Plant γ-Glutamyl Hydrolases and Folate Polyglutamates

γ-Glutamyl hydrolase (GGH, EC 3.4.19.9) catalyzes removal of the polyglutamyl tail from folyl and p-aminobenzoyl polyglutamates. Plants typically have one or a few GGH genes; Arabidopsis has three, tandemly arranged on chromosome 1, which encode proteins with predicted secretory pathway signal peptides. Two representative Arabidopsis GGH proteins, AtGGH1 and AtGGH2 (the At1g78660 and At1g78680 gene products, respectively) were expressed in truncated form in Escherichia coli and purified. Both enzymes were active as dimers, had low Km values (0.5–2 μm) for folyl and p-aminobenzoyl pentaglutamates, and acted as endopeptidases. However, despite 80% sequence identity, they differed in that AtGGH1 cleaved pentaglutamates, mainly to di- and triglutamates, whereas AtGGH2 yielded mainly monoglutamates. Analysis of subcellular fractions of pea leaves and red beet roots established that GGH activity is confined to the vacuole and that this activity, if not so sequestered, would deglutamylate all cellular folylpolyglutamates within minutes. Purified pea leaf vacuoles contained an average of 20% of the total cellular folate compared with ∼50 and ∼10%, respectively, in mitochondria and chloroplasts. The main vacuolar folate was 5-methyltetrahydrofolate, of which 51% was polyglutamylated. In contrast, the principal mitochondrial and chloroplastic forms were 5-formyl- and 5,10-methenyltetrahydrofolate polyglutamates, respectively. In beet roots, 16–60% of the folate was vacuolar and was again mainly 5-methyltetrahydrofolate, of which 76% was polyglutamylated. These data point to a hitherto unsuspected role for vacuoles in folate storage. Furthermore, the paradoxical co-occurrence of GGH and folylpolyglutamates in vacuoles implies that the polyglutamates are somehow protected from GGH attack.

␥-Glutamyl hydrolase (GGH, EC 3.4.19.9) catalyzes removal of the polyglutamyl tail from folyl and p-aminobenzoyl polyglutamates. Plants typically have one or a few GGH genes; Arabidopsis has three, tandemly arranged on chromosome 1, which encode proteins with predicted secretory pathway signal peptides. Two representative Arabidopsis GGH proteins, AtGGH1 and AtGGH2 (the At1g78660 and At1g78680 gene products, respectively) were expressed in truncated form in Escherichia coli and purified. Both enzymes were active as dimers, had low K m values (0.5-2 M) for folyl and p-aminobenzoyl pentaglutamates, and acted as endopeptidases. However, despite 80% sequence identity, they differed in that AtGGH1 cleaved pentaglutamates, mainly to di-and triglutamates, whereas AtGGH2 yielded mainly monoglutamates. Analysis of subcellular fractions of pea leaves and red beet roots established that GGH activity is confined to the vacuole and that this activity, if not so sequestered, would deglutamylate all cellular folylpolyglutamates within minutes. Purified pea leaf vacuoles contained an average of 20% of the total cellular folate compared with ϳ50 and ϳ10%, respectively, in mitochondria and chloroplasts. The main vacuolar folate was 5-methyltetrahydrofolate, of which 51% was polyglutamylated. In contrast, the principal mitochondrial and chloroplastic forms were 5-formyl-and 5,10-methenyltetrahydrofolate polyglutamates, respectively. In beet roots, 16 -60% of the folate was vacuolar and was again mainly 5-methyltetrahydrofolate, of which 76% was polyglutamylated. These data point to a hitherto unsuspected role for vacuoles in folate storage. Furthermore, the paradoxical co-occurrence of GGH and folylpolyglutamates in vacuoles implies that the polyglutamates are somehow protected from GGH attack.
In plants, as in other organisms, tetrahydrofolate (THF) 1 and its derivatives, collectively termed folates, usually have ␥-linked polyglutamyl tails of up to about seven residues attached to the first glutamate ( Fig. 1A) (1,2). The tail critically affects the cofactor activity and transport of folates because most folate-dependent enzymes prefer polyglutamates, whereas most carriers prefer monoglutamates (3,4). By favoring protein binding, the tail also enhances folate stability, since bound folate is less prone to oxidative degradation (1). The enzyme that adds the polyglutamyl tails, folylpolyglutamyl synthetase, has been well studied in plants and shown to have mitochondrial, plastidial, and cytosolic isoforms (5,6). Less is known for plants about the enzyme that removes the tails, ␥-glutamyl hydrolase (GGH), although soluble GGH activity (5,7) and GGH-like mRNAs (8,9) clearly occur widely.
Biochemical characterization of plant GGHs has so far been limited to crude extracts (9,10) or a mixture of isoforms (11) and has produced conflicting results on the enzyme structure and mode of action. Furthermore, the subcellular location of plant GGH is unclear; cytosolic (11) or extracellular (12) sites have been proposed, but the vacuole is another a priori possibility (11,13). Wherever it is, plant GGH must be sequestered away from folate polyglutamates, since extractable GGH activities are typically enough to deglutamylate all cellular folate within minutes (7,10,11), and treatments that disrupt cellular compartmentation cause massive deglutamylation of folates in situ (14).
Nor is much known about the subcellular distribution in plants of the folate polyglutamate substrates of GGH, i.e. about the glutamyl tail lengths, one-carbon (C 1 ) substituents, and oxidation state of the folates in different organelles. The only data are for pea leaves, in which penta-and tetraglutamates of 5-formyl-THF and THF appear to predominate in mitochondria and 5-methyl-THF (of unknown tail length) elsewhere (15)(16)(17). There are, however, some data for pea on the distribution of total folate. Mitochondria, plastids, and a fraction composed mainly of the cytosol and vacuole accounted for 11-30, 3-11, and 60 -80%, respectively, of leaf folate (17)(18)(19). Besides not distinguishing among folate forms, these data leave open the key question of whether vacuoles contain folates.
Knowing whether folates or GGH occur in vacuoles is important for several reasons. If vacuoles store folates, the transport or storage processes might be engineered to enhance folate accumulation (20 -22). However, if GGH is vacuolar, the situ-ation could be as in mammals, where lysosomes do not store folates but import folate polyglutamates, hydrolyze them, and export monoglutamates (23,24). GGH also hydrolyzes the paminobenzoate (pABA) polyglutamates formed by oxidative degradation of folates (11,25), yielding p-aminobenzoyl glutamate (pABAGlu 1 ). This may be the first step in the salvage of pABA for folate re-synthesis, a potentially crucial but unexplored process (22).
In this study we used recombinant Arabidopsis GGHs to clarify the features of this enzyme in plants and showed that GGH is confined to vacuoles. We also determined the types and tail lengths of folates in organelles. This analysis revealed high levels of 5-methyl-THF polyglutamates in vacuoles.
Organelle preparations for folate analysis were supplemented with 10 mM ␤-mercaptoethanol, frozen in liquid N 2 , and stored at Ϫ80°C. Enzymes were extracted from organelles by freezing in liquid N 2 and thawing at 30°C (3-5 cycles) then centrifuging at 16,000 ϫ g for 10 min at 4°C. These extracts and cytosol fractions were desalted on PD-10 columns (Amersham Biosciences) equilibrated in 100 mM potassium phosphate, pH 6.0, 10% glycerol, 10 mM ␤-mercaptoethanol, 10 mM ascorbic acid and stored at Ϫ80°C after freezing in liquid N 2 . GGH was assayed as outlined below. The marker enzymes ␣-mannosidase, NADP-linked glyceraldehyde-3-phosphate dehydrogenase, fumarase, and methylene tetrahydrofolate reductase (MTHFR) were assayed essentially as described (30 -32). Fumarase was assayed in a separate set of freshly prepared extracts not desalted or desalted in 50 mM Tricine-NaOH, pH 8.4.
Red Beet Root Vacuoles-Intact vacuoles were isolated essentially as described (33) except that dithiothreitol replaced ␤-mercaptoethanol, and Histodenz™ replaced metrizamide. Sigma plant protease inhibitor mixture was added to all media except when vacuoles were assayed for GGH activity. Routinely, 450 g of roots were peeled, cut into 5-mm slices, and loaded into the slicing apparatus, which was set to a slice thickness of 0.1 mm and operated at ϳ150 rpm. The slices and released cell contents were collected in 1 liter of ice-cold 50 mM Tris-HCl, pH 8.0, containing 1 M sorbitol, 5 mM EDTA, 15 mM dithiothreitol plus or minus 1 ml/liter protease inhibitor mixture and filtered through 2 layers of cheesecloth. The solid residue was sliced and filtered again, and the pooled filtrates were centrifuged at 2000 ϫ g for 20 min at 4°C. The vacuole-enriched pellets were resuspended in 20 ml of 15% (w/v) His-todenz™ dissolved in buffer A (1.2 M sorbitol, 1 mM EDTA, 15 mM dithiothreitol, 25 mM Tris-MES, pH 8.0, plus or minus 1 ml/liter protease inhibitor mixture). To prepare density gradients, 5-ml aliquots of the suspension in 15-ml disposable clinical centrifuge tubes were successively overlaid with 5 ml of 10% (w/v) Histodenz™ in buffer A and 2 ml of buffer A. Gradients were centrifuged at 650 ϫ g for 10 min at 4°C, and vacuoles were collected from the 0/10% Histodenz™ interface. To remove most of the Histodenz™, the suspension was added to at least 2 volumes of buffer A and centrifuged at 650 ϫ g for 5 min at 4°C. The final vacuole pellet was resuspended in 200 -500 l of resuspension medium or 200 l of GGH assay buffer (see GGH Assay section, below), frozen in liquid N 2 , and held at Ϫ80°C until analysis. Betanin contents of vacuoles and samples of the roots they came from were estimated spectrophotometrically (⑀ 550 nm ϭ 62,000 M Ϫ1 cm Ϫ1 ) (34). Betanin was extracted from freeze-dried root samples by grinding in 1 mM EDTA, 50 mM Tris-HCl, pH 7.5, and centrifuging to clarify.
Folate Analysis-Folates were extracted and analyzed by HPLC with electrochemical detection as described (21,35,36) with the following modifications. Subcellular fraction samples were routinely thawed, made to 10 ml with extraction buffer (50 mM Hepes-50 mM CHES, pH 7.9, 2% (w/v) sodium ascorbate, 10 mM ␤-mercaptoethanol), split in two, and processed plus or minus treatment with rat plasma conjugase. The folate binding column was scaled down from 5 to 1 ml, and the volumes of wash and eluting buffers were reduced proportionately. Beet root samples (8 g) were homogenized in 60 ml of extraction buffer; half the extract was conjugase-treated, and half was not. The HPLC column and mobile phase buffers were as described (21) with a 55-min (for monoglutamates) or 70-min (for polyglutamates) non-linear elution program. Detector response was calibrated using THF, 5-methyl-THF, 5,10methenyl-THF, 5-formyl-THF, and PteGlu 1 standards. Tri-and pentaglutamate forms of 5-methyl-, 5-formyl-, and 5,10-methenyl-THF were used to identify retention times of polyglutamates.
cDNAs and Expression in Escherichia coli-ESTs for genes At1g78660 (GenBank™ CF653065) and At1g78680 (GenBank™ AY096428) were obtained from the Deutsches Ressourcenzentrum fü r Genomforschung (Berlin, Germany) and the Arabidopsis Biological Resource Center (Columbus, OH), respectively. Based on sequence alignments, constructs lacking ϳ20 or ϳ45 N-terminal residues were designed for each protein (see Supplemental Fig. 1). The corresponding cDNAs were amplified using Pfu polymerase and the primers shown in Table I, digested, and cloned between the NdeI and XhoI sites of pET28b (Novagen, Madison, WI). This fused the hexahistidine tagcontaining sequence MGSSHHHHHHSSGLVPRGSHM to the N terminus of the proteins. Constructs were made in E. coli DH10B cells, sequence-verified, and introduced into E coli BL21-CodonPlus ® (DE3)-RIL cells (Stratagene), which were grown at 27°C in LB medium containing 50 g ml Ϫ1 kanamycin and 20 g ml Ϫ1 chloramphenicol until A 600 reached 0.6. Isopropyl-D-thiogalactopyranoside was then added (final concentration 1 mM), and incubation was continued for 3 h at 27°C.
Protein Purification-Operations were at 0 -4°C. Cells from a 50-ml culture were harvested by centrifugation, resuspended in 1 ml of 50 mM potassium phosphate, pH 8.0, 1.5 M NaCl, 10 mM ␤-mercaptoethanol, and broken in a Mini-BeadBeater (Biospec, Bartlesville, OK) using 0.1-mm zirconia/silica beads. After centrifugation (16,000 ϫ g, 15 min), the supernatant was subjected to Ni 2ϩ chelate affinity chromatography (0.5-ml column) following the manufacturer's protocol. Bound proteins were eluted using 250 mM imidazole and desalted on PD-10 columns in 100 mM potassium phosphate, pH 6.0, 10% glycerol, 10 mM ␤-mercaptoethanol. After storage at 4°C or at Ϫ80°C after freezing in liquid N 2 , the purified enzymes maintained full activity for several weeks and months, respectively. Protein was estimated by the Bradford method (37) using bovine serum albumin as the standard. Analytical Ultracentrifugation-Experiments were run using a Beckman XL-I analytical ultracentrifuge and an AN-60-Ti rotor at 25°C. The proteins were dialyzed into 10 mM sodium acetate, pH 5.5, 1 M NaCl, 1 mM dithiothreitol, 1 mM octyl ␤-glucoside. Buffer viscosity and density and protein partial specific volume ( ) were obtained from SEDNTERP software (www.jphilo.mailway.com). The values for At-GGH1 and AtGGH2 were, respectively, calculated from amino acid content to be 0.7324 and 0.7351 ml/g at 20°C. In sedimentation velocity studies sample volume was 0.42 ml, and the reference volume of dialysis buffer was 0.44 ml. The samples were run at 50,000 rpm. Absorption measurements were made at 280 nm for AtGGH1 (0.8 mg/ml) and 230 nm for AtGGH2 (0.1 mg/ml). A single sample was run in each experiment with zero time between scans, R min was set at 6.0, and the samples were scanned from the earliest time until the boundaries reached the cell bottom. The samples were at thermal equilibrium before starting to spin the rotor, which was accelerated directly to 50,000 rpm. The data were analyzed using the c(s) and c(M) methods found in Sedfit (38) (www.analyticalultracentrifugation.com). The sedimentation coefficients calculated from the Sedfit program were converted to s 20,w values using Sedfit.
GGH Assay-Standard assay mixtures contained folate pentaglutamate (PteGlu 5 ) or pABAGlu 5 (0.2 mM unless otherwise indicated), 100 mM potassium phosphate, pH 6.0, 10% (v/v) glycerol, 10 mM ␤-mercaptoethanol, and enzyme in a final volume of 100 l. After incubation for 0 -6 h, the reaction was stopped by boiling for 3 min, then centrifuged at 16,000 ϫ g for 10 min. Incubation was at 37°C; no substantial difference in reaction rate was observed in the range 30 -42°C. Folateor pABA-containing analytes in the supernatant were separated by HPLC using a C18 Symmetry column (Waters, 7.5 ϫ 0.46 cm; 3-m particle size) and quantified by absorption at 282 nm (292 nm for 5-methyl-THF) or by fluorescence (270-nm excitation, 350-nm emission), respectively. The column was eluted at 1.5 ml min Ϫ1 with a 10-min linear gradient from 15 to 45% (v/v) methanol in 50 mM sodium phosphate buffer, pH 6.0, containing 8 mM tetrabutylammonium bisulfate (buffer B). ␥-Glu n and glutamate were derivatized with o-phthalaldehyde (39), separated using the column above connected in series to a C18 Spherisorb ODS2 column (Waters, 15 ϫ 0.46 cm; 5-m particle size) and isocratic elution with 40% methanol buffer B, and quantified fluorometrically (365-nm excitation, 420-nm emission). For kinetic studies initial rates of substrate disappearance were measured (before an average of 25% of PteGlu 5 or 3% of pABAGlu 5 had been used). Rates were proportional to enzyme concentration and time. Kinetic constants were calculated from Hanes-Woolf plots. To determine specificity for the various ␥-glutamyl bonds in PteGlu 5 or pABAGlu 5 , residual substrate and the various PteGlu n or pABAGlu n products were determined, and the extent of hydrolysis (E) was plotted versus the relative concentration (R) of each PteGlu n or pABAGlu n species (40). These terms are defined as follows.
where XGlu n ϭ PteGlu n or pABAGlu n . The slope of the line corresponding to each product measures the relative extent of its formation and so indicates the ␥-glutamyl bond specificity of the enzyme.

GGH-like Genes in Arabidopsis and Other Plants-BLAST searches of genome and EST databases at GenBank™ and The
Institute for Genomic Research indicated that higher plants have one or a few genes encoding proteins similar to mammalian GGHs. For instance, rice and maize appear to have one gene, soybean and cotton two, Arabidopsis three, and tomato four. Arabidopsis is, thus, typical in having a small GGH gene family.
The three Arabidopsis genes are tandemly arranged on chromosome 1 and specify proteins (AtGGH1, -2, and -3) that are ϳ35% identical to human GGH and 70 -80% identical to each other. As shown in Supplemental Fig. 1, the Arabidopsis proteins share many features with mammalian GGHs, including a predicted secretory pathway signal peptide (41,42), at least one N-glycosylation motif (13), and the conserved cysteine and histidine residues that are catalytically essential in human GGH as well as six other conserved residues that may contribute to catalysis or substrate binding (43,44). In contrast to these similarities, the AtGGH3 sequence has two short insertions near the N terminus that have no clear-cut counterparts in mammalian GGHs and are absent from all other available plant GGH sequences. Because AtGGH3 appears to be unique to Arabidopsis and, thus, not of general interest it was not further studied. This enzyme is also far less strongly expressed than AtGGH1 or -2. 2 Kinetic Characterization of GGH Proteins-Recombinant At-GGH1 and AtGGH2 were expressed in E. coli, removing the signal peptide region and replacing it with a sequence containing a histidine tag as was done for human GGH (43). This strategy gave high levels of both proteins, which were purified by Ni 2ϩ affinity chromatography to Ͼ90% homogeneity as judged by SDS-PAGE (see Supplemental Fig. 2). Two versions of each enzyme were made, differing in how many residues were removed (see Supplemental Fig. 1). Because results for both versions were similar, we report data only for the shorter one. Enzyme assays were made at pH 6.0, which was shown to be near the optimum for both enzymes (6.5-7.0 for AtGGH1 and 6.5 for AtGGH2) and is also close to the pH of the vacuole (in which GGHs are located, see Subcellular Localization of GGH Activity section, below).
The mode of action and specificity of purified AtGGH1 and AtGGH2 were studied using the pentaglutamates of folic acid (PteGlu 5 ) and pABA (pABAGlu 5 ) as substrates (Fig. 1, B and  C). In the initial stage of PteGlu 5 hydrolysis, AtGGH1 formed PteGlu 2 and PteGlu 3 at similar rates, whereas AtGGH2 formed PteGlu 1 and a small amount of PteGlu 2 (Fig. 1B). This indicates that both enzymes have an endopeptidase action, with AtGGH1 showing an almost equal preference for the second and third ␥-glutamyl bonds (counting from the folate moiety) and AtGGH2 preferring the first. This difference in specificity was also evident when pABAGlu 5 was the substrate, with an additional cleavage at a lower frequency of the fourth ␥-glutamyl bond catalyzed by AtGGH2 (Fig. 1C).
The kinetic constants for AtGGH1 and AtGGH2 with Pte-Glu 5 or pABAGlu 5 as substrate are shown in Table II. The K m values for PteGlu 5 for both enzymes were lower than those reported for a purified pea GGH preparation and similar to those found for mammalian GGHs, which are typically Ͻ2 M. As judged from the K cat /K m ratios, both enzymes hydrolyzed pABAGlu 5 less efficiently than PteGlu 5 .
Further distinctions between the two enzymes appeared when long incubation periods were used to define the end products of the reaction (Fig. 2). When AtGGH1 acted on Pte-Glu 5, the PteGlu 3 formed initially, but not the PteGlu 2 , was further hydrolyzed to PteGlu 1 , so that the final folate products included PteGlu 1 as well as PteGlu 2 and PteGlu 3 (Fig. 2A). The other products were ␥-Glu 3 and ␥-Glu 2 , which were not attacked (Fig. 2B). In contrast, AtGGH2 gave only PteGlu 1 as the final folate product, although some PteGlu 2 accumulated transiently ( Fig. 2A). The other products of AtGGH2 cleavage were ␥-Glu 5 , ␥-Glu 4 , ␥-Glu 3 , ␥-Glu 2 , and glutamate, with ␥-Glu 5 and ␥-Glu 4 disappearing as the reaction progressed (Fig. 2B). The unexpected formation of ␥-Glu 5 is considered below. The results with pABAGlu 5 as substrate resembled those with Pte-Glu 5 except that AtGGH1 did not further hydrolyze pABAGlu 3 , and AtGGH2 produced a small, transient peak of pABAGlu 4 and did not further hydrolyze pABAGlu 2 (Fig. 2, C and D). The kinetics of appearance of glutamate and ␥-Glu n (Fig. 2, B and D) indicated that AtGGH1 does not attack ␥-glutamyl peptides released from PteGlu 5 or pABAGlu 5 , whereas AtGGH2 does so via exopeptidase action, the early accumulation of ␥-Glu 4 being followed by ␥-Glu 3 and later by ␥-Glu 2 and ␥-Glu 1 . This was confirmed by using ␥-Glu 5 as the substrate (not shown).
Formation of ␥-Glu 5 by AtGGH2 (Fig. 2, B and D) was not due to cleavage of the bond between glutamate and pABA because there was no decline in total PteGlu n or pABAGlu n during the reaction (Fig. 2, A and C), and no pteroic acid was released from PteGlu 5 or pABA from pABAGlu 5 . This was confirmed by demonstrating that, like mammalian GGHs (13), neither plant enzyme attacks PteGlu 1 or pABAGlu 1 (not shown). The ␥-Glu 5 , thus, presumably came from transpeptidation between the ␥-Glu n products released from Pte-or pABAGlu 5 . Transpeptidase activity is common among peptidases although not previously reported for GGH. Transpeptidation can also explain why no glutamate counterpart to the pABAGlu 4 formed was seen for AtGGH2 acting on pABAGlu 5 (Fig. 2, C and D).
Native Molecular Mass-Purified recombinant enzymes were analyzed by analytical ultracentrifugation in sedimentation velocity experiments. Each protein yielded a single component in the analysis. For AtGGH1 the s 20,w value from the c(s) method was 4.57 S, corresponding to a molecular mass of 69.8 kDa calculated using the c(M) method. For AtGGH2, the s 20,w was 4.43 S, and the calculated mass was 66.0 kDa. These values are close to twice the calculated masses of the recombinant AtGGH1 and AtGGH2 polypeptides (36.7 and 36.4 kDa, respectively), indicating that both enzymes exist as dimers, as does human GGH (44). In this connection it is noteworthy that the dimer interface regions of the human enzyme (44) are substantially conserved in the plant sequences (see Supplemental Fig. 1).
Subcellular Localization of GGH Activity-GGH was localized by cell fractionation and enzyme assay. Pea leaves were used because they are the tissue of choice for obtaining high yields of intact vacuoles and other organelles (27)(28)(29) and have been the object of much prior work in folate biochemistry (11,(15)(16)(17)(18)(19). The distribution of marker enzyme activities confirmed that purified chloroplasts, mitochondria, and vacuoles were essentially uncontaminated by other fractions (Fig. 3). The vacuole fraction contained Յ1-2% of the total activities of the cytosol marker MTHFR and the peroxisomal marker catalase (not shown) (such slight contamination of vacuole preparations is too small to complicate interpretation of the vacuolar folate data to be presented below). The distribution of GGH activity closely paralleled that of the vacuolar marker ␣-mannosidase (Fig. 3), indicating that GGH is an exclusively vacuolar enzyme. To corroborate this result, we analyzed vacuoles from red beet roots, a classical plant vacuole system. The ratios between GGH activity and the solely vacuolar pigment betanin (34) in vacuoles and whole roots were, respectively, 29.8 Ϯ 3.5 and 21.0 Ϯ 3.3 pmol min Ϫ1 mol Ϫ1 (mean Ϯ S.E.), indicating that GGH activity is entirely vacuolar. GGH activities in beet vacuoles were far lower than those in pea vacuoles (11 Ϯ 1 versus 2320 Ϯ 420 pmol min Ϫ1 mg Ϫ1 protein, respectively). Subcellular Localization of Pea Leaf Folates-Purified organelle preparations and mesophyll protoplasts were first analyzed after removing polyglutamyl tails to enable identification and quantification of the types of folate (i.e. THF and its C 1 -substituted forms). The protoplast values (Fig. 4, top frame) agree with published total folate contents of pea leaves (16 -19) and show a typical distribution of folate types for leaves of peas and other plants (5,16) (the acidic HPLC mobile phase used in the analysis converts 10-formyl-THF to 5,10-methenyl-THF so that the 5,10-methenyl-THF measurements include 10-formyl-THF plus any preexisting 5,10-methenyl-THF). The total folate levels found in mitochondria and chloroplasts (about 50 and 10%, respectively, of cellular folates) also agree with literature values (16 -19), as does the finding that 5-formyl-THF is the major folate in mitochondria (16,17). The most striking result is that vacuoles contain a substantial amount of folate, almost all as 5-methyl-THF (Fig. 4, bottom frame). That this is not contamination by other fractions is attested by the enzyme marker data above, and by the distribution of folate types, which differs from those of other fractions (Fig. 4). The absence of significant contamination was further confirmed by adding tracer [ 3 H]folic acid to the protoplast lysates from which vacuoles were isolated; the purified vacuoles accounted for Յ0.5% of the 3 H added (data not shown).
To estimate the percentage of total cellular folate and 5-methyl-THF present in vacuoles, the folate contents of vacuole preparations and the protoplasts from which they came were expressed relative to the activity of ␣-mannosidase, which is specific to vacuoles in plants (45). Among three separate preparations, an average of 20% of total folate was vacuolar and 38% of the 5-methyl-THF (Table III).
Polyglutamylation of Pea Leaf Organellar Folates-The extent of polyglutamylation was determined for the major folate in each organelle (Fig. 5). The main mitochondrial folate, 5-formyl-THF, existed mainly as penta-and hexaglutamates (similar to a previous report, Ref. 15) as did 5,10-methenyl-THF in chloroplasts. Half the vacuolar 5-methyl-THF was also polyglutamylated, although the average tail length was shorter than in the other organelles. The survival of any polyglutamyl folate in the vacuole was unexpected as GGH activity is in theory high enough to deglutamylate most of the vacuolar folate in less  2. Kinetics of product formation during the prolonged action of AtGGH1 or AtGGH2 on pentaglutamate substrates. Initial substrate concentrations were 0.2 mM. Data are presented in units of nmol/20-l reaction. A, folate-containing products formed from PteGlu 5 . B, glutamate and ␥-Glu n products formed from PteGlu 5 . C, pABA-containing products formed from pABAGlu 5 . D, glutamate and ␥-Glu n products formed from pABAGlu 5 . Numerals (1)(2)(3)(4)(5) by curves indicate the number of glutamyl residues in each product. Data are from representative individual experiments, which were repeated at least three times with similar results.

FIG. 3. Localization of GGH in pea leaf vacuoles by subcellular fractionation. Chloroplasts (CP), mitochondria (M), and vacuoles (V)
were purified by density gradient centrifugation. A fraction enriched in cytosol and vacuole contents (CSϩV) was prepared from pea leaf protoplasts by pelleting intact organelles. The specific activities (units mg Ϫ1 protein) of GGH (measured using pABAGlu 5 ) and marker enzymes were assayed in each fraction. Markers were ␣-mannosidase (vacuole), NADP-linked glyceraldehyde-3-phosphate dehydrogenase (GAPDH, chloroplast), fumarase (mitochondrion), and MTHFR (cytosol). The asterisk indicates that a trace of MTHFR activity was detectable in vacuoles but was too low to quantify; it represented Յ2% of the total activity of the protoplasts. Data are the means and S.E. of data from 3 to 10 independent preparations of each fraction. than a minute (see "Discussion" for the calculation).
To confirm that the persistence of vacuolar 5-methyl-THF polyglutamates is not simply because plant GGHs do not attack them, we compared 5-methyl-THF pentaglutamate and Pte-Glu 5 as substrates for purified AtGGH1 and -2 and for the GGH activity in pea vacuole extracts. At the physiological concentration of 1 M, both substrates were hydrolyzed at similar rates by all three enzyme preparations. The observed activities (nmol min Ϫ1 mg Ϫ1 protein, with 5-methyl-THF pentaglutamate and PteGlu 5 , respectively) were: AtGGH1, 3690 and 5200; At-GGH2, 3590 and 4120; pea GGH, 1.9 and 2.8.
Analysis of Beet Root Folates-To explore the generality of the findings above, we analyzed red beet roots and their vacuoles. The principal folate in root tissue was 5-methyl-THF, in accord with published data for beets and other storage organs (46) (Fig. 6A). As with pea vacuoles, beet vacuoles contained substantial levels of 5-methyl-THF (Fig. 6A). The amounts of 5-methyl-THF in beet vacuoles were quantified by relating the folate contents of vacuole preparations and root tissue to their betanin contents (Table IV). In five independent preparations, from 16 to 60% of the 5-methyl-THF was vacuolar. The vacuolar 5-methyl-THF was 76% polyglutamylated, with di-and pentaglutamates predominant (Fig. 6B). Because beet vacuoles also contain GGH, the situation in beet is like that in pea; folyl polyglutamates and the enzyme that hydrolyzes them co-exist in vacuoles. DISCUSSION Our results establish that plant GGHs are broadly similar in structure and catalytic properties to those of mammals, as suggested by other reports (8 -11). Like mammalian GGHs (13,47), the Arabidopsis enzymes studied are dimers of ϳ300residue polypeptides, have mildly acidic pH optima, attack both pABA and folate polyglutamates, and show different cleavage patterns despite having similar amino acid sequences. At-GGH1 and AtGGH2 both act as endopeptidases, but they prefer different bonds, and only AtGGH2 attacks PteGlu 2 . Because PteGlu 2 is a major product of AtGGH1 action, the two enzymes in a sense complement each other. Such complementarity may have confounded specificity studies of plant GGHs made with crude extracts or mixed isoforms (8,9). AtGGH2 also shows exopeptidase and transpeptidase action on ␥-Glu n products.
The Arabidopsis GGH proteins and all other plant GGH sequences available have an N-glycosylation motif in a conserved position, suggesting that they exist naturally as glycoproteins, as do mammalian GGHs (13). Even though mammalian GGHs have four or more potential N-glycosylation sites and are heavily glycosylated, the recombinant proteins produced in E coli have similar biochemical characteristics to their glycosylated counterparts (48 -50). It is, therefore, likely that the E. coli-derived Arabidopsis GGHs used in our study faithfully mirror the properties of the enzymes as they exist in planta.
Our finding that pea GGH activity is vacuolar is supported by the recent detection of AtGGH2 and AtGGH3 proteins in Arabidopsis leaf vacuoles (42) and fits with the presence of signal peptides, which occur in many vacuole-associated proteins (42,51) and participate in vacuolar sorting (52). The vacuolar site parallels the lysosomal location of intracellular mammalian GGH (13,47), because lysosomes and central vacuoles of plants are both lytic compartments (53). An exclusively vacuolar location seems hard to reconcile with previous reports that GGH is cytosolic or extracellular, but this is not the case. The cytosolic location was inferred by fractionating extracts prepared by chopping, which breaks many vacuoles, and in fact 19% of the GGH activity was in the vacuole fraction (11). The apoplastic location was based on there being a GGH-like protein in extracellular wash fluid of soybean leaves; GGH activity was not assayed (12). Isolation of the corresponding cDNA (8) and cognate ESTs has since shown that this protein, unlike all other plant GGH sequences, lacks two catalytically essential residues (Cys-110 and His-220 in the human enzyme) (43). It, thus, appears that soybean cells secrete into the apoplast an unusual GGH-like protein, not an active GGH enzyme.
Our analyses of organellar folates show that each organelle has a distinct folate profile. The mitochondrial profile is dominated by 5-formyl-THF, which is not a C 1 donor but an inhibitor of folate-dependent enzymes (54). 5-Formyl-THF is recycled to the active C 1 folate pool by a cycloligase that is mitochondrial in plants (55). The chloroplast folate pool is richest in 5,10-methenyl-THF (which is generated during our analysis from 10-formyl-THF) but also contains 5-methyl-THF, whose occurrence in chloroplasts has previously been inferred from the presence of the enzyme that uses it, methionine synthase (56). The folate pools of both leaf and storage root vacuoles proved to consist almost entirely of 5-methyl-THF, suggesting a storage role for this folate. 5-Methyl-THF seems a plausible candidate for storage because it is quite stable and is readily converted to other C 1 folates via the reversible MTHFR reaction in plants (32). Although vacuolar storage pools of folate have not been reported before, their existence might have been predicted from the lack of correlation between folate levels and C 1 metabolic activity and the wide variation in folate levels between comparable tissues from different species (57). Beet roots are a good example. Although metabolically quiescent their folate levels are as high as those in green leaves and far higher than those of other storage roots (57). FIG. 4. Types of folates in organelles from pea leaves. Mitochondria, chloroplasts, and vacuoles were purified by density gradient centrifugation and shown to be free of significant contamination by other fractions, as in Fig. 3. Mesophyll protoplasts were analyzed for comparison. Folates were deglutamylated before HPLC analysis with electrochemical detection. Data are the means and S.E. values from three independent organelle or protoplast preparations. In addition to the folates shown, small amounts (ϳ2% of total folate) of folic acid and 10-formyldihydrofolate were found in chloroplasts and mitochondria, respectively. 5-CH 3 -THF, 5-methyl-THF; 5-CHO-THF, 5-formyl-THF; 5,10ACH-THF, 5,10-methenyl-THF.
The observed polyglutamylation of vacuolar folates is paradoxical because the glutamyl tail is not expected to survive long in the presence of the GGH activity, as the following calculation shows. For pea leaves, assuming that the vacuole is 70% of the water volume (58) and that total folate content is 5 nmol g Ϫ1 fresh weight (17), the vacuolar folylpolyglutamate level (Table  III and Fig. 5) would be ϳ1 nmol ml Ϫ1 . The vacuolar GGH activity (2.3 nmol min Ϫ1 mg Ϫ1 protein, see Fig. 3) is equivalent to ϳ8 nmol min Ϫ1 ml Ϫ1 vacuolar contents at V max or ϳ4 nmol min Ϫ1 ml Ϫ1 at a folate concentration of ϳ1 nmol ml Ϫ1 (estimated from the vacuolar folate concentration above, the folate and GGH contents expressed per unit protein, and the K m values for the Arabidopsis enzymes). Under steady state conditions, the vacuolar folyl polyglutamates, therefore, have a predicted half-life of ϳ7 s. A similar calculation for beet root vacuoles indicates a polyglutamate half-life of ϳ5 min. These calculations assume that pea and beet GGHs behave similarly to a mixture of the Arabidopsis enzymes. Partial characterization of the activities in pea and beet vacuole extracts supported this assumption. The calculations also assume that GGH and folyl polyglutamates show no tendency to be segregated into distinct vacuole subpopulations within cells.
Chloroplasts and mitochondria contain isoforms of folylpolyglutamate synthetase (6) as well as the ATP that this enzyme requires. The presence of polyglutamates in these organelles is, therefore, easily explained by in situ synthesis from monoglutamyl folates. Not so for vacuoles, which most probably contain neither the synthetase (6) nor ATP (59). For vacuoles it is thus necessary to invoke, as for mammalian lysosomes (23,24), import of folylpolyglutamates by a carrier-mediated process.
The co-occurrence of folylpolyglutamates and GGH in vacuoles could be explained by the presence of a potent GGH inhibitor or by folate-binding proteins that protect polyglutamates from hydrolysis. There is so far no experimental evidence for or against either possibility. However, inhibiting GGH action on TABLE III Total folate and 5-methyl-THFcontents of pea leaf mesophyll protoplasts and vacuoles Total folate and 5-methyl-THF were determined in three independent protoplast preparations and in the vacuoles derived from them and expressed relative to the activities of the vacuolar marker ␣-mannosidase. One ␣-mannosidase unit (U) ϭ 1 mol of substrate hydrolyzed min Ϫ1 .   folylpolyglutamates would also stop pABA polyglutamate hydrolysis and so could disrupt folate recycling. Folate-binding proteins are, therefore, perhaps a more attractive solution. Various such proteins have been characterized in mammals (60) and shown to protect bound folates against both cleavage of the polyglutamyl tail and oxidative degradation (1,39).