Expression Cloning of a Human G T3 Synthase G D3 AND G T3 ARE SYNTHESIZED BY A SINGLE ENZYME*

Gangliosides of the C series such as G T3 are polysialy- lated glycosphingolipids whose synthesis is develop-mentally regulated. Here we report the expression cDNA cloning and characterization of G T3 synthase that adds the second (cid:97) -2,8-sialic acid to G D3 , Neu- NAc (cid:97) 2 3 8NeuNAc (cid:97) 2 3 3Gal (cid:98) 1 3 4Glc 3 Cer, thus forming G T3 , 4Glc 3 Cer. Unexpectedly, the cloned cDNA was found to be identical to the cDNA that encodes G D3 synthase. The newly identified enzyme was therefore named G D3 /G T3 synthase (G D3 /G T3 ST). G D3 /G T3 ST synthesized G T3 most efficiently when G M3 , NeuNAc 2 3 3Gal (cid:98) 1 3 4Glc 3 Cer, was incubated as an acceptor, indicating that G D3 /G T3 ST is a polysialyltransferase that can transfer more than one sialic acid residue via (cid:97) -2,8 linkage

Gangliosides of the C series such as G T3 are polysialylated glycosphingolipids whose synthesis is developmentally regulated. Here we report the expression cDNA cloning and characterization of G T3 synthase that adds the second ␣-2,8-sialic acid to G D3 , Neu-NAc␣238NeuNAc␣233Gal␤134Glc3 Cer, thus forming G T3 , NeuNAc␣238NeuNAc␣238NeuNAc␣233Gal␤13 4Glc3 Cer. Unexpectedly, the cloned cDNA was found to be identical to the cDNA that encodes G D3 synthase. The newly identified enzyme was therefore named G D3 /G T3 synthase (G D3 /G T3 ST). G D3 /G T3 ST synthesized G T3 most efficiently when G M3 , NeuNAc␣233Gal␤134Glc3 Cer, was incubated as an acceptor, indicating that G D3 /G T3 ST is a polysialyltransferase that can transfer more than one sialic acid residue via ␣-2,8 linkage to gangliosides. Moreover, a longer period of incubation of G D3 with G D3 /G T3 ST produced a significant amount of G T3 and higher polysialogangliosides. Among various cell lines expressing G D3 /G T3 ST, higher polysialogangliosides including G T3 were detected only in cell lines where the amount of G D3 /G T3 mRNA is sufficiently high. The expression of G D3 /G T3 ST mRNA among human tissues is highly restricted to fetal and adult brains. The G D3 / G T3 ST gene was found to be located at chromosome 12, region p12. Taken together, these results indicate that C series polysialogangliosides are synthesized by a ganglioside-specific polysialyltransferase, G D3 /G T3 ST, that is specifically expressed in neural tissues.
Glycoconjugates are major components of the plasma membrane of mammalian cells, and their carbohydrate structures change dramatically during development. Specific sets of carbohydrates are expressed in different stages of differentiation, and many of those carbohydrates are recognized by specific antibodies, thus providing differentiation antigens (Feizi, 1985;Fukuda, 1985). During the course of development, expression of distinct carbohydrates is eventually restricted to specific cell types, and aberrations in these cell surface carbohydrates are frequently observed in malignant cells (Hakomori, 1984). The functional significance of these cell type-specific carbohydrates and their alterations in malignancy is not well understood, although various reports suggest that some of these carbohydrates are involved in cell adhesion processes (Fukuda, 1992;Lowe, 1994).
Among glycosphingolipids, gangliosides comprise a structurally diverse set of sialylated species and are enriched in nervous tissues. Gangliosides have been found to act as receptors for growth factors, toxins, and viruses and are apparently involved in cell adhesion. For example, cholera toxin binds to G M1 , 1 Gal␤133GalNAc␤134(NeuNAc␣233)Gal␤134Glc-␤13 Cer, before its entry into cells (Spiegel and Fishman, 1987). Influenza A virus binds to sialylparagloboside, Neu-NAc␣233Gal␤134GlcNAc␤133Gal␤134Glc␤13 Cer (Higa et al., 1985;Suzuki et al., 1986). In addition, there have been reports suggesting that gangliosides, G D3 in particular (see Fig.  1 for its structure) may play roles in cell-cell interaction. Cheresh et al. (1986) found that G D3 and G D2 facilitate the attachment of human melanoma and neuroblastoma cells to extracellular matrix proteins. Epithelial-mesenchymal interactions in embryonic kidney formation were perturbed by anti-G D3 antibody, which reacted with G D3 on the mesenchymal cells (Sariola et al., 1988). Gangliosides also modulate enzymatic activities. For example, G M3 was found to inhibit epidermal growth factor receptor-mediated phosphorylation (Bremer et al., 1986), and G Q1b␣ was shown to inhibit ADP-ribosyltransferases (Hara-Yokoyama et al., 1995).
Among gangliosides, increasing attention has been directed to the so-called C series polysialogangliosides, which have unique trisialosyl residues, NeuNAc␣238NeuNAc␣238Neu-NAc␣233Gal3 R (Fig. 1). C series polysialogangliosides were found to be major constituents in adult fish brain. In higher vertebrates the C series polysialogangliosides comprise a minor proportion of total gangliosides present in the brain (Ando and Yu, 1979). However, a substantial amount of C series polysialogangliosides are present in fetal brain of higher vertebrates including human. They are also found in various neuroectodermal tumors, such as melanoma and glioma (Yates, 1988;Nakayama et al., 1993). In the early stages of neural development, G D3 is predominantly expressed in the neural tube that consists of progenitor cells for neurons and macroglial cells. During the later stage of development, progenitor cells migrate and extend processes and finally differentiate to postmitotic neurons. In this developmental period, G D3 decreases, and C series polysialogangliosides, such as G T3 , increase (Rösner et al., 1985). * This work was supported by Grants RO1 CA33895 (to M. F.) and DK 37016 (to M. N. F.) from the National Institutes of Health. 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 /EMBL Data Bank with accession number(s) L43494.
It has been generally accepted that each glycosyltransferase involved in the synthesis of gangliosides transfers only one sugar residue to form a specific linkage (Pohlentz et al., 1988) (Fig. 1). Until recently, the studies of C series polysialogangliosides have been limited to their structural analysis, since the enzyme responsible for G T3 synthesis (STIII) was not purified or cloned. In order to understand the roles and synthesis of C series polysialogangliosides, it is critical to isolate a cDNA clone of STIII that forms G T3 .
In this report, we describe the cloning of cDNA encoding G T3 synthase, STIII, using a mammalian expression cloning with a newly devised modification. Surprisingly, the newly isolated cDNA was found to be identical to that of G D3 synthase, STII. By transfecting the newly isolated cDNA into HeLa and MeWo cells and assaying the activity of the soluble form of the enzyme, we demonstrated that a single enzyme encoded by the isolated cDNA forms both G D3 and G T3 . We also found that G T3 synthase transcripts are expressed exclusively in neural tissues and that its gene is located at chromosome 12, region p12. These results, taken together, indicate that polysialogangliosides are synthesized by a single enzyme, G D3 /G T3 synthase, which is specifically expressed in neural tissues.
Construction of Stably Transfected COS-1 Cells Expressing G D3 -COS-1 cells were transfected with pAMo-GD3 (Sasaki et al., 1994b) and selected for G418 resistance. The transfected COS-1 cells expressing G D3 were selected by immunofluorescent staining with anti-G D3 antibody, R24, and one clone named COS-1⅐G D3 cells was isolated.
Expression Cloning of a Human G T3 ST cDNA-A mammalian expression vector, pcDNAI-based cDNA library, pcDNAI-SK-MEL-28 constructed from poly(A) ϩ RNA isolated from human melanoma SK-MEL-28 cells, was purchased from Invitrogen (San Diego, CA). SK-MEL-28 cells express a significant amount of G T3 (Dubois et al., 1986). COS-1⅐G D3 cells (1.2 ϫ 10 7 ) were transfected with 20 g of pcDNAI-SK-MEL-28 using lipofectamine (Life Technologies, Inc.). After 62 h, the transfected cells were detached at 37°C in Hanks'-based cell dissociation solution (Specialty Media, Lavallette, NJ). The detached cells were pooled and resuspended in cold phosphate-buffered saline (pH 7.4) containing 1% bovine serum albumin and were reacted with mouse monoclonal antibody M6703 at 1:200 dilution. After a 30-min incubation on ice, the cells were washed, and then fluorescein isothiocyanate (FITC)-conjugated (FabЈ) 2 fragment of goat anti-mouse IgG (Cappel, Durham, NC) was added. After a 30-min incubation on ice, the cells were washed and subjected to fluorescence-activated cell sorting (FACS) using FACStar (Becton-Dickinson, San Jose, CA). The sorting region was set where only strongly positive COS-1 cells were recovered. Plasmid DNAs were rescued from the positive cells (Hirt, 1967) and transformed into the host Esherichia coli MC1061/p3 cells by electroporation using Cell-Porator (Life Technologies, Inc.). The transformed cells were placed into 20 plates, each containing about 500 colonies. Plasmid DNAs prepared from each plate were separately used for transfection by lipofectamine into HeLa cells, and the transfected HeLa cells were examined by immunofluorescent staining using M6703 antibody. Sibling selection with sequentially smaller active pools identified a single plasmid, pcDNAI-G T3 ST, that determined the expression of G T3 at the cell surface.
Nucleotide Sequence Analysis-The cDNA insert of pcDNAI-G T3 ST was sequenced by the dideoxy nucleotide chain termination method (Sanger et al., 1977) using oligonucleotide primers, as described (Bierhuizen et al., 1993). The sequencing was initially carried out by using DyeDeoxy terminator cycle sequencing kit and DNA autosequencer (Applied Biosystems, Foster City, CA). The sequence was confirmed by using [ 35 S]dATP and a Sequenase sequencing kit (Amersham Corp.).

Construction of a Truncated Form of G D3 /G T3 ST, G D3 /G T3 ST-S-
The cDNA encoding a truncated form of G D3 /G T3 ST was prepared by polymerase chain reaction (PCR) using pcDNAI-G T3 ST as a template. Upstream and downstream primers used were 5Ј-cccaagctt-GAGGGGCC-3Ј (HindIII site shown by underline) and 5Ј-atagtt-tagcgggcgcCCATTGTTCC-3Ј (NotI site shown by underline), respectively. The PCR product encompasses the sequence from nucleotide 38 to nucleotide 1,080. The nucleotide 38 is 8 nucleotides upstream from the second initiation methionine, and the nucleotide 1080 resides 9 nucleotides downstream from the stop codon. PCR was performed in a final volume of 100 l using the primers (0.5 pmol each) for 30 cycles of 94°C for 1 min, 50°C for 1 min, and 72°C for 2 min. The amplified DNA fragment was digested with HindIII and NotI and cloned into the same sites of pcDNAI.
Establishment of Stable Transfectants Expressing G T3 -Human cervical epitheloid carcinoma HeLa cells and human melanoma MeWo cells were cotransfected with pcDNAI-G D3 /G T3 ST and pSV2neo (10:1) and selected by G418 resistance. The transfected cells expressing G T3 were screened by immunofluorescent staining with M6704 antibody, and two clones, named HeLa⅐G T3 and MeWo⅐G T3 , were established.
Isolation of gangliosides from cells and TLC-immunostaining were performed as reported previously (Hirabayashi et al., 1988). The purified gangliosides were applied onto a plastic plate (Poligram Sil G, Nagel, Doren, Germany) and developed under the same conditions as described above. The plate was subjected to immunostaining with R24 or M6703 antibody, followed by peroxidase-conjugated goat anti-mouse IgG antibody (Cappel). The peroxidase activity was visualized with 4-chloro-1-naphthol/H 2 O 2 .
In Vitro Sialyltransferase Assays and Product Characterization-The expression vector for protein A-G D3 /G T3 ST fusion protein was constructed using pAMoA vector as described (Sasaki et al., 1994b). The cDNA in this pAMoA-GD3 was excised by SalI and Asp718, filled in by the Klenow fragment, and subcloned into the EcoRV site of pcDNAI, resulting in pcDNA-proA-G D3 . After confirming the correct orientation by sequencing, pcDNAI-proA-G D3 or pPROTA (Kukowska-Latallo et al., 1990) as a control was transfected into COS-1 cells. The protein A-G D3 / FIG. 1. Synthetic pathways of C series polysialogangliosides. ␣-2,8-sialic acid residues are shown in boldface type. The rest of the sialic acid residues are ␣-2,3-linked. In MeWo cells, the synthesis of G T1c and G Q1c does not take place, since ␤-1,4-N-acetylgalactosaminyltransferase is absent. G Q3 and STVI are newly proposed in the present study. STII and STIII were found to be the same enzyme in the present study, and STVI is probably the same enzyme as STII.
Sialyltransferase activity was measured as described previously (Sasaki et al., 1994b). Briefly, after a 1-min sonication of 25 l of 0.1 M sodium cacodylate buffer (pH 6.0) containing 20 mM MgCl 2 , 1% Triton CF-54, 2.4 nmol of CMP-[ 14 C]NeuNAc, and 10 g of a substrate with or without a competing substrate, 25 l of the enzyme solution was added and incubated for 4, 12, or 24 h at 37°C. At the end of the incubation period, 200 l of phosphate-buffered saline was added to the incubation mixture, and the contents were applied to an Aspec Pak tC18 cartridge (M & S, Tokyo, Japan), according to the procedure described (Williams and McCluer, 1980). After washing the column with water, glycosphingolipids were eluted with 3 ml of chloroform-methanol (2:1 by volume). The sample was dried under nitrogen stream and then subjected to chromatography using an HPTLC plate under the same conditions as described above. Radioactive materials were visualized by fluorography after spraying an autoradiography enhancer (DuPont NEN). Standard and acceptor gangliosides were visualized by the resorcinol/HCl method.
Quantitation of G D3 /G T3 ST Transcripts Using Competitive PCR-The level of G D3 /G T3 ST transcript was measured by the competitive PCR using the cDNAs, which were prepared by reverse transcription of total RNA, as detailed in the previous report (Sasaki et al., 1994a). For distinction of a target cDNA from its competitor DNA, G D3 /G T3 ST cDNA was truncated by deleting a 125-base pair EcoT22I-PvuII fragment of the cDNA from pUC-GD3R (Sasaki et al., 1994b). The 5Ј and 3Ј primers were 5Ј-ACAGTTACATCTACATGCCTGCCTT-3Ј and 5Ј-CATGAAA-CAACTTGACCATTCCCTC-3Ј, respectively. The amount of amplified cDNAs was calculated from the respective standard curves, converted into the values of molar numbers. As a control, the ␤-actin transcript was measured in the same cDNA samples.
Northern Blot Analysis of Various Human Tissues-Poly(A) ϩ RNA from human fetal (19 -23 gestational weeks) and adult brains purchased from Clontech (Palo Alto, CA) were electrophoresed in a 1.2% agarose gel containing 2.2 M formaldehyde and transferred to a nylon filter (Micron Separation, Westboro, MA). Human multiple-tissue Northern blots of poly(A) ϩ RNA were purchased from Clontech, and these blots were hybridized with a gel-purified cDNA insert of pcDNAI-G T3 ST after labeling with [␣-32 P]dCTP by random oligonucleotide priming (Feinberg and Vogelstein, 1983) (Prime-It II labeling kit, Stratagene, San Diego, CA).
Fluorescence in Situ Hybridization Analysis of G T3 ST Gene-Human genomic P1 plasmid library was screened by PCR as described (Onda and Fukuda, 1995). The 5Ј and 3Ј primers for PCR correspond to the sequence of the nucleotides 1184 -1203 and that of nucleotides 1424 -1443 of the G T3 ST sequence (see Sasaki et al., 1994b).
Purified DNA from one of the isolated P1 clones, clone 5459, was labeled with digoxigenin-dUTP by nick translation. Labeled probe was combined with sheared human DNA and hybridized to normal metaphase chromosomes derived from phytohemagglutinin-stimulated pe-ripheral blood lymphocytes in a solution containing 50% formamide, 10% dextran sulfate and 2 ϫ SSC. Specific hybridization signals were detected by incubating the hybridized slides in FITC-labeled antidigoxigenin antibody followed by counterstaining with propidium iodide for one color experiment. Probe detection for two-color experiments was accomplished by cohybridizing the slides with a biotin-labeled probe, D12Z1-specific for centromere of chromosome 12 and the digoxigeninlabeled clone 5459. After incubating these slides with Texas Red-labeled avidin and FITC-labeled anti-digoxigenin antibody, they were counterstained with 4Ј,6-diamidino-2-phenylindole (Rouquier et al., 1995).

RESULTS
Isolation of a cDNA Clone Encoding G T3 Synthase-In order to clone G T3 synthase, it was necessary to employ cells expressing a precursor ganglioside G D3 but lacked G T3 itself as recipients for transfection. The parent COS-1 cells did not react with M6703 (anti-G T3 ) or R24 (anti-G D3 ) monoclonal antibodies, indicating that G D3 and G T3 are not synthesized by COS-1 cells. Therefore, we transfected COS-1 cells with pAMo-G D3 , which harbors cDNAs encoding the G D3 synthase and G418 resistance gene (Sasaki et al., 1994b), and isolated COS-1⅐G D3 cells that were strongly stained by R24.
When the COS-1⅐G D3 cells were tested for the presence of G T3 by M6703 antibody, however, 3.5% of COS-1⅐G D3 cells showed a strong positive signal for G T3 judged in FACS analysis. We thus isolated COS-1⅐G D3 cells, which barely reacted with M6703 antibody by FACS. The freshly sorted COS-1⅐G D3 cells expressed only G D3 and were expanded once up to 1.2 ϫ 10 7 cells in culture. Although a few of them still expressed G T3 (Fig. 2C), they were used as recipient cells for expression cloning of G T3 synthase.
COS-1⅐G D3 cells were transfected with the SK-MEL-28 cDNA library in pcDNAI. Sixty-two h after transfection, COS-1⅐G D3 cells expressing G T3 were enriched by FACS using M6703 antibody under the conditions where only highly positive cells were selected. From 2.6 ϫ 10 6 COS-1⅐G D3 cells applied, 4,044 cells were sorted. Plasmid DNAs were rescued from these M6703-positive cells.
When COS-1⅐G D3 cells were transiently transfected with a mixture of the above isolated plasmids, it was not possible to distinguish the cells that newly became G T3 -positive from the cells that were endogenously G T3 -positive by immunofluorescent staining with M6703 antibody. We reasoned that this failure was due to the high background expression of G T3 in COS-1⅐G D3 cells (see Fig. 2C). In order to overcome this problem, the plasmid DNAs were transfected into HeLa cells. The wild-type HeLa cells expressed detectable amounts of G D3 as judged by immunofluorescent staining using another anti-G D3 antibody, KM641 (Ohta et al., 1993) but were completely neg- ative for G T3 (Fig. 2, G and I).
The transformed bacteria obtained after the Hirt procedure were thus divided into 20 pools, and the plasmid DNA from each plate was transfected separately into HeLa cells. The transfectants were screened by immunofluorescent staining using antibody M6703. Because of no background staining for M6703 in HeLa cells, we could identify two out of 20 plasmid pools that directed the expression of G T3 in HeLa cells. One of the plasmid pools, which produced strongly positive cell staining by M6703, was selected, and subsequent rounds of sibling selection with sequentially smaller, active pools identified a single plasmid, pcDNAI-G T3 ST, that directed the expression of G T3 at the cell surface of HeLa cells.
COS-1⅐G D3 and HeLa cells were transiently transfected with pcDNAI-G T3 ST, and the transfected cells were examined for the expression of G D3 and G T3 by immunofluorescent staining. Fig. 2, D and J, show that the expression of G T3 , detected by M6703, was clearly seen on both COS-1⅐G D3 and HeLa cells after the transfection. The expression of G D3 was also notably increased in some of the transfected HeLa cells (Fig. 2, F  and H).
Sequencing of the isolated cDNA in pcDNAI-G T3 ST revealed an insert of 1622 base pairs in size encoding a single open reading frame in the sense orientation with respect to the pcDNAI promoter. The open reading frame predicts a protein of 356 amino acids in length with a calculated molecular mass of 40,517. When this cDNA sequence was compared with other cDNAs in the data base, it was found to be identical to that encoding another ␣-2,8-sialyltransferase, G D3 synthase (Haraguchi et al., 1994;Nara et al., 1994;Sasaki et al., 1994b) (Fig.  3A). The newly isolated sequence is, however, shorter in the 5Ј-flanking sequence and starts with 11 base pairs upstream from the initiation methionine (Fig. 3A). The 3Ј-flanking sequence just before poly(A) was shorter by 3 base pairs compared with the reported G D3 synthase (Sasaki et al., 1994b). The deduced amino acid sequence predicts that this protein has a type II membrane topology, which has been found in all mammalian glycosyltransferases cloned (Schachter, 1994).
These results indicate that G D3 synthase (STII) and G T3 synthase (STIII) are the same enzyme and suggests a possibility that a single enzyme catalyzes the reactions for the formation of disialosyl and trisialosyl groups. The newly identified enzyme is thus called G D3 /G T3 synthase or G D3 /G T3 ST hereafter.
The above cDNA sequence shows a second initiation methionine, which resides 16 codons from the first initiation methionine. We synthesized the shorter cDNA encoding nucleotides 37-1,080 of the cDNA sequence by PCR, allowing the translation initiation only from the second initiation methionine (Fig.  3A). This truncated cDNA encoding G D3 /G T3 ST-S (Fig. 3B) was cloned into pcDNAI and expressed in both COS-1⅐G D3 and HeLa cells. The results obtained by immunofluorescent staining of the transfected cells clearly indicate that this truncated cDNA is also capable of G D3 and G T3 expression (data not shown). In fact, the nucleotide sequence surrounding the second initiation methionine, GCCATGG is consistent with the consensus sequence, (A/G)CCATGG for optimal translation initiation (Kozak, 1991). In contrast, the nucleotide sequence surrounding the first methionine GCGATGA does not conform with the Kozak sequence. Moreover, the size of the cytoplasmic sequence in this shorter translated product is reasonably short (12 residues), which is characteristic for all glycosyltransferases cloned to date (Schachter, 1994). These results suggest that actual initiation of translation starts probably from this downstream methionine coding for a protein of 341 amino acids with a molecular mass of 38,901.
Synthesis of G D3 , G T3 , and Possibly Higher Polysialogangliosides by G D3 /G T3 Synthase-In order to determine whether or not G D3 /G T3 ST catalyzes the formation of G D3 and G T3 , HeLa and MeWo cells were transfected with pcDNAI-G D3 /G T3 ST, and the resultant transfected cells, termed HeLa⅐G T3 and MeWo⅐G T3 , were analyzed for the synthesis of gangliosides. The results shown in Fig. 4, C and D, indicate that the HeLa⅐G T3 cells synthesized not only G T3 but also G D3 , although the expression of G T3 was stronger. Similarly, the MeWo⅐G T3 cells synthesized a significant amount of G T3 together with G D3 (Fig. 4, G and H). Since the parent MeWo cells barely synthe- size G T3 even though a substantial amount of G D3 is synthesized (Fig. 4, E and F), the expression of G T3 in the MeWo⅐G T3 cells should be solely due to the newly introduced G D3 /G T3 ST cDNA, but not the accumulation of newly synthesized G D3 . On the other hand, the enhanced expression of G D3 and the new synthesis of G T3 in the HeLa⅐G T3 cells were due to the newly introduced G D3 /G T3 ST cDNA.
In order to confirm that the transfected cells synthesize both G D3 and G T3 , gangliosides were isolated from the parent HeLa and MeWo cells and their stable transfectants. The thin-layer chromatogram of the gangliosides, detected by resorcinol reaction, which reacts with sialic acid, showed that the HeLa⅐G T3 cells contained both G D3 and G T3 , while the parent HeLa cells contained no G T3 (Fig. 4I, lanes 1 and 2). Similarly, the MeWo⅐G T3 cells contained G T3 , whereas the parent MeWo cells did not contain G T3 (see lanes 3 and 4 in Fig. 4I). These results were confirmed by immunostaining of gangliosides after separation by thin-layer chromatography. The parent HeLa cells express a very small amount of G D3 , but HeLa⅐G T3 cells express a substantially increased amount of G D3 , detected by R24 antibody (Fig. 4J, lanes 1 and 2). The parent MeWo cells, on the other hand, express a significant amount of G D3 (Fig. 4J, lane  3), but no G T3 was detected by M6703 antibody (Fig. 4K, lane  3). In contrast, both HeLa⅐G T3 and MeWo⅐G T3 cells express a large amount of G T3 (Fig. 4K, lanes 2 and 4). These results, taken together, clearly indicate that G D3 /G T3 ST transfers an ␣-2,8-linked sialic acid to G M3 and G D3 , forming G D3 and G T3 , respectively. The immunofluorescent stainings of HeLa⅐G T3 and MeWo⅐G T3 cells by M6703 were completely abolished by pretreatment of chloroform-methanol (2:1) extraction (data not shown), indicating that all of the newly formed trisialosyl groups are attached to glycosphingolipids. If some of them were attached to glycoproteins, some staining should remain because M6703 also reacts with trisialosyl residues in glycoproteins (Nakayama et al., 1993).
In order to formally prove if G D3 synthase also has G T3 synthase activity, a putative catalytic domain of this protein was expressed as a protein fused with the IgG-binding domain of Staphylococcus aureus protein A preceded by a signal peptide sequence (Sasaki et al., 1993) (see ProA-G D3 /G T3 ST in Fig.  3B). The cDNA encoding this chimeric protein was cloned into pcDNAI and expressed in COS-1 cells. The fusion protein secreted into the culture medium was absorbed to IgG-Sepharose and then incubated with G M3 or G D3 and the donor substrate CMP-[ 14 C]NeuNAc. As shown in Fig. 5, lane 3, the soluble form of G D3 /G T3 ST synthesized both G D3 and G T3 when incubated with G M3 . These results establish that the newly identified enzyme, G D3 /G T3 ST, is a polysialyltransferase that adds more than one sialic acid residue in ␣-2,8-linkage.
The above experiments also suggested that G D3 /G T3 ST added sialic acids much less efficiently when G D3 was used as an acceptor (Fig. 5, lane 5). However, the enzyme added sialic acid residues to G D3 (136 M final concentration) after a longer period of incubation (Fig. 5, lanes 7 and 8). Under these conditions, G D3 /G T3 ST also synthesized higher polysialogangliosides, which presumably have more than three sialic acid res- idues, as shown in lanes 7 and 8 of Fig. 5.
We also tested if the product or an intermediate inhibits the enzymatic reaction as competing substrates. The results shown in Fig. 5 indicate that G D3 (680 M final concentration) inhibits the formation of G D3 and G T3 from G M3 (Fig. 5, lane 10), while G T3 (567 M final concentration) inhibits the formation of G T3 and higher polysialogangliosides from G D3 (Fig. 5, lane 11). These results taken together support the above conclusion that G D3 /G T3 ST synthesizes G D3 from G M3 and then G D3 /G T3 ST utilizes G D3 as an acceptor to form G T3 .
The above results also suggested that the amount of G D3 / G T3 ST mRNA transcript may be proportional to the amount of polysialylated gangliosides synthesized. In order to test this hypothesis, G D3 /G T3 ST transcript was quantitated in the parent HeLa, HeLa⅐G T3 , parent MeWo, and MeWo⅐G T3 cells. Fig.  4L shows that the HeLa⅐G T3 cells express a significant amount of the G D3 /G T3 ST transcript (330 fg), while the parent HeLa cells scarcely express it. The MeWo⅐G T3 cells express approximately the same amount (370 fg) of the transcript as the HeLa⅐G T3 cells and about 15 times more than that in the parent MeWo cells (23 fg). As shown in Fig. 4, I, J, and K, the parent MeWo cells express G D3 but barely express G T3 , while the MeWo⅐G T3 and HeLa⅐G T3 cells express both G D3 and G T3 . These results clearly indicate that G T3 is synthesized only when G D3 / G T3 ST is abundantly present.
G D3 /G T3 ST Is Expressed in both Fetal and Adult Brains-To determine the tissue distribution of G D3 /G T3 ST mRNA, Northern blots of poly(A) ϩ RNA derived from various human tissues were examined. As shown in Fig. 6, a band of 2.3 kb was detected in the poly(A) ϩ RNA isolated from the fetal and adult brains. The transcript was also detected in fetal lung. In adult tissues, the G D3 /G T3 ST transcripts were detected in brain and very weakly in lung. Among different parts of the adult brain, a substantial amount of G D3 /G T3 ST mRNA was detected invariably in different parts of brain (Fig. 6, right side). In some regions, a band of 9.5 kb was also detected. These two different sizes of the transcript might be produced due to the alternate usage of polyadenylation sites. The expression pattern of G D3 / G T3 ST is different from that of the neural cell adhesion molecule (N-CAM)-specific polysialyltransferase (Nakayama et al., 1995), and G D3 /G T3 ST expression is more restricted to brain.
G D3 /G T3 ST Gene Is Mapped to Chromosome 12p12-The previous studies showed that the G D3 ST gene is present in chromosome 12, but no precise chromosomal location of this gene was reported (Sasaki et al., 1994b).
In order to localize precisely the G D3 /G T3 ST gene, we utilized fluorescence in situ hybridization (FISH) procedures. First, P1 plasmid harboring G D3 /G T3 ST gene, named clone 5459 was isolated, and genomic DNA was prepared from this P1 clone. Using this genomic DNA as a probe, the initial experiment resulted in specific labeling of the short arm of a group C chromosome. A second experiment was conducted in which a biotin-labeled probe (D12Z1) specific for the centromere of chromosome 12 was cohybridized with the digoxigenin-labeled clone 5459. This experiment resulted in the specific labeling of the centromere of chromosome 12 in red and the short arm of the same chromosome in green (Fig. 7A). Measurements of 10 specifically hybridized chromosome 12 demonstrated that the clone 5459 is located at a position that is 43% of the distance from the centromere to the telomere of short arm, an area that corresponds to 12p12 (Fig. 7B). This location is close to that of the c-Ki-ras protooncogene (Muleris et al., 1993) and human islet amyloid polypeptide gene (Christmanson et al., 1990). DISCUSSION The present study describes the isolation of a cDNA clone encoding G T3 synthase, the key enzyme responsible for C series polysialogangliosides using expression cloning with a newly devised modification. We utilized COS-1 cells as recipient cells for the enrichment of plasmids that directed the expression of G T3 . Since the large T antigen is synthesized in COS-1 cells, plasmids such as pcDNAI that contain the replication origin of SV40 are amplified in the cells. In order to test whether or not plasmids isolated from sorted COS-1⅐G D3 cells could convert G D3 to G T3 , COS-1⅐G D3 cells were not suited as recipient cells because of a substantial background of G T3 expression. In contrast, HeLa cells do not express G T3 but synthesize a small amount of G D3 . Therefore HeLa cells should be able to synthesize G T3 once a cDNA encoding G T3 synthase is introduced. By using HeLa cells, we thus could identify a pool of plasmids that directed G T3 expression. This is the first report where recipient cells for enriching plasmids differ from cells used for testing the enriched plasmids that direct the expression of a desired gene.
It was also noted that the stable transfectants of both HeLa and MeWo cells acquired gangliosides larger than G T3 (see the arrowhead in Fig. 4K). Since MeWo cells lack ␤-1,4-N-acetylgalactosaminyltransferase (Yamashiro et al., 1995), G T1c , or G Q1c can not be formed in MeWo⅐G T3 cells (see Fig. 1 for the structure of G T1c and G Q1c ). In addition, M6703 recognizes not only trisialosyl residues but also tetrasialosyl residues. 2 When G D3 was incubated with the soluble G D3 /G T3 ST, slow migrating gangliosides were also produced in addition to G T3 (Fig. 5, lanes  7 and 8). Since this assay was performed in the absence of other enzymes, these higher gangliosides are most likely polysialogangliosides having more than three sialic acid residues. The results, taken together, strongly suggest that the slow migrating band may represent G Q3 shown in Fig. 1. It is possible that G Q3 is present as a very minor component so that it has escaped attention. Further studies are necessary to confirm the presence of this glycosphingolipid.
The present study indicates that the same enzyme is apparently capable of adding all of the ␣-2,8-linked sialic acid, forming disialosyl, trisialosyl, and possibly tetrasialosyl residues in gangliosides. This finding is very similar to those reported for polysialylation of N-CAM. We and others have recently cloned a polysialyltransferase, which is responsible for polysialylation of N-CAM (Eckhardt et al., 1995;Nakayama et al., 1995). Although it has been suggested that the first disialosyl linkage is separately formed by an initiation enzyme (see Kitazume et al., 1994), the results obtained on mutant Chinese hamster ovary cells lacking polysialylation strongly suggest that polysialyltransferase carries out all reactions that form ␣-2,8linked sialic acid polymer in N-CAM (Eckhardt et al., 1995). Moreover, it has been demonstrated recently that polysialyltransferase can add all ␣-2,8-linked sialic acid residues necessary for polysialic acid formation (Nakayama and Fukuda, 1996). These studies, taken together, strongly suggest that the polysialylation is catalyzed by single enzymes in both N-CAM glycoprotein (polysialyltransferase) and glycosphingolipids (G D3 /G T3 ST).
The present study demonstrated that G D3 /G T3 ST synthesizes G T3 more efficiently from G M3 than from G D3 . It is tempting to speculate that G D3 /G T3 ST first binds to G M3 and then continuously adds sialic acid residues until the binding of the enzyme to the product is weakened. In fact, the excess amount of G D3 inhibited the formation of G T3 from G M3 , confirming that G D3 is an intermediate in the polysialylation. Once the enzyme is released from the enzyme-acceptor complex, it is likely that the enzyme has much less affinity with the product, which would be an acceptor for another reaction. Similarly, N-CAMspecific polysialyltransferase was shown to scarcely add a sialic acid on the polymerized sialic acid residues such as colomic acid (McCoy et al., 1985). These results suggest that termination of polysialylation takes place when the enzyme no longer binds to 2 Y. Hirabayashi, unpublished results. FIG. 7. Chromosomal localization of G D3 /G T3 ST gene as revealed by FISH. A, two-color FISH analysis on metaphase chromosomes using a digoxigenin-labeled P1 clone 5439 encoding G T3 ST (stained as green signal) and biotin-labeled D12Z1, a specific probe for the centromere of chromosome 12 (stained as red signal). A discrete signal is discernible on all chromosome 12 chromatids for each probe. DNA was counterstained with 4Ј,6-diamidino-2-phenylindole. B, the specific hybridization occurs on chromosome 12, region p12. the polysialylated acceptor. It is apparent that G D3 /G T3 ST terminates its reaction early, most likely due to its inefficiency in binding to trisialosyl or tetrasialosyl residues.
It was shown recently that Neuro2a cells exhibited better neurite extension after the cells were stably transfected to express G D3 /G T3 ST (Kojima et al., 1994). Although these authors thought that this effect was due to the synthesis of G D3 and B series gangliosides, the effect may be due to the synthesis of G T3 and C series polysialogangliosides in the transfected Neuro2a cells. We have shown recently that neurite outgrowth in substratum cells is enhanced by the presence of polysialic acid in N-CAM (Nakayama et al., 1995). Previous studies showed that the presence of polysialic acid not only exhibits antiadhesive properties on homophilic N-CAM interaction but also influences cell-cell interactions carried by other cell surface receptors (Edelman, 1985;Rutishauser et al., 1988;Jessell et al., 1990). Further studies are thus needed to determine if the presence of polysialylated gangliosides influence the cellcell interaction carried by other adhesive molecules.
Together with previously cloned polysialyltransferase cDNA that forms polysialic acid in glycoproteins, the cDNA cloned in the present study will be a powerful tool to dissect the intricate and complex roles of polysialic acids attached to glycoproteins and glycosphingolipids in cell-cell interactions during development.