Chondroitin Sulfate Synthase-3

Recently, it has become evident that chondroitin sulfate (CS) glycosyltransferases, which transfer glucuronic acid and/or N-acetylgalactosamine residues from each UDP-sugar to the nonreducing terminus of the CS chain, form a gene family. We report here a novel human gene (GenBank™ accession number AB086062) that possesses a sequence homologous with the human chondroitin sulfate synthase-1 (CSS1) gene, formerly known as chondroitin synthase. The full-length open reading frame consists of 882 amino acids and encodes a typical type II membrane protein. This enzyme contains a β3-glycosyltransferase motif and a β4-glycosyltransferase motif similar to that found in CSS1. Both the enzymes were expressed in COS-7 cells as soluble proteins, and their enzymatic natures were characterized. Both glucuronyltransferase and N-acetylgalactosaminyltransferase activities were observed when chondroitin, CS polymer, and their corresponding oligosaccharides were used as the acceptor substrates, but no polymerization reaction was observed as in the case of CSS1. The new enzyme was thus designated chondroitin sulfate synthase-3 (CSS3). However, the specific activity of CSS3 was much lower than that of CSS1. The reaction products were shown to have a GlcUAβ1–3GalNAc linkage and a GalNAcβ1–4GlcUA linkage in the nonreducing terminus of chondroitin resulting from glucuronyltransferase activity and N-acetylgalactosaminyltransferase activity, respectively. Quantitative real time PCR analysis revealed that the transcript level of CSS3 was much lower than that of CSS1, although it was ubiquitously expressed in various human tissues. These results indicate that CSS3 is a glycosyltransferase having both glucuronyltransferase and N-acetylgalactosaminyltransferase activities. It may make a contribution to CS biosynthesis that differs from that of CSS1.

Recently, it has become evident that chondroitin sulfate (CS) glycosyltransferases, which transfer glucuronic acid and/or N-acetylgalactosamine residues from each UDP-sugar to the nonreducing terminus of the CS chain, form a gene family. We report here a novel human gene (GenBank TM accession number AB086062) that possesses a sequence homologous with the human chondroitin sulfate synthase-1 (CSS1) gene, formerly known as chondroitin synthase. The full-length open reading frame consists of 882 amino acids and encodes a typical type II membrane protein. This enzyme contains a ␤3-glycosyltransferase motif and a ␤4-glycosyltransferase motif similar to that found in CSS1. Both the enzymes were expressed in COS-7 cells as soluble proteins, and their enzymatic natures were characterized. Both glucuronyltransferase and N-acetylgalactosaminyltransferase activities were observed when chondroitin, CS polymer, and their corresponding oligosaccharides were used as the acceptor substrates, but no polymerization reaction was observed as in the case of CSS1. The new enzyme was thus designated chondroitin sulfate synthase-3 (CSS3). However, the specific activity of CSS3 was much lower than that of CSS1. The reaction products were shown to have a GlcUA␤1-3GalNAc linkage and a GalNAc␤1-4GlcUA linkage in the nonreducing terminus of chondroitin resulting from glucuronyltransferase activity and Nacetylgalactosaminyltransferase activity, respectively. Quantitative real time PCR analysis revealed that the transcript level of CSS3 was much lower than that of CSS1, although it was ubiquitously expressed in various human tissues. These results indicate that CSS3 is a glycosyltransferase having both glucuronyltransferase and N-acetylgalactosaminyltransferase activities. It may make a contribution to CS biosynthesis that differs from that of CSS1.
Chondroitin sulfate (CS) 1 proteoglycans are located in the extracellular matrix and on cell surfaces in various kinds of human tissues. Some CS proteoglycans provide high osmotic pressure and water retention, and others may modulate cell adhesion to the extracellular matrix, proliferation, and morphogenesis (1,2). The biosynthetic assembly of chondroitin sulfate proteoglycans is characterized by the following sequential processes: (i) synthesis of the core protein; (ii) xylosylation of specific Ser moieties of the core protein; (iii) addition of two Gal residues to the Xyl; (iv) completion of common tetrasaccharide linkage region by the addition of a glucuronic acid (GlcUA) residue; (v) addition of an GalNAc residue to initiate the chondroitin/dermatan sulfate biosynthesis; (vi) repeated addition of GlcUA residues alternating with GalNAc residues to grow the large heteropolymer glycosaminoglycan chains; and (vii) modification of these growing glycosaminoglycan chains by variable O-sulfation and by variable epimerization of GlcUA to IdoUA.
The assembly of the linkage region on the core protein followed by glycosaminoglycan polymerization and modification occurs in the intracellular membrane system, which is composed of the endoplasmic reticulum and Golgi apparatus (3,4). With the exception of the polysaccharide chain-initiating Xyl transferase, which is found in the endoplasmic reticulum (5), all of the enzymes are firmly attached to the Golgi membranes and may work in an orchestrated manner. Some of these biosynthetic enzymes are found in serum or in the culture medium of the cells (4,6). The enzymes responsible for the synthesis of the linkage regions in proteoglycans, Xyl transferase (6), Gal transferase I (7,8), Gal transferase II (9), as well as GlcUA transferase I (10,11), which act sequentially to transfer Xyl, Gal, Gal, and GlcUA from their respective sugar nucleotide precursors to the acceptor core protein, have been cloned. We have been interested in characterizing the modification reactions, especially sulfations, because specific regional structures created by these modifications allow chondroitin sulfate to interact with other molecules, including cytokines. The modifications also allow regulation of the assembly and activities of other proteins in extracellular and pericellular matrices (12-* This work was performed as part of the R & D Project of the Industrial Science and Technology Frontier Program (R & D for Establishment and Utilization of a Technical Infrastructure for Japanese Industry) supported by the New Energy and Industrial Technology Development Organization. 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) AB086062.
¶ Both authors contributed equally to this work. ¶ ¶ To whom correspondence should be addressed.  (20), and Nacetylgalactosamine 4-sulfate 6-O-sulfotransferase (21). The sulfation of chondroitin sulfate ordinarily proceeds along with polymerization at the Golgi apparatus. Thus, in order to address control mechanisms of the sulfation reaction, the enzymes involved in the chain synthesis should also be studied, especially the chondroitin sulfate elongation enzymes.
Recent progress with the human genome project and the expansion of other data bases, such as expressed sequence tags (ESTs) and full-length cDNAs, have enabled a wider search for novel genes that are homologous to known genes. Kitagawa et al. (22) identified a human chondroitin synthase from the HUGE (human unidentified gene-encoded large proteins) protein data base by screening with the keywords: "one transmembrane domain" and "galactosyltransferase family." This enzyme had the dual glycosyltransferase activities of glucuronyltransferase II (GlcUAT-II) and N-acetylgalactosaminyltransferase II (GalNAcT-II) that are responsible for synthesizing the repeated disaccharide units of chondroitin sulfate (22). By a similar search of the data base for homologues, five enzymes were found, including chondroitin synthase, which have been cloned and characterized. Thus the enzyme that contributes to the synthesis of repeated disaccharide units on chondroitin sulfate was designated chondroitin sulfate synthase-1 (CSS1), and the others were named CSGalNAcT-1, CSGalN-AcT-2, CSGlcUAT, and CSS2. Chondroitin sulfate GalNAcT-1 (CSGalNAcT-1) and chondroitin sulfate GalNAcT-2 (CSGalN-AcT-2), the second and fourth chondroitin sulfate glycosyltransferases cloned, respectively, exhibit both GalNAcT-II activity for chain elongation and GalNAcT-I activity that determines and initiates the synthesis of the common linkage region of chondroitin sulfate (23)(24)(25)(26). Chondroitin sulfate GlcUA transferase (CSGlcUAT), the third chondroitin sulfate glycosyltransferase cloned, has only GlcUAT-II activity, which is involved in chain elongation (27). Chondroitin sulfate synthase-2, the fifth chondroitin sulfate glycosyltransferase cloned, has both GlcUAT-II and GalNAcT-II activities, which are responsible for synthesis of the repeated disaccharide units of chondroitin sulfate (41). Therefore, more than five enzymes are likely responsible for chondroitin/dermatan sulfate biosynthesis, and they form a gene family analogous to the EXT family for heparin/heparan sulfate biosynthesis (28).
In the present study, a search of several data bases using the amino acid sequences of CS glycosyltransferases revealed a novel gene whose product was characterized as the sixth enzyme having a high homology to CSS1. Interestingly, as implied by its homology to CSS1, this enzyme, designated CSS3, shows both GlcUAT-II and GalNAcT-II activities toward the nonreducing terminal residue of chondroitin/chondroitin sulfate, which has a specific linkage structure but shows no polymerization reaction activity. However, its expression level was much lower than that of CSS1, and its specific activity was about one-tenth the activity of CSS1.

Materials
Cloning of CSS3 and Construction of CSS3 and CSS1 Expression Vectors-A BLAST search of the EST databases was performed using the amino acid sequences of the cloned human chondroitin sulfate glycosyltransferases, CSS1 (GenBank™ accession number AB023207), CSS2 (GenBank™ accession number AB086063), CSGlcUAT (Gen-Bank™ accession number AB037823), CSGalNAcT-1(GenBank TM accession number AB081516), and CSGalNAcT-2 (GenBank TM accession number AB079252) as a query, and a novel EST clone was found (GenBank TM accession number AC004219). As the sequence was not complete, a GENSCAN search of human genomic databases was performed. The predicted sequence was confirmed by PCR with the following two sets of primers: set 1, 5Ј-ATGGCTGTGCGCTCTCGCCGCCC-GT-3Ј and 5Ј-CGTCCCCGCTGCCGTTGTGGCTACT-3Ј; set 2, 5Ј-AGT-AGCCACAACGGCAGCGGGGACG-3Ј and 5Ј-TCAGGAGAGAGTTCG-ATTGTACCT-3Ј (GenBank TM accession number AB086062). The putative catalytic domain of CSS3 (amino acids 130 -883) was expressed as a secreted protein fused with a FLAG peptide in COS-7 cells. An ϳ2.3-kb DNA fragment was amplified by PCR using the Marathon-Ready TM cDNA derived from human brain (Clontech) as a template, with two primers, 5Ј-CCCAAGCTTGCCGAGGGGGAGCCCGA-3Ј and 5Ј-GCTCTAGACTGTCAGGAGAGAGTTCGATT-3Ј. The amplified fragment was inserted between the HindIII and XbaI sites of pFLAG-CMV-1. The putative catalytic domain of CSS1 (amino acids 47-802) was expressed as a secreted protein fused with a FLAG peptide in COS-7 cells. An ϳ2.3-kb DNA fragment was amplified by PCR using a cDNA clone, Kazusa DNA Research Institute number, KIAA0990, with 5Ј-AAGGAAAAAAGCGGCCGCGGGCTGCCGGTCCGGGCAG-3Ј and 5Ј-GCTCTAGACATTAGGCTGTCCTCACTGA-3Ј. The amplified fragment was inserted between the NotI and XbaI sites of pFLAG-CMV-1.
Purification of FLAG-tagged Recombinant Enzymes from Culture Supernatant-COS-7 cells (ATCC CRL-1651) were co-transfected with the expression plasmid and pcDNA3.1 by using TransFast TM (Promega, Madison, WI) according to the manufacturer's instructions. Stable transfectants were selected using 600 g/ml G418 in Dulbecco's modified Eagle's medium containing 10% (v/v) fetal bovine serum (HyClone Laboratories, Logan, UT), 100 g/ml streptomycin sulfate, and 100 units/ml penicillin G and cloned by limiting dilution. Cloned cell lines were tested for synthesis and secretion of the recombinant protein by immunoprecipitation and Western blotting of supernatants using an anti-FLAG BioM2 antibody (Sigma). The secreted enzyme was purified by affinity chromatography using an anti-FLAG M2-agarose gel (Sigma). The conditioned medium and gel were mixed overnight at 4°C and centrifuged for 5 min, and the supernatant was aspirated. The gel was washed five times with 10 ml of 20% (v/v) glycerol in 50 mM Tris-HCl, pH 7.4, and resuspended in the same buffer containing protease inhibitors (1 mM phenylmethylsulfonyl fluoride, 1 g/ml leupeptin, and 1 g/ml pepstatin A) to produce a 50% slurry. The immobilized enzyme was stable at 4°C for at least 4 weeks. The amount of recombinant protein recovered was estimated by immunoblotting. It was separated by SDS-PAGE (10% gel), transferred onto polyvinylidene difluoride membranes (Immobilon, Millipore), and probed with an anti-FLAG peptide antibody BioM2 (Sigma). FLAG-tagged bacterial alkaline phosphatase (FLAG-BAP, molecular mass 49 kDa) was used as a standard to estimate the relative amount, as described previously (27). Blotting and probing were performed according to the manufacturer's instructions, followed by horseradish peroxidase-conjugated streptavidin. Immune complexes were detected as positive bands using the ECL detection system (Amersham Biosciences) with 10-s exposures. The CSS3, CSS1, and BAP protein bands were quantified by densitometric scanning of the digitized image using NIH image (version 1.61) software. The standard curve for each substrate was generated by increasing the

FIG. 2. Estimation of the amount of FLAG epitope-tagged CSS3 and CSS1 proteins.
A and B, Western blot analyses of FLAG-tagged BAP protein and FLAG-tagged CSS3 and CSS1. The proteins were isolated from serial dilutions of the culture medium of COS-7 transfectants that were stably expressing FLAG-tagged CSS3 and CSS1 protein, respectively. The intensity of the 49-kDa band (BAP protein) and the 90-kDa bands (CSS3 and CSS1 proteins) increased with increasing concentrations of FLAG-tagged BAP protein and the volume of the medium, respectively. C and D, depiction of the relationship between the content of BAP protein and the band density; linear correlations were noted (R 2 ϭ 0.991 and 0.967 for CSS3 and CSS1, respectively). E and F, depiction of the relationship between the volume of the medium and the concentrations of CSS3 and CSS1 proteins, respectively, as derived from the BAP standard curve; linear correlations were again observed (R 2 ϭ 0.992 and 0.972 for CSS3 and CSS1, respectively). The amounts of recombinant soluble CSS3 and CSS1 proteins are expressed in arbitrary units of intensity, with 1 unit equivalent to 10 ng of FLAG-tagged BAP protein.
amount of FLAG-tagged BAP protein on the same blotting membrane as the CSS samples. The band intensity and the concentration of the recombinant CSS proteins (90 kDa) in the medium exhibited a linear correlation. The amounts of recombinant CSS proteins could therefore be estimated accurately from the standard curve, which was generated using known amounts of FLAG-tagged BAP protein (49 kDa). The amount of recombinant enzyme protein is expressed in arbitrary units, with each unit of intensity equivalent to 10 ng of FLAG-BAP protein (27).
Preparation of Acceptor Substrate-Glycosaminoglycan polymers were purchased from Seikagaku Corp. For the GlcUAT-II assay, chondroitin sulfate A-E, chondroitin, hyaluronan, heparan sulfate, and N-acetyl heparosan were digested with ␤-glucuronidase prior to the assay (27). Even-and odd-numbered oligosaccharides of chondroitin sulfate, chondroitin, and hyaluronan were prepared as described previously (27).  (27). The labeled products recovered were quantified by liquid scintillation counting. For the ac-ceptor oligosaccharide substrates with an aromatic residue (methoxyphenyl-, benzyl-, or 4-nitrophenyl-) at the reducing terminus, the reaction products were diluted with 1 ml of 0.5 M NaCl and applied to a Sep-Pak C18 cartridge (100 mg; Waters, Milford, MA) (27). The cartridge was washed with 3 ml of 0.5 M NaCl and then 3 ml of water, the product was eluted with 50% methanol, and the radioactivity of all fractions was measured by liquid scintillation counting.
Identification of the Enzyme Reaction Products-Each product of the GlcUAT-II reaction using chondroitin or CS11 and of the GalNAcT-II reaction using chondroitin was isolated by gel filtration column chromatography using the Superdex TM Peptide HR10/30 column. The radioactive peak containing the product was pooled and desalted with the Fast Desalting column HR10/10 using distilled water as an eluant, and was lyophilized. In order to identify the linkage structure, the dried sample (about 20 pmol of radiolabeled material) from GlcUAT-II reaction was incubated with one of the following: 1) 100 milliunits of chondroitinase ACII in a total volume of 100 l of 100 mM Tris-HCl, pH 7.4, containing 30 mM sodium acetate at 37°C overnight; or 2) 1 unit of ␤-glucuronidase in a total volume of 100 l of 100 mM sodium acetate buffer, pH 5.0, at 37°C overnight. To confirm the linkage structure, assays were carried out to determine whether the product could serve as an acceptor for E. coli K4 strain chondroitin polymerase, which synthesizes chondroitin, and if the resultant products could be digested completely with chondroitinase ACII (29) Briefly, 20 pmol of the radiolabeled material was lyophilized and served as a substrate for E. coli K4 strain chondroitin polymerase. The reaction was performed at 30°C overnight in a 50-l solution containing 50 mM Tris-HCl, pH 7.2, 20 mM MnCl 2 , 0.1 M (NH 4 ) 2 SO 4 , 1 M ethylene glycol, 20 pmol of radiolabeled [ 14 C]CS12, 30 nmol each of UDP-GlcUA and UDP-GalNAc, and 0.8 g of the enzyme preparation. The solution was then boiled for 5 min to stop the reaction. The radioactive peak containing the product was pooled and desalted with the Fast Desalting column HR10/10 using distilled water as an eluant and was then lyophilized. The dried sample (about 20 pmol of radiolabeled material) containing the E. coli K4 strain chondroitin polymerase reaction products was incubated with 100 milliunits of chondroitinase ACII in a total volume of 100 l of 100 mM Tris-HCl, pH 7.4, containing 30 mM sodium acetate at 37°C overnight. The enzyme digests were analyzed using the same Superdex TM Peptide HR10/30 column described above. In order to identify the linkage, the dried sample (about 20 pmol of radiolabeled material) from GalNAcT-II reactions was incubated with 100 milliunits of chondroitinase ACII in a total volume of 100 l of 100 mM Tris-HCl, pH 7.4, containing 30 mM sodium acetate at 37°C overnight, or with 100 milliunits of ␣-N-acetylgalactosaminidase in a total volume of 100 l of 50 mM sodium citrate buffer, pH 4.5, at 37°C overnight. The enzyme digests were analyzed using the same Superdex TM Peptide HR10/30 column described above.
Quantitative Analysis of the CSS3 Transcript in Human Tissues by Real Time PCR-For quantification of CSS3 transcripts, the real time PCR method was employed, as described in detail previously (30). Marathon Ready cDNA derived from various human tissues was purchased from Clontech. Standard curves for the endogenous control, glyceraldehyde-3-phosphate dehydrogenase (GAPDH) cDNA, were generated by serial dilution of pCR2.1 (Invitrogen) DNA containing the GAPDH gene. The primer set and probe for CSS3 are as follows: the forward primer, 5Ј-CCCAGAAAAAGTCCTTCATGATG-3Ј, and the reverse primer, 5Ј-AACTCTTCTAATTTGTCACCTTTGATGTAG-3Ј, and the probe, 5Ј-ATGAGTGGTTCATGCGC-3Ј, which contains a minor groove binder. The primer sets and probes for CSS1 are as follows: the forward primer for CSS1, 5Ј-AGTGTGTCTGGTCTTATGAGATGCA-3Ј, and the reverse primer for CSS1, 5Ј-AGCTGTGGAGCCTGTACTGG-TAG-3Ј, and the probe for CSS1, 5Ј-ATGAGAATTACGAGCAGAAC-3Ј, which contains a minor groove binder. PCR products were measured continuously with an ABI PRISM 7700 Sequence Detection System (Applied Biosystems, Foster City, CA). The relative amounts of the transcripts were normalized to the amount of GAPDH transcript in the same cDNA sample.

Molecular
Cloning of CSS3 and Determination of Its Nucleotide and Amino Acid Sequences-A BLAST search of the EST data bases was performed using the amino acid sequences of the cloned human chondroitin sulfate glycosyltransferases as a query, and a novel EST was found (GenBank TM accession number AC004219). As the sequence was incomplete, a GENSCAN search of human genomic data bases was performed. The predicted sequence was confirmed by PCR with two sets of primers as follows: set 1, 5Ј-ATGGCTGTGCGCTCTCGCCGCCCGT-3Ј and 5Ј-CGTCCCCGCTGCCGTTGTGGCTACT-3Ј; set 2, 5Ј-AG-TAGCCACAACGGCAGCGGGGACG-3Ј and 5Ј-TCAGGAGAG-AGTTCGATTGTACCT-3Ј (GenBank TM accession number AB086062). The putative amino acid sequences are shown in Fig. 1A. They contained an open reading frame of 2646 bp, corresponding to 882 amino acids, encoding a typical type II membrane protein with three possible N-glycosylation sites (arrowheads).
Comparison of Amino Acid Sequences between CSS3 and CSS1-The deduced amino acid sequence of the clone exhibited high homology (62%) with CSS1, as shown in Fig. 1A. Hydropathy plots of the amino acid sequence revealed one hydrophobic stretch, located at position 9 -29, as in CSS1 (Fig. 1A, underlined). Two DXD motifs, which are conserved in many glycosyltransferases and function as key sequences for divalent cation binding, as well as other motifs that are conserved in ␤1,3-glycosyltransferases (␤3GTs) and ␤1,4-glycosyltransferases (␤4GTs), were found, with one in the N-terminal and one in the C-terminal region, respectively (Fig. 1A, boxed). A comparison of the cysteine residue locations in the predicted proteins encoded by CSS3 (11 cysteines) and by CSS1 (11 cysteines) showed good conservation of 10 of these cysteines (Fig. 1A, boldface). Two of the three potential N-glycosylation sites in CSS3 appeared to be conserved in both enzymes (Fig.  1A, closed arrowheads). The third potential N-glycosylation site was located close to the C terminus of both enzymes (Fig.  1A, open arrowheads). A remarkable difference between the two amino acid sequences was seen in the lengths of their N-terminal stem regions. The proline-rich stem region of CSS3 was about 90 amino acids in length and was somewhat longer than the corresponding region in CSS1. Genome Organization and Chromosome Localization-A comparison of the cDNA sequence with the genomic sequence on chromosome 5 revealed that the CSS3 gene spans over 287 kb, because of its long second intron (ϳ280 kb), and that it consists of at least three exons (Fig. 1B, uppercase). In contrast, the CSS1 gene spans over 75 kb. Its genomic organization, including the exon-intron boundaries, is similar to that of the CSS3 gene (Fig. 1B, lowercase) and consists of three discrete exons in the coding region (27). The CSS3 and CSS1 genes were located on human chromosome 5q31 and 15q26.3, respectively. Estimation of the Amount of FLAG Epitope-tagged CSS3 and CSS1 Proteins-To facilitate functional analysis of the putative glycosyltransferase, a soluble form of the protein was generated by replacing the first 129 and 46 amino acids of CSS3 and CSS1, respectively, with the preprotrypsin signal sequence and a FLAG tag, as described under "Experimental Procedures." The soluble putative glycosyltransferases were expressed in COS-7 cells as recombinant enzymes fused with the FLAG tags. The fused enzymes expressed in the medium were adsorbed onto anti-FLAG M2 antibody-conjugated agarose gels to eliminate endogenous glycosyltransferases, and the enzymebound gels were then used for the various reactions. The amounts of FLAG-tagged CSS3 and CSS1 were estimated using FLAG-BAP as a standard (Fig. 2, A and B), as described previously (27). The amount of FLAG-BAP was correlated with the densitometric units obtained by measurement of each band intensity (R 2 ϭ 0.991 and 0.967 for CSS3 and CSS1, respectively), as shown in standard curves in Fig. 2, C and D. The amounts of recombinant CSS3 and CSS1 proteins were determined by defining 10 ng of the FLAG-BAP as 1 unit. One unit of CSS3 and of CSS1 was obtained from 19.9 (Fig. 2E) and 134.9 ml (Fig. 2F) of the pooled medium, respectively.
Acceptor Substrate Specificities of CSS1 and CSS3-The acceptor specificities of the truncated CSS3 and CSS1 proteins recovered from COS-7 transfectant cells were determined by testing a variety of glycosaminoglycans and their oligosaccharides as acceptor substrates. As expected from the homologies noted above, preliminary experiments showed that the CSS3 glycosyltransferase had dual enzymatic activities, a GlcUAT-II activity for CS-C11 and a GalNAcT-II activity for CS-C10. Therefore, the effects of buffers and pH on the enzymatic activities of both recombinant glycosyltransferases were examined and compared because those for CSS1 have not been determined (22). As shown in Fig. 3, CSS3 exhibited optimum activity at pH 5.8 and 6.2 in MES buffer with GlcUAT-II and GalNAcT-II, respectively (Fig. 3, A and C). CSS1 exhibited optimum activities at pH 5.8 in MES buffer with both Gl-cUAT-II and GalNAcT-II (Fig. 3, B and D). The reactions catalyzed by each of these enzymes were carried out with the appropriate buffer conditions. Divalent cations were essential for the activity of both enzymes, and 10 mM EDTA completely abolished all enzymatic activities (Fig. 4). For CSS3, Co 2ϩ evoked the highest level of activity under standard assay conditions, and Mn 2ϩ was 31 and 64% as effective as Co 2ϩ for GlcUAT-II and GalNAcT-II activity, respectively. Cd 2ϩ was also effective, as was Mn 2ϩ (Fig. 4A) In the case of CSS1, Co 2ϩ evoked the highest level of activity under standard assay conditions with the substrate GlcUAT-II, whereas Cd 2ϩ and Mn 2ϩ were 85 and 70% as effective as Co 2ϩ , respectively. Mn 2ϩ exhibited the highest activity for GalN-AcT-II under standard assay conditions, and Co 2ϩ and Cd 2ϩ were 70 and 53% as effective as Mn 2ϩ , respectively (Fig. 4B). The optimal concentrations for both enzymes were 20 mM Mn 2ϩ for the GlcUAT-II substrate and about 10 mM Mn 2ϩ for the GalNAcT-II substrate (Fig. 5, A and B). Under the established standard incubation conditions described under "Experimental Procedures," GlcUA incorporation into CS-C11 and GalNAc incorporation into CS-C10 were proportional to the incubation time for up to 4 h, for each of the enzymes (data not shown). The specificity of the recombinant CSS3 and CSS1 toward each UDP-sugar donor substrate was analyzed using a series of radiolabeled molecules UDP-GlcUA, UDP-Gal, UDP-GalNAc, and UDP-GlcNAc under optimized conditions. Both CSS3 and CSS1 were able to catalyze efficiently the transfer of GlcUA from UDP-GlcUA to the acceptor CS-C11 and of GalNAc from UDP-GalNAc to the acceptor CS-C10. In contrast, the other radiolabeled nucleotide sugars tested (UDP-Gal, UDP-GalNAc, and UDP-GlcNAc for CS-C11; UDP-GlcUA, UDP-Gal, and UDP-GlcNAc for CS-C10) were not substrates of the recombinant CSS3 and CSS1 (data not shown). Furthermore, various monosaccharides including GlcUA␤-O-4-nitrophenyl, Glc-NAc␣-O-benzyl, GlcNAc␤-O-benzyl, Gal␣-O-benzyl, Gal␤-Obenzyl, GalNAc␣-O-benzyl, and GalNAc␤-O-benzyl were not efficient acceptor substrates for any of the UDP-sugars tested as donors (data not shown).
Acceptor Activity of Polymer Chondroitin and Various Chondroitin Sulfate Isoforms-To characterize the substrate specificity of the purified recombinant CSS3 and CSS1, chondroitin, chondroitin sulfate isoforms, and other glycosaminoglycans were tested as acceptor substrates. As shown in Table I, chondroitin was the best substrate and the other polymer chondroitin sulfate isoforms were poor acceptors for enzymatic reactions catalyzed by both CSS3 and CSS1. The specific activity of CSS1 for chondroitin was about 18-and 10-fold more than the activity of CSS3 for GlcUAT-II and GalNAcT-II, respectively.
Acceptor Activity of Chondroitin and Chondroitin Sulfate Oligosaccharides-As shown in Table I, both enzymes apparently showed GlcUAT-II activity toward undecasaccharides having GalNAc in their nonreducing termini. These substrates were prepared from the CS isoforms and chondroitin. The activities of CSS3 for the CS-A and CS-C undecasaccharides were 1.6-and 1.7-fold higher than its activity for the chondroitin undecasaccharide, and for CSS1 these activities were 2.0-and 2.4-fold higher. The activity of each enzyme toward the hyaluronan undecasaccharide was negative. On the other hand, both enzymes apparently showed GalNAcT-II activity toward decasaccharides having GlcUA at their nonreducing termini, which were prepared from CS isoforms and chondroitin. The activities of CSS3 for the CS-A and CS-C decasaccharides were 1.9-and 2.1-fold higher than that for the chondroitin decasaccharide, and these activities of CSS1 were 1.6and 1.7-fold higher, whereas the activity for the hyaluronan decasaccharide was negative for each enzyme. The specific activity of CSS1 for chondroitin sulfate oligosaccharide was about 10-and 32-fold higher than those of CSS3 for transferring Gl-cUAT-II to CS-C11 and for transferring GalNAcT-II to CS-C10, respectively. The apparent K m values for UDP-GlcUA for GlcUA- Effects of the length of oligosaccharides on CSS3 and CSS1 activities were determined using CS-C and chondroitin oligosaccharides as acceptors (Fig. 6). For both the GlcUAT-II and the GalNAcT-II activity of CSS3, the CS-C oligosaccharides were better acceptor substrates than the chondroitin oligosaccharide (Fig. 6, A and C). In the case of chondroitin oligosaccharides, the longer oligosaccharides served as better acceptors for CSS3 than did the shorter oligosaccharides (Fig. 6, A and C, closed circles). On the other hand, in CSS3 activity toward CS-C oligosaccharides, CS-C9 was the most efficient acceptor for GlcUAT-II, and CS-C8 was the most efficient acceptor for GalNAcT-II (Fig. 6, A and C, open circles). A similar tendency was also observed in CSS1 activity toward CS-C oligosaccharides, although CS-C11 is the most efficient acceptor for Glc-UAT-II, and CS-C12 is the most efficient acceptor for Gal-NAcT-II (Fig. 6, B and D, open circles). To examine whether the CSS3 enzyme has other glycosyltransferase activities, several substrates were tested as acceptors, Gal␤1-3Gal␤1-4Xyl␤ 1 -Omethoxyphenyl and GlcUA␤1-3Gal␤1-3Gal␤1-4Xyl␤ 1 -O-methoxyphenyl (with GlcUAT-I and GalNAcT-I for the glycosaminoglycan linkage region, respectively) (Table I), Gal␤1-3GalNAc␣-O-benzyl (human natural killer cell-1 epitope synthase), and Gal␤1-4GlcNAc␤1-3Gal␤1-4GlcNAc (lactosamine tetrasaccharide), but no activity was detected (data not shown). These results again suggested that CSS3 is responsible for the chondroitin sulfate elongation but not for the linkage tetrasaccharide or activity toward other substrates, as described previously for CSS1 (22). Both CSS3 and CSS1 have dual glycosyltransferase activities of GlcUAT-II and GalNAcT-II, which are responsible for synthesizing the repeating disaccharide units of chondroitin sulfate. However, incubations of the recombinant soluble enzymes with CS-C10 in the presence of UDP-[ 3 H]GalNAc and UDP-GlcUA in vitro, despite the individual transferase activities observed for CSS3 and CSS1, did not result in polymerization (Fig. 7, A and B, closed circles). In contrast, E. coli K4 strain chondroitin polymerase yielded radiolabeled polymer chondroitin chains when the incubation was carried out with UDP-[ 3 H]GalNAc and UDP-GlcUA in vitro (Fig. 7C, closed circles). All of the enzymes demonstrated [ 3 H]GalNAc monosaccharide transfer to CS-C10 in the presence of UDP-[ 3 H]GalNAc alone (Fig. 7, open circles). These results suggest that CSS3 and CSS1 cannot produce chondroitin polymer upon incubation with UDP-sugars and oligosaccharide acceptor substrates in vitro, whereas the E. coli K4 strain chondroitin polymerase can carry out this reaction.
Analysis of CSS3 Reaction Products-To identify the Gl-cUAT-II reaction products, chondroitin polymer was labeled with [ 14 C]UDP-GlcUA by CSS3 under optimized conditions, and the products were isolated and then subjected to gel filtration analysis after treatment with chondroitinase AC-II or ␤-glucuronidase. As shown in Fig. 8A, the labeled products were completely digested by chondroitinase AC-II or ␤-glucuronidase, quantitatively yielding two 14 C-labeled peaks at the positions of [ 14 C]GlcUA␤1-3GalNAc and of free [ 14 C]GlcUA, respectively. These findings indicate that a GlcUA residue was transferred to the nonreducing terminal GalNAc residue of the chondroitin polymer through a ␤-linkage. Furthermore, 14 Clabeled chondroitin sulfate dodecasaccharide was used as an acceptor for a chondroitin polymerase from an E. coli K4 strain (Fig. 8B) (29). This reaction was performed in the presence of the enzyme and two donor substrates, UDP-GalNAc and UDP-GlcUA. The reaction products showed a molecular mass of ϳ3000 Da that was speculated to be the product resulting from transfer of approximately five sugar residues. These products were digested by chondroitinase AC-II, yielding unsaturated GlcUA␤1-3GalNAc disaccharides. These results indicated the  To identify the GalNAcT-II reaction products, chondroitin polymer was labeled with [ 3 H]GalNAc by CSS3, and the products were isolated and then subjected to gel filtration analysis after chondroitinase AC-II treatment. As shown in Fig. 8C, the labeled products were completely digested by chondroitinase AC-II, quantitatively yielding a 3 H-labeled peak at the position of free [ 3 H]GalNAc, which was separable from GlcUA␤1-3GalNAc. In contrast, the products were inert to the action of ␣-N-acetylgalactosaminidase. These findings clearly indicated that a GalNAc residue had been transferred exclusively to the nonreducing terminal GlcUA residue of the polymer chondroitin through a ␤1-4 linkage. These findings taken together suggest that the identified protein is chondroitin sulfate synthase, which has both CSGl-cUAT-II and CSGalNAcT-II activity.
Quantitative Analysis of CSS3 and CSS1 in Human Tissues by Real Time PCR-The tissue distribution and expression levels of the CSS3 transcripts were determined and compared with those of the transcripts by the real time PCR method. Expression levels of CSS3 and CSS1 in various tissues are shown as relative amounts versus the GAPDH transcripts in Fig. 9. Both enzymes were expressed ubiquitously, albeit differentially, in all tissues examined. Notably, expression of CSS3 was very low compared with CSS1 expression levels. The expression levels of CSS3 were less than 1/10 of CSS1 expression levels, except in brain, cerebral cortex, heart, skeletal muscle, and small intestine (Fig. 9).

DISCUSSION
In this report, we have described the cloning and characterization of a novel human chondroitin sulfate synthase-3 (CSS3), which is homologous with the human chondroitin sulfate synthase-1, formerly known as chondroitin synthase (22). We have designated these enzymes, which exhibit both Gl-cUAT-II activity and GalNAcT-II activity for chondroitin chain elongation in vitro, CSS (chondroitin sulfate synthase). The CSS3 enzyme is the third CSS and the sixth chondroitin sulfate glycosyltransferase identified.
The high amino acid sequence similarity, which includes many characteristic motifs shared between CSS3 and CSS1, suggests that the enzymes have related functions and evolutionary origins. A remarkable difference in two amino acid sequences, however, was seen in the lengths of their Nterminal stem regions. The stem region in CSS3 was about 90 amino acids longer than that in CSS1. This extra region consists of a proline-rich sequence that has no sequence sim-ilarity to any other proteins found in various data bases. 2 The longer stem region presumably not only separates the catalytic domain from the membrane proximal segment by an appropriate distance, but it also may have a function specific to CSS3. For example, it may determine the localization of CSS3 in the Golgi membrane or may allow it to participate in interactions with other molecules (31). Alternatively, the protein chain may be cleaved in this region (32). Chromosomal assignments of the six human chondroitin glycosyltransferases, CSS1, CSS3, CSS2, CSGlcUAT, CSGalNAcT-1, and CSGalNAcT-2, indicate that these genes are localized on different chromosomes, 15q26.3, 5q31, 2q36.1, 7q36, 8q21.3, and 10q11.22, respectively, despite the significant similarities of the nucleotide and predicted amino acid sequences among the six genes (22)(23)(24)(25)(26). Homologues of human CSS1 and CSS3 have been identified in Caenorhabditis elegans and in Drosophila, suggesting the duplication might have occurred after the evolutionary branch point leading to Drosophila. Interestingly, a homologue of human CSGalNAcT-1 and -2 is absent from C. elegans but is present in Drosophila, and a homologue of human CSGlcUAT or CSS2 is absent from both C. elegans and Drosophila. Differences in amino acid sequences were used to measure the evolutionary relationships between human CSS3 and other members of the CS glycosyltransferase family (Fig. 10C). The resulting dendrogram indicated that human CS glycosyltransferase genes may be classified into three groups (CSS1/CSS3, CSS2/ CSGlcUAT, and CSGalNACT-1/CSGalNAcT-2), with all groups in the family roughly equidistant from each other (Fig. 10C). This observation suggests that the CS glycosyltransferase family emerged early in evolution and has undergone recent duplication.
In this study we purified CSS3 and CSS1 as soluble enzymes, determined the amounts of these proteins, and investigated and compared the properties of purified CSS3 and CSS1. As was expected from its high homology with CSS1, CSS3 exhibited both GlcUAT-II and GalNAcT-II activity. Under the optimal conditions for each activity, the substrate specificity of CSS3 was shown to be quite similar to that of CSS1 toward the substrates for GlcUAT-II and GalNAcT-II. In summary, both enzymes prefer the nonsulfated isomer as a polymer substrate and the sulfated isomers as oligosaccharide substrates in vitro (22,27). The same specificities were observed for all of the cloned CS glycosyltransferases, CSS2 (41), CSGlcUAT (27), CSGalNAcT-1 (25), CSGalNAcT-2 (24), as well as for bovine serum ␤-glucuronyltransferase and bovine serum N-acetylgalactosaminyltransferase (34,35). Thus the specificities of the enzymes were similar, but their specific activities were quite different. The specific activities of CSS1 were much higher domains in ␤3GalTs or ␤4GalTs are indicated by open or closed boxes, respectively. The putative ␤4GalT-like domains observed in the C-terminal halves of CSGlcUA-T and CSS2 are indicated by dotted boxes. However, CSGlcUA-T has no GalNAcT-II activity for resulting in CS elongation, although CSS2 does have this activity. This is indicated by demerit marks in the dotted box. aa, amino acids. B, xylosyltransferase (XT), ␤1,4-galactosyltransferase VII (␤4GalT7), ␤1,3-galactosyltransferase VI (␤3GalT6), and ␤1,3-glucuronyltransferase-I (GlcUAT-I) work sequentially on the synthesis of tetrasaccharide linkages. In the processes of CS synthesis, CSS1, CSS2, and CSS3, which are CS synthases having both ␤1,3-glucuronyltransferase and ␤1,4-N-acetylgalactosaminyltransferase activities, as well as CS ␤1,3-glucuronyltransferase (CSGlcUAT) and CS ␤1,4-N-acetylgalactosaminyltransferase 1 and 2 (CSGalNAc-T1 and CSGalNAc-T2), participate. The sizes of the letters in the names of CS glycosyltransferases indicate their relative enzymatic activities. C, phylogenetic trees. Unrooted phylogenetic trees were produced by phylogenetic analysis using ClustalW, using the full coding sequences of all CS glycosyltransferase genes (human CSS1, GenBank TM accession number AB023207; human CSS2, GenBank TM accession number AB086063; human CSS3, GenBank TM accession number AB086062; human CSGlcUAT, GenBank TM accession number AB037823; human CSGalNAcT-1, GenBank TM accession number, AB071403; human CSGalNAcT-2, GenBank TM accession number AB090811; Drosophila CSS, GenBank TM accession number AE003499; C. elegans CSS, GenBank TM accession number T25233; Drosophila CSGalNAcT, GenBank TM accession number CG12913).
(10 -40-fold) than those of CSS3, CSS2, and CSGlcUAT. Furthermore, in comparing the expression level of CSS3 with those of the other five recently cloned chondroitin sulfate glycosyltransferases, quantitative PCR analyses has revealed that CSS3 is expressed at the lowest level of all the enzymes. Therefore, it is hard to consider that CSS3 may play a major role in the biosynthesis of chondroitin sulfate in all tissues. CSS3 activity is probably most significant in tissues where its expression level is relatively high compared with CSS1, as in the cerebral cortex and small intestine.
In order to express constructs of these enzymes as soluble proteins in mammalian cells, their transmembrane domains and some of their stem regions were removed. These approaches are used routinely in the study of glycosyltransferases, but we are not able to rule out possible deleterious effects of these alterations, which may have caused conformational changes and a decrease or elimination of some of their activities. In future studies it will be desirable to analyze their activities when they are expressed as full-length molecules.
The presence of at least six distinct enzymes that are involved in CS initiation and elongation suggests that CS synthesis takes place via a mechanism similar to heparan sulfate synthesis, where five homologous EXT family members with distinct but overlapping acceptor specificities are involved (28). In the case of heparan sulfate biosynthesis, EXT1 and EXT2, both of which are polymerases, can form complexes (36). When either EXT1 or EXT2 are overexpressed by transfection, the bulk of the protein remains in the endoplasmic reticulum. When the enzymes are expressed together, they move to the Golgi, where polymerization and sulfation are thought to occur (36,37). Importantly, co-expression greatly enhances GlcUA transferase activity in mammalian cells (36), and both GlcNAc and GlcUA transferase activities increase when both enzymes are expressed in yeast (37). EXT2 can also catalyze the transfer of GlcNAc and GlcUA to an oligosaccharide acceptor in vitro, but transfection of mammalian cells with EXT2 has very little effect on measured enzyme activities. Paradoxically, transfection with EXT2 decreases GlcUA transferase activity in COS cells (38). In addition, transfection of EXT1-deficient cell lines with EXT2 does not restore HS synthesis (36,39). Thus, despite its dual catalytic properties, EXT2 appears to work more as a molecular chaperone than as an HS polymerase that plays a role redundant to that of EXT1. Recently, a homologue of core 1 ␤3-galactosyltransferase (C1Gal-T2) was shown to act as a specific molecular chaperone (Cosmc: core 1 ␤3-Gal-T-specific molecular chaperone) in assisting the folding/stability of C1␤3Gal-T1 (33,40). Cosmc is required for the expression of C1␤3Gal-T1 activity in mammalian cells, and a mutation of Cosmc in Jurkat cells causes loss of C1␤3Gal-T1 activity and targeting of the inactive protein to the proteasome (40). Thus, heparan sulfate co-polymerase may consist of a complex containing EXT1 and EXT2, in which both subunits are essential for activity in vivo.
Experimentally, EXT1 was seen to have a more robust activity than EXT2 when expressed independently, but both subunits apparently can catalyze individual reactions. In the case of CS biosynthesis, CSS3 may work with other chondroitin sulfate glycosyltransferases as a co-polymerase in vivo. In fact, none of the purified, soluble versions of CSS1, CSS2, and CSS3 can produce chondroitin polymer upon incubation with UDPsugars and oligosaccharide acceptor substrates in vitro, whereas the E. coli K4 strain chondroitin polymerase can carry out this reaction (Fig. 7) (29).
In conclusion, at least six CS glycosyltransferases, CSS1, CSS3, CSS2, CSGlcUA-T, CSGalNAcT-1, and CSGalNAcT-2, form a gene family and can be classified into three groups based on their structures and enzymatic specificities (Fig. 10A). They may play roles at different stages of CS synthesis in a co-operative or orchestrated manner (Fig. 10B). The recent cloning, expression, and characterization of many glycosyltransferases have led to great progress in understanding chondroitin sulfate biosynthesis. However, none of these enzymes alone can form an entire glycosaminoglycan chain, and the coordination of complex cellular machinery may be required for CS chain elongation. To understand the nature of the complexes formed by these molecules, further information is required on interactions between the various membrane-bound enzymes including CS sulfotransferases, their appropriate substrates, and membrane-bound nascent proteoglycans. Such information will be critical to determine the specific structural basis underlying the efficiency of the reactions carried out during formation of the CS chain.