GDP-Mannose 3′,5′-Epimerase Forms GDP-L-gulose, a Putative Intermediate for the de Novo Biosynthesis of Vitamin C in Plants*

Despite its importance for agriculture, bioindustry, and nutrition, the fundamental process of l-ascorbic acid (vitamin C) biosynthesis in plants is not completely elucidated, and little is known about its regulation. The recently identified GDP-Man 3′,5′-epimerase catalyzes a reversible epimerization of GDP-d-mannose that precedes the committed step in the biosynthesis of vitamin C, resulting in the hydrolysis of the highly energetic glycosyl-pyrophosphoryl linkage. Here, we characterize the native and recombinant GDP-Man 3′,5′-epimerase of Arabidopsis thaliana. GDP and GDP-d-glucose are potent competitive inhibitors of the enzyme, whereas GDP-l-fucose gives a complex type of inhibition. The epimerase contains a modified version of the NAD binding motif and is inhibited by NAD(P)H and stimulated by NAD(P)+. A feedback inhibition of vitamin C biosynthesis is observed apparently at the level of GDP-Man 3′,5′-epimerase. The epimerase catalyzes at least two distinct epimerization reactions and releases, besides the well known GDP-l-galactose, a novel intermediate: GDP-l-gulose. The yield of the epimerization varies and seems to depend on the molecular form of the enzyme. Both recombinant and native enzymes co-purified with a Hsp70 heat-shock protein (Escherichia coli DnaK and A. thaliana Hsc70.3, respectively). We speculate, therefore, that the Hsp70 molecular chaperones might be involved in folding and/or regulation of the epimerase. In summary, the plant epimerase undergoes a complex regulation and could control the carbon flux into the vitamin C pathway in response to the redox state of the cell, stress conditions, and GDP-sugar demand for the cell wall/glycoprotein biosynthesis. Exogenous l-gulose and l-gulono-1,4-lactone serve as direct precursors of l-ascorbic acid in plant cells. We propose an l-gulose pathway for the de novo biosynthesis of vitamin C in plants.

Vitamin C (L-ascorbic acid (L-AA) 1 ) acts as an enzyme cofactor and an antioxidant. In plants it may represent one of the major soluble carbohydrates and is involved in crucial physiological processes such as biosynthesis of the cell wall, phyto-hormones, and secondary metabolites, cell division and growth, and stress resistance and photoprotection (for review, see Ref. 1). Large variations in vitamin C content (from 0.003 to 1% of fresh weight; w/w), reported for different plant species, organs, and tissues (2), are intimately linked to the vitamin biosynthesis, stability, and function. Plants, algae, and the majority of animals are able to synthesize vitamin C. Humans, however, lack L-gulono-1,4-lactone oxidase, the last enzyme of the vitamin C pathway in animals, and require L-AA as an essential micronutrient. L-AA biosynthetic genes can be engineered to increase vitamin C content of plants (3)(4), in view of improving the nutritional value and stress resistance of crops, but also potentially exploited for the industrial production of vitamin C (5).
The biosynthesis of vitamin C in plants is not completely elucidated, and its regulation is largely unknown. Two distinct pathways for vitamin C biosynthesis in plants were proposed (6 -7). The salvage pathway involves pectin-derived D-galacturonic acid (6) that is reduced at C1 to L-galactonic acid by the recently identified D-galacturonic acid reductase (4), and the resulting L-galactono-1,4-lactone is oxidized to L-AA by the mitochondrial L-galactono-1,4-lactone dehydrogenase (8 -9). Conversion of D-galacturonic acid into L-galactonic acid results in the inversion of carbon numbering. However, labeling studies of Loewus and Kelly (10) indicate that a non-inversion pathway, in which a hexose is converted into L-AA without reversion of the carbon chain, predominates in plants. The second pathway (7) is a non-inversion, energy-dependent biosynthesis that involves the conversion of GDP-D-mannose to GDP-L-galactose catalyzed by a GDP-D-Man 3Ј,5Ј-epimerase (11). L-Galactose, released from the nucleotide through some poorly understood steps, is then oxidized at C1 to L-galactono-1,4-lactone by an L-galactose dehydrogenase (12); the latter compound is converted to vitamin C by the L-galactono-1,4lactone dehydrogenase.
Recently, we obtained a highly purified GDP-Man 3Ј,5Ј-epimerase preparation from Arabidopsis thaliana cell suspensions and identified the corresponding gene (11). Only one copy of the epimerase gene is present in the Arabidopsis genome, and the gene is highly conserved among plant species (Ͼ88% identity at the protein level and Ͼ78% identity at the DNA level). A data base search revealed the presence of the epimerase sequence in cDNA libraries obtained from mature tomato fruits and potato tubers but also from salt-and pathogen-stressed ice plant and potato leaves, respectively.
The unique double epimerization reaction of the activated form of D-Man catalyzed by GDP-Man 3Ј,5Ј-epimerase precedes the committed step in the biosynthesis of L-AA and glycoconjugates, which results in the irreversible hydrolysis of the highly energetic glycosyl-pyrophosphoryl linkage. Therefore, the epimerization step must be tightly controlled. Here, we show that the GDP-Man 3Ј,5Ј-epimerase of A. thaliana undergoes a complex regulation that is linked to the cell wall biosynthesis and the redox state of the cell. In particular, we demonstrate the formation of a novel intermediate, GDP-L-gulose, and propose that this compound could be dedicated for the de novo biosynthesis of vitamin C in plants.
Plasmids-The GATEWAY TM (Invitrogen) plasmids containing the GDP-Man 3Ј,5Ј-epimerase gene of A. thaliana, pDEST15_Epim and pDEST17_Epim, were prepared as described (11) for the bacterial expression of GST-and His-tagged epimerase (N-terminal fusions), respectively.
GDP-Man 3Ј,5Ј-Epimerase Assay and Ascorbic Acid Determination-The GDP-Man 3Ј,5Ј-epimerase activity and L-AA were measured by the HPLC method as described (13), with the exception that the concentration of methanol in solvent A was 0.5%, and the flow rate was 0.8 ml/min.
In Vivo Labeling of A. thaliana Cell Suspensions with D-[U-14 C]-Man-In vivo labeling of A. thaliana cells was performed as described (13). Cell suspensions were pre-adapted to labeling conditions for 20 h in the presence or absence of exogenous L-AA or its precursors (2.5 mM) followed by a 2-h labeling with 1 Ci of D-[U-14 C]Man. L-AA was extracted with 5% metaphosphoric acid containing 2 mM dithiothreitol and 1 mM EDTA.
Bacterial Expression of the Recombinant Epimerase-Heterologous expression of the recombinant epimerase in Escherichia coli submitted to a "reversed" heat shock (a shift from 37 to 26°C just before the induction) was performed as described (11). Cells were re-suspended in 3 volumes of 50 mM Tris-HCl buffer, pH 7.7, containing 0.5 mM dithiothreitol, 1 mM phenylmethylsulfonyl fluoride, and 20% glycerol (buffer A). Crude extracts and 90% ammonium sulfate precipitates were prepared as described (11).
Nickel Nitrilotriacetic Acid Metal Affinity Chromatography of the Recombinant GDP-Man 3Ј,5Ј-Epimerase-A crude extract containing the His-tagged epimerase protein was loaded on a 2-ml nickel nitrilotriacetic acid Superflow column equilibrated with 5 mM imidazole in 25 mM Tris-HCl buffer pH 7.7 containing 1 mM phenylmethylsulfonyl fluoride (buffer B). The column was washed with 10 volumes of the equilibration buffer followed by 5 volumes of 20 mM imidazole in buffer B. The elution was carried out with 3 volumes of 300 mM imidazole in buffer B.
GST Affinity Chromatography of the Recombinant GDP-Man 3Ј,5Ј-Epimerase-A crude extract containing the GST-tagged epimerase was applied to a 2-ml GST-affinity column equilibrated with buffer A. The column was washed with 15 volumes of buffer A, and the recombinant epimerase was eluted with 3 volumes of 10 mM glutathione (reduced form) in buffer A.
Extraction and Assay of L-Gulono-1,4-lactone Dehydrogenase Activity-The L-gulono-1,4-lactone dehydrogenase activity was extracted from white potato tubers essentially as described (14), except that gel filtration was performed on NAP-25 columns (Amersham Biosciences) and the obtained high molecular weight fraction was separated by centrifugation (20,000 ϫ g for 20 min) into the cytosolic (supernatant) and the mitochondrial (pellet) fractions. The dehydrogenase activity was measured spectrophotometrically at 550 nm by following the L-gulono-1,4-lactone dependent reduction of cytochrome c (14).
PAGE-Proteins were separated by SDS/PAGE using 12.5% minigels and the buffer system described by Laemmli (15). Gels were stained with Coomassie Brilliant Blue R-250.
Peptide Sequencing and Protein Identification-Tryptic peptides prepared from in-gel digested protein bands were analyzed by nanoelectrospray tandem mass spectrometry, and the obtained sequence information was submitted to database searching as described (11).
Protein Determination-Protein concentration was determined by the method of Bradford (16) using bovine serum albumin as standard.
Sugar Analysis-GDP-14 C-labeled hexoses of the epimerase reaction mixtures were hydrolyzed in 50 mM HCl at 100°C for 20 min. For HPLC analysis the acid-released 14 C-labeled hexoses together with cold sugar standards were converted to the corresponding 1-phenyl-3-methyl-5pyrazolone (PMP) derivatives (17). The HPLC system with on-line UV and radioactivity detection (13) was used. The solvent was 18% acetonitrile in 0.1 M phosphate buffer, pH 7.0 or pH 5.0 (for separation of PMP-derivatives of Man and Gul) at a flow rate of 0.8 ml/min. To identify altrose, free sugars were separated by TLC on silica gel 60 aluminum sheets (pre-impregnated with 0.3 M NaH 2 PO 4 ) in acetone/nbutanol/water (8:1:1, v/v/v) and detected as described (13).

RESULTS
Characterization of the GDP-Man 3Ј,5Ј-Epimerase-To gain an insight into the regulation of the de novo biosynthesis of vitamin C, we have characterized the native and recombinant epimerase of A. thaliana. The epimerase belongs to the short chain dehydrogenase/reductase family (18). The native enzyme is a homodimer of 43 kDa subunits (11) and possesses two potential NAD binding sites and two potential substrate binding sites per dimer (19). The epimerase has a low K m for the GDP-Man substrate (4.4 M) (Fig. 1A) compared to the Chlorella epimerase (96 M) (20) and to the related bacterial enzymes, GDP-Man 4Ј,6Ј-dehydratase (19) and GDP-L-Fuc synthetase (21-22) (280 and 38.6 M, respectively). In contrast to the GDP-Man 4Ј,6Ј-dehydratase (23) and GDP-L-Fuc synthetase (24) of E. coli, the epimerase is strongly inhibited by GDP and GDP-Glc in a competitive manner, with respective K i values of 0.7 M (Fig. 1C) and 5 M (data not shown). The enzyme recognizes the purine moiety of GDP-derivatives since an adenosine derivative, ADP-D-Glc, had no effect on the enzyme activity. Surprisingly, only a partial inhibition was observed with GDP-L-Fuc (I 50 ϭ 70 M; Fig. 1D), even at 1 mM concentration ( Table I). The sigmoidal inhibition curve with GDP-L-Fuc ( Fig.  1D) is reminiscent of a feedback regulation observed in the biosynthesis of NDP-6-deoxyhexoses in bacteria (25)(26)(27). Like CDP-Glc 4Ј,6Ј-dehydratase of Yersinia pseudotuberculosis (28), the purified epimerase was stimulated by exogenously added oxidized forms of nicotinamide-adenine dinucleotides, NAD ϩ and NADP ϩ (145 and 110% of control, respectively), and inhibited by their reduced forms, NADH and NADPH (78 and 88% of control, respectively) ( Table I). Physiological concentrations (1 mM) of L-AA, a reducing agent and the end product of the pathway, inhibited the epimerase activity by 15% (Table I). In agreement with that, a feedback inhibition of the L-AA biosynthesis was clearly observed in vivo, since feeding A. thaliana cells with exogenous L-AA resulted in an increased level of the intracellular L-AA and a decreased incorporation of the [ 14 C]Man label into L-AA (Table II). The partial inhibition of the epimerase by L-AA might be explained in terms of "reductive inhibition" (29), i.e. as an L-AA-dependent reduction of the enzyme-NAD ϩ complex in the absence of the nucleotide sugar. However, the same degree of inhibition (14%) was observed in the presence of 1 mM L-galactono-1,4-lactone (Table I), whereas D-isoascorbic acid and L-galactose had no effect. These facts suggest the existence of a stereospecific mechanism of the enzyme inhibition by sugar lactones.
We cloned the epimerase gene of A. thaliana (11) and affinity-purified the recombinant His-and GST-tagged proteins from E. coli host submitted to a "reversed" heat-shock. Of the recombinant epimerase protein (3 mg/liter of culture), 90% were found in the soluble fraction, but only 30% of the total activity were retained on the affinity columns probably because of interactions with other proteins. Analysis of the affinitypurified enzyme by SDS-PAGE (Fig. 1B) followed by mass spectroscopy identification of the protein bands revealed the presence of the major GDP-Man 3Ј,5Ј-epimerase band and of a weak 70-kDa band corresponding to DnaK, a Hsp70 chaperone of E. coli (the tryptic peptides identified are TTPSIIAYTQDG-ETLVGQPAK, IINEPTAAALAYGLDK, and SLGQFNLDGIN-PAPR) Thus, the recombinant epimerase could interact with a Hsp70 molecular chaperone. During purification of the native GDP-Man 3Ј,5Ј-epimerase from A. thaliana cell suspensions (11), a 70-kDa chaperone (DnaK ortholog) co-purified with the epimerase throughout the whole procedure and was identified in the NaCl eluate from Blue-Sepharose as a cytosolic Hsc70.3 heat-shock cognate protein of Arabidopsis (At3g-09440; the identified tryptic peptides are NQVAMNPINTVFD-AK, NAVVTVPAYFNDSQR, DAGVIAGLNVMR, VQQLLVDF-FNGK, and FELSGIPPAPR). The majority of the Hsc70 protein was separated from the epimerase by gel filtration.    14 C-labeled and cold total L-AA in acid extracts was measured by HPLC. fw, fresh weight.

Compound
Cold L-AA (nmol/g fw) 14 C-labeled L-AA (10 23  This step resulted in a 10-fold decrease of the epimerase activity, possibly because of not only the partial loss of the NAD cofactor (11) but also the disruption of interactions with the Hsc70 chaperone. The steady-state parameters of the recombinant GST-and His-tagged epimerase were comparable with those of the native enzyme, although the recombinant enzymes were catalytically less efficient (Fig. 1A). The V max of the native enzyme was low, suggesting a rate-limiting enzyme. The overall catalytic efficiency of GDP-Man 3Ј,5Ј-epimerase, expressed as k cat /K m (9.1 s Ϫ1 mM Ϫ1 ), was fairly good and comparable with

FIG. 2. Formation of GDP-L-gulose by GDP-Man 3,5-epimerase of A. thaliana.
Shown is HPLC analysis of the reaction products obtained with the affinity-purified (A) and with a crude (90% ammonium sulfate fraction) GSTtagged recombinant epimerase (B). A, the affinity-purified epimerase (3 g) was incubated with GDP-[ 14 C]Man. The reaction products at the equilibrium were analyzed either directly as sugar nucleotides (panel a) or first submitted to mild acid hydrolysis followed by the conversion of the released 14 C sugars to the corresponding PMP derivatives and HPLC analysis at pH 7 (panel b) and pH 5 (panel c). The elution of unlabeled sugar standards is indicated. At pH 5, the PMP-Gal derivative elutes at 75 min (not shown). B, the GST-epimerase-containing crude fraction (90% ammonium sulfate precipitate; 75 g) was incubated with GDP-[ 14 C]Man for different intervals of time. Half of each reaction mixture was analyzed directly by HPLC (left panel). The second half was acid-hydrolyzed, the obtained 14 C-labeled hexoses were converted to the corresponding PMP derivatives, and their relative ratios were determined by HPLC (right panel), as described above.
that reported for the GDP-L-Fuc synthetase of E. coli (21).
Formation of GDP-L-Gulose by the GDP-Man 3Ј,5Ј-Epimerase of A. thaliana-The most intriguing observation was the variation of the apparent equilibrium constant K eq Ј (from 0.1 to 1.5) measured for different preparations of the recombinant epimerase. Similar variations of K eq Ј but in a narrower range (from 0.1 to 0.4) were observed with preparations of the native epimerase from A. thaliana. Also, an unexplained anomaly with the measured K eq Ј values for the reverse reaction was reported for the epimerase of Chlorella sp. (20). Fig. 2A, panel  a, shows the HPLC profile of the reaction products at the equilibrium obtained with the affinity-purified epimerase. The measured ratio (K eq Ј) of the epimerization product(s) to the GDP-Man substrate was 0.6. If GDP-L-Gal were the only epimerization product, then a similar ratio (0.6) should be obtained for the mild acid-released 14 C-labeled L-Gal versus D-Man. The measured L-Gal to D-Man ratio was only 0.35 ( Fig.  2A, panel b), fact that indicated that the Man peak contained an unknown component. This component was separated from D-Man and co-migrated with L-gulose standard ( Fig. 2A, panel  c). Therefore, we conclude that the epimerase reaction mixture contained at equilibrium GDP-D-Man, GDP-L-Gal, and GDP-L-Gul in a respective ratio of 1:0.4:0.2. A similar analysis of the epimerization products obtained with an epimerase-containing 55-70% ammonium sulfate fraction from A. thaliana cell suspensions revealed the presence of GDP-D-Man, GDP-L-Gal, and GDP-L-Gul in a ratio of 1:0.18:0.09 (results not shown). Fig. 2B shows the HPLC profiles of the epimerase reaction products obtained with a crude recombinant enzyme (90% ammonium sulfate fraction) and the relative ratios of the GDPhexoses formed. In this case the reaction was shifted toward the GDP-L-Gul formation and the relative ratios of GDP-D-Man, GDP-L-Gal, and GDP-L-Gul at the equilibrium were 1:0.4: 1.1 (Fig. 2B).
Our results demonstrate that the GDP-Man 3Ј,5Ј-epimerase reaction can be dissected into at least two distinct epimerization reactions leading to the formation of two discrete products: GDP-L-Gal and GDP-L-Gul (Fig. 3). The fate of the epimeriza- tion seems to depend on the molecular form of the enzyme, probably as a result of its interactions with other proteins.
Synthesis of L-AA from L-Gulose and L-Gulono-1,4-lactone by Plant Cells-In contrast to L-Gal, which is a minor constituent of plants, L-Gul, as far as we know, has never been reported in plants. Therefore, GDP-L-Gul might be dedicated to the biosynthesis of L-AA. To test this hypothesis, we supplied A. thaliana cells with cold L-Gul or L-gulono-1,4-lactone and measured the level of cold L-AA in the cells. As reported earlier for cress seedlings, preincubation with L-gulono-1,4-lactone (30) but also with L-Gul resulted in an increased level of vitamin C, and L-gulono-1,4-lactone was as efficient a precursor of L-AA as was L-galactono-1,4-lactone (Table II). Moreover, in the presence of cold precursors, a decreased incorporation of the [ 14 C]Man label into vitamin C was observed (Table II), as expected for biosynthetic intermediates. L-AA inhibited its own biosynthesis (Table II) by a feedback mechanism apparently at the level of GDP-Man 3Ј,5Ј-epimerase (Table I).
We demonstrated, therefore, that L-Gul and L-gulono-1,4lactone are converted into L-AA by A. thaliana cell suspensions. The L-Gal dehydrogenase can use L-Gul as substrate (12). However, the mitochondrial L-galactono-1,4-lactone dehydrogenase is highly specific and does not oxidize L-gulono-1,4-lactone (8 -9). Therefore, A. thaliana cells must possess another enzyme able to convert L-gulono-1,4-lactone to L-AA. L-Gulono-1,4-lactone dehydrogenase activity was reported in the cytosolic fraction of Euglena sp. (31) and in the mitochondrial fraction obtained from potato tubers (14). In agreement with Ô ba et al. (14), we could detect the L-gulono-1,4-lactone dehydrogenase activity (0.66 milliunit/g of tissue) in the mitochondrial fraction from potato tubers (Table III). This activity represented about 30% that of L-galactono-1,4-lactone dehydrogenase measured with L-galactono-1,4-lactone as substrate (2.16 milliunits/g of tissue) . Surprisingly, we found that the majority (75%) of L-gulono-1,4-lactone dehydrogenase activity (4.51 milliunits/g of tissue) was present in the cytosolic fraction from potato tubers (Table III). The cytosolic enzyme(s) could use L-galactono-1,4lactone as substrate, and this activity represented 90% that measured with L-gulono-1,4-lactone. These facts point to the existence of differently localized isozymes. DISCUSSION The biochemical characterization of the GDP-Man 3Ј,5Ј-epimerase of A. thaliana has brought new insights into the de novo biosynthesis of L-AA and its regulation. The unexpected partial inhibition of the epimerase by GDP-L-Fuc (Fig. 1D) could be of paramount importance because, even at high concentration of GDP-L-Fuc in the cell, the epimerase would still supply GDP-L-Gal/GDP-L-Gul substrates necessary for the de novo synthesis of L-AA. The complex type of inhibition by GDP-L-Fuc could also play a role in the regulation of the cell wall biosynthesis in plants. In the presence of high concentrations of GDP-L-Fuc, the cellular level of the GDP-L-Gal precursor will be low, thus resulting in a lesser incorporation of L-Gal into glycoconjugates. Indeed, in the L-Fuc-deficient mur1 mutant of Arabidopsis, which lacks the GDP-Man 4Ј,6Ј-dehydratase activity (32) catalyzing the first step in the GDP-L-Fuc pathway (Fig. 4), L-Gal replaces L-Fuc in xyloglucans (33) and N-glycans (34).
We observed that the epimerase is stimulated by the oxidized forms of nicotinamide adenine dinucleotides (NAD ϩ /NADP ϩ ) but inhibited by their reduced forms (NADH/NADPH) and by L-AA (Table I). Because the NAD binding motif (GAGGFI) present within the epimerase sequence is a modified version of the common Rossmann consensus (GXGXXG) found in other members of the short chain dehydrogenase/reductase family (35), this feature could be responsible for a lower affinity of the enzyme for the dinucleotides and the corresponding stimulation/inhibition effects. Moreover, our in vivo experiments demonstrated a feedback inhibition in the vitamin C pathway (Table II). A feedback inhibition was suggested by others (36), but no potential target for it was indicated. The late steps of the   (Table II). Therefore, the observed feedback inhibition must take place earlier in the pathway. GDP-Man 3Ј,5Ј-epimerase is partially inhibited by both L-AA and L-galactono-1,4-lactone in vitro (Table I). Inhibition of the reversible epimerase reaction by the end product could be of great importance because it would save the energy of the glycosyl-pyrophosphoryl linkage of the sugar nucleotide, otherwise irreversibly lost during the subsequent hydrolytic steps. We propose, therefore, that the GDP-Man 3Ј,5Ј-epimerase catalyzing the first specific step in the biosynthesis of vitamin C could undergo a feedback inhibition, sense the redox state of the cell, and play an important role in the regulation of vitamin C and cell wall/glycoproteins biosynthesis.
We have shown that GDP-Man 3Ј,5Ј-epimerase produces a novel intermediate, GDP-L-Gul, in addition to the well known GDP-L-Gal (Fig. 2). Thus, the unique epimerization catalyzed by the epimerase is a complex reaction involving at least two distinct epimerization steps (Fig. 3). To our knowledge, this is the first report of a sugar epimerase releasing two different epimerization products (37).
Both recombinant and native GDP-Man 3Ј,5Ј-epimerase of A. thaliana co-purified with Hsp70 heat-shock proteins (E. coli DnaK and A. thaliana Hsc70.3, respectively). The Hsc70 heatshock protein of Arabidopsis is constitutively expressed and stress-inducible (38). The highly conserved, ubiquitous Hsp70 chaperones play a key role in protection and adaptation to stress by participating in folding and unfolding of misfolded and native-state proteins (39), disassembly of regulatory complexes (40), and regulation of protein/enzyme activity (41)(42)(43)(44). Significantly, increased vitamin C levels of Arabidopsis leaves in response to heat shock were reported (45), and an overexpression of the bacterial DnaK chaperone improved the salt and heat tolerance of transgenic tobacco (46 -47). On the basis of its chromatographic behavior and enzymatic properties (K eq Ј), we could detect different molecular forms of the epimerase, although the nature of these forms is not understood. Given the known functions of molecular chaperones and the fact that epimerase could interact with Hsp70 heat-shock proteins, we speculate that the Hsc70.3 protein of A. thaliana might be implicated in folding and/or regulation of the epimerase.
L-Gulose is an extremely rare sugar. It is present in two bacterial products, a glycolipid of Thermoplasma acidophilum (archaeobacteria) (48) and in the bleomycin of Streptomyces verticillus (Actinomycetales) (49), the synthesis of which involves the BlmG gene product, a close homolog of the plant epimerase (11). Significantly, gulose (of unknown absolute configuration) was found also in algae, for example in a glycoprotein of Volvox (50) and in the cell walls of Chlorophyta (51). Because L-Gul is apparently not a structural component of plants, we searched for a possible role of GDP-L-Gul in the biosynthesis of vitamin C. The hydrolytic step(s) leading from GDP-L-Gal to free L-Gal had not been identified (7) and seems unspecific in vitro because totally different products compared with those reported (7), namely L-Fuc 1-phosphate and L-Fuc, were obtained with crude extracts of A. thaliana incubated with GDP-D-[ 14 C]Man (13). Therefore, we speculate that GDP-L-Gul might undergo an enzymatic hydrolysis and release free L-Gul. Indeed, L-Gul and L-gulono-1,4-lactone are intermediates in the L-AA biosynthesis (Table II), and L-gulono-1,4-lactone dehydrogenase activity is present in plant extracts (Table  III). Significantly, both gulonic and galactonic acids were present in metabolic profiles of plants (52). Moreover, the Arabi-dopsis genome contains a family of closely related genes that show 22-24% identity with the rat L-gulono-1,4-lactone oxidase (At1g 32300, At2g 46740, At2g 46750, At2g 46760; At5g 11540, At5g 56470, and At5g 56490 (www.arabidopsis.org)). The predicted gene products contain at the N terminus a non-covalent FAD binding domain and are targeted to different cellular compartments (secretory pathway, cytosol, and mitochondria). Consequently, some of these proteins could be responsible for the observed conversion of L-gulono-1,4-lactone to L-AA.
In summary, the first step of the de novo pathway for vitamin C in plants, catalyzed by GDP-Man 3Ј,5Ј-epimerase, undergoes a complex control and supplies two distinct products: GDP-L-Gal and GDP-L-Gul. We propose that, in contrast to GDP-L-Gal, which is used for the biosynthesis of glycoconjugates, GDP-L-Gul would be channeled directly into the vitamin C pathway (Fig. 4). After release, L-Gul could be oxidized to L-gulono-1,4lactone by the L-galactose dehydrogenase (12) or a similar enzyme. The last step of the proposed L-Gul branch, oxidation of L-gulono-1,4-lactone by an organelle-specific dehydrogenase, may take place at different cellular locations and, thus, produce L-AA in situ with no need for its intracellular transport. Further studies will be necessary to unravel the nature and specificity of the hydrolytic step(s) responsible for L-Gal/L-Gul release and to determine whether both branches, involving either L-Gal or L-Gul, function in the de novo biosynthesis of vitamin C in planta.