The CMP-sialic Acid Transporter Is Localized in the Medial-Trans Golgi and Possesses Two Specific Endoplasmic Reticulum Export Motifs in Its Carboxyl-terminal Cytoplasmic Tail*

The addition of sialic acid to glycoproteins and glycolipids requires Golgi sialyltransferases to have access to their glycoconjugate substrates and nucleotide sugar donor, CMP-sialic acid. CMP-sialic acid is transported into the lumen of the Golgi complex through the CMP-sialic acid transporter, an antiporter that also functions to transport CMP into the cytosol. We localized the transporter using immunofluorescence and deconvolution microscopy to test the prediction that it is broadly distributed across the Golgi stack to serve the many sialyltransferases involved in glycoconjugate sialylation. The transporter co-localized with ST6GalI in the medial and trans Golgi, showed partial overlap with a medial Golgi marker and little overlap with early Golgi or trans Golgi network markers. Endoplasmic reticulum-retained forms of sialyltransferases did not redistribute the transporter from the Golgi to the endoplasmic reticulum, suggesting that transporter-sialyltransferase complexes are not involved in transporter localization. Next we evaluated the role of the transporter's N- and C-terminal cytoplasmic tails in its trafficking and localization. The N-tail was not required for either endoplasmic reticulum export or Golgi localization. The C-tail was required for endoplasmic reticulum export and contained di-Ile and terminal Val motifs at its very C terminus that function as independent endoplasmic reticulum export signals. Deletion of the last four amino acids of the C-tail (IIGV) eliminated these export signals and prevented endoplasmic reticulum export of the transporter. This form of the transporter supplied limited amounts of CMP-sialic acid to Golgi sialyltransferases but was unable to completely rescue the transporter defect of Lec2 Chinese hamster ovary cells.

Biochemical and genetic analyses have shown that glycans modifying proteins and lipids mediate and modulate critical cell and protein interactions during development, in the adult animal, and during disease (1)(2)(3)(4)(5). They play important roles in protein folding and quality control (6); in protein targeting and clearance (7)(8)(9); in cell adhesion, cell signaling, and cell-matrix interactions (3,5,10); and in neural cell and immune cell function (4, 10 -14). Many glycans are terminated with negatively charged sialic acid residues that act as critical functional determinants. For example, glycans terminating in ␣2,6-sialic acid interact with CD22 (Siglec 2) and are necessary for B cell function (15). ␣2,8-Polysialylated glycans on the neural cell adhesion molecule are critical for brain development, promote axon guidance and pathfinding, and mediate neurite outgrowth and synaptic plasticity (10,13,16,17). Other sialylated glycans are receptors for viruses (5), regulate the clearance of glycoproteins (7), and impact the growth, motility, and invasiveness of cancer cells (2, 18 -21). Finally, the overall importance of glycoconjugate sialylation is highlighted by the finding that inactivation of the UDP-N-acetylglucosamine 2-epimerase, a bifunctional enzyme that catalyzes the first two steps in sialic acid biosynthesis, causes early embryonic lethality in mice (22).
The processes of protein and lipid glycosylation are compartmentalized in the endoplasmic reticulum (ER) 2 and Golgi complex of the cell, with the sialylation of glycoproteins and glycolipids confined to the Golgi cisternae (23,24). The addition of sialic acid residues to terminal positions of glycan structures is catalyzed by different substrate-and linkage-specific sialyltransferases (25). The functional efficiency of these enzymes depends upon their ability to encounter both their glycoprotein substrates, which are made in the same or earlier compartments, and their nucleotide sugar donor, CMP-sialic acid. CMP-sialic acid is synthesized in the nucleus by CMP-sialic acid synthetase (26,27). It diffuses into the cytoplasm and is specifically transported into the lumen of the Golgi apparatus by the CMP-sialic acid transporter (CST) (reviewed in Ref. 28).
The CST is an antiporter that facilitates both the influx of CMP-sialic acid into the Golgi complex and efflux of CMP, a product of the transfer reaction, out of the Golgi complex (29 -32). Since CMP is an inhibitor of sialyltransferases, its rapid removal from the Golgi complex is important to optimize sialyltransferase activity (33,34). The mammalian CST has been cloned by phenotypic correction of the Lec2 CHO cell CMPsialic acid transport mutant (31), and its identity has been further confirmed by functional expression in organisms that lack an endogenous CST, such as Saccharomyces cerevisiae and Escherichia coli (32,35). Work by Eckhardt et al. (36) demonstrated that the CST is a multipass membrane protein with 10 membrane-spanning regions and both N-and C-terminal sequences in the cytoplasm. Although recent studies show that some nucleotide sugar transporters have specificity for multiple substrates (37)(38)(39)(40)(41), the CST is very specific for its substrates CMPsialic acid and CMP, despite its 43% sequence identity with the UDP-galactose transporter (30). Using chimeric transporters containing sequences from both the CST and UDP-galactose transporter, Aoki et al. (29,30) demonstrated that transmembrane helix 7 of the CST is required for nucleotide sugar specificity, whereas transmembrane helices 2 and 3 are required for the efficiency of nucleotide sugar transport. Recent studies by Tiralongo et al. (32) suggest that the CST is a simple mobile carrier with a binding site that alternates between both sides of the membrane.
Genetic abnormalities in nucleotide sugar transporters, including the CST, are the basis for several developmental defects and diseases. For example, in C. elegans, mutations in the SQV7 nucleotide sugar transporter compromise chondroitin sulfate and heparan sulfate biosynthesis, leading to developmental defects in vulval formation (42,43). In humans, mutations in nucleotide sugar transporters are responsible for some congenital disorders of glycosylation (CDG) in which large groups of glycans are reduced or missing, leading to multisystemic health problems that include increased infections, hypotonia, psychomotor and growth retardation, and bleeding disorders (1). For example, leukocyte adhesion deficiency II/CDG IIc is caused by defects in the GDP-fucose transporter that lead to a severe reduction in fucosylated glycans, including selectin ligands, which are important in the inflammatory response and lymphocyte homing (44 -46). More recently, a new CDG type II, initially identified in a child lacking sialyl Lewis X glycans on polymorphonuclear cells, was revealed to be the result of inactivating mutations in the patient's two CST alleles (47).
Sialyltransferases are found in different Golgi cisternae, reflecting their functional order in the glycosylation pathway in which they participate. The sialyltransferases modifying Asnlinked glycans of glycoproteins are localized primarily in the distal Golgi cisternae, because they are among the last enzymes to act in this pathway (24,48). This is exemplified by the ST6Gal I enzyme that has been localized to the trans cisternae and trans reticular network (trans Golgi network (TGN)) of rat hepatocytes (49). Our laboratory has localized this enzyme to the medial and trans cisternae of the HeLa cell Golgi (50). In ganglioside biosynthesis, sialic acid is added both early and late in the synthetic sequence, and thus the sialyltransferases involved in this pathway are localized across the Golgi stack. The work of Maccioni and colleagues (23,51) shows that the sialyltransfer-ase responsible for the biosynthesis of the GM3 ganglioside is localized earlier in the Golgi (cis-medial) than the sialyltransferase required for the biosynthesis of the GT3 ganglioside (trans) and that a third sialyltransferase that converts GM1 to GD1 is probably found in the TGN. Most recently, work by Sewell et al. (52) demonstrated that the ST6GalNAc-I enzyme, which synthesizes the tumor-associated sialyl-Tn O-glycan structure, is localized throughout the Golgi stack, providing an explanation for why this sialyltransferase, when expressed in breast cancer cells, is able to preempt the activities of the glycosyltransferases that synthesize the normal core 1 and core 2 glycan structures.
Since the CST supplies CMP-sialic acid to all sialyltransferases, it seems that the exchange of CMP-sialic acid and CMP between these enzymes and the CST would be more efficient if these proteins were localized in the same Golgi subcompartments and existed as functional complexes. Previous studies demonstrated that a portion of the UDP-galactose transporter relocalizes to the ER by association with the UDP-galactose:ceramide galactosyltransferase to allow UDP-Gal import into the ER lumen for galactosyl-ceramide synthesis (53). In this work, we have tested the hypothesis that the CST is localized across the Golgi stack in order to more efficiently supply the sialyltransferases involved in glycoprotein and glycolipid biosynthesis with CMP-sialic acid. Using immunofluorescence and deconvolution microscopy, we find that the CST is restricted to the medial-trans cisternae of the HeLa cell Golgi complex, mirroring the localization of the exogenously expressed ST6Gal I STtyr isoform. Redistribution experiments provide no evidence for physical association of either ST6Gal I STtyr isoform or ST8Sia IV polysialyltransferase with the CST, making it unlikely that sialyltransferase association is responsible for the restricted CST localization pattern or its overall Golgi localization. Studies of the sequence requirements for CST trafficking and localization showed that the CST C-terminal tail sequences contain two independent ER export signals. An ER-retained CST mutant lacking the export signals provides a limited amount of CMP-sialic acid to Golgi sialyltransferases but is unable to completely restore sialylation to Lec2 CHO cells lacking functional CST. These data suggest that although an ERretained CST is active and CMP-sialic acid is free to diffuse throughout the ER-Golgi system, efficient sialylation occurs only when both the CST and sialyltransferases are compartmentalized in the Golgi apparatus.

MATERIALS AND METHODS
Tissue culture media and reagents, including Dulbecco's modified Eagle's medium (DMEM), minimal essential medium/␣-medium (MEM), Opti-MEM I, Lipofectin, and fetal bovine serum (FBS), as well as oligonucleotides, Taq polymerase SuperMix, and restriction endonucleases were purchased from Invitrogen. Targefect transfection reagent was from Targeting Systems (Santee, CA). Fugene 6 transfection reagent was purchased from Roche Applied Science. Prestained protein molecular mass standards were from Bio-Rad. The QuikChange site-directed mutagenesis kit was purchased from Stratagene. 35 S-Express protein label and [␣-35 S]dATP were obtained from PerkinElmer Life Sciences.
Protein A-Sepharose was purchased from Amersham Biosciences. Mouse anti-GM130 antibody was from BD Biosciences. The sheep anti-TGN46 antibody was from Serotec Ltd. (Oxford, UK). Fluorescein isothiocyanate (FITC)-and TRITC-conjugated goat anti-rabbit and goat anti-mouse, Cy2-conjugated donkey anti-sheep, and Cy3-conjugated donkey anti-mouse secondary antibodies were obtained from Jackson ImmunoResearch (West Grove, PA). The rabbit anti-Erv46 antibodies were a kind gift from Dr. Stefan Otte (Department of Biochemistry and Molecular Genetics, University of Illinois). The sheep anti-calreticulin antibodies were a kind gift from Dr. Hans-Dieter Soeling (Max-Planck-Institute of Biophysical Chemistry, Goettingen, Germany). The rabbit antibodies against ␣-mannosidase II were generously provided by Dr. Marilyn Farquhar (University of California, San Diego). The mouse CST cDNA and rabbit anti-CST antibodies were graciously provided by Drs. Nobuhiro Ishida and Masao Kawakita (Tokyo Metropolitan Institute for Medical Science). Other chemicals and reagents were purchased from Sigma and Fisher.
Construction of the HA-tagged CST Protein-Murine CST cDNA was inserted into the pCIneo expression vector (Promega Corp., Madison, WI) using EcoRI and XbaI restriction sites. The wild type (WT) CST-pCIneo construct was used as a template to insert a hemagglutinin (HA) tag (YPYDBPDYASL) following Ser 202 in the cytoplasmic loop between transmembrane helices 6 and 7. Site-directed mutagenesis was performed using the sense primer 5Ј-GTC TTA AAG AGT TCC TAC CCT TAT GAC GTC CCC GAT TAC GCC AGC CTG GAC ACT TCC CTT TG-3Ј and the antisense primer 5Ј-CA AAG GGA AGT GTC CAG GCT GGC GTA ATC GGG GAC GTC ATA AGG GTA GGA ACT CTT TAA GAC-3Ј. The sequence of the new cDNA was confirmed using the Quick Denature plasmid sequencing kit.
Construction of the N-and C-terminal Tail Mutants of the CST-All tail mutant cDNAs were generated from the HAtagged murine CST by either PCR amplification using the Taq polymerase SuperMix (for ⌬N-tail, ⌬C-tail, ⌬IIGV, and ⌬GV constructs) or the QuikChange site-directed mutagenesis kit (for ⌬II and II 3 AA constructs). For ⌬N-tail mutant cDNA, the 3Ј-pCIneo-T3 primer 5Ј-CAA TTA ACC CTC ACT AAA GG-3Ј was used. For ⌬C-tail, ⌬IIGV, and ⌬GV constructs, the 5Ј-pCIneo-T7 primer 5Ј-TAA TAC GAC TCA CTA TAG GG-3Ј was used. The PCR products of all the deletion mutants were digested with restriction enzymes EcoRI and XbaI and ligated into pCIneo vector cut with the same enzymes. The sequences of the cDNAs of all of the mutants were confirmed by sequencing using the Quick Denature plasmid sequencing kit.
Transfection of COS-1, HeLa, and Lec2 CHO Cells-Cells were maintained in DMEM (for COS-1 and HeLa cells) or MEM (for Lec2 CHO cells) containing 10% FBS and antibiotics and grown in a 37°C, 5% CO 2 incubator until 50 -70% confluent in 100-mm tissue culture dishes or in 24-well plates containing 12-mm glass coverslips. For COS-1 cells, Lipofectin transfections were performed according to the manufacturer's protocol. Briefly, 30 l of Lipofectin and 20 g of DNA were incubated separately in 1.5 ml of Opti-MEM in polystyrene culture tubes at room temperature for 35 min. They were then mixed together and incubated for another 15 min at room temperature. The transfection mixtures were then added to cells washed with 2 ϫ 5 ml of Opti-MEM. The cells were then incubated in a 37°C, 5% CO 2 incubator. Seven milliliters of DMEM with 10% FBS and antibiotics was added to the plates at 6 h, and the incubation continued for another 18 h. For HeLa cells, Targefect transfection was performed following the manufacturer's protocol. Briefly, 1.5 l of Targefect F-2 and 0.8 g of DNA were incubated together in 50 l of DMEM, 10% FBS in polystyrene culture tubes at room temperature for 20 min. The mixtures were then added to the cells on coverslips together with 450 l of DMEM, 10% FBS. The cells were incubated in the 37°C, 5% CO 2 incubator for 24 h. For Lec2 CHO cells, Fugene 6 transfection was performed following the manufacturer's protocol. Briefly, 3 l of Fugene 6 was incubated in 100 l of MEM in polystyrene culture tubes at room temperature for 5 min. Then 1.5 g of DNA was added, and the mixture was incubated for another 30 min and added to the cells on coverslips. After cells were incubated in the 37°C, 5% CO 2 incubator for 5 h, 1 ml of MEM, 10% FBS was added to cells, and the incubation continued for another 24 h.
Metabolic Labeling of Cells-16 -20 h following transfection, cells were incubated with 10 ml of Cys/Met-free DMEM for 1 h. After incubation, the medium was replaced with 4 ml of fresh Cys/Met-free DMEM containing 100 Ci/ml 35 S-Express protein labeling mix (PerkinElmer Life Sciences). Cells were radiolabeled for 2 h in a 37°C, 5% CO 2 incubator. After labeling, the cells were washed with 10 ml of PBS and then lysed with 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). Cell lysates and medium samples were stored at Ϫ20°C overnight.
Immunoprecipitation of Metabolically Labeled HA-tagged CST Proteins-CST proteins were immunoprecipitated from cell lysates using 3 l of anti-HA epitope tag antibody (Sigma H9658) and 50 l of a 50% slurry of protein A-Sepharose, as described previously (54). The immunoprecipitation beads were resuspended in 50 l of Laemmli sample buffer, and the entire sample, including Sepharose beads, was directly loaded into the gel well. The proteins were separated on a 10% SDSpolyacrylamide gel. Radiolabeled proteins were visualized by fluorography using 10% 2,5-diphenyloxazole in Me 2 SO, and gels were exposed to Eastman Kodak Co. BioMax MR films. COS-1 cells were used for these experiments because they are easily transfected with high efficiency and yield the levels of expressed proteins required for immunoprecipitation analysis.

Indirect Immunofluorescence Localization of WT and Mutant CST Proteins-HeLa cells were plated on glass coverslips
and transiently transfected with the CST cDNA expression constructs, as described above. Cells were processed for immunofluorescence 24 h post-transfection, as described previously (50). Briefly, cells were fixed and permeabilized with Ϫ20°C methanol to visualize the internal structures. Cells were washed with 2 ϫ 1 ml of PBS, blocked for 1 h with 5% goat serum or donkey serum (according to the secondary antibodies used) in PBS, and incubated with a 1:100 dilution of the mouse anti-HA (or affinity-purified rabbit anti-CST antibody in the case of GM130 co-staining) in PBS for 1 h. After washing the coverslips with PBS (4 ϫ 1 ml), cells were incubated with a TRITC-or Cy3-conjugated secondary antibody for 45 min. Cells were washed a second time with in PBS (4 ϫ 1 ml). To stain with ER/Golgi markers, cells were incubated with a 1:100 dilution of the antibody against an ER/Golgi marker for 1 h. After washing cells with PBS (4 ϫ 1 ml), cells were incubated with an FITC-or Cy2-conjugated secondary antibody for 45 min. Cells were washed with PBS (4 ϫ 1 ml) and mounted on glass slides using 20 l of mounting medium (15% (w/v) Vinol 205 polyvinyl alcohol, 33% (v/v) glycerol, 0.1% sodium azide, pH 8.5, in PBS). Cells were visualized and photographed using a Leica DMRB microscope equipped with a Hamamatsu charge-coupled device camera driven by the OpenLab imaging program (Improvision, Coventry, UK). For the Golgi localization of the WT CST, images at different optical sections were collected and deconvolved into the final images.
Assessment of CST Activity in Lec2 CHO Cells-This analysis was performed according to Aoki et al. (29,30). Lec2 CHO cells were transfected with the WT or mutant CSTs as described above. Cells were fixed with Ϫ20°C methanol to visualize the internal structures. The fixed cells were stained first with a 1:100 dilution of an FITC-conjugated peanut agglutinin (PNA) for 30 min. Cells were then washed with PBS (4 ϫ 1 ml) and incubated with a 1:100 dilution of the mouse anti-HA antibody for 1 h. After again washing the coverslips with PBS (4 ϫ 1 ml), cells were incubated with a TRITC-conjugated goat anti-mouse IgG for 45 min. Cells were washed with PBS (4 ϫ 1 ml) and mounted on glass slides using 20 l of mounting medium. Cells were again visualized and imaged using the Leica DMRB microscope.
Lectin Blotting-Lec2 CHO cells plated on 100-mm culture dishes were co-transfected with the WT or mutant CST and STtyr-Myc or Iip33 (major histocompatibility complex class II associated invariant chain protein)-STtyr-Myc. 24 h following transfection, cells were lysed with 100 l 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), and resolved on 10% SDS-polyacrylamide gel. Proteins were then transferred to a nitrocellulose film and probed with a horseradish peroxidase-conjugated Sambucas nigra agglutinin (SNA) in Tris-buffered saline with 1% Tween 20 at room temperature for 2h. The blot was washed four times before being developed using the SuperSignal West Pico Chemiluminescence kit and exposed to Kodak Bio-Max film at room temperature.

RESULTS
The CST is localized in the Golgi complex, where it functions to supply CMP-sialic acid to Golgi sialyltransferases ( Fig. 1) (29,30,36,54). We initiated these studies to understand how the sialyltransferases and the CST are organized in the Golgi apparatus to promote efficient sialylation. Here we evaluate the Golgi compartmentation of the CST, its ability to form stable complexes with sialyltransferases, and the role of its cytoplasmic sequences in trafficking and Golgi localization.
The CST Is Concentrated in the Medial-Trans Cisternae of the Golgi Complex-Using immunofluorescence and deconvolution microscopy, we tested the prediction that the CST may be more diffusely localized than any one sialyltransferase and that it may even be localized across the Golgi stack. The available anti-CST antibodies are directed against the C-terminal cytoplasmic tail of the CST (29). So that we could delete and mutagenize the cytoplasmic tail sequences to determine their role in CST trafficking and localization, we inserted an HA tag in the cytoplasmic loop between transmembrane helices 6 and 7 (see Fig. 4) and cloned the CST-HA cDNA into the pCIneo mammalian expression vector. The localization of the CST-HA protein was compared with the untagged CST to ensure that the epitope tag did not compromise localization. CST-pCIneo and CST-HA-pCIneo were transiently expressed in HeLa cells. We chose HeLa cells for these studies because they have a more elaborated Golgi complex with well defined subcompartments (55). CST localization was evaluated following methanol fixation using anti-HA tag antibodies (for CST-HA) or anti-C-terminal tail antibodies (for untagged CST) and FITC-conjugated secondary antibodies. We found that both forms of the CST were localized in a juxtanuclear Golgi compartment and exhib- ited little ER localization even when overexpressed (Fig. 1). Notably, no significant cell surface or lysosomal accumulation was observed for either overexpressed protein (data not shown), suggesting that the CST may not depend upon interaction with other low abundance proteins for its Golgi localization.
In order to determine the sub-Golgi localization of the CST, we transiently expressed CST-HA in HeLa cells and compared its localization to known ER and Golgi subcompartment markers (Fig. 2). Erv46 was used as a marker of the ER-Golgi intermediate compartment (ERGIC) and cis Golgi (56), GM130 was used as a marker of the cis Golgi network (57), ␣-mannosidase II (ManII) was used as a marker of the medial Golgi (58), and TGN46 was used as a marker of the trans Golgi network (59). We also co-expressed CST-HA with the ST6Gal I STtyr iso-form and compared their localization. Each of these Golgi markers appeared to co-localize with the CST when analyzed by traditional fluorescence microscopy (see Erv46 co-staining in Figs. 6 and 7 and data not shown). However, when images at different optical sections were obtained and deconvolved into a final image, we found that these markers and the CST exhibited more restricted sub-Golgi localization. The CST exhibited little to no co-localization with endogenous markers of the ERGIC/cis Golgi (Erv46), cis Golgi network (GM130), and trans Golgi network (TGN46). The CST showed a partial overlap with the medial Golgi marker, ManII, but co-localized most extensively with the ST6Gal I, which we had previously shown by both immunoelectron microscopy and deconvolution microscopy to be localized in the medial and trans cisternae of the HeLa cell Golgi complex (50). However, it was clear that CST was less broadly distributed than the ST6Gal I. These data indicate that the CST is more restricted in its pattern of sub-Golgi localization than we expected and imply that CMP-sialic acid and CMP move more freely throughout the Golgi stacks.
The Formation of CST-Sialyltransferase Complexes Is Not Supported by Redistribution Assays-Previous studies elegantly showed that the UDP-galactose transporter forms complexes with the ER-localized UDP-galactose:ceramide galactosyltransferase in order to redistribute a portion of this transporter from the Golgi complex to the ER (53). To evaluate the possibility that the CST forms complexes with sialyltransferases, we used a similar method employed by Nilsson et al. (60) to demonstrate interactions between Golgi glycosylation enzymes. We co-expressed CST-HA with either our previously analyzed, ER-localized Iip33-STtyr-Myc (59) or newly constructed Iip33-␣2,8-polysialyltransferase (Iip33-PST-Myc) and Iip33-galactosyltransferase (Iip33-␤1,4-galactosyltransferase-Myc). These constructs possess the cytoplasmic tail of Iip33 that includes its ER retrieval/ retention signal (61). Using immunofluorescence to analyze COS-1 cells co-expressing the CST and an ER-retained glycosyltransferase, we found that, like the ER-retained Iip33-␤1,4galactosyltransferase-Myc, neither of the two ER-localized sialyltransferases was able to redistribute the Golgi-localized CST to the ER (Fig. 3). We also found no evidence for stable complexes of the CST and ST6Gal I STtyr isoform using co-immunoprecipitation assays with or without cross-linking (data not shown). Although we cannot rule out weak interactions between the CST and sialyltransferases, these data suggest that it is unlikely that complexes formed between the CST and the STtyr protein lead to its restricted localization in the medial-trans regions of the Golgi complex and raise the question of what sequences mediate the trafficking and localization of the CST.
The C-terminal Cytoplasmic Tail of the CST Is Required for Its Exit from the ER-The transmembrane helices and cytoplasmic tails of both glycosyltransferases and viral multipass membrane proteins play important roles in their trafficking and Golgi localization (62)(63)(64)(65)(66)(67). The CST is predicted to have 10 transmembrane regions and N-terminal and C-terminal cytoplasmic tails (36) (Fig. 4, WT CST). We first generated and analyzed tailless CST mutants (⌬N-tail CST and ⌬C-tail CST) (Fig. 4). These proteins and other CST mutants described below were transiently expressed in COS-1 cells, metabolically labeled with [ 35 S]Met/Cys, immunoprecipitated with anti-HA antibodies, and electrophoresed on SDS-polyacrylamide gels to evaluate their molecular masses and expression levels (Fig. 5). The HA-tagged WT CST and mutant CST proteins were all well expressed and migrated relative to the unaltered WT CST as expected, based upon the changes made in their sequences. Interestingly, the unaltered WT CST migrates as two bands of 28 and 32 kDa, both of which are of lower molecular mass than predicted for CST based on amino acid sequence (39 kDa). All of the mutant CST proteins also migrated as doublets, including the two tail deletion mutants. This observation suggests that cleavage of either tail cannot explain the presence of two CST bands. CST Golgi localization is also not required for the generation of the two bands, because mutant proteins that are halted in the ER still migrate as a doublet (⌬C-tail CST and ⌬IIGV CST; see below). In addition, the lack of consensus N-glycosylation sites in the CST extracytoplasmic sequences rules out the possibility that heterogeneous N-glycosylation is involved in the observed size difference between CST bands. Tiralongo et al. (32) observed a similar CST doublet in their studies. The origin of these two CST bands is unclear.
We compared the localization of the two tailless mutants with the WT CST-HA. The ⌬N-tail CST localized in the Golgi   OCTOBER 13, 2006 • VOLUME 281 • NUMBER 41 JOURNAL OF BIOLOGICAL CHEMISTRY 31111 (Fig. 6, top panels), suggesting that the N-terminal tail of the CST has no role in mediating the protein's Golgi localization or trafficking. In contrast, the ⌬C-tail CST was always found in reticular structures reminiscent of the ER, and in some cells it was also observed in juxtanuclear Golgi structures (Fig. 6, top  and bottom panels). The ER localization of the ⌬C-tail CST was confirmed by its partial co-localization with the ER marker, calreticulin (68) (Fig. 6, bottom panels). These results suggest that the C-terminal cytoplasmic tail of CST is important for its efficient export from the ER.

The Di-Ile and Val Motifs on the C-terminal Cytoplasmic
Tail of the CST Mediate Its Export from the ER-The predominant ER localization of the ⌬C-tail CST suggested that the deleted tail sequences contained ER export signals. Several different types of ER export signals have been identified in the cytoplasmic sequences of transmembrane proteins (69,70). Inspection of the CST C-terminal cytoplasmic tail sequences revealed that the very C-terminal four amino acids, Ile-Ile-Gly-Val (IIGV), contain both a dihydrophobic ER export motif (Ile-Ile at Ϫ3 and Ϫ4) and a C-terminal Val motif that have been shown to function as ER export signals in ERGIC-53 (70). To test the role of these residues in CST ER export, we made a series of C-tail mutants that included the following: ⌬IIGV, in which both putative ER export signals were eliminated; ⌬II and ⌬GV, in which the two putative signals were deleted independently; and II 3 AA, in which the two Ile residues were replaced with two Ala residues (Fig. 4). All mutants were expressed well in COS-1 cells and migrated with the expected molecular masses (see Fig. 5).
To evaluate the impact of these mutations on CST trafficking, we transiently expressed the mutant proteins in HeLa cells and evaluated their localization relative to the Erv46 Golgi marker and calreticulin ER marker by immunofluorescence analysis (Fig. 7). We found that the ⌬IIGV mutant, lacking both the dihydrophobic (Ile-Ile) and C-terminal Val motifs, was localized in the ER (Fig. 7). Unlike the overexpressed ⌬C-tail CST protein, which on occasion was found in small amounts in the Golgi complex (Fig. 6, inset), we rarely observed any ⌬IIGV protein in the Golgi complex. This difference in localization was supported by Lec2 CHO cell rescue experiments in Fig. 8 (see below). By contrast, the ⌬GV, ⌬II, and II 3 AA CST mutants, in which only one of the two ER export motifs was eliminated, were localized in the Golgi like the WT CST (Fig. 7), indicating that either of these motifs alone was sufficient to mediate the ER export of the protein.
An ER-retained CST Mutant Is Unable to Completely Rescue Sialylation in Lec2 CHO Cells-In order to evaluate the impact of tail deletions and changes in localization on CST activity, we employed an in vivo activity assay previously devised by Aoki et al. (29,30). PNA is a lectin that binds to terminal galactose residues on glycoconjugates. In Lec2 CHO cells, there is little to no CST activity due to a mutation in the endogenous CST protein (71,72). All sialylation is therefore compromised, and as a consequence, Lec2 CHO cells express glycoconjugates with terminal galactose residues that can be recognized by fluorescently  tagged PNA. If an active CST protein is expressed in these cells, sialylation occurs, covering exposed galactose residues and eliminating PNA staining.
To assess how changes in location and mutations in the C-terminal cytoplasmic tail might alter in vivo activity, we evaluated the ability of CST and the CST C-tail mutants to rescue the CST deficiency in Lec2 CHO cells. CST proteins were transiently expressed in Lec2 CHO cells, and cells were analyzed by fluorescence microscopy for CST protein expression (using anti-HA antibodies) and CST activity (loss of PNA staining). The ST6Gal I STtyr isoform was expressed as a negative control. Untransfected Lec2 CHO cells or Lec2 CHO cells expressing ST6Gal I were uniformly stained in the Golgi complex and on the cell surface with PNA-FITC (Fig. 8, ST6Gal I and untransfected cells in every panel). In contrast, cells expressing the Golgi-localized WT CST, ⌬N-tail CST, ⌬II CST, ⌬GV CST, and II 3 AA CST proteins (indicated by arrows) lost PNA-FITC staining (Fig. 8), indicating that global sialylation was restored. Interestingly, whereas the sialylation defect was not rescued in the majority of ⌬C-tail CST-expressing cells (see ⌬C-tail (ER) panels) and in all ⌬IIGV CST-expressing Lec2 cells, it was rescued in a small proportion of cells expressing ⌬C-tail CST. This result is consistent with our observations that the DC-tail mutant exhibits a dual ER-Golgi localization in selected cells (Fig. 8, ⌬C-tail CST (ER/Golgi)). The ability of the Golgi-localized ⌬C-tail CST to function as a transporter also suggested that the CST C-terminal tail sequences are not required for transporter activity.
The inability of the ER-retained ⌬IIGV CST to rescue the Lec2 cell phenotype led us to ask whether any CMP-sialic acid was being delivered to the Golgi complex from the ER and whether any sialylation was occurring. Studies of nucleotide transporter mutants that maintained low levels of activity showed that various enzymes/pathways responded differently to limited nucleotide sugar levels. It was suggested that this finding reflected differences in the affinity (K m values) of specific enzymes for their nucleotide sugar donors (73,74). Based on these previous studies, we wondered whether CMP-sialic acid was actually transported into the ER lumen and then trafficked to the Golgi complex, but the exchange and return of CMP to the ER was too slow to maintain an optimal antiporter activity and sufficient CMPsialic acid levels in the Golgi complex. If this were the case, some sialyltransferases might be able to function and others not, leading to a situation where only some of the terminal galactose residues would be capped with sialic acids that could be recognized by specific lectins. To evaluate this possibility, we asked whether Lec2 CHO cells co-expressing ER-or Golgi-localized forms of CST and the ST6Gal I STtyr isoform produced ␣2,6sialylated glycoproteins (note that Lec2 CHO cells express neither CST nor ST6Gal I). Lec2 CHO cells transiently expressing different combinations of STtyr-Myc (Golgi) or Iip33-STtyr-Myc (ER-Golgi recycling) together with WT CST (Golgi) or ⌬IIGV CST (ER), were lysed, and cell lysates were subjected to lectin blotting with the ␣2,6-sialic acid-specific lectin, SNA (75,76). We found that all combinations of CST forms and STtyr forms produced SNA-reactive glycoconjugates (Fig.  9). Whereas co-expressing WT CST and STtyr-Myc produced the most SNA-reactive glycoconjugates, cells in which either STtyr or CST was ER-retained or -retrieved produced somewhat less. Not unexpectedly, cells co-expressing ⌬IIGV CST and Iip33-STtyr-Myc produced the least SNA-positive glycoconjugates. The presence of SNA reactive material in the STtyr-Myc-expressing Lec2 CHO cells probably reflects residual CST activity in these cells. Taken together, these results and those in Fig. 8 suggest that the ER-retained ⌬IIGV CST was func- tional but unable to provide sufficient CMP-sialic acid to Golgi sialyltransferases to completely reverse the Lec2 CST defect and completely eliminate PNA staining. In summary, these results demonstrate that the coordinated compartmentation of the CST and sialyltransferases in the Golgi complex is critical for efficient sialylation.

DISCUSSION
Sialyltransferases that function in glycoprotein and glycolipid biosynthesis are localized throughout the Golgi stack. The CST provides CMP-sialic acid to these sialyltransferases and removes CMP, a product of the transfer reaction and inhibitor sialyltransferase activity, from the Golgi lumen. We found in this work that the CST is restricted to the medial-trans cisternae of the Golgi complex in HeLa cells. This restricted sub-Golgi compartmentation does not hamper its ability to completely rescue the Lec2 CHO cell CMP-sialic acid transport defect and restore full sialylation to Lec2 CHO cell glycoconjugates. The CST localization mirrors that of the ST6Gal I STtyr isoform, but our data suggest that this does not reflect the ability of the CST to associate with this or other sialyltransferases.
Analysis of cytoplasmic tail mutants reveals that the CST N-terminal tail is not required for ER export or Golgi localization, whereas its C-terminal tail contains two independent ER export signals. Simultaneous elimination of the dihydrophobic (Ile-Ile) and terminal Val export signals generates a CST protein (⌬IIGV CST) that is localized to the ER and unable to rescue the global sialylation defect in Lec2 CHO cells. Co-expression of ⌬IIGV CST and ST6Gal I in Lec2 CHO cells, which lack both of these activities, reveals that CMP-sialic acid is being transported into the ER and then to the Golgi stacks, where it is used by the ST6Gal I to make SNA-reactive ␣2,6-sialylated glycoproteins. Taken together, these results suggest that CMP-sialic acid is likely to move freely through the Golgi system but that its movement, and that of Golgi-generated CMP, between the ER and Golgi complex is not sufficient to allow an ER-retained CST antiporter to provide sufficient CMP-sialic acid to Golgi glycosyltransferases.
We initially predicted that the CST would form complexes with sialyltransferases and that these functional complexes would make glycoconjugate sialylation more efficient by enhancing the exchange of CMP-sialic acid and CMP between enzyme and transporter. A second related possibility is that CST ER export and/or Golgi localization and subcompartmentation might depend on forming complexes with sialyltransferases. Previous work has identified complexes of Golgi glycosylation enzymes and complexes of glycosyltransferases and their sugar nucleotide transporters. Nilsson et al. (60) first provided evidence for Golgi enzyme association by demonstrating that ER-retained/retrieved Golgi glycosylation enzymes could redistribute partner enzymes that catalyze sequential reactions in the same pathway (e.g. ␣-mannosidase II and N-acetylglucosaminyltransferase I). Glycosyltransferase interactions in glycosaminoglycan and glycolipid biosynthesis pathways have also been identified and predicted to increase the efficiency of these glycosylation pathways. McCormick et al. (77) showed that an interaction between the EXT1 and EXT2 enzymes involved in heparan sulfate biosynthesis is required for their ER export and Golgi function. Pinhal et al. (78) demonstrated that two different enzymes in the heparan sulfate biosynthesis pathway, uronsyl 5-epimerase and 2-O-sulfotransferase, form complexes. Evaluating the organization of enzymes of glycolipid biosynthesis, Giraudo et al. (79,80) provided evidence for two functional enzyme complexes of sequentially acting glycosyltransferases in the early Golgi and late Golgi. More recently, Sprong et al. (53) showed that the UDP-galactose transporter, found in the Golgi of most cells, is relocalized to the ER by the expression of the ERlocalized galactosyltransferase that catalyzes the biosynthesis of galactosylceramide. Other work by Kabuss et al. (81) identified a glycosyltransferase-independent ER localization mechanism for a second splice variant of this transporter FIGURE 8. Activity of the WT and mutant CST proteins. HA-tagged WT and mutant CST proteins were transiently expressed in Lec2 CHO cells. Cells were fixed with ice-cold methanol and stained with FITC-conjugated PNA (green), followed by localization of CST using an anti-HA antibody (red). CST-expressing cells, few in a background of many nonexpressing cells, are indicated by arrows. The upper panel for ⌬C-tail CST highlights cells where the ⌬C-tail protein is found in both the ER and Golgi complex and is able to rescue the Lec2 CST deficiency (⌬C-Tail (ER/Golgi)). In the lower panel for ⌬C-tail CST-expressing, cells are shown to contain the ⌬C-tail mutant only in the ER and do not rescue the Lec2 CST deficiency (⌬C-tail (ER)). Calibration bar, 10 m. All of the proteins were also expressed alone as controls. Cells were lysed with a lysis buffer, resolved by 10% SDS-PAGE, and subjected to lectinblotting analysis using a horseradish peroxidase-conjugated SNA (top panel). The relative expression levels of STtyr and Iip33-STtyr proteins were determined by subjecting an aliquot of each cell lysate to immunoblotting with the anti-Myc antibody (bottom panel). The size difference between STtyr and Iip33-STtyr proteins reflects the replacement of the 9-amino acid cytoplasmic tail of the STtyr with the 33-amino acid cytoplasmic tail of Iip33. Molecular mass markers were as follows: 204 kDa, myosin; 119 kDa, ␤-galactosidase; 100 kDa, bovine serum albumin; 52 kDa, ovalbumin.
(UGT2) involving a di-Lys ER retention/retrieval signal. The latter two examples demonstrate that multiple mechanisms are capable of mediating nucleotide sugar transporter localization.
The signals that direct protein ER export and Golgi localization have been evaluated for both glycosyltransferases and to a lesser extent for nucleotide sugar transporters. Specific signals found in the cytoplasmic sequences of transmembrane proteins have been found to enhance or be required for their ER export. Giraudo and Maccioni (82) found that cytoplasmic sequences rich in Arg and Lys residues adjacent to the membrane function as glycosyltransferase ER export signals. They showed that these signals directly interact with the Sar1 GTPase that is required for COPII coat assembly on ER export sites. Other work has shown that diacidic, dihydrophobic, and Tyr-containing motifs function as ER export signals in various proteins, and some proteins require multiple signals for efficient export (reviewed in Ref. 69). Nufer et al. (70), in their studies of ERGIC-53 ER export, showed that the di-Phe motif in the cytoplasmic tail of this protein can be replaced by a single Phe or Tyr at position Ϫ2, two hydrophobic residues (e.g. Leu, Val, or Ile) at positions Ϫ1 and Ϫ2, or a single Val at position Ϫ1. Each of these signals is capable of interacting with specific COPII coat components (Sec23 and Sec24 proteins, p125). Interestingly, although the other motifs functioned to enhance the efficiency of ERGIC ER export, only the C-terminal Val motif functioned as a true export signal and could mediate the ER export of a reporter protein. The di-Ile and C-terminal Val ER export signals found in the cytoplasmic tail of the CST are similar to those identified by Nufer et al. (70), and both independently stimulate CST ER export sufficiently to rescue the Lec2 CHO cell CST deficiency.
The difference in the ER-to-Golgi trafficking of the ⌬C-tail CST lacking 20 C-terminal amino acids and the ⌬IIGV CST lacking only 4 amino acids at the very C terminus was puzzling. Both proteins lacked the two ER export signals, but the ⌬C-tail CST exhibited some Golgi localization, whereas the ⌬IIGV did not. One possibility is that the CST C-tail may contain sequences that function as a weak ER retention signal, so that in the absence of the export signals (⌬IIGV CST), the CST is retained in the ER. In the absence of both the putative retention and export signals, as in the case of the ⌬C-tail CST, the protein is allowed to leave the ER in a bulk-flow, receptor-independent fashion. Gao and Dean (83) demonstrated that the N-terminal cytoplasmic tail sequences of the yeast GDP-Man transporter (Vrg4) are involved in its ER export. They showed that deletion of the 44-amino acid N-terminal cytoplasmic tail (Vrg4 ⌬44N), but not the first 15 amino acids of this region (Vrg4 ⌬15N), led to retention of Vrg4 in the ER. Other deletion mutants show that the most C-terminal transmembrane region of Vrg4 is required for oligomerization and stability, whereas its 13-amino acid C-terminal cytoplasmic tail is not involved in oligomerization, stability, or localization. Interestingly, work performed by Abe et al. (84) suggests that the C-terminal cytoplasmic tail of this transporter contains a Lys-rich ER retention/retrieval signal that interacts with COPI coat components and is necessary for preventing trafficking of the transporter to the vacuole for degradation.
A new mutation in the GDP-fucose transporter has been identified in a leukocyte adhesion deficiency II/CDG IIc patient by Helmus et al. (85). This mutation leads to premature termination of the protein and the loss of the 10th transmembrane domain and C-terminal cytoplasmic tail of the transporter. This form of the transporter is localized in the ER and inactive. In this case, ER export requires the 10th transmembrane domain rather than C-terminal tail sequences.
The signals and mechanisms mediating Golgi protein localization have been studied intensively for glycosyltransferases (reviewed in Refs. 63, 66, and 67). Recent work from our laboratory suggests that multiple signals are likely to be involved in Golgi localization of the ST6Gal I isoforms (54). We predict that the enzyme's transmembrane region may mediate a lipid partitioning event that prevents the protein from exiting the late Golgi stacks as predicted by the bilayer thickness model (86), that its cytoplasmic tail sequences may be involved in coat protein interactions that cluster the enzymes for the retrograde transport process predicted by the cisternal maturation model (87), and that oligomerization is a secondary event that relies on the concentration of the enzyme by the other two mechanisms and, in turn, enhances these mechanisms and potentially leads to retention. Our finding that the overexpressed CST does not exhibit detectable extra-Golgi localization (ER, plasma membrane, lysosome) suggests that its trafficking and localization does not depend on low abundance proteins, such as the sialyltransferases. The restricted localization of the CST in the medial-trans Golgi and its inability to stably associate with later-acting sialyltransferases (ST6Gal I STtyr isoform or ␣2,8-polysialyltransferase) suggest that its localization is mediated by mechanisms independent of these enzymes. We cannot, however, rule out the possibility that weak, Golgispecific interactions between the CST and sialyltransferases do function to lead to more efficient exchanges of CMPsialic acid and CMP between enzymes and the transporter. Since the cytoplasmic tails of the transporter are not obviously involved in mediating Golgi localization per se, and we see no evidence of pH-dependent insoluble oligomers similar to those that we observe for the stably localized ST6Gal I STcys isoform (data not shown), we suspect that the CST transmembrane regions or cytoplasmic loops are the critical determinants for CST Golgi localization.
The restricted Golgi localization of the CST implies that imported CMP-sialic acid is likely to move freely throughout the Golgi stack. Recent evidence that the Golgi cisternae are more interconnected than once believed (88) is consistent with this premise and eliminates the need for precise cisternal cocompartmentation of the CST and sialyltransferases. The ability of an ER-retained CST protein to supply limited amounts of CMP-sialic acid to a Golgi sialyltransferase also suggests that CMP-sialic acid is able to move between the ER and Golgi complex. However, the inability of this CST mutant to rescue global sialylation in Lec2 CHO cells contrasts with the ability of the wild type, Golgi-localized CST to do so and suggests that the transport of CMP-sialic acid and CMP between the ER and Golgi complex is not efficient. These data highlight the importance of maintaining adequate levels of Golgi CMP-sialic acid for optimal glycoconjugate sialylation and the critical role CST Golgi localization plays in this process.