Purification, Characterization, and Cloning of a Spodoptera frugiperda Sf9 β-N-Acetylhexosaminidase That Hydrolyzes Terminal N-Acetylglucosamine on the N-Glycan Core*

Paucimannosidic glycans are often predominant in N-glycans produced by insect cells. However, a β-N-acetylhexosaminidase responsible for the generation of paucimannosidic glycans in lepidopteran insect cells has not been identified. We report the purification of a β-N-acetylhexosaminidase from the culture medium of Spodoptera frugiperda Sf9 cells (Sfhex). The purified Sfhex protein showed 10 times higher activity for a terminal N-acetylglucosamine on the N-glycan core compared with tri-N-acetylchitotriose. Sfhex was found to be a homodimer of 110 kDa in solution, with a pH optimum of 5.5. With a biantennary N-glycan substrate, it exhibited a 5-fold preference for removal of the β(1,2)-linked N-acetylglucosamine from the Manα(1,3) branch compared with the Manα(1,6) branch. We isolated two corresponding cDNA clones for Sfhex that encode proteins with >99% amino acid identity. A phylogenetic analysis suggested that Sfhex is an ortholog of mammalian lysosomal β-N-acetylhexosaminidases. Recombinant Sfhex expressed in Sf9 cells exhibited the same substrate specificity and pH optimum as the purified enzyme. Although a larger amount of newly synthesized Sfhex was secreted into the culture medium by Sf9 cells, a significant amount of Sfhex was also found to be intracellular. Under a confocal microscope, cellular Sfhex exhibited punctate staining throughout the cytoplasm, but did not colocalize with a Golgi marker. Because secretory glycoproteins and Sfhex are cotransported through the same secretory pathway and because Sfhex is active at the pH of the secretory compartments, this study suggests that Sfhex may play a role as a processing β-N-acetylhexosaminidase acting on N-glycans from Sf9 cells.

The hexosaminidase activity of insects and insect cells is of particular interest because of the role that the enzyme may play in altering the structures of N-glycans generated by these cells. The N-glycan synthesis pathway in insects differs from that in mammals in that insects and insect cells produce appreciable amounts of paucimannosidic glycans (reviewed in Ref. 5). The intracellular N-glycan processing pathway in the endoplasmic reticulum of insects has been observed to include the addition of a Glc 3 Man 9 GlcNAc 2 group to the acceptor Asn residue, followed by the subsequent trimming of the initial oligosaccharide to generate Man 5 GlcNAc 2 . Insect cells also contain significant levels of N-acetylglucosaminyltransferase I, which adds a GlcNAc residue to the Man␣(1,3) branch, followed by the removal of two Man residues to produce the GlcNAc␤(1,2)Man␣(1,3)(Man(1,6))-Man␤(1,4)GlcNAc␤(1,4)GlcNAc structure. Unlike mammalian cells, which subsequently modify this intermediate to yield complex, often sialylated oligosaccharides, insect cells typically hydrolyze the nonreducing terminal GlcNAc residue attached to the Man␣(1,3) branch of the N-glycan core by the action of a hexosaminidase, leading to generation of paucimannosidic structures with one or two Man attachments. Indeed, N-glycans that do not contain this GlcNAc residue are often prevalent on glycoproteins expressed in insect lines such as MB-0503 cells from Mamestra brassicae (cabbage moth), Sf21 cells from Spodoptera frugiperda (fall armyworm), and BmN cells from Bombyx mori (silk moth) (6); cells from Trichoplusia ni (cabbage looper) (7,8); and Ld652Y cells from Lymantria dispar (gypsy moth) (9); and in adult bodies of Drosophila melanogaster (10).
Hexosaminidases are widely distributed in insects, including species of Lepidoptera, Coleoptera, Hemiptera, and Orthoptera (11) and Diptera (12). This activity was found not only in tissues, but also in the blood and molting fluid of B. mori L. (13) and in the secretion fluid of female accessory glands of Ceratitis capitata (mosquito) (14). Several studies have reported hexosaminidase activity in culture media as well as in cell extracts of lepidopteran insect cell lines such as S. frugiperda Sf9, T. ni TN-368, Malacosoma disstria MD108, and B. mori cells (15); T. ni Tn-5B1-4 cells (16); and a dipteran insect cell line, D. melanogaster Kc cells (17,18). The culture medium of Culex quinquefasciatus (southern house mosquito) cells was also found to contain hexosaminidase activity (19). A few studies have reported the purification of a hexosaminidase from the hemolymph (20) and larval integument tissue (21) of B. mori; the larval or pupal molting fluid, hemolymph, and integument tissue of Manduca sexta (tobacco hornworm) (11,22); the secretion fluid of female accessory glands of C. capitata (14); the culture medium and cell extract of D. melanogaster Kc cells (17); and the culture medium of C. quinquefasciatus (19). Two closely related genes that encode insect hexosaminidases were cloned from B. mori (21) and M. sexta (23). The former is known to be an exochitinase. However, because these studies used synthetic substrates and not N-glycan substrates in the enzyme assay, it is not known whether or not previously purified or cloned insect hexosaminidases can hydrolyze the terminal GlcNAc residue linked to the N-glycan core. In addition to the above-mentioned studies, a microsomal membrane-associated hexosaminidase activity in lepidopteran insect cells that can catalyze such a reaction has been reported (24), and the importance of a cellular hexosaminidase activity in N-glycan processing in insect cells was supported by the finding of an inverse relationship between the level of cellular hexosaminidase activity and the level of GlcNAc-containing N-glycans in glycoproteins expressed in S. frugiperda Sf9 cells and Estigmene acrea cells (25).
In this study, we describe the purification of a hexosaminidase from the culture broth of S. frugiperda Sf9 cells (Sf hex). This enzyme is a homodimer of 110 kDa with maximum activity at pH 5.5 and was found to preferentially remove terminal GlcNAc residues from the Man␣(1,3) branch of the N-glycan core. The N-terminal sequence of the purified enzyme was subsequently used to isolate two corresponding cDNA clones from S. frugiperda mRNA. When introduced into insect cells, the Sf hex cDNA resulted in enhanced hexosaminidase activity in the medium. On the basis of our characterization of Sf hex, we believe that this hexosaminidase from lepidopteran insect cells is capable of removing the terminal GlcNAc residue linked to the N-glycan core and therefore of generating paucimannosidic N-glycans. The identification of this type of hexosaminidase not only contributes to a better understanding of N-glycan processing in insect cells, but will be important in engineering insect cells capable of generating complex N-glycans that can be used for baculovirus expression of heterologous mammalian proteins (26,27).
Oligosaccharides-The structures of oligosaccharides used in this work are shown in Table 1. The oligosaccharides were derivatized with 2-aminopyridine (PA) by the method of Kondo et al. (28). GnGn-PA was prepared from a PA derivative of a desialylated biantennary oligosaccharide prepared from human transferrin (Sigma). Briefly, sialic acids were removed by treating the PA-derivatized sialylated biantennary oligosaccharide in 20 mM HCl for 1 h at 80°C, and then terminal Gal residues were removed with jack bean ␤-galactosidase. MGn-PA and GnM-PA were prepared from PA derivatives of two positionally isomeric monogalactosylated biantennary oligosaccharides from human IgG (29) by sequential digestion with jack bean ␤-N-acetylglucosaminidase, ␤-galactosidase, and bovine kidney ␣-L-fucosidase. M 3 -PA and M 5 -PA were prepared from quail ovomucoid (30) and bovine pancreatic ribonuclease B, respectively. M 5 Gn-PA was synthesized from M 5 -PA by the action of N-acetylglucosaminyltransferase I (a kind gift from Dr. H. Schachter). (GlcNAc) 3 -PA was prepared from tri-N-acetylchitotriose. All PA-derivatized oligosaccharides were successively purified using normal-phase (Amide-80) and reversed-phase (Shim-pack CLC-ODS) HPLC columns before use.
Enzyme Assay-When 4-methylumbelliferyl glycosides were used as substrates, the substrates (10 nmol) were incubated in a 96-well plate with an appropriately diluted enzyme in 100 l of 50 mM sodium citrate/phosphate buffer (pH 5.5) at 37°C for 30 min. The reaction was quenched by the addition of 200 l of 0.4 M glycine/NaOH buffer (pH 10.5), and released 4-methylumbelliferone was measured by fluorescence (excitation ϭ 355 nm and emission ϭ 460 nm) using a VICTOR Model 1420 multilabel counter (Wallac, Gaithersburg, MD) in a fluorometric mode. When p-nitrophenyl 2-acetamido-2-deoxy-␤-D-glucopyranoside (pNP-GlcNAc) was used as the substrate, the substrate was incubated in a 96-well plate with an appropriately diluted enzyme in 100 l of 30 mM sodium citrate/phosphate buffer (pH 5.5) at 37°C for 30 min. The reaction was terminated as describe above, and released p-nitrophenol was measured at 415 nm using a Benchmark microplate reader (Bio-Rad). When PA-derivatized oligosaccharides were used as substrates, the reaction mixture contained substrates (500 pmol in 5 l), an appropriately diluted enzyme solution (10 l), and 0.1 M sodium citrate/phosphate buffer (pH 5.5; 10 l). The mixture was incubated at 37°C for predetermined periods of time. The reaction was terminated by heating the sample for 5 min in boiling water and centrifuged to obtain a clear supernatant. Substrates and products were analyzed using a normal-phase HPLC column (Amide-80). Enzyme activity was determined by the decrease of a substrate or the production of a product upon reaction. When GnGn-PA was used to determine the branch specificity of hexosaminidase, the enzyme reaction mixture was first subjected to normal-phase HPLC to separate the substrate, mono-N-acetylglucosaminylated M 3 -PA products, and the M 3 -PA product. A fraction containing mono-N-acetylglucosaminylated M 3 -PA products was isolated and then subjected to reversed-phase HPLC to separate MGn-PA and GnM-PA. When human asialo-or aglactotransferrin was used as the substrate, the protein (24 g) was incubated with an appropriately diluted enzyme solution (1 l) in 20 mM sodium citrate/phosphate buffer (pH 5.5) at 37°C for 5.5 h. After the reaction, 6-Omethylgalactose was added to the sample as an internal standard, and the sample was heated for 5 min in boiling water and centrifuged to obtain a clear supernatant. Released GlcNAc was measured by high-performance anion-exchange chromatography using a CarboPac PA20 column (3 ϫ 150 mm; Dionex Corp., Sunnyvale, CA) and 10 mM sodium hydroxide as an eluent. Sugars were detected by pulsed amperometry. One unit of hexosaminidase is defined as the amount of enzyme required to catalyze the release of 1 mol of terminal GlcNAc residue from substrates in 1 min at 37°C.
Cell Culture-Serum-free adapted Sf9 cells (Invitrogen) were routinely grown in serum-free Sf-900 II SFM medium in shaker flasks at 140 rpm and 27°C. The cells were passaged every 4 days at a seeding density of 0.8 ϫ 10 6 cells/ml.
Cell Extract and Supernatant Preparation-The suspension culture was typically harvested at 96 h post-seeding, and a clarified cell culture supernatant and a cell pellet were obtained by centrifugation at 350 ϫ g for 10 min. The cell pellet was washed with chilled phosphate-buffered saline (PBS) (Invitrogen). Cells were lysed by resuspending the cell pellet in chilled PBS containing 0.5% Nonidet P-40, followed by two cycles of sonication with a Tekmar sonic disruptor for 30 s at a 50% duty cycle and a power setting of 5. The cell debris was removed by centrifugation, and the clear extract was used for analysis.
Purification of Hexosaminidase-At 96 h post-seeding, the suspension culture (1.8 liters) was centrifuged at 350 ϫ g for 10 min, and the cell-free supernatant was collected. Purification was carried out at 4°C. To the clarified cell culture supernatant were added sodium chloride and sodium citrate to final concentrations of 1 M and 10 mM, respectively, and the pH was adjusted to 6.0 with a dilute sodium hydroxide solution. The sample was loaded at a flow rate of 5 ml/min onto the GlcNAcamidine affinity column (2.5 ϫ 6 cm, 30 ml) pre-equilibrated with 10 mM sodium citrate buffer (pH 6.0) containing 1 M sodium chloride. After washing the column with 3 bed volumes of the same buffer, bound protein was eluted with 10 mM sodium citrate/phosphate buffer (pH 7.0) containing 1 M D-Gl-cNAc at a flow rate of 5 ml/min by pumping the elution buffer in the reverse direction, and 5-ml fractions were collected. Hexosaminidase activity in each fraction was measured with MU-GlcNAc as the substrate, and the protein concentration was determined by the method of Bradford (34) using bovine serum albumin as a standard.
SDS-PAGE-The purified Sf hex protein (0.5 g) was analyzed by SDS-PAGE (10% acrylamide gel) under reducing conditions, and proteins were visualized by staining with Coomassie Brilliant Blue R-250.
Gel Filtration Chromatography-The purified Sf hex protein (30 g in 0.5 ml) was applied to a Sephacryl S-200 HR column (1.6 ϫ 60 cm) pre-equilibrated with 10 mM sodium citrate/ phosphate buffer (pH 6.5) containing 0.15 M sodium chloride. After loading a sample, protein was eluted with the same buffer at a flow rate of 25 ml/h, and 1-ml fractions were collected. Hexosaminidase activity in each fraction was measured with MU-GlcNAc as the substrate as described above. The column was calibrated with bovine thyroglobulin, human IgG, bovine serum albumin, chicken ovalbumin, bovine ␤-lactoglobulin, bovine ribonuclease A, and uridine.
Deglycosylation-N-Glycans attached to Sf hex was released by digesting the purified Sf hex protein with peptide N-glycosidase F. The reaction was performed according to the manufacturer's protocol.
Sequencing of N-terminal Amino Acids-N-terminal sequence analysis of the protein sample was carried out by Edman degradation using an Applied Biosystems Procise Model 494A protein sequencer.
Isolation of a cDNA Clone Encoding Sfhex and DNA Sequencing-Aligning the amino acid sequences of hexosaminidases from human, mouse, D. melanogaster, B. mori, M. sexta, and T. ni showed a highly conserved internal peptide (H(L/M)GG-DEV, amino acids (aa) 318 -324 of the human hexosaminidase ␣-chain) that forms part of the catalytic domain. This segment is common to all of these hexosaminidases and many other hexosaminidases in the GH20 family. Degenerate primers derived from both the N-terminal (LSIVNPGPQYPPTKGSIWPRP) and internal sequences were used for reverse transcription-PCR of Sf9 RNA to amplify a cDNA corresponding to a portion of the Sf hex gene. We reasoned that, although the downstream primer is common to all known hexosaminidase proteins, the upstream primer is specific for the Sf hex gene, and therefore, only the cDNA corresponding to the Sf hex gene would be amplified. The forward primer SF1Ј (5Ј-CACTAAGCTTAAYCCNG-GNCCNCARTAYCC) contained a HindIII site (shown in italics) and sequence corresponding to aa 23-29 (Supplemental Fig. S3). The reverse primer SF5Ј (5-AGTGAAGCTTACYT-CRTCNCCNCCNADRTG) contained a HindIII site (shown in italics) and sequence corresponding to aa 333-339. Total RNA prepared by the TRIzol method (Invitrogen) from Sf9 cells treated with amplification-grade DNase I (Invitrogen) was used as the template. Reverse transcription-PCR was performed using the 3Ј-SMART RACE kit (Clontech) to perform firststrand cDNA synthesis with 0.6 g of template RNA. Subsequently, 2.5 l (of the 110-l reverse transcription reaction) was introduced into a 100-l PCR at the following cycle settings: 94°C for 5 min; 40 cycles at 94°C for 1 min, 55°C for 1.5 min, and 72°C for 2 min; 72°C for 10 min; and hold at 4°C. PCR reagents were purchased from Applied Biosystems, and PCR was performed using AmpliTaq Gold in 3.5 mM MgCl 2 with an Applied Biosystems GeneAmp 2400 thermal cycler. The 948-bp product was subcloned into pUC18 and sequenced on both strands using BigDye terminators (PerkinElmer Life Sciences) by the Nucleic Acid/Protein Core Research Facility of the Children's Hospital of Philadelphia. The DNA sequence of the partial cDNA fragment matched the seven N-terminal amino acids used for designing the upstream degenerate primer, as well as the next 10 amino acids of the determined N-terminal sequence of Sf hex, confirming that the desired cDNA had been isolated. The full-length cDNA for the Sf hex gene was obtained by performing both 5Ј-and 3Ј-RACE using the SMART RACE cDNA amplification and BD Advantage 2 PCR kits (Clontech). For 3Ј-RACE, first-strand cDNA synthesis was performed as described above, and 2.5 l (of 110 l) was introduced into a 50-l PCR using the gene-specific upstream primer SF6 (5Ј-GCTCCTGGGGCGTTGCGTATCCAA, aa 280 -288) and the universal primer UPM as the downstream primer at the following cycle settings: 94°C for 5 min; five cycles at 94°C for 5 s and 72°C for 3 min; five cycles at 94°C for 5 s, 70°C for 10 s, and 72°C for 3 min; 30 cycles at 94°C for 5 s, 68°C for 10 s, 72°C for 3 min, and 72°C for 10 min; and hold at 4°C. A 900-bp fragment was obtained with high background. Nested PCR was performed using 1 l of the initial PCR in a 50-l reaction with the upstream primer NSF6 (5Ј-CGCGAATTGG-GATTGGGACCAATGGA, aa 289 -306) and the UPM universal primer as the downstream primer at the same cycle settings as before, except that only 25 cycles were performed. The resultant 840-bp fragment was purified using Geneclean (BIO 101, Inc., Vista, CA) and subcloned into pTOPO (Invitrogen) according to the manufacturers' instructions, except that the ligation reaction was for 1 h at room temperature. For 5Ј-RACE, first-strand synthesis was performed according to the manufacturer's instructions in the presence of the SMART II oligonucleotide, and 2.5 l (of 110 l) was used in a 50-l PCR with the UPM universal primer as the upstream primer and primer SF7 (5Ј-AGC-CCTCTTAGCACGCCCCATATGGAGT, aa 147-156) as the downstream primer at the following cycle settings: 94°C for 5 min; five cycles at 94°C for 5 s and 72°C for 3 min; five cycles at 94°C for 5 s, 70°C for 10 s, and 72°C for 3 min; 35 cycles at 94°C for 5 s, 68°C for 10 s, and 72°C for 3 min; 72°C for 10 min; and hold at 4°C. The resultant 519-bp fragment was subcloned into pTOPO. DNA sequencing of the fragments generated by both 3Ј-and 5Ј-RACE yielded the sequence of a full-length 1784-bp cDNA that terminated in a poly(A) tail. A full-length coding region of the Sf hex gene was obtained by reverse transcription-PCR using the forward primer SF9 (5Ј-CACTGGATCCGCCA-TCATGTTACGGCACGTAATATTGTTATTC, containing a BamHI site (shown in italics), a Kozak sequence (shown in boldface), and sequence corresponding to the first nine codons of the Sfhex gene) and the reverse primer SF10 (5Ј-AGTGGAATTCT-CACTAAAAGTAATTCCCTGTTACGCAAAA, containing an EcoRI site (shown in italics), two in-frame stop codons (shown in boldface), and sequence representing the last eight codons of the Sf hex gene) with the first-strand cDNA from the 3Ј-RACE as the template. PCR was performed in a 100-l reaction with AmpliTaq Gold polymerase and 3.5 mM MgCl 2 using reagents from Applied Biosystems in a GeneAmp 2700 thermal cycler at the following cycle settings: 94°C for 5 min; 30 cycles at 94°C for 1 min, 55°C for 1.5 min, and 72°C for 2 min; 72°C for 10 min; and hold at 4°C. The 1696-bp product was subcloned into the baculovirus vector pBlueBac4.5 (Invitrogen). DNA sequencing of multiple Sf hex cDNA clones revealed that Sf9 cells contain two closely related genes that encode nearly identical proteins (versions A and B) (Supplemental Fig. S3).
Expression of a Recombinant Hexosaminidase-Sf9 cells in mid-log phase were plated in tissue culture-treated 100-mm dishes at a density of 1 ϫ 10 7 cells/dish. After the cells adhered to the plate, the medium was removed by aspiration, and 4 ml of fresh medium and 1 ml of a recombinant baculovirus (AcSf hex) stock solution were added. After 30 min, 5 ml more medium was added. A similar procedure was also followed for control cultures with a mock baculovirus A-35 stock (1 ml) or with Sf-900 II SFM medium (1 ml). At Day 5, a culture supernatant was collected, and hexosaminidase activity was measured with GnGn-PA as the substrate.
V5-His 6 -tagged Sf hex-Construction of a plasmid expressing Sf hex protein tagged at the C terminus with both a V5 epitope and His 6 was performed by PCR using the forward primer SF9 and the reverse primer SF13 (5Ј-AGTGGAATTCTGAAAG-TAATTCCCTGTTACGCAAAA-3Ј, containing an EcoRI site (shown in italics), an extra two bases (shown in boldface) to keep the reading frame, and sequence representing the last eight codons of the Sfhex gene without the stop codon) with pBlueBac-Sfhex-3 (version A) as the template. PCR was performed as described above, except that the reaction contained 2.5 mM MgCl 2 , and only 12 cycles were carried out. The resultant Sfhex coding region was cloned into the BamHI/EcoRI sites of the pIB/V5-His vector (Invitrogen), creating pIB/V5-His-Sfhex.
Confocal Microscopy-Sf9 cells in mid-log phase were seeded onto Lab-Tek Chambered coverglass (Nalge Nunc International, Naperville, IL) at a density of 0.4 ϫ 10 6 cells/chamber and allowed to adhere for 40 min at room temperature. A transfection mixture was made by adding and gently mixing 1.2 l of Cellfectin (Invitrogen) to 200 l of Sf-900 II SFM containing 2 l of 1 g/l pIB/V5-His-Sfhex plasmid or control pIB/V5-His plasmid. The medium was aspirated; the transfection mixture (preincubated for 15 min at room temperature) was added; and the plates were rocked at six side-to-side motions/min for 4 h at room temperature. After 4 h, fresh Sf-900 II SFM (0.4 ml) was added, and the slides were incubated at 27°C for an additional 44 h. The medium was then removed, and the cells were gently washed with PBS. The cells were fixed with 10% (v/v) neutral buffered formalin (Richard-Allan Scientific) for 20 min and washed with PBS, and any remaining aldehyde was quenched by incubation for 4 min in 50 mM NH 4 Cl in PBS (1 ml). The cells were then washed twice with PBS and permeabilized with 0.05% Triton X-100 in PBS for 3 min. The fixed cells were washed again with PBS and blocked overnight with PBS and 8% bovine serum albumin (BSA), followed by incubation with a 1:1000 dilution of rabbit anti-V5 polyclonal antibody (Abcam, catalog no. ab9116) in PBS and 2% BSA for 2 h. The cells were then washed four times with PBS and incubated in a 1:1000 dilution of Alexa Fluor 546-conjugated goat anti-rabbit IgG antibody in PBS and 2% BSA for 1 h. The Golgi apparatus was stained by incubating permeabilized cells with 5 M 6-((N-(7-nitrobenz-2-oxa-1,3-diazol-4-yl)amino)hexanoyl)sphingosine⅐BSA complex (Molecular Probes, Eugene, OR) for 30 min at 4°C, followed by a PBS wash and incubation in PBS and 2% BSA for 1 h. After staining, the cells were examined under a Carl Zeiss LSM 510 META confocal laser scanning microscope equipped with an imaging system.
Immunoblotting-Sf9 cells in mid-log phase were seeded into 60-mm tissue culture-grade Petri plates at a density of 2.25 ϫ 10 6 cells/plate and allowed to adhere for 40 min at room temperature. The medium was removed by aspiration; 1 ml of the transfection mixture containing the pIB/V5-His-Sf hex plasmid or the control pIB/V5-His plasmid (prepared as described above) was added; and the plates were rocked at six side-to-side motions/min at room temperature. After 4 h, fresh Sf-900 II SFM medium (2 ml) was added, and the plates were incubated at 27°C. Cells and medium were harvested at 1, 2, or 3 days post-transfection. A cell extract was prepared by resuspending the cells in 0.2 ml of 2ϫ SDS denaturation buffer (0.1 M Tris-HCl (pH 6.8) and 4% (w/v) SDS containing 5% (v/v) 2-mercaptoethanol) using a cell scraper, diluted with an equal volume of water, and sonicated for 30 s at a 50% duty cycle using a Tekmar sonic disrupter. The cell extract was then immediately boiled for 5 min. The medium (3 ml) was incubated with phenylmethylsulfonyl fluoride and N-ethylmaleimide (each 1 mM) at room temperature for 15 min and dialyzed overnight against 5 mM Tris-HCl (pH 6.8) containing the same protease inhibitors at 4°C to remove excess salts. The sample was lyophilized, reconstituted in 0.2 ml of ice-cold water, mixed with an equal volume of 2ϫ SDS denaturation buffer, and boiled for 5 min.
Proteins were separated by SDS-PAGE (12% acrylamide gel) and transferred onto a nitrocellulose membrane. V5-His 6tagged Sf hex was detected with His 5 -horseradish peroxidase conjugate (Qiagen Inc.) and a SuperSignal West Pico chemiluminescence kit (Pierce) according to the manufacturers' instructions.
Homology Search and Phylogenetic Analysis-A BLAST search was performed against the complete ExPASy/UniPro-tKB Database of the Swiss Institute of Bioinformatics using the BLAST network service. A multiple sequence alignment was performed with the ClustalX program (35) using the Gonnet series protein weight matrix provided with the program. An unrooted phylogenetic tree was generated using the neighborjoining method (36), and 1000 bootstrapping trials were performed.

RESULTS
Hexosaminidase Activity in the Culture Medium of Sf9 Cells-Insects or insect cells are known to contain exo-type chitinolytic hexosaminidases (exochitinases) that release GlcNAc residues from the nonreducing terminus of chito-oligosaccharides (11,17,20,21). It is also known that several insect cell lines contain hexosaminidases that can hydrolyze the terminal GlcNAc residue linked to the N-glycan core (24). Because both chitinolytic and non-chitinolytic enzymes can hydrolyze synthetic substrates such as pNP-GlcNAc and MU-GlcNAc, it was essential to measure activities against an N-glycan substrate, GnGn-PA, and a chito-oligosaccharide substrate, (GlcNAc) 3 -PA (see Table 1 for structures), to differentiate these two types of enzyme activities. The PA-derivatized oligosaccharide substrates were incubated with either the culture medium or the extract of Sf9 cells harvested at 96 h post-seeding, and the reaction products were analyzed on an Amide-80 HPLC column. Shown in Fig. 1 (A and B) are HPLC chromatograms of the hydrolysis products of GnGn-PA and (GlcNAc) 3 3 -PA were hydrolyzed in the culture medium of Sf9 cells in a time-dependent manner, with initial velocities of 0.14 pmol/min (t1 ⁄ 2 ϭ 40 h) and 14 pmol/min (t1 ⁄ 2 ϭ 0.4 h), respectively. The data shown in Fig. 1 (C and D) fit well (R 2 Ͼ 0.99 for both cases) with simulated curves assuming pseudo first-order reactions for the hydrolysis. Consequently, we found that the hexosaminidase activity in the culture medium putatively responsible for the hydrolysis of the GlcNAc residue(s) in GnGn-PA was stable for up to 50 h at 37°C. Whereas the hexosaminidase(s) in the culture medium showed 1.5-fold higher activity at pH 5.5 than at pH 6.5 (an actual pH of the culture medium) against GnGn-PA, it showed 4.3-fold higher activity at pH 6.5 than at pH 5.5 against (GlcNAc) 3 -PA.
To determine whether the hexosaminidase in question is normally secreted by viable Sf9 cells, the level of this enzyme activity in the medium was monitored at various time points of the culture in relation to cell viability and total cell numbers (Fig. 2). The number of viable cells continually increased up to Day 6. However, whereas cell viability was nearly 100% until Day 4, it started to decline on Day 5. In agreement with a previous study with T. ni Tn-5B1-4 cells using pNP-GlcNAc as the substrate (16), hexosaminidase activity against GnGn-PA increased steadily until Day 6 in nearly direct proportion to the increase in number of viable cells. This direct proportion between the hexosaminidase activity and number of viable cells is reflected in the relative constant level of the specific activity (on a per cell basis) up to Day 6. After that time point, the enzyme activity remained at a constant level until Day 8, whereas the number of viable cells dropped quickly after Day 6. These data suggest that the hexosaminidase(s) that hydrolyzes GnGn-PA is secreted into the medium by viable Sf9 cells and not released by the cells upon lysis.
The hexosaminidase activity measured at pH 6.5 with GnGn-PA was 0.8 Ϯ 0.06 and 1.4 Ϯ 0.2 microunits/1 ϫ 10 6 cells in the culture medium and cell extract, respectively, whereas the exochitinase activity measured at the same pH with (GlcNAc) 3 -PA was found almost exclusively in the cell culture broth (190 microunits/1 ϫ 10 6 cells in the medium versus 8 microunits/1 ϫ 10 6 cells in the cell extract). This suggests that Sf9 cells express at least two different hexosaminidases: one that prefers N-glycans and one that prefers chito-oligosaccharides, with the latter being preferentially secreted.
Purification of a Secreted Hexosaminidase from Sf9 Cells-Although the extract of Sf9 cells contained more hexosaminidase activity capable of hydrolyzing the terminal GlcNAc residue on GnGn-PA compared with the Sf9 culture broth, we were unable to purify the responsible enzyme from either a detergent extract of cells or microsomes. This may have been due to competitive N-glycans present in the extracts that may have interfered with the affinity purification or the instability of the enzyme activity caused by the lysis methods used. On the other hand, a secreted hexosaminidase showing similar activity against GnGn-PA could be purified to homogeneity.
Glycosylamidine derivatives have been developed as affinity adsorbents for several kinds of ␤-glycosidases (31,32). We synthesized several types of GlcNAc-containing affinity adsorbents for purification of a hexosaminidase from Sf9 cells. After we examined their efficacy, we found that a GlcNAc-amidine affinity adsorbent could capture a secreted hexosaminidase from Sf9 cells efficiently. Shown in Fig. 3 is an elution profile of hexosaminidase activity from the GlcNAc-amidine column monitored with a fluorogenic substrate, MU-GlcNAc. Bound  hexosaminidase activity eluted as a sharp peak with 1 M GlcNAc in the elution buffer. A low level of activity was detected in f low-through fractions. This activity was again recovered in f low-through fractions, and no activity was detected in a 1 M GlcNAc eluate when it was applied to a regenerated affinity column. Thus, the activity that did not bind to the affinity column would represent a different enzyme. The activities of hexosaminidase in the unbound fraction and the eluate with 1 M GlcNAc were measured with GnGn-PA and (GlcNAc) 3 -PA as substrates. Because GlcNAc acts as an inhibitor, it was removed, prior to the assay, from the latter fraction by repeating ultrafiltration and dilution with 10 mM sodium citrate/ phosphate buffer (pH 6). The hexosaminidase activity in each purification step was measured at pH 5.5, at which the enzyme showed its maximum activity against the N-glycan substrate. The relatively lower activity against (GlcNAc) 3 -PA in the initial culture medium in Table 2 compared with the 100-fold difference in the initial velocities described above was due a difference in the assay pH and a difference in pH dependence of the activity against GnGn-PA and (GlcNAc) 3 -PA. As shown in Table 2, ϳ40% of the hexosaminidase activity in the medium measured with GnGn-PA was found in the eluate with 1 M GlcNAc. The hexosaminidase in this fraction showed 11-fold higher activity for GnGn-PA than for (GlcNAc) 3 -PA. The bound fraction was concentrated and subjected to SDS-PAGE, followed by Coomassie Blue staining (Fig. 4). The purified enzyme (lane 2) appeared as a single band with an apparent molecular mass of 67 kDa under reducing conditions. We named this N-glycan-specific hexosaminidase Sf hex. To determine whether Sf hex is glycosylated, a portion of the purified Sf hex protein was digested with peptide N-glycosidase F, an enzyme that removes N-linked glycans, and the digest was analyzed by SDS-PAGE. As shown in lane 3, peptide N-glycosidase F-digested Sf hex migrated faster (61 kDa) than untreated Sf hex, indicating that native Sf hex is N-glycosylated.
In contrast to Sf hex, the hexosaminidase activity in the culture medium detected with (GlcNAc) 3 -PA did not bind to the affinity resin and was recovered in an unbound fraction exclusively. The enzyme(s) in this fraction showed 42-fold higher activity for (GlcNAc) 3 -PA than for GnGn-PA (Table 2). Furthermore, it could release GlcNAc residues sequentially from the nonreducing terminus of (GlcNAc) 6 -PA (data not shown).
pH Optimum, K m , and V max -Sf hex showed its maximum activity at pH 5.5 with GnGn-PA, which was 1 pH unit higher compared with bovine kidney hexosaminidase (pH 4.5) (Supplemental Fig. S1). The same optimum pH (pH 5.5 for Sf hex and pH 4.5 for the bovine enzyme) was obtained with MU- and hexosaminidase activity in the culture medium was measured using GnGn-PA as the substrate as described in the legend of Fig. 1. E and F, hexosaminidase activity expressed in microunits/1 ϫ 10 6 cells and microunits/ ml, respectively; f, total cell number in the culture; Ⅺ, cell viability. FIGURE 3. Elution profile of hexosaminidase from a GlcNAc-amidine affinity column. The culture medium of Sf9 cells (1.8 liters) was adjusted to 1 M sodium chloride and 10 mM sodium citrate, and the pH of the sample was adjusted to 6.0. The mixture was applied to a GlcNAc-amidine-immobilized Toyopearl 650M column (2.5 ϫ 6 cm) equilibrated with 1 M NaCl and 10 mM sodium citrate (pH 6.0). After loading the sample, the column was washed with the same buffer, and bound proteins were eluted with the same buffer containing 1 M GlcNAc. The point of GlcNAc addition is indicated by the arrow. Hexosaminidase activity was measured at pH 5.5 with MU-GlcNAc as the substrate, and the protein concentration was determined by the method of Bradford (34). Solid line, hexosaminidase activity; dotted line, protein concentration.

TABLE 2
Purification of hexosaminidase from the culture medium of Sf 9 cells Enzyme activity was measured at pH 5.5 with the GnGn-PA and (GlcNAc) 3 -PA substrates. Recovery of enzyme activity is shown, and recovery percentages are in parentheses.

Hexosaminidase from S. frugiperda
GlcNAc and pNP-GlcNAc. K m and V max for pNP-GlcNAc at pH 5.5 were determined to be 0.51 mM and 68.4 nmol/min/g, respectively. With GnGn-PA and other N-glycan substrates, the reaction was proportional to the substrate concentration from 4 to 20 M, suggesting that K m values for N-glycans are much higher than these concentrations. We were not able to determine the precise kinetic parameters for N-glycan substrates because of limited availability, and the assay with N-glycan substrates had to be performed below saturating substrate concentrations. Substrate Specificity-The activities of purified Sf hex for substrates with N-acetylglucosaminide, N-acetylgalactosaminide, and sulfated N-acetylglucosaminide were compared using the f luorogenic substrates MU-GlcNAc, MU-GalNAc, and MU-GlcNAc-6SO 4 (supplemental Table SI). Purified Sf hex (0.2 g) hydrolyzed MU-GlcNAc and MU-GalNAc at rates of 5.9 and 3.0 pmol/min, respectively. However, no detectable hydrolysis was observed with MU-GlcNAc-6SO 4 .
The substrate specificity of purified Sf hex was further investigated with several N-glycan substrates with terminal ␤(1,2)linked GlcNAc residues (Table 3; see Table 1 for structures). The activities shown in Table 3 were measured with 20 M substrate. Because the reaction was proportional to the substrate concentration from 4 to 20 M, relative activity will be constant within this substrate concentration range. When we compared two positionally isomeric mono-N-acetylglucosaminylated substrates (MGn-PA and GnM-PA), the terminal GlcNAc residue on the Man␣(1,3) branch was released two times faster (1.55 pmol/min/g of protein) than the GlcNAc residue on the Man␣(1,6) branch (0.71 pmol/min/g of protein). Sf hex also released the GlcNAc residue in M 5 Gn-PA at a slower rate (1.14 pmol/min/g of protein) than the corresponding GlcNAc residue in MGn-PA. When a biantennary N-glycan substrate with two terminal GlcNAc residues (GnGn-PA) was digested with Sf hex, the GlcNAc residue on the Man␣(1,3) branch was released five times faster (2.39 pmol/min/g of protein) than the GlcNAc residue on the Man␣(1,6) branch (0.46 pmol/ min/g of protein) and faster than the GlcNAc residue at the same position on the monoantennary substrate MGn-PA. In contrast, the GlcNAc residue on the Man␣(1,6) branch of GnGn-PA was released at a slower rate compared with the Glc-NAc residue at the same position in GnM-PA (0.46 versus 0.71 pmol/min/g of protein). We compared the activities of Sf hex (the hexosaminidase activity in the medium and the extract of Sf9 cells) and hexosaminidases from bovine kidney and jack bean for different N-glycan substrates (supplemental Table  SII). In all cases, MGn-PA was hydrolyzed at a faster rate than GnM-PA. The relative activity of the hexosaminidase in the culture medium of Sf9 cells showed essentially the same pattern as that obtained with the purified enzyme. In agreement with a previous study (24), the enzyme activity in the extract of Sf9 cells was highly specific for the GlcNAc residue on the Man␣(1,3) branch, with very low activity for the GlcNAc residue in GnM-PA and M 5 Gn-PA.
The activity of purified Sf hex was also examined with a glycoprotein substrate by measuring released GlcNAc after incubation of human asialo/agalactotransferrin (12 M) with purified Sf hex (3 microunits) at 37°C for 5.5 h at pH 5.5. Human transferrin has two N-glycosylation sites. The human asialo/agalactotransferrin used in this experiment contained biantennary glycans that accounted for 90% of the total N-glycans (data not shown). Therefore, the assay contained ϳ20 M biantennary glycans that terminated in GlcNAc residues. Sf hex released 25% of the terminal GlcNAc residues from human asialo/agalactotransferrin. The same amount of enzyme released 60% of the terminal GlcNAc residues from GnGn-PA under the same reaction conditions.
Inhibitors-The effect of D-GlcNAc and 2-acetamido-1,2dideoxynojirimycin on the activity of Sf hex was measured using pNP-GlcNAc as the substrate. D-GlcNAc and 2-acetamido-1,2-dideoxynojirimycin showed competitive inhibition, with K i values of 2.0 mM and 0.22 M, respectively. D-GlcNAc and 2-acetamido-1,2-dideoxynojirimycin also inhibited the activity of Sf hex, with IC 50 values of 1.8 mM and 0.23 M, respectively, when GnGn-PA was used as the substrate (Supplemental Fig. S2).
N-terminal Sequence-Protein sequencing of purified Sf hex identified the N-terminal 21 amino acids to be LSIVNPG-PQYPPTKGSIWPRP. A BLAST search using this sequence as a

TABLE 3 Substrate specificity of Sfhex for N-glycan substrates with terminal GlcNAc residues
N-Glycan substrate (500 pmol) was incubated with purified Sf hex (0.1 g) in 40 mM sodium citratephosphate buffer (pH 5.5, 25 l) at 37°C, and then reaction velocity was determined by analyzing the substrates and products by HPLC. See Table 1 for structures. Standard error in the assay was Ͻ5%.

Substrate
Product v k bits, and expect ϭ 7e Ϫ06 ). We also noticed that one of the previously purified hexosaminidases from the integument tissue of B. mori larva, called ␤ 2 -enzyme (21), has a similar N-terminal sequence, LXIVEPPEYPAXKGAIWP (where X is an unidentified amino acid).
Sf hex Subunit Characterization-The apparent molecular mass of Sf hex in solution was determined to be ϳ110 kDa by gel filtration using a Sephacryl S-200 HR column (data not shown). Because amino acid sequencing of the N terminus of purified Sf hex showed only one species of N-terminal sequence, the results suggest that Sf hex is most likely to be present as a homodimer in solution.
Isolation of a cDNA Encoding Sf hex-Information on the N-terminal amino acid sequence of purified Sf hex and a predicted internal peptide derived from a highly conserved region in the catalytic domain (Motif 3) (Fig. 5; see below) of mammalian and insect hexosaminidases was used to design degenerate primers, which were used for reverse transcription-PCR of Sf9 RNA. A 948-bp cDNA fragment was obtained whose encoded amino acid sequence perfectly matched the N-terminal sequence of purified Sf hex. A full-length cDNA clone encoding Sf hex was obtained by performing both 5Ј-and 3Ј-RACE. The full-length cDNA was 1784 bp and was predicted to encode a protein of 555 amino acids, shown in Supplemental Fig. S3 (Swiss-Prot accession number Q3LS76), with a molecular mass of 63.5 kDa. We obtained the same sized fragments from independent 5Ј-and 3Ј-RACE, and six separate full-length cDNA clones were sequenced. All full-length cDNA clones initiated the same number of base pairs upstream of the empirically determined N terminus of purified Sf hex and terminated in a poly(A) tail, confirming that they were full-length. As shown in Supplemental Fig. S3, the Sf hex protein has a characteristic hydrophobic signal sequence for entry into the secretory pathway. N-terminal sequence analysis of the mature protein indicated that an 18-aa (or 17-aa in version B; see below) signal sequence was cleaved, yielding a 61.4-kDa mature protein. This predicted molecular mass is in close agreement with the molecular size of de-N-glycosylated Sf hex (Fig. 4). The Sf hex protein has four potential N-glycosylation sites (Asn-X-(Ser/Thr)) at Asn 116 -Leu-Ser, Asn 174 -Ala-Thr, Asn 309 -Ile-Thr, and Asn 357 -Met-Thr. The protein also has an Asn 51 -Pro-Ser sequence, but Asn 51 is probably not occupied because X is a Pro residue. Because peptide N-glycosidase F treatment of purified Sf hex reduced its apparent molecular mass on SDS-polyacrylamide gels (lane 3), it is likely that the Sf hex protein is indeed glycosylated at some or all of these sites. Although we selected the conserved internal peptide sequence (H(L/M)GGDEV) to design the degenerate reverse strand primer, the sequence of Sf hex in this region is actually HVGGDEV (aa 333-339). We found two closely related genes in cDNA clones. They encode nearly identical proteins, which we will call versions A and B in this work (Supplemental Fig. S3). Versions A and B differ at 24 nucleotide positions within the 1664-bp coding region, but FIGURE 5. Multiple sequence alignment of Sf hex and mammalian hexosaminidases. A multiple sequence alignment was performed with the ClustalX program (35). Eight conserved motifs in GH20 family hexosaminidases are underlined. Amino acids involved in substrate binding and the catalytic reaction (42,43) and in the formation of disulfide bonds (44) are indicated by residue number. these result in amino acid substitutions at only three positions and one deletion in the cleaved leader sequence. Therefore, the mature Sf hex proteins share 99.5% amino acid identity. We recovered cDNA clones for versions A and B at equal frequencies, and therefore, their RNA levels were comparable.
Expression of Sf hex from Its cDNA in Sf9 Cells-To confirm that the cloned cDNA (Sf hex) actually encodes the protein that was isolated from the culture medium of Sf9 cells, we expressed Sf hex by infecting Sf9 cells with a recombinant baculovirus containing the full-length Sf hex gene (AcSf hex). Expression of Sf hex resulted in a 60-fold increase in the hexosaminidase activity measured with the GnGn-PA substrate in the culture medium (65 microunits/1 ϫ 10 6 cells) compared with the activity in the medium from Sf9 cells with no virus infection (1.1 microunits/1 ϫ 10 6 cells) and the medium from Sf9 cells infected with a blank baculovirus lacking a foreign gene (1.8 microunits/1 ϫ 10 6 cells). Recombinant Sf hex displayed the same substrate specificity with GnGn-PA, GnM-PA, MGn-PA, and M 5 Gn-PA and the same optimum at pH 5.5 as Sf hex isolated from the culture medium. These data confirm that the Sf hex gene encodes a functional hexosaminidase that is capable of hydrolyzing the terminal ␤(1,2)-linked GlcNAc residue on the N-glycan core.
Distribution of Sf hex-In the above experiment, 20 and 80% of the total (intracellular and secreted) hexosaminidase activity was found in the cell extract and medium, respectively. Because baculovirus infection may disturb normal intracellular sorting, which could potentially result in an atypical enzyme distribution, we further examined the distribution of Sf hex by expressing V5-His 6 -tagged Sf hex in Sf9 cells using the pIB/V5-His expression vector and Cellfectin. The medium and cell extract samples obtained at 1, 2, and 3 days post-transfection were subjected to SDS-PAGE, followed by Western blot analysis (Fig.  6). Although a number of background bands were evident in the cell extract from the control infection (pIB/V5-His), a significant Sf hex band was observed at ϳ70 kDa in both the cell extract and medium. Its estimated molecular mass is in close agreement with the expected size of purified Sf hex plus a C-ter-minal tag. Although Sf hex was present in both the cells and medium, a significantly higher fraction accumulated in the medium with days post-transfection, consistent with our results for the baculovirus infection. In addition, a number of proteolytically processed fragments appeared upon prolonged culture (see Days 2 and 3) for Sf hex in both the medium and cells. Interestingly, fragments at ϳ60 kDa were observed only in the medium, suggesting differences in processing of the intracellular and secreted Sf hex proteins.
Localization of Intracellular Sf hex-The intracellular distribution of Sf hex was investigated by expressing V5-His 6 -tagged Sf hex in Sf9 cells. Sf9 cells were transfected with the pIB/V5-His-Sf hex DNA and immunostained with rabbit anti-V5 antibody, followed by Alexa Fluor 546-conjugated goat anti-rabbit IgG antibody and a Golgi-specific dye, 6-((N-(7-nitrobenz-2oxa-1,3-diazol-4-yl)amino)hexanoyl)sphingosine. Stained cells were examined with a confocal microscope. Sf9 cells expressing V5-His 6 -tagged Sf hex exhibited punctate red or green fluorescence, representing tagged Sf hex (Fig. 7, A and E) or the Golgi apparatus (Fig. 7, B and F), respectively, throughout the cytoplasm. However, there was no significant overlap between the two patterns (Fig. 7, C and G). These results suggest that intracellular Sf hex localizes within vesicles throughout the cytoplasm, but that the majority of these vesicles are not part of the Golgi apparatus. No red fluorescent signal from Sf hex was observed at the cell surface.
Sequence Homology-Hexosaminidases are found in three GH families: GH3, GH20, and GH84 (1-4). The amino acid sequence of Sf hex was searched against the Pfam Protein Families Database (37) to find regions of the sequence that belong to known domain families. Sf hex was found to have a catalytic domain that is conserved in the GH20 family hexosaminidases such as mammalian lysosomal hexosaminidases. A pairwise sequence alignment using the BLAST 2 sequences program (38) at NCBI detected an identity of 52% (228 of 552 aa; score ϭ 595 bits and expect ϭ e Ϫ168 ) between Sf hex and a putative B. mori hexosaminidase (Swiss-Prot accession number Q3L6N4). Furthermore, it detected identities of 39% (213 of 535 aa; score ϭ 379 bits and expect ϭ e Ϫ103 ) and 42% (239 of 562 aa; score ϭ 418 bits and expect ϭ e Ϫ115 ) between Sf hex and the human hexosaminidase ␣-chain (39,40) and ␤-chain (41), respectively. A multiple sequence alignment was performed with the precursor forms of Sf hex (version A) and hexosaminidases from human and mouse (both ␣and ␤-chains) (Fig. 5). Amino acids in the eight motifs known as a signature for GH20 family enzymes (SPRINTS, available at umber.sbs.man.ac.uk/ dbbrowser/sprint/) are highly conserved in Sf hex. Furthermore, Arg 211 , Asp 354 , Tyr 450 , Asp 452 , and Glu 491 , which are known to be involved in substrate binding, and Asp 240 , His 294 , and Glu 355 , which form the catalytic triad in human hexosaminidase B (HexB) (42,43), are conserved. Six cysteine residues are known to form three disulfide bonds (Cys 91 -Cys 137 , Cys 309 -Cys 360 , and Cys 534 -Cys 551 ) in human HexB (44). Of these, the third disulfide bond is known to be essential for the enzyme activity of human HexB (45). All of these cysteine residues are completely conserved in Sf hex. Tyr 456 plays an important role in dimerization of two subunits in human HexB (42), where Tyr 456 of one subunit forms hydrophobic interactions with Ile 454 and Tyr 492 of the other subunit. The importance of Tyr 456 was indicated by the absence of active HexB in cultured fibroblasts of a Sandhoff disease patient with a Y456S mutation and a defect in homodimerization of the mutant ␤-chains (46). As shown in Fig. 5, Tyr 492 is replaced by Gly 436 in Sf hex, which is a homodimer of two identical polypeptides and is active.
Phylogenetic Relationship to Other Hexosaminidases-We performed an initial phylogenetic analysis using the ClustalX program (35) with ϳ200 proteins that cover almost of all known proteins in the GH20 family and found that hexosaminidases in this family can be categorized into several groups (data not shown). We then performed the analysis on selected proteins that are closely related to Sf hex. Shown in Fig. 8 is an unrooted phylogenetic tree created by the neighbor-joining method (36), based on the result of a multiple sequence alignment. Because of the lack of an accepted common ancestor, the relationship between groups as shown by the branching order at the center of the tree is unreliable, but we found that there are two groups of hexosaminidases in the 32 proteins. Each of Group 1 (branches with thick solid lines) and Group 2 (branches with dashed lines) is defined by a highly significant node, found in Ͼ90% of bootstrap trials, near the center of the tree. Whereas Group 1 includes hexosaminidases from a wide range of organisms such as mammals, insects, nematodes, slime molds, and Ascidia, Group 2 includes hexosaminidases mostly from insects and fungi. As shown in Fig. 8, Sf hex was found to be more closely related to mammalian lysosomal hexosaminidases than several other insect hexosaminidases previously reported from B. mori (21) and M. sexta (23) and in sequence data bases for T. ni, Bombyx mandarina (wild silk moth), D. melanogaster, and Anopheles gambiae.
In addition to the above-mentioned proteins, a putative hexosaminidases from B. mori (Swiss-Prot accession number Q3L6N4) and A. gambiae (Swiss-Prot accession number Q7Q0Z2) were identified by performing BLAST searches. These two Sf hex orthologs were also found to belong to Group 1, similar to Sf hex. The genome of Caenorhabditis elegans contains a single gene that encodes a hexosaminidase (Swiss-Prot accession number Q22492), and it also belongs to Group 1. However, all three Drosophila hexosaminidases were found to belong to Group 2.

DISCUSSION
Insect cells are known to produce substantial amounts of paucimannosidic N-glycans that lack the GlcNAc residue on the Man␣(1,3) branch of the N-glycan core, in contrast to mammalian cells, which often produce complex-type Nglycans with sialic acid groups present on both antennae (5). A previous study (24) reported that lepidopteran insect cell lines contain a membrane-bound "N-glycan-processing ␤-Nacetylglucosaminidase," but that study did not use purified enzyme or examine whether a similar activity was also secreted.
In this study, we have reported the purification and characterization of an insect hexosaminidase (Sf hex) that is active against N-glycans from the culture medium of Sf9 cells and have cloned two corresponding cDNAs for the enzyme. Using epitope-tagged Sf hex, we investigated its distribution between intracellular and secreted forms and its intracellular localization. We detected possible differences in intracellular processing between the secreted and intracellular forms. Because we used purified enzyme and substrates, the characterization and properties of the enzyme may differ from those determined in previous studies that used crude extracts of cells or enriched microsomal fractions for enzyme assays.
We have shown that an N-glycan-active hexosaminidase was secreted by viable Sf9 cells and accumulated in the culture medium in a time-dependent manner (Fig. 2). We found that Sf9 cells secrete at least two different hexosaminidases. One enzyme that bound to the GlcNAc-amidine affinity column showed a preference for GnGn-PA, whereas the activity in the unbound fraction showed a preference for (GlcNAc) 3 -PA. Because the unbound fraction contained a high level of exochitinase activity, this enzyme may also be responsible for the relatively low activity observed with GnGn-PA, or the unbound fraction may include another enzyme that is capable of hydrolyzing the terminal GlcNAc residue on N-glycans.
A previously cloned B. mori hexosaminidase (Swiss-Prot accession number P49010) is known to be an exochitinase (21). D. melanogaster HEXO1 and HEXO2 are also exochitinases (47). In contrast, purified Sf hex showed higher activity for the terminal GlcNAc residue on the N-glycan core than for the GlcNAc residue on chitotriose. Sf hex could hydrolyze both MU-GlcNAc and MU-GalNAc at a comparable rate. Therefore, Sf hex should be called a hexosaminidase rather than a ␤-N-acetylglucosaminidase. Humans possess three forms of hexosaminidase: HexA is a heterodimer of ␣and ␤-polypeptide chains; HexB is a homodimer of two ␤-polypeptide chains; and HexS is a homodimer of two ␣-polypeptide chains. Of these, HexA and HexS show activity with negatively charged substrates such as MU-GlcNAc-6SO 4 , but HexB does not (48,49). Sf hex showed no detectable activity with the sulfated substrate. This result, together with Sf hex sequence similarity to human lysosomal hexosaminidases, suggests that Sf hex may be an S. frugiperda ortholog of human lysosomal HexB.
Similar to a previously reported hexosaminidase activity present in Sf21 cells (24), Sf hex showed a preference for the terminal GlcNAc residue linked to the Man␣(1,3) branch of the N-glycan core rather than the GlcNAc residue linked to the Man␣(1,6) branch (Table 3). Interestingly, Sf hex showed higher activity for the GlcNAc residue on the Man␣(1,3) branch of GnGn-PA than for the GlcNAc residue at the same position in MGn-PA (Table 3). This suggests that a GlcNAc residue on the Man␣(1,6) branch either increased the affinity for the substrate or enhanced the catalytic rate. Jack bean hexosaminidase, but not bovine kidney hexosaminidase, also exhibited higher activity with GnGn-PA than with MGn-PA. On the other hand, Sf hex showed lower activity with M 5 Gn-PA compared with MGn-PA. Because the former substrate has two additional Man residues on the Man␣(1,6) branch of MGn-PA, the result suggests that one or both of the two extra Man residues sterically hinder the action of Sf hex.
We cloned a gene that encodes Sf hex using sequence information from the N terminus of purified Sf hex and a predicted internal peptide derived from a highly conserved region in the catalytic domain of mammalian and insect hexosaminidases. We identified two closely related genes encoding proteins that share Ͼ99% amino acid identity (Supplemental Fig. S3). This high degree of similarity suggests that they represent polymorphic alleles of the locus and not two different genes, as is the case for the ␣and ␤-polypeptide chains of human hexosaminidase, which share only 56% identity. Indeed, insect genomes such as the genome of D. melanogaster are noted for having a high degree of polymorphism. The amino acids of Sf hex versions A and B are different at three positions in their mature forms. We inspected their locations in a three-dimensional structure of Sf hex modeled by a homology modeling method and based on the crystal structure of human HexB (43) as the template. These three amino acids were found to be far away from the substrate-binding site (data not shown). Thus, it is unlikely that the substitution at these three positions would affect the activity and specificity of Sf hex.
C. elegans, an organism whose N-glycans are also primarily of the oligomannosidic and paucimannosidic type, has been reported to possess a membrane-associated, Man␣(1,3) branch-specific hexosaminidase activity (50). Our data base search of sequences that share homology with known GH20 family hexosaminidases revealed that C. elegans has only a single gene encoding a hexosaminidase, and therefore, it likely encodes the enzyme responsible for this activity. Our phylogenetic analysis suggested the presence of two distinct groups of hexosaminidase in the GH20 family. Group 1 contains Sf hex, its orthologs in B. mori and A. gambiae, the above C. elegans enzyme, and several mammalian lysosomal hexosaminidases. Several enzymes in this group, including Sf hex, are known to be capable of hydrolyzing the terminal GlcNAc residue on the N-glycan core. On the other hand, Group 2 contains a B. mori exochitinase and its orthologs in M. sexta, B. mandarina, T. ni, and D. melanogaster and several fungal enzymes such as a Trichoderma harzianum exochitinase and a Trichoderma virens chitobiose. Our phylogenetic analysis suggested that lepidopteran insects contain both exochitinases (in Group 2) and lysosomal enzyme-type hexosaminidases (in Group 1).
Following our initial manuscript submission and subsequent to our depositing the Sf hex sequences in the GenBank TM Data Bank, two other studies reported the characterization and cloning of hexosaminidases active against N-glycans from insect cells (47,51). One of these studies reported that the D. melanogaster FDL protein (in Group 2 in our classification) is active against an N-glycan substrate, but not a chito-oligosaccharide, and is responsible for producing paucimannosidic N-glycans in Drosophila (47). We note that the FDL protein is atypical of other Group 2 hexosaminidases in having His and Tyr at positions 212 and 550 (numbers refer to the positions in the human hexosaminidase ␤-chain), respectively, similar to Group 1 enzymes, but in contrast to the conserved Asn and Trp at these positions typical of Group 2 enzymes. Substitution of these amino acids may affect its substrate specificity. It is possible that the FDL protein evolved a separate specificity from other Group 2 proteins or that the algorithm has miscategorized this protein.
We observed intracellular Sf hex as well as more abundant secreted Sf hex using V5-His 6 -tagged recombinant Sf hex expressed in Sf9 cells (Fig. 6). This distribution cannot be directly compared with the observation of 64% intracellular activity and 34% secreted activity for the endogenous hexosaminidase activity because of differences in cell culture conditions in the two experiments. However, it is clear that some of the newly synthesized V5-His 6 -tagged Sf hex resides in Sf9 cells. Furthermore, the overexpression of heterologous Sf hex results in a significantly higher fraction of secreted protein compared with endogenous ratios. If the intracellular sorting machinery is saturated, then the fraction of a protein that is secreted will increase, a situation that is amplified when proteins such as those targeted to the lysosome are overexpressed in insect cells using cDNA constructs such as human HexB (52), human GM2 (GalNAc␤1,4(NeuAc␣2,3)Gal␤1,4Glc␤1,1Ј-ceramide) activator protein (53), and human sphingomyelinase (54).
Interestingly, HexB, normally a lysosomal protein, is used as a marker for regulated secretion in some mammalian cell lines, and its extracellular release can parallel total protein secretion in stimulated cell lines (55). Thus, lysosomal proteins may be secreted into the medium, present in the secretory pathway compartment, or active in the lysosomes depending on cell type, culture conditions, and expression levels (55). As a result, it was not surprising to observe Sf hex, a homolog of mammalian hexosaminidase, in both the intracellular and secreted fractions in this study.
The transport mechanism and proteolytic processing of mammalian hexosaminidases have been well characterized (reviewed in Refs. 56 -59). As is typical of secretory and plasma membrane glycoproteins, a pro form of hexosaminidases is transported through the cis-Golgi 3 medial-Golgi 3 trans-Golgi to the trans-Golgi network, where hexosaminidases are sorted from secretory glycoproteins by mannose 6-phosphate receptors that recognize mannose 6-phosphate on hexosaminidases. Thereafter, hexosaminidases are routed to endosomes, where hexosaminidases are released from the receptors and subsequently transported to lysosomes. A series of proteolytic and glycosidic processing events occur to the pro from of hexosaminidases in endosomes or lysosomes to form their mature structure (60 -67). Lysosomal processing of the ␣and ␤-polypeptides by proteases and glycosidases results in smaller fragments ranging in size from 6 to 56 kDa held together by disulfide bonds (60).
Proteolytic fragmentation of Sfhex was observed in both the medium and extract of Sf9 cells upon prolonged culture periods.
Although some of the fragment sizes were similar in the cell extract and medium, the medium also contained an additional fragment (ϳ60 kDa) not observed in the cell extract. This difference in Sfhex protease digestion is due either to different processing pathways for proteins in the medium and Sf9 cells or to the presence of additional nonspecific proteases in the medium. The presence of multiple fragments in the cell extracts is also consistent with possible intracellular processing observed for other intracellular hexosaminidase proteins (63), although the principal intracellular band remains the full-length protein. The observation of some similar protein fragments in the cell extracts and medium may be due to leakage of intracellular fragments or similar protease digestion in the two environments.
Examination of the intracellular location of the tagged Sf hex protein by confocal microscopy and differential interference contrast microscopy revealed the presence of punctate cytoplasmic staining. However, Sf hex did not co-localize significantly with the Golgi marker to suggest another possible principal intracellular location. The Drosophila hexosaminidase FDL was observed only minimally in the Golgi, but was found to predominate in the plasma membrane, late endosomes, and extracellular medium. The localization and secretion of both Sf hex and D. melanogaster FDL are also different from those originally reported by Altmann et al. (24), who hypothesized that the Sf21 hexosaminidase would be a membrane-bound Golgi protein. Our findings and those of Léonard et al. (47) suggest that intracellular Sf hex may localize to endosomes/ multivesicular bodies and/or possibly lysosomes. Unlike D. melanogaster FDL (47), Sf hex was not observed at the cell surface. Aumiller et al. (51) also expressed Sf9 hexosaminidase in insect cells using a baculovirus vector and noted that the protein did not co-localize with a commercial LysoTracker probe (Molecular Probes). However, this probe labels a number of acidic organelles, and lysosomal fluorescence may represent only a small fraction of the total fluorescence obtained with the dye. These authors suggested that the protein is a degradation enzyme, but not an "N-glycan-processing enzyme" (51). We believe that Sf hex, like Drosophila FDL (47), is likely to be responsible for generating paucimannosidic N-glycans in insect cells for the following reasons. 1) Homologous lysosomal HexB is responsible for hydrolysis of the terminal GlcNAc residue on the Man␣(1,3) branch of the N-glycan core for the membrane protein BA-1 expressed in mouse brain (68). 2) Sf hex is active at the pH of secretory compartments such as the trans-Golgi (pH 6.17-6.36) (69), the trans-Golgi network (pH 5.91-5.95) (70,71), and secretory granules (pH 5.2) (72,73).
3) The majority of N-glycan-containing proteins, i.e. secretory glycoproteins, are cotransported with Sf hex. 4) Drosophila flies carrying a deficiency in the chromosomal region including the fdl gene show a 7-fold reduction in paucimannosidic N-glycans (47). 5) The optimum pH of D. melanogaster FDL with GnGn-PA (pH 5.5) (47) is same as that of Sf hex. 6) A single C. elegans hexosaminidase likely to be responsible for producing paucimannosidic glycans (50) is also homologous to Sf hex. Indeed, given the absence of other likely hexosaminidase candidates against N-glycans in the genomes of either D. melanogaster or C. elegans, we hypothesize that the Sf hex and D. melanogaster FDL proteins serve the multifunctional roles of both N-glycan processing and glycan degradation. The N-glycan-processing capability of Sf hex in insects and insect cells may result from the requirement that functional Sf hex be transported along the same pathway as that of secreted glycoproteins. Similar to what has been observed for many lysosomal proteins (52)(53)(54), we propose that a fraction of Sf hex likely fails to be marked for intracellular sorting and/or to be captured by the lysosomal receptor, resulting in its secretion into the extracellular medium from which it was purified. However, as the Spodoptera genome is not yet fully sequenced, another hexosaminidase is possible within the S. frugiperda genome that may function separately or in concert to modify secreted N-glycans.
If the same hexosaminidase is used for both N-glycan processing and degradation in insects and perhaps C. elegans, the question arises as to why mammalian N-glycans retain complex structures because they are also exposed to lysosomal hexosaminidases in the secretory pathway. The difference may be due to the fact that mammalian lysosomal hexosaminidases have a lower pH optimum, pH 4.5 (Supplemental Fig. S1), so they may not be as active in the Golgi compartments prior to arrival in the lysosome. This results in insect cells having a 14-fold higher hexosaminidase activity for a GnGn-PA substrate measured at pH 6.0 compared with mammalian tissues (see Table 5 in Ref. 24). In addition, the level of ␤1,4-galactosyltransferase in the Golgi is much higher in mammalian cells (74), serving to cap the N-glycans and to protect them from the action of hexosaminidases. Indeed, we have observed that overexpression of this galactosyltransferase protects the exposed GlcNAc residue from cellular hexosaminidases (7). Finally, the intracellular transport machinery of mammals may be more efficient than that of insects in delivering hexosaminidase and other degradation enzymes to the proper degradation compartments. Interestingly, Drosophila has a homolog of mammalian mannose 6-phosphate receptors, the lysosomal enzyme receptor protein, but ligand recognition by the lysosomal enzyme receptor protein does not depend on mannose 6-phosphate (75). Therefore, a different mechanism may be used for lysosomal targeting in insects. Clearly, evolution has favored the ability of mammalian N-glycans to retain the GlcNAc residues on both biantennary branches, allowing for the generation of more complex N-glycan structures, whereas insects trim these structures to the paucimannosidic form.
With our identification of a hexosaminidase (Sfhex) that is active against N-glycans from Sf9 cells and the recent report by Aumiller et al. (51) of a second Sf9 hexosaminidase with properties very similar to those of Sfhex, it should be possible to eliminate their activity in cells using targeted RNA interference to engineer insect cells with improved ability to reproduce mammal-type N-glycans. Such experiments are in progress in our laboratories.