The UDP-glucose:p-Hydroxymandelonitrile-O-Glucosyltransferase That Catalyzes the Last Step in Synthesis of the Cyanogenic Glucoside Dhurrin in Sorghum bicolor

The final step in the biosynthesis of the cyanogenic glucoside dhurrin in Sorghum bicolor is the transformation of the labile cyanohydrin into a stable storage form byO-glucosylation of (S)-p-hydroxymandelonitrile at the cyanohydrin function. The UDP-glucose:p-hydroxymandelonitrile-O-glucosyltransferase was isolated from etiolated seedlings of S. bicoloremploying Reactive Yellow 3 chromatography with UDP-glucose elution as the critical step. Amino acid sequencing allowed the cloning of a full-length cDNA encoding the glucosyltransferase. Among the few characterized glucosyltransferases, the deduced translation product showed highest overall identity to Zea maysflavonoid-glucosyltransferase (Bz-Mc-2 allele). The substrate specificity of the enzyme was established using isolated recombinant protein. Compared with endogenousp-hydroxymandelonitrile, mandelonitrile, benzyl alcohol, and benzoic acid were utilized at maximum rates of 78, 13, and 4%, respectively. Surprisingly, the monoterpenoid geraniol was glucosylated at a maximum rate of 11% compared withp-hydroxymandelonitrile. The picture that is emerging regarding plant glucosyltransferase substrate specificity is one of limited but extended plasticity toward metabolites of related structure. This in turn ensures that a relatively high, but finite, number of glucosyltransferases can give rise to the large number of glucosides found in plants.

flax (8), and the rubber tree (9) result in a reduced resistance to fungal attack. Unfortunately, foods containing cyanogenic glucosides have reduced nutritional value and may pose a health hazard if not properly processed (2,10). It is therefore desirable to engineer acyanogenic varieties of cyanogenic food crops, for example cassava (Manihot esculenta), given that traditional breeding methods so far have not achieved this aim (11). Conversely, it may be worthwhile to introduce the capability of cyanogenic glucoside accumulation into particular acyanogenic crop tissues in order to improve their pest or pathogen resistance.
The biosynthetic pathway of the cyanogenic glucoside dhurrin has been established in etiolated seedlings of Sorghum bicolor (Fig. 1) and is catalyzed by two membrane-bound multifunctional cytochrome P450s and an apparently soluble glucosyltransferase (reviewed in Ref. 1). The amino acid precursor L-tyrosine is N-hydroxylated twice by CYP79A1 (P450 Tyr ) forming (Z)-p-hydroxyphenylacetaldoxime, which subsequently is converted to (S)-p-hydroxymandelonitrile by CYP71E1 (P450 ox ) (Fig. 1). The S-enantiomer of the cyanohydrin is converted into a stable storage form, dhurrin, through conjugation to glucose by UDP-glucose:p-hydroxymandelonitrile-O-glucosyltransferase (12). The two first enzymes of the pathway have been isolated (13,14), their corresponding cDNAs isolated (15,16), and their function verified by heterologous expression in Escherichia coli and isolation of the recombinant enzyme (16,17).
To date there are no reports of the isolation of a cyanohydrin glucosyltransferase from a cyanogenic plant (12, 18 -20), and attempts to isolate cDNAs that encode for such enzymes have not been completed (21). Given the lability of p-hydroxymandelonitrile, the massive accumulation of dhurrin (22), and the absence of multiple p-hydroxymandelonitrile-O-glucosyltransferase activities in S. bicolor (12), it is most likely that a specific glucosyltransferase is responsible for the synthesis of the cyanogenic glucoside dhurrin, similar to that proposed for anthocyanin accumulation (23,34). Isolation of the cDNA encoding that specific glucosyltransferase would therefore improve the possibility to introduce the synthesis of cyanogenic glucosides into acyanogenic plants and hence to evaluate the biological role of that metabolic process. The availability of cDNAs and antibodies toward all three enzymes will enable in depth studies of (a) the biology of cyanogenic glucoside metabolism in planta, (b) the ecological role of cyanogenic glucosides, and (c) the possible association between cytochrome P450s and glucosyltransferases involved in secondary plant metabolism (24,25).
A large number of potential aglycone substrates can be present in any given plant, for example berries of grapevine (Vitis vinifera) contain over 200 different forms of glycosides (26,27).
Novel xenobiotic substances are also glucosylated by many plant species (28). Plants therefore have a large capability to glucosylate a wide range of different chemical structures. However, the number of glucosyltransferases present in plants and the range of substrate specificities exhibited by these are largely unknown. Whereas numerous glucosyltransferases have been partially purified, few have been characterized in an isolated state. Such earlier studies have indicated that both narrow and broad substrate specificities can be found (20, 29 -33), although the difficulty associated with the separation of glucosyltransferases with similar chromatographic properties has confused the picture (34). The multiplicity of glucosyltransferases can be assessed from the Arabidopsis thaliana genomic sequencing project. Eighteen putative secondary plant metabolism glucosyltransferase-encoding genes have been identified in the 28% of the genome sequenced to date, 1 implicating that around 65 genes may be present in the complete genome assuming an equal distribution. Altogether, over 100 different, putative, secondary plant metabolism glucosyltransferase-encoding cDNAs are available in international data banks. Only in few cases has the protein encoded by these gene sequences been verified (35)(36)(37)(38)(39), and only in two instances have authors (34,40) attempted to investigate the complete specificity of the recombinant protein. In order to gain further insight into the biology of secondary plant metabolism glucosyltransferases, it is therefore necessary to verify functionally the identity of the proteins encoded by these cDNAs by (a) heterologous expression and (b) conducting assays with aglycones of diverse chemical structures.
In the present study we report the isolation, cloning, functional heterologous expression, and characterization of the substrate specificity of a p-hydroxymandelonitrile-glucosyltransferase from S. bicolor. The isolation of this protein and the corresponding cDNA completes the biosynthetic pathway of cyanogenic glucoside biosynthesis.

EXPERIMENTAL PROCEDURES
Biochemicals and Reagents-All biochemicals were of analytical or higher grade. Substrates and authentic glucosides were obtained from Sigma and Extrasynthèse (Genay, France). Dye reagents were obtained from Amersham Pharmacia Biotech and Sigma.
Plant Materials-S. bicolor seeds were obtained from Pacific Seeds, Australia (cultivar MR31).
General Methods-Protein preparations were concentrated using a Speed Vac Concentrator (Savant) prior to electrophoresis. SDS-PAGE 2 was performed using high-Tris linear 8 -25% SDS-polyacrylamide gradient gels (Mini-Protean II, Bio-Rad) (41), and polypeptides were visualized by staining with Coomassie Brilliant Blue (R-250). DNA sequencing reactions were carried out using the Thermo Sequenase fluorescentlabeled primer cycle sequencing kit (7-deaza dGTP) (Amersham Pharmacia Biotech) and analyzed using an ALFexpress DNA Sequencer (Amersham Pharmacia Biotech). Sequence computer analysis was performed using programs in the Genetic Computer Group (Madison, WI) sequence analysis package and NCBI BLAST (42).
Enzyme Assays-General reaction mixtures (total volume 20 l) included 100 mM Tris⅐HCl (pH 7.9), 1-5 M [ 14 C]UDP-glucose (11.0 GBq⅐mmol Ϫ1 , Amersham Pharmacia Biotech), 0 -200 M UDP-glucose, 20 mM aglycone dissolved in water, 25 mM ␥-gluconolactone (␤-glucosidase inhibitor), and 0.5-10 l of protein preparation. At the end of the incubation period (10 min, 30°C), 2 l of 10% acetic acid were added to terminate the reaction. BSA (1 mg/ml) was included in assays for the assessment of sbHMNGT yield throughout the purification procedure. Qualitative analyses of recombinant sbHMNGT were performed as outlined above except for incubation period (20 min) and concentrations of reagents as follows: 1.25 mM aglycone (dissolved in ethanol except for flavonoids that were dissolved in ethylene glycol monoether), 1.25 M Reaction mixtures for analysis by NMR spectroscopy (total volume 0.5-1 ml) included 2 mM p-hydroxymandelonitrile or 6.5 mM geraniol, 3 mM UDP-glucose, 2.5 g recombinant sbHMNGT, 0.5 mg of BSA. After incubation (2 h), glucosides were extracted with ethyl acetate and lyophilized in speedy-vac prior to NMR analysis. For TLC, the reaction mixture was applied to Silica Gel 60 F254 plates (Merck), dried, and developed in a solvent containing ethyl acetate:acetone:dichloromethane:methanol:H 2 O (40:30:12:10:8, v/v) for 1 h. Plates were dried (1 h, room temperature) and exposed to storage phosphorimaging screens (Molecular Dynamics, Sunnyvale, CA) prior to scanning on a Storm 860 PhosphorImager (Molecular Dynamics). For LSC, reaction mixtures were extracted with 400 l of ethyl acetate to separate glucosides from unincorporated [ 14 C]UDP-glucose. Two ml of Ecoscint A (National Diagnostics, NJ) were added to 250 l of each ethyl acetate extract and analyzed using a liquid scintillation counter. Mandelonitrile was used as substrate to assay fractions generated by liquid chromatography.
Purification-All procedures were carried out at 4°C except where indicated. S. bicolor seeds (1 kg) were imbibed in water overnight at room temperature and grown at 30°C in darkness for 2 days (22). Seedling shoots were harvested and extracted in 2 volumes of ice-cold extraction buffer (250 mM sucrose; 100 mM Tris⅐HCl (pH 7.5); 50 mM NaCl; 2 mM EDTA; 5% (w/v) of polyvinylpolypyrrolidone; 200 M phenylmethylsulfonyl fluoride; 6 mM DTT) using mortar and pestle. The extract was filtered through a nylon mesh prior to centrifugation (20,000 ϫ g, 20 min). The supernatant fraction was subjected to differential ammonium sulfate fractionation (35-70%). The pellet was resus-pended in buffer A (20 mM Tris⅐HCl (pH 7.5); 5 mM DTT) and desalted using a Sephadex G-25 (Amersham Pharmacia Biotech) or Bio-Gel P-6 (Bio-Rad) column (2.5 ϫ 20 cm, flow rate: 20 ml/min) equilibrated in buffer A. The first UV-absorbing peak was collected and applied to a Q-Sepharose (Amersham Pharmacia Biotech) column (2.6 ϫ 23 cm, flow rate: 60 -80 ml/h) equilibrated in buffer B (buffer A ϩ 50 mM NaCl). The column was washed with buffer B until the base line had stabilized, and proteins were eluted with a linear gradient from 50 to 400 mM NaCl in buffer A (800 ml total). Fractions (10 ml) were collected and 3-5 l assayed for mandelonitrile and/or p-hydroxymandelonitrile glucosyltransferase activity by LSC. To reduce the salt concentration, combined active fractions (50 mg of protein, 20 ml) were diluted 5-fold in buffer B and concentrated 20-fold using an Amicon YM30 or YM10 membrane prior to storage at Ϫ80°C.
The remainder of the purification was carried out at room temperature. One-quarter of the concentrated material from the Q-Sepharose step (ϳ10 -15 mg protein, 5 ml) was applied to a column (1 ϫ 10 cm, flow rate: 10 -15 ml/hr) containing Reactive Yellow 3 cross-linked onto 4% beaded agarose (Lot 63H9502) (Sigma) equilibrated in buffer B. The column was washed with buffer B until the base line had stabilized. Proteins were eluted with 10 ml of 2 mM UDP-glucose in buffer B. Active fractions containing essentially pure sbHMNGT were combined and stored at Ϫ80°C with or without addition of 1 mg/ml BSA.
Peptide Generation and Sequencing-sbHMNGT (ϳ5 g) was subjected to N-terminal sequencing using an automated protein sequencer. For peptide digestion, sbHMNGT (ϳ100 g) was precipitated by addition of trichloroacetic acid (10% w/v final concentration), resuspended in 50 l of 50 mM Tris⅐HCl (pH 8.0), 5 mM DTT, and 6.4 M urea, incubated (60°C, 50 min), cooled to room temperature, and diluted with 3 volumes of 30 mM Tris (pH 7.7) and 1.25 mM EDTA. Endoproteinase Lys-C (Promega, Madison, WI) was added (proteinase:substrate ratio 1:25 (w/w)) and the reaction allowed to proceed for 24 h at 37°C. Peptides were purified with Beckman System Gold high performance liquid chromatography equipment fitted with a Vydac 208TP52 C8 column (2.1 ϫ 250 mm, flow rate: 0.2 ml/min). Peptides were applied in buffer C (0.1% trifluoroacetic acid) and eluted with a linear gradient from 0 to 80% acetonitrile in buffer C. Peptides were collected manually and sequenced as described above.
PCR Amplification, Cloning, and Library Screening-First round PCR amplification reactions (total volume: 40 l) were carried out using 2 units of Taq DNA polymerase (Amersham Pharmacia Biotech), 4 l of 10ϫ Taq DNA polymerase buffer, 5% (v/v) dimethyl sulfoxide, 1 l of dNTPs (10 mM), 80 pmol each of primers C2EF (5Ј-TTYGTIWSICAYT-GYGGITGGAA-3Ј) and T7 (5Ј-AATACGACTCACTATAG-3Ј), and ϳ10 ng of plasmid DNA template. The plasmid DNA template was prepared from a unidirectional pcDNAII (Invitrogen, Carlsbad, CA) plasmid library made from 1 to 2 cm high etiolated S. bicolor seedlings (16). Thermal cycling parameters were 95°C, 5 min, 3 times (95°C for 5 s, 42°C for 30 s, 72°C for 30 s), 32 times (95°C for 5 s, 50°C for 30 s, 72°C for 30 s), and a final 72°C for 5 min. Second round PCR amplifications were carried out as above, except using primers C2DF (5Ј-GARGCIACIGCIGCIGGICARCC-3Ј) and T7, and 1 l of first round reaction as DNA template. Thermal cycling parameters were 95°C, 5 min, 32 times (95°C for 5 s, 55°C for 30 s, 72°C for 30 s) and a final 72°C for 5 min. The PCR reaction mixtures were subjected to gel electrophoresis using a 1.5% agarose gel, and an ϳ600-bp band was excised and cleaned using a Qiaex II gel extraction kit (Qiagen). The cleaned PCR product was then ligated into the pGEM-T vector and used to transform the E. coli JM109 strain according to the manufacturer's instructions (Promega). The nucleic acid sequence of PCR clone 1544 encoded peptide sequences obtained in the purified sbHMNGT.
The PCR clone 1544 was employed as template to generate a 306-bp digoxigenin-11-dUTP-labeled probe by PCR using primers 441F (5Ј-GAGGCGACGGCGGCGGGGCAG-3Ј) and 442R (5Ј-CATGTCACTGCT-TGCCCCCGACCA-3Ј) according to the manufacturer's instructions (Roche Molecular Biochemicals). The labeled probe was cleaned using the Qiaex II gel extraction kit after 1.5% agarose gel electrophoresis and employed to screen approximately 50,000 colonies of the above mentioned plasmid library. Hybridizations were carried out overnight at 65°C in 5ϫ SSC, 0.1% (w/v) N-lauroylsarcosine, 0.02% (w/v) SDS, and 1% blocking reagent (Roche Molecular Biochemicals). Membranes were washed (3 times for 15 min) in 0.5ϫ SSC at 60°C. Seven hybridizing clones were isolated and one full-length clone, sbHMNGT, was chosen for further characterization.
Heterologous Expression-Primers EXF1 (5Ј-AATAAAAGCATATG-GGAAGCAACGCGCCGCC TCCG-3Ј) and EXR1 (5Ј-TTGGATCCTCA-CTGCTTGCCCCCGACCA-3Ј) were employed to amplify a 1500-bp fulllength sbHMNGT insert by PCR, using sbHMNGT plasmid as template. The primers contained 5Ј recognition sites for restriction endonucleases NdeI (EXF1) and BamHI (EXR1). PCR reaction conditions were essentially as above, except for thermal cycling parameters: 95°C, 3 min, 30 times (95°C for 5 s, 53°C for 30 s, 72°C for 90 s) and a final 72°C for 5 min. The PCR product was gel-purified, digested with NdeI and BamHI, gel-purified once again, and ligated into the plasmid expression vector pSP19 g10L (kindly supplied by Dr. Henry Barnes) (17) which also had been digested with the same restriction enzymes and gel-purified. The ligation reaction mixture was then used to transform E. coli JM109 cells according to manufacturer's instructions (Promega). After selection of successfully cloned cells, expression was initiated as per Ref. 34. Briefly, 600 l of an overnight 37°C culture was added to 300 ml of Luria broth (LB) containing ampicillin (100 g/ml). The culture was allowed to grow at 28°C (150 rpm) for 5 h, and isopropyl-1-thio-␤-D-galactopyranoside was then added to a final concentration of 0.4 mM. After induction, the culture was allowed to continue growing overnight and harvested by centrifugation (2500 ϫ g, 10 min). The pellet was resuspended in 9 ml of 200 mM Tris (pH 7.9), 1 mM EDTA, 5 mM DTT, and 0.1 mg/ml lysozyme. An equal volume of ice-cold water was added, and the mixture was allowed to incubate (10 min at room temperature, 20 min on ice). After the addition of 18 mol of phenylmethylsulfonyl fluoride and 100 units of DNase I/ml (Sigma), the suspension was subjected to three freeze and thaw cycles at Ϫ20°C. Phenylmethylsulfonyl fluoride was adjusted to 1.5 mM final concentration, and the preparation was centrifuged at 15,000 ϫ g for 15 min. Negative controls, containing no insert in the plasmid vector, were prepared as above.
The recombinant protein was isolated as the native protein using two 300-ml cultures lysed as above as starting material. The yield of recombinant protein varied between 0.1 and 1 mg/100 ml LB culture.

RESULTS
Yield and Stability of sbHMNGT-In preparation for sbHMNGT purification, sorghum seeds were germinated in darkness for 1.5-5 days, and extracts made from seedlings were tested for sbHMNGT activity. Under the conditions of growth, a 2-day germination period proved optimal with regard to total sbHMNGT activity, protein concentration, and extract volume (data not shown). The use of a Waring blender resulted in less than 50% of the activity as compared with extraction with mortar and pestle. sbHMNGT activity was largely unaffected by both freezing at Ϫ80°C and the addition of glycerol. The addition of relatively high concentrations of DTT were required to retain activity. Thus, lowering the concentration of DTT from 5 to 2 mM resulted in a 10-fold decrease in activity after storage at 4°C for 2 days. This pronounced effect of DTT was primarily found in crude preparations, whereas partially purified ion-exchange preparations were less responsive to the concentration of reducing agents in contrast to previous results (12).
Isolation of sbHMNGT-Mandelonitrile was employed as a substrate for the assay of sbHMNGT activity throughout purification, although the endogenous substrate of sbHMNGT is p-hydroxymandelonitrile. Previously, mandelonitrile had been shown to be an equally good substrate (12). Furthermore, the absence of a hydroxyl group at the para-position of the benzene ring ruled out the possibility of p-glucosyloxymandelonitrile synthesis, which would be indistinguishable from dhurrin when the convenient assay based on LSC was employed. Etiolated seedlings of S. bicolor were extracted with mortar and pestle, and the protein preparation was subjected to ammonium sulfate fractionation and desalted by gel chromatography. Whereas there was no measurable increase in the specific activity of sbHMNGT, low molecular weight solutes (including cyanide and cyanide precursors) were effectively removed. All sbHMNGT activity bound to Q-Sepharose and was eluted between 150 and 200 mM NaCl with an ϳ7-fold purification. Aliquots of combined fractions were stored at Ϫ80°C after desalting and concentrating.
Several pseudoaffinity reagents were subsequently tried in mini-column format, including Cibacron Blue 3G, Reactive Green 19, Reactive Yellow 3, and UDP-glucoronic acid crosslinked with 4% beaded agarose. Trials with elution using NaCl and UDP-glucose at varying salt concentrations identified Reactive Yellow 3 as the superior column material. sbHMNGT activity bound to Reactive Yellow 3 at 50 mM NaCl and could be eluted after washing with a slight increase in NaCl concentration, without any measurable UV absorbance in the eluate. sbHMNGT activity correlated with a polypeptide migrating around 50 -55 kDa by SDS-PAGE, but additional polypeptides were also present (data not shown). Elution with 2 mM UDPglucose instead of NaCl resulted in the elution of a similarly migrating polypeptide in apparent homogeneity ( Fig. 2A). Assuming that all of the polypeptide that was visualized by SDS-PAGE was active (and therefore that all inactive protein had been lost), sbHMNGT represented approximately 0.25% of the total protein in the ammonium sulfate extract and was purified 420-fold with a maximum yield of 22% ( Fig. 2A).
N-terminal sequencing of approximately 100 pmol of isolated sbHMNGT yielded phenylthiohydantoin-derivatives at a level 10 times lower than expected (sequence 1, Fig. 3A). The low response suggested either partial blockage of the N terminus or the presence of a co-migrating and fully blocked contaminant. The latter possibility was ruled out because protein digestion by endoproteinase Lys C yielded peptides (sequences 2-5, Fig.  3A) with sequences all contained in a single cDNA (see below). Only sequence 4 exhibited high similarity to other known and putative glucosyltransferases (Fig. 3A).
Cloning of Full-length sbHMNGT-Degenerate oligonucleotides derived from peptide sequence 4 and a plasmid T7 primer were employed in nested PCR reactions using a unidirectional S. bicolor seedling plasmid library as template. An ϳ600-bp PCR fragment representing the C-terminal portion of sb-HMNGT was isolated and shown to encode peptide sequences 4 and 5.
The partial PCR fragment was then employed to screen the library. Approximately 50,000 clones were screened, resulting in seven positive isolates, of which 4 were full-length. Preliminary sequencing indicated that they all represented an identical gene. One clone, sbHMNGT, was chosen for further study. The nucleic acid sequence of the sbHMNGT-encoding cDNA has been deposited in GenBank TM with the accession number XYYYYY.
The deduced translation product comprises 492 amino acid residues and has a predicted molecular mass of 52.9 kDa (Fig.  3A) and a theoretical pI of 5.3. Searching the PROSITE sequence motif data base revealed no extended stretches of identical sequence, except for a UDP-glucosyltransferase signature sequence between residues 368 and 411. Known and putative plant secondary metabolism glucosyltransferases in general exhibit a very low degree of overall similarity, with the exception of the C-terminal part which contains a postulated UDPglucose binding motif (Fig. 3B). Based on this observation the C-terminal region is thought to encode the UDP-glucose binding domain (21,44,45,57), whereas the N-terminal end of the protein may be responsible for binding the divergent and structurally dissimilar substrate aglycones. A comparison between sbHMNGT and a range of known and putative plant glucosyltransferases (listed in Fig. 3B) revealed that sbHMNGT shares highest overall identity (41.6%) and similarity (51.5%) with a partial putative glucosyltransferase-encoding cDNA from Pisum sativum (Table I). Among the few well characterized glucosyltransferases, sbHMNGT shared highest overall identity (36.7%) and similarity (41.5%) with Zea mays flavonoid-glucosyltransferase.
Functional Expression in E. coli and Characterization of Substrate Specificity-Active soluble recombinant sbHMNGT was synthesised in E. coli as described under "Experimental Procedures." The enzyme was isolated using the same procedure for the native protein (Fig. 2B), implicating that the recombinant and native protein shared those physical features that were necessary for this highly selective purification system.
The qualitative substrate specificity of recombinant sb-HMNGT was compared with the complement of glucosyltransferase activities present in crude desalted extracts of etiolated S. bicolor seedlings (Table II). Fifteen of the 22 substrates tested were glucosylated by the crude extract, whereas only 6 of these substrates were accepted by the recombinant enzyme. Hydroquinone(1,4-benzenediol) and p-hydroxybenzaldehyde were not glucosylated by sbHMNGT although both compounds reported to serve as substrates when non-homogenous preparations were used (12). In addition to mandelonitrile and phydroxymandelonitrile, benzyl alcohol, benzoic acid, and the  Table II include Cassava (M. esculenta putative glycosyltransferases, EMBL accession numbers X77459, X77459, X77460, X77461, X77462, and X77463), Gentiana triflora (Swiss-Prot Q96493), and Nicotiana tabacum (GenBank TM accession numbers AB000623 and U32644). lization between the differing aglycones. It is not possible to determine accurately the comparative K m values for the different substrates of sbHMNGT since p-hydroxymandelonitrile rapidly dissociates in aqueous solutions to p-hydroxybenzaldehyde and cyanide, forming an equilibrium that is dependent on the initial concentration of the substrate (12). Furthermore, p-hydroxybenzaldehyde and possibly p-hydroxymandelonitrile, dhurrin, mandelonitrile, sambunigrin, and benzaldehyde are potent concentration-dependent inhibitors of isolated sb-HMNGT, although this effect was negated by the inclusion of BSA (1 mg/ml) in the reaction mixture (data not shown). Accordingly, addition of BSA to the reaction mixtures rendered it possible to estimate V max values for these different substrates. Such quantitative measurements demonstrate a greater preference for p-hydroxymandelonitrile, the endogenous substrate, compared with mandelonitrile. This is in contrast to previous results obtained using nonhomogeneous preparations (12) (Table II). The other three non-cyanogenic substrates were only utilized at less than a fifth of the maximal rate observed with the cyanohydrins. However, the acceptance of benzyl alcohol and benzoic acid as substrates indicates that sbHMNGT is only partially specific for the nitrile group and that the stereochemistry and/or interactive chemistry of the additional groups present on the hydroxyl-bearing carbon also influence sbHMNGT acceptance. Acetone cyanohydrin, the non-aromatic precursor of the cyanogenic glucoside linamarin present in cassava (M. esculenta), was not glucosylated by sbHMNGT. This suggested that sbHMNGT is exclusively specific for the presence of a benzyl group. The acceptance of geraniol at rates comparable to benzyl alcohol was therefore surprising (Table II). Control reactions with crude extracts of E. coli transformed with the expression vector (pSP19 g10L) carrying no insert showed no evidence of geraniol conjugation (data not shown).

DISCUSSION
Thousands of glycosylated secondary plant metabolites exist and yet only recently have a handful of enzymes responsible for glycosidic transfer been purified and characterized in some detail. Glycosylation is the terminal step in the biosynthesis of cyanogenic glucosides, an important class of metabolites present in several major food crops (2). Significant progress has been made toward understanding the enzymology and molecular biology of the synthesis of dhurrin, the cyanogenic gluco-side present in S. bicolor. Although two cytochrome P450 enzymes responsible for the conversion of tyrosine into p-hydroxymandelonitrile have been isolated and their cDNAs cloned, the enzyme crucial to the stable accumulation of dhurrin, namely sbHMNGT, has like many other glucosyltransferases proven very difficult to identify at the molecular level (12, 18 -21). The difficulty associated with the isolation of plant secondary metabolite glucosyltransferases in general has been attributed to the apparent lability (46) and low concentration of these proteins (47). S. bicolor was therefore chosen as a model plant since young etiolated seedlings are capable of de novo synthesis of dhurrin up to a level corresponding to 30% of the dry weight matter (22). Furthermore, the two enzymes in the biosynthetic pathway that precede sbHMNGT, CYP79A1 (13) and CYP71E1 (14) (Fig. 1), had earlier been isolated from the same source. The critical step in sbHMNGT isolation was Reactive Yellow 3 dye chromatography in association with UDP-glucose elution, which resulted in more than a 100-fold purification in a single step. Recently, Reactive Yellow 3 with UDP-glucose elution was also employed in the purification of UDP-glucose:betanidin-6-O-glucosyltransferase with similar results as in the present study (48). The use of pseudo-affinity dye chromatography together with substrate-specific elution is now emerging as a highly selective key step in plant glucosyltransferase purification (47)(48)(49)(50)(51). This significantly shortens the purification procedure and minimizes the time-dependent loss of activity seen in earlier glucosyltransferase purification protocols (29,46,(52)(53)(54)(55)(56). The importance of optimized buffer conditions was highlighted by the strong effect of DTT concentration on sbHMNGT stability in crude extracts. The differential effect of DTT on the maintenance of a reduced environment at different stages throughout purification may be explained by a higher concentration of low molecular weight radicals and oxidative enzymes in the crude plant extracts. The mechanism by which sbHMNGT activity is affected remains unknown; however, it is interesting to note that the addition of DTT to purified Z. mays indole/acetic acid-glucosyltransferase inhibited the formation of inactive multimers (49). The strong inhibitory effect of p-hydroxybenzaldehyde on sbHMNGT activity was surprising, given that it is a degradation product of the acceptor substrate, p-hydroxymandelonitrile. The addition of BSA to purified sbHMNGT significantly enhanced the apparent activity. Whether this effect is non-selectively chemical or due to allosteric regulation is not known. p-Hydroxybenzaldehyde would only be present in high concentration at the site of synthesis, if p-hydroxymandelonitrile was released into the cytosol prior to conjugation by sbHMNGT. The consequence of such an in vitro effect is further argument for an intimate association between sbHMNGT and CYP71E1 (24), which would minimize any formation of tyrosine-derived p-hydroxybenzaldehyde. However, this effect may not be relevant in vivo since (a) local concentrations of p-hydroxymandelonitrile may be much less than in vitro (1-5 mM), and (b) the presence of other proteins may diminish the inhibition in a similar manner to BSA.
The sequences of glucosyltransferase-encoding cDNAs exhibit only moderate positional identities, and these are largely confined to discrete blocks specifying the C-terminal regions of the encoded enzymes. This block of relatively well conserved sequence in glucosyltransferases most likely represents a common UDP-glucose binding domain (21,39,57). At present it is not possible to predict glucosyltransferase acceptor substrate specificity from amino acid sequence, as no determinant residues or regions of residues have been established. A larger set of functionally verified glucosyltransferase-encoding cDNAs are required to further our understanding on this matter,    which may not be confirmed until the three-dimensional structure of a secondary plant metabolism glucosyltransferase has been presented. sbHMNGT shares highest degree of overall identity with a putative glucosyltransferase deduced from an unknown clone from P. sativum, a Z. mays flavonoid-glucosyltransferase (43), and a Z. mays indole/acetic acid-glucosyltransferase (39) (Table I). This illustrates that the large number of putative glucosyltransferase-encoding cDNAs tabulated in the nucleic acid data banks, which have been labeled as flavonoid or indole/acetic acid-glucosyltransferase homologues, may be expected to glucosylate a range of other substrates in vivo and thus to be functionally mislabeled at present. The strong sequence and structural similarity exhibited between the glucosinolate-degrading myrosinases and a cyanogenic ␤-glucosidase from Trifolium repens (58), and the presence of CYP79 homologues in glucosinolate-producing plants (59), suggest that there is a strong evolutionary link between cyanogenic glucosides and glucosinolates. It was therefore surprising to find that the deduced sequence of a thiohydroximate-glucosyltransferase from Brassica napus showed only moderate overall identity to sbHMNGT. Similarly, none of the putative glucosyltransferase-encoding cDNAs isolated from cassava (M. esculenta) (21) shared any strong identity with sbHMNGT. However, sbHMNGT exhibited no activity toward acetone cyanohydrin, the main cyanohydrin aglycone present in cassava. Given that the monocotyledonous maize flavonoid-glucosyltransferase and sbHMNGT both utilize aglycones with benzyl groups, this may simply indicate a stronger co-evolution between sbHMNGT and flavonoid glucosyltransferases than between cyanohydrin-glucosyltransferases of different species. Alternatively, none of the clones isolated by Hughes et al. (21) encode for an acetone cyanohydrin:glucosyltransferase.
Investigations into the quantitative and qualitative substrate specificity of recombinant sbHMNGT showed a strong preference for the cyanohydrin present in S. bicolor. Similarly, when recombinant V. vinifera anthocyanidin-glucosyltransferase was assayed against a wide range of different aglycones, it was found to be strictly specific for flavonols and anthocyanidins only, with a strong preference for the latter (34). Both sbHMNGT and V. vinifera anthocyanidin-glucosyltransferase are involved in the final stages of predominant secondary metabolite biosynthetic pathways. Their presence is necessary for the highly tissue-and development-specific accumulation of their respective glucosides (12,(22)(23). A possible scenario then is that the sole in vivo function of these enzymes is related to the glucosylation of unique and single secondary metabolites. It is, nevertheless, likely that enzymes present at the end of biosynthetic pathways have a broader substrate specificity than those preceding upstream, if there is to be any flexibility with respect to the evolution of novel secondary metabolite biosynthesis and xenobiotic catabolism. This is illustrated by the finding that CYP71E1 and sbHMNGT also accept phenylalanine-derived oximes and cyanohydrins (mandelonitrile), respectively, whereas the first enzyme of the pathway, CYP79A1, is exclusive for tyrosine ( Fig. 1) (60).
The stereochemistry of the cyanohydrin function is very important for substrate recognition. In sorghum the enzyme is stereospecific for the S-enantiomer of p-hydroxymandelonitrile (12). The wrong stereochemistry at the chiral carbon atom carrying the cyanohydrin function prevents acceptance of the nitrile group in the active site. On the other hand, the presence of a nitrile group is not necessarily required for substrate recognition by sbHMNGT, as demonstrated by the glucosylation of benzyl alcohol and benzoic acid, although at significantly lower rates compared with mandelonitrile. The above results indicate that sbHMNGT has high specificity for sub-strates that are closely similar to mandelonitrile, given that aglycones with only slight differences in chemical structure, such as hydroquinone, gentisic acid, and acetone cyanohydrin, do not serve as acceptor substrates. It was, therefore, surprising to find that sbHMNGT also conjugated the monoterpenoid geraniol, with equal efficiency compared with benzyl alcohol. To date there are no reports of the isolation of a monoterpenoid glucosyltransferase, despite the obvious importance of this enzyme class in relation to the aroma of processed fruits and vegetables (61). Initial tertiary structural modeling indicates that this unexpected and broadened specificity may be explained by the similarity of geraniol to benzyl alcohol in particular configurations.
The extensive characterization of sbHMNGT, V. vinifera anthocyanidin-glucosyltransferase (34), and a tobacco phenylpropanoid-glucosyltransferase (40), now allows us to address the question of whether glucosylation of the multitude of secondary plant metabolites results from the action of a relatively small number of highly promiscuous enzymes with broad substrate specificity or, at the other extreme, a large number of glucosyltransferases with a tight substrate specificity. The picture that is emerging, at least in vitro, is an intermediate situation. The in-depth characterization of the three glucosyltransferases reveals that a finite number of glucosyltransferases with some, but not very extended, plasticity toward structurally similar secondary metabolites exist. The effective substrate specificity can be further tightened in vivo, through the generation of only a specific set of aglycones. For example, if p-hydroxymandelonitrile and not geraniol is formed in etiolated seedlings of S. bicolor, then sbHMNGT exhibits tight specificity for the former metabolite. Alternatively, if a multitude of secondary metabolites, all of which can act as substrates for a particular glucosyltransferase, are present simultaneously, then it is possible that glucosyltransferase promiscuity is exhibited in vivo. The consequences of modulating glucosyltransferase activity, in planta, then becomes an intriguing one, and it remains to be seen whether the metabolism of a single class of metabolites can be influenced in vivo through the modulation of a specific glucosyltransferase expression. The tools and experimental systems available in the area of cyanogenic glucoside metabolism now allows us to address this important question. Hence, the availability of cDNAs encoding for all three enzymes of the cyanogenic glucoside biosynthetic pathway will now enable the preparation of transgenic plants, of acyanogenic cultivars, in which the capability to synthesize dhurrin has been introduced. This, together with antibodies directed toward these enzymes, will now permit in-depth studies of the biological role of this important class of secondary plant metabolites to take place.