The crystal structure of the hydroxynitrile lyase from the rubber tree Hevea brasiliensis suggests that this enzyme is structurally

Structure (1996) 4, 811-822

The crystal structure of the hydroxynitrile lyase from the rubber tree Hevea brasiliensis suggests that this enzyme is structurally and mechanistically related to a/b hydrolases.

U. G. Wagner^1*, M. Hasslacher^3,4, H. Griengl^2,4, H. Schwab^3,4, C. Kratky¹

^{1Institut für physikalische Chemie, Universität Graz, Heinrichstraße 28, A-8010 Graz, Austria,
²Institut für organische Chemie, Technische Universität Graz, Stremayrgasse 18,
A-8010 Graz, Austria,
³Institut für Biotechnologie, Petersgasse 12, A-8010 Graz, Austria, ⁴Spezialforschungsbereich Biokatalyse, Technische Universität Graz, A-8010 Graz, Austria

Running Title: 3D structure of Hydroxynitrile Lyase}

Keywords: cyanogenesis, X-ray diffraction, a/b hydrolase, cyanhydrin fomation, oxynitrilase

Background: Cyanogenesis, i.e. the liberation of HCN upon tissue damage, is a frequent defense mechanism against predation known to occur in several thousand plant species. One of the key enzymes of this process is hydroxynitrile lyase (Hnl), which mediates the decomposition of a a-cyanohydrin to form hydrocyanic acid (HCN) plus the corresponding aldehyde or ketone. A second physiological role for cyanogenesis is the utilization of the liberated HCN for the synthesis of asparagine. Hnl's are also of potential use for the industrial biocatalytic synthesis of chiral cyanohydrins, and they are a target to detoxify cyanogenic food crops. For these reason, elucidation of the mechanism of hydroxynitrile lyases at the molecular level is of fundamental biological and industrial relevance. 3D structure information is an indispensable prerequisite towards this objective.

Results: We have determined the crystal structure of the hydroxynitrile lyase (Hnl) from the rubber tree Hevea brasiliensis at 1.9 Å resolution. The structure belongs to the a/b hydrolase superfamily. Its active site is deeply buried inside the protein, and connected to the outside by a narrow tunnel. Electron density interpreted as a histidine molecule (or ion) was observed in the present crystal structure's active site. The catalytic triade consists of residues Ser80, His235 and Asp207. By analogy with other a/b hydrolases, the oxyanion hole is formed by the mainchain-NH of Cys81 and by the sidechains of Cys81 and Thr11.

Conclusions: The 3D structure suggests a mechanism for enzyme-catalyzed formation and cleavage of cyanohydrins. By analogy with the mechanism of related a/b hydrolases, a tetrahedral hemiketal or hemiacetal intermediate with the Ser80-OH is proposed. Cys81-SH participates in the stabilization of the negatively charged oxygen, and possibly mediates proton transfer between HCN and the hydroxynitrile-OH. It thus appears that the mechanism of Hnl-catalyzed cyanogenesis differs from the corresponding solution reaction, with the former belonging to the realm of "SN₂-chemistry" and the latter to the one of "carbonyl-chemistry".

Introduction

Cyanogenesis, the release of hydrogen cyanide (HCN) from injured tissue, has been observed in more than 3000 plant species including crop plants [1]. Liberation of HCN from stored cyanoglycosides is initiated by a b-glycosidase to produce free sugar and a relatively unstable a-hydroxynitrile (cyanohydrine), which decomposes into HCN and the corresponding aldehyde or ketone. This process occurs spontaneously at high pH, but requires assistance by the enzyme a-hydroxynitrile lyase at acidic pH [2]. Cyanogenesis is widely accepted as a defense mechanism against predation or fungal attack [3], and a second physiological role of cyanogenic glycosides for the biosynthesis of L-asparagine has been suggested [4].

Besides their biological interest, hydroxynitrile lyases (Hnl's) have attracted much attention as potent biocatalysts for the synthesis of optically pure cyanohydrins [5, 6, 7, 8], making use of the reverse in-vivo reaction. Chiral cyanohydrins are important synthetic intermediates for a wide range of pharmaceuticals and agrochemicals. Moreover, they are an interesting target to detoxify cyanogenic food crops forming a potential health risk to its consumers mainly in the third world. The most important of these crops is cassava (Manihot esculenta), which forms the basic nutriment for several hundred million people with an estimated annual production of 150 million metric tons [9].

Hnl's have been isolated and purified from a variety of plant species, and naturally fall into two different classes discernible by the presence or absence of FAD as cofactor. FAD dependent Hnl's are typically isolated from Rosaceae. They are highly glycosylated, single chain proteins with molecular masses between 50 and 80 kDa, accept (R)(+)mandelonitrile as natural substrate, and show sequence homology with FAD-dependent oxidoreductases [10]. FAD-independent Hnl's have been isolated from Olacaeae [11], Linacaeae [12], Graminae [13], Euphorbiaceae [9] and Polypodiaceae [14]. They show molecular masses between 20 and 42 kDa, contain up to 9% carbohydrate, occur as homo- or heterooligomers and accept a variety of (R) or (S) configured cyanohydrins as substrates. The (S)-Hnl from Sorghum bicolor (Graminae) is related to serine carboxypeptidase [15]. Thus, it appears that nature has invented enzymes with oxynitrilase activity at least twice, once derived from FAD-dependent oxidoreductases and once from a/b hydrolases. Both classes of Hnl's are the subject of intense efforts towards the elucidation of their 3D structures [16, 17].

(S)-Hydroxynitrile lyase (EC 4.1.2.39) from the Euphorbiacea Hevea brasiliensis is an unglycosylated protein with 29.2 kDa subunit molecular mass, as calculated from its sequence [19]. There is evidence from gel filtration that the protein occurs as a homodimer in solution at or near neutral pH [18]. It is homologous (74% sequence identity [19]) to the Hnl from cassava (Manihot esculenta [9]) and distantly related (35% sequence identity [19]) to two rice proteins of unknown function [20]. Its natural substrate is (S)-acetone cyanohydrine, but due to a low substrate specificity it can be used for the synthesis of chiral a-hydroxynitrils with aliphatic, aromatic and heterocyclic aldehydes or ketones [7, 8].

In spite of insignificant sequence homology (sequence identity between Hnl and other a/b hydrolases from the Swissprot data base, after manually aligning the secondaty structure elements of the 190 residue long peptide forming the a/b hydrolase motive: 6% with acetylcholinesterase ACES TORCA, P04058; 13% with carboxypeptidase CBPY YEAST, P00729; 12% with haloalkane dehalogenase HALO XANAU, P22643), a profile search in the Swissprot data base in combination with secondary-structure prediction led us to believe that, like the Hnl from Sorghum bicolor [15], the Hevea enzyme should belong to the a/b hydrolase superfamily [M. Hasslacher, unpublished result]. Sequence alignment [19] and site-directed mutagenesis (no detectable Hnl activity for the Ser80®Ala and His235®Ala mutants; M. Hasslacher, unpublished result) permitted the putative identification of the residues of the catalytic triade (Ser80, His235, Asp207). The strong sensitivity of the catalytic activity towards mercurials [20] as well as the low activity (<5%) of the Cys81®Ser mutant [20] point to an involvement of Cys81 in the mechanism of catalysis.

We have determined the crystal structure of recombinant Hnl from Hevea Brasiliensis at a resolution of 1.9Å. These data show that the enzyme indeed belongs to the a/b hydrolase family, and they suggest a mechanism for this class of cyanogenic enzymes. To our knowledge, it is the first report of the 3D structure of a member of this class of enzymes.

Results and discussion

Structure determination

Although sequence-based structure prediction indicated an a/b hydrolase fold, attempts to solve the structure by molecular replacement with various a/b hydrolases as search models failed. We therefore had to resort to ab initio multiple isomorphous replacement using 5 derivatives. Native and derivative data (see table 1) were initially collected at cryo-temperature on a rotating anode generator. While these data permitted structure solution and refinement to about 2.5 Å resolution, we also collected a native (room-temperature) synchrotron data set for the last stages of refinement, using partial data to 1.9 Å and complete data to about 2.2 Å resolution (completeness of data between 1.9 and 2.2 Å: 74.3% for all data and 54.1% for 3s data). Using these data, all residues could be observed and assigned, with the exception of the N-terminal methionine (which had probably been cleaved off), the C-terminal asparagine (which shows diffuse density and refines to large B-values) and the sidechains of Lys223 and Glu192, which protrude into the solvent region and are disordered. No electron density was observed for these residues beyond C_b. With 77 water molecules and one sulfate ion, this model refined to R=19.9% for data between 15-1.9 Å (3 s cutoff, see experimental section), and shows good stereochemistry by PROCHECK [21] analysis. Only one residue - the catalytic residue Ser80 - falls outside favourable regions of the Ramachandran plot (see below). Stereoscopic views of the molecule are shown in Fig.1, an indication of the map quality at the current state of structure refinement is shown in Fig.2a.

In one of the last difference maps, an extended cluster of electron density was noted within the putative active site of the enzyme. This density was interpreted as a bound histidine ion, (Fig.2b). Scrutiny of the purification protocol of the Hnl batch used for crystallization revealed that in the first purification step an ion exchange column was eluted with a histidine buffer. Evidently, the enzyme had picked up a molecule of histidine from this buffer and kept it through the remaining purification and crystallization steps. Thus, although we had set out to determine the crystal structure of native Hnl, we apparently ended up with a complex between Hnl and an inhibitor, presumably histidine.

Crystal packing

Hnl molecules of approximate dimensions 30x38x48 Å pack in an orthorhombic unit cell in space group C222₁, (a=47.4, b=109.0, c=128.2 Å at room temperature). Each molecule makes contacts with 6 neighbours:

(1) Pronounced dimers between two crystallographically related molecules are formed by the operation of the dyad along a (see Fig.3). The contact area between two related molecules consists of apolar residues of the loop connecting strand 2 and helix B, as well as of helices A and D₃' (see Fig.4). The apolar nature of the majority of interacting residues is compatible with evidence from gel filtration that Hnl is dimeric in aqueous solution at neutral pH [18]. Homodimers (involving, however, contacts between topologically different helices) were also observed in the crystal structure of acetylcholinesterase (AchE) [22].

(2) Molecules related by the translation along the crystallographic a-axis are connected through hydrogen bonds to and from a sulfate ion, leading to 1-dimensional stacks of molecules along a.

(3) Molecules related by a dyad along b make contacts through helix E, with symmetry - equivalent helices interacting at an angle of approximately 25°.

(4) Phe 3 and Gly30 make hydrophobic contacts with Pro69 and Pro70 of a molecule at x+1/2, -y+1/2, -z-1.

The result of interactions (1) to (3) is the formation of undulated 2-dimensional sheets perpendicular to the b-axis. Interaction (4) connects adjacent sheets in such a way, that extensive solvent channels along a are formed. The active site opens into these channels, making future substrate diffusion studies promising. The observation of a layer-arrangement of Hnl molecules correlates with the pronounced platelike morphology of the crystals and with the large decrease in the length of the b-axis upon cooling to cryotemperatures [17].

Hnl and the hydrolase family

Hnl belongs to the a/b hydrolase fold family [23, 24], which includes a variety of enzymes with different catalytic functions such as hydrolases [25], carboxypeptidases [26], lipases [27, 28], thioesterases [29], dehalogenases [30], oxidoreductases [31] and acetylcholinesterase (AchE) [22]. This fold is characterized by an extended central b-sheet flanked by a-helices on both sides.

Compared to the "prototypic hydrolase fold" [23], the topology of the Hnl (Fig.4a) lacks the first two b-strands and has an extended "cap" region between strand 4 and helix D, consisting of a three-stranded b-sheet and three helices. Fig. 4b gives an indication for the agreement of elements of secondary structure between Hnl and a typical member of the a/b hydrolase family (AchE). Evidently, while the central b-sheet appears to be quite well conserved between the two structures, corresponding helices of the a/b hydrolase motive deviate considerably.

Enzymes with a/b hydrolase fold share a number of structurally conserved features around the active site, which can be used to identify the residues involved in enzyme catalysis. In view of the similarity to AchE, we will concentrate on a comparison with this enzyme, which - together with a lipase from Geotrichum candidum (GCL) [28] - forms the prototype of a class of closely related a/b hydrolases with lipase- or esterase activity [24].

The nucleophile elbow.

Ser80 is the central residue in a sharp g turn between strand 3 and helix C. The sharpness of the turn results in its backbone phi and psi angles being the only non-Gly residue in an unfavourable region of the Ramachandran plot. Like in other a/b hydrolases, the potential steric problem resulting from the sharpness of the turn is avoided by small residues at key positions, leading to the (frequently observed) consensus sequence G-x-S-x-G-G arranged in a highly conserved structural motiv known as the nucleophile elbow. In AchE and the related lipase/estarase family, the residue preceding the central serine is usually a glutamate [24], which is also observed in the present structure.

The catalytic triade.

In the majority of known a/b hydrolases, the residues forming the catalytic triade (typically - as in the present case - Ser, His and Asp) are in topologically and structurally related loop regions. Identification of Ser80, His235 and Asp207 as the residues forming the catalytic triade is thus unambiguous, although Ser80 and His 235 are not directly connected by a hydrogen bond (His235Ne₂...Ser80 0g, 3.9 Å) due to the interference of the bound inhibitor. Asp207 and His235 Nd₁ are within hydrogen bonding distance.

The assignment of the residues forming the catalytic triade is corroborated by the following features observed similarly in other a/b hydrolases:

the active-site histidine residue frequently occurs [28] in a b-turn within the sequence G-Sm-x-H-Sm-x-E/D (Sm denotes a small, x any residue). In the present structure, the corresponding sequence is G-G-D-H235-K-L-Q.

A hydrogen bond from one of the aspartate side-chain carbonyl oxygens to the backbone NH of the residue two amino acids towards the C-terminus is characteristic for the third residue of the catalytic triade [23]. This interaction (in the Hnl structure between the corresponding atoms of residues Asp207 and Ile209) stabilizes the Asp207 sidechain and guides it into the correct orientation for hydrogen bonding to the catalytic His235.

A sharp bend occurs between strand 4 and the following peptide at residue Ser105. A similar sharp turn is observed at the topologically equivalent position in other a/b hydrolases, where the corresponding position is usually occupied by a small (frequently Gly) residue [23]. Both, the sharp turn and the small side chain are required to avoid interference with the catalytic triade and helix C.

The Oxyanion hole

Identification of the atoms and residues presumably involved in oxyanion stabilization is suggested by Fig. 5, showing a superposition of the catalytic site region between the structures of Hnl and of acetylcholinesterase (AChE) [22]. The two molecules were superimposed with the C_a-atoms adjacent to the catalytic serine (Ser80 in Hnl, Ser200 in AChE). For AChE, the oxyanion hole was proposed [22] to consist of the mainchain NH-groups of residues Gly118, Gly119 and Ala201. In the Hnl structure, only the mainchain NH of Cys81 is in a conformation corresponding to Ala201 of AChE, capable of hydrogen bonding to a putative oxyanion. The region of the "opposite" chain (AChE residues 117-199) is occupied in Hnl by much bulkier residues (Thr11 and Ile12), which are oriented such that their mainchain - NH are not within H-bonding distance of the oxyanion. Instead, Thr11-OH and Cys81-SH both point towards the catalytic Ser80, suggesting their involvement in oxyanion stabilization.

The active site

From the topological and structural analogy with other a/b hydrolases the location of the active site can be assigned with considerable confidence. Moreover, site-directed mutagenesis showed that replacement of Ser80 and His235 (by Ala) leads to complete (M. Hasslacher, unpublished result) and of Cys81 (by Ser) to substantial (>95%) [19] loss of catalytic activity.

The active site is deeply buried inside the protein, and it is connected to the protein surface by a narrow channel, which is flanked by predominantly apolar residues from the "cap" domain (see Fig. 6). The active site cavity is also formed by apolar residues, with the exception of the nucleophile Ser80 and the two residues proposed to be involved in oxyanion stabilization (Cys81 and Thr11). The apolar nature of the active site is illustrated by Fig. 7, which compares surface representations of Hnl and AchE, coloured according to electrostatic potential. As noted earlier [32], AchE has a large negative charge distributed in such a way that the cationic substrate is "pulled" into the active site. This fact reflects itself in a negative surface potential (relative to the rest of the molecule) for the active site region of AchE (Figs. 7c and 7d). The corresponding surface area of Hnl, on the other hand, shows no pronounced potential, neither positive nor negative (Figs. 7a and 7b). This is not surprising since all substrates and products of the Hnl catalyzed reaction are uncharged at or near neutral pH.

A histidine ion is bound to the active site.

Within the inferred active-site region, a final F_o-F_c map revealed extended electron density above the 3s level (see Fig.2b). Although this density could not be assigned with complete certainty, it was interpreted as a histidine molecule (or ion), which presumably had been picked up in the first purification step. One of the carboxylate oxygen atoms of the histidine is interposed between the sidechains of the triad residues Ser80 and His235, within hydrogen-bonding distance to each of them, remeniscent of the way how the edrophonium inhibitor is bound to AchE [33]. The other carboxylate oxygen is within H-bonding distance of Lys236. The imidazole end of the histidine points towards the putative oxyanion hole, with a hydrogen bond between its Nd and Cys81-SH.

Preliminary experiments have shown, that under the assay conditions [20], histidine does not or only weakly inhibit Hnl activity. However, the assay (rate of cleavage of mandelonitrile) is routinely carried out at low pH (pH 5.5) in order to suppress the non-enzymatic reaction, whereas the purification and crystallization was done close to neutral pH (pH 7.5). Since histidine has a pK around 6 in aqueous solution, it is conceivable that histidine only binds to the active site of Hnl in its deprotonated form. This would be consistent with the observation [18] that anionic species - such as citrate and acetate - inhibit Hnl activity, as well as with the fact that the active site of Hnl is almost uncharged (Fig. 7). While no comprehensive and unambiguous data concerning the binding of histidine to the active site and ist effect on the catalytic activity of Hnl are available at the moment, experiments are under way in our laboratories to determine the nature and function of the molecule bound to the presumed active site, and to study the binding and inhibition of histidine and related molecules.

The cap region.

While a high degree of structural and topological conservation of the "core" domain has been observed for all a/b hydrolases, the majority of structures of this family show at least one additional domain with considerable topological, structural and functional variability. This extra domain is frequently called the "cap". In the Hnl structure, it consists of a 3-stranded sheet and three helices (Fig. 1) surrounding and forming the entrance to the active site (Fig. 6). Topologically, the "cap" is inserted between strand four and helix D (Fig. 4), which is a frequent but not the only topological position for "cap" domains in a/b hydrolases [23]. It appears that nature has used the cap domains to adapt the core with its conserved catalytic residues to a variety of diverse functions. Thus, cap domains are involved in interfacial activation (lipases [34]), substrate recognition (dehalogenases [30] and carboxypeptidases [26]) and enzyme inactivation [35]. In acetylcholinesterase [22] residues of the cap domain contribute to the formation of the channel to the active site. In spite of considerable topological and structural differences between the cap domains of Hnl and AChE, the present work suggests an analogous function in the two enzymes.

The reaction mechanism of the Hnl enzyme.

Based on the crystal structure of the Hnl-inhibitor complex plus the available data about enzyme kinetics and inhibition, it is tempting to speculate about a possible mechanism for the Hnl-catalyzed reaction (Fig. 8). In doing so, we assume close correspondence to the (more or less) established mechanisms of other enzymes with a/b hydrolase fold. The reaction will be discussed in the direction of cyanohydrin formation (i.e. in the reverse direction to what is known to occur in cyanogenic plants), since the analogy to the mechanism of hydrolases is more evident there.

Nucelophilic attack of the substrate's carbonyl

Central to the hydrolase mechanism is the occurrence of a tetrahedral intermediate formed by attack of the (activated) nucelophile to the carbonyl group. In the case of Hnl, the nucleophile Ser80 (activated by incipient proton abstraction through a chain of hydrogen bonds to His235 and Asp207) attacks the carbonyl of the substrate aldehyde or ketone, leading to a hemi-acetal or hemi-ketal intermediate. As in corresponding hydrolases, the negative charge is stabilized in a site with high propensity for hydrogen-bonding known as oxyanion-hole, which in the case of Hnl consists of the mainchain amide of Cys81 and of the sidechains of Cys81 and Thr11 (Fig. 5).

The tetrahedral intermediate

Fig. 9 shows the energy-minimized result of a modelling study of the tetrahedral intermediate of the natural substrate (acetone), viewed from the outside through the channel to the active site. Nucleophilic substitution of the protein by an incoming CN^- is proposed as the next rational step in the cyanohydrin formation reaction. We note that the tetrahedral intermediate in its proposed orientation (Fig. 9) is ideally positioned for a "back-side" attack of a nucleophile advancing through the channel.

Displacement of Ser80 by CN^-

Under neutral or acidic conditions, cyanide occurs as HCN in aqueous solution. It is tempting to propose that the Cys81-SH - which would be ideally positioned for this task - mediates deprotonation of HCN and concommittant protonation of the oxyanion. This would be in agreement with the fact, that Hnl is completely inhibited by SH-active reagents such as mercurials, and with the low (<5%) activity of the Cys81 ® Ser mutant [19]. Like in related hydrolases, substitution is facilitated by back-protonation of the leaving group (Ser80) by the His235 imidazole. This last reaction step is followed by the release of the cyanohydrin product through the channel.

The fact, that the active site is deeply buried inside the protein with only one narrow channel to the outside suggests that the two substrates (ketone and HCN) have to enter (or leave) in a sequential fashion, in agreement with the kinetically derived ordered uni-bi mechanism [36, 18]. It is also remarkable that the active site is much larger than required to accommodate the biological substrate acetone cyanhydrin, and it is therefore not surprising that a wide range of other aliphatic or aromatic cyanhydrins are cleaved by this Hnl. The proposed sequence of reaction steps differs markedly from the reaction mechanism for (uncatalyzed) cynohydrin formation in solution, where direct attack of cyanide to the carbonyl is believed to occur. In contrast, according to the proposed mechanism, enzyme-catalyzed cyanohydrin formation involves SN₂-type attack of CN^- to the tetrahedral hemiacetal or hemiketal intermediate.

Biological Implications

Cyanogenesis is known to occur in several thousand plant species, and serves at least two purposes: defense against herbivoral or fungal attack and release of HCN from cyanogenic glucosides for the synthesis of asparagine. Hydroxynitrile lyase is one of the two key enzymes required for cyanogenesis, catalyzing the decomposition of a cyanohydrine into HCN and the corresponding aldehyde or ketone. Nature has invented at least two different classes of Hnl's, one related to FAD dependent oxidoreductases, the other to a/b hydrolases. They differ with respect to substrate specificity and stereoselectivity, and possibly also with respect to their catalytic mechanisms, although kinetic data suggest for both classes of enzymes a mechanism involving the sequential release of first HCN and then the aldehyde or ketone (ordered uni-bi mechanism [37, 36, 18]). Correspondingly, the reverse reaction would involve sequential binding of the aldehyde/ketone followed by HCN.

The crystal structure analysis for the Hnl from Hevea brasiliensis, which constitutes the first 3D-structure determination for an enzyme with oxynitrilase activity, confirms the predicted membership to the a/b hydrolase family, with a particularly close relationship to acetylcholinesterase. The structure suggest an enzyme mechanism analogous to other hydrolases, which accomodate a catalytic triade consisting of a serine nucleophile (Ser80 in Hnl) activated by a histidine (His235) and an aspartate (Asp207) residue. According to the proposed mechanism, nucleophilic attack of the Ser80-OH on the carbonyl (for the Hnl catalyzed cyanohydrin formation) leads to a hemiacetal or hemiketal intermediate, whose negative charge is stabilized by a group of hydrogen bond donors known as oxyanion hole. A key residue of the proposed oxyanion hole is Cys81, which is suggested to mediate proton transfer between the (incoming or outgoing) HCN and the hemiacetal/ketal oxygen. There is only one entrance to the (apolar) catalytic site, in agreement with the kinetically predicted ordered uni-bi mechanism.

The proposed mechanism of enzyme-catalyzed cyanohydrin formation (or cleavage) differes markedly from the unkatalyzed reaction in solution, suggesting that evolution has preferred to adapt a hydrolase to the task of cyanogenesis rather than invent a new enzyme on the basis of the solution mechanism. Thus, while the solution chemistry of cyanohydrins belongs to the realm of "carbonyl chemistry", the proposed mechanism for the corresponding Hnl-catalyzed reaction involves nucleophilic displacements between CN^- and (activated) Ser-OH.

Material and Methods

Enzyme preparation and crystallization.

The recombinant Hnl protein from Hevea brasiliensis was overexpressed in yeast (Pichia pastoris) and purified to homogeneity [18, 19].

The first step of the purification consisted of FPLC ion exchange chromatography of the crude extract on a Resource Q column (Pharmacia Biotech, Uppsala, Sweden). The column was equilibrated with loading buffer (10 mM histidine / H₂S0₄, pH 6.7) and Hnl was eluted with a linear salt gradient increasing from 0 to 0.6 M ammonium sulfate in loading buffer. At this stage, the Hnl had evidently picked up a histidine molecule (or ion), which remained attached to the active site during the second purification which consisted of a size exlusion chromatography on a Biogel P150 column (BioRad, Hercules, CA, USA), equilibrated with 100 mM potassium phosphate pH 6.5 [19]. Crystals were grown with the hanging-drop vapour diffusion technique from 2% PEG 400, 2.0M ammonium sulfate in 0.1M Na Hepes buffer, pH 7.5, and were enhanced by macro-seeding [17].

Data collection and data processing.

Since derivative crystals decayed at room temperature in the X-ray beam, we collected a native data set and several derivative data sets (Table 1) at cryogenic temperature (93K). Crystals were soaked in cryoprotectant (mother liquor plus 30% glycerol), picked up in a mounting loop and dumped into liquid nitrogen. Data were then collected on a Siemens rotating anode generator (operated at 40kV and 80mA) equipped with a Siemens multiwire area detector on a 3-circle goniometer. All frames were processed with the XDS software ([38, 39]). With this setup, a native data set could be collected to 2.4 Å and derivative data (prepared by soaking crystals for at least one day in mother liquor containing millimolar amounts of heavy atom compounds) to about 3 Å. These low-temperature rotating-anode data were used for phase determination by MIR and for map interpretation.

We also collected a native data set at the EMBL beamline X31 at Hasylab (DESY, Hamburg). Since, at the time, the X31 station was not equipped with cryocooling, the data were collected at room temperature (Table 1). The synchrotron data had better counting statistics and extended to higher resolution than the rotating anode data, and we therefore used them for most of the later stages of structure refinement. Data reduction for the SR data was done with the program DENZO [40].

Phase determination

The first mercury site could be located by manual inspection in the differnce Patterson map of the PCMB derivative, the second site was found in a difference Fourier map. All heavy atom sites from the other derivatives were determined by cross phasing with the mercury phases. The (fully occupied) positions of the Hg and the Sm derivatives are identical, whereas Pd and Pt bind to different and unique sites. DMA was observed at the Hg and Sm sites, with two heavy atoms at each of the two sites due to its chemical constitution. The parameters of the heavy atom derivatives were refined using MLPHARE [41]. For the Sm and Hg derivatives, anomalous differences were also taken into account. At this stage, the figure of merit at 3.0 Å resolution after phase refinement was 0.58. The MIR phases were enhanced by a solvent flattening procedure using the program DM, improving the figure of merit to 0.73. All crystallographic computing was carried out with the CCP4 program suite [42].

Model building and crystallographic refinement

The MIR map and the solvent flattend map clearly showed secondary structure features such as the central beta-sheet and several helices. The N-terminus was readily found by sequence alignment to the possible starting points, making use of the clear density for Trp17 and Trp19. The loop regions between residues 40 and 55 as well as between residues 110 and 190 could not be seen at this point. A map combination of solvent flattend maps at 4.0 Å, 3.0 Å and 2.4 Å (the latter calculated by phase extension with program DM) with the program COMAP [43] (unit weights for each map) eventually also revealed these regions. Model building and fitting was done with the graphic program O [44]. The initial model with 254 traced residues (out of 257) was used for a rigid- body refinement with X-PLOR [45] against the native (room temperature) data set collected at the synchrotron. A subsequent difference Fourier map immediately revealed several tracing errors, together with the majority of missing side chains. Subsequent cycles of positional refinement and manual model fitting disclosed all side chains exept those of Lys223 and Glu192. Moreover, the N-terminal methionine could not be observed, but it is likely that this residue has been cleaved off, and the density for the C-terminal asparagine is poor due to disorder. Individual B-factor refinement with X-PLOR led to the current R-factor of 19.9% (free R-factor=24.1%, [46], corr=0.92, corr_free=0.88, 20547 reflections, representing 76.7% of all reflections) with 77 water molecules and one sulfate ion using 3s (on Fobs) data between 15 and 1.9 Å resolution. The R-factor for all data between 15 and 1.9 Å was 21.6% (R_free = 25.6%, 89.7% completeness). The model displays rmds for bond lengths and angles of 0.018 Å and 1.99°, respectively. With the exception of Ser80, there are no q/y combinations in unfavourable regions of the Ramachandran plot.

The atomic coordinates of the refined structural model have been deposited with the Brookhaven Protein Data Bank.

Acknowledgements: We acknowledge support by the Österreichischer Fonds zur Förderung der wissenschaftlichen Forschung through the Spezialforschungsbereich Biokatalyse and through projects 9542 and 11599 (CK and UGW). The synchrotron data were collected at the X31 station of the Hamburg outstation of EMBL, and we acknowledge help from G. Evans. We also thank Karl Gruber, Andrea Hickel and Hans-Beat Bürgi for stimulating discussion and M. Hayn, S. Kohlwein and M. Schall for collaborative help in the early phases of the project.

References

1. Conn , E.E. (1981). In The Biochemistry of Plants: A Comprehensive Treatise. (Stumpf, P.K. & Conn, E.E., eds), Vol. 7, Secondary plant products, pp. 479-500,. Academic Press, NY.

2. Poulton, J.E. (1990). Cyanogenesis in Plants. Plant Physiology 94, 401-405.

3. Selmar, D., Lieberei, R., Conn, E.E. & Biehl, B. (1989). Alpha-Hydroxynitrile Lyase in Hevea brasiliensis and Its Significance for Rapid Cyanogenesis. Physiologia Plantarum 75, 97-101.

4. Selmar, D. (1986). Cyanogenese in Hevea - Zwei Wege zur Metabolisierung Cyanogener Glycoside. Ph.D. thesis, Braunschweig, Germany.

5. Kruse , C.G. (1992). Chiral Cyanohydrins- Their Manufacture and Utility as Chiral Building Blocks. In Chirality in Industry. (Collins, A.N., Sheldrake, G.N., & Crosby. J, eds), pp 279-299, Wiley.

6. Effenberger, F. (1994). Synthesis and Reactions of Optically-Active Cyanohydrins. Angew. Chem. Int. Ed. 33, 1555-1564.

7. Klempier, N., Griengl, H. & Hayn, M. (1993). Aliphatic (S)-cyanohydrins by enzyme catalyzed synthesis. Tetrahedron Lett. 34, 4769-4772.

8. Klempier, N., Pichler, U. & Griengl, H. (1995). Synthesis of a, b-unsaturated (S)-cyanohydrins using the oxynitrilase from Hevea brasiliensis. Tetrahedron: Asymmetry 6, 845-848.

9. Hughes, J., Decarvalho, & Hughes, M.A. (1994). Purification, characterization, and cloning of alpha-hydroxynitrile lyase from cassava (Manihot esculenta crantz). Arch. Biochem. Biophys. 311, 496-502.

10. Cheng, I.P., & Poulton, J.E. (1993). Cloning of cDNA of Prunus-Serotina (R)-(+)-Mandelonitrile Lyase and Identification of a Putative FAD-Binding Site. Plant Cell Physiol. 34, 1139-1143.

11. Kuroki, G.W., & Conn, E.E. (1989). Mandelonitrile lyase from Ximenia americana L.: stereospecificity and lack of flavin prosthetic group. Proc. Natl. Acad. Sci. USA 86, 6978-81.

12. Xu, L.L., Singh, B.K. & Conn, E.E. (1988). Purification and Characterization of Acetone Cyanohydrin Lyase from Linum usitatissimum. Arch. Biochem. Biophys. 263, 256-263.

13. Wajant, H., & Mundry, K.W. (1993). Hydroxynitrile Lyase from Sorghum-bicolor - A Glycoprotein Heterotetramer. Plant Sci. 89, 127-133.

14. Wajant, H., Förster, S., Selmar, D., Effenberger, F. & Pfizenmaier, K. (1995).

Purification and characterization of a novel (R)-mandelonitrile lyase from the fern Phlebodium aureum. Plant Physiol. 109, 1231-1238.

15. Wajant, H., Mundry, K.W. & Pfizenmaier, K. (1994). Molecular Cloning of Hydroxynitrile Lyase from Sorghum bicolor (L) - Homologies to Serine Carboxypeptidases. Plant Mol. Biol. 26, 735-746.

16. Lauble, H., Müller, K., Schindelin, H., Förster, S. & Effenberg, F. (1994). Crystallization and preliminary X-ray diffraction of mandelonitrile lyase from almonds. Proteins: Struct., Funct. Genet. 19, 343-347.

17. Wagner, U.G., et al. & Kratky, C. (1996). Crystallization and preliminary X-ray diffraction studies of a hydroxynitrile lyase from Hevea brasiliensis. Acta Crystallogr. D52, 591.

18. Schall, M. (1996). Isolation and characterization of a (S)-hydroxynitrile lyase from Hevea brasiliensis. Ph.D. thesis, University Graz, Austria.

19. Hasslacher, M., Schall, M., Hayn, M., Griengl, H., Kohlwein, S.D & Schwab, H. (1996). Molecular Cloning of the Full-length cDNA of (S)-Hydroxynitrile Lyase from Hevea brasiliensis. J. Biol. Chem. 271, 5884-5891.

20. Reimmann, C., Hofmann, C., Mauch, F. & Dudler, R. (1995). Characterization of a rice gene induced by Pseudomonas Syringae PV syringae - requirement for the bacterial lema gene-function. Physiol. Mol. Plant Pathol. 46, 71-81.

21. Laskowski, R.A., McArthur, M.W., Moss, D.S. & Thornton, J.M. (1993). PROCHECK: a program to check the stereochemical quality of protein structures. J. Appl. Crystallogr. 26, 283-291.

22. Sussman, J.L., et. al., & Silman, I. (1991). Atomic Structure of Acetyl-cholinesterase from Torpedo californica: A prototypic Acetylcholine-Binding Protein. Science 253, 872-879.

23. Ollis, D.L., et. al., & Goldman, A. (1992). The a/b hydrolase fold. Protein Eng. 5, 197-211.

24. Cygler, M., et al., & Doctor, B.P. (1993). Relationship between sequence conservation and three-dimensional structure in a large family of esterases, lipases and related proteins. Protein Sci. 2, 366-382.

25. Pathak, D. & Ollis, D. (1990). Refined structure of dienelactone hydrolase at 1.8 Å. J. Mol. Biol. 214, 497-525.

26. Endrizzi, J.A., Breddan, K. & Remington, S.J. (1994). 2.8 Å-structure of yeast serine carboxypeptidase. Biochemistry 33, 11106-11120.

27. Winkler, F.K., D'Arcy, A. & Hunziker, W. (1990). Structure of human pancreatic lipase. Nature 343, 771-774.

28. Schrag, J.D. & Cygler, M. (1993). The 1.8Å refined structure of the lipase from Geotrichum candidum. J. Mol. Biol. 230, 575-591.

29. Lawson, D.M. et al., & Derewenda, Z.S. (1994). Structure of a myristoyl-ACP-specific thioesterase from Vibrio harveyi. Biochemistry 33, 9382-9388.

30. Franken, S.M., Rozeboom, J., Kalk, K.H. & Dijkstra, B.W. (1991). Crystal structure of haloalkane dehalogenase: an enzyme to detoxify halogenated alkanes. EMBO J. 10, 1297-1302.

31. Hecht, H.J., Sobek, H., Haag, T., Peifer, O. & van Pée, K.-H. (1994). The metal-ion-free oxidoreductase from streptomyces aureofaciens has an a/b hydrolase fold. Nature struct. biol. 1, 532-537.

32. Ripoll, D.R., Faerman, C.H., Axelsen, P.H., Silman, I. & Sussman, J.L. (1993). An electrostatic mechanism for substrate guidance down the aromatic gorge of acetylcholinesterase. Proc. Natl. Acad. Sci. USA 90, 5128-5132.

33. Harel, M., et al., & Sussmann, J.L. (1993). Quaternary ligand binding to aromatic residues in the active site gorge of acetylcholinesterase. Proc. Natl. Acad. Sci. 90, 9031-9035.

34. Smith, L.C., Faustinella, F. & Chan, L. (1992). Lipases: three-dimensional structure and mechanism of action. Curr. Opin. Struct. Biol. 2, 490-496.

35. Rudenko, G., Bonten, E., d'Azzo, A. & Hol, W.G. (1995). Three-dimensional structure of the human protective protein: structure of the precursor form suggests a complex activation mechanism. Structure 3, 1249-1259.

36. Jorns, M.S. (1980). Studies of the kinetics of cyanhydrin synthesis and cleavage by the flavoenzyme oxynitrile lyase. Biochim. Biophys. Acta 613, 203-209.

37. Jorns, M.S. (1979). Mechanism of catalysis by the flavoenzyme oxynitrilase. J. Biol. Chem. 254, 12145-12152.

38. Kabsch, W. (1988). Evaluation of single crystal X-ray diffraction data from a position sensitive detector. J. Appl. Crystallogr. 21, 916-924.

39. Kabsch, W. (1988). Automatic indexing of rotation diffraction patterns. J. Appl. Crystallogr. 21, 67-71.

40. Otwinowski, Z. (1990). DENZO Data Processing Package. Yale University, CT.

41. Otwinowski, Z. (1991). Maximum likelihood refinement of heavy atom parameters. In Isomorphous Replacement and Anormalous Scattering. (Wolf, W., Evans, P.R., & Leslie, A.G.W., eds), pp 80-86. Science end Engineering Research Council, Darsbury Laboratory, Warrington, UK.

42. Bailey, S. (1994). The CCP4 suite-programs for protein crystallography. Acta Cryst. D50, 760-763.

43. Kleywegt, G.J. & Jones, T.A. (1994). Convenient single and multiple-crystal real-space averaging of macromolecular electron-density maps. To be published.

44. Jones, T.A., Zou, J.-Y., Cowan, S. & Kjeldgaard, M. (1991). Improved methods for building protein models in electron density maps and the location of errors in these models. Acta Crystallogr. A 47, 110-119.

45. Brünger, A.T. (1992). X-PLOR, a system for X-ray crystallography and NMR; version 3.2, Yale University, USA.

46. Brünger, A.T. (1992). The free R value: a novel statistical quantity for assessing the accuracy of crystal structures. Nature (London), 355, 472-474.

47. Kraulis, P.J. (1991). MOLSCRIPT: a program to produce both detailed and schematic plots of protein structures. J. Appl. Crystallogr. 24, 946-950.

48. Merritt,E.A. & Murphy,M.E.P. (1994). Raster3D, version 2.0, a program for photorealistic molecular graphics. Acta Crystallogr. D50, 869-873.

49. Nicholls, A.J. (1993). GRASP: Graphical Representation and Analysis of Surface Properties. Columbia University, New York.

Table 1 Data collection and phasing statistics
Data set	Max. resolution (Å)	Measured reflections	Unique reflections	R_merge	Number of sites	R_deriv(%)	R_Cullis (%)	Phasing power^e	used resolution (Å)
p-ClHg-phenylSO₃Na^a	2.9	20314	7297	9.0	2	37.1	63	2.16	25 - 3.0
(CH₃COO)₃Sm^a	3.0	12700	4705	11.3	2	31.6	60	2.34	25 - 3.0
K₂PtCl₄^a	2.9	21867	7314	6.8	2	28.3	79	1.23	25 - 4.0
DMA^a	3.0	11605	4567	11.2	2	32.7	94	1.17	25 - 4.0
PdCl₂^a	3.0	18705	6689	17.0	1	28.8	82	1.13	25 - 4.0
native^a	2.4	41707	11115	7.1
native^b	1.9	137615 (89.6% complete)	26663 (20547 above 3s)	7.3					15 - 1.9
^{a data collected at 95K ^{b data collected at ambient temperature ^{c R_deriv = S½FPH-FP½¤S½FP½ where FPH is the structure factor amplitude of the derivative crystal and FP is that of the native crystal. d R_Cullis = S½½FPH±FP½-½FH(calc)½½¤S½FPH-FP½, where FPH and FP are defined as above and RF(calc) is the calculated heavy atom structure factor amplitude. e Phasing power = F(H)¤E, the root mean square heavy atom structure factor amplitudes divided by the lack of closure.}}}

Figure 1: Chain fold of Hnl from Hevea brasiliensis. (a) Stereoscopic ribbon representation with the central b-sheet in red and the helices and loops of the "core" region in blue. The "cap" region is colored in green (b). Stereoscopic C_a-trace in the same orientation. (Figures generated with MOLSCRIPT [47]).

Figure 2: (a) Omit-map density (F_o-F_c map, 3 s level) of a portion of the Hnl structure to illustrate the quality of the final phases. (b) Stereoview of the electron density (F_o-F_c map, 3 s level) in the active site region, interpreted as a histidine ion (see text).

Figure 3:The dimers observed in the Hnl structure (Figure generated with MOLSCRIPT [47] and Raster 3D [48]).

Figure 4: (a) Topology of the Hnl molecule from Hevea brasiliensis. The color coding is identical to the one in Fig.1a. The upper-left insert shows the topology of the "prototypic hydrolase fold" [23]. (b) Result of a superposition of the crystal structures of Hnl from Hevea brasiliensis and AchE from Torpedo californica [22]. Plotted are the distances between corresponding C_a atoms for topologically corresponding elements of secondary structure comprising the "hydrolase fold". The labeling of a- helices and b-sheets corresponds to the one for Hnl (see above). Note that helix D has been omitted due to excessively large deviations.

Figure 5: Superposition of the structures of Hnl (white bonds) and AChE [22] (grey bonds). The figure shows the serine and histidine residues of the catalytic triade plus the residues suggested to be involved in oxyanion stabilization (Cys81 and Tyr11 in Hnl; Gly118, Gly119 and Ala201 in AChE; Figure generated with MOLSCRIPT [47]).

Figure 6: Color-coded model of the Hnl structure: the "core" region is red, the "cap" region green and the partially visible inhibitor molecule yellow. The picture shows that the entrance to the active site is formed predominantly by residues from the cap region. (Figure generated with program Sybyl, version 6.2)

Figure 7: Surface representations of Hnl ( (a) and (c) ) and AChE ( (b) and (d) ), coloured according to electrostatic potential (blue for positive, red for negative). The two upper views ((a) and (b)) are approximately along the direction of the channel to the active site, whose entrance is indicated by an arrow. Below ((c) and (d)) are two sections approximately along and bisecting the channel, showing the "inner surface" of the active-site gorge. These figures illustrate that AchE has - compared to Hnl - a more negatively charges active site which is connected to the outside by a wider channel (Figure generated with GRASP [49]).

Figure 8: The proposed steps in the Hnl - catalyzed synthesis of cyanohydrins.

Figure 9: Structure of the proposed tetrahedral reaction intermediate for the biological substrate acetone cyanhydrin. The structure was generated by modelling and subsequent energy minimization (program Sybyl, Version 6.2). The view is approximately doing the channel from the outside to the active site. (Figure drawn with MOLSCRIPT [47] and Raster 3D [48]).