Cloning of a Human UDP-galactose:2-Acetamido-2-deoxy-d-glucose 3β-Galactosyltransferase Catalyzing the Formation of Type 1 Chains*

Biochemical evidence suggests that the galactosyltransferase activity synthesizing type 1 carbohydrate chains is separate from the well characterized enzyme that is responsible for the synthesis of type 2 chains. This was recently confirmed by the cloning, from melanoma cells, of an enzyme capable of synthesizing type 1 chains, which was shown to have no homology to other galactosyltransferases. We report here the molecular cloning and functional expression of a second human β3-galactosyltransferase distinct from the melanoma enzyme. The new β3-galactosyltransferase has homology to the melanoma enzyme in the putative catalytic domain, but has longer cytoplasmic and stem regions and a carboxyl-terminal extension. Northern blots showed that the new gene is present primarily in brain and heart. When transfected into mammalian cells, this gene directs the synthesis of type 1 chains as determined by a monoclonal antibody specific for sialyl Lewisa. A soluble version of the cloned enzyme was expressed in insect cells and purified. The soluble enzyme readily catalyzes the transfer of galactose to GlcNAc to form Gal(β1–3)GlcNAc. It also has a minor but distinct transfer activity toward Gal, LacNAc, and lactose, but is inactive toward GalNAc.

Two types of carbohydrate chains are known to exist in the lacto-series of oligosaccharides, type 1 chains that contain the Gal(␤1-3)GlcNAc linkage and type 2 chains containing the topoisomer Gal(␤1-4)GlcNAc. Both types of carbohydrate structures are present in soluble oligosaccharides of human milk (1), are also found on glycoproteins (2) and glycolipids (3), and are important precursors of blood group antigens (4). The differences in function between type 1 and type 2 chains are not well understood. For example, during both embryogenesis (5) and carcinogenesis (6), the ratio of type 1 to type 2 chains produced by the cell changes. Furthermore, both type 1 and type 2 chains can be ligands for the selectin family of leukocyte extravasation receptors (7). The physiological significance of these observations is not yet known (8).
The biosynthesis of type 1 and type 2 structures is catalyzed by specific galactosyltransferases, which transfer galactose to GlcNAc terminating chains. The galactosyltransferase respon-sible for type 2 chain biosynthesis (␤4-Gal-T) 1 has been cloned and well characterized and was shown to be expressed in various tissues and cell types (reviewed in Refs. 9 and 10). This enzyme requires Mn 2ϩ for activity and is regulated by ␣-lactalbumin to change the kinetics of transfer to glucose, thus favoring the synthesis of lactose. Relatively little information is known about the type 1 elongating enzyme, UDP-galactose:2acetamido-2-deoxy-D-glucose 3␤-galactosyltransferase (␤3-Gal-T). This enzyme is clearly different from the ␤4-Gal-T and is expected to have a more restricted tissue distribution. The type 1 elongating enzyme is thought to be distinct from another ␤3-Gal-T activity detected in various sources and transferring to lactose or LacNAc (11)(12)(13), although this has not been molecularly established.
A ␤3-Gal-T enzyme catalyzing the synthesis of Gal(␤1-3)GlcNAc has been purified from pig trachea and shown to require Mn 2ϩ , not to be influenced by lactalbumin, and to have an acceptor specificity consistent with its role of being responsible for elongation of oligosaccharide chains on both mucins and glycolipids (14,15). Another ␤3-Gal-T enzyme capable of forming type 1 chains has been detected in colon carcinoma cell lines (16), as well as normal colonic mucosa (17). Moreover, DNA from COLO 205 cells when transfected into mammalian cells produced cell lines de novo synthesizing type 1 chains (18). No molecular information is available for any of these enzymes, and it is therefore difficult to judge their similarity. A ␤3-Gal-T was recently cloned from the human melanoma WM266 -4 cell line using an expression cloning strategy that relied on lectin resistance to identify clones (19). This enzyme, which has no homology to known glycosyltransferases, transfers galactose in vitro to produce Gal(␤1-3)GlcNAc(␤1-3)Gal(␤1-4)Glc and directs the synthesis of sLe a in transfected cells.
We report here the cloning and functional expression of a new ␤3-Gal-T from human brain distinct from the enzyme present in melanoma cells. The new enzyme is homologous to the melanoma cell enzyme in the putative catalytic domain, and is mainly expressed in brain and heart. When transfected in mammalian cells, the new gene directs the de novo synthesis of type 1 chains. A soluble version of the enzyme expressed in * 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.
The nucleotide sequence(s) reported in this paper has been submitted to the GenBank TM /EBI Data Bank with accession number(s) Y15014.

EXPERIMENTAL PROCEDURES
Materials-All cell culture media, sera, and antibiotics were from Life Technologies, Inc. CHO K1 cells (ATCC CCL-61) were maintained in ␣-minimal essential medium, 5% fetal bovine serum. Sf 9 insect cells were grown in Sf-900 II SFM medium. GlcNAc, Gal, Gal(␤1-3)GlcNAc, p-nitrophenyl ␤-D-lactopyranoside, GalNAc␣-benzyl, and UDP-Gal were purchased from Sigma. Antibody CSLEX was purchased from Becton Dickinson, and GSLA1 was a gift from Dr. J. Magnani (Gly-coTech Corp.). The cDNA for Fuc-T III in the expression vector pcDM7 was a gift from Dr. J. B. Lowe (Howard Hughes Medical Institute, Ann Arbor, MI).
EST Data Base Searches-The cDNA sequence of ␤3GalT2 was cloned by its homology to ␤3GalT1 (19). Using the complete 322-amino acid protein sequence of ␤3GalT1, a TBLASTN search was performed (20) on the dbest data base (release: Feb. 4,1996). The best aligning EST sequences (accession numbers R13867, H14861, and R13064) from infant brain showed sequence identities of about 67% over a stretch of about 68 amino acids to the putative catalytic domain of ␤3GalT1. Thirty-eight amino acids deduced from the cDNA sequence of R13867 were then used to search again the dbest data base (release: Feb. 29,1996) using TBLASTN. An additional human fetal EST clone D81474 was found, further extending the homology within the catalytic domain.
Screening of gt10 Library-The sequences of EST clones R13867 and D81474 were artificially combined, and primers were designed, allowing PCR amplification of a 307-bp DNA fragment specific for combined ESTs R13867 and D81474. These primers, designated galtdia2.pcr (ACT CGC CAG TGA TTG AAC ACA AAC) and galtdia3.pcr (TGA AGC CAG ATC TGC CTC CC) were then used to screen a collection of 16 heat-inactivated human cDNA libraries (QUICK-screen™, CLONTECH) by PCR for the presence of the 307-bp diagnostic fragment.
A viable form of the human brain gt10 cDNA library (CLONTECH, HL3002) was screened for the full-length cDNA of the putative new ␤3-Gal-T. Screening was done by preparing hierarchical phage pools and screening them by PCR using primers galtdia2.pcr and galtdia3.pcr, essentially as described previously by D'Esposito et al. (21). In brief, 9.5 ϫ 10 5 plaque-forming units from the human brain gt10 cDNA library were used to infect Escherichia coli C600 hfl cells, mixed with LB top agar, and plated onto 32 six-well tissue culture plates. After bacterial lysis, phage were eluted with SM buffer and extracted once with chloroform. The samples were briefly centrifuged and the clear lysates were subsequently stored at 4°C (top level pool). For each 96-well plate of the top level, column and row pools were prepared by mixing equal volumes of phage lysate from each well to obtain a total of 40 column and row phage pools. Aliquots (1 l each) of every well in each column and row were pooled, heat-treated (5 min at 95°C), and tested by PCR with primer pair galtdia2.pcr and galtdia3.pcr for the presence of the 307-bp DNA fragment. The intersections of each positive column and row pool were identified as a potentially positive wells possibly containing single independent clones and were individually re-tested by PCR. Two positive top level wells were selected and their phage titer determined. Level 1 pools were obtained by replating 13,000 plaque-forming units each from the selected top level wells onto 16 six-well plates and preparation of the corresponding phage lysates. Positive level 1 pools were identified by PCR using the strategy described above. To further reduce phage complexity, 450 plaque-forming units of several positive level 1 wells were subsequently plated onto four 24-well tissue culture plates each. Phage from positive level 2 wells were plated at a density so that single plaques could be isolated. Forty-eight single phage lysates were prepared and analyzed by PCR. Purity and insert size of each phage clone was determined by PCR using DNA specific oligonucleotide primers lgt10 -5Ј.seq (AGC AAG TTC AGC CTG GTT AAG T) and lgt10 -3Ј.rev (TTA TGA GTA TTT CTT CCA GGG). One of the 307-bp positive phage lysates, designated GA4/1, contained the largest insert (3 kb).
Subcloning and DNA Sequencing-After plaque purification, the insert of clone GA4/1 was subcloned into the single EcoRI site of expression vector pZEOSV (Invitrogen). Isolation of the insert of clone GA4/1 was done either by PCR amplification using the gt10-specific primers shown above or by excision from purified DNA using restric-tion endonuclease EcoRI. The DNA sequences of several clones of GA4/1 in pZEOSV (GA4/1.zeo) were determined. Both strands were sequenced using 21-mer and 22-mer oligonucleotide primers synthesized according to the sequence of the cDNA insert. The DNA sequences were assembled with the CAP program (22) and analyzed using Clone Manager (Scientific & Educational Software) and the sequence analysis software package of the University of Wisconsin Genetics Computer Group (23).
Northern Blot Analysis-A premade Northern blot of poly(A) ϩ RNA (multiple tissue Northern blot, CLONTECH) was prehybridized in 5 ϫ SSC, 5 ϫ Denhardt's, and 1% SDS solution at 65°C for 4 h and then hybridized overnight at 65°C with a 32 P-random prime-labeled (24) probe from the GA4/1 insert, containing the entire coding region of ␤3GalT2. Filters were washed at room temperature in 5 ϫ SSC, 0.2% SDS two times and then in 2 ϫ SSC, 0.2% SDS two times. Final wash was in 0.1 ϫ SSC, 0.2% SDS at room temperature.
Construction and Characterization of CHO/Fuc-T III Cell Line-To create a cell line capable of producing sLe a , CHO-K1 cells were transfected with the gene for Fuc-T III (25). Expression vector pcDNA(ϩ).hyg was constructed by inserting into the single NheI restriction endonuclease site present in vector pcDNAI/Amp (Invitrogen) a XbaI-restricted 1.68-kb DNA fragment encompassing the entire expression unit for hygromycin-B resistance. The fragment was obtained from vector pREP7 (Invitrogen) by PCR amplification using primers Hyg1.Xba (GCT CTA GAG CGT TTG CTG GCG GTG TCC) and Hyg2.Xba (GCT CTA GAC CAT GGG TCT GTC TGC TCA GTC CA). In pcDNA(ϩ).hyg both, the hygromycin-B resistance gene and inserted genes/cDNAs are transcribed in the same direction. The complete cDNA for human Fuc-T III was excised from pcDM7 using restriction endonuclease XhoI and ligated into pcDNA(ϩ).hyg cut with the same restriction endonuclease to create vector FTIII.hyg. Correct orientation of the Fuc-T III coding sequence was verified by restriction analysis.
CHO-K1 cells were seeded overnight in six-well plates and transfected with 4 g/well of FTIII.hyg, using LipofectAMINE according to the manufacturer's instructions. Cells were trypsinized into a T-175 flask, and 0.2 mg/ml hygromycin (Calbiochem) was added to the medium after 24 h. After 3 weeks of selection, surviving cells were sorted by FACS for the surface expression of the sLe x epitope by incubation with monoclonal antibody CSLEX-1 and staining with FITC-labeled polyclonal anti-mouse IgM (Jackson ImmunoResearch). The sorted cells were placed in 96-well plates at a density of 0.5 cells/well. Cells from wells containing single colonies were analyzed for surface expression of sLe x by FACS with CSLEX, and clones displaying the highest staining were used for further experiments.
Expression of Full-length ␤3GalT2 in CHO/Fuc-T III Cells and FACS Analysis-A mammalian expression vector was prepared containing the complete open reading frame of ␤3GalT2. Construction was done by PCR amplifying a DNA fragment from plasmid GA4/1.zeo using oligonucleotides NGalTATG.ECO (CGC GAA TTC GCC ACC ATG CTT CAG TGG AGG AGA AGA CAC TGC) and NGalTTAG.ECO (CGC GAA TTC CTA ATG TAG TTT ACG GTG GCG ATA CCT GCC). Oligonucleotide NGalTATG.ECO had a 5Ј extension containing an EcoRI restriction endonuclease site and consensus sequence for efficient translation (26). Oligonucleotide NGalTTAG.ECO included the putative stop codon of the ␤3GalT2 and an 5Ј extension containing an EcoRI restriction endonuclease site. The gel-purified, amplified 1.29-kb PCR fragment was directly ligated into plasmid pCR3™-uni (Invitrogen). Correct orientation of the DNA insert present in plasmid pGA4/1CDS.uni was verified by restriction analysis.
CHO/Fuc-T III cells seeded overnight in a six-well plate were transfected with 4 g of plasmid pGA4/1CDS.uni using LipofectAMINE. Cells were trypsinized in a T-175 flask, and after 24 h 0.5 mg/ml G418 was added to the culture medium. After 3 weeks of selection, cells were analyzed by FACS for the expression of sLe a on their surface. FACS analysis was performed by first incubating the cells with the monoclonal antibody GSLA1 and after washing with a FITC-labeled polyclonal anti-mouse IgG (Jackson ImmunoResearch).
Construction of Soluble ␤3-Gal-T2 Fused with Protein A-A basic mammalian expression vector for the expression of Staphylococcus aureus protein A fusions was prepared as follows. First an expression vector was constructed containing 20 amino acids of the human ␥-interferon signal sequence and the first 6 amino acids of mature human ␥-interferon (27)  Klenow fragment of DNA polymerase I, and the HindIII/SpeI-restricted DNA fragment was subcloned into vector pcDNAI/neo (Invitrogen) restricted with HindIII and XbaI, resulting in plasmid IFNG-new.neoI. A DNA fragment encoding for the Ig-binding domains of protein A was amplified from plasmid pRIT2T (Pharmacia Biotech Inc.) by PCR using oligonucleotides SPANEW1.SAL (GGT ACG GTC GAC TGG GAT CAA CGC AAT GGT TTT ATC) and SPANEW3.XHO (GGT GCA CTC GAG ATT TGT TAT CTG CAG ATC GAC). This DNA fragment was then cut with restriction endonucleases SalI and XhoI and subcloned into plasmid IFNG-new.neoI restricted with the same enzymes. At the carboxylterminal end of protein A were added XhoI and AgeI cloning sites for in-frame insertion of cDNAs. The resulting plasmid, designated sPROTA2.neoI, is capable of directing the expression of secreted protein A or protein A fusion proteins under control of the human cytomegalovirus promoter (data not shown).
A DNA fragment encoding for amino acids 125-422 (Fig. 1) of ␤3GalT2 was inserted into the XhoI and AgeI sites of vector sPROTA2.neoI. Construction was done by PCR amplification of the corresponding DNA fragment from GA4/1.zeo using oligonucleotides NGALTS1.XHO (GGT GCA CTC GAG AAA GGT ACT GGA CAT CCA AAT TCT TAC) and NGALTEND.AGE (GGT GCA ACC GGT TAC TAA TGT AGT TTA CGG TGG CGA TAC C). The amplified 0.92-kb PCR fragments was restricted with AgeI and XhoI and subcloned into sPROTA2.neoI, cut with the same enzymes. The presence of the ␤3GalT2 insert in the resulting protein A fusion vector SPA2GATS.neoI was verified by sequencing. The complete expression cassette for the protein A fusion of ␤3GalT2 of vector SPA2GATS.neoI was then transferred into an insect cell expression system to be able to purify larger quantities of the fusion protein for functional studies. This was achieved by transferring a 1.8-kb XbaI fragment from SPA2GATS.neoI into the unique XbaI site of donor plasmid pFastBac1 (Life Technologies, Inc.). Correct orientation of the DNA insert present in the resulting donor plasmid SPA2GATS.bac1 was verified by restriction analysis.
Expression and Purification of Protein A Fusion of ␤3GalT2 in Insect Cells-Donor plasmid SPA2GATS.bac1 was used to create in E. coli recombinant SPA2GATS bacmid DNA containing all the genetic elements for the production of recombinant virus particles, by utilizing Tn7-mediated transposition according to protocols given in the instruc- For large scale production of the ␤3GalT2/protein A fusion, Sf 9 cells grown at 28°C in flasks to a density of 2 ϫ 10 6 were infected with virus at a multiplicity of infection of 5, and culture supernatants were harvested 72 h post-infection. After centrifugation, filtration and concentration, supernatants were made 25 mM in sodium cacodylate, pH 6.5, and 100 mM in NaCl and loaded onto a SP-Sepharose column (Pharmacia). The column was washed with 25 mM sodium cacodylate, pH 6.5, 200 mM NaCl, and the ␤3GalT2/protein A fusion protein was eluted with 25 mM sodium cacodylate, pH 6.5, 500 mM NaCl. The eluted fractions were dialyzed against 40 mM sodium cacodylate, pH 6.5, 100 mM NaCl, 20 mM MnCl 2 and loaded onto a UDP-hexanolamine-Sepharose column, equilibrated in the same buffer. The column was washed with loading buffer, followed by washing with loading buffer without MnCl 2 . Elution of the enzyme was accomplished with buffer containing 50 mM sodium cacodylate, pH 6.5, 100 mM NaCl, 10 mM EDTA, 1 mg/ml UDP (Fluka). The enzyme was stored at 4°C.
Expression and Purification of Protein A Fusion of ␤3GalT1-A protein A fusion chimera of ␤3GalT1 was cloned and expressed in a manner analogous to that indicated above for ␤3GalT2. A portion of ␤3GalT1 representing the complete stem region and catalytic domains (amino acids 35-326) was amplified by PCR from the genomic DNA of Colo 205 cells (ATCC CCL-222), and recombinant virus were produced in a manner analogous to that described for ␤3GalT2. Cell culture and protein purification was performed as indicated above for ␤3GalT2.
Detection of ␤3GalT2/Protein A Fusion Protein by Enzyme-linked Immunosorbent Assay-The protein A portion of the ␤3GalT2/protein A fusion protein was used to semi-quantitatively determine the concentration of the soluble form of ␤3GalT2. Microtiter plates were coated overnight at 4°C with 120 l of human IgG (5 g/ml; Sigma) in PBS and blocked with 0.5% BSA in PBS for 60 min at room temperature. Samples as well as a protein A standard (0.5-50 ng/ml recombinant IgGbinding fragment of protein A; Sigma) diluted in 100 l of PBS containing 0.5% BSA (PBS/BSA) were added to the microtiter plates and incubated for 60 min at room temperature. Wells were washed with PBS containing 0.05% Tween 20 and incubated successively with 100 l of biotinylated goat anti-protein A antibody (1:100.000; Sigma) and 100 l of streptavidin-peroxidase conjugate (1:5000; Boehringer) in PBS/ BSA, for 60 min at room temperature. Wells were washed six times with PBS plus 0.05% Tween 20 and developed with TMB substrate solution (Bio-Rad), and absorbance at 450 nm was measured after stopping with 50 l of 1 M H 2 SO 4 .
␤3GalT2 Assays and Product Characterization-The linkage synthesized by the ␤3GalT2/protein A fusion protein was analyzed by HPAE/ PAD as follows: to 28 l of assay stock solution (180 mM sodium cacodylate, pH 6.5, 1 mg/ml BSA, and 0.26 mM UDP-Gal), 2 l of MnCl 2 (500 mM), 1 l of GlcNAc (500 mM), 14 l of H 2 O, and 5 l of enzyme were added. After incubation at 37°C for 2 h the reaction was stopped by freezing. An aliquot of 25 l of the reaction mix was analyzed by HPAE/PAD (Dionex) using the following conditions: 70% H 2 O, 30% 0.5 M NaOH, at a flow rate of 1 ml/min.
Enzymatic activity of the ␤3GalT2/protein A fusion protein was determined using a radioactive assay similar to the method of Palcic et al. (28) as follows; the appropriate amounts of enzyme were incubated with 14 l of assay stock solution (180 mM sodium cacodylate, pH 6.5, 1 mg/ml BSA, 0.26 mM UDP-Gal, 2 l of [ 3 H]UDP-Gal (Amersham Corp.)), 1 l of MnCl 2 (500 mM), 2.5 l of GlcNAc-Lemieux (37 mg/ml in dimethyl sulfoxide) at 37°C for 60 min. The reaction was quenched by the addition of 1 ml of water and loaded on a C18 Sep-Pack cartridge (Waters), and the column was washed twice with 5 ml of H 2 O and eluted with 5 ml of methanol. All fractions were counted in a BETAmatic ␤ counter (Kontron) after the addition of 10 ml of scintillation fluid. Enzymatic activity was determined by calculating percent UDP-Gal conversion ( 3 H activity in the methanol fraction versus total 3 H activity of all fractions).

RESULTS AND DISCUSSION
Cloning and Nucleotide Sequencing of the ␤3GalT2 cDNA-It is difficult to predict the sequence homology of glycosyltransferases based on their enzymatic activity. Some glycosyltransferases are grouped into families of homologous genes, as for example the ␣1-3 fucosyltransferases (29), or have characteristic motifs, as is the case for sialyltransferases (30,31). Many glycosyltransferases, however, have little or no homology even between enzymes that utilize the same activated sugar donor and carbohydrate acceptor. This is the case for galactosyltransferases, which seem to have no common motif except possibly for a hexapeptide (B-D-K-K-N-A, where A is either E or D and B is either R or K) identified by Joziasse et al. (32). TBLASTN (20) data base searches using all possible permutations of this peptide motif revealed no new homologous sequences.
The cloning of ␤3GalT1 by an elegant expression cloning method using lectin resistance for phenotype selection (19) provided the opportunity to search for novel galactosyltransferase genes by sequence similarity. Searches performed on the dbest data base using the TBLASTN algorithm revealed several ESTs with homology to portions of the ␤3GalT1 sequence. From a total of four ESTs identified (all of which were from human brain), a continuous cDNA fragment encoding for about 102 amino acids of a putative new galactosyltransferase gene was assembled. This cDNA fragment allowed the design of primers, which were used to provide a diagnostic PCR signal for the presence of the new putative gene. From the strength of the PCR signal, the new gene was found to exist mainly in libraries from human brain, heart, and cells of the immune system (data not shown). The human brain library gave the strongest PCR signal and was subsequently used as a starting point for the identification of the new gene. To clone the new gene, the method of D'Esposito et al. (21) was used, which relies on successive levels of splits, tests, and expansions of phage pools to obtain successively enriched fractions of the target gene. Starting with approximately 1 million clones, the first split to 192 wells (level 0) provided 17 wells with a positive PCR signal. This translates into an abundance of the cDNA of at least 1 in 84,000 calculated on the basis of 950,000 independent phage clones used for the primary infection. Each of the level 0 wells potentially contained a single independent clone from the putative gene. Because the diagnostic PCR amplified signal was from an internal portion of the putative new gene, no information was obtained regarding the length of the various level 0 clones and it was therefore not possible to judge which of the positive wells contained the complete gene. Two level 0 wells were chosen for subsequent expansion solely on the strength and clarity of the PCR signal and were split into 96 separate wells. After testing and splitting again into 96 separate wells, finally single plaques were evaluated for insert size and the largest gt10 clone GA4/1 was determined to contain an insert of ϳ3 kb.
DNA sequence analysis of gt10 clone GA4/1 revealed that it contained a single, long open reading frame encoding for a protein of 422 amino acids with a predicted M r of 49,202 (Fig.  1). Hydropathy analysis (Fig. 2) (33), as well as primary amino acid sequence analysis using the SAPS program (34), revealed a hydrophobic, putative transmembrane region at the aminoterminal end of the open reading frame (amino acids 27-43), predicting that this protein has a type II transmembrane topology, typical for all mammalian glycosyltransferases cloned to date. BLAST searches of public data bases using either the deduced amino acids sequence of ␤3GalT2 or the complete cDNA sequence of clone GA4/1 revealed no significant homology with any protein in the Swissprot (release: June 23, 1997) or with any gene in the GenEMBL (release: June 26, 1997) data base. Furthermore, besides the four human brain EST sequences, which were originally used to design the diagnostic PCR primer pair for the cloning of ␤3GalT2, no additional matching entries were found in the EST data base (release: June 23, 1997).
Comparison of the new protein sequence with the 326-amino acid protein sequence of ␤3GalT1 revealed an overall sequence identity of 46%. The highest level of sequence similarity between the two ␤3-Gal-T sequences was found between positions 47 and 326 for ␤3GalT1 and between positions 119 and 405 for ␤3GalT2, respectively. The sequence identity in this region was calculated as 51% (67% similarity), with four 1-3-amino acid insertions found in ␤3GalT2 as compared with ␤3GalT1 ( Fig.   FIG. 3. Sequence alignment of  ␤3GalT1 and ␤3GalT2. A, schematic representations of deduced amino acid sequences of ␤3GalT1 (19) and ␤3GalT2. The number of amino acids contained in each putative region are given above the schematic illustrations; putative N-glycosylation sites are indicated by a Y. The longer cytosolic region, stem region, and carboxyl-terminal extension of ␤3GalT2 are evident. B, amino acid sequence alignment of ␤3GalT1 and ␤3GalT2. The amino acids indicated in bold represent the putative transmembrane domain of the corresponding enzyme. 3B). Due to the high sequence conservation, we assume that these regions most likely represent the catalytic domains of the two enzymes. Both enzymes contain two conserved potential N-glycosylation sites (N-X-(S/T)) in the putative catalytic domains, with three additional sites present in the stem and putative catalytic domain of ␤3GalT2 (Fig. 3). The major differences between ␤3GalT1 and ␤3GalT2 are a 17-amino acid extension at the carboxyl terminus of ␤3GalT2 and differences in both the lengths and primary sequences of the putative cytoplasmic, transmembrane, and stem regions (Fig. 3, A and  B).
The high sequence homology within the catalytic domain between the two genes suggested that the new enzyme was also a ␤3-Gal-T. It remained, however, unclear whether the new enzyme transferred galactose to GlcNAc, as would be expected for a type 1 chain extending enzyme, or to other acceptors. Glycosyltransferase stem regions are known to influence enzyme acceptor specificity as has been reported for Fuc-T III and Fuc-T V (35,36), and cytoplasmic domains are known to influence Golgi localization (37,38). A distinct possibility therefore existed that ␤3GalT2 would transfer galactose to GalNAc to make the core 1 structure, or to other acceptors. Such an enzyme would be expected to have both different acceptor specificity and Golgi localization compared with ␤3GalT1. To test this possibility, we examined the acceptor specificity of ␤3GalT2 both in cell culture and in enzymatic assays.
Determination of Enzymatic Activity in Cell Culture-The ␤3GalT2 gene was subcloned in an expression vector and checked for its ability to direct the synthesis of type 1 chains in CHO cells, which do not normally synthesize type 1 chains (39). To readily detect de novo production of type 1 chains, CHO cells were transfected with the gene for Fuc-T III. This enzyme fucosylates both type 1 and type 2 chains so that cells expressing it produce the corresponding fucosylated and also sialylated oligosaccharides (25,39). CHO/Fuc-T III cells stained brightly with the anti-sLe x antibody CSLEX (data not shown) but not with the anti-sLe a antibody GSLA1 (Fig. 4, curve A). Transfection of CHO/Fuc-T III cells with a vector containing the newly cloned putative galactosyltransferase produced significant staining of these cells with GSLA1 (Fig. 4, curve B), indicating that the new gene is indeed a type 1 extension enzyme.
Production of Soluble ␤3GalT2/Protein A Fusion Protein and Enzymatic Assays-The acceptor specificity of ␤3GalT2 was also directly established in enzymatic assays. To purify sufficient amounts of enzyme for analysis, we constructed and expressed a soluble ␤3GalT2/protein A fusion protein by removing the putative cytoplasmic, transmembrane, and part of the stem region and replacing them with the IgG binding domain of S. aureus protein A. The ␤3GalT2/protein A fusion protein was expressed in Sf9 cells and purified by ion exchange and affinity chromatography. Enzymatic assays with GlcNAc as the acceptor produced a new peak upon HPAE/PAD analysis, which co-eluted with Gal(␤1-3)GlcNAc (Fig. 5). Activity assays were also performed using radiolabeled UDP-Gal and measuring the transfer of radioactivity to various sugar acceptors. Using this method, transfer of galactose to GlcNAc-Lemieux was readily observed. Using the radioactive assay with GlcNAc-Lemieux as the acceptor, the final purified ␤3GalT2/ protein A fusion protein had a specific activity of about 20 units/mg. This specific activity is typical for several other glycosyltransferase fusion proteins using simple oligosaccharides as acceptors. 2 Surprisingly, ␤3GalT2 transfers galactose to Gal-terminating acceptors. The results in Table I show Gal to be a relatively good acceptor, with LacNac and lactose being distinctly worse. The relative efficiency of the substrates in Table I,   attached on their anomeric sites are different. No transfer was observed to GalNAc␣-benzyl even when a 10-fold increase of enzyme was used (data not shown). As shown in Table I, ␤3GalT1 also transfers to Gal, but at a much lower relative rate and with more restricted specificity than ␤3GalT2. The reactivity pattern of ␤3GalT2 is partially reminiscent of the human kidney ␤3-galactosyltransferase (11). Although the assays for the kidney enzyme were performed at different conditions (pH 5, 4 mM Cd 2ϩ ), this enzyme was reported to accept a series of Gal terminating acceptors and to be influenced by the presence of hydrophobic groups at the anomeric site. Unfortunately, no assays with GlcNAc terminating acceptors were reported with the kidney enzyme leaving open the question whether it can synthesize type 1 chains. It is thus not clear if ␤3GalT2 and the kidney ␤3-galactosyltransferase are similar enzymes, although Northern blots show no significant signal for ␤3GalT2 in kidney (Fig. 6). The acceptor specificity of ␤3GalT2 is distinct from those reported for the snail and marsupial ␤3-galactosyltransferases (12, 13), and we expect that these enzymes are also molecularly different. Carbohydrate structures terminating in Gal(␤1-3)Gal␤-have been reported in both glycoproteins (40,41) and glycolipids (42). The in vitro specificity of ␤3GalT2 suggests that it is capable of synthesizing such structures.
Northern Blot Analysis-The complete open reading frame of ␤3GalT2 was used to probe polyadenylated mRNA from eight different human tissues. Two transcripts were identified by this procedure, with a strongly hybridizing transcript at 3.5 kb and a weaker one of 2.8 kb in size (Fig. 6). Both transcripts were detected only in heart and brain; no transcripts were detectable in placenta, lung, liver, skeletal muscle, kidney and pancreas. Thus, expression of this new ␤1-3or ␤3-galactosyltransferase is probably regulated in a tissue-specific and cellspecific manner. This is in contrast to the ␤4-Gal-T enzyme, which is transcribed in most tissues (reviewed in Refs. 29 and 43). No information is yet available about the distribution of ␤3GalT1. Preliminary PCR experiments using libraries from different tissues and cell types showed clear differences in the distribution of ␤3-Gal-T1 and ␤3GalT2. 3 The ␤3GalT2 described here represents the second type 1 chain extending enzyme cloned. The homology of the two genes in the catalytic domain suggests that they correspond to an evolutionary related family of ␤3-gal-T genes, but the wide divergence both in size and sequence in the cytosolic and stem regions implies different acceptors for the two enzymes and possibly different Golgi compartmentalization. It is plausible that one of the genes is specific for mucin and glycolipid acceptors and the other for N-linked glycoproteins. The restricted tissue specificity of ␤3GalT2 seems to indicate a specific role for this enzyme. At this time, it is not known whether the ␤3-Gal-T family is restricted to these two members or if more genes exist, as is the case for the ␣3-fucosyltransferases (29). The presence of additional enzymes would imply a diversity of type 1 bearing structures with perhaps varying roles. The availability of the ␤3GalT2 sequence will allow addressing these questions by molecular means and possibly uncovering the physiological functions of type 1 chains.  Acceptor substrates were assayed at the indicated concentrations using the radioactivity transfer assay as described under "Experimental Procedures." The same amounts of enzyme were used in all assays. The concentration of acceptor substrate in the assay mix were as indicated. Results are presented as percent conversion with transfer to GlcNAc having been normalized to 100%. ND indicates no detectable transfer under the assay conditions. Substrate ␤3GalT1 ␤3GalT2 GlcNAc␤-Lemieux (8 mM) 100 100 Gal␤-Lemieux (2 mM) 2 1 6 LacNAc␤-Lemieux (5 mM) N D 1 Lactose␤-p-nitrophenyl a (9 mM) N D 3 GalNAc␣-benzyl b (7 mM) N D N D a p-Nitrophenyl ␤-D-lactopyranoside. b Benzyl 2-acetamido-2-deoxy-␣-D-galactopyranoside.