β1,4-Galactosyltransferase I (GalT I) exists in two subcellular compartments where it performs two distinct functions. The majority of GalT I is localized in the Golgi complex where it participates in glycoprotein biosynthesis; however, a small portion of GalT I is expressed on the cell surface where it functions as a matrix receptor by binding terminal N-acetylglucosamine residues on extracellular glycoside ligands. The GalT I polypeptide occurs in two alternate forms that differ only in the length of their cytoplasmic domains. It is thought that the longer cytoplasmic domain is responsible for GalT I function as a cell surface receptor because of its ability to associate with the detergent-insoluble cytoskeleton. In this study, we demonstrate that the long GalT I cytoplasmic and transmembrane domains are capable of targeting a reporter protein to the plasma membrane, whereas the short cytoplasmic and transmembrane domains do not have this property. The surface-localized GalT I reporter protein partitions with the detergent-insoluble pool, a portion of which co-fractionates with caveolin-containing lipid rafts. Site-directed mutagenesis of the cytoplasmic domain identified a requirement for serine and threonine residues for cell surface expression and function. Replacing either the serine or threonine with aspartic acid reduces surface expression and function, whereas substitution with neutral alanine has no effect on surface expression or function. These results suggest that phosphorylation negatively regulates GalT I function as a surface receptor. Consistent with this, phosphorylation of the endogenous, full-length GalT I inhibits its stable expression on the cell surface. Thus, the 13 amino acid extension unique to the long GalT I isoform is required for GalT I expression on the cell surface, the function of which is regulated by phosphorylation.
Multicellular organisms utilize a variety of cell surface receptors to interact with the extracellular environment. The best-characterized of these is the integrin family (Giancotti and Ruoslahti, 1999; Schoenwaelder and Burridge, 1999), although other classes of matrix receptors have been described, including dystroglycans (Durbeej et al., 1998; Ervasti and Campbell, 1993), syndecans (Iba et al., 2000; Rapraeger, 2000), discoidin domain receptor tyrosine kinases (Schlessinger, 1997; Shrivastava et al., 1997; Vogel et al., 1997), leukocyte antigen-related protein (LAR) receptor-like tyrosine phosphatase (O'Grady et al., 1998; Schaapveld et al., 1997), and cell surface-associated β1,4-galactosyltransferase I (GalT I) (Evans et al., 1993; Gong et al., 1995; Shur et al., 1998).
GalT I is one of six known β4-galactosyltransferase polypeptides that transfer galactose in a β1,4 linkage from uridine diphosphate galactose (UDP-Gal) to specific glycoprotein and glycolipid substrates (Amado et al., 1999). The vast majority of GalT I is present in the Golgi complex where it participates in oligosaccharide synthesis. However, a portion of GalT I is also found on the cell surface where it functions as a cell adhesion molecule during a variety of cellular interactions by binding N-acetylglucosamine (GlcNAc)-containing oligosaccharide substrates, or ligands, in the extracellular matrix (Shur et al., 1998). Similar to other cell adhesion molecules and matrix receptors, GalT I requires association with the detergent-insoluble pool in order to function as a receptor for extracellular glycoside ligands (Appeddu and Shur, 1994a; Appeddu and Shur, 1994b; Eckstein and Shur, 1992; Evans et al., 1993). Furthermore, ligand-induced GalT I aggregation activates cell-specific intracellular signaling cascades, including transient activation of focal adhesion kinase leading to cytoskeletal reorganization during fibroblast migration (Wassler and Shur, 2000), and heterotrimeric G-protein activation leading to vesicle exocytosis in sperm and Xenopus oocytes (Gong et al., 1995; Shi et al., 2001). Although not as well studied as GalT I, there is evidence supporting the presence of other glycosyltransferases on the cell surface, including a sialyltransferase (Close and Colley, 1998), an N-acetylgalactosaminyltransferase (Mandel et al., 1999), and two fucosyltransferases (Borsig et al., 1996; Marker et al., 2001).
It has been a challenge to determine how a portion of GalT I is routed from the Golgi complex to the plasma membrane and how it functions as a receptor during cellular interactions. The level of GalT I on the cell surface is differentially regulated from that in the Golgi pool and is not the result of simple bulk flow of Golgi-derived vesicles to the cell surface (Lopez et al., 1989). This suggests that the Golgi and surface pools of GalT I result from distinct protein species and/or molecular pathways (reviewed by Shur et al., 1998). Insight into this issue came from the finding that the GalT I gene encodes two RNA transcripts that differ in the length of their 5′ ends, and which result in two GalT I polypeptides that differ only in the length of their N-terminal cytoplasmic domains (Russo et al., 1990; Shaper et al., 1988). Both GalT I proteins are type II integral membrane proteins with identical signal sequence/transmembrane domains and C-terminal catalytic domains. However, the cytoplasmic domain of the shorter protein is 11 amino acids in length, whereas that of the longer protein is 24 amino acids in length.
The synthesis of two GalT I polypeptides with distinct cytoplasmic domains suggested a mechanism to account for its dual subcellular distribution and function (Dinter and Berger, 1995; Lopez et al., 1991; Nilsson et al., 1991; Shur et al., 1998; Teasdale et al., 1992; Youakim et al., 1994). In this regard, previous results show that both GalT I isoforms can function biosynthetically in the Golgi complex. However, only the long GalT I isoform can function as a receptor on the cell surface because of its association with the detergent-insoluble cytoskeleton (Evans et al., 1993; Wassler et al., 2001; Wassler and Shur, 2000). The long cytoplasmic domain may influence GalT I subcellular distribution because of its increased length (Munro, 1995), or specific sequences within the long cytoplasmic domain may override putative Golgi retention signals thought to reside in the transmembrane domain (Colley, 1997).
It has previously been reported that expression of the GalT I cytoplasmic and transmembrane domain produces a dominant-negative phenotype by displacing the endogenous GalT I from its detergent-insoluble cytoskeleton. This indirectly illustrates the ability of the long cytoplasmic domain to function at the cell surface (Evans et al., 1993). In this study, we took a more direct approach by asking if the cytoplasmic and transmembrane domains of long GalT I are able to target a reporter construct to the cell surface, and if so, if mutating specific amino acid residues within the cytoplasmic domain abrogates cell surface expression and/or function. Results show that the long GalT I cytoplasmic domain, coupled with the transmembrane domain, is capable of transporting a reporter protein to the cell surface where it associates with the detergent-insoluble cytoskeleton. In contrast, the short GalT I cytoplasmic domain does not have this activity. We also determined that a portion of the GalT I associated with the detergent-insoluble pool reflects partitioning into lipid rafts, where it presumably is in equilibrium with the cytoskeletally associated pool of GalT I. Because others have shown that GalT I is serine-phosphorylated (Strous et al., 1987), we asked if phosphorylation of serine and/or threonine residues in the GalT I cytoplasmic domain influences its surface expression and/or function. Consistent with this, mutating serine 11 or threonine 18 to aspartic acid, which mimics the negative charge of phosphorylation, abolishes surface expression and function, whereas substitution with neutral alanine has no effect. Furthermore, phosphorylation of endogenous, full-length GalT I prevents its surface expression. These results suggest that GalT I expression on the cell surface and function as a matrix receptor is dependent upon the long cytoplasmic domain and is negatively regulated by phosphorylation.
Materials and Methods
Mouse embryo NIH 3T3 fibroblasts (American Type Culture Collection) were routinely cultured in Dulbecco's Modified Eagle Medium (DMEM, Gibco), supplemented with 10% bovine calf serum (BCS; Gibco) and antibiotics at 37°C with 5% CO2. Cells were passaged at ∼70% confluence.
Generation of GalT I-chloramphenicol acetyltransferase and Green Fluorescent Protein fusion proteins
Reporter constructs were produced by fusing the cytoplasmic and transmembrane domains of GalT I, referred to as `truncated GalT I', to chloramphenicol acetyltransferase (CAT). Construction of the truncated long (TLGT) and truncated short (TSGT) GalT I constructs has been previously described (Evans et al., 1993). The TL-CAT and TS-CAT fusions were generated by removing TLGT and TSGT from the EV142 vector (Evans et al., 1993), followed by digestion with Xba I, and cloning into the Xba I site of the pCAT-Basic vector (Promega). The TL-CAT and TS-CAT fusions were removed from pCAT-Basic by digestion with Hind III and Nar I followed by Ban I. As a control, the Xba I/Ban I fragment (728 bp) containing the coding sequence of CAT was purified from the pCAT-Basic vector. The gel-purified fragments were treated with T4 DNA polymerase to generate blunt ends, then each was cloned back into either Xba I-cut, T4 polymerase-blunted EV142 vector [containing the metallothionein I (MT-1)-inducible promoter], or Pst I-cut, blunted pKJ vector [containing the phosphoglycerokinase (PGK) constitutive promoter (Evans et al., 1993)]. Orientation of the fragments was confirmed by restriction endonuclease digestion analysis and by sequencing.
To visualize surface expression of the GalT I cytoplasmic domain fusion proteins, the TLGT and TSGT constructs were fused 5′ to an enhanced green fluorescent protein (EGFP, Clontech) under the control of a cytomegalovirus (CMV) promoter. The GalT I-encoding portions of the original clones were excised by digestion with Spe I, blunted, then digested with Acc I. The 170 bp fragment was gel-purified and subcloned into Acc I-Sma I-digested EGFP. Sequences were verified by restriction analysis and sequencing.
Transfection of GalT I reporter constructs
NIH 3T3 cells were plated 24 hours before transfection at a density of 5500/cm2. Transfection was done in serum-free, antibiotic-free DMEM using LipofectAMINE™ or LipofectAMINE Plus™ (Invitrogen), according to the manufacturer's instructions. For stable transfection and selection, cells were co-transfected with pKJ-neo (Evans et al., 1993). Twenty-four hours after transfection, cells were split 1:5 and stable transfectants were selected in medium containing geneticin (G418, 400 μg/ml, Invitrogen). After 10-14 days, individual G418-resistent colonies were selected and expanded. Expression of the MT-1 promoter was induced with 80 μM ZnSO4.
Transient expression of GalT I reporter constructs was performed by electroporation. Sub-confluent cells were harvested, washed once with DMEM, and resuspended in DMEM at a concentration of 1×107 cells/ml. For each transfection, 1 ml of cell suspension was mixed with 15 μg of DNA and electroporated following the manufacturer's instructions in a Bio-Rad Gene Pulsar Transfection Apparatus (Bio-Rad). The cells were plated onto coverslips in DMEM, and 4 hours later the medium was changed to complete medium containing BCS and antibiotics. The coverslips were used at 48 hours after transfection.
Determination of CAT activity
CAT activity in NIH 3T3 cells transfected with CAT, TL-CAT or TS-CAT was assayed with the CAT Enzyme Assay System (Promega). Briefly, 24-72 hours after LipofectAMINE™ transfection (see above), the cells were washed with phosphate buffered saline (PBS), and harvested by scraping into TEN buffer (40 mM Tris-HCl, pH 7.5, 1 mM EDTA, 150 mM NaCl). Cell pellets were resuspended in 110 μl of 0.25 M Tris-HCl, pH 8.0, and subjected to 3 rapid freeze/thaw/vortex cycles. The lysates were heated at 60°C for 10 minutes to inactivate the endogenous deacetylase activity. Fifteen μl of reaction mixture [0.25 μCi [3H] chloramphenicol (Dupont-NEN) 50 mCi/mmol, 5 μl n-butyryl CoA, 5 mg/ml] was added. A CAT standard curve (supplied in the kit) was used as a positive control and `no cell extract' or mock-transfected cells were used as negative controls. The reactions were incubated at 37°C for 5 hours. Reactions were terminated by adding 300 μl of mixed xylenes (Sigma-Aldrich), vortexed for 30 seconds, then centrifuged for 3 minutes at 16,000 g. The upper xylene phase was transferred to a new tube, 100 μl 0.25 M Tris-HCl, pH 8.0 was added, and the extraction was repeated. The upper xylene phase was quantitated by liquid scintillation spectroscopy.
Flow cytometric analysis of live cells in suspension
Cells stably transfected with the CAT fusion genes under the control of the MT-1 promoter were induced for 19 hours with 80 μM ZnSO4 in DMEM supplemented with 5% BCS. Cells in suspension, as well as the adherent cells, were collected and washed 3 times with PBS. Cells were blocked with PBS/3% goat serum (PBS-GS) for 30 minutes at 4°C and incubated with anti-CAT polyclonal rabbit antibody (5 Prime 3 Prime) in PBS-GS diluted to 1:1500 for 45 minutes at 4°C. After washing with PBS, cells were incubated for 45 minutes at 4°C with FITC-conjugated goat anti-rabbit antibody diluted 1:400 in PBS-GS. Cells were washed with PBS, fixed overnight in PBS/1% paraformaldehyde, washed, resuspended in PBS, and analyzed by flow cytometry on a Coulter Profile (Coulter). As a positive control for the detection of GalT I on the surface of NIH 3T3 cells, non-transfected cells were processed as above but incubated with polyclonal anti-recombinant GalT I catalytic domain (Nguyen et al., 1994).
Determination of GalT I-CAT fusion protein association with the detergent-insoluble fraction
Transient transfected cells were assayed for surface CAT activity associated with the detergent-insoluble cytoskeleton 48 hours after transfection essentially as previously described for surface GalT I (Eckstein and Shur, 1992), substituting anti-CAT antibody for anti-GalT I antibody.
Partitioning of GalT I with lipid rafts
Cells were washed and lysed with TEN buffer containing 1% Triton X-100. The lysate was sheared 10 times with a 21-gauge needle and incubated on ice for 15 minutes. The lysate was brought to 35% Optiprep by adding 585 μl of 60% Optiprep to 400 μl of lysate. For some assays, the detergent-insoluble material was collected by centrifugation (16,000 g, 30 minutes, 4°C), resuspended in TEN buffer with 1% Triton X-100, and used instead of the total detergent-solubilized cell lysate. In either case, the Triton X-100 suspension was placed in the bottom of a SW41 tube and overlaid with 8 ml of 30% Optiprep in TEN. One ml of TEN was added to the tube and centrifuged for 4 hours at 274,000 g in SW41 at 4°C. One ml fractions were removed from the top. The pellet was resuspended in 1 ml TEN. All fractions were subsequently resolved by SDS-polyacrylamide gel electrophoresis (SDS-PAGE) and analyzed by Western blotting using antibodies against the GalT I catalytic domain (Nguyen et al., 1994), the long GalT I cytoplasmic domain (Youakim et al., 1994), caveolin, GM130 and p115 (BD Transduction Laboratories).
An inverse PCR approach was used to mutate individual codons within the long GalT I cytoplasmic domain using TL-GFP as a template (Moreau et al., 1994). Mutagenic and complementary primer pairs (Genosys) were designed in inverted (5′ to 5′) orientation, resulting in amplification of the entire plasmid. Primers were phosphorylated with T4 kinase (USB). Amplification was in the presence of 2 mM Mg+2, 200 μM nucleotides, 0.5 μM each primer, and 2 units Vent DNA polymerase (New England Biolabs) in a 50 μl reaction volume. For some primer sets, addition of 5% formamide to the PCR reaction improved yield. PCR conditions were as follows: 95°C for 5 minutes, one time; 95°C for 30 seconds, 65-68°C for 30 seconds, 72°C for 5 minutes (25-35 times); 72°C for 15 minutes, one time. Cycle number and annealing temperature were varied to give optimal yield for each primer set. Following amplification, a band of the appropriate size was excised from an agarose gel, purified, ligated and transformed into E. coli. Colonies were screened by restriction enzyme digestion to detect the mutation, and positive clones were sequenced to verify the mutation, to ensure the fidelity of the ligation site, and to confirm that no other mutations had been introduced by the PCR.
Quantitation of cell spreading and expression of cell surface GFP fluorescence
Cells to be transfected were plated in 24-well plates 24 hours prior to transfection. Transfection was done using LipofectAMINE™ Plus as described above, and cells were replated onto tissue culture chamber slides (Falcon, Becton Dickinson) at the end of the transfection. Chambers were coated with either 10 μg/ml laminin (Sigma-Aldrich, to score for cell spreading) or 10 μg/ml fibronectin (Invitrogen, to monitor for cell surface fluorescence). Twenty four to 48 hours following transfection, cells were washed gently in two changes of PBS and fixed for 15 minutes at room temperature in 4% paraformaldehyde in PBS. Cells were then rinsed in three changes of PBS and mounted in Vectashield (Vector Labs).
Cell spreading was scored on a Zeiss inverted microscope (Carl Zeiss) connected to Metamorph imaging software (Universal Imaging). Transfected cells were identified using an FITC filter to detect GFP fluorescence. Cell area was determined using Metamorph software, or alternatively by opening images in NIH Image (NIH freeware). The experiment was repeated 3 times, and 25-30 cells were scored for each sample in each experiment. Statistical significance was determined using a modified Student's t test (Bonferroni method).
The presence of surface fluorescence at the tips of lamellipodia and/or filopodia was scored on a Zeiss 510 confocal microscope. The number of cells with fluorescence localized to tips was expressed as a percentage of the total number of cells transfected as judged by the detection of Golgi-localized GFP fluorescence. The experiment was repeated 4 times, and an average of 140 GFP-positive cells (no fewer than 30 cells) were scored for each sample in each experiment. Statistical significance was determined using a modified Student's t test (Bonferroni method).
Western blotting with GFP antibody
Transiently transfected cells were lysed (1% SDS, 1 mM DTT, 1 hour at 50°C), and insoluble material was removed by centrifugation (5 minutes at 16,000 g). Equal protein concentrations were resolved by SDS-PAGE under reducing conditions, then transferred to nitrocellulose. Membranes were blocked for 1 hour in TBS-T buffer (20 mM Tris, pH 7.6, 137 mM NaCl, 0.1% Tween-20) containing 5% BSA. Membranes were then incubated for 3 hours at room temperature in monoclonal anti-GFP antibody (Clontech) diluted in TBS-T/BSA. After washing exhaustively in TBS-T, membranes were incubated for 1 hour at room temperature in HRP-conjugated goat anti-mouse IgG (Vector), diluted 1/50,000 in TBS-T/BSA. After washing in TBS-T, membranes were developed using ECL Plus chemiluminescence detection (Amersham), and exposed to film.
Immunoprecipitation of radiolabeled GalT I
For metabolic labeling, cells were incubated in DMEM deficient in either methionine (for 35S) or phosphate (for 32P) for 30 minutes, after which the medium was replaced with fresh DMEM-deficient medium. Cells were labeled with either 10 μCi/ml [35S]methionine or 250 μCi/ml [32P]orthophosphate for 3 hours, washed, dissociated with TEN buffer, washed twice with medium B, and solubilized in lysis buffer (1% Triton X-100, 50 mM NaCl, 10 mM EDTA, 50 mM Na fluoride, 10 mM Na pyrophosphate, 1 mM glycerol phosphate, 2 mM Na vanadate, 60 μM okadaic acid, 10 mM HEPES pH 7.4). The cell lysate was precleared by incubating 1 ml aliquots overnight at 4°C with 10 μl preimmune rabbit sera, followed by 1-2 hours with Protein A-Sepharose. The Protein A-Sepharose-bound IgG complexes were removed by centrifugation and the supernatants incubated at 4°C with Protein A-Sepharose beads precoated with either preimmune rabbit sera, rabbit sera specific for the GalT I catalytic domain (Nguyen et al., 1994) or the long cytoplasmic domain (Youakim et al., 1994). After 2-3 hours of incubation, the beads were collected by centrifugation, washed twice and resuspended in SDS-sample buffer for SDS-PAGE. After electrophoresis and transfer to nitrocellulose, the blots were processed for autoradiography.
To detect surface expression, dissociated cells were resuspended in 5 ml of medium B supplemented with 450 μg NHS-SS-Biotin (Pierce). After 1-2 hours of incubation at 4°C, cells were washed twice with medium B, solubilized in lysis buffer, and applied to streptavidin agarose columns. The columns were washed (medium B, 0.25% Triton X-100) until radioactivity reached background levels, after which 1 ml of medium B containing 2% β-mercaptoethanol (BME) was applied to the column. After 30-45 minutes of incubation, the BME-released material was eluted from the column with 1 ml medium B/0.25% Triton X-100; this step was repeated and the eluted material pooled. After dialysis, radioactivity was determined in all fractions and the dialyzed retentate was used for immunoprecipitation of GalT I as described above. In some instances, GalT I enzyme-specific activity was determined in aliquots of cell lysates and biotinylated surface fractions as previously described (Evans et al., 1993), and used to calculate the amount of GalT I activity on the cell surface.
The cytoplasmic and transmembrane domains of the long GalT I isoform are sufficient to transport a reporter protein to the cell surface
Previous results support the hypothesis that GalT I is able to function as a cell surface receptor because of amino acid sequences contained within the long cytoplasmic domain and which facilitate GalT I association with the detergent-insoluble cytoskeleton. Here, this hypothesis was tested more rigorously by asking if the GalT I cytoplasmic and transmembrane domains could direct a reporter to the cell surface, and if so, does it partition with the detergent-insoluble pool.
NIH 3T3 cells were transfected with GalT I-CAT fusion constructs (TL-CAT and TS-CAT) under the control of the inducible MT-1 promoter, and stable transformants were selected. CAT assays were used to screen the expanded G418-resistant colonies for CAT-expressing clones. As controls, 3T3 cells were transfected with CAT alone, or were transfected with the neomycin (neo) resistance vector alone. Clones were assayed for CAT activity as described in Materials and Methods, and TL-CAT and TS-CAT clones with similar levels of CAT expression were chosen for subsequent analysis.
Consistent with previous results, overexpressing the cytoplasmic and transmembrane domains of the long form of GalT I caused the cells to become nonadhesive after Zn2+ induction (Evans et al., 1993) (and see below). In contrast, there was no altered morphology in any of the clones transfected with either CAT or TS-CAT. To assay CAT protein expression on the cell surface, neo-resistant, CAT-positive clones were analyzed by indirect immunofluorescence and flow cytometry using anti-CAT antibodies. Uninduced and induced 3T3 transfectants were labeled live in suspension with anti-CAT antibody and FITC-conjugated goat anti-rabbit antibody. Indirect immunofluorescence shows that TL-CAT was localized to the surface of cells, whereas mock-transfected cells, or cells transfected with either CAT or TS-CAT (data not shown) showed minimal surface immunoreactivity (Fig. 1).
Flow cytometry was used to quantify the amount of CAT protein expressed on the cell surface (Fig. 2). As a positive control, GalT I expression on the surface of non-transfected 3T3 fibroblasts was assayed using antibodies raised against the bacterially expressed GalT I catalytic domain (Nguyen et al., 1994) (Fig. 2 inset). Induced and non-induced live, intact cells were incubated with anti-CAT IgG, followed by FITC-conjugated goat anti-rabbit antibody. Mock, CAT, TS-CAT and TL-CAT transfectants showed very weak fluorescence in the uninduced (–) cultures (Fig. 1). However, the signal increased dramatically when TL-CAT cells were induced with Zn2+ (M+). The increase in immunoreactivity was apparent as two peaks of fluorescence because of the presence of two populations of cells: floating nonadherent cells (F+) and adherent cells (A+). The highest level of CAT expression at the cell surface was found on the nonadherent TL-CAT cells (F+), whose fluorescence intensity coincided with the higher peak in the mixed population (M+). The loss of cell adhesion in these TL-CAT cells probably results from the dominant-negative phenotype produced by expression of the truncated long protein (Evans et al., 1993). Furthermore, the fluorescence intensity of the cells that remain adherent after induction (A+), and which should be the lowest expressers, coincided with the lower peak in the mixed population (M+), having little surface immunoreactivity above control levels (Fig. 2). As expected, all CAT, TS-CAT and mock transfectants remained adherent after induction. These results are similar to previous studies showing that the truncated long GalT I protein creates a dominant-negative phenotype in 3T3 cells and are consistent with the long form of GalT I being expressed on the cell surface. Unfortunately, the phenotype of the TL-CAT cells was gradually lost, probably because of low, basal activity of the MT-1 promoter, such that cells expressing the TL transgene become nonadherent and are gradually lost during routine cell culture (Evans et al., 1993). To circumvent this problem, we turned to a transient transfection system in our subsequent analyses.
Surface-localized TL-CAT associates with the detergent-insoluble cytoskeleton
Transiently transfected cells were assayed for CAT expression on the cell surface by incubating live intact cells with anti-CAT antibody followed by 125I-goat anti-rabbit IgG (Fig. 3). Similar to that seen by indirect immunofluorescence (Fig. 1) and FACS (Fig. 2), CAT expression on the surface of TS-CAT transfectants was only slightly increased above controls (CAT transfectants, Fig. 3). In contrast, CAT protein expression on the surface of TL-CAT transfectants was approximately 7-fold higher than that found on control cells. Most of the CAT protein localized on the surface (81%) was associated with the detergent-insoluble cytoskeleton (Fig. 3). These results show that the cytoplasmic and transmembrane domains of the long form of GalT I are able to direct a reporter to the cell surface and anchor it to the detergent-insoluble cytoskeleton.
A portion of the detergent-insoluble GalT I co-fractionates with caveolin-containing lipid rafts
We have previously reported that GalT I exists in three pools on the cell surface: (1) a detergent-soluble pool; (2) a detergent-insoluble pool that is associated with the actin cytoskeleton; and (3) a detergent-insoluble pool that does not appear to be cytoskeletally associated because it resists conditions that release cytoskeletally associated proteins (Eckstein and Shur, 1992). The distribution of GalT I in these three pools varies with the migratory status of the cells and is similar to that reported for other matrix receptors (Rapraeger et al., 1986). Since these earlier studies, it has become clear that the detergent-insoluble pools can reflect association with either the cytoskeleton and/or detergent-resistant lipid rafts. We subsequently asked whether any of the detergent-insoluble, non-cytoskeletally associated pool is associated with lipid rafts.
NIH 3T3 cells were lysed and the detergent-insoluble material was subjected to Optiprep gradients to float the lipid rafts from the detergent-insoluble pellet. Fractions were collected and subjected to SDS-PAGE and immunoblotting. The position of caveolin-containing lipid rafts was determined by caveolin immunoblotting. As expected, the caveolin-containing rafts floated out of the pellet to the lighter density fractions (Fig. 4). Interestingly, a portion of the detergent-insoluble GalT I reproducibly cofractionated with caveolin, whereas the remainder stayed near the bottom of the gradient. This distribution is reminiscent of the two pools of detergent-insoluble GalT I previously described; one associated with the actin cytoskeleton and one that is not cytoskeletally associated (Eckstein and Shur, 1992). Similar results were obtained using either the detergent-insoluble material or the entire detergent-solubilized lysate, as well as when immunoblotting with antibodies raised against either the GalT I catalytic domain or the 13 amino acid sequence unique to the long cytoplasmic domain. The partitioning of GalT I into the lighter density fractions was not simply the result of nonspecific protein trapping in these fractions, because two proteins known not to be associated with rafts (GM130, p115) remained in the higher density, lower fractions of the Optiprep gradient (Fig. 4). Finally, treatment of cells with cyclodextran to destroy the lipid rafts prevented caveolin and GalT I partitioning into the lighter density fractions (data not shown).
The long GalT I cytoplasmic domain fused to GFP leads to expression on cell processes and a GalT I-dependent dominant-negative phenotype
In order to visualize the occurrence of GalT I fusion constructs on the surface of live cells, and to avoid the use of antibodies and enzyme assays, we replaced the CAT reporter with Enhanced GFP. This also simplified the detection of GalT I-GFP-expressing cells. Transient transfection with equal concentrations of DNA resulted in equal expression of GalT I-GFP fusion proteins, as detected by immunoblotting with anti-GFP antibodies (Fig. 5a). Similar to that seen with GalT I-CAT fusion proteins, both TS-GFP and TL-GFP were readily detected in the perinuclear region (i.e. Golgi) of transiently transfected 3T3 cells (Fig. 5b). GFP fluorescence was also detectable on the surface of specific transfectants in distinct patches on lamellipodia and filopodia (Fig. 5b), consistent with the previously described distribution of the endogenous GalT I (Youakim et al., 1994). The expression of GFP fluorescence on the cell surface was quantified for each transfectant; 38% of TL-GFP-expressing cells demonstrated GFP fluorescence on their surfaces, whereas only 13% of TS-GFP-expressing cells had any detectable surface expression (Fig. 6).
Previous studies have demonstrated that expression of the cytoplasmic and transmembrane domain of long GalT I (TLGT) leads to a loss of 3T3 cell spreading on laminin matrices by displacing the endogenous long GalT I from the detergent-insoluble cytoskeleton (Evans et al., 1993). These results have been confirmed here using the truncated long GalT I fused to a CAT reporter (TL-CAT) (Fig. 2). TL-GFP fusions produced a similar dominant-negative phenotype (Fig. 5c). 3T3 cell spreading on laminin was quantitated by measuring the surface area of transfected cells. Cells expressing TS-GFP had an average area of 1160 μm2, similar to mock-transfected or GFP-transfected cells, whereas the area of TL-GFP-expressing cells was approximately one-half that of controls (570 μm2) (Fig. 7).
Site-directed mutagenesis of serine and threonine residues prevents surface expression and the dominant-negative phenotype
We asked if specific amino acid residues could be identified in the GalT I cytoplasmic domain that are critical for the surface expression of the TL-GFP fusion protein. Inverse PCR was used to introduce specific amino acid substitutions within the cytoplasmic domain of long GalT I. We initially focused on putative phosphorylation sites (one Ser and one Thr), because human GalT I has been shown to be Ser-phosphorylated (Strous et al., 1987), and a Ser/Thr-specific kinase that phosphorylates GalT I has been identified (Bummell et al., 1990). Ser 11 (S11) and Thr 18 (T18) were mutated, either singly or together, to neutral alanine (A) to mimic the dephosphorylated state, or to aspartic acid (D) to mimic the phosphate negative charge. The mutated GalT I cytoplasmic and transmembrane domains were fused to a GFP reporter to determine both the effects on GFP expression on the surface of cells grown on fibronectin, and the ability of the construct to produce a dominant-negative phenotype; i.e. a loss of cell spreading on laminin. As before, cells were chosen for analysis that expressed similar levels of GFP as determined by fluorescence and immunoblotting.
Mutating either S11 or T18 to alanine, singly or together, had no effect on cell surface expression of GFP, relative to TL-GFP controls. In contrast, mutating S11 or T18 to aspartic acid significantly reduced GFP fluorescence on the cell surface (Fig. 6). Mutating S11 and T18, singly or together, to alanine had no effect on mean surface area relative to TL-GFP, suggesting that these amino acids in their neutral state function equally to non-mutated TL-GFP. In contrast, mutating S11 or T18 to aspartic acid leads to a loss of the dominant-negative phenotype–cells expressing these constructs spread similarly to TS-GFP controls (Fig. 7). TL-GFP mutants in which both S11 and T18 were mutated to aspartic acid were always expressed at low levels, rendering results with this construct uninterpretable.
The results from the site-directed mutagenesis collectively suggest that phosphorylation of the long GalT I cytoplasmic domain interferes with its expression on the cell surface, which would preclude its ability to function as a surface receptor. However, one mutant construct illustrated that GalT I expression on the cell surface can be uncoupled from its ability to function as a receptor. TL-GFP, in which the two phenylalanine residues were mutated to glycine [Phe 3 and 7 to Gly; F3G,F7G], was expressed on the cell surface at levels similar to TL-GFP (Fig. 6); however, it no longer produced a dominant-negative phenotype (Fig. 7), indicating that transport of GalT I to the cell surface does not necessarily lead to its function as a surface receptor. Clearly, specific amino acid residues, including F3 and F7, are somehow required for GalT I ability to function as a matrix receptor, possibly by interacting with cytosolic effector proteins, but do not play a critical role for expression on the cell surface.
Long GalT I is phosphorylated and confined to the intracellular compartment
Collectively, the results suggest that phosphorylation of the long GalT I cytoplasmic domain inhibits its expression on the cell surface. This would predict that endogenously phosphorylated GalT I should be reduced on the cell surface. We first confirmed the ability of murine GalT I to be phosphorylated by immunoprecipitating 32P-labeled GalT I from 3T3 cell lysates (Fig. 8b). Control immunoprecipitations were performed with 35S-labeled GalT I (Fig. 8a). Immunoprecipitation with antibodies against the GalT I catalytic domain or the long cytoplasmic domain produced similar results. Similarly, 32P-GalT I could be purified from 3T3 cell lysates by α-lactalbumin affinity chromatography, confirming the ability of endogenous GalT I to be phosphorylated (data not shown).
To determine if the 32P-labeled GalT I was expressed on the cell surface, the surfaces of live, intact 3T3 cells were biotinylated using disulfide-linked biotin, and the surface biotinylated material was isolated on streptavidin affinity columns. The avidin-bound biotinylated material was eluted by reducing the disulfide bond with β-mercaptoethanol. The presence of GalT I in total cell lysates, as well as in the biotinylated surface material, was determined by immunoprecipitation and autoradiography. As expected, 35S-labeled GalT I could be readily immunoprecipitated from the biotinylated surface fraction (Fig. 8c). GalT I enzyme-specific activity was assayed in the biotinylated surface fraction and compared with that in the total cell lysate. Interestingly, 7% of the total GalT I activity was recovered in the biotinylated surface fraction (19,371,000 cpm galactosylated product/hour in cell surface fraction versus 281,673,000 cpm galactosylated product/hour in total cell lysate), similar to that reported when GalT was quantified on 3T3 plasma membranes isolated by subcellular fractionation (Cummings et al., 1979). Although ∼7% of the total cellular 35S-labeled GalT I was expressed on the cell surface, none of the 32P-labeled GalT I was recovered in the biotinylated surface fraction (data not shown), consistent with results illustrating that phosphorylation of the GalT I cytoplasmic domain inhibits its stable expression on the cell surface.
Previous studies have shown that surface GalT I requires association with the detergent-insoluble cytoskeleton in order to function as a matrix receptor (Eckstein and Shur, 1992). Subsequent studies indicated that sequences within the long cytoplasmic domain, rather than in the short domain, are responsible for GalT I association with the cytoskeleton and function as a matrix receptor (Appeddu and Shur, 1994a; Appeddu and Shur, 1994b; Evans et al., 1993). In this study, we have shown that the cytoplasmic and transmembrane domains of the long GalT I isoform can direct the expression of two different reporter proteins to the cell surface, where it produces a dominant-negative phenotype during cell spreading on laminin. Fusion of the short GalT I cytoplasmic and transmembrane domains to the reporter proteins does not result in expression on the cell surface, nor does it produce a dominant-negative phenotype. An early report by others was unable to detect a reporter protein on the cell surface when fused to either the long or short GalT I cytoplasmic and transmembrane domains (Russo et al., 1992). However, because only a small portion of GalT I is expressed on the surface (∼7% in fibroblasts), it is not surprising that these studies failed to detect surface expression given the low level of transgene expression in their system. In any event, results reported in this study extend and confirm the suggestion that sequences within the long cytoplasmic domain are responsible for its ability to function as a surface receptor, and probably do so by facilitating association with the detergent-insoluble cytoskeleton (Evans et al., 1993). However, our early studies failed to consider the possibility that the detergent-insoluble pool may represent both cytoskeleton-associated GalT I as well as GalT I partitioning to detergent-insoluble lipid rafts (Eckstein and Shur, 1992).
A portion of GalT I distributes to lipid rafts
We therefore examined the possibility that the non-cytoskeletally associated pool of detergent-insoluble GalT I may, in fact, represent partitioning into lipid rafts. Consistent with this, a significant proportion (approximately one-third to one half) of the detergent-insoluble GalT I in nontransfected 3T3 fibroblasts cofractionated with caveolin, a marker for lighter density, detergent-insoluble lipid fractions (Liu et al., 2002). Results using antibodies against either the GalT I catalytic domain or against the long GalT I cytoplasmic domain produced similar results. The fact that other intracellular membrane and matrix proteins (GM130, p115) failed to cofractionate with caveolin-containing lipid fractions eliminated the possibility that GalT I was nonspecifically trapped in caveolin-containing fractions. Finally, depleting cholesterol pools with cyclodextran prevented the formation of rafts and prevented GalT I migration to the lighter density fractions. Consequently, these studies suggest that GalT I behaves similarly to other matrix receptors, such as the integrins (Berditchevski, 2001; Leitinger and Hogg, 2002), in that it exists in two interrelated detergent-insoluble pools; one associated with the actin cytoskeleton and the other associated with lipid rafts. Although previous studies indicate that GalT I requires association with the detergent-insoluble pool to function as a surface receptor, it is unknown at this time whether, in fact, both populations of detergent-insoluble GalT I participate in its adhesive function.
Phosphorylation of the cytoplasmic domain reduces surface expression and function
The ability of the long GalT I cytoplasmic domain to target a reporter protein to the cell surface and produce a dominant-negative phenotype enabled us to examine specific amino acid residues that may be critical for surface expression and function as a matrix receptor. We focused our initial efforts on the importance of the S11 and T18 residues present in the cytoplasmic domain, because GalT I has previously been shown to be Ser-phosphorylated (Strous et al., 1987), and Ser/Thr phosphorylation is a common mechanism for protein targeting (Leger et al., 1997; Steveson et al., 2001; Walter et al., 2001). Furthermore, a Ser/Thr-specific kinase, homologous to the p34cdc2 kinase, has been reported to co-purify with GalT I and phosphorylate GalT I in vitro (Bummell et al., 1990).
Mutating either S11 or T18 to an acidic residue (S11D, T18D) resulted in decreased expression on the cell surface and a concomitant decrease in the dominant-negative phenotype. In contrast, mutating either or both of these residues to a neutral alanine had little, if any, effect on cell surface expression or the dominant-negative phenotype. These results suggest that phosphorylation of GalT I interferes with its transport to and/or retention on the cell surface, such that it is unable to function as a matrix receptor. Consistent with this, the endogenous GalT I can be readily phosphorylated, but this phosphorylated GalT I is not detectable on the cell surface, although ∼7% of the non-phosphorylated GalT I is surface localized on 3T3 fibroblasts.
It is interesting that one (T18) of the phosphorylation sites that influences GalT I localization and function is in the region of the cytoplasmic domain common to both isoforms, whereas the other phosphorylation site (S11) is unique to the long cytoplasmic domain. If mutagenesis of T18 and S11 produce similar phenotypes, then why doesn't short GalT I demonstrate some of the same properties as long GalT I? One obvious possibility is that only one of the two potential phosphorylation sites within the long GalT I cytoplasmic domain is functional in vivo, and in this regard, mutagenesis of S11 proved to have more striking consequences than did mutagenesis of T18 (Fig. 6). However, a more probable possibility is that the phosphorylation state of specific residues in the cytoplasmic domain is relevant or functional only in the context of the full-length GalT I cytoplasmic domain. Consistent with this, only the full-length cytoplasmic domain is able to functionally interact with heterotrimeric G-proteins in sperm; neither the 11 amino acids common to the long and short domains, nor the 13 amino acids unique to long GalT I domain, were functional in this assay (Gong et al., 1995).
The notion that the length of the cytoplasmic domain may be critical for overriding any Golgi retention signals within the transmembrane domain comes from a study of asialoglycoprotein receptor mutants (Wahlberg et al., 1995). Mutations were created in the cytoplasmic domain of this normally cell surface-localized protein, and it was concluded that shortening the cytoplasmic domain to 17 amino acids or less results in increased Golgi retention. Interestingly, the short GalT I cytoplasmic domain is only 11 amino acids long, whereas the long cytoplasmic domain is 24 amino acids in length. Although the actual sequence of the asialoglycoprotein receptor cytoplasmic domain seems to have no effect on localization, it remains possible that all of the sequences tested have a feature in common other than length; in this regard, all of the sequences tested contain at least one Ser residue.
How might phosphorylation influence GalT I retention in the Golgi?
The subcellular targeting of glycosyltransferases has been studied in detail as a means to understand how proteins that pass through the endoplasmic reticulum/secretory pathway are targeted to, or retained in, specific compartments (Colley, 1997). Although there are subtle differences in methodology among the various studies, the consensus appears to be that the transmembrane domain is responsible for Golgi retention of Golgi glycosyltransferases, including GalT I (Bulbarelli et al., 2002; Masibay et al., 1993; Nilsson et al., 1991; Russo et al., 1992; Teasdale et al., 1992). Both the length and the overall hydrophobicity of the transmembrane domain play a role in Golgi retention, rather than the presence of specific sequences. The formation of oligomers (Yamaguchi and Fukuda, 1995) as well as higher order complexes could further facilitate Golgi retention (a property termed `kin recognition') (Nilsson et al., 1994). GalT I forms homodimers in the Golgi (Yamaguchi and Fukuda, 1995); however, no evidence for higher order complex formation has been demonstrated for GalT I (Opat et al., 2000).
In spite of the evidence suggesting that the transmembrane domain of glycosyltransferases mediates Golgi retention, several labs have demonstrated the cell surface localization of a variety of glycosyltransferases, including sialyltransferase (Close and Colley, 1998), fucosyltransferase (Borsig et al., 1996; Marker et al., 2001), N-acetylgalactosaminyl-transferase (Mandel et al., 1999) and GalT I (Shur et al., 1998). These observations argue that other domains within these proteins may modulate the Golgi retention signal in the transmembrane domain. Most probably, multiple sorting signals, possibly with competing activities, exist within most transmembrane proteins, and these act in combination to determine the final localization of a transmembrane protein, whether it resides in one or multiple cellular compartments (Bulbarelli et al., 2002; Mostov et al., 2000). With respect to GalT I in particular, studies reported here and elsewhere strongly suggest that surface expression and function requires signals within the full-length cytoplasmic domain, and these properties are negatively regulated by phosphorylation.
How might the phosphorylation state of long GalT I cytoplasmic domain influence surface localization and function? One possibility that awaits further study is that phosphorylation affects the association of long GalT I with a scaffolding protein required for its stable association with the Golgi or with the cell surface. In this regard, a yeast two-hybrid screen has shown that the long GalT I cytoplasmic domain binds to SSeCKS (Src Suppressed CKinase Substrate), a kinase and cytoskeleton scaffolding protein (Wassler et al., 2001). Both SSeCKS and long GalT I have similar dual subcellular localizations at the plasma membrane and in the Golgi region (Lin et al., 1996; Wassler et al., 2001). Another possibility is that phosphorylation affects GalT I internalization and/or recycling kinetics. GalT I has been detected in the endosomal compartment (Becich and Baenziger, 1991), and it is therefore of interest to determine the role of cytoplasmic domain sequences in GalT I recycling, as well as the role of internalization on surface GalT I function. In any event, the results reported here are consistent with the hypothesis that GalT I is able to function as a surface receptor by virtue of sequences present in the context of the long cytoplasmic domain, and which are negatively regulated by phosphorylation. Further studies are required to determine how phosphorylation/dephosphorylation of the cytoplasmic domain influences GalT I association with cytoplasmic effector proteins, and how this influences retention and/or release from the Golgi complex as well as stability on the cell surface.
We thank Sayadeth Khounlo for technical assistance, and Betty Skipper for assistance with statistical analysis. Supported by grants from the NIH (DE07120, HD22590 to B.D.S.) and from the ACS (RPG-99-179-01-CSM to H.J.H.). We acknowledge the use of the confocal microscope in the Fluorescence Microscopy Facility at the University of New Mexico Health Sciences Center, and the University of New Mexico Cancer Center (supported by NCRR 1 S10 RR14668, NSF MCB9982161, NCRR P20 RR11830 and NCI R24 CA88339).