Carbohydrate Structure of Erythropoietin Expressed in Chinese Hamster Ovary Cells by a Human Erythropoietin cDNA*

proper glycosylation of erythropoietin repeats methylation saccharides provided 2.4 mol of 3-substituted galactose, 4 mol of terminal galactose, and 1 mol each of 2,4-, 2,6-, and 3,6-substituted mannose. The same analysis that about 45% of the reducing terminal N-acetylglucosamine was un-reduced judging from the amount of 3-O-methyl-N-acetylglu-cosamine. This compound is produced when the reducing terminal N-acetylglucosamine is incompletely reduced before methylation.

The proper glycosylation of erythropoietin is essential for its function in vivo. Human erythropoietins were isolated from Chinese hamster ovary cells transfected with a human erythropoietin cDNA and from human urine. Carbohydrate chains attached to these proteins were isolated and fractionated by anion-exchange high performance liquid chromatography (HPLC) and HPLC employing a Lichrosorb-NHz column. The structures of fractionated saccharides were analyzed by fast atom bombardment-mass spectrometry and methylation analysis before and after treatment with specific exoglycosidases.
We have also shown that the carbohydrate moiety of urinary erythropoietin is indistinguishable from recombinant erythropoietin except for a slight difference in sialylation, providing the evidence that recombinant erythropoietin is valuable for biological as well as clinical use.
Erythropoietin is a glycoprotein which stimulates proliferation and differentiation of erythroid precursor cells to more mature erythrocytes (1). Erythropoietin is primarily produced in adult kidney and fetal liver cells (2-4). Patients with chronic renal failure are anemic as a result of impaired renal function which leads to a decreased production of erythropoietin (5).
Thus, availability of purified erythropoietin in quantity is essential to understand molecular mechanisms of erythropoiesis and for treatment of anemia. However, this has been hampered by the fact that only a very small amount of erythropoietin is present in starting sources, even in such cases as the urine of aplastic anemia patients (6).
In order to overcome this problem, cDNA clones for human erythropoietin have been isolated in several laboratories, and the expression of erythropoietin cDNA clones has been achieved (7-9). Furthermore, the recombinant erythropoietin has been successfully used to reverse the anemia of patients with endo-stage renal disease (10,ll). Interestingly, the erythropoietin produced in Escherichia coli or yeast was inactive or very weakly active i n uiuo. On the other hand, the erythropoietin produced in COS cells or Chinese hamster ovary cells was found to be fully active i n uiuo. In agreement with these results, it has been reported that desialylation of partially purified erythropoietin results in inactivation of erythropoietin activity (12-15). Thus, it is apparent that the proper glycosylation is essential for erythropoietin activity in uiuo.
These results prompted us to analyze carbohydrate structures of erythropoietin produced by transfection of recombinant DNA into Chinese hamster ovary cells. In addition, we compared those structures with carbohydrate units present in erythropoietin purified from human urine.

EXPERIMENTAL PROCEDURES
Erythropoietin-Chinese hamster ovary cells (dihydrofolate reductase-) were transfected with an expression vector which harbors the human erythropoietin cDNA as described (7). This expression vector also contains dihydrofolate reductase minigene so that stable transfects can be grown in the presence of methotrexate (16). Erythropoietin was purified from the spent medium of those cells as described (7). The purification procedure was slightly modified from that of Miyake et al. (6), and fractionation on a Vydac C, reverse-phase HPLC' column (The Separations Group) was included (7). This erythropoietin will be called recombinant erythropoietin hereafter. Erythropoietin was also purified from the urine of aplastic anemia patients according to Miyake et al. (61, with a similar modification as for purification of recombinant erythropoietin. The erythropoietin purified from urine will be called urinary erythropoietin hereafter. These erythropoietin samples were provided by Chugai Pharmaceutical Co., Ltd. (Tokyo).
Isolation of N-Linked Glycopeptides and O-Linked Oligosaccharides from Erythropoietins-Glycopeptides were prepared by Pronase digestion of erythropoietin (5 mg of recombinant erythropoietin and The abbreviations used are: HPLC, high performance liquid chromatography; FAB-MS, fast atom bombardment-mass spectrometry, Hex, hexose; HexNAc, N-acetylhexosamine; NeuNAc, N-acetylneuraminic acid; Lac, N-acetyllactosaminyl repeat. 1 mg of urinary erythropoietin, respectively) as described (17). The Pronase digest was applied to a small column (1.0 X 45 cm) of Sephadex G-15, which was equilibrated and eluted with water. The gylcopeptides, eluted at and near the void volume, where lyophilized and subjected to alkaline borohydride degradation, as described (17). Briefly, the glycopeptides were dissolved in 500 p1 of 0.05 M NaOH, 1 M NaBHI containing 5 mCi of NaB3H4 (9 Ci/mmol) and incubated at 45 "C for 24 h. After treatment, 1-2 ml of methanol which contained 1 drop of acetic acid was added to the sample and evaporated under nitrogen. The alkaline borohydride-treated sample was then applied to the same Sephadex G-15 column for desalting. The glycopeptides and oligosaccharides, which eluted between the void volume and the salt peak, were pooled and applied to a column (1.0 X 140 cm) of Bio-Gel P-4 (200-400 mesh). The column was eluted with 0.1 M NHlHC03 at a flow rate of 6 ml/h, and each fraction contained 1 ml. Under these chromatographic conditions, N-linked glycopeptides eluted near the void volume, whereas 0-linked oligosaccharides eluted in later fractions (see Fig. 2).
Isolation of N-Linked Saccharides from Glycopeptides-The glycopeptide fraction containing N-linked saccharides was digested with Flavobacterium meningosepticum N-glycanase (peptide-N4-(N-acetyl-8-glucosaminy1)aspargine amidase) (18) which was purchased from Genzyme (Boston, MA). The glycopeptides from 5 mg of erythropoietin were dissolved in 500 p1 of 0.1 M sodium phosphate buffer, pH 8.6, containing 20 mM EDTA and 20 mM 2-mercaptoethanol and incubated with 30 m units of the N-glycanase, which corresponds to 30 units of the enzyme expressed by Genzyme, at 37 "C for 20 h. When the glycopeptides from 1 mg of urinary erythropoietin were digested with N-glycanase, the incubation mixture was scaled down to one-fifth.
The digest was desalted by Sephadex G-15 gel filtration, and the saccharides were reduced at room temperature for 30 min with 5 mCi of NaB3H4 dissolved in 200 pl of 0.01 M NaOH followed by 3 mg of NaBH4 for 2 h. After reduction, the sample was neutralized with the addition of methanol containing a small amount of acetic acid and evaporated under nitrogen. The dried sample was dissolved in water, and the supernatant obtained after centrifugation was applied to the Sephadex G-15 column for desalting. The saccharides, radioactively labeled at reducing terminals, were applied to a column (1.0 X 45 cm) of Sephadex G-50 (superfine) eluted with 0.2 M NaCl (see Fig. 2B). Each fraction contained 0.5 ml.
Fractionation of Oligosaccharides by Anion-exchange High Performance Liguid Chromatography-Oligosaccharide fractions obtained after Sephadex G-50 gel filtration were subjected to anion-exchange HPLC with a Varian HPLC apparatus (Model 5000, Varian Associates, Inc.). The sample was applied to a Toyo Soda TSK-DEAE column purchased from Kratos Analytical Instruments (Ramsey, NJ). The column (4.6 mm X 24 cm) was equilibrated with 25 mM potassium phosphate buffer, pH 5.0, and after eluting with the same buffer for 10 min, the elution was programed by the linear gradient to 0.4 M potassium phosphate buffer, pH 5.0, over 80 min. The flow rate was constant at 1 ml/min, and each fraction contained 0.3 ml. The elution was monitored by measuring the absorbance at 206 nm with a Beckman 163 detector. Since potassium phosphate has absorbance at 202 nm, the absorbance at 206 nm was measured under these conditions. The base line of the absorbance at 202 nm was increased under these conditions. In order to estimate the elution positions of monosialosyl, disialosyl, trisialosyl, and tetrasialosyl saccharides, IgG (bovine,Sigma),fetuin (19,20), and nl-acid glycoprotein saccharides (21, 22) were prepared by hydrazinolysis and subjected to HPLC under the same conditions. The saccharides, separated by TSK-DEAE, were subjected to methylation analysis to confirm their structures.
Fractionation of Neutral Oligosaccharides by HPLC-The saccharides were desialylated by mild acid hydrolysis in 0.01 N HCl at 80 "C for 1 h. The desialylated sample was desalted by Sephadex G-15 gel filtration and evaporated. The neutral saccharides were then fractionated by HPLC with the same apparatus as described above. The saccharides were dissolved in acetonitrile, 10 mM potassium phosphate buffer, pH 4.5 (65:35), and applied to a column (0.4 X 25 cm) of Lichrosorb-NHz (5-pm particle size, Merck). The column was eluted with a linear gradient to acetonitrile, 10 mM potassium phosphate buffer, pH 4.5 (3961), over 60 min. In order to achieve this elution, the first solvent was acetonitrile, 10 mM potassium phosphate buffer, pH 4.5 (65:35, v/v), and the second solvent was 10 mM potassium phosphate buffer, pH 4.5. The ratio of these two solvents was 1oO:O at the beginning of the elution and 6040 at the end of elution. This elution system was used because it was preferable to premix acetonitrile and the buffer before the elution started, otherwise, mixing of two solutions causes cooling down of the solvent. The flow rate was 1 ml/min, and each fraction contained 0.5 ml.
Structural Analysis of Saccharides-Structures of saccharides were analyzed essentially as described previously for other saccharides. These methods include fast atom bombardment-mass spectrometry of permethylated saccharides (17), analysis of partially 0-methylated monosaccharides after methylation and hydrolysis ("methylation analysis") (17,231, and exoglycosidase digestion combined with methylation analysis (24). Methylation of saccharides and purification of methylated saccharides for FAB-MS and methylation analysis were carried out as described previously (17,23). Before methylation of saccharides, all samples were desalted by Sephadex G-15 gel filtration eluted with water.
Glycosidase digestion of saccharides was carried out as follows. For sequential digestion by @-galactosidase and @-N-acetylglucosaminidase, the saccharides were incubated with 5 milliunits of Charonia lampas &galactosidase in 40 pl of 0.1 M sodium citrate buffer, pH 4.3, at 37 "C for 24 h. After incubation, the incubation mixture was heated in a boiling water bath for 2 min. The sample was then incubated with 100 milliunits of beef kidney @-N-acetylglucosaminidase in 140 p l of 0.1 M sodium citrate buffer, pH 4.3. In order to inhibit a possible contaminating activity of @-galactosidase, 100 mM (final concentration) galactose was added to this incubation mixture. After further incubation at 37 "C for 24 h, the mixture was heated in a boiling water bath for 2 min. The digests were then purified by Sephadex G-50 gel filtration followed by HPLC employing a Lichrosorb-NHz column as described above. For extensive digestion by a mixture of @-galactosidase and @-N-acetylglucosaminidase, the saccharides were first incubated with C. lampas @-galactosidase and then with beef kidney @-N-acetylglucosaminidase without heat inactivation of 8galactosidase or the addition of galactose. After total incubation at 37 "C for 48 h, enzymes were inactivated by heating in a boiling water bath for 2 min, and digested saccharides were purified by anionexchange HPLC. @-Galactosidase from C. lumpas and @-N-acetylglucosaminidase from beef kidney were purchased from Sigma and Boehringer Mannheim, respectively.
Determination of Carbohydrate Composition-Sialic acid content was determined by the periodate-resorcinol reaction (25). Neutral sugars and hexosamines were determined after methanolysis in 0.5 N hydrochloric acid in anhydrous methanol at 80 "C for 4 h. Inositol was added as an internal standard. After methanolysis, the products were dried under a nitrogen stream and further in uacw under PB05 and NaOH. The dried products were trimethylsilylated with Tri-Si1 (Pierce Chemical Co.). The trimethylsilylated derivatives were then analyzed by gas-liquid chromatograph-mass spectrometry as described.* In parallel, fetuin was analyzed in order to obtain response factors.

RESULTS
Isolation of N-Linked Sacchurldes from Recornbinant Etythropoietin-Erythropoietin was purified from the spent medium of Chinese hamster ovary cells transfected with a human erythropoietin gene and from human urine of aplastic anemic patients as described under "Experimental Procedures." The purified proteins showed a major band with M, 38,000 and a faint band with M, -80,000 (Fig. 1). The latter band is probably the dimer of erythropoietin. The carbohydrate composition of this molecule is shown in Table I. Glycopeptides were prepared from 5 mg of recombinant erythropoietin by Pronase digestion and isolated by Sephadex G-15 gel filtration. Glycopeptides (1.8 mg) were then treated with alkaline borohydride to release 0-linked oligosaccharides, and the alkaline borohydride-treated samples were applied to a column of Bio-Gel P-4. As shown in Fig. 2 Greenwood et al. (26). The radioactively labeled proteins were then applied to sodium dodecyl sulfate-polyacrylamide gel electrophoresis (gel concentration, 15%) according to Laemmli (27), and the gel was directly autoradiographed with Kodak x-ray AR-5 film Fractions 25-29 were pooled and digested with N-glycanase. The digest, after reduction with NaBSH4, was subjected to Sephadex, G-50 gel filtration as shown in Fig. 2B. The saccharides eluting between fractions 29 and 41 were subjected to further analysis.
Methylation Analysis and FAB-MS of N-Linked Saccharides from Recombinant Erythropoietin-In order to determine the carbohydrate structures of N-linked saccharides, the saccharides were subjected to FAB-MS and methylation analyses. As shown in Fig
After desialylation, a majority of 3-substituted galactose residues were converted to terminal galactose, indicating that sialic acid is linked to galactose through an a2+3-linkage. However, 0.82 mol of 3-linked galactose, which corresponds to 17% of the total galactose derivatives, was still detected after desialylation. This amount of galactose is presumably derived from N-acetyllactosaminyl repeats. The same analysis also showed that 85% of reducing terminal N-acetylglucosamine is substituted with fucose at C-6, whereas the rest of the reducing terminal N-acetylglucosamine contains no fucose.
These results suggest that the N-linked saccharides of recombinant erythropoietin are mainly composed of tetraantennary saccharides with or without N-actyllactosaminyl repeats.

HexNAc+Hex+HexNAc+Hex+HexNAc+.
A small peak was detected also at m/z 464 for Hex+HexNAc+. B, in addition to the ions described above, a minor fragment ion was detected at m/z 913 for Hex+HexNAc+Hex+ HexNAc+. charides, obtained from 5 mg of recombinant erythropoietin (Batch 1 in Table I), were desialylated by mild acid hydrolysis and then desalted by Sephadex G-15 gel filtration. Neutral saccharides thus obtained were fractionated by HPLC employing a Lichrosorb-NH2 column. As shown in Fig. 5A, Nlinked saccharides from recombinant erythropoietin provided six peaks which correspond to biantennary (peak 2), triantennary (peak 3), tetraantennary (peak 4), tetraantennary with one N-actyllactosaminyl repeat (peak 5, Lacl), tetraantennary with two N-acetyllactosaminyl repeats (peak 6, Lacz), and tetraantennary with three N-acetyllactosaminyl repeats (peak 7, Lac3). The identical elution profile was obtained when Nlinked saccharides were desialylated by clostridial neuraminidase. No significant amount of carbohydrate was detected in other fractions. This result indicates that most of the glycopeptides were digested by N-glycanase since the glycopeptides would elute later than saccharides without amino acid residues in a Lichrosorb-NHz column.
Structures of Biantennary Saccharides and TTiantennary Saccharides-The elution position of the biantennary sac-charides (peak 2) was identical to the IgG saccharide which is mainly composed of Galz -GlcNAc, . Man, -GlcNAc,. Fuc. In addition, this saccharide bound to a concanavalin A-Sepharose and was eluted by 20 mM methyl-a-glucoside. Those results establish that this fraction (peak 2) is a typical complex saccharide with biantennary side chains.
In order to elucidate which of the outer a-mannosyl residues is disubstituted at C-2 and C-4, the N-linked saccharides (fractions 29-41 in Fig. 2B) were subjected to periodate oxi-  x R x x dation followed by reduction and mild acid hydrolysis (Smith degradation) as described (26). Methylation analysis of the product provided 2,4,6-tri-O-methylmannose with the concomitant loss of 3,4-di-O-methylmannose. These results indicate that the 2,4-disubstituted a-mannose is linked to C-3 of &mannose, as shown in Table 111.
Structure of Asialo Tetraantennary Saccharides and Triantennary Saccharides with One N-Acetyllactosaminyl Repeat-FAB-MS of the saccharides which eluted at peak 4 provided a molecular ion at m/z 3135 corresponding to (Fucl-Hex7. HexNAc6)R (Fig. 423). This ion was associated with A-type ions at m/z 2668 for Hex7-HexNAc; and at m/z 464 (and 432) for Hex-+HexNAc+. In addition, an ion at m/z 913 corresponding to Hex-HexNAc-Hex+HexNAc+ was detected. The latter result suggests that this peak contains a triantennary saccharide with one N-acetyllactosaminyl repeat. The presence of a triantennary saccharide was confirmed by methylation analysis (Table 11). A small amount (0.11 mol) of 2-substituted mannose (3,4,6-tri-O-methylmannose) was detected as well as 1 mol of 2,4-substituted and 0.9 mol of 2,6-substituted mannose. This result indicates that a triantennary saccharide with 2,4-and 2-substituted a-mannose is present in this saccharide fraction. This triantennary saccharide presumably contains one N-acetyllactosaminyl repeat. The presence of 3-substituted galactose supports the conclusion that this triantennary saccharide contains one Nacetyllactosaminyl repeat (Table 11).
The saccharides were sequentially digested with P-galactosidase and P-N-acetylglucosaminidase and subjected to HPLC. As shown in Fig. 6A, the products provided two peaks: the first peak eluted at 12 min corresponds to Man3. GlcNAc. (+Fuc)GlcNAcOH, and the second peak at 15 min corresponds to Gal. GlcNAc Man3. GlcNAc (+Fuc)GlcNAcOH. The ratio of two peaks was found to be 1.00.19. These results indicate that about 15% of the saccharides are triantennary saccharides with one N-acetyllactosaminyl repeat, whereas 85% of the saccharides are tetraantennary saccharides. These two molecular species, however, will provide the same molecular ion at m/z 3135 on FAB-MS analysis.
The saccharides of peak 5 were sequentially digested by Pgalactosidase and P-N-acetylglucosaminidase and subjected to HPLC. As shown in Fig. 623, the major product eluted at 15 min, which corresponds to Gal. GlcNAc' Man3 I GlcNAc.
Structure of Tetraantennary Saccharides with Two N-Acetyllactosaminyl Repeats (Lac&"ethylation analysis of peak 6 in Fig. 5A provided 1.7 mol of 3-substituted galactose, 4 mol of terminal galactose, and 1 mol each of 2,4-substituted, 2,6substituted, and 3,6-substituted mannose, in addition to other derivatives (Table 11). The results indicate that this saccharide fraction consists of tetraantennary saccharides with two N-acetyllactosaminyl repeats. FAB-MS of this oligosaccharide fraction supported the above conclusion since a fragment ion at m/z 913 corresponding to Hex-HexNAc-Hex-HexNAc' was detected (Fig. 40).
The saccharides were digested sequentially by P-galactosidase and P-N-acetylglucosaminidase to yield a major peak at 19 min (peak 2 in Fig. 6C), which corresponds to (Gal. GlcNAc)2. Mans. GlcNAc-(+Fuc)GlcNAcOH, confirming that the starting sample contains tetraantennary saccharides with two N-acetyllactosaminyl repeats. Methylation analysis of this product provided 1 mol of terminal mannose, 0.8 mol of 2,6-substituted mannose, and 0.2 mol each of 6and 4-substituted mannose, in addition to 1 mol each of 3,6substituted mannose, terminal galactose, and reducing terminal N-acetylglucosamine and 2 mol of 4-substituted Nacetylglucosamine. However, 2,4-substituted mannose or 3substituted galactose was not detected. These results indicate that each side chain arising from 2,6-substituted mannose was elongated by one N-acetyllactosaminyl repeat in 80% of the saccharides. In addition, 20% of the molecules have Nacetyllactosaminyl repeats in side chains elongating from C-6 and C-4 (Table 111).
Structure of Tetraantennary Saccharides with Three N-Acetyllactosaminyl Repeats (Lac&-The saccharides in peak 7 eluted at the position where saccharides containing seven Nacetyllactosaminyl units are expected to elute. This is because the difference of the elution time between peaks 7 and 6 is the same as that between peaks 6 and 5. The saccharides were sequentially digested with P-galactosidase and P-N-acetylglucosaminidase, and the products were analyzed by HPLC. As shown in Fig. 60 These results suggest that 65% of the peak 7 saccharides are tetraantennary saccharides with three N-acetyllactosaminyl repeats (Lac3), whereas 35% of the saccharides are tetraantennary saccharides with two N-acetyllactosaminyl repeats (Lac2). These results were confirmed by methylation analysis, as shown in Table 11. Methylation analysis of the saccharides provided 2.4 mol of 3-substituted galactose, 4 mol of terminal galactose, and 1 mol each of 2,4-, 2,6-, and 3,6substituted mannose. The same analysis indicates that about 45% of the reducing terminal N-acetylglucosamine was unreduced judging from the amount of 3-O-methyl-N-acetylglucosamine. This compound is produced when the reducing terminal N-acetylglucosamine is incompletely reduced before methylation.
In order to elucidate which side chains were elongated to form the N-acetyllactosaminyl repeat, the saccharides which eluted at 24 min (peak 3 in Fig. 6D) were methylated. The saccharides provided 0.6 mol of 2,6-substituted mannose, 0.4 mol of 6-substituted mannose, 0.3 mol of 4-substituted mannose, 0.3 mol of terminal mannose, 0.4 mol of 2-substituted mannose, and 1 mol of 3,6-substituted mannose as mannose derivatives. These results indicate that each side chain arising from 2,6-substituted mannose and the side chains arising from C-2 or C-4 of 2,4-substituted mannose were elongated by Nacetyllactosaminyl units.
Since FAB-MS of the starting materials afforded a fragment ion at m/z 1723 for NeuNAc-+Hex+HexNAc+Hex+ HexNAc-+Hex-+HexNAc' (see Fig. 3A), it is likely that Lacs saccharides contain three N-acetyllactosaminyl units in one of the side chains which are attached to C-2 or C-6 of 2,6substituted mannose or C-4 of 2,4-substituted mannose. This was confirmed by the fact that peak 3 in Fig. 6D provided a small amount (0.1 mol) of 3-substituted galactose on methylation analysis.
Structures of Asialo N-Linked Saccharides from Recombinant Erythropoietin-The results obtained above are summarized in Table 111. By measuring radioactivity in each fraction, the relative yields of saccharides were calculated. In some cases, it was necessary to obtain the ratio after exoglycosidase digestion. For example, the ratio of triantennary with one N-acetyllactosaminyl repeat and tetraantennary in peak 4 was obtained in Fig. 6A.
These results indicate that 1) the biantennary saccharide is almost exclusively in a monosialylated form; 2) the triantennary saccharides are in monosialylated or disialylated forms; 3) the tetraantennary saccharides are mostly in disialylated or trisialylated forms; and 4) the tetraantennary saccharides with one, two, or three N-acetyllactosamine repeats are mostly in disialylated or trisialylated forms. These results suggest that one of the side chains in the saccharides is almost alway terminated without a sialic acid residue (see below).
In order to know whether any difference exists among different batches of recombinant erythropoietin, two additional batches (Batches 3 and 4 in Table I) of recombinant erythropoietin were subjected to analysis. Interestingly, these samples contained more highly sialylated saccharides: the disialosyl form is 18-21% of the total saccharides; the trisialosyl form is 64-67%, the tetrasialosyl form is 10-13%; and the monosialosyl form is less than 6%. However, the relative ratios of asialo biantennary, triantennary, tetraantennary, and tetraantennary saccharides with one, two, or three Nacetyllactosaminyl repeats were almost identical among different samples. These results indicate that sialylation may vary among different batches of recombinant erythropoietin but their backbone structures are the same.
Separation of Saccharides with Different Backbone Structure but with the Same Number of Sialic Residues in Side Chains-The results of Fig. 7A suggested to us that each peak in the monosialosyl, disialosyl, or trisialosyl fraction may represent saccharides with different backbone structures but with the same number of sialic acid residues. In order to test this possibility, another 5 mg of recombinant erythropoietin (Batch 2 in Table I) was treated to yield N-linked saccharides, and these saccharides were fractionated by TSK-DEAE ionexchange chromatography. This sample provided an elution profile almost identical to that in Fig. 7A. Saccharides were divided into four (fraction 11) or three (fraction 111) fractions, desialylated, and subjected to another HPLC employing a Lichrosorb column. Fraction 11-1, which eluted earliest in TSK-DEAE chromatography, provided tetraantennary saccharides with two or three N-acetyllactosaminyl repeats (Fig.  8A, see Miniprint), whereas the last peak (fraction 11-4) mainly consists of triantennary and tetraantennary saccharides (Fig. 80). Similarly, the earliest peak in the trisialosyl fraction (fraction 111-1) provided Lac2 and a small amount of Lacl and Lac3 (Fig. 8E), whereas the last peak (fraction III-3) provided almost exclusively tetraantennary saccharides (Fig. 8G). These results indicate that the saccharides with higher numbers of N-acetyllactosamine units elute earlier than those with smaller numbers of N-acetyllactosamine units in TSK-DEAE ion-exchange chromatography.
Localization of a24-Linked Sialic Acid in the Side Chains-In order to know which side chains are preferentially sialylated, fractions 11-2, 11-3 and 111-3 were digested extensively with a mixture of P-galactosidase and P-N-acetylglucosaminidase, and the products were purified by Sephadex G-50 gel filtration followed by TSK-DEAE chromatography. The purified products were then subjected to methylation analysis, and the results are summarized as follows.
Disialosyl Tetraantennary Saccharides with One or Two N-Acetyllactosaminyl Repeats (Fraction II-2)"Methylation analysis on the exoglycosidase product of fraction 11-2 provided the following mannose derivatives: 0.9 mol each of 2,6- No other mannose derivatives, including 2,4-substituted mannose, were detected. The results indicate that the 2 sialosyl residues are almost exclusively linked to the side chains arising from C-2 and C-6 of 2,6-substituted mannose (see Fig.  11).
Structure of Sialylated Tetraantennary Saccharides with or without N-Acetyllactosamine Repeats-Based on the results described above, the structure of intact tetraantennary saccharides and those with N-acetyllactosaminyl repeats, which represent 85% of the total saccharides, can be proposed as shown in Fig. 11. The tetraantennary saccharides are mainly present as disialosyl or trisialosyl forms, and 2+3-linked sialic acid is attached to the side chains arising from C-6 and C-2 of 2,6-substituted mannose and C-4 of 2,4-substituted mannose. In the tetraantennary saccharides with N-acetyllactosaminyl repeats, sialic acid residues are always present in the side chain which contains N-acetyllactosaminyl repeats. This conclusion was supported by FAB-MS analysis. As shown in Fig. 3A, all of the fragment ions containing polylactosaminyl units are sialylated.
In order to delineate this further, trisialylated saccharides (fraction 111) were extensively digested by P-galactosidase and @-N-acetylglucosaminidase. The methylation analysis of this product showed that more than 90% of the side chain attached to C-2 of 2,4-substituted mannose was terminated without sialic acid. These combined results support the proposed structures shown in Fig. 11.
Structure of Carbohydrate Units of Urinary Erythropoietin-Since only a limited amount of urinary erythropoietin was available, the following experiments were carried out to analyze carbohydrate units of urinary erythropoietin. Glycopeptides, prepared by Pronase digestion of urinary erythropoietin, were subjected to alkaline borohydride treatment. The alkaline borohydride-treated sample was then applied to Bio-Gel P-4 gel filtration. Urinary erythropoietin saccharides provided almost the same elution profile as Fig. 2A. The glycopeptides containing N-linked saccharides (fractions 24-29) were digested by N-glycanase, and the digest was subjected to Sephadex G-50 gel filtration. Again, the elution profile of N-linked saccharides from urinary erythropoietin was almost identical to that from recombinant erythropoietin (see Fig.  2B).
Methylation analysis of N-linked saccharides (fractions 29-41 after Sephadex (3-50 gel filtration) provided partially 0methylated monosaccharide derivatives, which are almost identical to those produced from highly sialylated Batch 3 of recombinant erythropoietin (Table 11) Hex+HexNAc+ (m/z 1723) (Fig. 3B). These results are essentially the same as those obtained on recombinant erythropoietin (compare Fig. 3, A and B). N-Linked saccharides were then desialylated and subjected to HPLC employing a Lichrosorb-NHz column. As shown in Fig. 5B, urinary erythropoietin saccharides provided triantennary, tetraantennary, and tetraantennary saccharides with one, two, or three N-acetyllactosaminyl units. The relative proportion among these saccharides is almost identical to that obtained on recombinant erythropoietin except that urinary erythropoietin apparently lacks biantennary saccharides. In order to determine the relative amounts of sialylated N-linked saccharides, intact N-linked saccharides were subjected to TSK-DEAE ion-exchange chromatography. Fig. 7B shows that N-linked saccharides from urinary erythropoietin contain disialosyl(27% of the total saccharides), trisialosyl(56%), and tetrasialosyl (17%) saccharides. These results indicate that urinary erythropoietin and recombinant erythropoietin have an almost identical set of N-linked saccharide units but with slightly different sialylation depending upon the batches of recombinant erythropoietin (see above and Table I).

DISCUSSION
This paper reports the detailed structures of the carbohydrate moiety of human erythropoietin produced by recombinant DNA. The protein analyzed was produced in Chinese hamster ovary cells which were transfected with human erythropoietin cDNA (7). As far as we are aware, this is the first report on the detailed carbohydrate structure of a glycoprotein produced by recombinant DNA in comparison with the glycoprotein of natural origin. Although Mutsaers et al. (28) reported the carbohydrate structure of human y-interferon produced in Chinese hamster ovary cells, their studies did not investigate those of naturally occurring human y-interferon. The carbohydrate composition (Table I) showed that erythropoietin contains three N-linked saccharides and one 0linked saccharide, and these conclusions are consistent with the recent report on the amino acid sequence of human urinary erythropoietin (29).
The present study revealed that a large proportion of the carbohydrate moiety of recombinant erythropoietin is composed of tetraantennary saccharides with one (32.1% of the total saccharides), two (16.5%), and three (4.7%) N-acetyllactosaminyl repeats. The localization of these polylactosaminyl units was elucidated by sequential exoglycosidase digestion followed by methylation analysis, and the results are summarized as follows (see also Table 111).
When the saccharides contain one N-acetyllactosaminyl repeat, more than 70% of this repeat is preferentially attached to the side chain arising from C-6 of 2,6-substituted mannose, and 19% of the repeat is attached to that from C-4 of 2,4substituted mannose. When the saccharides contain two Nacetyllactosaminyl repeats, these repeats are attached to C-2 and C-6 of 2,6-substituted mannose in 80% of the molecules. The rest of the molecule contains N-acetyllactosaminyl repeats in the side chains arising from C-6 of 2,6-substituted mannose and C-4 of 2,4-substituted mannose. These results indicate that N-acetyllactosaminyl repeats are most preferentially added to C-6 of 2,6-substituted mannose and then to C-2 of 2,6-substituted mannose. These conclusions are consistent with previous reports on several cellular glycoproteins. For example, Cummings and Kornfeld (30) reported that the mouse lymphoma BW5147 cell line expressed a signX1cant amount of polylactosaminoglycan, whereas its mutant, which lacks the side chain arising from C-6 of 2,6-substituted mannose, expresses a minimum amount of polylactosaminoglycan. Li et al. (31) isolated polylactosaminoglycan from Chinese hamster ovary cells in which polylactosaminyl units are attached to C-2 and C-6 of 2,6-substituted mannose. Similarly, polylactosaminyl units were found in triantennary and tetraantennary saccharides of various origins (24,(32)(33)(34). These results appear to establish that N-acetyllactosaminyl repeats are preferentially added to C-6 of 2,6-substituted mannose and then to C-2 of 2,6-substituted mannose. These 2,6-substituted mannose residues are usually linked to the C-6 side of @-mannose. In human erythrocytes, polylactosaminyl elongation can be found in the side chain arising from C-2 of (Ymannose which is linked to C-6 of P-mannose (23, 35). It is likely that human erythroid cells contain very little activity of the N-acetylglucosaminyltransferase which forms a GlcNAcpl4Man branch. As a result, N-acetyllactosamine repeats are formed on the secondary preferable side chain, which is attached to C-2 of a-mannose linked to C-6 of 8mannose, in these erythroid cells.
This study showed that human erythropoietin exclusively contains a2-3-linked sialic acid. This fact allowed us to elucidate the locaiization of 2+3-linked sialic acid residues among different side chains. This was achieved by extensive digestion of intact saccharides with P-galactosidase and P-Nacetylglucosaminidase followed by methylation analysis of the products. These results can be summarized as follows. 1) When typical triantennary or tetraantennary saccharides contain 2 sialic acid residues, they are attached to C-2 and C-6 of 2,6-substituted mannose or C-6 of 2,6-substituted mannose and C-4 of 2,4-substituted mannose. 2) When saccharides with N-acetyllactosaminyl repeats contain 2 sialic acid residues, they are attached almost exclusively to C-2 and C-6 of 2,6-substituted mannose. This localization is essentially identical to that of N-acetyllactosamine repeats. Thus, it is apparent that polylactosamine is preferably sialylated through an aB-+S-linkage. These results are consistent with our previous results obtained on polylactosaminoglycans from chronic myelogenous leukemia cells; a2-3-linked sialic acid is present on side chains arising from C-6 and C-2 of 2,6-substituted mannose and C-4 of 2,4-substituted mannose, and those side chains are longer than that terminating with 24-linked sialic acid (24). Similar results were obtained in human erythrocyte Band 3 polylactosaminoglycans; polylactosaminyl side chains arising from the C-6 side of p-mannose are sialylated through a 2-+ 3-linkage, whereas the shorter chain arising from the C-3 side is sialylated through a 24-linkage (23,35). Similarly, Yamashita et al. (33) and Markle and Cummings (36) found that longer polylactosaminyl side chains are almost exclusively sialylated through a 2+3-linkage. Interestingly, short polylactosamine chains in thyroid cell glycoprotein Gp-1 (37) and BW5147 (36) are terminated with 2 4 -l i n k e d sialic acid. Our results also showed that almost no 2-3-linked sialic acid is attached to the side chain arising from C-2 of 2,4-substituted mannose. It is noteworthy that this side chain was found to be exclusively sialylated through a 24-linkage in many glycoproteins including normal and leukemic granulocyte polylactosaminoglycans (19-21, 24, 34, 38-41).
By using a bovine colostrum a2-&-sialyltransferase, Joziasse et al. (42) have shown that preferential sialylation takes place first on C-2 of 2,4-substituted mannose and then on C-4 of 2,4-substituted mannose. This branch (or side chain) specificity appears to be opposite to the distribution of 2-3linked sialic acid. Thus, it is likely that a2+=3-~ialyltransferase and a24-sialyltransferase have complementary specificity toward different side chains. Furthermore, our results raise the possibility that the side chains containing polylactosaminyl units would be preferable sites for 2+3-linked sialylation.
This study demonstrated that the carbohydrate moiety of human erythropoietin isolated from human urine is indistinguishable from that of recombinant erythropoietin except for a difference in degree of sialylation. Urinary erythropoietin has a similar degree of sialylation as the highly sialylated batch of recombinant erythropoietin (Tables I and 11). The recombinant erythropoietin was produced in Chinese hamster ovary cells, and urinary erythropoietin is presumably derived from human kidney cells. The results therefore suggest two possibilities. 1) Chinese hamster ovary and human kidney cells contain similar glycosyltransferases. 2) The protein acceptor itself influences glycosylation even when a similar set of glycosyltransferases are not present in two cell types. It will be interesting to see if the carbohydrate moiety of erythropoietin produced in other mammalian cells is similar to those elucidated in this study. This study also demonstrated that the major carbohydrate units of erythropoietin are tetraantennary saccharides with or without N-acetyllactosa-mine repeats. It has been shown that rat liver cells uptake the asialo form of glycoproteins which contain tri-or tetraantennary saccharides (43). It is therefore reasonable that the asialo form of erythropoietin is taken up by liver cells through a galactose-binding protein (15). Our preliminary studies showed that a portion of intact erythropoietin of both recombinant and urinary origins is taken up by rat liver cells, presumably because of the incomplete sialylation. It will be interesting to test if sialylation by a24-sialyltransferase elongates the serum concentration of erythropoietin and sustains in vivo activity longer than the starting erythropoietin.

B .
O e s i a l y l a t e d t e t r a -a n t e n n a l a n d t r i -a n t e n n a r w i t h one N-acet llactoramine raccharlde. except for the following. change ChPanatography as shown i n F i g .

4 .
Each peak obtained was d e l i a l y l a t e d and subj e c t e d t o HPLC M i t h a Lichrerorb-NH2 column I S the ram Condition of Fig. 5 . A. 11-1 ( f r a c t i o n s 70-77, i n Fig. la).