A Disulfide-bonded Dimer of the Golgi (cid:98) -Galactoside (cid:97) 2,6-Sialyltransferase Is Catalytically Inactive yet Still Retains the Ability to Bind Galactose*

The (cid:97) 2,6-sialyltransferase is a terminal glycosyltransferase localized in the trans Golgi and trans Golgi net- work. Here we show that 30% of the total rat liver Golgi (cid:97) 2,6-sialyltransferase forms a disulfide-bonded 100-kDa species that can be converted to the 50-kDa monomer form of the enzyme upon reduction. Limited proteolysis of both enzyme forms demonstrates that the 100-kDa species is a disulfide-bonded homodimer of the (cid:97) 2,6-sialyltransferase. The (cid:97) 2,6-sialyltransferase disulfide-bonded dimer is found in bovine liver Golgi membranes and in Golgi membranes prepared and solubilized in the presence of 100 m M iodoacetamide, suggesting that it is not unique to rat liver or formed aberrantly upon membrane lysis. The dimer form of the enzyme possesses no significant catalytic activity and has a much lower affinity for CDP-hexanolamine-agarose compared with the monomer form. In contrast, both the (cid:97) 2,6-sialyl-transferase monomer and the disulfide-bonded dimer bind strongly to galactose and galactose-terminated substrates. These results suggest that the (cid:97) 2,6-sialyl-transferase disulfide-bonded dimer lacks catalytic ac- tivity due to a weak affinity for its sugar nucleotide donor, CMP-NeuAc, and that this catalytically inactive form of the enzyme may act as a galactose-specific

The oligosaccharide structures found on proteins have been shown to be important for their correct folding, stability, and biological function (reviewed in Refs. 1 and 2). Specific oligosaccharide structures are also found to mediate cell-cell interactions during development and disease (reviewed in Refs. [3][4][5]. The activities of the terminal Golgi glycosyltransferases, including the N-acetylglucosaminyltransferases, galactosyltransferases, fucosyltransferases, and sialyltransferases, are responsible for the large variation in glycoprotein oligosaccharide structures. The levels of galactosyltransferase and the sialyltransferases are controlled primarily by transcriptional mechanisms (6 -9). Because these glycosyltransferases require specific sugar nucleotide donors and acceptor substrates, another way of controlling their activity is through compartmentation in the Golgi cisternae. The Golgi glycosyltransferases of the rat hepatocyte are localized across the Golgi stacks in the order in which they act to add sugar residues to the oligosac-charide chains of the nascent protein (10). This arrangement ensures that each enzyme has access to the correct substrates and sugar nucleotide donors. In other cell types, including cancer cells and immortalized cell lines, this strict compartmentation breaks down and many enzymes are found in overlapping cisternae (11)(12)(13). How this organization influences the type of oligosaccharide structures observed on the cell surface is not known.
The signals and mechanisms that mediate the localization of proteins throughout the secretory pathway have been studied intensely over the past several years (reviewed in Refs. 14 and 15). Unlike the localization signals of soluble and membraneassociated endoplasmic reticulum (ER) 1 proteins that are composed of specific linear amino acid sequences, the localization signals of Golgi glycosyltransferases are much more complex (16). Although some studies have concluded that the primary Golgi localization signals are found in the transmembrane regions of these enzymes (17)(18)(19), other studies have suggested that sequences in the cytoplasmic tails, transmembrane regions, and luminal regions are required for efficient Golgi localization (20 -24).
Several different mechanisms of Golgi protein localization have been suggested (25)(26)(27). A receptor-mediated retention of the glycosyltransferases has been ruled out by the observation that the overexpression of these proteins does not saturate a potential retention receptor leading to their expression on the cell surface (20). Bretscher and Munro (26) suggested the possibility that the relatively short transmembrane regions of Golgi proteins might prevent their partitioning into cholesterol-rich transport vesicles destined for the plasma membrane. Although early observations favored this model (22), more recent experiments (21) and the observation that the transmembrane regions of Golgi and plasma membrane proteins overlap in length suggest that this cannot be the sole mechanism for Golgi retention.
Because several Golgi proteins require sequences in their cytoplasmic tails, transmembrane regions, and luminal domains for their Golgi localization (20 -24), it seems likely that their localization signals are conformation-dependent. From this standpoint, one mechanism that seems attractive is oligomerization. In this model, the environment of a specific cisterna would induce an oligomerization of one protein or a set of colocalized Golgi proteins. Work by Nilsson et al. (28) has suggested that glycosyltransferases that overlap in localization may be able to form specific complexes leading to their retention. Other investigators studying early Golgi proteins, such as the M protein of the infectious bronchitis virus (29), have presented results that suggest that the formation of insoluble oligomers correlates with the Golgi retention of these proteins.
Little is known about the oligomerization status of the ␤-galactoside ␣2,6-sialyltransferase (␣2,6-ST), a terminal glycosyltransferase localized in the trans Golgi and trans Golgi network. Radiation inactivation experiments performed by Fleischer et al. (30) suggested that the active form of the ␣2,6-ST in Golgi membranes is a dimer, although these experiments cannot rule out the formation of larger oligomers. Here we report that approximately one third of the total ␣2,6-ST in rat and bovine liver Golgi membranes is found as an inactive, disulfide-bonded dimer. The disulfide-bonded dimer form of the ␣2,6-ST exhibits a low affinity for the sugar nucleotide donor, CMP-NeuAc. As a result, this form has little to no sialyltransferase activity. Interestingly, both the ␣2,6-ST monomer and disulfide-bonded dimer have similar affinities for galactose and galactose-terminated substrates. These results suggest that the ␣2,6-ST disulfide-bonded dimer may act as a galactosespecific lectin in the Golgi.

Materials
Tissue culture media and reagents, including Dulbecco's modified Eagle's medium, were purchased from Life Technologies, Inc. Fetal bovine serum was obtained from Irvine Scientific, Irvine, CA. FTO2B rat hepatoma cells were obtained from Dr. Carolyn Bruzdzinski (University of Illinois, Chicago, IL). Male Sprague-Dawley rats were obtained from Harlan (Indianapolis, IN). Nitrocellulose membranes were purchased from Schleicher & Schuell. The enhanced chemiluminescence (ECL) detection kit was obtained from Amersham Corp. Reagents for alkaline phosphatase development of immunoblots were purchased from Promega Corporation (Madison, WI). Protein molecular weight standards were purchased from Bio-Rad. Endoglycosidase H (Endo H) and V. cholerae neuraminidase were purchased from Boehringer Mannheim. CDP-hexanolamine-agarose was purchased from Genzyme (Cambridge, MA). Peptide-N-glycosidase F was purchased from Oxford Glycosystems (Rosedale, NY). 35 S-Express protein labeling mix was purchased from DuPont NEN. High molecular weight electrophoresis standards used to calibrate the glycerol gradients were purchased from Pharmacia Biotech Inc. Iodoacetamide, ␤-mercaptoethanol (BME), 6-aminohexyl ␤-D-thiogalacto-pyranoside agarose (galactose-Sepharose), asialofetuin, Staphylococcus aureus strain V8 protease, and all other chemicals were purchased from Sigma.

Methods
Subcellular Fractionation-Golgi, smooth ER, and rough ER membranes were prepared from rat and bovine liver as described previously by Fleischer and Kervina (31). ER and Golgi membranes were stored in 20 mM phosphate buffer, pH 7.1, containing 20% glycerol, 10 g/ml leupeptin, and 10 g/ml aprotinin.
Affinity Purified Anti-␣2,6-ST Antibodies-For all immunoblot analyses and immunoprecipitations, affinity purified rabbit anti-rat liver ␣2,6-ST antibodies were used. These antibodies were purified on columns of soluble ␣2,6-ST catalytic domain conjugated to Sepharose beads. The soluble ␣2,6-ST catalytic domain was obtained from the media of Chinese hamster ovary cells expressing the catalytic domain of the enzyme fused to a cleavable signal peptide (sp-ST) (32). This sp-ST protein was purified by CDP-hexanolamine-agarose affinity chromatography (32) prior to conjugation to the Sepharose beads.
SDS-Polyacrylamide Gel Electrophoresis and Immunoblot Analysis-Golgi membranes were solubilized in the presence of 100 mM iodoacetamide with 1.0% Triton X-100 for 30 min on ice, mixed with Laemlli sample buffer (33), and electrophoresed on 10% or 12.5% SDSpolyacrylamide gels. For nonreducing conditions, BME was omitted from the sample buffer and samples were loaded on the gel without heating. For reducing conditions, 1-10% BME was included and the sample heated to 100°C for 5 min before loading. Following electrophoresis, proteins were electrophoretically transferred to nitrocellulose membranes and processed for immunoblotting according to the Amersham protocol. Primary antibody was diluted 1:500 and secondary antibody, goat anti-rabbit IgG conjugated to alkaline phosphatase or horseradish peroxidase, was diluted 1:6,000 or 1:10,000, respectively. Immunoblots were developed using either the alkaline phosphatase method from Promega or the ECL method from Amersham Corp.
Prestained broad range protein standards (Bio-Rad) were used to estimate protein molecular masses. Each lot of protein standards differed in apparent molecular mass. Molecular mass ranges for standards were 205-208 kDa for myosin; 115-123 kDa for ␤-galactosidase; 79 -85 kDa for bovine serum albumin (BSA); 49.5-50.3 kDa for ovalbumin; 33.3-34.2 kDa for carbonic anhydrase; 27.8 -28.5 kDa for soybean trypsin inhibitor; and 18.9 -19.4 kDa for lysozyme. Quantitation of protein bands on immunoblots was performed by densitometry scanning. Films exposed to the chemiluminescent blots were scanned using a Color One Scanner (Apple Computer) connected to a Macintosh IIvx using the Ofoto R program (version 2.0) from Light Source Computer. Densitometric analysis was performed using Scan Analysis program (version 2.20) from Specom Research. To determine the relative amounts of monomer and dimer in the Golgi and ER membranes, standard curves of 16 -160 g of total membrane protein concentrations were analyzed by immunoblotting and densitometry. Densitometry values for monomer and dimer were plotted individually versus total protein concentration, and the ratio of monomer to dimer was determined at lower protein concentrations (16,32, and 48 g/ml) using values where both curves were linear.
Limited Protease Digestion-Limited protease digestion of the two ␣2,6-ST forms was performed as described previously (34). FTO2B rat hepatoma cells were metabolically labeled with 100 Ci/ml 35 S-Express protein labeling mix (DuPont NEN) for 6 h, solubilized, and the ␣2,6-ST immunoprecipitated (32). The 50-kDa monomer and 100-kDa disulfidebonded dimer from either rat liver Golgi membranes or radiolabeled cells were separated on nonreducing SDS-polyacrylamide gels, and the bands were excised from the gel. The gel slices were minced and incubated in 0.125 M Tris-HCl, pH 6.8, 0.5% SDS, 10% glycerol, and 0.001% bromphenol blue for 16 h at 37°C to elute the proteins. The eluted proteins were digested for 90 min with 1-200 g/ml of V8 protease at 37°C. The final concentrations of SDS and BME were adjusted to 2 and 10%, respectively. Proteolysis was then stopped by heating samples for 5 min at 100°C prior to SDS-polyacrylamide gel electrophoresis of the peptide mixtures. Radiolabeled peptides were visualized by fluorography and exposure to x-ray film at Ϫ80°C (32), whereas unlabeled peptides were visualized by immunoblotting, as described above.
Estimation of Native Molecular Mass-A 4.2-ml 12-35% continuous glycerol gradient (100 mM KCl, 0.1% Triton X-100 in 20 mM phosphate buffer, pH 7.1) was formed using a Hoeffer gradient maker in a Beckman SW60Ti ultracentrifuge tube. Golgi membranes were solubilized in 100 mM iodoacetamide and 1.0% Triton X-100 for 30 min on ice and applied to the gradient. The gradient was centrifuged at 55,000 rpm for 20 h at 4°C. 200-l fractions were collected from the gradient tube using a Buchler Instruments Auto Densi-Flow IIC. Protein standards (BSA (67 kDa) and lactate dehydrogenase (140 kDa)) were applied to identical gradients and detected by SDS-polyacrylamide gel electrophoresis and silver staining. The presence of the ␣2,6-ST in gradient fractions was determined by nonreducing SDS-polyacrylamide gel electrophoresis and immunoblot analysis, as described above.
Sialyltransferase Assays-Sialyltransferase assays were performed as described previously (35). CMP-[ 14 C]NeuAc was used as the sugar nucleotide donor and asialofetuin, asialotransferrin, or asialo-␣1-acid glycoprotein were used as acceptors. The reactions were carried out at 37°C for 2 h and stopped by freezing the reaction mixture in a dry ice-methanol bath. [ 14 C]NeuAc-labeled acceptor was separated from free CMP-[ 14 C]NeuAc by Sephadex G-50 gel filtration chromatography. Picomoles of NeuAc transferred per h/l of membrane preparation were calculated based on a specific activity of 7200 cpm/nmol CMP-NeuAc.
CDP-Hexanolamine-Agarose Affinity Chromatography-Affinity chromatography of solubilized rat liver Golgi membranes was performed on a small column (1 ml) of CDP-hexanolamine-agarose essentially as described previously (36,37). Columns were equilibrated with buffer E (10 mM sodium cacodylate, pH 6.5, 0.1% Triton CF-54, 0.25 M NaCl). Golgi membranes were solubilized in 0.5% Triton CF-54 for 30 min on ice and applied to the column (100 l diluted in 500 l of buffer E). Columns were washed with buffer E and buffer H (20 mM sodium cacodylate, pH 5.3, 0.1% Triton CF-54, 0.15 M NaCl) and eluted with buffer H containing 5 mM CDP. Fractions of 300 l were collected, and the presence of ␣2,6-ST in each fraction was determined by immunoblot analysis.
Galactose-Sepharose and Asialofetuin-Sepharose 4B Affinity Chromatography-Rat liver Golgi membranes were solubilized in 0.5% Triton CF-54 and 100 mM iodoacetamide for 30 min on ice and incubated with galactose-Sepharose or asialofetuin-Sepharose beads for 1 h at 4°C. Following centrifugation, the supernatants were collected, and the were beads washed extensively in 10 mM sodium cacodylate, pH 6.5, 0.1% Triton CF-54, and 0.25 M NaCl; this fraction plus the previous unbound material was collected as wash through (W). The beads were then incubated with 0.5 M galactose in 25 mM sodium cacodylate, pH 5.3, 0.1% Triton CF-54, and 0.15 M NaCl for 1 h at 4°C to elute bound proteins. The flow through plus wash fractions (W), elution fractions (E), and Sepharose beads (B) were electrophoresed on nonreducing SDSpolyacrylamide gels, and the ␣2,6-ST was detected by immunoblotting.
Endoglycosidase and Neuraminidase Digestions-Golgi membranes were centrifuged in a Beckman airfuge for 20 min at 35 psi. The pellets were resuspended in 20 l of either Endo H digestion buffer (0.1 M sodium citrate, pH 6.0, 0.075% SDS) or peptide N-glycosidase F digestion buffer (0.2 M sodium phosphate, pH 8.6, 1.25% Nonidet P-40) or 40 l of neuraminidase digestion buffer (0.1 M sodium acetate, pH 5.5, 0.05 mg/ml BSA). 10 milliunits of Endo H, 1 units of peptide N-glycosidase F, or 10 milliunits of V. cholerae neuraminidase were added to the reaction mixtures and incubated at 37°C for 2 h (neuraminidase) or overnight. Samples were electrophoresed on a nonreducing 10% SDSpolyacrylamide gel, and the ␣2,6-ST was detected by immunoblotting.

RESULTS
A High Molecular Mass Disulfide-bonded Form of the ␣2,6-ST Is Found in the Golgi-Rat liver Golgi, rough ER, and smooth ER membranes were isolated and electrophoresed on a SDS-polyacrylamide gel under reducing and nonreducing conditions. The ␣2,6-ST was detected in these membrane preparations by immunoblot analysis using an affinity purified anti-␣2,6-ST antibody. Under reducing conditions, the ␣2,6-ST found in the Golgi and ER membranes migrated with the expected molecular mass of ϳ50 kDa (Fig. 1, ϩBME). Under nonreducing conditions, two forms of the ␣2,6-ST were observed in Golgi membranes. In addition to the 50-kDa monomer band, a second band at ϳ100 kDa was detected by the anti-␣2,6-ST antibody ( Fig. 1, ϪBME, Golgi). The molecular mass of the 100-kDa immunoreactive protein band and its disappearance in the presence of reducing agent suggested that it may represent a disulfide-bonded dimer of the ␣2,6-ST. Quantitation of the two ␣2,6-ST forms in the Golgi and ER membranes indicated that the higher molecular mass form of the ␣2,6-ST accounts for ϳ30% of the total enzyme in the Golgi and ϳ6 -7% of the total enzyme in the ER. To rule out the possibility that artifactual disulfide bonds are formed during membrane preparation or lysis, we included 100 mM iodoacetamide in buffers we used in these procedures. Iodoacetamide is an alkylating agent that should react with free, reactive sulfhydryl groups that could potentially form aberrant intermolecular disulfide bonds (38,39). Preparation of Golgi membranes in the presence and the absence of 100 mM iodoacetamide and lysis of membranes in 100, 200, 300, and 400 mM iodoacetamide yielded the same ratio of 50 -100-kDa forms and suggested that the disulfide bonding was not an artifact of membrane preparation and lysis (data not shown).
The reduced form of the ␣2,6-ST appeared to be very sensitive to proteolytic degradation, especially if the sample was stored in sucrose containing buffers ( Fig. 1, ϩBME, sucrose). This was particularly noticeable in the reduced rough ER and smooth ER membrane fractions. To ensure that the 100-kDa band contains the 50-kDa ␣2,6-ST monomer and is not an antigenically related protein that is being preferentially degraded upon heating under reducing conditions, we separated the 50-and 100-kDa bands on a SDS-polyacrylamide gel, excised these bands, and extracted both proteins from the gel slices by incubating for 16 h at 37°C in Laemlli sample buffer containing BME (Fig. 2). Immunoblot analysis of the extracted proteins demonstrates that the 100-kDa band collapses to the 50-kDa band after heating the sample to 100°C in the presence of BME (Fig. 2, Excised Monomer and Excised Dimer). In addition, simply heating the detergent-solubilized Golgi membranes to 100°C for 5 min (Fig. 2, GM, ϩ100°C, ϪBME) or treating the detergent solubilized Golgi membranes with 10% BME without heating (Fig. 2, GM, Ϫ100°C, ϩBME) do not deplete the amount of the 100-kDa form. These results suggest that the 100-kDa immunoreactive band observed in Golgi membranes contains the 50-kDa ␣2,6-ST monomer.
To determine whether the 100-kDa form of the ␣2,6-ST is a dimer of two identical ␣2,6-ST monomers, we performed limited proteolysis of both enzyme forms (Fig. 3). First, rat liver Golgi membrane 50-(M) and 100-kDa (D) immunoreactive bands were excised from gels and subjected to proteolysis with 10 -200 g/ml V8 protease, as described previously (34). Immunoblots of the resulting peptide maps were identical, confirming that the 100-kDa immunoreactive band contains the FIG. 1. A portion of the Golgi ␣2,6-ST forms a higher molecular mass disulfide-bonded form. Golgi, rough ER (rER), and smooth ER (sER) membranes were purified from rat liver according to the method of Fleischer and Kervina (31). Preparations were performed in the presence of 100 mM iodoacetamide to prevent aberrant disulfide bond formation. Membrane lysates were electrophoresed on SDS-polyacrylamide gels under reducing and nonreducing conditions. The gels were then transferred to nitrocellulose filters, and an affinity purified anti-␣2,6-ST antibody was used to detect the ␣2,6-ST by immunoblotting. All reduced samples contained 10% BME and were heated at 100°C for 5 min (ϩBME). Sucrose (0.25 M) and glycerol (20%) refer to the storage conditions for each membrane sample.
FIG. 2. The 100-kDa immunoreactive band contains the ␣2,6-ST monomer. Isolated rat liver Golgi membranes (GM) were solubilized and were prepared for SDS-polyacrylamide gel electrophoresis with (ϩBME) or without (ϪBME) 10% BME and with (ϩ100°C) or without (Ϫ100°C) heating to 100°C for 5 min. In addition, the 50-kDa form of the ␣2,6-ST (Excised Monomer) and the 100-kDa form of the ␣2,6-ST (Excised Dimer) were excised from the SDS-polyacrylamide gel and eluted with sample buffer containing 10% BME for 16 h at 37°C, and recovered proteins were prepared for electrophoresis as described previously. Following electrophoresis, ␣2,6-ST proteins were detected by immunoblotting as described under "Experimental Procedures." ␣2,6-ST monomer (Fig. 3A). Next, FTO2B rat hepatoma cells were metabolically labeled for 6 h with 35 S-Express protein labeling mix, and the ␣2,6-ST was immunoprecipitated from cell lysates. Immunoprecipitates were electrophoresed on a SDS-polyacrylamide gel under nonreducing conditions, and radiolabeled bands corresponding to the ␣2,6-ST 50-(M) and 100-kDa (D) forms were excised. Following a 10-g/ml V8 protease digestion of both the radiolabeled monomer and unreduced 100-kDa disulfide-bonded bands, we found that the 100-kDa disulfide-bonded form of the enzyme was more completely digested than was the monomer form and looked very similar to the immunoblots of the rat liver Golgi ␣2,6-ST digested with 200 g/ml V8 protease (in Fig. 3, compare B, left side (D) with A, 200 g/ml protease). In contrast, digestion of the radiolabeled ␣2,6-ST monomer form with 10 g/ml V8 protease led to a peptide pattern similar to that observed for the rat liver Golgi enzyme digested at lower protease levels (in Fig. 3, compare B, left side (M) with A, 10 and 50 g/ml protease). Because the 100-kDa form of the ␣2,6-ST was very difficult to recover by immunoprecipitation, we believe that this result simply reflects differing amounts of protein added to the proteolytic digests of the 50-and 100-kDa proteins. Using higher amounts of V8 protease to digest the more abundant 50-kDa monomer form, we found that the 50-kDa band was digested into a pattern similar to that of the 100-kDa form of the enzyme (in Fig. 3, compare the left side of B (D) with the right side of B, identical bands are indicated by asterisks). No obvious unique peptide bands were observed in the digestion pattern of the ␣2,6-ST 100-kDa form, suggesting that this form is comprised of only the ␣2,6-ST. These results strongly suggest that the 100-kDa form is a homodimer of two ␣2,6-ST 50-kDa monomers.
To ensure that the presence of the ␣2,6-ST disulfide-bonded dimer is not a rat liver-specific phenomenon, we purified bovine liver Golgi membranes and subjected these to immunoblot analysis under reducing and nonreducing conditions. The immunoblot analysis shows that the bovine liver ␣2,6-ST is also found as a high molecular mass form that is sensitive to BME reduction (Fig. 4). We also were able to detect the dimer form of the ␣2,6-ST in Cos-1 cells expressing the exogenous enzyme and in FTO2B and H-4-II-E rat hepatoma cells (Fig. 3 and data  not shown). Notably, the 100-kDa immunoreactive protein was only observed in Cos-1 cells transfected with ␣2,6-ST cDNA, demonstrating that the presence of the 100-kDa protein depends on the expression of the 50-kDa ␣2,6-ST monomer (data not shown). These data suggest that a significant amount of the ␣2,6-ST exists as a 100-kDa disulfide-bonded dimer in the  (34). Digestion patterns were visualized by immunoblotting. B, left side, FTO2B rat hepatoma cells were metabolically labeled for 6 h with 100 Ci/ml of 35 S-Express protein labeling mix, the ␣2,6-ST monomer (M) and 100-kDa disulfide-bonded form (D) were immunoprecipitated, and these bands excised from a nonreducing SDS-polyacrylamide gel. Proteins were eluted from the gel slices as described (34) and incubated with 10 g/ml V8 protease for 90 min at 37°C. Proteolysis was stopped by bringing the samples to final concentrations of SDS and BME of 2 and 10%, respectively, and heating for 5 min at 100°C. Proteolytic peptides were separated on 12.5% polyacrylamide gels and visualized by fluorography (32). Molecular masses of bands with asterisks (from the largest peptide) were 32, 30, 29, and 24.5 kDa. Right side, immunoprecipitated radiolabeled ␣2,6-ST monomer was subjected to proteolysis with 10 -200 mg/ml V8 protease, and peptides were analyzed by polyacrylamide gel electrophoresis and fluorography as described above. The asterisks indicate peptides common to both the monomer and disulfide-bonded forms of the enzyme. Molecular masses of bands with asterisks (from the largest peptide) were 32.5, 30, 29, and 24 kDa. Molecular mass standards indicated by hash marks in B are identical to those in A.

FIG. 4. The disulfide-bonded form of the ␣2,6-ST is also found in bovine liver Golgi membranes.
Bovine liver Golgi membranes were isolated and detergent solubilized as described under "Experimental Procedures." Prior to SDS-polyacrylamide gel electrophoresis, the bovine Golgi membrane lysates were prepared with (ϩBME) or without (ϪBME) 10% BME and with (ϩ100°C) or without (Ϫ100°C) heating for 5 min at 100°C. The ␣2,6-ST was detected by immunoblotting.
Golgi complex in the liver of different species and in different cell types.
Analysis of the Native Molecular Mass of the Monomer and Disulfide-bonded Dimer Forms of the ␣2,6-ST-To further characterize the native molecular mass of the ␣2,6-ST in Golgi membranes, we performed glycerol gradient sedimentation analysis. Rat liver Golgi membranes were solubilized and then loaded on a 12-35% continuous glycerol gradient. Gradient fractions were collected and electrophoresed on 10% SDS-polyacrylamide gels, and the ␣2,6-ST was detected by immunoblotting. We found that the monomer form of the enzyme migrates slower than BSA (67 kDa) on the glycerol gradient. This sedimentation position correlates well with its 50-kDa molecular mass determined by SDS-polyacrylamide gel electrophoresis (Fig. 5). Likewise, the ␣2,6-ST disulfide-bonded dimer appears between BSA (67 kDa) and lactate dehydrogenase (140 kDa) on the glycerol gradient, and this sedimentation position also correlates well with its ϳ100-kDa molecular mass determined by SDS-polyacrylamide gel electrophoresis (Fig. 5). Similar results were obtained using sucrose density gradients (data not shown). Using either glycerol or sucrose gradients, we were unable to detect any higher molecular mass oligomers of the ␣2,6-ST monomer or disulfide-bonded dimer in rat liver Golgi membranes. However, we cannot rule out the possibility that membrane lysis using 1.0% Triton X-100 breaks down weakly associated ␣2,6-ST oligomers.
The ␣2,6-ST Disulfide-bonded Dimer Possesses Little to No Catalytic Activity-To begin to understand the role of the two ␣2,6-ST forms in the Golgi, we performed a series of experiments to determine whether these two species differ in their catalytic activity, affinity for substrates and sugar nucleotide donors, or glycosylation. Rat liver Golgi membranes were fractionated on glycerol gradients, as described above. Sialyltransferase activity assays were performed on these glycerol gradient fractions, and activity was compared with levels of ␣2,6-ST monomer and dimer determined by immunoblot analysis of identical gradient fractions. Densitometric analysis of immunoblots produced two overlapping peaks of ␣2,6-ST immunoreactive material corresponding to the monomer and dimer forms (Fig. 6, top panel). Activity assays using [ 14 C]CMP-NeuAc as donor and asialofetuin as substrate revealed only a single activity peak corresponding to the monomer form of the enzyme (Fig. 6, lower panel). Similar activity assays were performed using asialotransferrin and asialo-␣1-acid glycoprotein with the same results (data not shown). Because asialofetuin can be used as an acceptor for both ␣2,6and ␣2,3-sialyltransferases (N-linked and O-linked) (37) and asialo-␣1-acid glycoprotein can be used as an acceptor for both ␣2,6and ␣2,3-sialyltransferases (N-linked) (35), these data also suggested that the ␣2,6-ST disulfide-bonded dimer does not have altered specificity.
The ␣2,6-ST Disulfide-bonded Dimer Binds to CDP-hexanolamine with Lower Affinity Than Does the Enzyme Monomer-Because the ␣2,6-ST disulfide-bonded dimer is not as catalytically active as the enzyme monomer, it is reasonable to determine whether its affinity for the CMP-NeuAc sugar nucleotide donor or galactose-terminated substrates differs from that of the active monomer. CDP-hexanolamine, which is To determine whether the monomer and dimer forms of the ␣2,6-ST associate into higher molecular mass oligomers, we performed glycerol gradient sedimentation analysis. Rat liver Golgi membranes were solubilized in 1.0% Triton X-100 and 100 mM iodoacetamide on ice for 30 min and then loaded on a 12-35% continuous glycerol gradient. Gradient fractions were collected and electrophoresed on SDS-polyacrylamide gels under nonreducing conditions. The ␣2,6-ST was detected by immunoblot analysis. BSA (67 kDa) and lactate dehydrogenase (140 kDa) were used as molecular mass markers and detected on SDS-polyacrylamide gels by silver staining.
FIG. 6. The ␣2,6-ST dimer is not as catalytically active as the monomer form. Sialyltransferase assays were performed on partially separated ␣2,6-ST monomer and dimer forms to determine whether both forms possessed catalytic activity. Rat liver Golgi membranes were solubilized in 1.0% Triton X-100 and loaded on a 12%-35% glycerol gradient. Gradient fractions were collected and analyzed by immunoblotting and for sialyltransferase activity. Asialofetuin was used as the acceptor substrate in this assay. Top panel, the separation of ␣2,6-ST monomer and dimer forms assessed by densitometric scanning of a film exposed to an immunoblot of gradient fractions developed using ECL reagents. Bottom panel, sialyltransferase activity found in these gradient fractions using asialofetuin as an acceptor substrate. chemically similar to CMP-NeuAc, has been used extensively to purify sialyltransferases (35,36). We used a CDP-hexanolamine-agarose column to test whether the ␣2,6-ST monomer and dimer forms differed in their affinity for this matrix. Solubilized Golgi membranes were applied to columns of CDPhexanolamine-agarose, and unbound material was removed by extensive washing. Specifically bound material was then eluted using 5 mM CDP. Both wash through and eluted fractions were collected and electrophoresed under nonreducing conditions, and the ␣2,6-ST was detected by immunoblotting. The ␣2,6-ST disulfide-bonded dimer is found predominantly in the wash through fractions, whereas the majority of the monomer specifically binds to the CDP-hexanolamine-agarose and is eluted with 5 mM CDP (Fig. 7). These results suggest that the ␣2,6-ST monomer has a higher affinity for CDP-hexanolamine than does the dimer, and it is likely that this affinity would also indicate the relative affinities of these two ␣2,6-ST forms for the sugar nucleotide donor, CMP-NeuAc. These results correlate well with the apparent inactivity of the ␣2,6-ST dimer and suggest that its lack of catalytic activity may be related to its inability to bind CMP-NeuAc with high affinity.
Both the ␣2,6-ST Monomer and Dimer Bind Strongly to Galactose and Galactose-terminated Substrates-To determine whether a low affinity for galactose-terminated substrates also contributed to the ␣2,6-ST disulfide-bonded dimer's lack of sialyltransferase activity, we tested the ability of both enzyme forms to bind to galactose-Sepharose and asialofetuin-Sepharose 4B (Fig. 8). Solubilized Golgi membranes were incubated with 200 l of the affinity beads, extensively washed (W) and eluted with 500 mM galactose (E). Following elution, the affinity beads were incubated with SDS-polyacrylamide gel loading buffer and loaded directly onto the gel to determine how much of the two ␣2,6-ST formed remained on the affinity beads (B). Both the ␣2,6-ST monomer and dimer forms bound very strongly to the galactose-Sepharose and asialofetuin-Sepharose beads (galactose-Sepharose and asialofetuin-Sepharose, E). This binding was so strong that a significant amount of both forms remained on the affinity beads following elution with 500 mM galactose (galactose-Sepharose and asialofetuin-Sepharose, B). Notably, the ratio of monomer to dimer in the wash through fraction (W), elution fraction (E), and the fraction remaining on the beads following elution (B) remained roughly the same at about 2:1. This is the same ratio of ␣2,6-ST monomer to dimer that is observed in the unfractionated Golgi membranes (see Fig. 1) and suggests that the monomer and disulfide-bonded dimer of the enzyme have a similar affinity for galactose and galactose-terminated substrates. From these re-sults and those above, it appears that the formation of the 100-kDa ␣2,6-ST disulfide-bonded dimer disrupts its ability to bind the sugar nucleotide donor, CMP-NeuAc, thus leading to its lack of sialyltransferase activity. In contrast, formation of the ␣2,6-ST dimer does not obviously influence its affinity for galactose and galactose-terminated substrates, suggesting that this form of the enzyme may act as a galactose-specific lectin in the Golgi.
The N-Glycosylation Pattern of the ␣2,6-ST Disulfide-bonded Dimer Is Similar to That of the Enzyme Monomer-An earlier study suggested that the degree of ␣2,6-ST glycosylation can affect the catalytic activity of the enzyme (40). We have previously demonstrated that there are three differentially glycosylated populations of the ␣2,6-ST in rat liver Golgi membranes (20). One population has two complex N-linked oligosaccharides (Endo H-resistant), a second population has one complex and one high mannose N-linked oligosaccharide (partially Endo H-resistant), and a third population has two high mannose N-linked oligosaccharides (Endo H-sensitive). To test the possibility that the ␣2,6-ST molecules that are found in the monomer and dimer forms are glycosylated differently, we treated solubilized rat liver Golgi membranes with Endo H, peptide N-glycosidase F, and V. cholerae neuraminidase and analyzed the treated samples using nonreducing SDS-polyacrylamide gel electrophoresis and immunoblotting. Following digestion with Endo H, which specifically cleaves high mannose N-linked oligosaccharides, the ␣2,6-ST monomer form migrated on the SDS-polyacrylamide gel as the three populations described above (Fig. 9, ϩEndo H). The ␣2,6-ST dimer migrated as several species that may correspond to different disulfide-bonded combinations of the differentially glycosylated monomer forms. Following treatment with peptide N-glycosidase F, which specifically cleaves all the N-linked oligosaccharides, the ␣2,6-ST monomer migrates as a single band with a molecular mass that is identical to the lowest molecular mass form of the Endo H-treated material (Fig. 9, ϩN-glycosidase F). Likewise, the peptide N-glycosidase F-treated disulfide-bonded dimer migrates with a molecular mass identical to the lowest FIG. 7. The ␣2,6-ST dimer binds CDP-hexanolamine-agarose much more weakly than the monomer form. To determine whether the ␣2,6-ST dimer form has altered affinity for its CMP-NeuAc donor, we applied solubilized rat liver Golgi membranes to a 1-ml CDP-hexanolamine-agarose column. The column was washed using buffer E (Wash Through Fractions 1-5) and buffer H (Wash Through Fractions 6 -24) and eluted with buffer H plus 5 mM CDP (5 mM CDP Elution Fractions) (35,36). Fractions were electrophoresed on a SDS-polyacrylamide gel, and the ␣2,6-ST was detected by immunoblotting. FIG. 8. Both the ␣2,6-ST monomer and dimer forms bind strongly to galactose-Sepharose and asialofetuin-Sepharose. To determine whether the ␣2,6-ST monomer and dimer forms have the same affinity for galactose-terminated substrates, we investigated their binding to galactose-Sepharose (galactose-Seph) and asialofetuin-Sepharose 4B (asialofetuin-Seph). Detergent solubilized rat liver Golgi membranes were incubated with 1 ml of each type of affinity beads for 2 h at 4°C. Then the flow through was collected, and the affinity beads were washed extensively in buffer E (flow through and wash through, W). Both sets of affinity beads were rotated with 500 mM galactose in buffer H for 30 min at 4°C (elution fraction, E). The wash through (W), elution (E), and 500 l of the affinity beads (B) were electrophoresed on a nonreducing SDS-polyacrylamide gel, and the ␣2,6-ST in these fractions was detected by immunoblotting. molecular mass form of the Endo H-treated ␣2,6-ST dimer. Treatment with V. cholerae neuraminidase, which cleaves ␣2,3-, ␣2,6-, and ␣2, , led to a more rapid migration of both monomer and dimer forms of the enzyme (Fig. 9, ϩNeuraminidase). We conclude from these results that the low activity of the ␣2,6-ST disulfide-bonded dimer is not due to any type of abnormal N-linked glycosylation. DISCUSSION Studies of the signals required for ␣2,6-ST Golgi localization have led to the identification of sequences in the cytosolic, transmembrane, and luminal regions that play a role in this process (17)(18)(19)(20)(21)(22)(23)(24). The lack of a saturable Golgi retention receptor coupled with the ambiguous retention "signal" suggests that the inherent characteristics of the ST may lead to its specific retention in the trans Golgi and trans Golgi network. Although several hypotheses concerning the mechanism of Golgi retention have been proposed (25)(26)(27), we have chosen to focus on one of these that suggests that the specific environment of the late Golgi causes oligomerization of the ␣2,6-ST and that this leads to its retention. Our studies of ␣2,6-ST oligomerization led to the identification of a 100-kDa form of the enzyme that comprises ϳ30% of the total Golgi ␣2,6-ST. Reduction of the 100-kDa form results in the appearance of the 50-kDa ␣2,6-ST monomer and suggests that this larger form is a disulfide-bonded dimer of the enzyme (Figs. 1 and 2). Limited proteolysis of isolated monomer and dimer forms confirmed that the 100-kDa form is a homodimer of two 50-kDa ␣2,6-ST monomers (Fig. 3). The dimer form of the enzyme is observed in rat and bovine liver, in H-4-II-E and FTO2B rat hepatoma cells, and in Cos-1 cells transfected with ␣2,6-ST cDNA (Figs. 1, 3, and 4 and data not shown), suggesting that dimerization is not a characteristic of a particular species of cell type. Analysis of the catalytic activity of both ␣2,6-ST forms demonstrated that the disulfide-bonded dimer has little to no sialyltransferase activity using asialofetuin, asialotransferrin, or asialo-␣1acid glycoprotein as substrates ( Fig. 6 and data not shown). Substrate and donor binding assays suggest that the disulfidebonded dimer's lack of catalytic activity probably reflects its relatively low affinity for its sugar nucleotide donor, CMP-NeuAc (Fig. 7). However, both the monomer and the disulfidebonded dimer bind very strongly to galactose-Sepharose and asialofetuin-Sepharose, suggesting that the disulfide-bonded dimer's primary activity in the Golgi may be as a galactosebinding lectin (Fig. 8).
What differs between the ␣2,6-ST molecules that remain as monomer and those that form a disulfide-bonded dimer is not clear. Analysis of both monomer and dimer forms with Endo H, peptide N-glycosidase F, and V. cholerae neuraminidase suggest that the pattern of glycosylation is generally the same in the two forms (Fig. 9). One distinct possibility is that the sub-Golgi localization of these two forms differs. The rat liver ␣2,6-ST was previously localized in the trans cisternae of the Golgi and the trans Golgi network by immunoelectron microscopy (42). It is possible that only 30% of the total rat liver ␣2,6-ST reaches the trans Golgi network and in this compartment forms a disulfide-bonded molecule. Unfortunately, at this time we have no antibodies that distinguish the monomer and dimer forms of the enzyme to test this hypothesis.
Our results suggest that the disulfide-bonded dimer's lack of catalytic activity reflects its low affinity for its sugar nucleotide donor, CMP-NeuAc. Consistent with this observation, preliminary experiments suggest that the disulfide bond involved in dimer formation occurs between two catalytic domains of ␣2,6-ST monomers. 2 All the sialyltransferases cloned to date contain a consensus sequence in the catalytic domain called the "sialyl motif" (43). Datta and Paulson have demonstrated that this region is involved in CMP-NeuAc binding (44). Within the sialyl motif is a Cys residue (Cys 181 ) that is conserved in all sialyltransferases (45). Alteration of this residue and other Cys residues in the ␣2,6-ST catalytic domain lead to inactivation of the enzyme (44). 3 It is possible that this Cys residue in the sialyl motif does participate in the disulfide bond formed between two catalytic domains and leads to inactivation of the enzyme. Future studies will focus on identifying the Cys residues involved in the ␣2,6-ST dimer's disulfide bond.
Fleischer et al. (30) used radiation target inactivation analysis to show that the active form of the Golgi ␣2,6-ST is a dimer. Because this technique relies on the inactivation of catalytic activity, we presume that the dimer identified is a noncovalent dimer of ␣2,6-ST monomers and that the inactive, disulfide-bonded dimer we have identified was not detected in this analysis. Taken together, our results and those of Fleischer et al. (30) suggest an in vivo situation where the ␣2,6-ST exists as both covalently and noncovalently associated dimers in the Golgi membrane. Our inability to detect a noncovalently associated dimer form by glycerol gradient or sucrose gradient sedimentation may be simply due to a very weak association of these dimers and again points to the possibility that larger oligomers are also formed but are unstable in the nonionic detergents used during the sedimentation analyses.
Several other Golgi localized proteins have been identified as dimers both biochemically and by radiation target inactivation. Moremen et al. (46) found that the active form of the Golgi ␣-mannosidase II is a disulfide-bonded homodimer. The disulfide bonds of the enzyme are formed by Cys residues in the luminal domain because proteolytic removal of the amino-ter-2 J. Ma and K. Colley, unpublished results. 3 A. Datta, personal communication. minal cytosolic tail and transmembrane region results in a disulfide-bonded, soluble form of the enzyme. A similar disulfide-bonded form has been observed for the G M3 /G D3 ␤1,4Nacetylgalactosaminyltransferase expressed in CHO cells. 4 Interestingly, purified Golgi ␤1,4-galactosyltransferase has been observed to exist as both a monomer and a disulfide-bonded dimer (47), and radiation target inactivation also demonstrated that the active form of this glycosyltransferase is a dimer (30). These observations are remarkably similar to those made by our laboratory and Fleischer et al. (30) for the ␣2,6-ST and suggest that this may be a common situation for glycosyltransferases localized in the late Golgi cisternae.
The presence of two forms of the Golgi ␣2,6-ST in vivo leads us to consider how they are related and what function the enzyme disulfide-bonded dimer might play. One possibility is that the ␣2,6-ST dimer sialylates other substrates or makes other types of anomeric linkages. However, because this form does not bind CMP-NeuAc with high affinity, this is not likely. A second possibility is that this disulfide bond formation is a mechanism of down-regulation that is controlled by the availability of donor or substrate molecules. If the availability of CMP-NeuAc in the Golgi cisternae is limiting, the ␣2,6-ST monomer form (noncovalent dimer) having a higher affinity for CMP-NeuAc will perform the transferase function. In this situation, the disulfide-bonded dimer form of the ␣2,6-ST, having a much lower affinity for CMP-NeuAc, remains essentially inactive but will still be able to bind galactose-terminated substrates. This leads to a third possibility that is that the disulfide-bonded ␣2,6-ST dimer acts as a galactose-specific lectin in the Golgi. This form of the enzyme might act to retain unsialylated molecules and pass them off to the active, noncovalently associated ␣2,6-ST dimers for sialylation. In this way, the disulfide-bonded dimer would be acting essentially as a chaperone molecule by preventing the exit of asialoglycoproteins from the late Golgi. Other lectin-like proteins are thought to play important roles in the secretory pathway (47)(48)(49). For example, the chaperone protein calnexin is also a lectin that recognizes the Glc 1 Man 9 GlcNAc 2 carbohydrate structure on unfolded or misfolded proteins and prevents their exit from the ER (49,50).
We predict that the cellular conditions may control the formation of the ␣2,6-ST disulfide-bonded dimer, its down-regulation, and potential conversion into a galactose-specific lectin. A somewhat similar situation is observed in the control of the dimerization and activity of the heme-regulated eIF-2␣ kinase by levels of heme in reticulocytes (reviewed in Ref. 51). In heme deficiency, the heme-regulated eIF-2␣ kinase is an active noncovalent dimer whose activity will ultimately lead to the inhibition of protein synthesis initiation. However, when levels of heme are high, the noncovalent dimers of heme-regulated eIF-2␣ kinase become covalently associated via disulfide bonds leading to an inactivation of the kinase and allowing the initiation of protein synthesis. In this way protein synthesis will only commence when the reticulocyte has adequate levels of heme. Analogously, low levels of CMP-NeuAc may lead to the formation of inactive disulfide-bonded dimers of the ␣2,6-ST. However, to compensate for a decreased sugar nucleotide donor pool, this covalent dimerization may also convert the enzyme into a galactose-specific lectin can that control the exit of undersialylated glycoproteins from the Golgi.
The significant amount of the ␣2,6-ST disulfide-bonded dimer form in the Golgi and its relative absence from the ER suggest that it may be formed in the Golgi. While studying the biosynthesis of the ␣2,6-ST in dexamethasone-treated H-4-II-E cells, Bosshart and Berger (52) identified a high molecular mass form of the enzyme that appeared after 10 min of pulse labeling and 1-2 h of chase. Analysis of the oligosaccharide structures on the ␣2,6-ST protein suggested that this high molecular mass form was made in the Golgi, and they suggested that it might be an enzyme dimer. Formation of the disulfide-bonded ␣2,6-ST dimer in the Golgi supports the above hypothesis that the ␣2,6-ST dimer may be formed in response to donor or substrate levels in the Golgi. One possibility is that Cys residues are intimately involved in the ␣2,6-ST catalytic mechanism. If some noncovalently associated dimers are not supplied with adequate sugar nucleotide donor or substrate, the reactive sulfhydryl groups in the catalytic domain of one molecule may interact with other reactive sulfhydryl groups in the catalytic domain of another molecule. Only those noncovalently associated dimers that are continuously supplied with donor and/or substrate resist disulfide bond formation and remain catalytically active. In this way enzyme activity and putative galactose-specific chaperone activities could respond to cellular conditions and availability of substrate and/or donor molecules. Further experiments must be performed to study the location and mechanism of ␣2,6-ST disulfide-bonded dimer formation and determine the role the dimer form of this enzyme plays in the process of N-linked glycosylation.