The minimal structural domains required for neural cell adhesion molecule polysialylation by PST/ST8Sia IV and STX/ST8Sia II.

A limited number of mammalian proteins are modified by polysialic acid, with the neural cell adhesion molecule (NCAM) being the most abundant of these. We hypothesize that polysialylation is a protein-specific glycosylation event and that an initial protein-protein interaction between polysialyltransferases and glycoprotein substrates mediates this specificity. To evaluate the regions of NCAM required for recognition and polysialylation by PST/ST8Sia IV and STX/ST8Sia II, a series of domain deletion proteins were generated, co-expressed with each enzyme, and their polysialylation analyzed. A protein consisting of the fifth immunoglobulin-like domain (Ig5), which contains the reported sites of polysialylation, and the first fibronectin type III repeat (FN1) was polysialylated by both enzymes, whereas a protein consisting of Ig5 alone was not polysialylated by either enzyme. This demonstrates that the Ig5 domain of NCAM and FN1 are sufficient for polysialylation, and suggests that the FN1 may constitute an enzyme recognition and docking site. Two other NCAM mutants, NCAM-6 (Ig1-5) and NCAM-7 (FN1-FN2), were weakly polysialylated by PST/ST8Sia IV, suggesting that a weaker enzyme recognition site may exist within the Ig domains, and that glycans in the FN region are polysialylated. Further analysis indicated that O-linked oligosaccharides in NCAM-7, and O-linked and N-linked glycans in full-length NCAM, are polysialylated when these proteins are co-expressed with the polysialyltransferases in COS-1 cells. Our data support a model in which the polysialyltransferases bind to the FN1 of NCAM to polymerize polysialic acid chains on appropriately presented glycans in adjacent regions.

␣2,8-Polysialic acid chains are found in neuroinvasive bacteria, fish, sea urchin eggs, and mammals where they play a variety of biological roles (1)(2)(3)(4). Although this unique carbohydrate structure is found in variable forms, in mammals polysialic acid chains are linear polymers of 5-N-acetylneuraminic acid (Neu5Ac) linked by ␣2,8-glycosidic bonds (3,5) and can extend to lengths of 50 -100 units (6,7). In mammalian cells, the expression of polysialic acid is developmentally regulated and is found at highest levels in the embryonic and neonatal brain attached to the neural cell adhesion molecule (NCAM) 1 (8). Polysialic acid is also expressed in other embryonic and neonatal tissues such as heart and kidney (8,9). In the adult animal, the levels of polysialic acid decrease, but are maintained in selected areas of the brain, such as the hippocampus and olfactory system (10 -12). Polysialic acid is also re-expressed on the surface of some very highly metastatic cancer cells such as small cell lung carcinoma (13,14), Wilms' tumor (15), and neuroblastomas (7,16,17). For these cancer cells, polysialic acid is believed to play an anti-adhesive role and promote cell growth and metastasis (18,19).
The role of polysialic acid as an anti-adhesive molecule has been studied most thoroughly in the central nervous system. Both in vivo and in vitro studies have shown that polysialic acid promotes plasticity in cell interactions and is critical for axon guidance and pathfinding, neurite outgrowth, and general cell migration (20 -24). Investigators suspected that polysialic acid would function primarily during development because of its high level expression at that time (25). Surprisingly, mice null for either NCAM or the polysialyltransferase PST/ST8Sia IV do not have developmental defects (25,26). Instead, the phenotypes of these null animals suggest that polysialylated NCAM and polysialic acid itself play more important roles in maintaining plasticity in specific areas of the adult nervous system (12,27). For example, studies done using mice null for either NCAM or PST/ST8Sia IV suggest that polysialic acid and NCAM promote long term potentiation, and as such may play an important role in spatial learning (25,26). Other functions for polysialylated NCAM have been proposed including those in which polysialylated NCAM modulates cellular signal transduction by organizing cell surface receptors and sequestering ligands (for a discussion see Ref. 23).
Polysialic acid chains are synthesized by the polysialyltransferases, PST/ST8Sia IV and STX/ST8Sia II (these enzymes will heretofore be called PST and STX, respectively), and the major substrate for the polysialyltransferases is NCAM (28 -34). Studies done by a number of groups have shown that both PST and STX can act alone to polysialylate NCAM (for examples Refs. 29 and 35). However, the enzymes have also been shown to differ in their preferences for oligosaccharide acceptors in the Ig5 domain of NCAM and the length of chains synthesized (36,37). Additional work by Fukuda and colleagues (37,38) suggests that the two polysialyltransferases work cooperatively to polysialylate NCAM and that they show differences in their abilities to polysialylate substrates of varying chain lengths. Early work by Kojima et al. (30) showed that while STX could polysialylate a variety of glycoprotein substrates in vitro, NCAM served as a 1500-fold better acceptor for polysialic acid than another glycoprotein, fetuin. More recent work by Angata et al. (39) compared the substrate specificity of PST and STX. They showed that while these enzymes could add sialic acid to free oligosaccharide acceptors, the polysialylation of oligosaccharides attached to NCAM was much more efficient. Taken together, these in vitro data suggest that the polypeptide chain of NCAM contributes to the specificity of polysialylation.
The limited number of polysialylated proteins identified to date suggests that polysialylation is a protein-specific event.
Only four other proteins besides NCAM have been found to be polysialylated in mammalian cells. These include the ␣ subunit of the voltage-dependent sodium channel (40), a form of the CD36 scavenger receptor found in milk (41), and the polysialyltransferases themselves, which appear to autopolysialylate their own oligosaccharides (42,43). Other data from our laboratory also support the idea that polysialylation is proteinspecific (44,45). Using mutagenesis, we identified the oligosaccharides on each polysialyltransferase that are polysialylated, and generated catalytically active, non-autopolysialylated forms of the enzymes, PST Mut 2.3 and STX Mut 2.4.5 (44,45). Following expression of PST Mut 2.3 and STX Mut 2.4.5 alone in COS-1 cells we detected no polysialic acid using the antipolysialic acid antibody OL.28 (44,45). Only after NCAM was co-expressed with these mutant enzymes did the cells express polysialic acid. These results indicate that the endogenous glycoproteins expressed by COS-1 cells are not substrates for the polysialyltransferases PST and STX, and further support the notion that polysialylation is a protein-specific event.
A few other carbohydrate addition/modification processes have been shown to be protein-specific including the biosynthesis of the mannose 6-phosphate recognition marker on lysosomal enzymes, which is required for enzyme trafficking to the lysosomes (46,47), the biosynthesis of the terminal GalNAc-4-SO 4 structure on the N-linked oligosaccharides of pituitary glycoprotein hormones, which is required for their clearance from the circulation (48), the biosynthesis of Glc 1 Man 9 GlcNAc 2 oligosaccharides on misfolded/unfolded glycoproteins in the endoplasmic reticulum, which mediates their interactions with the chaperones calnexin and calreticulin (49,50), and the addition of O-linked fucose to epidermal growth factor repeats in proteins such as Notch, which modulates Notch signaling (51). In each case, the initial step in the biosynthetic pathway is believed to involve protein-protein contact between enzyme and substrate. This has been directly shown for two of these modification processes. During mannose 6-phosphate biosynthesis, the first enzyme in the two-step pathway, the N-acetylglucosamine-1-phosphotransferase, recognizes a signal patch containing a specific Lys residue on the lysosomal enzyme to be modified (52,53), whereas during the biosynthesis of GalNAc-4-SO 4 , the N-acetylgalactosaminyltransferase recognizes an ␣-helical stretch of basic amino acids 6 -9 amino acids NH 2terminal to the N-linked oligosaccharide to be modified (54,55). In both cases, disruption of the site of interaction leads to an abrogation of the modification.
Based on these observations, we hypothesize that ␣2,8-polysialylation is a protein-specific event, and that the basis for this specificity is an initial protein-protein contact between the polysialyltransferase and the glycoprotein substrate. To test this hypothesis we have chosen to evaluate the sequence re-quirements for NCAM polysialylation by both PST and STX. In this study, we have constructed a series of NCAM domain deletion mutants and evaluated their polysialylation by each polysialyltransferase. We found that a minimal NCAM protein consisting of the fifth immunoglobulin-like domain (Ig5), which possesses the N-linked oligosaccharides that are polysialylated in the wild type protein, and the adjacent fibronectin type III repeat (FN) are sufficient for polysialylation by each enzyme. Surprisingly, we also found that a NCAM protein consisting of only the two FN regions was polysialylated and that the polysialic acid is modifying O-linked rather than N-linked glycans. These data suggest that the first fibronectin type III repeat (FN1) of NCAM may act as an initial recognition and docking site for the polysialyltransferases, and allow us to formulate a model of protein-specific polysialylation.  35 S-Express protein labeling mixture was purchased from PerkinElmer Life Sciences. Fluorescein isothiocyanate-conjugated and horseradish peroxidase-conjugated goat anti-mouse antibodies were purchased from Jackson Laboratories (West Grove, PA). Other chemicals and reagents were obtained from Sigma and Fisher Scientific (Hanover Park, IL).
Construction of mNCAM-7 N702Q Glycosylation Mutant-The consensus glycosylation site at Asn-702 in mNCAM-7-pcDNA 3.1/V5-His B was mutated to Gln using the QuikChange TM site-directed mutagenesis system per the manufacturer's protocol (Stratagene). The primers used for the mutagenesis were 5Ј-CCATCCCAGCCCAAGGCAGCCCC-3Ј and 5Ј-GGGGCTGCCTTGGGCTGGGATGG-3Ј. The mutation was confirmed by sequencing as described above.
Transfection of COS-1 Cells with NCAM cDNAs-COS-1 cells maintained in DMEM, 10% FBS were plated on 100-mm tissue culture plates or 12-mm glass coverslips and grown in a 37°C, 5% CO 2 incubator until 50 -70% confluent. Lipofectin transfections were performed according to the protocols provided by Invitrogen Corp. Thirty microliters of Lipofectin and 20 g of NCAM plasmid DNA in 3 ml of Opti-MEM I were used for transfection of each 100-mm tissue culture plate. Three microliters of Lipofectin and 0.5 g of NCAM plasmid DNA in 300 l of Opti-MEM I were used for transfection of each coverslip.
Metabolic Labeling of Cells and Immunoprecipitation of NCAM Proteins-Following transfection of COS-1 cells with soluble or membranebound NCAM cDNAs and expression of these proteins in the cells for 18 h, 100-mm tissue culture dishes of transfected cells were incubated with cysteine/methionine-free DMEM for 1 h. After incubation, this medium was replaced with 3.5 ml of fresh cysteine/methionine-free DMEM containing 100 Ci/ml 35 S-Express protein labeling mixture (PerkinElmer Life Sciences). Cells were incubated with the radiolabel for 1 h at 37°C in a 5% CO 2 incubator. After labeling, medium was removed and cells were washed, the labeled proteins were chased for either 0 (mNCAMs only) or 3 h (sNCAMs only) with 4 ml of unlabeled DMEM, 10% FBS. For the mNCAM proteins, cells were washed with 10 ml of PBS and lysed in 1 ml of immunoprecipitation buffer 2 (50 mM Tris-HCl, pH 7.5, 150 mM NaCl, 5 mM EDTA, 0.5% Nonidet P-40, 0.1% SDS). For the sNCAM proteins, following 3 h of chase, the cell medium was collected, debris removed by centrifugation, and the supernatant frozen overnight at Ϫ20°C.
The NCAM proteins were immunoprecipitated from the cell lysates and 3-h chase media using 2 g of anti-V5 epitope tag antibody and protein A-Sepharose (Amersham Biosciences), as previously described (57). The immunoprecipitation beads were resuspended in 50 l of Laemmli sample buffer containing 5% ␤-mercaptoethanol, heated for 10 min at 65°C, and directly loaded into the gel wells. Immunoprecipitated proteins were separated on 7.5% separating, 3% stacking SDSpolyacrylamide gels (58). Radiolabeled proteins were visualized by fluorography using 10% 2,5-diphenyloxalzole in dimethyl sulfoxide (59), and gels were exposed to Kodak Bio-Max MR film at Ϫ80°C.
Precipitation and Immunoblot Analysis of NCAM Proteins from COS-1 Cells Co-expressing PST and STX Proteins-COS-1 cells plated on 100-mm tissue culture plates were co-transfected with untagged PST or STX and sNCAM or mNCAM plasmid DNAs at a ratio of 4:1, respectively (20:5 g). Following transfection, 3 ml of transfection mixture was removed and 4 ml of DMEM, 10% FBS was added to the cells. After incubation for 18 h at 37°C in a 5% CO 2 incubator, cell lysates (mNCAMs only) and cell media (sNCAMs only) were collected. Cells were washed with 10 ml of PBS and lysed in 1 ml of immunoprecipitation buffer 2 (50 mM Tris-HCl, pH 7.5, 150 mM NaCl, 5 mM EDTA, 0.5% Nonidet P-40, 0.1% SDS), debris was removed from the cell medium by centrifugation, and the samples were frozen overnight at Ϫ20°C.
The NCAM proteins were immunoprecipitated from the cell lysates and media using 2 g of anti-V5 epitope tag antibody and 50 l of a 50% slurry of protein A-Sepharose as described above. The precipitated proteins were separated on 5% separating, 3% stacking SDS-polyacryl-amide gels after the addition of 50 l of Laemmli sample buffer containing 5% ␤-mercaptoethanol and incubation of the beads at 65°C for 5 min. Following electrophoresis, proteins were electrophoretically transferred to nitrocellulose membranes overnight at 500 mA. The membranes were processed for immunoblotting according to the manufacturer's protocol (Pierce). Anti-polysialic acid antibody, OL.28 (IgM), was diluted 1:500 in blocking buffer 1 (2% dry milk in Tris-buffered saline, pH 8.0). Horseradish peroxidase-conjugated secondary antibody, goat anti-mouse IgM, was diluted 1:8000 in blocking buffer 2 (5% dry milk in Tris-buffered saline, pH 8.0, 0.1% Tween 20). Immunoblots were developed using the SuperSignal West Pico chemiluminescence kit (Pierce) and exposed to Kodak Bio-Max MR film at room temperature.
Glycosidase Digestion of NCAM Proteins-NCAM proteins were immunoprecipitated from lysates of COS-1 cells following their co-expression with untagged PST. Following the final immunoprecipitation wash, the immune complexes bound to protein A-Sepharose beads were treated with glycosidase for 18 h with shaking at 37°C. Briefly, 3 l (1500 units) of PNGase F (New England BioLabs) was added to 77 l of dH 2 O, 10 l of 10% Nonidet P-40, and 10 l of 10ϫ reaction buffer (0.5 M sodium phosphate, pH 7.5). PNGase F cleaves between the innermost N-acetylglucosamine and asparagine residues of high mannose, hybrid, and complex N-linked oligosaccharides from glycoproteins, but does not cleave O-linked oligosaccharides (60). Digested samples were electrophoresed on a 5% separating, 3% stacking SDS-polyacrylamide gel after addition of 50 l of Laemmli buffer (58) and ␤-mercaptoethanol to 5% final concentration. Protein polysialylation was detected by immunoblotting using the OL.28 antibody as described above.
Immunofluorescence Localization of mNCAM Proteins-COS-1 cells were plated on glass coverslips, transfected with various mNCAM cDNAs, and processed for immunofluorescence microscopy as described previously (61). Briefly, cells were treated with either Ϫ20°C methanol to visualize internal staining or 3% paraformalydehyde to visualize cell surface staining. Anti-V5 epitope tag antibody was diluted 1:100 and the fluorescein isothiocyanate-conjugated goat anti-mouse IgG secondary antibody was diluted 1:200 in 5% normal goat serum/PBS blocking buffer prior to use. Coverslips were mounted on glass slides using 20 l of mounting medium (15% (w/v) Vinol 205 polyvinyl alcohol, 33% (v/v) glycerol, 0.1% azide in PBS, pH 8.5). Cells were visualized and photographed using a Nikon Axiophot microscope equipped with epifluorescence illumination and a ϫ60 oil immersion Plan Apochromat objective. Pictures were taken using an attached Spot RT Color digital camera and processed using Spot RT Software v.3.5.1 (Diagnostic Instruments, Inc., Sterling Heights, MI).
Indirect Immunofluorescence Microscopy of COS-1 Cells Co-expressing mNCAM and Non-autopolysialylated forms of PST and STX-COS-1 cells maintained in DMEM, 10% FBS were plated on 12-mm glass coverslips and grown in a 37°C, 5% CO 2 incubator until 50 -70% confluent and essentially transfected as described previously (44,45). For each coverslip, the ratio of polysialyltransferase to mNCAM plasmid DNA transfected was 1:1 (0.5:0.5 g). Here, catalytically active, mutant forms of the polysialyltransferases (PST Mut 2.3 (45) and STX Mut 2.4.5 (44)) that lack the oligosaccharides that are autopolysialylated were used so that only polysialic acid on the NCAM proteins would be visualized. Cells were fixed and processed for immunofluorescence microscopy as described before. Both primary antibodies, OL.28 anti-polysialic acid antibody and anti-mouse NCAM IgG, and secondary antibodies, fluorescein isothiocyanate-conjugated goat anti-mouse IgG and goat anti-mouse IgM, were diluted 1:200 in blocking buffer (5% normal goat serum in PBS) prior to use.

Expression and Localization of Membrane-associated and Soluble NCAM Domain Deletion Mutants-Previous studies by
Nelson et al. (62) suggested that the minimal NCAM extracellular domain structure required for polysialylation included Ig4-Ig5-FN1 (our mNCAM-2) and they also suggested that membrane association may be critical to position the NCAM correctly relative to the membrane-associated polysialyltransferases. These investigators expressed their truncated NCAM proteins in F11 rat/mouse hybrid cells that were capable of polysialylating the full-length NCAM (FL-NCAM) protein. At that time, the polysialyltransferase enzymes had not yet been cloned and characterized. In the present study we wanted to evaluate the sequence requirements for the polysialylation of NCAM by PST and STX individually, and to test the necessity of membrane association of the NCAM substrate in polysialylation. To do this we constructed eight NCAM domain deletion mutants using human NCAM-140 as a template (Fig. 1). Each protein was made as a membrane-associated form containing the transmembrane region and cytoplasmic tail (mNCAM proteins), and as a soluble form lacking these regions (sNCAM proteins). All proteins were cloned into the pcDNA3.1/V5-His B vector, which places a V5 epitope tag and a His 6 epitope tag at the carboxyl terminus of the protein.
Before analyzing the polysialylation of FL-NCAM and the domain deletion mutants (mNCAM 1-5 and sNCAM 1-5), we analyzed their expression and localization. The relative levels of expression of both the membrane-associated and soluble NCAM domain deletion mutants were evaluated following transient expression in COS-1 cells. Expressing cells were labeled for 1 h with 100 Ci/ml Expre 35 S 35 S protein labeling mixture (PerkinElmer Life Sciences) and chased for either 0 (mNCAM proteins) or 3 h (sNCAM proteins) with unlabeled media. Labeled mNCAM proteins were recovered from cell lysates and sNCAM proteins were recovered from cell media by immunoprecipitation with the anti-V5 epitope tag antibody and the immunoprecipitated proteins were separated by SDS-PAGE ( Fig. 2A). We found that all proteins, with the exception of sNCAM-5, were expressed at comparable levels. We also noticed that the membrane-associated proteins all migrated as doublets. This was especially noticeable for the mNCAM-5 protein. We suspected that the upper band of each doublet possessed more completely processed and modified N-linked glycans than the lower band. Treatment of the immunoprecipitated mNCAM-5 with PNGase F and Endo H demonstrated that the protein migrating with a higher molecular mass possessed complex type N-linked glycans (PNGase F sensitive, Endo H insensitive), whereas the protein migrating with a lower molecular mass possessed high mannose N-linked glycans (Endo H and PNGase F sensitive), confirming our earlier prediction (data not shown).
Next we evaluated the localization of the mNCAM proteins as a measure of their correct folding. FL-NCAM and the five mNCAM proteins were transiently expressed in COS-1 cells and processed for immunofluorescence microscopy with the anti-V5 epitope tag antibody, as described under "Experimental Procedures." We found that like the FL-NCAM, all of the membrane-associated domain deletion mutants were localized at the cell surface (Fig. 2B). In permeablized cells shown in these micrographs, Golgi staining is also apparent and reflects the movement of these proteins through this concentrated region of the cell. These results indicated that the membraneassociated NCAM domain deletion mutants were folding into forms that were readily transported of the endoplasmic reticulum, through the Golgi, and to the plasma membrane.

NCAM Domain Deletion Mutants Reveal an Important Role for FN1 in Recognition and Polysialylation by PST and STX-
The polysialylation of the NCAM domain deletion mutants by PST and STX was evaluated in two separate assays. Our first assay was an immunofluorescence assay employing the antipolysialic acid antibody, OL.28, and non-autopolysialylated forms of the two polysialyltransferases. In this assay, we transiently expressed FL-NCAM and the five membrane-associated domain deletion mutants (mNCAM 1-5) with catalytically active, non-autopolysialylated forms of PST and STX (PST Mut 2.3 and STX Mut 2.4.5) in COS-1 cells. Expressing cells were then fixed with 3% paraformaldehyde for surface staining or fixed and permeabilized with Ϫ20°C methanol for internal staining with the OL.28 anti-polysialic acid antibody and a fluorescein isothiocyanate-conjugated goat anti-mouse IgM secondary antibody, as described under "Experimental Procedures." Any fluorescence staining we observe must be because of the polysialylation of the NCAM proteins by the two polysialyltransferases and not enzyme autopolysialylation because the PST Mut 2.3 and STX Mut 2.4.5 proteins, while lacking the oligosaccharides that we have previously shown to be modified with polysialic acid, remain catalytically active (44,45). This analysis revealed that FL-NCAM and mNCAM-1, -2, -3, and -4 were all polysialylated by both PST Mut 2.3 and STX Mut 2.4.5 to similar levels. Here we show the data for FL-NCAM, mN-CAM-2, and mNCAM-4 (Fig. 3). In contrast, NCAM-5 was not polysialylated by either enzyme (Fig. 3, mNCAM-5). Because mNCAM-4, with an extracellular portion consisting of Ig5 and FN1, is polysialylated, whereas mNCAM-5, with an extracellular portion consisting of Ig5 alone, is not, we conclude that FN1 of NCAM is critical for its recognition and polysialylation by both PST and STX.
Our second assay confirmed the results we obtained above and showed that membrane association of the NCAM proteins is not necessary for polysialylation. In this assay, we co-expressed untagged, wild type PST or STX with V5-tagged membrane-associated and soluble forms of NCAM and the NCAM domain deletion mutants. NCAM proteins were recovered by immunoprecipitation using the anti-V5 epitope tag antibody, and immunoprecipitated proteins were subjected to SDS-PAGE and immunoblotting with the OL.28 anti-polysialic acid antibody. Fig. 4 shows that NCAM (FL-NCAM and sNCAM) and membrane-associated and soluble forms of domain deletion mutants 1-4 were polysialylated by PST and STX. In contrast, NCAM-5 proteins were not polysialylated by either enzyme. Whereas sNCAM-5 was not well expressed, its membraneassociated form, mNCAM-5, was well expressed and localized like FL-NCAM, and yet still not polysialylated (Figs. 2A, 3, and  4). These results show that mNCAM-5 is not recognized by the polysialyltransferases and show that FN1 of NCAM is necessary for the recognition and polysialylation of the Ig5 domain oligosaccharides. In addition, these data demonstrate that membrane-and soluble forms of NCAM-4, rather than those forms of NCAM-2, possess the minimal sequences for recognition and polysialylation by the polysialyltransferases, PST and STX.
It was also obvious from this analysis that, whereas membrane association was not necessary for polysialylation, it appeared to enhance the polysialylation of the NCAM proteins. It is likely that a prolonged Golgi residence time and enhanced interaction with membrane-associated enzymes contributes to the increased polysialylation observed for the membrane-associated NCAM proteins. In addition, we observed that while the molecular weight of the soluble NCAM proteins reflected differences in the number of Ig domains and FN repeats, the membrane-associated NCAM proteins were all migrating with similar maximum molecular masses, suggesting that some of the smaller NCAM proteins may actually contain more sialic acid than the larger proteins (Fig. 4). Preliminary studies to analyze the length of polysialic acid chains on these proteins suggest that indeed the smaller the NCAM protein the longer the average length of the added polysialic acid chains. 2 The reasons for this are unclear at present and now under investigation.

Separation of the Putative Enzyme Recognition Site from the Sites of Polysialylation on Ig5 Leads to a Significant Reduction in Polysialylation by PST and No Polysialylation by STX-To
further analyze the requirement for FN1 in the recognition and polysialylation of the oligosaccharides on the adjacent Ig5 domain, we wanted to separate this putative enzyme recognition site from the sites of polysialic acid addition. To do this we constructed mNCAM-6, consisting of all five Ig domains in its extracellular portion, and mNCAM-7, consisting of only FN1 and FN2 in its extracellular portion (Fig. 1, mNCAM-6 and  mNCAM-7). As an additional control, we also generated mN-CAM-8, which consists of FN2 in its extracellular portion (Fig.  1, NCAM-8). All three proteins were expressed at levels comparable with FL-NCAM (data not shown). Immunofluorescence microscopy using the anti-V5 epitope tag antibody demonstrated that these three NCAM proteins were expressed and localized in COS-1 cells like the FL-NCAM and mNCAM-4 proteins (Fig. 5, Localization). Using the OL.28 anti-polysialic acid antibody in our immunofluorescence assay for polysialylation (as done in Fig. 3), we found that while FL-NCAM and mNCAM-4 were highly polysialylated by PST Mut 2.3 and STX Mut 2.4.5, mNCAM-8 was not polysialylated by either enzyme, and mNCAM-6 and mNCAM-7 were only weakly polysialylated by PST Mut 2.3 (Fig. 5, Polysialylation). To compare the polysialylation of mNCAM-6, -7, and -8 to one another and to two minimal NCAM proteins, NCAM-2 and NCAM-4, we co-expressed each of these proteins with untagged, wild type PST, recovered the NCAM proteins by immunoprecipitation using the anti-V5 antibody and subjected the recovered proteins to immunoblot analysis with the OL.28 antibody as done previously (Fig. 6). While we cannot discern any information concerning the number or length of polysialic acid chains from this analysis, it was certain that both mNCAM-2 and mNCAM-4 were more immunoreactive with the OL.28 anti-polysialic acid antibody than either mNCAM-6 or mNCAM-7, and that no OL.28 immunoreactivity was detected for mNCAM-8 (Fig. 6). These results paralleled those in Fig. 5 and together with the results in Figs. 3 and 4 indicate that the combination of Ig5 and 2 C. Sato, B. Close, and K. Kitajima, unpublished results.

FIG. 2. Expression and localization of membrane-associated and soluble NCAM domain deletion mutants.
A, FL-NCAM and its domain deletion mutants were transiently expressed in COS-1 cells, metabolically labeled with Expre 35 S 35 S protein labeling mixture for 1 h, and immunoprecipitated from cell lysates (membrane-associated mNCAMs) or 3-h chase medium (soluble sNCAMs) with the anti-V5 epitope tag antibody. The immunoprecipitated samples were electrophoresed on 7.5% SDS-polyacrylamide gels and radiolabeled protein bands were visualized by fluorography. Molecular mass markers: 109-kDa, ␤-galactosidase; 78-kDa, bovine serum albumin; 46.7-kDa, ovalbumin. B, COS-1 cells transiently expressing FL-NCAM and its membrane-associated domain deletion mutants were analyzed by indirect immunofluorescence microscopy using the anti-V5 epitope tag primary antibody. Prior to staining, cells were fixed and permeabilized with methanol to visualize internal structures. Immunofluorescence was visualized using a Nikon Axiophot fluorescence microscope and a ϫ60 oil immersion Plan Apochromat objective. Magnification, ϫ750.  3. mNCAM-4, consisting of Ig5 and FN1, is sufficient for polysialylation: indirect immunofluorescence microscopy. COS-1 cells transiently coexpressing FL-NCAM or selected deletion mutants and either of the non-autopolysialylated, catalytically active enzymes, PST Mut 2.3 or STX Mut 2.4.5, were analyzed by indirect immunofluorescence microscopy using the OL.28 anti-polysialic acid primary antibody. Prior to staining, cells were fixed with methanol to visualize internal structures (permeabilized) or fixed with 3% paraformaldehyde to visualize only cell surface (non-permeabilized). The insets for mNCAM-5 show a phase-contrast image of the non-fluorescing cells. Immunofluorescence was visualized using a Nikon Axiophot fluorescence microscope and a ϫ60 oil immersion Plan Apochromat objective. Magnification, ϫ750. 4. mNCAM-4, consisting of Ig5 and FN1, is sufficient for polysialylation: immunoblot analysis. FL-NCAM and its soluble and membrane-associated domain deletion mutants were transiently co-expressed in COS-1 cells with non-epitope-tagged PST or STX and immunoprecipitated from unlabeled COS-1 cell lysates and medium 18 h post-transfection. Immunoprecipitated proteins were separated on 5% SDS-polyacrylamide gels, electrophoretically transferred to nitrocellulose membranes, and subjected to immunoblot analysis using the OL.28 anti-polysialic acid antibody. Molecular mass markers: 203 kDa, myosin; 109 kDa, ␤-galactosidase. are polysialylated only on N-linked glycans and that no Nglycans are present in the FN region (62,63). An inspection of the sequences in the ectodomain of mNCAM-7 revealed a consensus N-glycosylation site at Asn-702 close to the transmembrane region (Fig. 7A, bottom, NGS). To determine whether this Asn was glycosylated and the attached oligosaccharide polysialylated, we replaced the Asn at position 702 with a Gln in the mNCAM-7 protein. Analysis of protein molecular mass indicated that the mNCAM-7/N702Q protein migrated with a slightly lower molecular mass than unaltered mNCAM-7, and we found that the molecular mass difference between the mutant and unaltered protein could be eliminated by treatment of mNCAM-7 with PNGase F, an enzyme that removes all Nlinked oligosaccharides (60) (data not shown). This indicated that this N-linked glycosylation site was used in mNCAM-7. Next we analyzed the polysialylation of FL-NCAM, mNCAM-7, and mNCAM-7/N702Q following co-expression with PST. We found that mutation of the N-linked glycosylation site at Asn-702 and the absence of a glycan at that site did not eliminate the polysialylation of mNCAM-7 by PST. These results strongly suggested that O-glycans in the FN region are polysialylated by PST. Reconsidering the inability of mNCAM-8 to be polysialylated, it seems that this protein may lack not only the likely enzyme recognition site (FN1), but also could be missing sites of O-glycan attachment, which are polysialylated in the presence of FN1.

FIG.
The presence of polysialylated O-glycans in the mNCAM-7 protein led us to ask whether the FL-NCAM protein and our minimal polysialylated protein, mNCAM-4, also possess polysialylated O-glycans in addition to the polysialylated N-glycans within the Ig5 domain. To investigate this possibility we analyzed the OL.28 reactivity of FL-NCAM, mNCAM-4, mN-CAM-7, and mNCAM-7/N702Q before and after treatment with PNGase F (Fig. 7B). We found that polysialylation of FL-NCAM was not completely abolished following treatment with PNGase F. While we cannot at present rule out the possibility that core fucosylation of some of the polysialylated N-glycans is inhibiting PNGase F cleavage (64), control PNGase F digests of nonpolysialylated FL-NCAM suggests that all its N-glycans are removed by this treatment (data not shown). Based on our results with mNCAM-7, we believe it is more likely that Olinked glycans are polysialylated when the wild type protein is co-expressed with PST in COS-1 cells (Fig. 7B). Interestingly, the polysialylation of mNCAM-4 was dramatically reduced after PNGase F treatment, while the polysialylation of mN-CAM-7 and its N702Q glycosylation mutant were unchanged after enzyme treatment (Fig. 7B). Considered together, these results suggest two possibilities. First, the majority of the polysialylated O-linked glycans may reside outside the region encompassed by mNCAM-4 (Ig5-FN1). Second, that the Nglycans in the Ig5 domain may be better substrates for PST than accessible O-glycans. Additional experiments will be necessary to determine the sites of O-glycan attachment and polysialylation.

DISCUSSION
The limited number of polysialylated proteins in mammalian cells led us to hypothesize that polysialylation is a proteinspecific event and that the protein specificity of the polymerization process is mediated by an initial protein-protein contact between polysialyltransferase and glycoprotein substrate. Using the major polysialylated protein in mammalian cells, NCAM, we evaluated the sequence requirements for the recognition and polysialylation by the two polysialyltransferases, PST and STX. Co-expression of the two polysialyltransferases individually with the various NCAM deletion mutants revealed that FN1 of NCAM was necessary for the polysialylation of either N-linked oligosaccharides on the adjacent Ig5 domain (the major sites of polysialylation in the full-length protein) or O-linked oligosaccharides found within the FN region. We found that while a protein consisting of Ig5 and FN1 (NCAM-4) was polysialylated efficiently by both PST and STX, another consisting of only the Ig5 domain (NCAM-5) was not (Figs. 3  and 4). These results suggest that the polysialyltransferases may recognize and "dock" on FN1 of NCAM positioning themselves to polysialylate the N-glycans on the adjacent Ig5 domain. When we separated the major sites of polysialylation on Ig5 (mNCAM-6 (Ig1-Ig5)) from the FN repeats (mNCAM-7 (FN1 and FN2)), we unexpectedly found that these proteins were polysialylated at low levels by PST but not by STX (Figs. 5 and 6). While the polysialylation of mNCAM-6 suggested there may be a weaker enzyme recognition site within the Ig domain region, the polysialylation of mNCAM-7 by PST suggested that glycans other than the fifth and sixth N-linked oligosaccharides modifying Ig5 could be polysialylated. We found that while a previously unreported N-linked glycosylation site (Asn-702) was actually modified in the mNCAM-7 protein, its elimination did not abrogate polysialylation, strongly suggesting that O-linked oligosaccharides in this region were polysialylated (Fig. 7A). Further analysis showed that O-linked glycans as well as N-linked glycans were polysialylated when FL-NCAM is co-expressed with PST in COS-1 cells (Fig. 7B).
Nelson et al. (62) previously evaluated the sequences of NCAM necessary for polysialylation in F11 rat/mouse hybrid cells. At the time this work was published, the two polysialyltransferases had not been identified and cloned. Using domain deletion mutants and chimeric proteins they showed that FN1 of NCAM was critical for NCAM polysialylation, and also suggested that the Ig4 domain was important in this process because its elimination either by deletion or by replacement with sequences from the L1 adhesion molecule substantially decreased polysialylation (62). Later studies by this group, in which they evaluated the interplay between NCAM, polysialic acid, and cadherins in cell adhesion, suggested that the Ig4 domain of NCAM was not necessary for polysialylation by PST stably expressed in Chinese hamster ovary cells (65). This agrees with our studies presented in this article in which we found no requirement for Ig4 for polysialylation by either PST or STX.
Interestingly, more recent work by Angata et al. (38) showed that Ig4 was absolutely necessary for polysialylation by either PST or STX, and that polysialylation by STX was extremely weak even in the presence of Ig4, Ig5, FN1, and FN2. Why do these results differ from ours and from the more recent findings of Fujimoto et al. (65)? It is likely that a combination of factors contribute to these differences. One possibility is that the results of Angata et al. (38) reflect their use of soluble forms of both the polysialyltransferases and the various NCAM proteins (NCAM-IgG proteins) in an in vitro assay system. In contrast, our studies and those by Fujimoto et al. (62) were performed in cells using polysialyltransferases that were membrane-associated. Whereas our studies showed that membrane association of the NCAM proteins was not absolutely required for polysialylation, we did detect enhanced polysialylation of the membrane-associated forms of certain NCAM domain deletion mutants (see Fig. 4). Indeed, Nelson et al. (62) suggested that the distance of the Ig5 domain from the membrane and its positioning relative to the polysialyltransferase catalytic site may be critical for polysialylation. However, they also acknowledged that their results showed that in fact there was no strict requirement for Ig5 spacing from the membrane. One could envision that membrane association of both polysialyltransferases and NCAM substrates could enhance polysialylation by either optimizing the positioning of the enzymes with respect to the substrates and/or by increasing Golgi residence time for both enzymes and substrates. In an in vitro assay system residence time would not be an issue, but optimal positioning of enzyme and substrate might be important. In the absence of the naturally restricted movement of enzyme and substrate in the membrane, any factor that increased the opportunity for the polysialyltransferases to recognize and bind to NCAM would be expected to enhance polysialylation. Whereas our studies and the recent work of Fujimoto et al. (65) indicated that Ig4 was not necessary for polysialylation, the fact that the mNCAM-6 protein did exhibit low levels of polysialylation by PST hinted that there may be an additional weaker enzyme recognition site with the Ig domain region. Because of its position adjacent to Ig5, a likely candidate for a secondary docking site could be Ig4. Perhaps the presence of both Ig4 and FN1 in the soluble NCAM proteins used in the in vitro studies of Angata et al. (38) was necessary to increase the interactions of the soluble polysialyltransferases with these proteins. Likewise, it is possible that there are other factors found in cells that optimize the polysialylation process and these are missing in the in vitro assay.
Our results also suggest that the polysialic acid found on mNCAM-7 is exclusively found on O-glycans and that FL-NCAM expressed in COS-1 cells is found on both N-linked and O-linked glycans (Fig. 7). Previous work analyzing NCAM expressed in F11 rat/mouse hybrid cells and newborn mouse brains has demonstrated that polysialic acid is found on the fifth and sixth N-linked oligosaccharides found to modify Ig5 (62,63). However, there is also evidence for O-linked glycans on NCAM and O-linked polysialic acid on NCAM and other proteins. Dickson et al. (66) found that myotubes express an additional sequence between the two FN repeats called MSD1 that is generated by tissue-specific and developmentally regulated alternative splicing. Further work by this group demonstrated that this region in the glycosylphosphatidylinositol-anchored form of NCAM expressed in myotubes is enriched in O-linked glycans (67). Later work by Suzuki et al. (68) suggested that the O-linked glycans in this muscle-specific region could be polysialylated in the C21C12 myotube cell line. The results presented here strongly suggest that mNCAM-7, which has an ectodomain consisting of only FN1 and FN2 and adjacent sequences, is polysialylated on O-glycans. Sequencing of fulllength human NCAM 140 demonstrated that our clone does not contain the MSD1 sequences that are inserted into the FN region. This suggests that other O-glycans in this region are being polysialylated.
Other examples of O-linked polysialic acid have also been published. Recently, Kitajima and colleagues (41) showed that a small proportion of the CD36 scavenger receptor found in milk is polysialylated, and that the polysialic acid on this protein is exclusively modifying O-glycans. Earlier in our laboratory, we identified proteins expressed in rat basophilic leukemia cells and MCF-7 human breast carcinoma cells that possessed both N-linked and O-linked polysialic acid (69). We demonstrated that the polysialylated proteins in these cells are distinct from NCAM and the voltage-dependent sodium channel ␣ subunit. As Yabe et al. (41) note, it is also unlikely that CD36 is the polysialylated molecule in rat basophilic leukemia and MCF-7 cells because of significant differences in the molecular masses of CD36 and the other as yet unidentified rat basophilic leukemia and MCF-7 proteins. Taken together the results of these reports suggest that the presence of O-linked polysialic acid is cell type-specific and likely dependent on the type of O-linked structures generated by a cell and whether they are acceptable substrates for polysialylation.
The data that we present in this article demonstrate that FN1 is necessary for the polysialylation of the N-linked glycans on the fifth Ig domain of NCAM, and that Ig5-FN1 is the minimal protein that is recognized and polysialylated by the polysialyltransferases, PST and STX. From these data we propose a model where, for polymerization of the polysialic acid chain to occur, the polysialyltransferases must be able to make a relatively stable contact with the NCAM protein that enables them to access appropriate glycan structures and "hold on" while polymerization occurs (Fig. 8). In our model this contact or docking site is FN1 of NCAM. From this model we predict that the interaction between FN1 and the enzyme is relatively strong, that the arrangement of glycans with respect to this docking site is critical, and that the affinity of the enzyme for the docking site may influence chain length.
Can our model explain the polysialylation of other proteins such as the sodium channel ␣ subunit and CD36, or the autopolysialylation of the polysialyltransferases? It is predicted that FN repeats are found in ϳ2% of all animal proteins, and yet of that group, only NCAM is polysialylated (70). In many cases, the FN repeat is a region of structural, not sequence homology (71,72). Taking this into consideration, it seems likely that specific amino acid contacts might be critical for enzyme-NCAM interaction. These could come in the form of a linear amino acid sequence or a signal patch generated by noncontiguous amino acids. Because there is no obvious homology between the linear sequences of NCAM FN1, the ␣ subunit of the sodium channel, and CD36, it seems that a signal patch within FN1 of NCAM and the other proteins may mediate the interactions of the polysialyltransferases with these substrates. Autopolysialylation of the polysialyltransferases may be somewhat different mechanistically and yet still rely on protein-protein interactions in the form of enzyme dimerization or oligomerization. In support of this idea, many glycosyltransferases form dimers (73)(74)(75)(76), and another ␣2,8-sialyltransferase, ST8Sia III, is also capable of autopolysialylation even though it does not polysialylate NCAM or any other glycoprotein substrate (39). We predict that these three ␣2,8-sialyltransferases form dimers or higher order oligomers, and this type of protein-protein interaction allows them to autopolysialylate their own oligosaccharides. We are currently testing these models of NCAM polysialylation and polysialyltransferase autopolysialylation, and refining the NCAM FN1 sequence requirements for polysialyltransferase recognition and binding.