Carbohydrate Structures of Recombinant Human α-l-Iduronidase Secreted by Chinese Hamster Ovary Cells*

α-l-Iduronidase is a lysosomal hydrolase that is deficient in Hurler syndrome and clinically milder variants. Recombinant human α-l-iduronidase, isolated from secretions of an overexpressing Chinese hamster ovary cell line, is potentially useful for replacement therapy of these disorders. Because of the importance of carbohydrate residues for endocytosis and lysosomal targeting, we examined the oligosaccharides of recombinant α-l-iduronidase at each of its sixN-glycosylation sites. Biosynthetic radiolabeling showed that three or four of the six oligosaccharides of the secreted enzyme were cleaved by endo-β-N-acetylglucosaminidase H, with phosphate present on the sensitive oligosaccharides andl-fucose on the resistant ones. For structural analysis, tryptic and chymotryptic glycopeptides were isolated by lectin binding and reverse phase high pressure liquid chromatography; their molecular mass was determined by matrix-assisted laser desorption-time of flight mass spectrometry before and after treatment with endo- or exoglycosidases or with alkaline phosphatase. Identification of the peptides was assisted by amino- or carboxyl-terminal sequence analysis. The major oligosaccharide structures found at each site were as follows: Asn-110, complex; Asn-190, complex; Asn-336, bisphosphorylated (P2Man7GlcNAc2); Asn-372, high mannose (mainly Man9GlcNAc2, some of which was monoglucosylated); Asn-415, mixed high mannose and complex; Asn-451, bisphosphorylated (P2Man7GlcNAc2). The Asn-451 glycopeptide was unexpectedly resistant to digestion byN-glycanase unless first dephosphorylated, but it was sensitive to endo-β-N-acetylglucosaminidase H and to glycopeptidase A. The heterogeneity of carbohydrate structures must represent the accessibility of the glycosylation sites as well as the processing capability of the overexpressing Chinese hamster ovary cells.

␣-L-Iduronidase is a lysosomal hydrolase that is deficient in Hurler syndrome and clinically milder variants. Recombinant human ␣-L-iduronidase, isolated from secretions of an overexpressing Chinese hamster ovary cell line, is potentially useful for replacement therapy of these disorders. Because of the importance of carbohydrate residues for endocytosis and lysosomal targeting, we examined the oligosaccharides of recombinant ␣-Liduronidase at each of its six N-glycosylation sites. Biosynthetic radiolabeling showed that three or four of the six oligosaccharides of the secreted enzyme were cleaved by endo-␤-N-acetylglucosaminidase H, with phosphate present on the sensitive oligosaccharides and L-fucose on the resistant ones. For structural analysis, tryptic and chymotryptic glycopeptides were isolated by lectin binding and reverse phase high pressure liquid chromatography; their molecular mass was determined by matrix-assisted laser desorption-time of flight mass spectrometry before and after treatment with endo-or exoglycosidases or with alkaline phosphatase. Identification of the peptides was assisted by amino-or carboxyl-terminal sequence analysis. The major oligosaccharide structures found at each site were as follows: Asn-110, complex; Asn-190, complex; Asn-336, bisphosphorylated (P 2 Man 7 GlcNAc 2 ); Asn-372, high mannose (mainly Man 9 GlcNAc 2 , some of which was monoglucosylated); Asn-415, mixed high mannose and complex; Asn-451, bisphosphorylated (P 2 Man 7 GlcNAc 2 ). The Asn-451 glycopeptide was unexpectedly resistant to digestion by N-glycanase unless first dephosphorylated, but it was sensitive to endo-␤-N-acetylglucosaminidase H and to glycopeptidase A. The heterogeneity of carbohydrate structures must represent the accessibility of the glycosylation sites as well as the processing capability of the overexpressing Chinese hamster ovary cells.
␣-L-Iduronidase (EC 3.2.1.76), a lysosomal enzyme that participates in the degradation of dermatan sulfate and heparan sulfate, is deficient in the Hurler, Hurler/Scheie, and Scheie syndromes, collectively known as mucopolysaccharidosis I (1). In the absence of ␣-L-iduronidase, lysosomal accumulation of partially degraded glycosaminoglycans causes characteristic clinical manifestations that include corneal clouding, skeletal abnormalities, cardiovascular disease, limited joint mobility, and organomegaly. Mental retardation and death in childhood characterize the Hurler syndrome, while intelligence is normal and life span nearly so in the Scheie syndrome. There exist canine and feline forms of ␣-L-iduronidase deficiency, and a murine form has recently been generated by homologous recombination (2). The disorders have been extensively reviewed, as have recent studies of their molecular basis (1,3).
Early work in cell culture had suggested that mucopolysaccharidosis I might be amenable to enzyme replacement therapy, since exogenous enzyme could be taken up by receptormediated endocytosis and delivered to lysosomes (1). To provide sufficient enzyme, we isolated a stably transfected Chinese hamster ovary (CHO) 1 cell line that synthesized and secreted large amounts of recombinant human ␣-L-iduronidase (4). The secreted enzyme had properties desirable for replacement purposes, including efficient endocytosis by cultured fibroblasts through a mannose 6-phosphate-dependent system and a 5-day half-life within the cells. When used in replacement trials for the canine model of ␣-L-iduronidase deficiency, the recombinant human enzyme was taken up to the largest extent by liver, in lesser amounts by lung, kidney, and spleen, and little if at all by brain, cartilage, myocardium, and cornea (5,6). Similar results were found in replacement trials for the feline model of the disease (7). It is not known whether this distribution is the result of accessibility of the circulating enzyme to tissues, of uptake of the enzyme by specific receptor systems, or of some combination of these factors.
Overexpressing CHO cell lines have been engineered for production of a number of other soluble lysosomal enzymes (8 -13). In contrast to normal cultured cells, which secrete very little lysosomal enzyme except in the presence of NH 3 or lysosomotropic amines, engineered CHO lines secrete a substantial fraction of the newly synthesized recombinant enzymes even in the absence of these weak bases (4,8,11,13). Such secretion occurs although the enzymes appear to have the mannose 6-phosphate signal for targeting to lysosomes, as evidenced by mannose 6-phosphate inhibition of their uptake. There is no detailed information on the structures of the carbohydrate constituents of recombinant lysosomal enzymes. We undertook * This work was supported in part by National Institutes of Health (NIH) Grant DK38857 (to E. F. N.). The UCLA Protein Microsequencing Facility is supported in part by NCI (NIH) Cancer Center Support Grant CA 16042-20 to the Jonsson Comprehensive Cancer Center. The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked "advertisement" in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.
A preliminary account of this work has been published in abstract form (17).
¶ To whom all correspondence should be addressed: an analysis of the N-linked carbohydrates of secreted ␣-L-iduronidase to shed light on processing by overexpressing CHO cells as well as to facilitate interpretation of enzyme uptake in vivo. Since the structure of carbohydrates at each of the six glycosylation sites (Asn residues 110, 190, 336, 372, 415, and 451 (14)) would be more useful for both purposes than the structure of pooled carbohydrates, we used MALDI-TOF mass spectrometry of isolated glycopeptides before and after treatment with glycosidases and phosphatase (15,16).

EXPERIMENTAL PROCEDURES
Materials-␣-L-Iduronidase was collected from secretions of the overexpressing CHO cell line 2.131 and purified to apparent homogeneity as described previously (4). Antiserum to this enzyme was raised in rabbits. Reagents were purchased from the following vendors: 33 P i from Amersham Corp.; L- [5,  Metabolic Labeling and Immunoprecipitation of ␣-L-Iduronidase-The CHO cell line 2.131 was maintained in minimal essential medium-␣ (with nucleosides) supplemented with 10% fetal bovine serum, nonessential amino acids, 20 mM Hepes, 0.4 mg/ml Geneticin, and 1% penicillin/streptomycin, with a final pH of 6.8, at 37°C and 5% CO 2 . Cells were transplanted to 6-well plates (35-mm well diameter) and used after reaching confluence. They were preincubated under similar conditions in methionine-or phosphate-free Dulbecco's modified Eagle's Medium supplemented as above except for the use of dialyzed fetal bovine serum and of 2.5 mM Hepes (starvation medium). After 30 min at 37°C, the starvation medium in each well was replaced with 0.5 ml of medium of similar composition containing 25 Ci of 35 S protein labeling mix or 25 Ci of 33 P i . After 1 h of labeling, a chase was initiated by the addition of unlabeled methionine to a final concentration of 0.1 mg/ml or of unlabeled phosphate to 1 mM. Where deoxymannojirimycin was used, it was added to the medium 6 h before labeling and kept through the labeling and chase period. No change was made in medium components when labeling with L-[5,6-3 H]fucose.
Immunoprecipitation of ␣-L-iduronidase with rabbit antiserum and Pansorbin was carried out essentially as described for ␤-hexosaminidase (18), except that 10 mM sodium phosphate, pH 5.8, was substituted for Tris-HCl. Because of the high level of expression, the enzyme could be immunoprecipitated from the medium without prior concentration.
The immunoprecipitates were dissolved in 0.1% SDS, 0.01% NaN 3 and heated at 98°C for 3 min; after centrifugation, 0.1 volume of 7.5% Nonidet P-40 and 1.5% ␤-mercaptoethanol were added to the supernatant fluids. The mixtures were heated again at 98°C for 3 min and centrifuged prior to endoglycosidase digestion. Immunoprecipitates were divided into aliquots for treatment at 37°C with one of the following: 50 milliunits of PNGase-F in 0.1 M Tris-HCl, pH 7.8; 100 units of endo-H in 0.1 M citrate buffer, pH 5.5 (the amount of enzyme was reduced to 50 units to observe glycosylation intermediates); 5 milliunits of neuraminidase in 0.1 M sodium acetate buffer, pH 4.6; or 250 milliunits of alkaline phosphatase in 0.1 M Tris-HCl, pH 8.2, for the indicated period of time. The immunoprecipitates, with or without enzymatic treatment, were subjected to SDS-polyacrylamide gel electrophoresis on a 7.5% gel (19), and radiolabeled ␣-L-iduronidase was visualized by fluorography.
Protease Digestion of Purified ␣-L-Iduronidase-Enzyme purified to apparent homogeneity (4) was stored in 0.15 M NaCl buffered to pH 5.8 with 10 mM sodium phosphate (buffer A). To change the pH for proteolysis, 2 nmol of enzyme (140 g of protein, excluding the weight of the carbohydrate chains) was buffer-exchanged in a Centricon-30 microconcentrator that had been pretreated overnight with 5% ethylene glycol to block nonspecific binding to the plastic wall. The exchange buffer was 100 mM Tris-HCl containing 10 mM sodium phosphate, 2 mM EDTA, and 0.02% NaN 3 , pH 8.2 (buffer B). After three cycles of dilution and concentration, the final protein solution in the reservoir (ϳ100 l) was recovered by reverse centrifugation for 5 min.
To prepare [ 33 P]␣-L-iduronidase for protease digestion, biosynthetic labeling of CHO 2.131 cells was carried out in a 100-mm plate in phosphate-free medium as described above, except for absence of fetal bovine serum during both the starvation and labeling periods. After 30 min preincubation, the medium was changed to 2 ml of the same serum-free starvation medium containing 200 Ci of 33 P i . At the end of a 6-h labeling period, the medium was collected and centrifuged, and the supernatant fluid was applied to a small column (0.8 ml) of heparinacrylic beads prewashed with buffer A containing 0.02% NaN 3 . ␣-L-Iduronidase was eluted in fractions of 0.5 ml with 0.6 M NaCl, 10 mM sodium phosphate, pH 5.8, 0.02% NaN 3 . The purified radiolabeled ␣-L-iduronidase was concentrated and combined with 2-3 nmol of unlabeled ␣-L-iduronidase for further processing.
The concentrated samples of ␣-L-iduronidase were denatured under reducing conditions by the addition of 4 mol of dithiothreitol, 1.2 mmol of solid guanidinium HCl, and buffer B to a total volume of 200 l. After 2 h at ambient temperature, solid iodoacetamide (18 mol) was added, and the volume was adjusted to 0.4 ml with buffer B. After further incubation for 3 h in the dark, the samples were dialyzed overnight at 4°C against 2 liters of 50 mM NH 4 HCO 3 , 10 mM sodium phosphate, pH 8.2. The samples were then incubated with TPCK-treated trypsin or TLCK-treated chymotrypsin for 12-16 h at 37°C (protease to substrate ratio ϳ1:50, w/w).
Lectin Separation of Glycopeptides-Phenylmethylsulfonyl fluoride was added to the tryptic or chymotryptic digests to a final concentration of 1 mM; the mixture was adjusted to pH 5.8 by the addition of 1 M HCl. After the addition of CaCl 2 and MgCl 2 to a final concentration of 1 mM each, the digest was passed over a series of lectin columns, each having a 0.5-ml bed volume, beginning with ConA-agarose. The column was extensively washed with buffer A until A 280 was negligible. Glycopeptides were eluted from the ConA-agarose column with 5 ml of 10 mM ␣-methylglucoside and 5 ml of 0.5 M ␣-methylmannoside in buffer A, in fractions of 0.5 ml. Unbound fractions were pooled, adjusted to pH 7.4, and passed over a column of wheat germ agglutinin-agarose joined to a column of castor bean agglutinin-agarose. After washing with buffer A, the columns were disconnected and eluted with 0.1 M N-acetylglucosamine or 0.1 M galactose in buffer A, respectively.
Separation of Glycopeptides on Reverse Phase HPLC-Lectin column eluates were pooled, concentrated to about 0.5 ml in a Speedvac concentrator, and applied to a reverse phase HPLC column (Keystone Scientific betasil, 2 ϫ 250 mm, particle size 5 m, and pore size 100 Å). The column was developed with 0.1% trifluoroacetic acid for 10 min and eluted with a gradient of 0 -50% acetonitrile, 0.1% trifluoroacetic acid for 100 min in fractions of 0.2 ml/min. All fractions were stored at 4°C.
Deglycosylation and Dephosphorylation of Glycopeptides-Immediately before use, endo-H was diluted to 20 units/l with water; PNGase-F to 10 milliunits/l and alkaline phosphatase to 2 milliunits/l with 50 mM NH 4 HCO 3 ; ␣-mannosidase and sialidase to 0.2 milliunits/l with buffers provided by the supplier; and glycopeptidase A to 0.2 milliunits/l with 50 mM (NH 4 ) 2 H citrate. ␣-Glucosidase II, provided in 100 mM Mes buffer, pH 6.8, containing 0.1% Triton X-100 (20) was diluted 300-fold with 50 mM sodium citrate buffer, pH 7.2. In each case, 1 l of enzyme was incubated with 1 l of HPLC fraction overnight at ambient temperature.
Mass Spectrometric Analysis-HPLC eluates or their deglycosylation products, 1 l, were co-spotted on a stainless steel plate with 1 l of matrix (␣-cyano-4-hydroxycinnamic acid (5 mg/ml) in water/acetonitrile/trifluoroacetic acid (50:50:0.1) containing 0.5 nmol/ml bradykinin). The bradykinin served as an internal standard for calibration. After the spots had been air-dried, each was washed for 30 s with 5 l of water and dried again. The wash step improved the detection efficiency and resolution of the resulting signals. The dried samples were examined by laser desorption mass spectrometry using a reflection time of flight instrument in the linear mode. Mass signals were analyzed without smoothing. The measured mass of deglycosylated peptides was compared with that calculated from the deduced amino acid sequence of ␣-L-iduronidase cDNA (4,14) using the MacBiospec computer program (PE Sciex Instruments, Ontario, Canada).
Amino Acid Sequence Analysis-Amino-terminal sequence analysis was carried out on a Porton 2090E Sequencer at the UCLA Microsequencing Facility. Cleavage of amino-terminal pyroglutamate was carried out by incubating 1 l of pyroglutamate aminopeptidase (20 ng/l in 50 mM NH 4 HCO 3 , 5 mM EDTA, 5 mM dithiothreitol) with 1 l of PNGase-F-treated HPLC fraction overnight at ambient temperature. Carboxyl-terminal sequence analysis was carried out by graded cleavage of the peptide with carboxypeptidase Y and determination of the mass of the resulting peptides (21). Equal volumes (0.5 l) of carboxypeptidase Y (serially diluted in (NH 4 ) 2 H citrate to give 150 -0.3 g/l) and PNGase-F-treated HPLC fractions were mixed on the steel plate. After incubation for 2 h at 37°C in a sealed Petri dish humidified with wet cotton wads, the reaction was stopped, and the samples were crystallized by the addition of 1 l of 10 mg/ml ␣-cyano-4-hydroxycinnamic acid in 50% acetonitrile, 0.1% trifluoroacetic acid. After the wash step as above, the peptides were analyzed by MALDI-TOF mass spectrometry.

Carbohydrate Structures of Metabolically Radiolabeled ␣-L-
Iduronidase-Susceptibility of radiolabeled ␣-L-iduronidase to cleavage by endo-H and PNGase-F provided information on the ratio of high mannose to complex oligosaccharides. Newly made intracellular enzyme differed from secreted enzyme in that the former was completely cleaved by endo-H (Fig. 1A, lanes 1 and  3), whereas secreted enzyme was partially resistant (Fig. 1A,  lanes 7 and 9). The carbohydrate chains not removed from the secreted enzyme by endo-H could be cleaved, although not completely, by PNGase-F (Fig. 1A, lanes 11 and 12); a possible explanation for the incomplete hydrolysis by PNGase-F will be provided below. Only after the labeling had been performed in the presence of dMM, an inhibitor of the Golgi ␣-mannosidase I and hence of the mannose trimming required for complex sugar formation (22), were the migration and endo-H susceptibility of the secreted enzyme equal to those of newly made intracellular enzyme (Fig. 1A, lanes 2, 4, 8, and 10). Thus, the difference is due to complex oligosaccharides on the secreted enzyme.
Intermediate deglycosylation steps were made visible by par-tial endo-H digestion. Treatment of newly made intracellular ␣-L-iduronidase with a reduced amount of endo-H gave a ladder of six distinct bands, not including the starting material (Fig.  1B, left panel), showing that all six sites were of the high mannose type. Similar treatment of the secreted enzyme with endo-H resulted in only 3 or 4 bands (Fig. 1B, right panel), indicating that two or three sites had been processed to the complex form. The bands derived from the secreted enzyme were diffuse and poorly resolved, although the sample had been pretreated with neuraminidase to reduce heterogeneity due to different degrees of sialylation. The diffuse appearance and poor resolution of the bands are attributed to incomplete desialylation and/or glycoform heterogeneity of secreted ␣-L-iduronidase.
When the secreted ␣-L-iduronidase was biosynthetically labeled with 33 P, all of the phosphate label could be removed by treatment with endo-H ( Fig. 2A, lanes 1-4), indicating that it was associated with the carbohydrate and not with the protein.
The deglycosylated protein itself could be visualized with [ 35 S]methionine labeling ( Fig. 2A, lanes 7 and 8). The secreted enzyme became strongly labeled after continuous exposure of the cell to 33 P i for 6 and 10 h (Fig. 2B, lanes 3 and 4), but the intracellular enzyme contained a surprisingly low level of radioisotope at all times (Fig. 2B, lanes 1 and 2), suggesting that it was dephosphorylated after delivery to lysosomes.
On the other hand, L-[ 3 H]fucose was incorporated only into endo-H-resistant, PNGase-F-sensitive oligosaccharides of secreted ␣-L-iduronidase (Fig. 2C), showing that it was present solely on N-linked complex oligosaccharides.
Analysis of ␣-Methylmannoside-eluted Glycopeptides of Secreted ␣-L-Iduronidase by MALDI-TOF Mass Spectrometry- Fig. 3, A and C, shows the HPLC profile of ␣-L-iduronidase glycopeptides produced by tryptic and chymotryptic digestion and isolated by ␣-methylmannoside elution from ConA-agarose; Fig. 3, C and D, show the corresponding radioactivity profile of 33 P-labeled ␣-L-iduronidase glycopeptides similarly treated with trypsin and chymotrypsin, respectively.
MALDI-TOF analysis of HPLC fractions 69 -70 of the tryptic digest (Fig. 3A) gave a prominent signal at m/z 3802.8. This signal disappeared when the fractions were treated with endo-H or PNGase-F and was replaced by signals of m/z 2306.2 or 2103.7, respectively. This decrease in mass resulted from the loss of carbohydrate, the structure of which could be deduced from the magnitude of the decrease (Table I). The carbohydrate itself was not detected, since it does not crystallize with the matrix and/or fails to protonate under the conditions used. Treatment of fractions 69 -70 with ␣-mannosidase produced no change in mass, whereas treatment with alkaline phosphatase resulted in a prominent signal at m/z 3644.7. The changes in mass observed upon treatment with endo-H, PNGase-F, and phosphatase were compatible with the structure P 2 Man 7 GlcNAc 2 ( Table I). The 33 P radiolabel in fractions 69 -70 (Fig. 3B) confirmed the phosphorylated structure of the oligosaccharide. Some minor signals could also be seen in the intact sample, of m/z 3638.9, 3556.7, and 3390.1, compatible with structures P 2 Man 6 GlcNAc 2 , P 1 Man 6 GlcNAc 2 , and P 1 Man 5 GlcNAc 2 ; these accounted for ϳ6%, 6%, and 2% of the total, respectively. Amino-and carboxyl-terminal sequence analyses as well as the mass of the deglycosylated glycopeptide identified it as Val-325 to Tyr-343 (Table II). The phosphorylated high mannose oligosaccharide was therefore assigned to Asn-336.
Mass spectrometric analysis of HPLC fraction 66 from the tryptic digest (Fig. 3A) showed a series of signals starting with m/z 3693.4 and decreasing in steps of 162 Da to m/z 2719.5 (Fig.  4). This fraction contained no 33 P (Fig. 3B). The major signal of m/z 3530.7 was compatible with a structure of Man 9 GlcNAc 2 (Table I), the signals of lower m/z were compatible with progressively fewer mannose residues down to Man 3 GlcNAc 2 , and the signal of m/z 3693.4 was presumed to represent a structure with an additional hexose. Treatment with ␣-glucosidase II reduced the latter to 3530.5 (Fig. 4), identifying the m/z 3693.4 species as Glc 1 Man 9 GlcNAc 2 (Table I). Treatment with jack bean ␣-mannosidase resulted in the expected Man 3 GlcNAc 2 (m/z of 2557.7), with some further removal of one or two mannose residues. However, some of the material was digested only to m/z 3368.5; this signal was presumably derived from the loss of 2 uncovered mannose residues from the glucosylated species. Amino-and carboxyl-terminal sequence analysis as well as the mass of the PNGase-F treated glycopeptide of fraction 66 identified it as Phe-369 to Leu-382 (Table II). The unphosphorylated high mannose chain was therefore assigned to Asn-372. The distribution of the glycoforms was as follows: 13% GlcMan 9 GlcNAc 2 , 62% Man 9 GlcNAc 2 , 19% Man 8 GlcNAc 2 , and 6% Man 4 -7 GlcNAc 2 .
Similar glycoforms on Asn-372 peptides with different COOH termini were also found in fractions 48 -50, 58 -60, and 61-62 (Fig. 3A, Table II). Further analysis of the tryptic digests failed to yield information about any glycosylation site other than Asn-372 and Asn-336. Since endo-H digestion of biosynthetically labeled ␣-L-iduronidase had predicted at least three glycosylation sites of high mannose structure (Fig. 1B), we reasoned that some tryptic glycopeptides might not crystallize with the matrix used and therefore would not have been detected in the mass spectrometric analysis.
Similar analyses were therefore performed on chymotryptic digests of secreted ␣-L-iduronidase. The major peak in HPLC fractions 50 -51 (Fig. 3C) was also strongly labeled with 33 P (Fig. 3D). Analysis of this fraction by MALDI-TOF mass spectrometry revealed a major signal of m/z 3515.0, with additional strong signals of m/z 3254.6, 3092.3, and 2930.6 as well as a number of minor signals (Fig. 5A). After endo-H cleavage of the intact glycopeptides, two strong signals of m/z 2015.4 and 1590.5 appeared, suggesting the presence of two different peptide backbones. This was confirmed by amino-terminal sequence analysis, which indicated a mixture of two peptides, one containing the Asn-372 and the other the Asn-451 glycosylation site. Since the glycoforms at Asn-372 had already been identified (see above), several signals could be unambiguously assigned to that site: m/z 3254.6, 3092.3, and 2930.6 of the intact sample, corresponding to high mannose forms with 9, 8, and 7 mannose residues, and m/z 1590.5 of the endo-H-treated sample. By the process of elimination, the signals of m/z 3515.0 (intact) and m/z 2015.4 (endo-H-treated) could be assigned to the Asn-451 glycosylation site.
Treatment of the intact sample with ␣-mannosidase confirmed these assignments and also resolved the glycopeptide mixture. The signal of m/z 3515.0 remained unchanged (Fig.  5B), whereas the signals assigned to Asn-372 were replaced by signals of m/z 2279.2 and 2443.4. Phosphatase digestion of the ␣-mannosidase-treated glycopeptides (after reisolation on a Sep-Pak cartridge) caused the m/z 3515.0 signal to shift to m/z 3433.6 and 3353.8, corresponding to a loss of one and two phosphates, respectively. Therefore, the m/z 3515.0 oligosaccharide on the Asn-451 site could be assigned the structure of P 2 Man 7 GlcNAc 2 .
In contrast to digestion with endo-H, which gave two distinct products (Fig. 5A), digestion of fraction 50 -51 with PNGase-F yielded only one new signal, m/z 1387.8, corresponding to the Asn-372 peptide (Fig. 5B); the signal derived from the phosphorylated oligosaccharide at Asn-451, m/z 3512.8, remained untouched by PNGase-F (Fig. 5B).
The unexpected resistance of the phosphorylated glycopeptide at Asn-451 was investigated further (Fig. 6). Fractions 50 -51 were first treated with ␣-mannosidase to reveal the phosphorylated Asn-451 glycopeptide at m/z 3515 (Fig. 6A), which was resistant to PNGase-F (Fig. 6, B and C); but if the mannosidase-treated mixture was reisolated and exposed to phosphatase, the resulting monophosphorylated and dephosphorylated glycopeptides (m/z 3433 and 3354, respectively) were rapidly hydrolyzed by PNGase-F (Fig. 6, D-F). Resistance of bisphosphorylated glycopeptides to PNGase-F is not a general phenomenon, since the bisphosphorylated glycopeptide at Asn-336 (tryptic fractions 69 -70, Table I) was sensitive to   Fig. 7. b Found in mixture with Asn-415 peptide. c Glycopeptide data in Table I. d Glycopeptide data in Fig. 4 and Table I. e Glycopeptide data in Fig. 5; mixture with Asn-451 glycopeptide. f Peptide seen in mixture with Asn-190 peptide by amino-terminal sequence analysis; not seen in mass spectrometry. g Glycopeptide data in Fig. 5; mixture with Asn-372 glycopeptide. h Endo-H-cleavable oligosaccharide same as that of methylmannoside-eluted glycopeptide.
The PNGase-F-resistant glycopeptide was readily cleaved by glycopeptidase A (not shown). Some Asn-372 glycopeptides in chymotryptic fractions 50 -51 showed doublet signals differing in mass by 17 Da; difficulty in obtaining amino-terminal sequence occurred when the smaller member of the doublet predominated. The problem was traced to cyclization of the amino-terminal glutamine to pyroglutamate with loss of NH 3 (23). The presence of pyroglutamate was demonstrated by the reduction in mass by 111 Da when the PNGase-F-deglycosylated peptide was treated with pyroglutamate aminopeptidase (data not shown). Additional cyclized peptides were also found ( Table II).
Analysis of Tryptic and Chymotryptic Complex Glycopeptides-Tryptic and chymotryptic glycopeptides eluted from ConA-agarose with ␣-methyl glucoside were subjected to similar separation on HPLC and analysis by MALDI-TOF mass spectrometry before and after treatment with sialidase or PNGase-F. These fractions would be expected to contain only biantennary complex structures. Assignment of one or more N-acetyl neuraminic acid residues was made on the basis of mass before and after treatment with sialidase. Amino-and carboxyl-terminal amino acid sequence analysis confirmed the identity of the peptides. Fig. 7 shows a glycopeptide from tryptic fraction 84, of mass assignable to Gly-106 to Leu-121 and therefore containing glycosylation site Asn-110. The carbohydrate structure is compatible with NeuAc 0 -2 Gal 2 Glc-NAc 4 -Man 3 Fuc 1 . Similar structures were also observed in methylglucoside-eluted chymotryptic fractions 54 (Asn-190), 58 (Asn-190 mixed with Asn-415), 51-52 (Asn-451), and 60 -61 (Asn-110) (not shown). Analysis of the structure of the Asn-415 oligosaccharide is incomplete, since we were not able to resolve it from that of the admixed Asn-190 glycopeptide (Table II).
Attempted analysis of HPLC fractions eluted from wheat germ agglutinin-agarose or castor bean agglutinin-agarose did not provide any identifiable signals on mass spectrometry. Glycopeptides present in these fractions would have contained more branched complex structures, with additional neuraminic acid residues, and may have been undetectable under the conditions used for mass spectrometry. Table II summarizes the data identifying the major glycopeptides found in HPLC fractions of tryptic and chymotryptic digests. The peptide mass after PNGase-F deglycosylation and its aminoand/or carboxyl-terminal amino acid sequence formed the basis of glycosylation site assignment. Analysis of the glycoforms has been described above for the major fractions.

DISCUSSION
Glycosylation of the secreted recombinant ␣-L-iduronidase is shown schematically in Fig. 8. The structural analyses showed that all of the glycoforms at Asn-372, most glycoforms at Asn-336 and Asn-451, and some unknown fraction at Asn-415 are of the high mannose or phosphorylated high mannose type and therefore endo-H-sensitive. This is in good agreement with the results of biosynthetic experiments, which showed that three or four of the six glycosylation sites were occupied by endo-Hsensitive oligosaccharides. The biosynthesis experiments were carried out under conditions (medium composition and 12-h collection time) that simulated the preparative protocol.
The presence of phosphorylated and unphosphorylated high mannose oligosaccharides, as well as of partially desialylated complex oligosaccharides, suggests that ␣-L-iduronidase could be endocytosed by the widely distributed mannose 6-phosphate receptor, by the mannose receptor of macrophages, or by the asialoglycoprotein receptor of hepatocytes. In addition, if antibodies are produced against the administered enzyme, immune complexes could be taken up by the Fc receptor of macrophages. On the other hand, the recombinant ␣-L-iduronidase did not appear to have the ␣-L-fucosyl-threonine that has been proposed as a recognition signal (24) and that is known to be synthesized by CHO cells (25), since all of the L-[ 3 H]fucose incorporated into ␣-L-iduronidase was found on N-linked (PNGase-F-sensitive) structures. Even one O-linked L-fucose per ␣-L-iduronidase molecule would have been easily detected on Fig. 2C.
Synthesis of the mannose 6-phosphate targeting signal did not appear to be rate-limiting in the overexpressing CHO cells. The oligosaccharides on Asn-336 and Asn-451 were primarily of the bisphosphorylated variety, P 2 Man 7 GlcNAc 2 , with no evidence of any blocked phosphate. The presence of only a minor component of complex chains at these positions suggests that in the race between phosphorylation and mannose trimming, the phosphorylation reaction predominated most of the time. Therefore, the absence of any phosphorylated oligosaccharide at Asn-372 indicates that this site was not accessible to the GlcNAc1-phosphotransferase that catalyzes the first step in synthesis of that signal.
It was surprising to find Glc 1 Man 9 GlcNAc 2 , accounting for about 13% of the oligosaccharides on Asn-372. We are not aware of other instances of such glucosylated structures in secreted proteins. The single glucose residue is thought to be part of a quality control system that keeps deglucosylating and reglucosylating newly synthesized glycoproteins to ensure their retention in the endoplasmic reticulum until folding is completed (26). The cause of incomplete deglucosylation of the secreted recombinant ␣-L-iduronidase is not known, but it could be a result of its overexpression.
The CHO cell line used here had been shown to secrete half of the recombinant ␣-L-iduronidase that it synthesized (4). This propensity for secreting a large fraction of overexpressed lysosomal enzymes makes engineered CHO cells particularly useful, since secreted enzymes are relatively easy to purify. We unexpectedly found another advantage of using secreted enzyme for preparative purposes; only the secreted ␣-L-iduronidase contained phosphorylated oligosaccharides. Presumably, the intracellular enzyme of CHO cells was dephosphorylated soon after reaching lysosomes, whereas the secreted form was not exposed to phosphatases or protected by P i present in the medium.
The natural history of recombinant ␣-L-iduronidase in CHO cells is different from that of the enzyme in diploid human fibroblasts. Fibroblasts secrete very little ␣-L-iduronidase except in the presence of NH 3 (27,28); furthermore, the intracellular enzyme retains phosphate groups (27). Because of the very low abundance of ␣-L-iduronidase in fibroblasts, no detailed structural analyses are available. Our concepts of the biosynthesis and targeting of soluble lysosomal enzymes (e.g. Ref. 29) are based on naturally occurring enzymes, including ␣-L-iduronidase; they do not explain why or how these processes differ for recombinant enzymes made by transfected cells.
A high degree of intrasite heterogeneity was found in the glycopeptides isolated from proteolytic digests of the secreted human ␣-L-iduronidase. Intrasite heterogeneity of the oligosaccharide structures (Figs. 4, 5, and 8) can be ascribed to the carbohydrate processing capabilities of the CHO cells. On the other hand, the intrasite heterogeneity of the peptides (Table  II) must be ascribed to sample preparation. Most of the tryptic peptides had carboxyl termini at other than trypsin cleavage FIG. 8. Oligosaccharides at the six glycosylation sites of recombinant human ␣-L-iduronidase secreted by CHO cells. C, complex; M, high mannose; P, phosphorylated high mannose. Capital letters denote well identified, major oligosaccharides, whereas lowercase letters denote minor or incompletely characterized components.
FIG. 6. Resistance of phosphorylated oligosaccharide at Asn-451 to cleavage by PNGase-F. Chymotryptic fractions 50 -51 were treated with ␣-mannosidase (panels A-C) or by ␣-mannosidase followed by reisolation of the glycopeptide and treatment with alkaline phosphatase (panels D-F) before being subjected to incubation with PNGase-F for 1 or 24 h, as indicated. Tryptic fractions 69 -70, with a phosphorylated oligosaccharide at Asn-336, was treated in the same manner (panels G-I). -110). HPLC fraction 84 of tryptic glycopeptides, eluted from ConA-agarose with ␣-methylglucoside, was subjected to MALDI-TOF mass spectrometry either directly (intact) or after treatment with sialidase or PNGase-F as indicated.

FIG. 7. Identification of a complex oligosaccharide in methylglucoside-eluted tryptic fraction 84 (Asn
sites, indicating the action of peptidase or protease contaminants in trypsin; the chymotryptic peptides showed alternate cleavage sites due to the broad specificity of chymotrypsin. Cyclization of amino-terminal glutamine residues and oxidation of methionine residues added to the complexity of the mixtures. Since this problem of peptide heterogeneity may be intrinsic to the procedures used, assignment of glycosylation sites based solely on peptide mass and presumptive protease cleavage sites is likely to lead to errors. Sequence of one or both termini of the peptide is needed for correct assignment. The protocol used here was least successful in the detection and analysis of complex carbohydrates; no tri-or tetra-antennary structures were found in mass spectrometry, and even the biantennary structures gave broad peaks before desialylation. A more complete study of the complex glycopeptides of the secreted enzyme was not attempted; the known processing pathways make it likely that the more highly branched structures would exist only at the sites at which we had identified the biantennary forms. Analysis of purified ␣-L-iduronidase secreted by the same CHO cells (under slightly different conditions), using fluorophore-assisted carbohydrate electrophoresis (30), had shown a preponderance of triantennary and tetraantennary complex oligosaccharides. 2 In the course of this study, we found that PNGase-F was unable to remove the bisphosphorylated high mannose oligosaccharide from Asn-451, although it readily removed it after partial or complete dephosphorylation. This does not represent a general inability to hydrolyze bisphosphorylated high mannose structures, since the same structure on Asn-336 was sensitive to PNGase-F. Perhaps some interaction between the two phosphate groups and the basic residues surrounding Asn-451 hindered the action of PNGase-F, although the glycopeptide could be cleaved by endo-H and by glycopeptidase A. Whatever the mechanism of PNGase-F resistance, this finding should prompt some caution in interpreting the results of PNGase-F action on glycoproteins bearing phosphorylated carbohydrate chains.