Crystal Structure of 1-Aminocyclopropane-1-carboxylate Deaminase from Hansenula saturnus *

The pyridoxal 5′-phosphate (PLP)-dependent enzyme 1-aminocyclopropane-1-carboxylate deaminase (ACCD) catalyzes a reaction that involves a ring opening of cyclopropanoid amino acid, yielding α-ketobutyrate and ammonia. Unlike other PLP-dependent enzymes, this enzyme has no α-hydrogen atom in the substrate. Thus, a unique mechanism for the bond cleavage is expected. The crystal structure of ACCD fromHansenula saturnus has been determined at 2.0 Å resolution by the multiple wavelength anomalous diffraction method using mercury atoms as anomalous scatterers. The model was built on the electron density map, which was obtained by the density averaging of multiple crystal forms. The final model was refined to an R-factor of 22.5% and an R free-factor of 26.8%. The ACCD folds into two domains, each of which has an open twisted α/β structure similar to the β-subunit of tryptophan synthase. However, in ACCD, unlike in other members of the β family of PLP-dependent enzymes, PLP is buried deep in the molecule. The structure provides the first view of the catalytic center of the cyclopropane ring opening.

biosynthesis starts with the S-adenosylation of methionine in order to give S-adenosylmethionine. This step is followed by the closing a cyclopropane ring to form ACC, which is then oxidatively cleaved to give ethylene (3). The introduction of ACCD in higher plants by gene technology reduces the production level of ethylene and delays the ripening progression of fruits (4,5). Thus, this enzyme provides a way to regulate ethylene biosynthesis and plant ripening.
PLP-dependent enzymes catalyze many important reactions that act upon amino acids, including transamination, decarboxylation, ␤,␥-replacement/elimination, and racemization (6,7). In all of these reactions (except in the case of the glycogen phosphorylase family), the two basic chemical properties of the PLP are conserved; it forms an external aldimine between its aldehyde group and the ␣-amino group of the substrates and withdraws electrons from the substrate by serving as an electron sink (7). As a PLP-dependent enzyme, the ACCD's ringopening reaction starts with a transformation reaction from an internal aldimine between the PLP and the enzyme to an external aldimine. In most of the PLP-dependent enzymes, the next step is the nucleophilic abstraction of the ␣-substituent, either an ␣-proton or a carboxylate group, to form an ␣-carbanionic intermediate. This reaction mechanism cannot be applied to ACCD because the substrate (ACC) does not contain ␣-hydrogen and the carboxyl group is retained in the product. Therefore, the ring-opening reaction of ACC must be initiated without obvious accessibility to an ␣-carbanionic intermediate, which is, for PLP-dependent enzymes, the common entry for catalysis. One proposed reaction mechanism is the nucleophilic addition to C ␥ followed by the cleavage of the C ␣ -C ␥ bond and ␤-proton abstraction. Because PLP acts as an electron sink, external aldimine is fairly electrophilic, and the nucleophilic addition to C ␥ to rupture the cyclopropane ring of ACC is mechanistically feasible (8,9).
To understand such a reaction mechanism on a molecular basis, a knowledge of the three-dimensional structure of the enzyme is necessary. The ACC deaminases from the bacterium Pseudomonas sp. (bACCD; EC 4.1.99.4) and the yeast Hansenula saturnus (yACCD; EC 4.1.99.2) have been well characterized (1,10). These two kinds of ACCD perform the same activities and have amino acid sequences that are 60% identical. Although the bACCD has an estimated molecular weight that corresponds to the trimeric form of a single polypeptide chain of 338 amino acid residues, the yACCD has a molecular weight that corresponds to the dimeric form of a single polypeptide chain of 341 amino acid residues with a molecular mass of 37,500 Da (11). Both proteins have a tightly bound PLP as a cofactor per monomer. We have also purified both proteins from an overexpressing clone of Escherichia coli and crystallized them in earlier studies (12,13). The crystals of yACCD are the more appropriate for diffraction study. Here we present the structure of the yACCD, which was determined by the multiple wavelength anomalous diffraction (MAD) method using mercury atoms as anomalous scatterers.

EXPERIMENTAL PROCEDURES
Crystallization and Data Collection-The yACCD was purified and crystallized as described previously (11,13). Mercury derivatives of yACCD were obtained by co-crystallization with 0.5 mM p-hydroxymercuribenzene sulfonate (PHMBS) using the hanging drop vapor diffusion method. Two crystal forms appeared under crystallization conditions that were identical except for their protein concentrations; the orthorhombic form appeared with a protein concentration of 10 mg/ml, and the trigonal form appeared with a protein concentration of 20 mg/ml. The orthorhombic and the trigonal form diffract 2.8 and 2.5 Å resolutions, respectively.
Since only one useful mercury derivative of yACCD was found, and because non-isomorphism between the native and derivative crystals is considerable, we decided to use the MAD technique rather than the single isomorphous replacement method with anomalous scattering. The native data set of 2.0 Å and the MAD data set of 2.8 Å of the orthorhombic form were collected on the MAD beamline BM14 using a MAR imaging plate at the European Synchrotron Radiation Facility, Grenoble, France. Based on the fluorescence spectrum of the mercury atom, three different wavelengths (1.0063 Å (peak), 1.0086 Å (edge), and 0.9183 Å (remote)) were chosen for MAD phasing.
For the mercury derivative of the trigonal form, MAD data were collected at the BL18B station of Photo Factory, Tsukuba, Japan, with a Weissenberg camera (14) using two different wavelengths (1.0090 Å (edge) and 0.9800 Å (peak)), which were chosen at the L-III edge. All data sets were collected from frozen crystals at 100 K and integrated by DENZO (15). Scaling was calculated by SCALEPACK (15) for the orthorhombic form and by the CCP4 program package (16) for the trigonal form. The data collection and processing statistics are shown in Table I.
Structure Determination and Refinement-Initial phasing was carried out independently for each form of mercury derivative by MAD analysis. Because neither mercury derivative was isomorphous with the native crystals, the "remote data" and "peak data" were each treated as "native" in the orthorhombic and trigonal forms, respectively. Three of four independent mercury atoms were located in the orthorhombic form by the SHELXS (17), and two independent mercury atoms were located in the trigonal form by the RSPS (18) program. The maximumlikelihood program SHARP (19) was used for the heavy atom refinement and phasing. The initial electron density map was obtained after phase improvement by the program SOLOMON (20) implemented in SHARP (19). The operators of non-crystallographic symmetry in both forms were obtained by the program RAVE (21). The transformation matrix between the two forms was calculated by the AmoRe (22) program based on real-space molecular replacement using the two different dimer masks, which were built by the graphics program O (23).
After electron density averaging of multiple crystal forms using the program RAVE (21) combined with SOLOMON (20) and the CCP4 program package (16), the atomic model was built using the baton and lego in the O program (23). Excluding two C-terminal residues, 339 out of 341 residues and the PLP cofactor were identified for each monomer on the modified electron density map. After one cycle of dynamics refinement using MAD data, the model was transferred to the native crystal by rigid body refinement at 3.0 Å, and the sigmaa (24) map was calculated as the basic map for fitting model.
Crystallographic refinement was carried out with molecular dynamics using the program CNS (25). 10% of the reflection data were set aside for the calculation of the free R-factor. Several cycles of simulated annealing at 3000 K, including minimization and B-factor refinement, were performed. In each cycle, manual model modification was carried out referring to -weighted (2F obs Ϫ F cal ) and (F obs Ϫ F cal ) maps using the program O (23). The initial structure and topology files of PLP were obtained from the Uppsala web site, HIC-Up (26). The refinement was initiated with non-crystallographic symmetry restraints. However, after several cycles of the refinement, the four monomers in the asymmetric unit showed a significant difference in the conformation at N and C termini, as well as loops between helices and ␤-sheets. Thus, noncrystallographic symmetry restraints were not used throughout the refinement. After 824 water molecules and a sulfoxide ion were assigned, R-factor and R free -factor were 22.5% and 26.8% for 101,968 reflections (F obs Ͼ 3.4, 90.3% completeness) at resolutions from 15.0 to 2.0 Å. The phasing and the refinement statistics are given in Tables II  and III.

RESULTS AND DISCUSSION
Structure Description-Native yACCD was crystallized in the orthorhombic form of the space group C222 1 with unit cell dimensions of a ϭ 65.7 Å, b ϭ 268.5 Å, and c ϭ 187.2 Å. Attempts to co-crystallize it with mercury derivatives (PHMBS) have led to two crystal forms; the first form is nearly isomorphous with the native crystals (cell dimensions are a ϭ 65.4 Å, b ϭ 268.3 Å, and c ϭ 186.6 Å), and the second belongs to trigonal space group P3 2 21, with cell dimensions of a ϭ b ϭ 79.4 Å, c ϭ 243.6 Å, and ␥ ϭ 120°. The asymmetric unit of C222 1 crystals contains two dimers with a solvent content of 59% (V M ϭ 2.98), while P3 2 21 crystals contain one dimer of yACCD in the asymmetric unit with a solvent content of 62% (V M ϭ 3.21). The structure of yACCD was determined by the MAD method using mercury atoms as described under "Experimental Procedures." The structure was refined against the diffraction data of native C222 1 crystals.
The yACCD monomer consists of two domains of different sizes. We refer to them as the small domain (residues 58 -169) and the PLP-binding domain (residues 1-57 and 170 -341) ( where ϽIϾ h is the mean intensity of symmetry-equivalent reflections. Friedel pairs were merged as individual data. four surrounding helices (helices 3-6) (Fig. 2). The PLP-binding domain contains seven ␣-helices (helices 1, 2, and 7-11), one 3 10 -helix 12, and six ␤-strands (A, B, and G-J). The core of this domain is an open twisted ␣/␤ structure with a central ␤-sheet surrounded by four helices (helices 7-9 and 11), similar to the small domain. The central ␤-sheet is basically parallel, but one terminal short strand, strand A, is antiparallel. These strands are strongly twisted so as to make the sixth strand nearly perpendicular to the first. There is a crevice at the carboxyl ends of the ␤-strands G and J, which makes a space for bound PLP. Except for ␣-helices 2 and 10, all ␣-helices of the two domains have a chain direction toward the molecular surface, and all ␤-strands are directed toward the molecular center. The 3 10 -helix 12 at the C terminus consists of 11 residues with a kink in the middle. The 3 10 -helix is energetically unfavorable, and only short pieces are found in protein structures. The kinked 3 10 -helix of yACCD is at the interface region between two domains. The two domains are connected by two linkers (residues 56 -58 and 162-173). Between the two do-mains, there is a large internal gap of 5.8 Å ϫ 6.8 Å ϫ 12 Å that includes the crevice mentioned above. This large internal gap provides a space for the active site (Fig. 2).
Two monomers of yACCD interact closely with each other at their nearly flat surfaces, forming a dimer with a dyad axis running through the interacting surfaces (Figs. 4 and 5B). The interface between the two monomers is formed through hydrogen bonds and hydrophobic interactions. The buried accessible surface area between the two monomers is 14.0% of the total molecular accessible surface (calculated by the CNS program (Ref. 25)). The monomer-monomer contacts involve 18 pairs of hydrogen bonds by 17 residues (Arg 23 -Ala 89 , Arg 38    monomer interface, a hydrophobic core is formed by five hydrophobic residues from both monomers: Phe 47 , Leu 121 , Leu 290 , Pro 327 , and Ala 326 . Most of the residues involved in the contacts are at helices 1, 4, and 12 of the N and C termini. The tightly bound dimer may contribute to the stability of the yACCD. Crystal Packing-Two crystallographically independent dimers (AB and CD) of yACCD have different packing interactions in the crystal; the CD dimer has more extensive interactions with five symmetry-related CD dimers and two symmetry-related AB dimers, while the AB dimer has contact with four symmetry-related AB dimers and two symmetry-related CD dimers. Because of this obvious difference in the environment in the crystal, the features of the electron densities in the two types of dimers were significantly different; the electron density in the AB dimer was poor in parts. Therefore, the number of water molecules assigned through the refinement were also different: 685 for the CD dimer and 254 for the AB dimer. Main-chain temperature factors of the two crystallographically independent dimers were also significantly different (Table III). The AB and CD dimers were also related by the pseudo-dyad axis. However, this axis was not perpendicular to the individual dimer axis, and the interactions between the two dimers were weak.
Comparisons with Other PLP-dependent Enzymes-Threedimensional structure comparisons were carried out using the program DARI (28) with entries from the Protein Data Bank (29) in order to search for all of the related structures. Among all of the PLP-dependent enzymes that have been published in the Protein Data Bank, the ␤-subunit of tryptophan synthase from Salmonella typhimurium (TRPS␤) (30), O-acetylserine sulfhydrylase from S. typhimurium (OASS) (31), and threonine deaminase from E. coli (TD) (32) have some similarity to yACCD (Fig. 5B). All of these molecules belong to the ␤ family of the PLP-dependent enzymes (6). It has been previously suggested that the PLP-binding regions in these enzymes share the same fold, which consist of two domains (33). Some of the earliest descriptions of the structural attributes of yACCD based on a sequence comparison with TRPS␤ (16% identities in sequences) also suggested that the fold of yACCD would bare some similarity to that of TRPS␤ (11). On the other hand, the sequence comparisons of yACCD with TD and OASS detected (if any) very limited similarity (Fig. 3). The TRPS␤ was the first member of the ␤ family of the PLP-dependent enzymes to have its three-dimensional structure determined by x-ray crystallography (30). Tryptophan synthase exists as an ␣␤␤␣ type tetramer with a dyad axis between the two ␤ subunits. TD is the allosteric enzyme that exists as a homotetramer with 222 symmetry. Each monomer of TD contains one catalytic domain (the PLP-binding domain) and one regulatory domain. The catalytic domain of TD is structurally similar to yACCD. OASS is a dimeric molecule with a dyad axis at the center.
Although the chemistries involved in these enzymatic reactions and quaternary association are different (two of the four members are tetramers), the overall folding topologies of these molecules are clearly related each other (Fig. 5A). Also, the dimer interface is more or less conserved in the four enzymes (Fig. 5B). A careful comparison of the tertiary structures and folding topologies of these molecules reveals several important regions in the enzymes. The largest difference in the folding topology of ACCD and other members in the ␤ family of the PLP-dependent enzymes is at helix 9. In three other members (TRPS␤, TD, and OASS), helix 9 of ACCD is replaced by a region consisting of a long loop and helices (Fig. 5A). This inserted region makes close contact with the re face of PLP and may be important for positioning the PLP at an appropriate place for adapting to each substrate. This inserted region is actually one of the most diverted regions in these three enzymes. Although ACCD does not contain this region, the PLP is nonetheless exposed to the molecular surface due to the presence of a loop between strand I and helix 10 (residues 262-273) and two extra loops mentioned in the following section. Topologically, the folding of ACCD is the simplest and probably represents a more ancient fold in this family of proteins.
Coenzyme Binding Site-As a typical open twisted ␣/␤ structure, the PLP cofactor is positioned at a crevice between the ␤-strands (G and J) of the PLP-binding domain and lies on the large internal gap between the two domains (Fig. 2). The PLP cofactor is covalently bound through a Schiff base linkage to NZ of Lys 51 with C4Ј (the internal aldimine) (Fig. 6). This mode of binding is consistently observed in the PLP-dependent enzymes. However, the environments surrounding PLP are fairly different from those of other PLP-dependent enzymes. The re side of the internal aldimine of the ␤ family of PLP-dependent proteins, whose tertiary structures have been determined, faces the active-site entrance and is open. At the front side of the PLP pyridine of the yACCD, however, the three extra loops bury the PLP deep in the interior of the molecule (Fig. 7). These are the loops between strand D and helix 5 (residues 101-116), and between strand E and helix 6 (residues 132 and 141) in the small domain, and the loop between strand I and helix 10 (residues 262-273) in the PLP-binding domain.
Behind these loops, two cavities with different sizes were formed, as calculated by VOIDOO (34) of the Uppsala program package and the MS 2 program (Fig. 7). The small (inner diameter, 3.2 ϫ 7.5 Å) one is located along the direction of the substrate tunnel of the tryptophan synthase. The residues Arg 240 , Gln 237 , and Gln 167 are positioned at the entrance. However, this cavity does not pass through to the active center without a movement of the peptide chain, since Ala Tyr 295 ) and hydrophobic residues (Val 103 , Ile 138 , and Val 294 ). These aromatic and hydrophobic residues favor the formation of hydrophobic environments for the enzymatic reaction and of a tight channel for substrate/product transportation.
A more detailed inspection of the structure around the PLP revealed further differences between this enzyme and other members of the ␤ family of the PLP-dependent proteins. In three other members of the ␤ family of the PLP-dependent proteins, the residues immediately preceding PLP-bound Lys are aromatic or bulky hydrophobic, such as His 86 of TRPS␤ (30), Phe 61 of TD (32), and Val 40 of OASS (31). These residues may function to prevent a backward tilt of the coenzyme pyridine ring (31). In the case of ACCD, the Asn 50 immediately preceding Lys 51 is involved in the hydrogen-bonding network mediated by water molecules (Fig. 8). This residue and Leu 323 are close to the PLP and seem to play a role in supporting the pyridinium plane from the backside. The front side (re face) is also different from other members of the ␤ family of PLP-dependent proteins. The phenol group of Tyr 295 is stacked to the plane of the pyridinium ring with a distance of 3.9 Å, and an angle of about 20° (Fig. 6). Although this is the first observation of such stacking in the ␤ family of the PLP-dependent proteins, the stacking of two rings has been found in other types of PLP-dependent enzymes (36 -38). This finding may represent convergent evolutions, since the sequence comparison suggested that there is no evolutionary relationship between the ␤ family and other types of PLP-dependent enzymes (6). As discussed in a previous study, the stacking structure favors the formation of the quinonoid intermediate (39).
The N1 nitrogen (pyridinium N) of the cofactor pyridine ring is within hydrogen bonding distance of the OE1 of Glu 296 . This The sequence of yACCD has 17% identity to TRPS␤ but has very low (if any) identity to TD or OASS. However, the sequences of the latter two have more similarity to TRPS␤. The amino acid residues are shaded as follows: completely identical (red), conserved change (light blue), first three are identical (purple), last three are identical (light brown). The star represents the Lys residue to which PLP is bound. The secondary structure elements indicated are those defined by the present work for the yACCD, using the program DSSP (27). position is occupied by Ser in three other members of the ␤ family of PLP-dependent enzymes whose tertiary structures have been determined. Since Glu 296 is negatively charged, it stabilizes the positively charged PLP, thereby increasing the electron-withdrawing properties of the cofactor. OE1 of Glu 296 also forms strong hydrogen bonds to the main-chain nitrogen atom of Ile 323 and to a water molecule (Fig. 8). The O3Ј of the cofactor is in close contact with the side chain of Asn 79 , which also forms strong hydrogen bonds to ND of Asn 118 and to NH of Arg 82 . Moreover, OD of Asn 50 is involved in a hydrogen-bond- ing network to O3Ј of the cofactor via a bridged water molecule. These hydrogen bonds would impede any rotation of these residues and allow charge dissipation over the hydrogen network.
Despite these major differences in the cofactor binding, these enzymes have a common recognition mechanism as well. The phosphate group of PLP was tightly fixed in a pocket formed by the residues Lys 54 , Val 201 , Thr 202 , Gly 203 , and Thr 205 with their hydrogen bonding networks around the loop region between ␤-strand G and helix 8 (Fig. 8). Val 201 and Gly 203 made hydro-  (30 -32). The re face of the cofactor is stacked to the plane of phenol group of Tyr 295 with a distance of 3.9 Å and an angle of about 20°. The bound sulfoxide ion is also drawn in blue wire, and water molecules are drawn as blue spheres. There was a water molecule located between Op2 and Cys 199 / Thr 206 within hydrogen-bonding distance. The three hydrogen bonds of Op3 were formed by N and OG of Thr 202 and a water molecule. There were three hydrogen bonds between Op1 and NZ of Lys 54 , and between N and OG of Thr 205 . These environments of the phosphate group-binding pocket are conserved in three other PLP-dependent enzymes of the ␤ family (TRPS␤, TD, and OASS).
Reaction Mechanism-An earlier experiment showed that ACC deaminase opens the cyclopropane ring at a bond between ␣ carbon and pro-S methylene carbon (Fig. 1) (40). Furthermore, ACC deaminase showed reactivity toward D-amino acids such as D-alanine and D-serine (but not L-isomers) (8), which suggested the presence of a base at the active site for the removal of a proton from the ␣ carbon of the D-amino acid. Previous chemical and genetic studies have also implicated residues that contribute to catalysis. The enzymatic activity of ACCD is inhibited by modifying Cys 162 of bACCD (which corresponds to Cys 165 of yACCD) with sulfhydryl-modifying reagents (41). However, further experiments have shown that substitution of this residue to Ala does not affect the enzymatic activity, suggesting that Cys 165 is not directly involved in the enzymatic activity (42). On the other hand, the replacement of Lys 51 with Ala at the PLP-binding site caused a loss of detectable ACC deamination activity (42). The present structure analysis is consistent with these earlier observations. Cys 162 is positioned at the internal gap between the two domains. The chemical modification by the bulky reagents may have altered the relative orientation of the two domains, thereby causing the loss of enzymatic activity.
Although the crystals of yACCD were grown without any substrates or inhibitors, the present analysis has shown an important clue to the enzymatic reaction mechanism, namely that the electron density map (with phases calculated by MAD and density modification) or the difference Fourier map revealed a significant peak with a trigonal bipyramid shape that could be interpreted to be a sulfoxide or phosphorous ion (Fig.  9). This is the largest peak in the difference Fourier map. The position of the peak is close to the PLP and within hydrogenbinding distance of the OG of Ser 78 , OH of Tyr 295 , N of Gln 80 , and N of Asn 79 . It is thus very likely that the substrate pocket is occupied by the sulfoxide or phosphorous ion in the crystallization buffer. A model of the substrate ACC was built into the peak position of the electron density map using the graphical software O (23). In order to form an external aldimine between the ACC and PLP, the PLP should be rotated by about 20°w here the pyridinium ring of PLP is parallel to the phenol group of Tyr 295 (Fig. 9). The resulting complex structure suggests that Tyr 295 at the re side of the PLP may play a role in binding the substrates by the hydrogen bond.
The reaction mechanism of the ACCD was studied previously by NMR method using modified substrates including cyclic and acyclic substrates. Such studies have suggested two mechanistic routes: 1) a nucleophilic attack on pro-S to cleave the bond between pro-S and C␣ of ACC followed by the abstraction of proton from pro-R (nucleophilic route), 2) direct pro-S-proton abstraction to initiate the cyclopropane ring fragmentation (direct abstraction route) (8). The first route is more likely because the second route requires abstraction of an inert proton from the ring for initial reaction and such an anion-induced ring cleavage is stereoelectronically unfavored (9). In the light of these previous descriptions, we sought for the residues that are involved in the enzymatic reaction. Although no clear-cut reaction mechanism could be proposed from the present analysis alone, the most likely candidate for the nucleophile is Ser 78 . This residue is hydrogen-bonded to Thr 81 and is close to the putative ACC molecule. In this case, the released Lys 51 is the most likely candidate for base for proton abstraction. In order to clarify such reaction mechanism, the site-directed mutagenesis studies are currently under way.