Specific Sequences in the Signal Anchor of the P-Galactoside a=2,6=Sialyltransferase Are Not Essential for Golgi Localization MEMBRANE FLANKING SEQUENCES MAY SPECIFY GOLGI RETENTION*

The Pgalactoside a-2,6-sialyltransferase is a trans Golgi/trans Golgi network glycosyltransferase which adds sialic acid residues to Asn-linked oligosaccharides of glycoproteins. Previous results suggested that the sialyltransferase stem and signal anchor including flank- ing sequences may be two independent Golgi retention regions. However, other experiments demonstrated that the sequence of the signal anchor itself was not impor- tant. To investigate whether the sialyltransferase signal anchor was necessary and sufficient for Golgi retention, several mutant and chimeric proteins were expressed and localized in Cos-1 and Chinese hamster ovary cells. We found that the signal anchor and flanking sequences were able to retain the sialyltransferase catalytic do- main in the Golgi. However, efficient Golgi retention was still observed when the signal anchor was altered or entirely replaced in either the presence or absence of most of the luminal stem region. Chimeric proteins con- sisting of the sialyltransferase cytoplasmic tail and signal anchor fused to the extracellular domains of two different cell surface proteins demonstrated poor Golgi retention. Asignificant increase in the Golgi retention of one of these chimeras was observed when two lysines were placed next to the signal anchor on the luminal side. Taken together these results suggest that the sia- lyltransferase signal anchor is not

cent secretory proteins are co-translationally translocated into the lumen of the ER where they are core glycosylated on Asn residues and correctly folded and oligomerized (reviewed in Refs. 1-31. The transport of secretory proteins from the ER, through the Golgi complex, and to the plasma membrane is thought to occur in a bulk flow fashion, with nascent proteins moving in vesicular carriers with the bulk flow of lipid until they are retained in a specific compartment (1). Retention or targeting signals have been identified for a variety of ER proteins (2, 4-71, Golgi proteins (reviewed in Refs. 8 and 91, and lysosomal proteins (reviewed in Refs. [10][11][12]. Pelham and colleagues (2,4) have elucidated the retentiodretrieval signal and mechanism for soluble ER proteins, while other groups (5)(6)(7) have identified specific signals for the retention of ER transmembrane proteins. Several researchers have found that Golgi glycosyltransferases and Golgi-localized viral proteins require sequences in and around their membrane spanning regions for Golgi retention (reviewed in Refs. 8 and 9). Soluble lysosomal enzymes are targeted to lysosomes by receptors which recognize mannose 6-phosphate recognition markers found on the Asn-linked oligosaccharides of these proteins (10, l l ) , while membrane associated lysosomal proteins have targeting signals within their cytoplasmic tails (12).
The Golgi complex is central to the sorting and modification of proteins in the secretoly pathway. Morphologically the Golgi complex is composed of a series of cisternae (cis, medial, trans, and trans Golgi network (TGN)) which have been functionally defined by the specific localization of glycosidases and glycosyltransferases that sequentially process the Asn-linked oligosaccharides of secretory proteins as they pass through these compartments (13,14). Recently, Mellman and Simons (15) have suggested viewing the Golgi complex as a three-compartment structure consisting of the cis Golgi network, the medial Golgi, and the TGN. The cis Golgi network, consisting of the cis Golgi cisternae and the transitional elements between the ER and cis Golgi, is involved in the trafic from ER to Golgi and also in the recycling of proteins and lipids from this region back to the ER. The medial Golgi is thought to function as a glycosylation compartment and contain the majority of the enzymes involved in these processes. Presumably, the sequential, and at times overlapping, organization of Golgi glycosyltransferases would be maintained in this medial compartment (13, 16). The final Golgi compartment, the TGN, is known to play the central role in the sorting of plasma membrane, lysosomal, and regulated secretory proteins (13,17). Recent results suggest that retention signals and mechanisms employed by proteins exclusively localized in the TGN could be different from those in the other Golgi compartments (18,19).
While the terminal Golgi glycosyltransferases have similar type I1 domain structures, they possess no obvious sequence homology which would suggest a common Golgi retention signal (9,14). Despite this limitation, several investigators have identified sequences in Golgi glycosyltransferases and other Golgi proteins which are important for their localization (reviewed in Refs. 8 and 9). Machamer and Rose (20) demonstrated that the first membrane spanning region of the infectious bronchitis virus (IBV) E l glycoprotein is necessary for this protein's cis Golgi retention. Later work demonstrated that uncharged polar residues within the transmembrane a-helix are important components of this retention signal (21). Construction and expression of soluble forms of the p-galactoside a-2,6-sialyltransferase (ST) (22) and the p-1,4-galactosyltransferase (23), demonstrated that their Golgi retention signals reside in their amino-terminal cytoplasmic tail, signal anchor, andor luminal stem regions. Further analysis demonstrated that the sequences in and around the signal anchor regions of the p-l,4-galactosyltransferase (23)(24)(25)(26), N-acetylglucosaminyltransferase I (27), and ST (28)(29)(30) are required for the Golgi retention of these proteins. Evidence from our laboratory (29) and that of Munro (28) suggest that the Golgi retention signal of the the ST may be somewhat more complicated than those of other glycosyltransferases. Unlike other Golgi glycosyltransferases, changes in the sequences of the signal anchor of the ST do not result in the cell surface expression of the protein (28,29). In addition, other experiments have demonstrated that other ST sequences, such as those which flank the signal anchor and those within the luminal stem region, may participate in Golgi retention or constitute independent retention regions (28,29).
To determine the relative importance of the signal anchor region in the Golgi localization of the ST, we constructed a series of mutant and chimeric proteins to determine what sequences of the ST are necessary and sufficient for Golgi localization. Our results suggest that the precise sequences of the ST signal anchor region are not necessary for Golgi retention since they can be altered or completely replaced with no change in Golgi retention. Since the signal anchor region and flanking sequences are able to retain the ST catalytic domain in the Golgi, it is likely that correctly spaced membrane flanking sequences constitute the ST Golgi retention signal. Localization of chimeric proteins demonstrated that the ST cytoplasmic tail and signal anchor region alone are not sufficient for Golgi retention. More efficient Golgi retention required 2 lysine residues in the luminal sequences flanking the ST signal anchor region. These results suggest that the signal anchor region of the ST is not necessary or sufficient for Golgi retention and that it can be replaced by any transmembrane region which allows correct spacing and folding of sequences flanking the membrane.

In Vitro Oligonucleotide-directed Mutagenesis
In vitro oligonucleotide-directed mutagenesis was performed as previously described (29) according to the method of Zoller and Smith (31).
For the ATAS mutant and the ASSAmutants, single-stranded DNAwas made from AStem-ST cDNA (29) subcloned into the bs+(KS) phagemid (Stratagene). Following selection of positive bacterial colonies, DNA was prepared by the Qiagen method and the mutant cDNA was subcloned into the XbaI and SmaI sites of the pSVL expression vector (Pharmacia) for expression in Cos-1 cells. The following oligonucleotides were used in the construction of the ATAS mutant and the ASSA

Construction of Signal Anchor Replacements and Chimeric Proteins
Polymerase chain reaction (PCR) using Vent polymerase (New England BioLabs) was performed according to manufacturer's instructions using ST, NA, and TR coding sequences in either bs' or pSVL as templates. The PCR fragments encoding the specific sequences of the chimeric proteins were gel-purified using low melting point agarose (Life Technologies, Inc.) and were then ligated into either pSVL for expression in Cos-1 cells or into a vector possessing the T7 polymerase promoter, such as bs+ or pGEM-2, for expression in CHO cells using the recombinant vaccinia virus. Specific restriction enzyme sites were incorporated into the oligonucleotides used in the PCR reactions and allowed the ligation of DNA fragments into these expression vectors and with other DNA fragments. The sequence of each mutant construct was verified by DNA sequencing using Sequenase enzyme (U. S. Biomedical Signal Anchor Replacements-Initially, SA29 containing the entire NA signal anchor region was constructed using these oligonucleotides: primer 1, GTAATACGACTCACTATAGGG; primer 2, AAGCTTCCA-GATC'I3TTTClTCAAGTTCGTAT; primer 3, GGG"TAAAAA4GG-GAGCGACTATG; primer 4, CCCGGGCCCCATTMCCTCAGAA; primer 5, GGGAGATCTTAACCATTGGGTCAATCT; primer 6, AAGCT-TAGTACTAATCCATA'ITGAGATTATATT. Primers 1 and 2 were used to synthesize the ST cytoplasmic tail, primers 3 and 4 were used to synthesize the ST stem and a portion of the catalytic domain, and primers 5 and 6 were used to synthesize the signal anchor region of the NA. After sequential construction of the SA29 construct into bs', the remaining portion of the ST catalytic domain was added by cleavage of ST-bs+ and the partial SA29 construct with MscI (within the stem region) and SacI (polylinker of bs') and insertion of this region of the wild type ST into the partial SA29 construct. Both SA23 and SA17 were constructed using SA29 as a template and simply truncating the signal anchor region which was ligated to the stem and catalytic domain of the ST. The T7 primer (primer 1) and SA23 primer (GATTGAAGTACTAT-TGCCTATITG) were used to synthesize the ST tail and 23 amino acids of the NA signal anchor, the T7 primer (primer 1) and SA17 primer (GTTGAG-TACTTAGGCTAATTATTC) were used to synthesize the ST tail and 17 amino acids of the NA signal anchor, and the T3 primer and primer 3 (shown above) were used to synthesize the ST stem and catalytic domains for both constructs. Initially the SA23 and SA17 tail and signal anchor fragments (cut with XbaI and ScaI) were subcloned into the XbaI and RsaI sites in the pGEM-7 vector. The ScaIIRsaI ligation regenerates the RsaI site which was used with the Hind111 site in the pGEM-7 polylinker to subclone the DraI (also compatible with RsaI ends)-Hind111 ST stem and catalytic domain fragment to complete both constructs. Both SA23 and SA17 constructs were subcloned into the XbaI and SacI sites of pSVL for expression in Cos-1 cells.
ST-NA Chimeric Proteins-The STNA chimera was generated by fusing the ST tail and signal anchor (generated from the ST-bs+ plasmid Corp.).
using the T7 primer (primer 1) and primer 9 (CTCCCCATATWCAAA-CACAGATG)) to the NA stalk and head regions (generated from the NA-bs' plasmid using the T3 primer and primer 10 (GGA'ITAGC-CATATGATT'CAAACCGG)) at an oligonucleotide-encoded N&I site. The STKKNA construct was ligated into the XbaI and Sac1 sites in the polylinker regions of both pSVL and pGEM-2. STKKNA chimera was constructed by fusing the ST tail and signal anchor plus two luminal lysines (generated from the ST-bs' plasmid using the T7 primer (primer 1) and primer 7 (GTGG'ITGACTT'C'ITCCAAACAC)) to the NA stalk and head regions (generated from the NA-bs' plasmid using the T3 primer and Primer 8 (GGCAGTCAACTCAAACCGG)) at an oligonucleotide encoded HincII site. Initially in the bs+ at the SacI and KpnI sites, the STNA construct was subcloned into pSVL in the XbaI and SacI sites. NA was subcloned into the EcoRI site in the bs+ polylinker region and into the XbaI and SmaI sites in the pSVL expression vector for expression in either CHO cells or Cos-1 cells.
ST-TR Chimeric Protein-Using ST-bs+ as a template, the T7 primer (primer 1) and primer 10 (GTCGCTGGCGCCCTT'CCAAACACAG) were used to generate the ST tail and signal anchor regions including a NarI site at the carboxyl-terminal end. Using the TR coding sequence (32) subcloned into bs' as a template, the T3 primer and Primer 11 (GT-CAAGGCGCCGAACCAAAAACTG) were used to generate a fragment encoding the extracellular domain of the TR including a NurI site at the amino-terminal end. These fragments were cloned into bs+ in the XbaI and BamH I sites and then into pSVL in the BamHI site. The TR coding sequence (32) was ligated into the BamHI site of pSVL for expression in Cos-1 cells.
"Funsfection of Cos-1 Cells Cos-1 cells were maintained in DMEM (Dulbecco's modified Eagle's medium), 10% fetal bovine serum (FBS), plated on 100-mm dishes, and incubated at 37 "C in a 5% C02 incubator. Cells were transfected using lipofectin in Opti-MEM I with 55 p~ P-mercaptoethanol according to the Life Technologies, Inc. instructions and as described previously (29). Expression of transfected proteins was typically allowed to continue for 1 6 3 6 h.

Infection of CHO Cells with Recombinnnt Vaccinia Virus
To achieve the higher level of expression required by the various NA constructs, we used a recombinant vaccinia virus possessing the T7 RNA polymerase (33). Cells were infected as described previously (33). Briefly, CHO cells were plated on 100 mm dishes and maintained in a-MEM, 10% FBS. When these cells became 70% confluent, they were washed with a-MEM containing no serum and infected with the recombinant vaccinia virus in 1.5 ml of a-MEM containing no serum at a multiplicity of infection of approximately 5. Infection was allowed to continue for 1 h and at this time the cells were washed with 10 ml of Opti-MEM I, 55 p~ 6-mercaptoethanol in preparation for transfection. The target gene under the control of the T7 promoter in either bs' or a pGEM vector, was introduced into the infected cells by the lipofectin transfection procedure described above. In this case however, the transfection was allowed to continue for 14-16 h and fresh a-MEM, 10% FBS was not added. Pulse-chase analysis of expressed proteins proceeded immediately, and labeling and chase times could not exceed 10-12 h because the cytopathic effect of the virus began to kill the cells.

Immunofluorescence Microscopy
Cos-1 and CHO cells expressing wild type, mutant, and chimeric proteins were processed for immunofluorescence microscopy as previously described (29). Following fixation, permeabilization (if required), and blocking steps, cells were incubated for 45 min with a 1:lOO dilutions of rabbit affinity-purified antibodies raised against the rat liver ST, rabbit polyclonal antibodies raised against the WSN NA (Nl), or a monoclonal antibody (IgG1) against the human TR (Boehringer Mannheim) in blocking buffer. Following four PBS washes, appropriate secondary antibodies conjugated to fluorescein isothiocyanate were then incubated with the cells at a 1:lOO dilution in blocking buffer for 45 min. Again, cells were washed four times with PBS prior to mounting on glass slides using 20 pl of mounting media (15% (w/v) Vinol205 polyvinyl alcohol, 33% (vh) glycerol, 0.1% azide, pH 8.5 in PBS). Cells were visualized and photographed using a Nikon Axiphot microscope equipped with epifluorescence illumination and a 60x oil emersion Plan Apochromat objective.

Cell Surface Biotinylation
Cells were transfected and labeled as described above, culture dishes were then placed on ice and washed with cold PBS. Cells were then biotinylated as previously described (30). Three milliliters of cold PBS containing 1 mdml sulfosuccinimidyl (biotinamidyl) hexonate was added to the cells, and the incubation was continued for 30 min with gentle rocking at 4 "C. Cells were washed four times with 5 ml of PBS containing 50 m~ lysine to block any unreacted reagent. Cells were lysed and proteins immunoprecipitated with appropriate antibodies and immobilized secondary reagents as described above. The proteins were eluted from Protein Gor Protein A-Sepharose beads by boiling for 5 min in 100 pl of 0.2 M Tris-HC1, pH 8.8, 1.0% SDS, 0.5 l~l~ EDTA. Pellets were washed further in 150 pl of lysis buffer containing 3% Nonidet P-40. Both the elution and wash were combined and 25 pl was reserved as "total." The remaining 225 pl were rotated for 1 h with 40 pl of a 50% suspension of streptavidin-agarose (Pierce Chemical Co.) at 4 "C. Complexes were pelleted, washed, and eluted as described above in preparation for SDS-polyacrylamide gel electrophoresis. In the CHOrecombinant vaccinia expression system, we and others2 have found that the level of wild type NA found on the cell surface is very low; however, we attempted to quantitate these results by densitometry scanning to try to compare the levels of the NA, STNA, and STKKNA at the cell surface relative to the amount of total protein expressed in each case.  DNA fragments were ligated together using the restriction enzyme sites encoded by the specific oligonucleotides and the constructs were subcloned into the pSVL expression vector as described under "Methods." C, construction of the S T R , STNA, and STKKNA chimeric proteins. PCR was performed, as described under "Methods," using TR, ST, or NA coding sequences in bs+ as templates and oligonucleotides encoding specific restriction enzymes sites as primers. STNA was constructed by fusing the ST tail and signal anchor regions to the NA stalk and head regions and subcloning these into the pGEM-2 vector. STKKNA was constructed as STNA but the two lysines at the beginning of the ST stem were retained. This chimera was subcloned into the bs+ vector. S T R was constructed by fusing the tail and signal anchor regions of the ST to the extracellular domain of TR and then subcloning these into the pSVL expression vedor.

DWAXN
side of the membrane in order to prevent the protein from assuming a type I orientation (positive-in rule (36)) and to prevent any signal anchor cleavage on the luminal side (37,38). The ASSA mutant proteins all contained the full nine-amino acid cytoplasmic tail, but only the stem sequences, KKGSD, bordering the signal anchor on the luminal side of the membrane.
When expressed in Cos-1 cells, metabolically labeled with 35S-Express protein label, and immunoprecipitated with anti-ST antibodies, the ATAS and the four ASSA mutant proteins migrated on SDS-polyacrylamide gels with the approximate molecular mass of the AStem-ST protein (-43-44 kDa) (Fig. 2). Localization of the ATAS protein in Cos-1 cells by indirect immunofluorescence microscopy demonstrated that it is localized in the Golgi complex just like the wild type ST enzyme (Fig. 3). These experiments demonstrated that the signal anchor region of the ST, MKKK flanking this region on the cytoplasmic side of the membrane, and KKGSD flanking the signal anchor region on the luminal side of the membrane, are able to retain the ST catalytic domain in the Golgi complex. scope. No cell surface immunofluorescence was detected for any of these the ASSA mutants (Fig. IA). In these mutants, the signal anchor region of the AStem-ST was altered by the replacement of 4-5 amino acids along its entire length using amino acids from the signal anchor region of the type I1 plasma membrane protein, influenza NA. Analysis of the localization of the these mutants by indirect immunofluorescence microscopy also demonstrated that these proteins are localized in the Golgi complex like the wild type ST (Fig. 3). These results demonstrated that even in the absence of the stem region, the sequence of the ST signal anchor region is not critical for Golgi retention.

Replacement of the Sialyltransferase Signal Anchor Region with Other Signal Anchor Sequences of Different Lengths Does
Not Interfere with Golgi Retention-To determine whether that signal anchor region of the ST was necessary for Golgi localization and to investigate the impact of the length of the signal anchor region on G o l g i localization, we constructed a series of replacements of the ST signal anchor region using PCR techniques (Fig. 1B ). In these mutant proteins, we replaced the ST signal anchor region with the entire signal anchor region of the plasma membrane protein, influenza NA (29 amino acids), in the SA29 protein, or the first 23 and 17 amino acids in this region in the SA23 and SA17 proteins, respectively. Following construction of these altered coding sequences, the mutant proteins were transiently expressed in Cos-1 cells and analyzed by metabolic labeling of the expressing cells, immunoprecipitation of the expressed proteins, analysis of these immunoprecipitated proteins by SDS polyacrylamide gel electrophoresis, and fluorography. The SA29 and SA23 proteins were immunoprecipitated by the anti-ST antibody and migrated with molecular masses predicted based on an increase in length of 12 and 6 amino acids, respectively (Fig. 4). Analysis of the localization of both the SA29 and SA23 by indirect immunofluorescence microscopy demonstrated that they are expressed in the Golgi complex like the wild type ST (Fig. 5). Overexpression of these mutant ST proteins in Cos-1 cells did not lead to expression at the cell surface (data notBshown). In contrast, the SA17 protein appeared to express very poorly and was difficult to detect following immunoprecipitation (Fig. 4). Not surprisingly, indirect immunofluorescence microscopy demonstrated that this mutant ST protein is weakly expressed in the ER (Fig. 5). It is possible that the first 17 amino acids of the NA signal anchor region is not sufficient to span the membrane and this results in a grossly misfolded protein which does not leave the ER and is either poorly recognized by anti-ST antibodies and/or rapidly degraded. Similar changes in the intact NA protein demonstrated that 23 amino acids of the signal anchor region are adequate to span the membrane, whereas 17 amino acids ofthe signal anchor are not.2 These results strongly suggest that the signal anchor region of the ST can be replaced by a signal anchor region which adequately spans the membrane no matter what its length in amino acids. are the primary determinants in the Golgi localization signal. If this is the case, then these two chimeric proteins would be predicted to be inefficiently or partially localized in the Golgi apparatus.

Sialyltransferase Cytoplasmic Tail and Signal Anchor Region Are Not Sufficient for Golgi Retention of Chimeric Proteins
The STl'R and wild type TR in the pSVL expression vector were transiently expressed and localized in Cos-1 cells, while the STNA and wild type NA in either the pGEM-2 or bs' vector were transiently expressed in CHO cells using the recombinant vaccinia virus expressing the T7 polymerase (33). The S'ITR and STNA proteins and the wild type TR and NA proteins were immunoprecipitated from lysates of metabolically labeled cells using the appropriate anti-TR or anti-NA antibodies and immunoprecipitated proteins analyzed by SDS-polyacrylamide gel electrophoresis and fluorography. The S'ITR and STNA proteins and the wild type TR and NAare efficiently expressed and folded in Cos-1 or CHO cells, as demonstrated by the immunoprecipitation of radiolabeled proteins of the appropriate molecular mass from these cells (Fig. 6).
Indirect immunofluorescence microscopy of permeabilized (internal) and unpermeabilized (surface) cells expressing either the TR, NA, or SlTR and STNA chimeric proteins was used to determine whether the ST tail and signal anchor were suffi- The significant surface expression of these chimeric proteins, particularly the S"R chimera, demonstrated that the ST cytoplasmic tail and signal anchor regions alone are not sufficient to localize these chimeric proteins in the Golgi (Figs. 7 and 8).
The level of surface expression of these chimeras appeared to be related to the level of their expression in the cells. This is particularly well illustrated in the immunofluorescence micrographs of the STTR chimera (Fig. 8), in which two cells expressing different levels of this protein are presented. The cell expressing lower protein levels demonstrates less observable surface staining, while the cell expressing higher levels of protein demonstrates significant surface staining. These results contrast with what has been observed in the overexpression of the wild type ST. As has been previously demonstrated by our laboratory (29) and others (28), the overexpressed ST does not move to the cell surface, but appears to backup along the cisternae of the Golgi and into the ER. It is clear from these results that chimeric proteins possessing only the ST cytoplasmic tail and signal anchor region are poorly retained in Golgi apparatus and are lacking sequences andlor characteristics required for efficient, nonsaturable Golgi retention.
Luminal Lysine Residues Significantly Improve Golgi Retention of Sialyltransferase-Neuraminidase Chimeric Proteins-To reconstitute the complete retention signal containing the luminal flanking sequences of the stem region, in addition to those in the cytoplasmic tail and a spacer signal anchor region, we constructed the STKKNA chimeric protein which differed from the STNA chimeric protein in possessing KKSTQ rather than HMIQ flanking the ST signal anchor region on the luminal side of the membrane. Following expression in CHO cells using the recombinant vaccinia-T7 polymerase expression system and transfection with the STKKNA construct in pGEM-2, we analyzed the localization of this chimera by indirect immunofluorescence microscopy using anti-NA antibodies. From these experiments, it is clear that the STKKNA protein is much more efficiently retained in the Golgi and demonstrates little to no surface expression even when highly expressed in CHO cells (Fig. 7).
To demonstrate the difference in surface expression biochemically, we biotinylated cell surface proteins of metabolically labeled and chased CHO cells expressing NA, STNA, or STKKNAin order to compare intracellular and surface forms of each protein (Fig. 9). Quantitation of the protein bands by densitometry scanning showed that the proportion of total STNA protein at the CHO cell surface significantly higher than the proportion of total STKKNA protein at the cell surface. If we calculate the percentage of total protein which is found a t the cell surface for NA, STNA, and STKKNA, and we arbitrarily set the surface percentage for NA a t loo%, we calculate that surface levels of STNA and STKKNA are 71% and 20% of the surface NA level, respectively. These results suggest that the stalk and head regions of NA are not efficiently retained in the Golgi apparatus by the tail and signal anchor regions of the ST unless two stem lysine residues are present in the chimeric protein.
The processing of these proteins' Asn-linked oligosaccharides also suggested that the STKKNA protein was retained more efficiently in the Golgi than the STNA protein (Figs. 6 and 9). The wild type ST in rat liver and expressed in Cos-1 cells possesses a high proportion of Asn-linked carbohydrate structures which lack terminal processing (29). In contrast, the secreted form of the ST catalytic domain (sp-ST (29)) expressed in either Cos-1 cells (22) or CHO cells possesses more completely processed Asn-linked oligosaccharides. These results suggest that the wild type ST protein which is retained in the Golgi apparatus is unable to be completely terminally glycosylated, while the soluble form of the protein may be more accessible to glycosyltransferases and therefore more completely processed. teins on the CHO cell surface. CHO cells were grown on coverslips and infected with the recombinant vaccinia virus expressing the T7 polymerase at an multiplicity of infection of 5. These cells were then transfected with the cDNA encoding the chimeric proteins using lipofectin method. After expression for 16 h, cells were fixed with 3% paraformaldehyde followed by permeabilization with 0.1% Triton X-100 if internal staining was desired. The proteins were detected by incubation with rabbit anti-NA antibodies followed by incubation with fluorescein isothiocyanate-conjugated goat anti-rabbit I& second antibody. Cells were visualized using a Nikon Axiphot fluorescence microscope. Magnification, x 750.
We observe the same phenomena with the incompletely retained STNA protein and the more completely retained STK-KNA protein. As demonstrated by the differences in migration on SDS-polyacrylamide gels (Fig. 6) and the increased sensitivity of the STKKNA to endo-13-N-acetylglycosaminidase H digestion (data not shown), the oligosaccharides of the STK-KNA, which is efficiently retained in the Golgi, are less well processed than those of the STNA protein, which is observed a t the cell surface and migrates more slowly on SDS-polyacrylamide gels (Figs. 6 and 9).

DISCUSSION
In this study we provide evidence that the sequences of the ST signal anchor are not necessary or sufficient for the Golgi retention of the wild type ST or chimeric proteins. Replacement of all or some of the signal anchor in the presence or absence of most of the stem region results in no change in ST Golgi retention (SA29 and SA23 (Fig. 5) and ASSA mutants (Fig. 3)).
These results, and the Golgi localization of the ATAS protein, suggest that the sequences flanking the signal anchor region may be the most critical portion of the ST Golgi retention signal and that the signal anchor region plays the role of a spacer between these two portions of the retention signal. Localization of chimeric proteins, consisting of the ST cytoplasmic tail and signal anchor fused to either the TR or NA extracellular domains, demonstrates that the tail and signal anchor are not sufficient for efficient Golgi retention of these plasma membrane proteins (STNA and STTR, Figs. 7 and 8). More efficient cells. Cells were infected with the recombinant vaccinia virus expressing the T7 polymerase, transfected using the lipofectin method, then pulsed labeled for 2 h with Y3-Express protein label in methionine-free DMEM and chased for 6 h in DMEM containing 10% fetal calf serum. Cells were then incubated with sulfosuccinimidyl (biotinamidyl) hexonate for 30 min at 4 "C, washed extensively with 50 mM lysine in PBS, and then lysed, and the expressed proteins were immunoprecipitated with goat anti-NA antibodies. Following several washes of the Protein G-Sepharose beads and elution of the immunoprecipitated proteins, VI0 of the eluate was reserved as the "total * and 9/10 ("surface") was incubated with streptavidin-agarose for 1 h a t 4 "C. The complexes were pelleted, washed, and eluted by boiling in sample buffer. Proteins in each fraction were analyzed on SDS-polyacrylamide gels and visualized by fluorography. 'T-Methylated molecular mass markers shown are ovalbumin (46 kDa) and bovine serum albumin (69 kDa).
retention of a STNA chimera is achieved only after two lysines are placed adjacent to the signal anchor on the luminal side of the membrane (STKKNA, Fig. 7). These data suggest that these flanking lysines may have reconstituted a ST retention signal which consists of cytoplasmic KKK, a spacer transmembrane region and luminal KK or KKGSD. Immunofluorescence microscopy is a good method for obtaining general information on the localization of proteins in cells. However, this technique does not allow a precise cisternal localization of proteins in the Golgi complex (39). It is therefore possible that the Golgi-localized mutant and chimeric proteins analyzed in this work are localized in different Golgi cisternae than the wild type protein, and that changes within, or replacement of, the signal anchor region may result in slightly altered cisternal localization. Because of the unusual dual localization of the ST in the trans Golgi and TGN in the rat hepatocyte (401, its differential localization in the Golgi in intestinal absorptive and goblet cells (41), and the potential for more than one independent ST Golgi retention signal (291, it will be ultimately important to determine the precise localization of the ST mutant and chimeric proteins using immunoelectron microscopy. The transmembrane regions of many Golgi proteins appear to be completely sufficient for Golgi retention, while other proteins require additional sequences flanking these transmembrane regions for efficient Golgi retention (reviewed in Refs. 8 and 9). The first transmembrane region of the IBV E l glycoprotein is necessary and sufficient for Golgi localization, and specific uncharged polar residues in this region are required for the retention of a chimeric protein containing this transmembrane region (20,21). The ~-1,4-galactosyltransferase, which is localized in the trans cisternae of the Golgi, also possesses a Golgi retention signal which appears to be entirely contained within its signal anchor region (23)(24)(25)(26). Golgi retention signals of the N-acetylglucosaminyltransferase I and the ST appear to require sequences flanking their transmembrane regions for Golgi retention. Tang et al. (27) demonstrated that the cytoplasmic tail and signal anchor of the medial Golgi protein, N-acetylglucosaminyltransferase I, are not sufficient for Golgi retention of a dipeptidyl peptidase IV reporter protein, and that complete Golgi retention is achieved only when 12 amino acids from the N-acetylglucosaminyltransferase I stem region are included. Similar requirements have been demonstrated for the ST by our laboratory (29) and that of Munro (28) (this work).
For protein sequences to be considered retention signals they must be shown to be both necessary and sufficient. Replacement of the ST signal anchor with either 29 or 23 amino acids of the influenza NA signal anchor does not change the Golgi localization of the ST. Since the influenza NA is a plasma membrane protein, no Golgi retention signals should be present in its sequences, and therefore this experiment demonstrates that the ST signal anchor is not necessary for its retention in the Golgi. This being the case, what does constitute the ST Golgi retention signal(s)? Previous results suggested that the stem region might contain a separate retention signal (28,29), and results in this work suggest that the sequences flanking the signal anchor region may constitute a separate Golgi retention signal (ASSA mutants). Munro (28) replaced the ST signal anchor region with the 23-amino acid transmembrane region of dipeptidyl peptidase IV and found that the resulting chimeric protein exhibited cell surface staining. He also demonstrated that 17 leucines plus the ST stem were sufficient for Golgi localization of a chimeric protein containing a lysozyme carboxyl-terminal domain. However, the presence of 23 leucines plus the ST stem resulted in the cell surface expression of a similar chimera. We suggest that these chimeras may not allow for the correct presentation of the ST sequences which flank the membrane on the luminal side. This would lead to a masking of the retention signal and the appearance of these chimeras at the cell surface. In our SA29 and SA23 proteins, its seems that the two lengths of the NA transmembrane region fold so that they are able to correctly present the ST luminal (and perhaps cytoplasmic) flanking sequences in a way that they can achieve retention in the Golgi.
To determine whether the ST signal anchor region is suffcient for Golgi retention, we and others (28,30) have constructed a series of chimeric proteins containing this region. Both our STNA and STTR chimeras which possess the ST cytoplasmic tail and signal anchor region fused to the extracellular domains of NA and TR, are poorly retained in the Golgi. Golgi retention is significantly increased when two luminal lysine residues flanking the ST signal anchor region are included in the STNA chimera (STKKNA), suggesting that we have (partially) reconstituted a Golgi retention signal which consists of lysine residues flanking a spacer transmembrane region. In order to achieve complete Golgi retention, we attempted to construct chimeras which included the entire ST stem region, however these chimeras were misfolded and rapidly degraded.3 Consistent with our results, Munro (28) demonstrated that the signal anchor of the ST did not completely retain a chimeric protein in the Golgi. In contrast, Wong et al. (30) observed no cell surface expression of a chimera in which the 17-amino acid ST signal anchor region replaced the transmembrane region of the plasma membrane protein, dipeptidyl peptidase IV. It was suggested that the cell surface expression that Munro observed with his chimera, possessing only the ST signal anchor region, was a result of the aberrantly high level of expression in Munro's transient expression system (30). R evious results have demonstrated that the wild type ST protein is never expressed on the cell surface of Cos-1 cells or CHO cells even when overexpressed in a transient expression system (28,29) (this work). In contrast, the cell surface levels of the STNA and STTR chimeras are increased with increased expression (Fig. 8). These observations suggest that the ST cytoplasmic tail and signal anchor are not adequate for the efficient Golgi retention. Addition of two luminal flanking lysines significantly improves Golgi retention, however some of the STKKNA chimera is still observed at the cell surface (Fig. 7). Interestingly, the STTR protein naturally possesses a lysine flanking the ST signal anchor region and yet this protein is very poorly retained in the Golgi (Fig. 8). The inefficient retention of the chimeric proteins could reflect both the lack of complete retention signals and the presence of carboxyl-terminal regions which either cause the improper folding of the amino-terminal retention regions andor may not accommodate the ST Golgi retention mechanism.
It is likely that certain characteristics of the :Item and catalytic domain of the ST perfectly accommodate the Golgi retention mechanism and allow for absolute retentior.. Variations in Golgi retention efficiency of chimeras may result from how well the non-Golgi reporter proteins can accommodate the Golgi retention mechanism. The ability to contact ST' or other resident Golgi proteins may be important, especially if the Golgi retention mechanism involves oligomerization With this in mind we chose the NA as a reporter protein. NA has basically the same type I1 domain structure as the ST and it also forms disulfide-bonded dimers which ultimately associate into tetramers (42,431. The contacts which exist in these NA oligomers are known to occur between NA transmembrane, stalk and head regions (42,431. We reasoned that if the Golgi retention mechanism involved interactions in carboxyl-terminal regions, then the NA stalk and head groups would most l:.kely be able to make these contacts and ultimately be retained in the Golgi, when fused with the correct ST retention se,quences. This worked to a large degree in the STKKNA protein, however this chimera was still not completely retained in the Golgi like the wild type ST. Until a mechanism for Golgi retention is elucidated, it will be difficult to know what features in these carboxyl-terminal regions are important for retention. Based on the requirements for ST Golgi retcmtion one can speculate on the type(s) of Golgi retention mechanisms which may be likely for this glycosyltransferase and potentially other Golgi proteins (see Ref. 8 for review). One potential retention mechanism is a protein or lipid receptor-mediated retention mechanism. Specific association with membranes of a defined lipid composition is a logical model since the sequences re-K. Colley, unpublished results. quired for the Golgi retention of glycosyltransferases and other Golgi proteins are found in and around their membrane spanning regions. A gradient of cholesterol is known to exist across the Golgi cisternae, and similarly, particular lipids are thought to be enriched in either the cis or trans face of this organelle, with the cis cisterane having a lipid composition most like the ER and the trans cisternae having a lipid composition most like the plasma membrane (4447). Lipid gradients could be the basis for the gradient of glycosyltransferases and glycosidases observed across the Golgi, and gradual alterations in these lipid gradients would also accommodate the variations in glycosyltransferase and glycosidase localization observed in different cell types (16, 41, 48). Several examples of cell type dependent variation in glycosyltransferase localization have now been reported. These results were particularly surprising since a cis Golgi localization has been always assumed for the a-mannosidase I in the traditional ordered compartmentation of enzymes in the Asn-linked glycosylation pathway. Although ER, Golgi, and plasma membrane lipid compositions are available for rat liver and kidney cells, further support for a lipid-Golgi protein interaction model must await the precise analysis of Golgi cisternae lipids in a variety of cell types.
A second potential mechanism is that of self-aggregation or hetero-oligomerization in the specific microenvironments of the Golgi cisternae. Possibly, the glycosyltransferases or other Golgi proteins are induced to form self-aggregates or oligomers with other resident Golgi proteins because of the specific lipid composition, pH levels, and/or calcium concentrations in their resident Golgi cisternae. One potential problem with this mechanism is that at this time only dimers of the ST and It is clear that the ST does not bypass the Golgi when overexpressed, but rather appears to backup along the Golgi cisternae and into the ER (28,29). This observation suggests that the ST Golgi localization mechanism may not exclusively involve a receptor-mediated process because it is not saturable. We can contrast this observation with the behavior of a group of mammalian and yeast proteins which contain Golgi localization signals in their cytoplasmic tails and upon overexpression are either found at the cell surface (TGN38) (18) or in the yeast vacuole (Kex2p) (19). This group of Golgi proteins may interact with clathrin and/or clathrin-associated proteins via tyrosinebased signals in their cytoplasmic tails (51). The ambiguous nature of the glycosyltransferase Golgi retention signal(s1 also draws into question an exclusively receptor-mediated retention mechanism. One possibility is that a combination of events occurs to achieve efficient Golgi localization. Resident Golgi proteins could first associate with specific lipid compositions via sequences in and around their membrane-spanning regions, then due to either the microenvironment of the particular Golgi cisternae or perhaps a certain critical protein concentration, self-aggregate or interact with other resident Golgi proteins or lipids resulting in the formation of stable proteo-Localization Sequences lipid complexes and their retention. It seems probable that the sequences required for the initial retention and the formation of resident Golgi complexes could be identical or overlapping so that different Golgi proteins would appear to have slightly different requirements for Golgi retention. The Golgi protein sequences spanning and flanking the membrane are most likely to make contacts with lipid head groups and fatty acid chains, while both these regions and luminal stem sequences would be involved in any further protein-protein contacts. The results presented in this paper suggest that sequences flanking the ST signal anchor region may actually make membrane contacts while the signal anchor sequences themselves may function to appropriately space these sequences across the membrane and ultimately act with the stem region in complex formation.
The current challenge is now to uncover evidence for specific Golgi retention mechanisms and to further characterize the regions of the ST and other glycosyltransferases and Golgi proteins which participate in these mechanisms. Electron microscopic and biochemical analysis of wild type, mutant, and chimeric proteins should provide insight into these problems and ultimately lead to a greater understanding of the structure and function of the Golgi complex.