The Absence of Fucose but Not the Presence of Galactose or Bisecting N-Acetylglucosamine of Human IgG1 Complex-type Oligosaccharides Shows the Critical Role of Enhancing Antibody-dependent Cellular Cytotoxicity*

An anti-human interleukin 5 receptor (hIL-5R) humanized immunoglobulin G1 (IgG1) and an anti-CD20 chimeric IgG1 produced by rat hybridoma YB2/0 cell lines showed more than 50-fold higher antibody-dependent cellular cytotoxicity (ADCC) using purified human peripheral blood mononuclear cells as effector than those produced by Chinese hamster ovary (CHO) cell lines. Monosaccharide composition and oligosaccharide profiling analysis showed that low fucose (Fuc) content of complex-type oligosaccharides was characteristic in YB2/0-produced IgG1s compared with high Fuc content of CHO-produced IgG1s. YB2/0-produced anti-hIL-5R IgG1 was subjected to Lens culinaris aggulutin affinity column and fractionated based on the contents of Fuc. The lower Fuc IgG1 had higher ADCC than the IgG1 before separation. In contrast, the content of bisecting GlcNAc of the IgG1 affected ADCC much less than that of Fuc. In addition, the correlation between Gal and ADCC was not observed. When the combined effect of Fuc and bisecting GlcNAc was examined in anti-CD20 IgG1, only a severalfold increase of ADCC was observed by the addition of GlcNAc to highly fucosylated IgG1. Quantitative PCR analysis indicated that YB2/0 cells had lower expression level of FUT8 mRNA, which codes α1,6-fucosyltransferase, than CHO cells. Overexpression of FUT8 mRNA in YB2/0 cells led to an increase of fucosylated oligosaccharides and decrease of ADCC of the IgG1. These results indicate that the lack of fucosylation of IgG1 has the most critical role in enhancement of ADCC, although several reports have suggested the importance of Gal or bisecting GlcNAc and provide important information to produce the effective therapeutic antibody.

An anti-human interleukin 5 receptor (hIL-5R) humanized immunoglobulin G1 (IgG1) and an anti-CD20 chimeric IgG1 produced by rat hybridoma YB2/0 cell lines showed more than 50-fold higher antibody-dependent cellular cytotoxicity (ADCC) using purified human peripheral blood mononuclear cells as effector than those produced by Chinese hamster ovary (CHO) cell lines. Monosaccharide composition and oligosaccharide profiling analysis showed that low fucose (Fuc) content of complex-type oligosaccharides was characteristic in YB2/0-produced IgG1s compared with high Fuc content of CHO-produced IgG1s. YB2/0-produced anti-hIL-5R IgG1 was subjected to Lens culinaris aggulutin affinity column and fractionated based on the contents of Fuc. The lower Fuc IgG1 had higher ADCC than the IgG1 before separation. In contrast, the content of bisecting GlcNAc of the IgG1 affected ADCC much less than that of Fuc. In addition, the correlation between Gal and ADCC was not observed. When the combined effect of Fuc and bisecting GlcNAc was examined in anti-CD20 IgG1, only a severalfold increase of ADCC was observed by the addition of GlcNAc to highly fucosylated IgG1. Quantitative PCR analysis indicated that YB2/0 cells had lower expression level of FUT8 mRNA, which codes ␣1,6-fucosyltransferase, than CHO cells. Overexpression of FUT8 mRNA in YB2/0 cells led to an increase of fucosylated oligosaccharides and decrease of ADCC of the IgG1. These results indicate that the lack of fucosylation of IgG1 has the most critical role in enhancement of ADCC, although several reports have suggested the importance of Gal or bisecting GlcNAc and provide important information to produce the effective therapeutic antibody.
Antibody-dependent cellular cytotoxicity (ADCC), 1 a lytic attack on antibody-targeted cells, is triggered upon binding of lymphocyte receptors (Fc␥Rs) to the constant region (Fc) of the antibodies. ADCC is considered to be a major function of some of the therapeutic antibodies, although antibodies have multiple therapeutic functions (e.g. antigen binding, induction of apoptosis, and complement-dependent cellular cytotoxicity) (1,2).
One IgG molecule contains two N-linked oligosaccharide sites in its Fc region (3). The general structure of N-linked oligosaccharide on IgG is complex-type, characterized by a mannosyl-chitobiose core (Man3GlcNAc2-Asn) with or without bisecting GlcNAc/L-fucose (Fuc) and other chain variants including the presence or absence of Gal and sialic acid. In addition, oligosaccharides may contain zero (G0), one (G1), or two (G2) Gal.
Recent studies have shown that engineering the oligosaccharides of IgGs may yield optimized ADCC. ADCC requires the presence of oligosaccharides covalently attached at the conserved Asn 297 in the Fc region and is sensitive to change in the oligosaccharide structure. In the oligosaccharide, sialic acid of IgG has no effect on ADCC (4). The relationship between the Gal residue and ADCC is controversial. Boyd et al. (4) have shown that obvious change was not found in ADCC after removal of the majority of the Gal residues. However, several reports have shown that Gal residues enhance ADCC (5,6).
Several groups have focused on bisecting GlcNAc, which is a ␤1,4-GlcNAc residue transferred to a core ␤-mannose (Man) residue, and it has been implicated in biological activity of therapeutic antibodies (7). N-Acetylglucosaminyltransferase III (GnTIII), which catalyzes the addition of the bisecting Glc-NAc residue to the N-linked oligosaccharide (8), has been expressed in a Chinese hamster ovary (CHO) cell line with an anti-neuroblastoma IgG1 and resulted in greater ADCC (9). Moreover, expression of GnTIII in a recombinant CHO cell line has led to the increase in ADCC of the anti-CD20 antibody (10).
Recently, Shields et al. have revealed the effect of fucosylated oligosaccharide on antibody effector functions, including binding to human Fc␥R, human C1q, human FcRn, and ADCC (11). The Fuc-deficient IgG1s have shown 50-fold increased binding to Fc␥RIIIa and enhanced ADCC. Nevertheless, there are no data on comparison of the effect of Fuc, Gal, and GlcNAc or the combined effect of Fuc and bisecting GlcNAc.
Here, we describe the correlation between glycosylation of human IgG1 and ADCC and demonstrate that Fuc showed the critical role for enhancing ADCC out of several sugar residues reported previously. We unexpectedly found that human IgG1 produced by rat hybridoma YB2/0 cells showed extremely high ADCC at more than 50-fold lower concentration of those pro-duced by CHO cells. YB2/0-produced IgG1 had lower Fuc content than CHO-produced IgG1. IgG1 containing lower fucosylated oligosaccharides, which was fractionated by Lens culinaris aggulutin (LCA) lectin affinity chromatography, showed higher ADCC before separation. In contrast, the addition of bisecting GlcNAc to IgG1 enhanced ADCC much less effectively than defucosylation. The effect of bisecting GlcNAc was only observed in highly fucosylated IgG1. YB2/0 cells expressed a lower level of FUT8 (␣1,6-fucosyltransferase gene) mRNA than CHO cells, and overexpression of FUT8 in YB2/0 led the increase of fucosylation of IgG1 and the decrease of ADCC.

EXPERIMENTAL PROCEDURES
Cell Lines-Rat hybridoma YB2/0 cells were purchased from the American Type Culture Collection (ATCC; CRL-1662). CHO cell line DG44 (12), for wild type IgG1 production, was kindly provided by Dr. Lawrence Chasin (Columbia University). LEC10, a variant CHO cell line overexpressing GnTIII (13), was kindly provided by Dr. Pamela Stanley (Albert Einstein College of Medicine).
Expression of IgG1s-For the generation of human IgG1 of humanized anti-human interleukin-5 receptor (hIL-5R) ␣-chain and chimeric anti-CD20, the appropriate humanized or murine VL and VH cDNAs were subcloned into the previously described pKANTEX93 vector (14). The cDNA coding for the VL and VH region of each antibody was constructed by the PCR-based method (14). In the case of chimeric anti-CD20 antibody, the cDNA sequence of each V region was designed as the same with that of Rituxan TM (VL: GenBank TM accession number AR015962; VH: GenBank TM accession number AR000013). Establishment of anti-hIL-5R humanized IgG1 will be described elsewhere. 2 Antibody expression vectors were introduced into YB2/0 cells or DG44 cells via electroporation and selected for gene amplification in methotrexate-containing medium (14).
Production of IgGs-The anti-hIL-5R IgG1-producing YB2/0 cell line was suspended in GIT medium (Wako, Osaka, Japan) containing 0.5 mg/ml G418 and 200 nM methotrexate to give a density of 3 ϫ 10 5 cells/ml and dispensed in suspension culture flasks (Greiner, Frickenhausen, Germany). The anti-hIL-5R IgG1-producing CHO cell line was suspended in the EX-CELL302 medium (JRH, Kansas City, MO) containing 3 mM L-Gln, 0.5% chemically defined lipid concentrate (Invitrogen), and 0.3% PLURONIC F-68 (Invitrogen) to give a density of 3 ϫ 10 5 cells/ml and cultured using spinner flasks (Asahi Techno Glass, Tokyo, Japan) under agitating at a rate of 100 rpm. The anti-CD20 IgG1producing YB2/0 cell line was suspended in the hybridoma-SFM medium (Invitrogen) containing 5% Daigo's GF21 (Wako) and 200 nM methotrexate to give a density of 1 ϫ 10 5 cells/ml and dispensed in suspension culture flasks. The flasks were incubated under conditions of 37°C in humid air containing 5% CO 2. After 8 or 10 days of incubation, the culture supernatants were recovered.
Purification of IgG1s-The culture supernatants containing anti-hIL-5R IgG1 from YB2/0 cells and CHO cells and anti-CD20 IgG1 from YB2/0 cells were clarified by centrifugation and passed through a 0.2-m filter. The IgG1 bound to a PROSEP-A (Millipore) column was eluted with 0.1 M citrate buffer (pH 3.5). Then the antibody was subjected to a Sephacryl S-300 (Amersham Biosciences) column. The buffer composition of the YB2/0-produced anti-CD20 IgG1 was changed to that for Rituxan TM (9.0 mg/ml sodium chloride, 7.35 mg/ml sodium citrate dihydrate, 0.7 mg/ml polysorbate 80). The purity of IgG1 was confirmed by SDS-PAGE. YB2/0-and CHO-produced humanized anti-hIL-5R IgG1 were designated as KM8399 and KM8404, respectively. YB2/0produced chimeric anti-CD20 IgG1 was designated as KM3065. Rituxan TM (chimeric mouse/human anti-CD20 monoclonal antibody derived from the CHO cell line) was purchased from Genentech (South San Francisco, CA)/IDEC Pharmaceutical (San Diego, CA).
ADCC Assay for the Anti-hIL-5R IgG1-An ADCC assay was performed by 51 Cr release assay as reported previously (15). Briefly, target cells (1 ϫ 10 6 ), a murine T cell line CTLL-2 (h5R) expressing hIL-5R ␣-chain and ␤-chain (16), were labeled with 3.7 MBq of Na 2 51 CrO 4 at 37°C for 1.5 h. Human effector cells were peripheral blood mononuclear cells (PBMC) purified from healthy donors using Polymorphprep (Nycomed Pharma AS, Roskilde, Norway). Aliquots of the 51 Cr-labeled target cells were dispensed into 96-well U-bottomed plates (1 ϫ 10 4 /50 l) and incubated with serial dilutions of antibodies (50 l) in the presence of human effector cells (100 l) at an E/T ratio of 90/1. After 4 h of incubation at 37°C, the plates were centrifuged, and the radioactivity in the supernatants was measured using a ␥ counter. The percentage of specific cytolysis was calculated from the counts of samples according to the formula, % specific lysis ϭ 100 ϫ ͑E Ϫ S͒/͑M Ϫ S͒ (Eq. 1) where E represents the experimental release (cpm in the supernatant from target cells incubated with antibody and effector cells), S is the spontaneous release (cpm in the supernatant from target cells incubated with medium alone), and M is the maximum release (cpm released from target cells lysed with 1 mol/liter HCl). ADCC Assay for the Anti-CD20 IgG1-An ADCC assay was performed by a lactate dehydrogenase release assay. Aliquots of target cells, a human B lymphoma cell line Raji (number 9012, purchased from JCRB, Tokyo, Japan), were distributed into 96-well U-bottomed plates (1 ϫ 10 4 /50 l) and incubated with serial dilutions of antibodies (50 l) in the presence of human effector cells (100 l) at an E/T ratio of 25:1 or 20:1. Human effector cells were PBMC purified from healthy donors using Lymphoprep (Axis Shield, Dundee, UK). After a 4-h incubation at 37°C, the plate was centrifuged, and the lactate dehydrogenase activity in the supernatants was measured using a nonradioactive cytotoxicity assay kit (Promega, Madison, WI). The percentage of specific cytolysis was calculated from the activities of samples according to the formula, where E represents the experimental release (activity in the supernatant from target cells incubated with antibody and effector cells), S E is the spontaneous release in the presence of effector cells (activity in the supernatant from effector cells), S T is the spontaneous release of target cells (activity in the supernatant from target cells incubated with medium alone), and M is the maximum release of target cells (activity released from target cells lysed with 9% Triton X-100).
Profiling Analysis of N-Linked Oligosaccharides-Oligosaccharides were prepared from 100 g of IgG1 by hydrazinolysis and N-acetylation by the methods reported previously (17). Oligosaccharides were pyridylaminated (17) and analyzed by reverse-phase HPLC using a Shimpack CLC-ODS column (60 ϫ 150 mm; Shimadzu, Kyoto, Japan) (18) with slight modifications. Elution was performed at a flow rate of 1.0 ml/min at 55°C using two solvents, A and B. Solvent A was 10 mM sodium phosphate buffer (pH 3.8), and solvent B was 10 mM sodium phosphate buffer (pH 3.8) containing 0.5% 1-butanol (Sigma). The column was equilibrated with solvent A. After injection of sample, the ratio of solvent B to A was increased with a linear gradient to 60:40 in 80 min. The elution profile was monitored by fluorescence detection with excitation at 320 nm and emission at 400 nm. The oligosaccharide peak assignments were made according to retention time comparison with PA-labeled oligosaccharide standards (TaKaRa Bio, Otsu, Japan). The peak area was used to calculate the percentage of each oligosaccharide, since the relative fluorescence was the same on a molar basis for each component (17). To identify the structure, each oligosaccharide fraction that separated as a peak was collected and evaporated to dryness under vacuum prior to matrix-assisted laser desorption/ionization time-of-flight mass spectrometry.
Monosaccharide Composition of IgG1s-Monosaccharides were released from an aliquoted quantity of IgG1 by heating with 4 M trifluoroacetic acid at 100°C for 2 h. Monosaccharides were dried under vacuum and reconstituted in water prior to high performance anion exchange chromatography analysis. Monosaccharides were analyzed using a waveform and DX500 system (DIONEX, Sunnyvale, CA) described previously (19). A CarboPac PA-1 column (DIONEX) was used to resolve monosaccharides with a flow rate of 0.8 ml/min at 35°C. After injection of samples, the monosaccharides were resolved with 18 mM NaOH for 30 min, and the column was regenerated by elution with 500 mM NaOH for 10 min. The column was held 18 mM NaOH for 30 min prior to the next injection.
Lectin Affinity HPLC-LCA or Phaseolus vulgaris E 4 (PHA-E 4 ) were used as ligands of lectin affinity HPLC for separation of IgG1 based on the content of Fuc or bisecting GlcNAc, respectively. An LA-LCA (4.6 ϫ 150 mm; HONEN, Tokyo, Japan) column was installed in the LC-6A HPLC system (Shimadzu). Purified antibodies dissolved in 10 mM KH 2 PO 4 were applied to the column previously equilibrated with 50 mM Tris-H 2 SO 4 (pH 7.3). The column was eluted with a linear gradient of the buffer containing 0.2 M ␣-methyl-D-mannoside (Nacalai Tesque, Kyoto, Japan) in 60 min. An LA-PHA-E 4 (4.6 ϫ 150 mm; HONEN) column was installed in the LC-6A HPLC system (Shimadzu). Purified antibodies were applied to the column, previously equilibrated with 50 mM Tris-H 2 SO 4 (pH 8.0) (A buffer). The column was eluted with the buffer containing 0.1 M K 2 B 4 O 7 (B buffer). Elution was followed by linear gradients from 0 to 58% B buffer in 35 min and then 100% B buffer for 5 min. The column was equilibrated with 100% A buffer for 20 min before the next injection. Each chromatography was performed at room temperature and a flow rate of 0.5 ml/min. Establishment of ␣1,6-Fucosyltransferase-overexpressing YB2/0 Cells-A mammalian expression vector, pAGE249, which was derived by excision of a 2.7-kb SphI-SphI fragment containing the dihydrofolate reductase gene expression cassette from pAGE249 (20), was employed. This plasmid contained a hygromycin-resistant gene driven by the herpes simplex virus thymidine kinase gene promoter. The murine FUT8 cDNA (1728 bp) (21) was inserted into pAGE248 under the control of the Moloney murine leukemia virus 3Ј-LTR promoter to form a plasmid designated as pAGEmfFUT8. Two FspI restriction sites located within the plasmid backbone enabled linearization of the expression vector construct, prior to transfection of cells. FUT8 expression vector, pAGEmfFUT8, was introduced into anti-CD20-IgG1-producing YB2/0 cells via electroporation, and FUT8-overexpressing YB2/0 cells, 3065ft8 -72, were selected in 0.5 mg/ml hygromycin-B (Sigma)-containing medium.
Competitive RT-PCR Analysis of FUT8 -Total RNA was isolated from 1.0 ϫ 10 7 YB2/0 cells, FUT8-overexpressing YB2/0 cells, or CHO/ DG44 cells using the RNeasy minikit (Qiagen, Tokyo, Japan) and incubated for 1 h at 37°C with 20 units of RNase-free DNase (RQ1; Promega) to degrade genomic DNA. After DNA digestion, the total RNA was purified again using the RNeasy minikit. The single-strand cDNA was synthesized from 3 g of each total RNA using the Superscript first strand synthesis system for RT-PCR (Invitrogen). The 50-fold diluted reaction mixture was used as a template for the following competitive RT-PCR. Quantification of FUT8 transcripts was carried out using competitive RT-PCR in which a 979-bp partial fragment of rat FUT8 cDNA was used as a standard DNA and a 772-bp fragment deleting a 207-bp ScaI (blunt)-HindIII (blunt) fragment from standard DNA was used as a competitor. Standard DNA was amplified from singlestranded cDNAs of YB2/0 cells by PCR using primers 5Ј-ACTCATCT-TGGAATCTCAGAATTGG-3Ј and 5Ј-CTTGACCGTTTCTATCTTCTCT-CG-3Ј. PCRs for detection of FUT8 were carried out by heating at 94°C for 3 min and subsequent 32 cycles of 94°C for 1 min, 60°C for 1 min, and 72°C for 1 min in 20 l of reaction mixture containing 5 l of the 50-fold diluted single-stranded cDNA, 10 fg of linearized competitors, 10 pmol of FUT8-specific primers, 4 nmol of dNTP mixture, 5% dimethyl sulfoxide, and ExTaq polymerase (Takara Bio). The sense primer 5Ј-GTCCATGGTGATCCTGCAGTGTGG-3Ј and antisense primer 5Ј-CACCAATGATATCTCCAGGTTCC-3Ј were employed to amplify an FUT8 fragment. Aliquots of PCR products (7 l) were subjected to electrophoresis in 1.75% agarose gel and stained with SYBR Green I nucleic gel stain (Molecular Probes, Inc., Eugene, OR). The amount of products was quantified by measuring luminescence intensity using a FluoroImager SI (Amersham Biosciences), calculated from standard curves, and converted into molar numbers. To normalize the synthesis efficiency of first-strand cDNAs, the amount of ␤-actin transcripts was also quantified by competitive RT-PCR in which a 1128-bp partial fragment of rat ␤-actin cDNA was used as a standard DNA and a 948-bp fragment deleting a 180-bp DraIII (blunt)-DraIII (blunt) fragment from standard DNA as a competitor were employed. Standard DNA was amplified from single-stranded cDNAs of YB2/0 cells by PCR using primers 5Ј-ATTTAAGGTACCGAAGCATTTGCGGTGCACGATGGAG-GGG-3Ј and 5Ј-AAGTATAAGCTTACATGGATGACGATATCGCTGCG-CTCGT-3Ј. PCRs for detection of ␤-actin were carried out by heating at 94°C for 3 min and subsequent 17 cycles of 94°C for 30 s, 65°C for 1 min, and 72°C for 2 min in 20 l of reaction mixture containing 5 l of the 50-fold single-stranded cDNAs, 1 pg of linearized competitors, 10 pmol of ␤-actin-specific primers, 4 nmol of dNTP mixture, 5% dimethyl sulfoxide, and ExTaq polymerase (TakaRa Bio). The sense primer 5Ј-GATATCTGCTGCGCTCGTCGTCGAC-3Ј and antisense primer 5Ј-CA-GGAAGGAAGGCTGGAAGAGAGC-3Ј were designed to amplify an ␤-actin fragment. Aliquots of PCR products (7 l) were subjected to electrophoresis in 1.75% agarose gel for analysis.

RESULTS
ADCC of YB2/0-and CHO-produced IgG1-The purified humanized anti-hIL-5R IgG1 antibodies, KM8399 (YB2/0-produced) and KM8404 (CHO-produced), were compared for their ability to induce ADCC against a murine T cell line CTLL-2 (h5R) expressing hIL-5R ␣-chain and ␤-chain. Human periph-eral blood mononuclear cells were used as effector cells for ADCC. Both humanized KM8399 and KM8404 showed high affinity to the soluble hIL-5R ␣-chain antigen and had no differences in antigen binding in enzyme-linked immunosorbent assay (data not shown). In contrast, the ADCC of KM8399 was ϳ50-fold higher than that of KM8404 at the concentration of antibody of 25% cytotoxicity that was the maximum activity of KM8404 (Fig. 1A), indicating that YB2/0-produced IgG1 promoted killing of IL-5R-positive cells at an ϳ50-fold lower concentration than the CHO-produced IgG1.
To confirm the reproducibility of this result, we assessed the ADCC of another antibody, chimeric anti-CD20 IgG1. Rituxan TM was CHO-produced chimeric anti-CD20 IgG1 approved as a therapeutic agent in non-Hodgkin's lymphoma. We originally established YB2/0-produced chimeric anti-CD20 IgG1, KM3065, that had the same V region amino acid sequences as Rituxan TM . Both chimeric Rituxan TM and KM3065 exhibited the same antigen binding activity in flow cytometric analysis using CD20-positive cell lines (data not shown). On the other hand, the ADCC of KM3065 was at least 100-fold higher than that of Rituxan TM at the concentration of antibody of 20% cytotoxicity that was the maximum activity of Rituxan TM (Fig. 1B). Moreover, the maximum cytotoxicity of YB2/0-produced KM8399 and KM3065 was 2-3-fold higher than that of CHO-produced KM8404 and Rituxan TM , respectively (Fig. 1, A and B).
Oligosaccharide Analysis-To elucidate the molecular basis of the difference of ADCC between YB2/0-and CHO-produced IgG1, we analyzed the protein portion and oligosaccharide portion of KM8399 and KM8404. There was no significant difference in SDS-PAGE, peptide mapping, and CD (data not shown), suggesting the importance of oligosaccharides for controlling ADCC. Hydrazinolysis-derived oligosaccharides were labeled with 2-aminopyrizine and separated by HPLC (Fig. 2, A and B). As shown in Fig. 2, A and B, peak patterns and content of oligosaccharide between YB2/0-produced KM8399 and CHO-produced KM8404 were quite different. KM8399 contained nine major oligosaccharides (peaks a, b, c, e, f, g, h, k, and l); in contrast, KM8404 contains five major oligosaccharides (peaks a, e, f, g, and h). The oligosaccharide structure of each peak is shown in Fig. 2C, and all structures have been found in human IgG N-linked oligosaccharides as natural structures (22,23).
To clear the difference of oligosaccharides found in Fig. 2, quantitative monosaccharide compositions of IgG1s were determined (Table I). The contents of Fuc, Gal, and GlcNAc were different between YB2/0-produced IgG1s and CHO-produced IgG1s. KM8399 and KM3065 (YB2/0-produced) contained 0.8and 0.09-fold lower content of Fuc than KM8404 and Rituxan TM (CHO-produced), respectively. In contrast, the two YB2/ 0-produced IgG1s showed a higher content of GlcNAc than the two CHO-produced IgG1s. The difference in the content of Gal between YB2/0-produced IgG1s and CHO-produced IgG1s was not consistent. These results suggest the difference of ADCC between YB2/0-produced IgG1s and CHO-produced IgG1s is caused by that of oligosaccharide structure, especially Fucand/or GlcNAc-containing oligosaccharides.
Lectin Affinity Chromatography-To analyze the effect of Fuc-containing and bisecting GlcNAc-containing oligosaccharides on ADCC, YB2/0-produced anti-hIL-5R IgG1 KM8399 was fractionated based on the content of Fuc or bisecting Glc-NAc by two lectin affinity chromatographies, LCA (Fig. 3) or PHA-E 4 (Fig. 4). KM8399 was fractionated by LCA lectin affinity chromatography to the fraction I with lower Fuc content and the fraction II with higher Fuc content before separation ( Fig. 3A and Table II). The contents of nonfucosylated IgG1 of the fraction I, II, and unseparated fraction were 100, 15, and 34%, respectively. As shown in Fig. 3B, ADCC of the fraction I was enhanced 10-fold before separation; nevertheless, that of the fraction II was decreased 10-fold. These results indicated that Fuc content is inversely correlated to ADCC of anti-hIL-5R IgG1.
Several groups have focused on the relationship between ADCC and bisecting GlcNAc, although KM8399 had few contents of bisecting GlcNAc-binding oligosaccharides (4% , Table  II). To examine the relationship, KM8399 was separated to fractions III and IV based on the content of bisecting GlcNAc by PHA-E 4 column (Fig. 4A) followed by sequential separation of fraction IIIЈ and IVЈ by LCA chromatography (Fig. 4, B and C) to reduce the effect of nonfucosylated oligosaccharides. As shown in Table III, fraction IIIЈ contained no bisecting GlcNAcbinding oligosaccharide (0%), and fraction IVЈ had more content of bisecting GlcNAc-binding oligosaccharides (30%) than before separation (4%). On the other hand, the content of Fuc of each fraction was consistent (fraction IIIЈ, 88%; fraction IVЈ, 90%). As shown in Fig. 4D, fractions IIIЈ and IVЈ showed no significant difference in ADCC, suggesting bisecting GlcNAc at least under 30% content in the oligosaccharides has little effect in ADCC of anti-hIL-5R IgG1.
YB2/0-produced KM3065 contained bisecting GlcNAc-binding nonfucosylated oligosaccharides with a relatively high percentage of 16%. Although oligosaccharides of such structure are quite small in quantity (less than 1%), those oligosaccharides are detectable in serum IgG derived from normal human (22,24). To analyze the combined effect of Fuc and bisecting Glc-NAc, KM3065 was separated to four fractions, V, VI, VII, and VIII, based on the content of bisecting GlcNAc by PHA-E 4 lectin affinity chromatography (Fig. 5A). As shown in Table IV, each fraction had a different content of bisecting GlcNAc-binding oligosaccharides (from 0 to 45%); however, the content of nonfucosylated oligosaccharides was not significantly different. All four fractions showed almost the same ADCC to unseparated KM3065 (Fig. 5B), indicating that the content of bisecting GlcNAc at least under 45% did not affect the ADCC of anti-CD20 chimeric IgG1, which contained around 90% content of the nonfucosylated oligosaccharides.
To verify the combined effect of bisecting GlcNAc with Fuc, LEC10 cells, a variant CHO cell line overexpressing GnTIII (12), were used to produce chimeric anti-CD20 monoclonal IgG1. In oligosaccharide analysis, bisecting GlcNAc-binding fucosylated oligosaccharides were the majority on LEC10-produced anti-CD20 IgG1 (74% of bisecting GlcNAc and 100% of Fuc; data not shown), whereas bisecting GlcNAc-nonbinding FIG. 2. Oligosaccharide profiles of anti-hIL-5R IgG1. Oligosaccharides derived from IgG1s produced by YB2/0 (A) and CHO (B) cell cultures were pyridylaminated and applied to reverse-phase HPLC as described under "Experimental Procedures." The alphabetical peak codes correspond to those of oligosaccharide structure in C. A conserved oligosaccharide core, linked to the Asn, is composed of three Man and two GlcNAc monosaccharide residues. Additional GlcNAcs are normally ␤1,2-linked to the ␣6 Man and ␣3 Man (␣6 and ␣3 arms, respectively), whereas the monosaccharide residues in boldface type, Gal, Fuc, and the bisecting GlcNAc (boxed), can be present or absent. Peaks indicated with asterisks are artifacts of pyridylamino derivatives (33). The percentage of nonfucosylated oligosaccharides was calculated by the following equation: total peak area of a, b, c, d, i, and j)/(total peak area) ϫ 100. fucosylated oligosaccharides were the majority of Rituxan TM (0% of bisecting GlcNAc and 94% of Fuc; data not shown). In the ADCC assay, LEC10-produced IgG1 showed only severalfold enhancement of ADCC compared with Rituxan TM (Fig.  5B). In addition, YB2/0-produced KM3065 with low Fuc content (19% of bisecting GlcNAc and 9% of Fuc; Table IV) exhibited ϳ100-fold higher ADCC than LEC10-produced IgG1. These results suggest that a relatively high content of bisecting GlcNAc (74%) improves ADCC when IgG1 has a high content of fucosylated oligosaccharides, although the impact of the content of bisecting GlcNAc in ADCC is much less than that of Fuc. Expression Levels of ␣1,6-Fucosyltransferase in YB2/0 Cells and CHO Cells-FUT8 is considered to be the only gene coding ␣1,6-fucosyltransferase, which catalyzes the transfer of Fuc from GDP-Fuc to GlcNAc in ␣1,6-linkage of complex-type oligosaccharides, because no homologous gene has been found (25). To elucidate the mechanism of lower Fuc content of YB2/ 0-produced IgG1 than that of CHO-produced IgG1, we examined expression levels of FUT8 in each cell line.
Quantification of FUT8 transcripts was performed using competitive RT-PCR. To normalize the synthesis efficiency of first-strand cDNAs, the amounts of ␤-actin transcripts were also quantified by competitive RT-PCR. These PCR analyses, which were performed independently three times, revealed that an expected FUT8 fragment from YB2/0 cells has poor intensity compared with CHO cells (Fig. 6A). The expression level of FUT8 transcripts in YB2/0 cells was shown to be 0.1% relative to ␤-actin transcripts, whereas CHO cells have 2.0% FUT8 transcripts to ␤-actin transcripts. There was no significant difference observed when we performed PCR analysis using primers and competitor DNAs specific for Chinese hamster FUT8 and ␤-actin. These results suggest that lower expression of FUT8 mRNA is the cause for lower content of Fuc of IgG1 produced by YB2/0 cells.
Overexpression of ␣1,6-Fucosyltransferase in YB2/0 Cells-To confirm the effects of FUT8 expression on the ADCC of IgG1 produced by YB2/0 cells, FUT8-overexpressing YB2/0 cells were established by transfection of murine FUT8 cDNA to anti-CD20 IgG1-producing YB2/0 cells. The most highly FUT8expressing clone that showed 145.5% FUT8 transcripts relative to ␤-actin transcripts was selected and designated as 3065ft8-72 (Fig. 6A). The ADCC of anti-CD20 IgG1 produced by 3065ft8 -72 cells was 100-fold lower than that produced from the original YB2/0 cells and equivalent to the commercially available anti-CD20 IgG1 Rituxan TM produced by normal CHO cells (Fig. 6B). Monosaccharide analysis of each IgG1 showed that there was no significant difference between the IgG1 from FUT8-overexpressing cells and the original YB2/0 cells except that a higher amount of Fuc (81%; data not shown) was detected in the IgG1 from FUT8-overexpressing cells. These findings strongly suggest that FUT8 acts as a key gene in YB2/0 cells to affect the Fuc content as well as ADCC of IgG1. DISCUSSION In this study, we analyzed the molecular basis of extremely high ADCC of recombinant IgG1 produced by rat hybridoma YB2/0 cells, which produced IgG1 with at least 50-fold higher ADCC than that produced by CHO cells, one of the most widely used host cell lines for production of recombinant antibodies (Fig. 1). Our conclusion of the present study is that nonfucosylated oligosaccharide of YB2/0-produced IgG1 has a more critical role in enhancing ADCC than Gal-binding or bisecting GlcNAc-binding oligosaccharides according to the following evidence. First, monosaccharide composition and oligosaccharide profiling analysis showed that high content of nonfucosylated complex-type oligosaccharides were characteristic in YB2/0produced anti-hIL-5R IgG1 (34% , Table II) and anti-CD20 IgG1 (91% , Table IV) compared with low content of those in CHOproduced anti-hIL-5R IgG1 (9%, data not shown) and anti-CD20 IgG1 (6%, data not shown). Second, ADCC assay of the anti-hIL-5R IgG1 separated by LCA affinity chromatography demonstrated that Fuc content of IgG1 was inversely correlated with ADCC (Fig. 3, A and B). Third, quantitative PCR analysis indicated that YB2/0 cells had a 10-fold lower expression level of FUT8 mRNA than CHO cells (Fig. 6A). Fourth, overexpression of FUT8 in YB2/0 cells increased the content of fucosylated oligosaccharides and also decreased ADCC of anti-CD20 IgG1 (Fig. 6B).
L-Fuc residues in an ␣1,6-linkage to the GlcNAc of the reducing end ("core Fuc") are relatively common in mammalian N-linked oligosaccharide. FUT8, considered to be the only gene that codes ␣1,6-fucosyltransferase, catalyzes the transfer of Fuc from GDP-Fuc to GlcNAc of the reducing end. Therefore, we focused on FUT8 as a key gene controlling the low Fuc content of IgG1 produced by YB2/0 and indicate that the cells produce low Fuc content IgG1 simply due to the low expression level of FUT8. Since biosynthesis of N-linked oligosaccharides is controlled by a number of glycosyltransferases, their acceptors and substrates, etc., it remains to be determined whether there is possible involvement of the other factor in biosynthesis of nonfucosylated IgG1 in YB2/0 cells.
The importance of nonfucosylated oligosaccharide on ADCC has been reported very recently by Shields et al. (11). They have shown that nonfucosylated anti-Her2 humanized IgG1 and anti-IgE humanized IgG1 produced by a variant of CHO cells, Lec13, had enhanced ADCC relative to fucosylated IgG1s produced by normal CHO cells. However, they have only focused on Fuc, because no appreciable differences in the content of the other sugar residues have been found in Lec13-produced IgG1 and normal CHO-produced IgG1. Until now, the effects of Fuc, Gal, or bisecting GlcNAc on ADCC have been analyzed independently (4 -7, 9 -11); therefore, comparison of the effect of each sugar residue or the combined effect of each sugar residue has not yet been reported.
In this report, we could not find any correlation between the content of Gal and ADCC. A difference in the content of Gal between YB2/0-produced IgG1s and CHO-produced IgG1s was not correlated to ADCC (Table I). As a result of the separation of KM8399 using LCA lectin affinity chromatography, the compositions of G0, G1, and G2 of fraction II were very similar to that of KM8399 (Table II); nevertheless, ADCC of fraction II and KM8399 was quite different (Fig. 3B). Our results show a good coincidence with the report of Boyd et al. (4), in which obvious change was not found in ADCC after removal of the majority of the Gal residues of anti-CDw52 IgG1 produced by CHO cells. In contrast, Kumpel et al. (5,6) reported that highly galactosylated anti-D-antigen IgG1s have higher ADCC, although that effect was only 2-3-fold.
Two groups independently reported that increasing the level of bisecting GlcNAc of anti-neuroblastoma IgG1 and anti-CD20 IgG1 could enhance ADCC (9, 10). GnTIII-transfected CHO cells produced anti-CD20 IgG1 with a high content of bisecting GlcNAc (48 -71%), which showed a 10 -20-fold enhancement of ADCC compared with that with no content of bisecting GlcNAc (0%). In the present study, we carefully examined the effect of bisecting GlcNAc in ADCC in comparison with that of Fuc. We prepared anti-hIL-5R IgG1 with different content of bisecting GlcNAc (0 -30%). To avoid the effect of fucosylation, nonfucosylated IgG1s were depleted by LCA lectin affinity chromatography. To our surprise, we could not detect any correlation between ADCC and content of bisecting GlcNAc. One possible explanation of the discrepancy with the results of Umana et al. (9) and Davies et al. (10) might be that 30% content of bisecting GlcNAc is not enough to enhance ADCC. We next produced anti-CD20 IgG1 by LEC10 cells, a variant CHO cell that over-expressed GnTIII. The resultant IgG1 (74% bisecting GlcNAc and 100% Fuc) had shown only severalfold higher ADCC than normal CHO-produced IgG1 (0% bisecting GlcNAc and 94% Fuc); in contrast, YB2/0-produced IgG1 (19% bisecting GlcNAc and 91% non-Fuc) had 100-fold higher ADCC than LEC10produced IgG1 (Fig. 5B). These results suggest that an extremely high content of bisecting GlcNAc (74%) has a relatively weak effect for enhancing ADCC. More importantly, nonfucosylated oligosaccharide was shown to have a prominent effect in enhancement of ADCC of IgG1 compared with bisecting GlcNAc-containing oligosaccharide. We further evaluated the combined effect of bisecting GlcNAc with nonfucosylated oligosaccharides. YB2/0-produced anti-CD20 IgG1 was separated based on the content of bisecting GlcNAc-binding oligosaccharides (0, 8, 33, and 45%), which contained the same content of nonfucosylated oligosaccharides (around 90%). These four fractions did not show any significant difference in ADCC, indicating that the presence of bisecting GlcNAc-binding oligosaccharides, at least under 45%, does not have any additional effect in ADCC of highly nonfucosylated IgG1 (90%). To our knowledge, this is the first report that shows the effect of Fuc, Gal, and bisecting GlcNAc simultaneously and also shows the combined effect of Fuc and bisecting GlcNAc.
The ADCC have been believed to be a result of specific killing of antigen-positive cells by natural killer cells through binding of the IgG Fc domain to Fc␥RIIIa. Recently, Shields et al. (11) have revealed that binding of the Fuc-deficient IgG1 (produced by Lec13 cells) to Fc␥RIIIa was enhanced up to 50-fold. They have shown that improved binding to Fc␥RIIIa has translated into improved ADCC in vitro, using PBMC or natural killer cells. These results suggest that Fuc-deficient IgG1 may require a lower concentration of antibody on the surface of the target cell to activate an effector cell. There are a few possible explanations of why the antibodies with nonfucosylated oligosaccharides give rise to stronger binding to Fc␥RIIIa than those in which the glycoforms are absent. A core Fuc has been shown to influence the conformational flexibility of biantennary oligosaccharides (26,27). The oligosaccharides of IgG appear to be largely sequestered between the CH2 domains and may help to stabilize the CH2 domain (28). In the co-crystal structure of IgG1 Fc:Fc␥RIIIb, Fuc is orientated away from the interface and making no specific contacts with the receptor (29); nevertheless, Harris et al. (30,31) have supposed that Fuc could have influence on the binding by the receptor. We speculated that the absence of Fuc provided a more suitable conformation for the binding of IgG1 to Fc␥RIII than the presence of bisecting GlcNAc or Gal, although structural analyses of a series of IgG1 with or without Fuc, bisecting GlcNAc, or Gal are needed for further discussion.
Several recombinant monoclonal antibodies are being used as human therapeutics. Some of these are blocking monoclonal antibodies to receptors or soluble ligands and therefore may function without utilizing antibody effector functions. However, ADCC is still considered to be one of the most important anti-tumor mechanisms of clinically effective anti-Her2 hu-

TABLE IV
Oligosaccharide composition of PHA-E 4 -separated fractions ND, not detected. Fuc(Ϫ), total percentage of nonfucosylated oligosaccharides. Bis(ϩ), total percentage of bisecting GlcNAc-binding oligosaccharides. G0, G1, and G2, total percentage of nongalactosylated, monogalactosylated, and digalactosylated oligosaccharides, respectively. manized IgG1 and anti-CD20 chimeric IgG1 at least in animal models, since they are supposed to have multiple anti-tumor mechanisms (2). More importantly, Cartron et al. (32) have reported recently that therapeutic activity of anti-CD20 chimeric IgG1 in patients with non-Hodgkin's lymphoma has been correlated with polymorphism in the Fc␥RIIIa gene. They have shown that FCGR3A-158V patients showed a better response to anti-CD20 chimeric IgG1 because they have higher ADCC against lymphoma cells. These reports have suggested that therapeutic antibodies with enhanced ADCC including the nonfucosylated IgG1 would result in the improvement of clinical response. Moreover, these findings may allow for use of the nonfucosylated IgG1 at lower doses with no reduction in efficacy. Antibody therapeutics effective in lower doses might reduce the cost of antibody therapy.