Identification and characterization of three novel beta 1,3-N-acetylglucosaminyltransferases structurally related to the beta 1,3-galactosyltransferase family.

We have isolated three types of cDNAs encoding novel β1,3-N-acetylglucosaminyltransferases (designated β3Gn-T2, -T3, and -T4) from human gastric mucosa and the neuroblastoma cell line SK-N-MC. These enzymes are predicted to be type 2 transmembrane proteins of 397, 372, and 378 amino acids, respectively. They share motifs conserved among members of the β1,3-galactosyltransferase family and a β1,3-N-acetylglucosaminyltransferase (designated β3Gn-T1), but show no structural similarity to another type of β1,3-N-acetylglucosaminyltransferase (iGnT). Each of the enzymes expressed by insect cells as a secreted protein fused to the FLAG peptide showed β1,3-N-acetylglucosaminyltransferase activity for type 2 oligosaccharides but not β1,3-galactosyltransferase activity. These enzymes exhibited different substrate specificity. Transfection of Namalwa KJM-1 cells with β3Gn-T2, -T3, or -T4 cDNA led to an increase in poly-N-acetyllactosamines recognized by an anti-i-antigen antibody or specific lectins. The expression profiles of these β3Gn-Ts were different among 35 human tissues. β3Gn-T2 was ubiquitously expressed, whereas expression of β3Gn-T3 and -T4 was relatively restricted. β3Gn-T3 was expressed in colon, jejunum, stomach, esophagus, placenta, and trachea. β3Gn-T4 was mainly expressed in brain. These results have revealed that several β1,3-N-acetylglucosaminyltransferases form a family with structural similarity to the β1,3-galactosyltransferase family. Considering the differences in substrate specificity and distribution, each β1,3-N-acetylglucosaminyltransferase may play different roles.

A family of human ␤1,3-galactosyltransferases (␤3Gal-Ts) 1 consisting of five members (␤3Gal-T1, -T2, -T3, -T4, and -T5) was recently identified (1)(2)(3)(4). The first ␤1,3-galactosyltransferase (␤3Gal-T1), which catalyzes the formation of type 1 oligosaccharides, was isolated by us using an expression cloning approach (1). Expression patterns of ␤3Gal-T1 and type 1 oligosaccharides strongly suggested the existence of ␤3Gal-T1 homologs. For instance, type 1-derived oligosaccharides such as sialyl-Le a were known to be expressed in colon and pancreatic cancer cell lines, whereas expression of ␤3Gal-T1 was detected in brain, but not in cancer cells. Our early approach using Southern hybridization failed to detect the existence of ␤3Gal-T1 homologous genes. However, recent accumulation of nucleotide sequence information on human cDNAs and genes such as expressed sequence tags (ESTs) enabled us to search homologous genes that do not have high similarity as detected by hybridization, but show significant similarity. A homology search based on the nucleotide or amino acid sequence of ␤3Gal-T1 led to the isolation of ␤3Gal-T2, -T3, and -T4, indicating that ␤3Gal-Ts form a family (1)(2)(3).
␤3Gal-T2 catalyzed a similar reaction, but showed different substrate specificity compared with ␤3Gal-T1. The activity of ␤3Gal-T3 has not been detected, whereas the corresponding mouse enzyme exhibits weak ␤3Gal-T activity for both GlcNAc and GalNAc (5). On the other hand, ␤3Gal-T4 transfers galactose to the GalNAc residue of asialo-G M2 or G M2 to catalyze the formation of asialo-G M1 or G M1 , respectively (3). ␤3Gal-T4 may be a human homolog of rat G M1 and G D1 synthases (6) since these enzymes shows 79.4% identity at the amino acid level.
A PCR cloning approach using degenerate primers corresponding to conserved regions in the ␤3Gal-T family has enabled us to isolate a fifth member (␤3Gal-T5) of this family, which catalyzes the synthesis of type 1 oligosaccharides and is the most probable candidate involved in the biosynthesis of a cancer-associated sugar antigen, sialyl-Le a , in gastrointestinal and pancreatic cancer cells (4).
Very interestingly, a ␤1,3-N-acetylglucosaminyltransferase (designated ␤3Gn-T1) has been recently isolated based on the structural similarity to the ␤3Gal-Ts (7). ␤3Gn-T1 shows sig-* 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.
§ These authors contributed equally to this work and should be considered as first authors.
During the course of study to isolate ␤3Gal-T1 homologs, we have identified three additional types of putative members of the ␤3Gal-T family. In this study, we show additional examples that glycosyltransferases structurally related to the ␤3Gal-T family exhibit ␤1,3-N-acetylglucosaminyltransferase activity, but not ␤1,3-galactosyltransferase activity. These results indicate that ␤1,3-N-acetylglucosaminyltransferases (␤3Gn-Ts) form a family having structural similarity to the ␤3Gal-T family. Alignment of primary sequences of all members of the ␤3Gn-T and ␤3Gal-T families revealed that the members are clustered into four subgroups, probably reflecting enzymatic activity and substrate specificity. Transfection experiments and in vitro enzymatic analysis have demonstrated that ␤3Gn-T2, -T3, and -T4 are able to catalyze the initiation and elongation of poly-N-acetyllactosamine sugar chains; however, they exhibit different substrate specificity. These results, taken together with the different distributions of these enzymes, indicate that ␤3Gn-T2, -T3, and -T4 each exert distinct roles in physiological and pathological processes.
Cell Lines-Namalwa KJM-1, a subline of the human Burkitt lymphoma cell line Namalwa, was cultivated in serum-free RPMI 1640 medium as described (9,10). Cell lines SK-N-MC and Colo205 were obtained from the American Type Culture Collection. These cell lines were cultured in RPMI 1640 medium containing 10% fetal calf serum. Sf9 and Sf21 insect cells were cultured at 27°C in TNM-FH insect medium (Pharmingen) as described previously (11).
Preparation of cDNA Libraries and Single Strand cDNAs-cDNA libraries of human gastric mucosa and human placenta were constructed as described previously (12). Single strand cDNAs were synthesized from total RNA prepared from the neuroblastoma cell line SK-N-MC.
DNA Sequencing-DNA sequences were determined by the dideoxynucleotide chain termination method using an ABI PRISM TM 377 DNA sequencer (Applied Biosystems, Inc.).
Construction and Purification of ␤3Gn-T Proteins Fused to the FLAG Peptide-The putative catalytic domain of each ␤3Gn-T2, -T3, and -T4 was expressed as a secreted protein fused to the FLAG peptide in insect cells. A 1.1-kb DNA fragment encoding a COOH-terminal portion of ␤3Gn-T2 (amino acids 31-397) was amplified by PCR using primers 5Ј-CACGGATCCAGCCAAGAAAAAAATGGAAAAGGGGA-3Ј and 5Ј-A-TCCGATAGCGGCCGCTTAGCATTTTAAATGAGCACTCTGCAAC-3Ј, digested with BamHI and NotI, and inserted between the BamHI and NotI sites of pVL1393-F2 to yield pVL1393-F2G2. pVL1393-F2 is an expression vector derived from pVL1393 (Pharmingen) and contains a fragment encoding the signal peptide of human immunoglobulin (MHFQVQIFSFLLISASVIMSRG) and the FLAG peptide (DYKD-DDDK). Joining in-frame a cDNA fragment with a unique BamHI site of pVL1393-F2 just downstream of the COOH terminus of the FLAG peptide enables the cDNA product to be secreted as a protein fused to the FLAG peptide. A 1.0-kb DNA fragment encoding a COOH-terminal portion of ␤3Gn-T3 (amino acids 38 -372) was amplified by PCR using primers 5Ј-CGCGGATCCTCCCCACGGTCCGTGGACCAG-3Ј and 5Ј-A-TAGTTTAGCGGCCGCGGAAGGGCTCAGCAGCGTCG-3Ј, digested with BamHI and NotI, and inserted between the BamHI and NotI sites of pVL1393-F2 to yield pVL1393-F2G3. A 0.9-kb DNA fragment encoding a COOH-terminal portion of ␤3Gn-T4 (amino acids 56 -378) was amplified by PCR using primers 5Ј-ATAAGATCTGCAGGAGACCCCA-CGGCCCACC-3Ј and 5Ј-ATAGTTATGCGGCCGCCTCAGGCTGTTGC-  H78875  N58174  R77875  R77780  AA133340  AA133381  H47990  N66915  R75815  R75816  R18612  R41690  W26453  N51037  W25364  R31722  H93550  H13125  H47991  R74552  H80116  R82733  G23485 CCAACCCAC-3Ј, digested with BglII and NotI, inserted between the BamHI and NotI sites of pVL1393-F2 to yield pVL1393-F2G4. The PCR-amplified portions of pVL1393-F2G2, pVL1393-F2G3, and pVL1393-F2G4 were sequenced to confirm the absence of possible PCR errors. Sf9 insect cells were cotransfected with BaculoGold viral DNA (Pharmingen) according to the manufacturer's instruction and each of plasmids pVL1393-F2G2, pVL1393-F2G3, and pVL1393-F2G4 and were incubated for 3 days at 27°C to produce individual recombinant viruses. These viruses were amplified three times to reach titers of ϳ10 9 plaque-forming units/ml. Sf21 insect cells (4 ϫ 10 7 cells; Pharmingen) were infected at a multiplicity of 10 and incubated in 30 ml of TNM-FH insect medium at 27°C for 72 h to yield conditioned medium including recombinant ␤3Gn-T proteins fused to the FLAG peptide, which were readily purified by anti-FLAG M1 antibody resin (Sigma) according to the protocol of the manufacturer. Briefly, the culture medium (30 ml) was collected by centrifugation and added to NaCl (150 mM final concentration), NaN 3 (0.1% final concentration), and M1 antibody resin (30 l) to adsorb the recombinant ␤3Gn-T proteins on the resin. The resin was recovered by centrifugation and washed three times with buffer (1 ml) consisting of 50 mM Tris-HCl (pH 7.4), 150 mM NaCl, and 1 mM CaCl 2 . The recombinant ␤3Gn-T proteins were eluted with buffer (90 l) consisting of 50 mM Tris-HCl (pH 7.4), 150 mM NaCl, and 2 mM EDTA, followed by the addition of CaCl 2 (4 mM final concentration), and stored at 4°C until use. The amount of the purified proteins was not enough for accurate quantification.
Silver Staining and Western Blot Analysis-The enzymes purified above (3 l) were subjected to SDS-polyacrylamide gel electrophoresis, followed by silver staining or Western blot analysis. Silver staining was performed using a silver staining kit (Wako Bioproducts). Proteins separated on 6% SDS-polyacrylamide gel were transferred to an polyvinylidene difluoride membrane (Immobilon, Millipore Corp.) in a Trans-Blot SD cell (Bio-Rad). The membrane was blocked with phosphate-buffered saline containing 5% skim milk at 4°C overnight and then incubated with 10 g/ml M2 antibody (Sigma). The membrane was stained with ECL Western blot detection reagents (Amersham Pharmacia Biotech) according to the manufacturer's instructions.
To further confirm the structure of the reaction product derived from lactose, it was digested with endo-␤-galactosidase or modified by ␤1,4galactosyltransferase. To avoid the inhibitory effect of MOPS on endo-␤-galactosidase, an N-acetylglucosaminyltransferase assay was done without MOPS using lactose as an acceptor. The reaction was stopped by boiling, and the reaction mixture was recovered by centrifugation. The reaction mixture (15 l) was added to acetate buffer (50 mM final concentration; pH 5.8) and incubated with 250 milliunits/ml Escherichia freundii endo-␤-galactosidase (Seikagaku Kogyo) (14) in a total volume of 41 l at 37°C for 16 h, followed by analysis with HPAE/PAD as described above. Alternatively, the reaction mixture (15 l) was added to Tris-HCl (50 mM final concentration; pH 8.0) and incubated with 750 milliunits/ml bovine milk ␤1,4-galactosyltransferase (Sigma) in a total volume of 40 l at 37°C for 16 h, followed by analysis with HPAE/PAD as described above.
Endo-␤-galactosidase digestion of the product yielded two peaks that comigrated with the standard oligosaccharides GlcNAc␤1-3Gal␤ and glucose at a 1:1 molar ratio. These results clearly indicated that the product derived from lactose was GlcNAc␤1-3Gal␤1-4Glc.
Since the amount of the purified proteins was not enough for accurate quantification, enzymatic activity is defined as picomoles of acceptor substrate N-acetylglucosaminylated per ml of culture medium/h. The amounts of reaction products were determined from their absorbance intensities using individual standards.
Alternatively, the N-acetylglucosaminyltransferase activities of the purified proteins (15 l) were assayed in 200 mM MOPS (pH 7.5), 20 mM MgCl 2 , 20 mM UDP-GlcNAc, and 50 M pyridylaminated acceptors (total volume of 30 l). As acceptors, the following pyridylaminated oligosaccharides were used: . After incubation at 37°C for the appropriate times (2 h for ␤3Gn-T2 and 15 h for ␤3Gn-T3 and -T4), the reactions were terminated by boiling and analyzed by HPLC as described previously, with exception that HPLC was performed at 50°C with a flow rate of 0.5 ml/min (9,10). Parallel reactions were done in the absence of UDP-GlcNAc to identify products and to check hydrolysis of substrate and product. The oligosaccharides were purchased from Oxford Glycosystems and pyridylaminated according to the method of Kondo et al. (15). The amounts of products were determined from their fluorescence intensities using pyridylaminated lactose as a standard.
The reaction product derived from pyridylaminated LNnT was identified by comparison of the retention time on HPLC with that of the pyridylaminated standard oligosaccharide GlcNAc␤1-3Gal␤1-4GlcNAc␤1-3Gal␤1-4Glc, which was prepared from pyridylaminated p-LNnH by digestion with jack bean ␤-galactosidase. To further confirm the structure of the reaction product, the reaction product was modified by ␤1,4-galactosyltransferase to examine whether p-LNnH was produced or not. The reaction mixtures (20 l) was incubated with 20 milliunits of bovine milk ␤1,4-galactosyltransferase in the presence or absence of UDP-Gal (20 mM) in a total volume of 30 l at 37°C for 15 h according to manufacturer's recommendations. The product further modified by ␤1,4-galactosyltransferase comigrated with pyridylaminated p-LNnH on HPLC. These results indicated that the product was pyridylaminated GlcNAc␤1-3Gal␤1-4GlcNAc␤1-3Gal␤1-4Glc. The galactosyltransferase activities of the purified proteins (15 l) were assayed using pyridylaminated oligosaccharides (GlcNAc␤1-3Gal␤1-4Glc and LNnT) as substrates as described previously (4).
Preparation and Fractionation of Blood Leukocytes-Human polymorphonuclear leukocytes, monocyte-enriched population, and lymphocyte-enriched population were obtained as described previously (10).
Quantitative Analysis of the Three ␤3Gn-T Transcripts in Human Tumor Cell Lines and Human Tissues by Competitive RT-PCR-The levels of the ␤3Gn-T2, -T3, and -T4 transcripts were measured by competitive RT-PCR as described in detail previously (4,16). Competitor DNA plasmids carrying a small deletion within the respective cDNA were constructed by appropriate restriction endonuclease digestion as shown in Table II. For instance, a competitor DNA plasmid for measuring ␤3Gn-T2 transcripts was prepared by deleting the 227-bp Eco81I-PflMI fragment in ␤3Gn-T2 cDNA from the standard DNA plasmid pBS-␤3Gn T2.
Single strand cDNAs were synthesized with an oligo(dT) primer from 6 g of DNase I-treated total RNA from human tissues (colon, jejunum, stomach body, stomach antrum, and esophagus) and cell lines (HL-60 and Colo205) as described previously (4). Single strand cDNAs from human leukocytes were prepared as described (10). In addition, single strand cDNAs were synthesized with an oligo(dT) primer from 1 g of poly(A) ϩ RNAs from 35 human tissues (CLONTECH) using a Superscript TM pre-amplification system for first strand cDNA synthesis (Life Technologies, Inc.) according to manufacturer's instructions. After cDNA synthesis, the reaction mixture was diluted 50-fold with H 2 O and then stored at Ϫ80°C until use.
Competitive RT-PCR was performed with AmpliTaq Gold TM (PerkinElmer Life Sciences). The annealing temperatures and specific primers used are listed in Table II. The amount of each of the ␤3Gn-T transcripts was normalized by the amount of ␤-actin transcripts (4,16).
Determination of Chromosomal Localization-The chromosomal localizations of the ␤3Gn-T2, -T3, and -T4 genes were determined using 3Ј-EST mapping data (NCBI Protein Database). The chromosomal localization of the ␤3Gal-T1 gene was determined by PCR analysis using a series of genomic DNAs from hamster-human somatic hybrids (BIO-SMAP TM Somatic Cell Hybrid PCRable TM DNAs, BIOS Laboratories) and specific primers (5Ј-TTCAGCCACCTAACAGTTGCCAGG-3Ј and 5Ј-ATACCTTCTTCGTGGCTTGGTGGAG-3Ј). The predicted fragment of 495 bp was amplified only when genomic DNA from a hybrid containing human chromosome 2 (hybrid 852) was used, indicating that this gene is located on chromosome 2.

RESULTS
Identification and Isolation of ␤3Gn-T2, -T3, and -T4 -A homology search in the EST division of the GenBank TM /EBI Data Bank using the FrameSearch algorithm revealed the existence of six types of cDNAs encoding proteins with low but significant similarity to ␤3Gal-T1, three of which have been reported recently to be ␤3Gal-T2, -T3, and -T4 (2, 3). Based on the nucleotide sequence of the ESTs shown in Table I, we prepared specific probes and isolated three types of full-length cDNAs encoding novel proteins (designated ␤3Gn-T2, -T3, and -T4) of 397, 372, and 378 amino acid residues, respectively, with structural similarity to ␤3Gal-T1. ␤3Gn-T2 and -T3 cDNAs were obtained from human gastric mucosa, and ␤3Gn-T4 cDNA was from the neuroblastoma cell line SK-N-MC.
A Kyte-Doolittle hydropathy analysis (17) revealed that ␤3Gn-T2, -T3, and -T4 show type 2 transmembrane topology typical of most glycosyltransferases. ␤3Gn-T2 is predicted to consist of an N-terminal cytoplasmic domain of 9 residues, a transmembrane segment of 19 residues, and a stem region and catalytic domain of 369 residues. ␤3Gn-T3 is predicted to consist of an N-terminal cytoplasmic domain of 11 residues, a transmembrane segment of 21 residues, and a stem region and catalytic domain of 340 residues. The predicted coding region of ␤3Gn-T4 has two potential initiation codons, both of which are in agreement with Kozak's rule (18). Therefore, it is predicted that ␤3Gn-T4 is composed of two different N-terminal cytoplasmic domains of 29 and 4 residues, a transmembrane segment of 20 residues, and 329 residues containing the stem region and catalytic domain (Fig. 1a). Fig. 1A shows a multiple alignment of the amino acid sequences of ␤3Gn-T2, -T3, and -T4 as well as ␤3Gn-T1 and five members of the ␤3Gal-T family. ␤3Gn-T2, -T3, and -T4 show 19 -24, 22-26, and 22-25% identities, respectively, to the ␤3Gal-T family (␤3Gal-T1, -T2, -T3, -T5, and -T5), whereas they show 15, 18, and 15% identities to ␤3Gn-T1. ␤3Gn-T2, -T3, and -T4 show 40 -45% identity one another. The sequence similarities are limited to the putative catalytic regions. Several sequence motifs conserved in the ␤3Gal-T family are also shared by ␤3Gn-T2, -T3, and -T4 as well as ␤3Gn-T1. Twentyfive amino acid residues located separately in the putative catalytic regions are identical among all the proteins. Three cysteine residues conserved in all members of the ␤3Gal-T family are also maintained in ␤3Gn-T2, -T3, and -T4, whereas two of these are not conserved in ␤3Gn-T1 (Fig. 1A, white arrows), indicating that ␤3Gn-T1 is relatively distinct from other members, especially in the context of the three-dimensional structure. There are five potential N-linked glycosylation sites in ␤3Gn-T2, three in ␤3Gn-T3, and three in ␤3Gal-T4. One site in a highly conserved motif is maintained among all the proteins (Fig. 1A, black arrow). The phylogenetic tree of these proteins generated using the amino acid sequences of the putative catalytic domains demonstrates that ␤3Gn-T2, -T3, and -T4 form a subgroup, indicating that they have similar enzymatic activity (Fig. 1B).
Production of Secreted Recombinant Proteins Fused to the FLAG Peptide-To examine the enzymatic activities of ␤3Gn-T2, -T3, and -T4, we expressed the putative catalytic domain of each enzyme (amino acids 31-397 of ␤3Gn-T2, amino acids 38 -372 of ␤3Gn-T3, and amino acids 56 -378 of ␤3Gn-T4) as a secreted protein fused to the FLAG peptide in Sf21 insect cells. The FLAG-fused recombinant proteins were partially purified using anti-FLAG M1 antibody resin and analyzed by SDSpolyacrylamide gel electrophoresis, followed by silver staining ( Fig. 2A) or Western blotting using anti-FLAG monoclonal antibody (Fig. 2B). Two major bands with apparent molecular masses of 45.5 and 48 kDa, broad bands of 42-45 kDa, and two major bands of 37.6 and 40 kDa were observed specifically for ␤3Gn-T2, -T3, and -T4, respectively. The FLAG-fused recombinant proteins for ␤3Gn-T2, -T3, and -T4 have predicted molecular masses of 43,674, 39,507, and 37,608 Da for the respective polypeptides, indicating glycosylation of the recombinant proteins produced by insect cells.
␤3Gn-T2, -T3, and -T4 Are ␤1,3-N-Acetylglucosaminyltransferases-The glycosyltransferase activities of the partially purified FLAG-fused recombinant proteins were examined. When lactose was used as an acceptor, ␤3Gn-T2, -T3, and -T4 showed a significant amount of N-acetylglucosaminyltransferase activity, whereas no activity was detected in a sample prepared from the conditioned medium of insect cells infected with empty vector virus. The structure of the product was estimated to be GlcNAc␤1-3Gal␤1-4Glc by comparing the retention time on HPLC with that of the standard oligosaccharide (Fig. 3A). To further confirm the structure of the product, it was digested with endo-␤-galactosidase or modified by ␤1,4-galactosyltransferase. Digestion of the product by E. freundii endo-␤-galactosidase yielded two peaks comigrating with the standard oligosaccharides GlcNAc␤1-3Gal and glucose at a 1:1 molar ratio (Fig. 3, compare A and B). Modification of the product by bovine milk ␤1,4-galactosyltransferase yielded a peak comigrating with LNnT (Fig. 3, compare A and C). These results clearly indicated that the product was GlcNAc␤1-3Gal␤1-4Glc.
Substrate Specificity of ␤3Gn-T2, -T3, and -T4 -Analysis of the substrate specificity of ␤3Gn-T2, -T3, and -T4 revealed that these enzymes utilized common oligosaccharides as substrates, but the substrate preference was significantly different (Tables  III and IV). ␤3Gn-T2 and -T4 showed more preferential activity for LNnT than for LNT, which is consistent with the nature of
Since Den, LEA, and PWM are likely to recognize nonsialylated poly-N-acetyllactosamines more preferentially than sialylated ones, the transfected cells were treated with neuraminidase before staining. As shown in Fig. 4, expression of ␤3Gn-T2, -T3, or -T4 increased the levels of poly-Nacetyllactosamines recognized by Den, LEA, or PWM compared with the vector (pAMo) transfectant, consistent with in vitro enzymatic activity. In particular, expression of ␤3Gn-T3 or -T4 led to a remarkable increase in reactivity to Den, in contrast to the slight increase in the ␤3Gn-T2 transfectant. On the other hand, reactivity to LEA or PWM was increased in the ␤3Gn-T2 transfectant more clearly than in the other two transfectants. These results indicate that ␤3Gn-T2, -T3, and -T4 each are involved in the biosynthesis of poly-N-acetyllactosamine sugar chains in transfected cells.
Expression Levels of the ␤3Gn-T2, -T3, and -T4 Transcripts-The expression levels of the ␤3Gn-T2, -T3, and -T4 transcripts were examined by competitive RT-PCR. These genes were differentially expressed in human tissues and cells (Table V). ␤3Gn-T2 was ubiquitously expressed in the tissues and cells tested, but expression of ␤3Gn-T3 and -T4 was relatively restricted. ␤3Gn-T3 was expressed in colon, jejunum, stomach (body and antrum), esophagus, placenta, and trachea. ␤3Gn-T4 was mainly expressed in brain tissues such as whole brain, hippocampus, amygdala, cerebellum, and caudate nucleus, as well as in colon, esophagus, and kidney. ␤1,3-N-Acetylglucosaminyltransferase activities were detected in several tissues, cells, and sera, some of which were characterized using partially purified enzymes (19 -28). Poly-N-acetyllactosamines are known to serve as backbone oligosaccharides for presenting the sialyl-Le X and sialyl-Le a determinants, which function as selectin ligands in leukocytes and several cancer cells such as colon cancer cells (29 -40). The human promyelocytic leukemia cell line HL-60 and the human colon adenocarcinoma cell line Colo205 are known to express ␤1,3-N-acetylglucosaminyltransferase activities as well as poly-N-acetyllactosamines presenting the sialyl-Le a and sialyl-Le X determinants. Therefore, it was of significant interest to examine the expression levels of ␤3Gn-T2, -T3, and -T4 in leukocytes and cancer cells such as HL-60 and Colo205. ␤3Gn-T2, but not ␤3Gn-T3 and -T4, was significantly expressed in HL-60 and Namalwa KJM-1 cells (Table V) as well as in human peripheral polymorphonuclear cells and lymphocytes (data not shown). On the other hand, ␤3Gn-T2 and -T3, but not ␤3Gn-T4, were highly expressed in the colon cancer cell line Colo205 (Table V). DISCUSSION In this study, we identified three novel ␤1,3-N-acetylglucosaminyltransferases (␤3Gn-T2, -T3, and -T4) that show structural similarity to ␤3Gn-T1 as well as the ␤3Gal-T family, including five members (␤3Gal-T1, -T2, -T3, -T4, and T5), demonstrating the existence of a ␤3Gn-T family now consisting of four members (␤3Gn-T1, -T2, -T3, and -T4). The existence of the multiple enzymes showing similar activity is a common feature of glycosyltransferases, which was demonstrated for  FIG. 3. HPLC analysis of the reaction product generated from lactose by recombinant ␤3Gn-T2. A, the N-acetylglucosaminyltransferase activity of the purified ␤3Gn-T2 protein was assayed using lactose (Gal␤1-4Glc) as an acceptor. The reaction mixture was analyzed using high-pH anion-exchange chromatography with pulsed amperometric detection. The peaks for substrate lactose and the generated product are labeled S and P, respectively. Arrow 1 indicates the elution position of the standard oligosaccharide GlcNAc␤1-3Gal␤1-4Glc. Based on the elution position, the peak indicated by the asterisk seems to be GlcNAc, which may be a degradation product of UDP-GlcNAc. The peak with a retention time of 2-3 min may be glycerol, which appeared in the absence of UDP-GlcNAc. B, the reaction mixture described in A was analyzed after digestion with endo-␤-galactosidase. Arrows 2 and 3 indicate the elution positions of the standard oligosaccharides GlcNAc␤1-3Gal and glucose, respectively. C, the reaction mixture described in A was analyzed after galactosylation with ␤1,4-galactosyltransferase. Arrow 4 indicates the elution position of the standard oligosaccharide LNnT (Gal␤1-4GlcNAc␤1-3Gal␤1-4Glc). Based on the elution position, the peak indicated by the double asterisks seems to be Gal␤1-4GlcNAc, which would be a galactosylation product of GlcNAc indicated by the asterisk in A.
We constructed the secreted recombinant proteins for ␤3Gn-T2, -T3, and -T4 fused to the FLAG peptide. Western blot analysis using anti-FLAG antibody revealed that the secreted enzymes were successfully produced by insect cells and were readily recovered by anti-FLAG M1 antibody resin. The FLAGfused proteins adsorbed to the resin were eluted under mild conditions using buffer containing 2 mM EDTA. Since the eluted proteins showed activity comparative to that of the adsorbed proteins, it was confirmed that EDTA treatment did not damage the enzymes (data not shown). The molecular masses of the recovered proteins were equal to or larger than the predicted ones for their polypeptides, indicating some glycosylation and no significant degradation of the recovered proteins.
All of the recombinant proteins showed Gal ␤1,3-N-acetylgalactosaminyltransferase activity for common oligosaccharides, whereas their substrate preference was significantly different. Since the amount of the recombinant proteins used in this study was not enough for determination of the protein concentration, we could not precisely compare the relative activities of the enzymes. However, the relative activities of ␤3Gn-T2 for LNnT, lactose, Gal␤1-4GlcNAc, and p-LNnH seemed to be higher than those of other enzymes (Tables III  and IV). Considering the variety of acceptor substrates and the different reactivities of the transfected cells to anti-i-antigen antibody or PWM and LEA lectins, the higher activity of ␤3Gn-T2 for these oligosaccharides may reflect substrate specificity.
To date, ␤3Gn-T activities have been detected in several tissues, cells, and sera, some of which were characterized using partially purified enzymes (19 -28). Based on the substrate specificity, ␤3Gn-T1, but not ␤3Gn-T2, -T3, and -T4, may correspond to a ␤3Gn-T partially purified from calf serum. ␤3Gn-T2 and -T4 showed more preferential activity for LNnT than for LNT, which was similar to the nature of the calf serum enzyme as well as ␤3Gn-T1 and iGnT (7,8,26); however, ␤3Gn-T2 and -T4 were distinguished from the calf serum enzyme and ␤3Gn-T1 by the activities for lactose (Gal␤1-4Glc) and N-acetyllactosamine (Gal␤1-4GlcNAc) (Table IV). On the other hand, ␤3Gn-T3 is quite unique since it showed activity for LNT comparable to LNnT. It has been reported that human colon cancer tissues and the colon cancer cell line Colo205 contain cancerassociated glycosphingolipids with dimeric Le a antigens (Gal␤1-3(Fuc␣1-4)GlcNAc␤1-3Gal␤1-3(Fuc␣1-4)GlcNAc) (101). Considering the substrate specificity and expression in colon tissues and Colo205 cells, ␤3Gn-T3 is likely to be the most probable candidate involved in the biosynthesis of the backbone structure of dimeric Le a (Gal␤1-3GlcNAc␤1-3Gal␤1-3GlcNAc). On the other hand, it is difficult to ascribe the ␤3Gn-T activities detected in crude samples to the isolated ␤3Gn-Ts (␤3Gn-T1, -T2, -T3, and -T4 or iGnT) because of the following reasons: differences in experimental conditions such as substrates used, the possibility of the existence of multiple ␤3Gn-Ts in the crude samples, and the possibility of the existence of additional unidentified ␤3Gn-Ts.
Analysis of substrate specificity revealed that ␤3Gn-T2, -T3, and -T4 each could be involved in the initiation and elongation of poly-N-acetyllactosamine synthesis by itself, which was demonstrated by increased expression of poly-N-acetyllactosamines in the transfected cells. The different reactivities of the respective transfectants to the anti-i-antigen antibody or LEA and PWM lectins may reflect the preference of the antibody and lectins as well as substrate specificity of these enzymes. On the other hand, expression of two or more ␤3Gn-Ts in the same cell, which was clearly demonstrated for Colo205 cells, indicates that poly-N-acetyllactosamine sugar chains might be synthe-
In this study, we isolated three types of novel ␤3Gn-T genes, which enabled us to discriminate the respective enzymes at the molecular level. Considering the enzymatic activities in vitro and in vivo as well as the expression patterns of the ␤3Gn-Ts, the respective enzymes are likely to play different roles. The poly-N-acetyllactosamine or GlcNAc␤1-3Gal structure appears in glycolipids, keratan sulfate proteoglycans, and human milk oligosaccharides, in addition to N-and O-glycans of glycoproteins. Therefore, the existence of multiple ␤3Gn-Ts is not strange. It remains to be determined which ␤3Gn-T makes which types of sugar chains. Poly-N-acetyllactosamines are known to be synthesized at various positions by the concerted action of several glycosyltransferases required for the elongation or formation of specific sugar branches preferred by elongation enzymes. For example, poly-N-acetyllactosamines are preferentially formed in the specific branch in complex type N-glycans, which are formed by ␤1,6-N-acetylglucosaminyltransferase V (102). In addition, core 2 ␤1,6-N-acetylglucosaminyltransferases 1 and 2 and large I ␤1,6-N-acetylglucosaminyltransferases are branching enzymes critical for elongation with poly-N-acetyllactosamines (48,80,81,(103)(104)(105). Discovery of the multiple ␤3Gn-Ts in addition to other multiple glycosyltransferases involved in the biosynthesis of poly-N-acetyllactosamines (e.g. ␤1,4-galactosyltransferases, ␤3Gal-Ts, ␤1,6-Nacetylglucosaminyltransferase V, core 2 ␤1,6-N-acetylglucosaminyltransferases, and large I ␤1,6-N-acetylglucosaminyltransferases) indicates that regulation of poly-N-acetyllactosamine synthesis may be more complex than previously recognized. Definitive determination of the enzymatic activities and expression patterns of the ␤3Gn-Ts as well as experiments using knockout mice may provide insight into their functions in physiological and pathological processes.