The active principle of garlic at atomic resolution.

Despite the fact that many cultures around the world value and utilize garlic as a fundamental component of their cuisine as well as of their medicine cabinets, relatively little is known about the plant's protein configuration that is responsible for the specific properties of garlic. Here, we report the three-dimensional structure of the garlic enzyme alliinase at 1.5 A resolution. Alliinase constitutes the major protein component in garlic bulbs, and it is able to cleave carbon-sulfur bonds. The active enzyme is a pyridoxal-5'-phosphate-dependent homodimeric glycoprotein and belongs to the class I family of pyridoxal-5'-phosphate-dependent enzymes. In addition, it contains a novel epidermal growth factor-like domain that makes it unique among all pyridoxal-5'-phosphate-dependent enzymes.

class I of PLP-dependent enzymes (6). With its C-S lyase activity, alliinase is able to cleave the C ␤ -S ␥ bond of sulfoxide derivatives of cysteine to produce allicin (Fig. 1B). Because of the vacuolar location of the enzyme and the presence of its substrate in the cytosol, the alliinase/alliin system has been discussed as a primitive defense mechanism of the plant (7). Based on multiple sequence alignments of various alliinase sequences, the presence of a unique epidermal growth factor (EGF)-like domain was proposed (8). The protein also contains four putative N-glycosylation sites at Asn 19 , Asn 146 , Asn 191 , and Asn 328 (9). Recently, diffraction quality crystals of the natural form of the enzyme have been obtained (8,10).
Here, we report the three-dimensional structure of garlic alliinase at 1.5 Å resolution. Alliinase is only the second garlic protein for which a structure determination has been carried out. We confirm the presence of the proposed EGF-like domain, rationalize the reaction mechanism, and provide some explanation for the substrate selectivity of alliinase.
Diffraction Data Collection, Structure Determination, and Refinement-The structure was determined using the single isomorphous replacement with anomalous scattering (SIRAS) method. The native data set was collected to a resolution of 1.53 Å at the EMBL beamline BW7B (DESY, Hamburg, Germany) and the gold derivative data set to a resolution of 2.1 Å at the x-ray diffraction beamline at the ELETTRA synchrotron (Trieste, Italy). Three gold sites were identified from an anomalous difference Patterson synthesis. Phases were calculated using the program MLPHARE (11) and improved by solvent flattening using the program DM (11). The figures-of-merit of the phases belonging to the correct hand were 0.33 and 0.73 before and after solvent flattening, respectively. The initial structure model was built using the ARP/wARP procedure (12), and structure refinement was carried out using the program REFMAC version 4 (11).
Structure Comparisons-Structural homology searches against other known structures were performed using the DALI server (13). The programs ALIGN (14) and LSQKAB (11,15) were used to calculate the three-dimensional structural alignments and the root mean square differences between the aligned structures.

RESULTS AND DISCUSSION
The Overall Structure-The refined model of alliinase consists of residues 2-425 in subunit A, 2-427 in subunit B, 4 sugar chains, 830 water molecules, 2 chloride ions, and 11 sulfate ions as well as 2 partially occupied HEPES buffer molecules and 1 partially occupied aminoacrylate (AA) moiety. With R and R free values of 19.3 and 22.1% (Table I) 1 The abbreviations used are: PLP, pyridoxal-5Ј-phosphate; aAT, ar-resolution of 1.53 Å, it can be considered well refined. An example for the quality of the final electron density is shown in Fig. 2. The quaternary structure of alliinase is a dimer consist-ing of two identical subunits related by a rotation of 180°( Fig.  3) with a root mean square deviation between the two subunits of the homodimer of 0.36 Å based on the superposition of all 424 C ␣ -pairs. Each alliinase monomer consists of three distinct domains, a unique N-terminal domain, the central PLP-binding domain, and the C-terminal domain. Comparisons of alliinase with 1-aminocyclopropane-1-carboxylate (ACC) synthase (16) as a representative for C-S lyases and aromatic amino acid-aminotransferase (aAT) (17) as a typical aminotransferase show that the overall folds of the central and the C-terminal domains of the three enzymes are similar (Fig. 4). The sevenstranded mixed ␤-sheet in the central domain and the threestranded antiparallel ␤-sheet in the C-terminal domain flanked by three ␣-helices all on one side are present in all three cases. Three-dimensional alignments with the open (18) and the closed form of chicken mitochondrial aspartate aminotransferase (19) reveal that the conformation of alliinase is more akin to the closed than to the open form of aspartate aminotransferase.
The EGF-like Domain-A peculiar and distinguishing fea- f Phasing power for centric (cen) and acentric (acen) reflections: ture of the alliinase structure is the presence of an EGF-like domain in its N-terminal part (Fig. 5A). EGF-like domains are small disulfide-rich structures (20) that often constitute modules for binding to other proteins. Although they occur frequently in animal proteins, they are rather uncommon in plant proteins. When they do occur, they are usually found in the extracellular portion of membrane-bound or secreted proteins. Alliinase constitutes the first example of a catalytic domain fused to an EGF-like domain in a plant enzyme. Comprised of 47 amino acids, this domain ranges from Glu 13 to Ala 59 and contains six cysteine residues arranged in a disulfide pattern of the type (C1-C2, C3-C5, C4-C6). This pattern is different from the one (C1-C3, C2-C4, C5-C6) found in the canonical EGFs. The first and second disulfide bond of alliinase corresponds to the second and third of canonical EGFs (Fig. 5B). A search for structural homologues revealed that the closest known relative is the heparin-binding epidermal growth factorlike growth factor, also known as the diphtheria toxin receptor (21). A superposition of the two molecules is shown in Fig. 5B. The function of this EGF-like domain in alliinase is unclear.
One may speculate that it is a binding site for other proteins or the docking site for a hypothetical alliinase receptor. An inter-  (29) and RASTER3D (28). In (A), the 2-fold axis of the dimer is vertical in the paper plane, whereas in (B) the view is down the 2-fold axis. The typical S-shaped structure of the dimer is clearly discernible. esting observation along this line is that garlic consumption leads to the appearance of anti-alliinase antibodies in human blood serum (22). This seems to imply that alliinase must be resorbed into the circulatory system with its three-dimensional structure still intact.
The Chloride Binding Loop-A strong peak in the anomalous difference electron density map revealed the presence of a chloride ion bound to the loop 92-100 in each of the subunits. The chloride ion is hydrogen-bonded to three main-chain amide NH groups (Phe 94 , Ser 98 , Phe 100 ) and one water molecule. It stabilizes the loop 92-100 (including a cis-peptide bond between Asn 95 and Pro 96 ) in a conformation that orients the aromatic amino acids Tyr 92 , Phe 93 , and Phe 100 toward the active site of the neighboring subunit. These residues presumably bind the hydrophobic part of the substrate. The presence of chloride in this position provides a convincing explanation for the previously made observation that NaCl stabilizes the enzymatic activity and enhances dimer stability (8,23).
The Glycosylation Sites-Only two of the four predicted glycosylation sites per alliinase subunit (Asn 146 and Asn 328 ) were observed to be utilized, although all four Asn residues are located at the surface of the molecule. The core sugar structure is Man(␤1-4)GlcNAc(␤1-4)[Fuc(␣1-3)]GlcNAc(␤1-N)Asn, which is quite typical for plant glycoproteins (Man: mannose; GlcNAc: N-acetylglucosamine; Fuc: fucose). Up to four sugar rings are visible in the electron density map at the four sites in the dimer. The sites at Asn 146 are located at the dimer interface, and the sugar chains bind to both subunits, thereby stabilizing the dimer. In contrast, the sites at Asn 328 point into the solvent area and do not contact any protein atoms.
The Active Site (Cofactor Binding)-The active site of alliinase is located at the dimer interface and consists of residues from both subunits (Fig. 6). The PLP moiety is covalently bound to Lys 251 via a Schiff base, forming a so-called internal aldimine. The pyridine ring of the PLP is sandwiched between the aromatic ring of Tyr 165 and the isopropyl group of Val 227 , which forms a C-H . . . hydrogen bond (24) to the PLP heterocycle (not shown in Fig. 6). Other PLP binding features include the hydrogen bonds from Asp 225 to the pyridinium nitrogen and from Asn 207 and Tyr 228 to the phenolic oxygen atom of the PLP cofactor. The hydrogen bond formed by the amide side chain of Asn 207 appears to be an especially interesting feature, because Asn 207 belongs to the so-called "strained loop" (see below). The negative charge of the PLPphosphate group is stabilized by the helical dipole of ␣-helix 132-143, by Arg 259 , and by four further hydrogen bonds involving the side chains of Thr 133 , Thr 248 , Ser 250 , and Tyr 92 of the neighboring subunit, a structural motif which has been termed the phosphate-binding cup (25). Taken together, all parts of the PLP molecule are held very tightly by the enzyme, reflecting the need to not lose the cofactor during the reaction cycle when the covalent bond between the enzyme and the cofactor is broken.
The Active Site (Substrate Binding)-Additional electron density in the active site of alliinase suggested the presence of other chemical entities (Fig. 7). The electron density was interpreted as HEPES molecules from the crystallization medium binding to both active sites of the dimer, albeit at partial occupancy. In addition, in one of the active sites an AA moiety was found to be covalently linked to the PLP cofactor but also at partial occupancy only. The AA most likely originated from the inhibitor S-ethyl-L-cysteine (SEC), which was present during the crystallization of alliinase (8). Apparently, SEC is 17) (right). A, topology diagrams: the spheres and triangles represent ␣-helices and ␤-strands, respectively. The small symbols represent ␣or 3 10 -helices containing four to seven residues and ␤-strands of three residues. B, MOL-SCRIPT diagrams (29) of the monomers of the three enzymes in the same orientation. The overall root mean square deviations between alliinase and ACC synthase and aAT are 1.96 and 1.93 Å based on 300 and 285 aligned C ␣ pairs, respectively. not an inhibitor but a poor substrate. The current electron density is consistent with the presence of a 50:50 mixture of internal aldimine (Lys 251 -PLP with HEPES present) and external aldimine (Lys 251 plus PLP-AA), although the formation of a geminal diamine (Lys 251 -PLP-AA) cannot be ruled out. In any case, the observed architecture enabled us to model the real substrate alliin into the active site (Fig. 6). The carboxylate group (bound tightly by the guanidinium group of Arg 401 ), the C ␣ , and the nitrogen atom of the AA moiety were assumed to occur in the identical positions in the substrate. The sulfoxide and the allyl group of alliin were then oriented so that the sulfoxide oxygen was close to a hydrogen bond donor. The only conformation without sterical clashes turned out to be the one in which hydrogen bonds between the sulfoxide oxygen of a (ϩ)-alliin and Ser 63 -OH and the backbone amide NH of Gly 64 are formed (Fig. 6B). The allyl group would then be in close contact to the side chain of Tyr 92 of the neighboring subunit. For (Ϫ)-alliin in the same conformation, the hydrogen bond to Ser 63 -OH is not possible, but instead a weak hydrogen bond to the amide side chain of Gln 388 could be formed (Fig. 6B). In these alliin conformations, the sulfoxide oxygens assume approximately  the same positions as the sulfone oxygen atoms in the bound HEPES molecules, which lends further support to our model. In addition, the sulfoxide oxygens of alliin are relatively far from the carboxylate oxygens of Glu 283 of the neighboring subunit (5.8 and 6.0 Å for (ϩ)-alliin and 5.6 and 6.3 Å for (Ϫ)-alliin), thus minimizing electrostatic repulsion. These pieces of evidence, taken together, provide a possible explanation for the selectivity of alliinase toward (ϩ)-alliin.
The Strained Loop Thr 203 to Glu 211 -In both subunits of the dimer, a highly strained loop was observed to be involved in the binding of both the PLP cofactor and the substrate. This loop occurs between ␤-strands 3 and 4 of the central seven-stranded ␤-sheet in the PLP-binding domain and is located at the interface of the central PLP-binding and the C-terminal domains. It contains the two cis-peptide bonds, Ser 204 -Pro 205 and Asn 207 -Pro 208 . With -angles of Ϫ4°and ϩ18°in subunit A and Ϫ10°a nd ϩ25°in subunit B, both these peptide bond conformations deviate significantly from planarity. In addition, the transpeptide bonds Asn 206 -Asn 207 ( ϭ 170°and 168°) and Pro 208 -Glu 209 ( ϭ Ϫ168°and Ϫ164°) are also significantly non-planar, and Pro 208 makes a van der Waals contact to Pro 163 of the neighboring loop, which is preceded by another peptide bond in cis-conformation. Because the side chain of Asn 207 forms a hydrogen bond to the PLP cofactor (see above) and presumably to the substrate as well (Fig. 6), it may be that this strained conformation has some influence on the conformational change that is expected to happen during the reaction cycle. This change presumably involves a reorientation of the C-terminal domain relative to the central PLP-binding domain. Because of its location at the interface of the two domains, the strained loop is a prime candidate for triggering such a reorientation. In a mutational study in aspartate aminotransferase from Escherichia coli it could be shown that the cis-conformation corresponding to the Asn 207 -Pro 208 peptide bond is retained when the Pro is replaced by an Ala, whereas the cis-conformation corresponding to the Ala 162 -Pro 163 peptide bond is not (26). Furthermore, based on the reducibility of the PLP-Lys aldimine it was indicated that the mutation of Pro 195 (which corresponds to Pro 208 in alliinase) to Ala may affect the open/ closed equilibrium of the enzyme (26). This is in full accord with our hypothesis.
In conclusion, the elucidated structure of alliinase opens the door to a more rational understanding of garlic sulfur chemis-try, and it may even provide a first step toward the elucidation of the reaction mechanism and the design of altered selectivities and therefore altered therapeutic features of garlic and other Allium plants.