3.5 mg/ml, and addition of 5 mM Tris(2-carboxyethyl)phosphine hydrochloride and 2.5 mM DHU.
Similar conditions were used to obtain crystals for Sk DHPase-DHU. The only differences are the absence of DHU and the exchange of the ZnCl 2 against 1 mM CaCl 2 in the reservoir solution. Prior to cryoprotection, the crystals were soaked for 3 min in 100 mM Tris, pH 8.0, 100 mM ammonium sulfate, 22% PEG 3350, 10 mM NaHCO 3 , 20 M ZnCl 2 , 50 mM NC␤A, and 2.5 mM DHU.
Sk DHPase-NC␤A was collected from crystals grown by equilibration against 100 mM Bistris, pH 6.5, 20% (w/v) PEG 3350, and 100 mM trisodium citrate. The protein solution was prepared as follows: 4 ml of a 1.5 mg/ml Sk DHPase solution were transferred into a dialysis cassette (Slide-A-Lyzer; Pierce) with a 3.5-kDa molecular mass cut-off and dialyzed overnight against 1.5 liters of 17 mM Hepes, pH 8.0, 0.4 mM ZnCl 2 , 8.3 mM KHCO 3 , and 3% (v/v) glycerol. After dialysis, the sample was concentrated to 20 mg/ml and stored at Ϫ20°C until further use. For crystallization, the stock of dialyzed Sk DHPase was diluted with 50 mM Bicine, pH 9.0, 100 mM NaCl, 10 M ZnCl 2 , and 5 mM KHCO 3 to reach a final protein concentration of 3.5 mg/ml. 5 mM Tris(2carboxyethyl)phosphine hydrochloride and 50 mM NC␤A were added. Sk DHPase crystals belong to space group P2 1 and contain one homotetramer per asymmetric unit.
Expression, purification, and crystallization of Dd DHPase have also been described previously (36). Crystallographic data to a 2.05-Å effective resolution were collected from a single crystal, which belongs to space group I222 with one molecule per asymmetric unit.
Structure Determination-Prior to data collection, Sk DHPase crystals were drawn through a drop of paraffin oil for cryoprotection and flashfrozen in a nitrogen gas stream at 100 K. For cryoprotection of the Dd DHPase crystals, 10% (w/v) of PEG 1000 were added to the mother liquor. Details for the data collection and processing are given in Table  1. The data sets were processed with MOSFLM and merged and scaled with SCALA; structure factors were derived with TRUNCATE (37,38).
For Sk DHPase, phases were obtained by molecular replacement using a D-enantiospecific hydantoinase (D-Hyd) as a search model (35). Initial rigid body refinement of the polyalanine model gave a rather high free R-factor (54.1%), but the observed packing of the four protomers corresponded to the tetramer structure of Hyds and suggested that the correct solution was found. SCHEME 1. a, the reductive degradation pathway for uracil and thymine. b, hydrolysis of 5-monosubstituted hydantoin as catalyzed by hydantoinases.
For all three Sk DHPase structures, iterations of model building using the program O (39) were alternated with TLS and restrained refinement in REFMAC5 (40). All reflections in the given resolution range (Table 1) were used with the exception of 5% randomly selected reflections for monitoring R free (41). For most residues, tight and medium NCS restraints were applied for main chain and side chain atoms, respectively. Restraints were removed for amino acids at the N and C terminus or at the borders to a flexible loop region, for the carboxylated Lys 167 , and for residues adopting alternate main chain conformations between the monomers (residues 104 -105, 141-151, and 380 -381). Waters were added using the ARP/wARP water cycling routine in REFMAC5.
The final model for Sk DHPase-nat contains residues 2-294 and 303-541 for chains A and B, 2-293 and 303-541 for chain C, and 2-294 and 305-541 for chain D, 8 zinc ions, and 681 water molecules. Occupancies for the zinc ions and the carboxyl group attached to Lys 167 were set to the limit of appearance of negative density for these atoms, lying in the range 0.6 -0.8.
For Sk DHPase-DHU, the ͉F o ͉ Ϫ ͉F c ͉ map after initial rigid body refinement using the refined structure of the ligand-free enzyme showed clear electron density for Zn-␣ and DHU and indicated low occupancies for Zn-␤ and the Lys 167 modification. Zinc ions, substrate, and water molecules were added. The final model contains the same amino acid residues as for Sk DHPasenat, two zinc ions and one DHU molecule per subunit, as well as 1043 water molecules. Lys 167 remains unmodified in chains B and C. Occupancies were set below 1.0 for all zinc ions (0.7-0.8 for Zn-␣ and 0.4 -0.5 for Zn-␤) and the carboxyl group at Lys 167 in chain D.
For SkDHPase-NC␤A, the initial ͉F o ͉ Ϫ ͉F c ͉ map revealed additional electron density in the active sites of chains A and B, indicating partial binding of NC␤A. The corresponding molecules were fitted into the density. The final model contains residues 2-294 and 303-541 for chains A and B, 2-293 and 303-541 for chain C, and 2-294 and 304 -542 for chain D, 8 zinc ions, 2 NC␤A, and 794 water molecules. Occupancies were set below 1.0 for the carboxyl group at Lys 167 and for the side chain of Tyr 172 (chains A and B).
Phases for Dd DHPase were obtained by molecular replacement using CRMP1 as the search model (36). Iterations of model building using the program WinCoot (42,43) were alternated with refinement in REF-MAC5. All reflections in the resolution range 45.5 to 2.05 Å were used, with the exception of 5% randomly selected reflections for monitoring R free . Waters were added using WinCoot. The final model contains residues 7-490, 2 zinc ions, 280 water molecules, and 1 molecule of malonate originating from the crystallization solution. The occupancies for Zn-␣, Zn-␤, and the carboxyl atoms attached to Lys 158 were set to 0.5, 0.25, and 0.5, respectively.
All models have good stereochemistry, as determined by the program PROCHECK (44). Refinement statistics are given in Table 1.

Structure Determination and Quality of the Model
To elucidate the mechanism of pyrimidine catabolism in molecular detail, and to reveal potential structural and functional differences between eukaryotic DHPases, microbial hydantoinases, and DHOases, we determined the crystal structures of DHPases from the yeast S. kluyveri ( Sk DHPase) and from the slime mold D. discoideum ( Dd DHPase). These organisms belong to two very different eukaryote branches, and comparison of their enzymes should be able to shed light on the diversity of eukaryotic DHPases.
Recombinant Sk DHPase (residues 2-542 plus 18 tag residues at the C terminus) yielded crystals belonging to space group P2 1 , with one homotetramer per asymmetric unit. The ligand-free structure was determined at 2.4 Å resolution by molecular replacement (35). Complexes of the yeast enzyme with the substrate ( Sk DHPase-DHU, 2.4 Å) and product ( Sk DHPase-NC␤A, 2.45 Å) were obtained by soaking and co-crystallization, respectively. Data collection and refinement statistics are given in Table 1. The Ramachandran plot indicates only one nonglycine residue per chain in disallowed regions, Tyr 172 , which belongs to stereochemistry gate loop 3 (SGL-3) (see below) and is located near the active site. Two distinct rotamers can be observed for Tyr 172 ; for both, the main chain atoms are well defined in the electron density. His 255 , a metal ligand, is found in the generously allowed regions. Electron density was visible for residues 2-293 and 304 -541 in all four polypeptide chains. The 10 residues of a loop not resolved by electron density are mostly hydrophilic and surface-located; the lack of density can be attributed to structural disorder.
Recombinant Dd DHPase (residues 2-503 plus 18 tag residues at the C terminus) was crystallized in space group I222 with one polypeptide chain per asymmetric unit. The structure was also solved by molecular replacement to a maximum resolution 2.05 Å (36). Crystallographic refinement yielded 20.4 and 25.3% for R cryst and R free , respectively. Additional refinement statistics are given in Table 1. Only residues 7-490 could be traced; 5 amino acids at the N terminus, the 13 C-terminal residues, and the polyhistidine tag are not visible in the electron density map because of structural flexibility. As for Sk DHPase, one histidine residue serving as a metal ligand, His 247 , was found in the generously allowed region of the Ramachandran plot. Ser 166 belonging to SGL-3 and Cys 240 located at the C-terminal end of a barrel helix also show generously allowed (⌽, ⌿) combinations.

Subunit Structure
The structures of Sk DHPase and Dd DHPase are illustrated in Fig. 1, a and b, respectively, and the secondary structure assignment is given in Fig. 2. As for most members of the amidohydrolase superfamily, the subunit of both enzymes can be divided into two domains, a core catalytic domain and a smaller ␤-sandwich domain.
For Sk DHPase, the catalytic domain includes residues 57-436, which form a TIM-like barrel of eight parallel ␤-strands (␤8 -15) flanked on the outer face by eight ␣-helices (␣1-4, 7, 8, and 10, 11). Because of high variability in the lengths of barrel strands and helices, one side of the barrel appears compressed. Furthermore, the hydrogen-bonding pattern within the barrel remains incomplete, as strands ␤8 and ␤15 are not interacting with each other via main chain atoms. Their N-terminal ends are separated from each other by the C-terminal end of helix ␣12. The observed barrel distortion is not unique to Sk DHPase but has been described previously for Hyds (25,26,28,29). The top of the barrel is decorated by long loops, six 3 10 -helices (2-7), three ␣-helices (␣5, ␣6, and ␣9), and two ␤-strands (␤16 and ␤17). Helices ␣12 and ␣13 are found at the barrel bottom, accompanied by 10 from the C-terminal tail.
The ␤-sandwich domain includes residues of both the N and the C terminus (2-56 and 437-542). The larger of the two sheets is formed by seven strands in the order ␤3-2-7-18-19-20-21. Only ␤2 and ␤7 are parallel to each other. The topology of the four-stranded sheet is ␤5-4-1-6, with ␤1 and ␤6 being parallel. Helical turns (1 and 8) are inserted between ␤5/6 and ␤18/19. Starting after the last strand, the long C-terminal tail follows a path along the bottom side of the barrel to the opposite face of the molecule, where it turns toward ␣3. It contains two additional 3 10 -helical turns (9 and 10).
There are three cis-prolines per subunit, Pro 142 , Pro 332 , and Pro 436 , of which the latter two are conserved in Dd DHPase (Pro 298 and Pro 389 , respectively) and other members of the amidohydrolase superfamily. Pro 142 is located on the surface and directly precedes helix ␣3. It belongs to one of the inserted amino acid stretches that distinguish Sk DHPase from other DHPases (see below). The cis-peptide bond between Ser 331 and Pro 332 is of functional importance because it forms part of the substrate-binding site. The other conserved cis-proline, Pro 436 , is buried in the interface between molecules A and C and B and D, respectively.
In Dd DHPase, the catalytic core domain includes residues 61-389. Again, the barrel is distorted, and most of the secondary structure elements are conserved between both DHPases (Fig. 2). However, the residue stretch corresponding to barrel strand ␤15 in Sk DHPase does not form a ␤-strand in Dd DHPase. Although the position of the corresponding amino acids is highly equivalent, they do not form the necessary interactions with the neighboring strands. Structure elements unique to Dd DHPase comprise a ␤-hairpin (␤9 and ␤10) and helix ␣12 located at the top of the barrel, and ␣16 and 5 of the C-terminal tail. Except for this tail, a few loop regions, and the prolongation of strands ␤17 and ␤18, the structure and topology of the ␤-sandwich domain of Dd DHPase (residues 2-60, 390 -503) correspond well to that of Sk DHPase.

Quarternary Structure
The asymmetric unit of Sk DHPase crystals contains a homotetramer (Fig. 3), which corresponds to the native state of the enzyme in solution (16). The four protomers are related by internal 222 symmetry. The resulting three noncrystallographic 2-fold axes are hereby oriented almost parallel to the crystallographic axes.
The structure of the four subunits is nearly identical, with the highest root mean square deviation (r.m.s.d.) of 0.3 Å observed between chains B and D. Two distinct sets of interfaces are formed within the tetramer. The larger of the two, observed between subunits A and B and subunits C and D, respectively, buries an accessible surface area of 1870 Å 2 per subunit, which is equivalent to 9.5% of the monomer surface area. Monomer-monomer interactions are mostly hydrophobic but also include 15 hydrogen bonds. The interface is dominated by coiled-coil-like interactions between helices ␣4 -5 and ␣7 from each subunit. In addition, residues from the C-terminal tail engage in van der Waals interactions and three hydrogen bonds with ␣5 and ␣6, the N terminus of ␣7, and the loops connecting these structure elements, in total accounting for almost 20% of the buried surface.
The other interface is formed between monomer A and C (B and D), by extension of the smaller of the sheets in the corresponding ␤-sand-wich domains to an eight-stranded ␤-sheet. Additional inter-subunit contacts are formed by residues from helices ␣8, ␣10, ␣13, and 10 and from adjacent loops. Again, the interface is dominated by hydrophobic interactions; it buries an area corresponding to 6% of the monomer surface (1140 Å 2 ).
The asymmetric unit of the Dd DHPase crystals contains one polypeptide chain; the physiological relevant tetramer is formed by the crystallographic 222 symmetry. The interface between subunits A and B (C and D) is well conserved when compared with the corresponding interface in Sk DHPase, and has with 1740 Å 2 (9.6% of the monomer accessible surface area) also approximately the same size. However, the C-terminal tail (including 5) contributes with ϳ30% of the accessible surface area more than in the yeast enzyme.
In contrast, the contact area between subunits A and C (B and D) is altered significantly, although some of the interactions are found in both enzymes. Among them are those involving the conserved cis-peptide bond between Tyr 388 -Pro 389 (Tyr 435 -Pro 436 in Sk DHPase), as well as the interactions of helices ␣8, ␣11, and ␣15. However, the hallmark of the smaller interface in Sk DHPase, the combination of two four-stranded ␤-sheets to an eight-stranded sheet, is not observed in Dd DHPase. When subunits A of both enzymes are superimposed, the position of the interacting strand ␤5 in subunits C differs by as much as 9.5 Å, caused by an ϳ15°rotation of the C/D pair of subunits relative to the corresponding pair in Sk DHPase. Besides the C-terminal tail, only three residues from the ␤-sandwich domain of Dd DHPase are contributing to the formation of the smaller interface, which allocates 990 Å 2 (5.4%) of the monomer surface area. There are no direct contacts between diagonally opposite subunits, i.e. the A/C and B/D pairs, in the DHPase tetramers.

The Di-metal Center
The DHPase active site is confined by the C-terminal ends of the barrel strands and harbors a dinuclear zinc center (Fig. 4a). According to the classification of structurally characterized metal centers within the amidohydrolase superfamily (51), it belongs to subtype I. The geom-FIGURE 1. Subunit structures. a, stereo view of the subunit of Sk DHPase. The following color coding is used for the secondary structure elements: green and dark green, barrel strands and helices, respectively; blue and light blue, strands of the larger and smaller sheets of the ␤-sandwich domain, respectively; yellow, additional ␤-strands; orange and dark orange, additional ␣and 3 10 -helices, respectively. Spheres in magenta represent the zinc ions of the metal center. b, stereo view of the subunit of Dd DHPase. The same color coding as in a is used. c, stereo view of the superimposed subunits of Sk DHPase (light blue), Dd DHPase (blue), CRMP1 (gold), and B9 D-Hyd (magenta). The orientation is changed for better visualization of the structural differences, especially at the C terminus, and is related to that from a and b by a 45°upward rotation around a horizontal axis lying within the paper plane.
etry of the di-zinc center and the type and position of the amino acid side chains involved in its ligation are strictly conserved between Sk DHPase and Dd DHPase. In the following, and unless otherwise mentioned, the first residue number will always refer to the yeast enzyme and the second to Dd DHPase.
The two zinc ions are positioned 3.6 -3.9 Å from each other and bridged by the hydrolytic water molecule. The close proximity of the positively charged metal ions makes it likely to be deprotonated to a hydroxyl ion. A second bridge is formed by the carbamyl group of a carboxylated lysine residue (Lys 167/158 ). Because it is possible to retain full catalytic activity by reconstitution of the DHPase apoenzymes with zinc ions (16), it is likely that the post-translational modification is acquired via reaction with CO 2 /HCO 3 Ϫ in a process of spontaneous self-assembly of the metal center similar to that described for phosphotriesterase (52). The more buried of the two ions, Zn-␣, is furthermore ligated by an aspartate (Asp 358/325 ) and by the two histidine residues (His 62/66 and His 64/68 ) from the HXH motif that is well conserved within the amidohydrolase superfamily. Two additional histidines (His 199/191 and His 255/247 ) interact with Zn-␤. All six residues participating in the ligation of the metal center to the protein originate from barrel strands or stretches directly upstream. The pentacoordination sphere of Zn-␣ has distorted trigonal bipyramidal geometry, with Lys 167/158 and Asp 358/325 as apical ligands. In contrast, the geometry of the Zn-␤ coordination is distorted tetrahedrally. It is expected to be completed to an almost square pyramidal coordination sphere upon binding of the carbonyl O-4 of the substrates to Zn-␤. The DHPase metal centers strongly resemble those found in other amidohydrolases with a subtype I metal center, such as Hyd and DHOase. In D-hydantoinases from Thermus sp. ( Ts D-Hyd) (26) and Burkholderia pickettii ( Bp D-Hyd) (29), the empty slot in the coordination sphere of Zn-␤ is occupied by a water molecule. In ligand-free Sk DHPase and Dd DHPase, no electron density for such a water molecule is visible. However, this might be caused by the observed incompleteness of the active centers and low occupancy of the Zn-␤ sites (see below), which prevents the stable binding of a water molecule.

Active Site Integrity and Saturation Experiments
As revealed by the electron density maps for more than 20 collected data sets, the Sk DHPase samples used for crystallization were not fully saturated with zinc ions and did not contain a covalent modification at all Lys 167 , leaving the majority of the active sites incomplete. Previous metal exchange experiments had already strongly suggested that at least one of the zinc ions is loosely bound (16). 3 The specific activity for Sk DHPase after purification has been reported to 7.5 units/mg (15), which increases to 23.2 units/mg when the enzyme is saturated with zinc. 4 The same behavior has been observed for Dd DHPase and calf liver DHPase, which exhibit generally higher specific activities than the yeast enzyme (13). 5 After incubation in the presence of 0.4 mM Zn 2ϩ and 8 mM bicarbonate, the specific activity of Sk DHPase samples used for crystallization is 4 -6 times higher and comparable with values observed previously. 4 This indicates that less than 20% of the subunits contained both zinc ions in the active site, corresponding well with the occupancies estimated from the observed electron densities. These usually lie between 0.6 and 1.0 for the deeper buried Zn-␣, whereas Zn-␤ occupancies vary significantly between data sets. On average, Zn-␤ and thus also the bridging water molecule are absent in about 75% of the active sites, thereby showing a clear correlation between occupancy and the position of the corresponding polypeptide chain within the asymmetric unit. Similarly, Lys 167 has a lower tendency to be carboxylated in chain D and often completely lacks the modification in chains B and C. It seems that Sk DHPase tetramers with partially occupied active sites are preferentially packed during the crystallization process, although it is unclear exactly how it is achieved because no significant structural differences could be observed between the four crystallographically independent monomers.
Although it is not the data set with the best obtained resolution, we used Sk DHPase-nat, collected from crystals grown in presence of 1 M ZnCl 2 and 2.5 mM DHU, to describe the ligand-free Sk DHPase structure. It shows the highest Zn-␤ occupancy and provides the most complete picture of the active sites of the enzymes. The bridging hydroxyl is present in all four chains, and Lys 167 modification and Zn-␤ are observed in an estimated 70% of the monomers. All metal-ligand distances refine to values usually observed for di-zinc centers, with the exception of the Asp 358 -Zn-␣ distance in subunits A and C, which is ϳ2.5 Å and therefore modestly out of range. It should be noted that no electron density for DHU was observed in the active sites. At the pH of the crystallization conditions, the backward reaction (of the anabolic type) is favored, and the used DHU concentration is most likely not sufficient to form a stable complex.
To obtain Sk DHPase-DHU, a crystal was soaked in a solution of pH 8.0 containing ZnCl 2 at a higher concentration (20 M). Nevertheless, average occupancies for Zn-␣ and Zn-␤ are only estimated to 0.7 and 0.4, respectively. The active sites in subunits B and C are least complete because of the missing carboxylation of the corresponding Lys 167 . It is likely that the observed incompleteness of the active sites allowed us to find DHU bound at the pH optimal for catalytic activity. Only in chain C has a small peak of negative density for the scissile C-N been observed, which might be interpreted as a sign for occurring catalysis.
Addition of more than 10 M ZnCl 2 and 2 mM bicarbonate interfered with the ability of the enzymes to crystallize under otherwise unchanged conditions. Thus, we attempted to saturate the active sites of Sk DHPase prior to crystallization by dialysis of the enzyme against a solution containing 0.4 mM ZnCl 2 and 8.5 mM sodium bicarbonate. The dialyzed samples failed to crystallize under the original conditions, most likely because of the altered protein buffer composition. However, very similar conditions, with the salt component changed from ammonium sulfate to trisodium citrate, did yield crystals. Subsequently, the data set Sk DHPase-NC␤A was collected from crystals of dialyzed Sk DHPase grown in the presence of NC␤A. Here refinement with full occupancies for Zn-␣, Zn-␤, and Lys 167 modification does not lead to the appearance of peaks of negative electron density. However, their significantly higher B-values as compared with those observed for the protein atoms indicate that the performed dialysis failed to fully saturate the enzyme.
With the determination of the crystal structure of Dd DHPase, it became apparent that looser Zn-␤ binding might be a general feature of DHPases. A rough estimation of the occupancies for Zn-␣, Zn-␤, and the carboxyl group attached to Lys 158 for the one polypeptide chain present in the asymmetric unit gives values of 0.5, 0.25, and 0.5, respectively. D-Hydantoinase from the Bacillus sp. AR9 ( B9 D-Hyd) (25) is the only other member of the amidohydrolase superfamily for which lacking carboxylation of the active site-lysine has been reported. The metal center of this enzyme does not contain the usually observed zinc ions but instead two manganese ions that are positioned equivalently.

Substrate Binding
Soaking of Sk DHPase crystals prior to data collection in a mother liquor-like solution of altered pH, which contained both the substrate DHU and the product NC␤A, led to the appearance of additional electron density in the active sites of all four protomers. A close-up view of the active site of subunit A is shown in Fig. 4b. The electron density can clearly be attributed to a bound DHU molecule. Only two amino acid residues are directly involved in substrate binding. The backbone carbonyl of Asn 392 forms a hydrogen bond with the N-1 of the dihydropyrimidine ring. Enabled by the cis-peptide bond with the following proline, Ser 331 forms two hydrogen bonds with the substrate; its amide nitrogen interacts with O-2, and the backbone oxygen interacts with N-3 of DHU. The remaining ring oxygen of the substrate points toward the binuclear metal center and is located about 2.0 Å from Zn-␤, and 2.4 -2.8 Å from Zn-␣, which is close to the position of the bridging activated water as observed in Sk DHPase-nat. Thus not surprisingly, refinement of a model containing both DHU and bridging water molecule leads to the appearance of peaks of negative density for either one and displacement of the substrate from the associated density. We assume that the observed binding mode is not entirely representative for the actual substrate binding prior to catalysis, because it does not allow the concomitant presence of the hydroxyl moiety performing the nucleophilic attack. However, a slight tilt of the DHU molecule brings it into position, with its O-4 occupying the empty slot in the Zn-␤ coordination sphere and its C-4 aligned for attack by the hydrolytic water. This tilt does not require changes in the distances to primary substrate binding residues. By taking into account that catalysis should proceed under the conditions present in the crystal, it is possible that only substrate molecules trapped in an unfavorable binding mode due to the partial incompleteness of the active sites can be observed.
Hydrophobic residues line the wall of the active site facing the ring carbons C-5 and C-6. The side chain of Cys 360 is the closest, which is placed within van der Waals distance (3.6 -3.8 Å) to N-1 and C-6 of DHU. Leu 72 faces both carbons at a distance of about 4.2 Å, and its neighbor Leu 71 is another 4 Å further away. The hydrophobic wall is completed by Phe 169 , Leu 175 , and Ile 135 at a 4.3-, 6.9-, and 7.7-Å distance to substrate atoms, respectively. The pocket created by these residues is more than sufficient to harbor the methyl group of the thymine substrate and appears to be large enough to accommodate even slightly bulkier hydrophobic substituents at C-5. Based on the observed orientation of the bound DHU, it can be predicted that 5-monosubstituted dihydropyrimidines with L-configuration are preferred, because C-5substituents in D-configuration would cause steric clashes with active site residues if they are larger than a methyl group. There are no solvent molecules within hydrogen-bonding distance from the substrate. Nevertheless, the active site is freely solvent-accessible, as evident from the surface representation of its entrance (Fig. 5). A narrow channel leads from the protein surface straight down to the di-zinc center. All loop segments defining the entrance to the active site are clearly resolved in electron density. Furthermore, they are positioned identically in the enzyme complexes with substrate and product and in the ligand-free enzyme, which indicates that control of active site accessibility via conformational changes of loop segments is not required in Sk DHPase. The only active site residue with observed conformational flexibility is Tyr 172 . Its side chain is oriented toward the bulk solvent, i.e. away from the active site in chains A, B, and D of ligand-free Sk DHPase. Ill-defined electron density indicates high mobility for this side chain conformation. In chain C as well as in all active sites of Sk DHPase-DHU and Sk DHPase-NC␤A, the majority of the Tyr 172 side chains points toward the di-zinc center, placing the hydroxyl group within 3.2 Å distance from atoms N-3, C-4, and O-4 of the substrate.

Product Binding
Co-crystallization of recombined Sk DHPase with 50 mM NC␤A at pH 6.5 led to the appearance of additional electron density in the active sites of molecules A and B, in which NC␤A molecules could be fitted (Fig.  4c). As expected, the DHPase reaction product is bound in a manner very similar to the substrate molecule, and again no solvent molecules are involved in ligand binding. The lengths of the hydrogen bonds between the NC␤A atoms corresponding to N-1, O-2, and N-3 of DHU and the primary substrate binding residues Asn 392 and Ser 331 , respectively, are unaltered within the error limit of the observed resolution. A modest movement of the carbamyl moiety of the product molecule brings the O-2 atom into hydrogen-bonding distance to the backbone amide of Gly 393 ; however, the geometric requirements for the formation of such a bond are not fulfilled. However, we do observe weak hydrogen bonding interaction (Ͼ3 Å) of N-3 with the hydroxyl group of Tyr 172 . C-5 and C-6 remain at similar distance to the surrounding hydrophobic residues. The bridging solvent molecule of the bi-nuclear metal center is replaced by one of the oxygen atoms of the NC␤A carboxyl group. The distances to Zn-␣ and Zn-␤ refine to 2.8 and 2.4 Å, respectively. The other carboxyl oxygen is placed ϳ3.1 Å from Zn-␣ and not involved in any interactions with protein residues.
The relatively weak electron density and high B-values for the ligand indicate that only 50 -80% of the active sites of chains A and B have NC␤A bound. For chains C and D, the estimated ligand occupancy lies below 0.3, and therefore the product molecule was not modeled into the corresponding active sites.

The Active Site of Dd DHPase
Comparison of the Dd DHPase active site with that of the yeast enzyme reveals that the vast majority of the amino acid side chains within a radius of 8 Å from the substrate molecule is identical (Fig. 4d). The substitution of Ser 331 in Sk DHPase by a glycine (Gly 297 ) in Dd DHPase should not affect substrate binding, because only main chain atoms of this residue interact with the ligand. All other exchanges concern residues located in more than van der Waals distance from the ligand. Among them are most residues forming the hydrophobic wall, such as Ile 135 and Leu 175 , which are in Dd DHPase exchanged to alanine (Ala 133 ) and phenylalanine (Phe 167 ), respectively. An altered loop conformation positions the amino acids corresponding to Leu 71 and Leu 72 of Sk DHPase differently in Dd DHPase, but the resulting gap in the hydrophobic lining of the active site is closed by two other residues, Leu 71 and Ile 102 , respectively.

Reaction Mechanism
Based on the high level of active site conservation to DHOase and the pH dependence of kinetic parameters and kinetic solvent deuterium isotope effects (13), there is general agreement in the literature that enzymatic catalysis in Hyds and DHPases follows a mechanism very similar to that proposed for DHOase (30,53). According to this mechanism (Scheme 2), a water molecule in the coordination sphere of the more solvent exposed Zn-␤ is replaced by the 4-oxo group of the incoming substrate. This interaction polarizes the carbonyl group and prepares the carbon of the scissile bond for hydrolysis. Asp 358 /Asp 325 has a dual function as it ligates Zn-␣ and also acts as a general base required for activation/deprotonation of the zinc-bridging hydroxyl ion. As the latter performs a nucleophilic attack at C-4, the oxygen O-4 is expected to change its position, leading to the formation of a geminal diol in the tetrahedral transition state. We modeled such a geminal diol into the active sites of Sk DHPase-DHU (not shown), assuming that the hydrogen bonds to Ser 331 and Asn 392 are maintained during catalysis and that the nascent hydroxyl group will occupy approximately the same location as the bridging hydroxyl moiety. This places the other hydroxyl group closer to Tyr 172 (2.6 -2.8Å) than the original carbonyl oxygen O-4 of DHU had been. Thus, stabilization of the tetrahedral transition state over the substrate is most likely achieved via bidentate ligation to both zinc ions, and via the formation of a hydrogen bond between one of the oxygens of the nascent carboxyl group and Tyr 172/163 . Such a function in transitions state stabilization has also been suggested recently for the corresponding tyrosine residue of another amidohydrolase, Escherichia coli isoaspartyl dipeptidase, whose exchange against phenylalanine led to a 3 orders of magnitude reduced rate of catalysis (54,55) and caused a complete loss of catalytic activity for Bs D-Hyd (56). Concomitant protonation of the neighboring amide by Asp 358/325 assists the cleavage of the carbon-nitrogen bond, leading to collapse of the transition state, product formation, and eventually its release.

Comparison to other DHPases and Hyds
Overall Structure-A sequence alignment of DHPases and structurally characterized Hyds reveals that Sk DHPase contains several insertions that are not found in the other sequences (Fig. 2). A biological function of the inserted stretches is not clearly evident, as all are located on the periphery of the molecule far from the active site. Insertion 1 (residues 141-147) forms the N terminus of helix ␣3 and contributes to the formation of the monomer-monomer interfaces. The second and longest insertion (residues 295-320) is largely unstructured, as the first 8 to 9 residues are not resolved in the electron density map and the only secondary structure element formed is the helical turn 3. The third insertion (residues 365-370) is separated from the fourth insertion (residues 378 -384) by seven residues that structurally align between Sk DHPase and Dd DHPase. The latter insertion forms ␤16, ␤17, and 16, whereas the third and also the last inserted stretch (residues 459 -466) are devoid of any secondary structure elements.
Compared with Hyds of known structure, the C termini of all known DHPase sequences are extended. In fact, the extension is shortest in Sk DHPase (18 amino acids); Dd DHPase has a 38-residue extension, and the enzyme from Drosophila melanogaster contains as much as 110 additional residues. Although the structures of the C-terminal tails of Sk DHPase and Dd DHPase do not correspond to each other, they both pack against the periphery of the barrel domain and are deeply engaged in inter-subunit contacts. It can be expected that this feature is conserved in other eukaryotic DHPases, although it is possible that the C terminus of the fruit fly enzyme forms an additional domain. Truncation experiments performed on Hyds from Bacillus thermocatelunatus GH2 revealed that deletion of 11 residues at the C terminus did not significantly affect the catalytic activity of the enzyme but caused a change from tetrameric to dimeric state (57). Thus, it cannot be excluded that also in some Hyds the C-terminal tail is involved in subunit interactions.
Most DHPase and Hyd elements of secondary structure are conserved, and a majority of them can be structurally aligned (Fig. 1c and Table 2), emphasizing the immediate evolutionary and functional relationship between both enzymes. Surprisingly, both Sk DHPase and Dd DHPase show higher similarity to D-enantioselective Hyds from two Bacillus sp. ( B9 D-Hyd and Bs D-Hyd) than to each other. For Dd DHPase, the noncatalytic CRMP1 from mouse is the closest structural relative, with an r.m.s.d. of 0.84 Å for 468 matching C-␣ atoms (Fig. 1c). It is also noteworthy that the Hyd from Arthrobacter aurescens ( Aa L-Hyd), the only structurally characterized Hyd with an L-substrate preference, is the least similar of all hydantoinases.
Comparison of the quaternary assemblies shows that Sk DHPase and Hyd tetramers superimpose well (Fig. 3). Dd DHPase can clearly be distinguished as it is the only enzyme of the group of close relatives that does not form an extended ␤-sheet across a dimer interface, a feature it shares with CRMP1 (Fig. 3). In both DHPases the involvement of the extended C terminus in subunit interactions leads to a clear increase in oligomerization interface area as compared with D-Hyd and L-Hyd (Table 3). Again, the L-enantioselective Hyd deviates from the group as it deploys a much smaller interface area.
Slime mold is thought to be more closely related to man than yeast, and yeast are more closely related to man and slime mold than are bacteria. Based on the phylogenetic analysis of amidohydrolase sequences (16,58), it had been suggested that Dd DHPase, animal DHPases, and DHPase-related proteins such as CRMPs originate from the same progenitor. Surprisingly, bacterial Hyds, except Aa L-Hyd, are more closely related to this group than the yeast enzyme (16). This relationship is supported by the structural data. Therefore, from the SCHEME 2. Reaction mechanism. Similar to the DHOase-catalyzed reaction (31,61), hydrolysis of dihydrouracil is proposed to proceed via the following steps: a and b, binding of the substrate DHU; b and c, nucleophilic attack of its C-4 atom by the zinc-bridging hydroxyl and formation of the tetrahedral intermediate; c and d, breakage of the N-3-C-4 bond and formation of NC␤A; d and e, product release and binding of a new solvent molecule; and finally e-a, its activation to a hydroxyl moiety. The given residue numbering corresponds to Sk DHPase. The complete zinc coordination is shown only in a. phylogenetic and structural perspectives, the divergence of L-Hyd from D-Hyd/DHPases/DHPase-related proteins is the earliest event.
The yeast enzyme appears to have diverged from the other studied eukaryotic DHPases before DHPase and D-Hyd branches separated, which contradicts the phylogenetic relationship between bacteria, yeast, and other eukaryotes. It seems more likely that the Sk DHPase gene underwent a retrograde evolution similarly as described for fruit fly deoxyribonucleoside kinases (59). Whereas D-Hyds seem to have remained closest in structure to the common ancestor protein, both DHPases underwent more pronounced although distinct structural changes, with the extension of the C terminus and increase of oligomerization interface area as the only commonalities. These structural changes seem to be an advantage and perhaps a necessity for the pyrimidine-degrading enzymes. Subsequently, CRMPs must have originated rather recently and developed their new, noncatalytic function after duplication of the DHPase gene in the higher animal lineage, explaining the high sequence identity and structural conservation to Dd DHPase. It has been suggested that the two CRMPs identified for Caenorhabditis elegans (CeCRMP/ DHP-1 and -2) are representative for the common ancestor of both protein lineages, as they show higher sequence similarity to DHPs than to vertebrate CRMPs, have retained all zinc-binding residues, and exhibit low amidohydrolase activity (60). Stereochemistry Gate Loops and Substrate Specificity-Hyds can differ significantly in their substrate and stereo specificities (61) and are commonly classified as D, L, and nonselective (62). This classification became ambiguous with the identification of Aa L-Hyd, which shows a substrate-dependent enantioselectivity (63). DHPases have been described as D-selective for the hydrolysis of 5-monosubstituted hydantoins (19).
The emergence of structural models for several Hyds enabled comparative studies with the aim to reveal the structural basis for their distinct substrate specificities and stereoselectivities. Cheon et al. (27) identified three hydrophobic loop regions involved in the binding of the exocyclic ring substituents in D-Hyds, whose amino acid compositions define the particular substrate preferences of the enzymes. The conformations of these so-called "stereochemistry gate loops" (SGLs) are well conserved between Bs D-Hyd, Bp D-Hyd, Ts D-Hyd, and B9 D-Hyd, however, amino acid exchanges are frequent although usually retaining the hydrophobic character of the ligands of the exocyclic substrate side chains (Scheme 3). Fig. 6 shows a superposition of the SGLs for Hyds and both DHPases. Not surprisingly, SGL-3 shows the least structural deviation. It harbors the well conserved tyrosine residue (Tyr 172 /Tyr 163 ), which is predicted to be crucial for transition state stabilization and to engage in hydrophobic interactions with the substrate ring. Its exchange to phenylalanine or glutamate in Bs D-Hyd produced inactive mutant proteins (56), whereas the affinity of the enzymes toward substrates with a larger aromatic exocyclic substituent could be gradually increased by mutating Phe 159 (corresponding to Leu 175 /Phe 167 ) four residues upstream to amino acids with hydrophobic side chains of decreasing size (64). In Sk DHPase, the shortening of SGL-3 by one amino acid, as compared with D-Hyds, does not have significant structural consequences. In contrast, the insertion of two residues and less curving brings the corresponding loop in Aa L-Hyd closer to the ligand-binding site, constricting its dimensions on this side of the active site cavity. The structural change of SGL-3 in this enzyme is accompanied by a movement of SGL-1 and SGL-2 away from the di-zinc center. As a consequence, the corresponding part of the active site pocket is enlarged, eventually leading to a completely altered shape of the cavity available for ligand binding, and thus to a substrate spectrum very different from that of D-Hyds and DHPases (63). It is also noteworthy that the transition state-stabilizing tyrosine is not conserved in Aa L-Hyd and that the SGL-3 backbone occupies most of the space available for its side chain. There is, however, another hydroxyl group provided near this position by Ser 153 . When the DHU-and Zn-␣-ligating residues of Sk DHPase are superimposed with the corresponding residues of Aa L-Hyd, the position of the hydroxyl groups of Tyr 172 and Ser 153 differ by about 3 Å. Based on the almost identical positioning of all other catalytic and ligand-binding residues, we can assume that the 5-or 6-membered ring of the Aa L-Hyd substrate itself must be bound in a way similar to DHU in Sk DHPase. Thus, to be involved in transition state stabilization, Aa L-Hyd-Ser 153 has to move closer to the ligand, which could be achieved by a conformational change of SGL-3 upon substrate binding.  The structure of SGL-1 differs between both DHPases. In Dd DHPase the loop basically follows the same path as observed for D-Hyds. In the yeast enzyme an altered loop structure brings leucines 71 and 72 closer to the bound substrate, restricting the space available for potential exocyclic ring substituents at carbons 5 and 6 (Figs. 5 and 6). Hence, the previous finding of a very narrow substrate spectrum for Sk DHPase (16) is supported by the structural data.
SGL-2 is the least structurally conserved stereochemistry gate loop and shows deviations even among the D-Hyds. Although it is involved in binding of the carboxyl group of dihydroorotate in DHOase (30), it does not contribute any ligands for the substrate in the Sk DHPase-DHU complex. Nevertheless, because of its closer placement toward the di-zinc center in Dd DHPase, Bp D-Hyd, and Ts D-Hyd, it cannot be excluded that SGL-2 might have a greater influence on substrate specificity and stereoselectivity in those enzymes than it appears to have in Sk DHPase.
In search for potential determinants for preferential hydrolysis of dihydropyrimidine versus hydantoin, we superimposed the di-zinc centers and ligating residues of all structurally characterized Hyds, Dd DHPase and Sk DHPase-DHU, and modeled the transition state binding for DHU and hydantoin. The superposition results in a remarkably good agreement in the placement of the primary substratebinding residues, with Aa L-Hyd showing the largest but still modest deviations. For both DHPases, keeping the distances of the geminal hydroxyl groups of the transitions state models to the zinc ions constant and the hydrogen bond between O-2 and Ser 331 /Gly 297 at the same length results in an ϳ0.2 Å longer hydrogen bond between Asn 392/346 and the N-3 of the hydantoin analog in comparison to the same inter-action for the DHU analog. The resulting inefficient binding of hydantoin might explain why DHPases prefer dihydropyrimidines. A similar analysis for Hyds is hampered by the incompleteness of the available biochemical data regarding their substrate specificities and a lack of complex structures. For Bs D-Hyd, Bp D-Hyd, and Ts D-Hyd, no information could be found regarding their ability to hydrolyze dihydropyrimidines. B9 D-Hyd hydrolyzes DHU and dihydrothymine with a 4 and 16 times lower rate than hydantoin (65), respectively, whereas these two compounds are not substrates for Aa L-Hyd (28). The superposition of the active sites reveals no obvious trend in placement of the substratebinding residues from DHPases to Hyds. However, the differences required for an efficient discrimination between the 6-membered ring of DHU and the 5-membered ring of hydantoin are expected to be small, and their identification is complicated by the resolution limits of the crystal structures.
In conclusion, the level of correspondence in active site architecture and composition between D-Hyds and DHPases emphasizes once more their immediate evolutionary relationship. Although not identical enzymes, they are highly homologous in structure as well as function.
Dihydropyrimidine Synthesis Versus Degradation-In the cell, several reversible reactions are catalyzed by an anabolic and a catabolic enzyme, which originate from a progenitor enzyme upon gene duplication. In general, each of these two enzymes preferentially catalyzes the reaction in one direction. In addition, the active site of each duplicated enzyme has a configuration that would preferentially accept only one of the substrates. Therefore, the direction and/or substrate specificity of both parameters represent a control mechanism to efficiently separate the FIGURE 6. Determinants of substrate specificity and enantioselectivity. a, stereochemistry gate loops in DHPases and Hyds. The superimposed backbones of the SGLs as found in Sk DHPase-DHU (yellow), Dd DHPase (orange), B9 D-Hyd (blue), Bs D-Hyd (dark blue), Bp D-Hyd (cyan), Ts D-Hyd (light blue), and Aa L-Hyd (green) are shown. The location of the active site is indicated by the ball-and-stick representation of the two zinc ions (black spheres), dihydrouracil, substrate-binding residues, as well as Tyr 172 (all with carbon atoms in yellow) for Sk DH-Pase-DHU. b, stereoview of the superimposed active sites of Sk DHPase-DHU and DHOase with bound DHO. For the yeast enzyme, SGLs 1-3 and selected amino acid side chains originating from them are depicted in magenta, blue, and green, respectively, although DHU-binding residues and Tyr 362 are shown with yellow carbon atoms. For DHOase, the corresponding loops and amino acid side chains are shown in light gray. SGL-3 adopts a different conformation when the DHOase substrate carbamyl-L-aspartate is bound. For comparison, it is depicted here in a darker gray. Both ligands are represented as stick models, DHU with yellow and DHO with gray carbon atoms. Hydrogen bonding and electrostatic interactions to surrounding amino acids as well as coordinative bonds to the di-zinc center are indicated by dotted lines. All residues and loops are labeled.
two reactions in the cell. Enzymes involved in metabolism of nucleic acid precursors, such as DHPase and DHOase, represent an interesting model to study the structural specificities, which have arisen during the evolution of duplicated enzymes. Here, interference of the anabolic with the catabolic pathway is prevented by the employment of distinct substrate/ product pairs. DHOase catalyzes the reversible cyclization of carbamyl-Laspartate to L-dihydroorotate (DHO), a dihydropyrimidine with an exocyclic carboxyl group at carbon 6, which is removed later on in the anabolic pathway.
DHOase shows a different oligomerization state than DHPase. In higher organisms, it is found within a large polyfunctional protein (CAD), whereas it is a homodimeric and monofunctional enzyme in bacteria and lower eukaryotes. Nevertheless, a superposition of the active sites of both DHPases and DHOase from E. coli (30) aligns the zinc-ligating residues of the enzymes almost perfectly. It also reveals the determinants for their distinct substrate specificities (Fig. 6b). The dihydropyrimidine ring is bound in a very similar way via hydrogen bonding interactions of atoms N-1, O-2, and N-3 with backbone atoms. The exocyclic carboxyl group is anchored by three DHOase residues, Arg 20 , Asn 44 , and His 254 . In contrast, the active sites of both DHPases do not support binding of DHO. Although the replacement of His 254 by a tyrosine in the yeast enzyme would still allow the formation of a hydrogen bond to this compound, it is excluded by the substitution with a phenylalanine in Dd DHPase. Furthermore, there are no additional anchors equivalent to Arg 20 and Asn 44 . The latter residue belongs to SGL-2, which is positioned further away from the dihydropyrimidine ligand in the DHPases. Because of the deviating conformation of SGL-1, the aspartate and glutamine that replace Arg 20 in Sk DHPase and Dd DHPase, respectively, point away from the active site. Furthermore, the immediate proximity of the hydrophobic SGL-1 residues Leu 71 /Pro 78 and Leu 72 /Phe 79 to the charged carboxyl group hinders sterically as well as electrostatically DHO binding to the DHPases.
For DHOase, an "open" and a "closed" conformation of the SGL-3 has been observed depending on whether the reaction product DHO or the substrate carbamyl-L-aspartate are bound in the active site (Fig. 6b). The structure of the SGL-3 in Sk DHPase (and in Dd DHPase) is more similar to the closed conformation, and as mentioned before, no indications have been found that loop conformational changes occur upon ligand binding/release in the catabolic enzyme. It is also noteworthy that in DHOase the SGL-3 does not contain a tyrosine at the position equivalent to that of Tyr 172 in Sk DHPase but rather three residues downstream (Tyr 104 ). Nevertheless, a simple rotation of its side chain from the observed conformation in the crystal structure should enable Tyr 104 to fulfill a similar function in transition state stabilization as proposed for Sk DHPase-Tyr 172 .
In conclusion, the efficient separation of two very similar reactions steps in the de novo pyrimidine biosynthesis and the corresponding catabolic pathway was mostly achieved upon a limited number of amino acid exchanges in three loop regions of the DHPase/DHOase/Hyd progenitor. Interestingly, the same three loop regions determine substrate specificity and enantioselectivity within the subgroup of DHPase/Hyd enzymes.