Retrograde transport of the mannosyltransferase Och1p to the early Golgi requires a component of the COG transport complex.

The yeast COG complex has been proposed to function as a vesicle-tethering complex on an early Golgi compartment, but its role is not fully understood. COG complex mutants exhibit a dramatic reduction in Golgi-specific glycosylation and other defects. Here we show that a strain carrying a COG3 temperature-sensitive allele, cog3-202, clearly exhibited the glycosylation defect while exhibiting nearly normal secretion kinetics. Two Golgi mannosyltransferases, Och1p and Mnn1p, were mislocalized in cog3-202 cells. In cog3-202 cells Och1-HA was found in lighter density membranes than in wild type cells. In sed5(ts) and sft1(ts) strains, Och1p rapidly accumulated in vesicle-like structures consistent with the delivery of Och1p back to the cis-Golgi on retrograde vesicles via a Sed5p/Sft1p-containing SNARE complex. In contrast to cog3-202 cells, the membranes in sed5(ts) cells that contained Och1p were denser than in wild type. Together these results indicate that Och1p does not accumulate in retrograde vesicles in the cog3-202 mutant and are consistent with the COG complex playing a role in sorting of Och1p into retrograde vesicles. In wild type cells Och1p has been shown previously to cycle between the cis-Golgi and minimally as far as the late Golgi. We find that Och1p does not cycle via endosomes during its normal itinerary suggesting that Och1p engages in intra-Golgi cycling only. However, Och1p does use a post-Golgi pathway for degradation because a portion of Och1p was degraded in the vacuole. Most surprisingly, Och1p can use either the carboxypeptidase Y or AP-3 pathways to reach the vacuole for degradation.

The yeast COG complex has been proposed to function as a vesicle-tethering complex on an early Golgi compartment, but its role is not fully understood. COG complex mutants exhibit a dramatic reduction in Golgi-specific glycosylation and other defects. Here we show that a strain carrying a COG3 temperature-sensitive allele, cog3-202, clearly exhibited the glycosylation defect while exhibiting nearly normal secretion kinetics. Two Golgi mannosyltransferases, Och1p and Mnn1p, were mislocalized in cog3-202 cells. In cog3-202 cells Och1-HA was found in lighter density membranes than in wild type cells. In sed5 ts and sft1 ts strains, Och1p rapidly accumulated in vesicle-like structures consistent with the delivery of Och1p back to the cis-Golgi on retrograde vesicles via a Sed5p/Sft1p-containing SNARE complex. In contrast to cog3-202 cells, the membranes in sed5 ts cells that contained Och1p were denser than in wild type. Together these results indicate that Och1p does not accumulate in retrograde vesicles in the cog3-202 mutant and are consistent with the COG complex playing a role in sorting of Och1p into retrograde vesicles. In wild type cells Och1p has been shown previously to cycle between the cis-Golgi and minimally as far as the late Golgi. We find that Och1p does not cycle via endosomes during its normal itinerary suggesting that Och1p engages in intra-Golgi cycling only. However, Och1p does use a post-Golgi pathway for degradation because a portion of Och1p was degraded in the vacuole. Most surprisingly, Och1p can use either the carboxypeptidase Y or AP-3 pathways to reach the vacuole for degradation.
The trafficking of lipid and protein cargo between membrane-enclosed compartments involves the formation of vesicular carriers that subsequently fuse with an appropriate compartment (1). The specificity of vesicle fusion is regulated on multiple levels. An initial contact between the vesicle and target membrane appears to be mediated by "tethering factors" that consist either of long coiled-coiled proteins or large, multisubunit complexes (2,3). The subsequent fusion reaction, mediated by interaction between SNARE proteins on the vesicle and target membranes, adds additional specificity to the process (4,5).
Because each organelle of the secretory pathway contains resident proteins necessary for compartment function, resident proteins must be sorted from secretory cargo during vesicular transport. For example, resident glycosyltransferases of the Golgi apparatus are maintained in an asymmetric steady-state distribution within Golgi cisternae despite a high rate of secretory protein trafficking through these compartments (6). Both yeast and mammalian Golgi glycosyltransferases are typically type II integral membrane proteins with a short cytosolic domain, a single transmembrane domain, and a relatively large luminal domain. Two main models have been proposed to account for localization of Golgi glycosyltransferases. The kin recognition model posits that oligomerization of glycosyltransferases within Golgi cisternae prevents their forward transport by preventing entry into anterograde vesicles (7,8). However, in recent years new evidence strongly supports the idea that Golgi cisternae themselves represent anterograde trafficking structures and that most, if not all, vesicles that bud off of Golgi cisternae mediate retrograde transport (9 -12). Thus the role of oligomerization in Golgi localization remains obscure. Golgi membrane bilayers are thinner than that of the plasma membrane, and a relatively short transmembrane domain is an important determinant for localization of the medial and trans-Golgi residents (13). Based on these observations, it has been proposed that glycosyltransferases localize to Golgi compartments because the transmembrane domain exhibits the best "fit" with the bilayer thickness of Golgi membranes (14). Indeed, glycosyltransferases have been mislocalized to the plasma membrane as a result of modest increases in their transmembrane domain length. In addition, a role for cytosolic domain sorting determinants has also been noted in several cases (15)(16)(17). Golgi localization appears to be a dynamic process with some Golgi glycosyltransferases cycling between their principal site of residence and distal compartments. Dynamic localization is exemplified by yeast Och1p, an ␣1,6-mannosyltransferase localized to the cis-Golgi (18 -20). As part of its normal itinerary, Och1p is transported at least as far as the late Golgi (trans-Golgi network) with a half-time of ϳ5 min and is then actively recycled back to the cis-Golgi (21). In contrast, certain other early Golgi membrane proteins maintain their localization by constantly cycling between the Golgi and endoplasmic reticulum (ER) 1 (22)(23)(24). These proteins utilize COPI vesicles for retrograde transport to the ER and COPII vesicles to cycle back to the Golgi.
Although much is known regarding the trafficking patterns and sorting signals of resident late Golgi membrane proteins (25)(26)(27), relatively little is known about the distal compartments that Golgi glycosyltransferases visit or how they are retrieved from these compartments for transport back to their site of function. Glycosyltransferases have been found in Golgiderived COPI-coated vesicles thought to be mediating intra-Golgi retrograde transport (10) suggesting that the retrieval compartment may be a late Golgi cisternae. However, it is also possible that these cargoes may be cycled back from a more distal compartment and use intra-Golgi retrograde traffic in the final stage of their journey.
A Golgi-associated tethering complex necessary for Golgi trafficking and morphology has been characterized in both yeast and mammalian cells (3). This complex, originally called the Sec34-Sec35 complex and now referred to as the conserved oligomeric Golgi (COG) complex, contains eight subunits. Yeast COG2/SEC35 and COG3/SEC34/GRD20 exhibit genetic interactions with genes involved in ER-to-Golgi trafficking, are pleiotropically required for efficient secretion, and encode proteins that are required for optimal transport for in vitro ERto-Golgi transport assays (28 -31). However, yeast strains carrying mutations in COG complex subunits exhibit a number of other phenotypes including a loss of retention of the late Golgi resident protein Kex2p, missorting of the vacuolar hydrolase carboxypeptidase Y (CPY), failure to recycle the late secretory v-SNARE Snc1p, and a striking decrease in Golgi-specific glycosylation (32)(33)(34). These results suggest that the COG complex is involved in both intra-Golgi retrograde transport and transport from the endosomal system back to the Golgi. This idea is further supported by the observation that the COG complex associates with the Golgi vesicle coat complex COPI and with SNAREs involved in intra-Golgi recycling (31). It is possible that a defect in retrograde transport of anterograde trafficking machinery in cog mutants could explain the ER-to-Golgi defects. Taken together it appears that the COG complex is either directly involved in both ER-to-Golgi transport and retrograde transport to the Golgi or, alternatively, that the primary role is in retrograde trafficking with an indirect role in ER-to-Golgi transport.
In this study we have investigated the trafficking pathways and degradation of the glycosyltransferase Och1p. We show that Och1p is mislocalized in a cog3 mutant consistent with the observed glycosylation defects in cog mutants. Our results suggest that in wild type cells Och1p is retrieved from the late Golgi back to the early Golgi via vesicular transport and does not reach more distal compartments such as endosomes or the plasma membrane.

MATERIALS AND METHODS
General Methods and Antibodies-The production of minimal (synthetic dextrose) and rich (YPD) yeast media, the genetic manipulation of yeast strains, and all general molecular biology methods were performed as described (35) or as otherwise noted. Rabbit polyclonal antibodies against alkaline phosphatase (ALP) (36) and Kex2p (33) have been described previously. Rabbit polyclonal antibodies against invertase, Tlg1p, and phosphoglycerol kinase were gifts from S. Emr (University of California, San Diego), Hugh Pelham (MRC, Cambridge, UK), and T. Stevens (University of Oregon, Eugene, OR), respectively. Rabbit and mouse anti-hemagglutinin (HA) epitope and mouse anti-c-Myc epitope antibodies were purchased from Covance (Richmond, CA). Mouse antibodies against Dpm1p and Vph1p were from Molecular Probes Inc. (Eugene, OR).
Rabbit antibodies were raised against the product of the VPS10 gene. Plasmid pAAC209 (37) was used to express a protein consisting of the maltose-binding protein fused to a region from the luminal domain of Vps10p in Escherichia coli. The fusion protein was purified by using an amylose resin column (New England Biolabs, Beverly, MA), and antibodies were raised against this protein in New Zealand White rabbits.
Plasmids and Yeast Strains-Plasmids and yeast strains used this study are described in Table I. Plasmids pSN335 and pSN336 were generated by inserting the 2.7-kbp XhoI-SacII fragments from pSB1-25 (cog3-201 allele in pRS316) and pSB1-212 (cog3-202 allele in pRS316), respectively, into the XhoI/SacII site of pRS313 (38). pSB29 was created by inserting the 4.3-kbp EcoRI fragment from pRB58 (YEp24 containing the SUC2 gene) into the EcoRI site of pRS314. pSN390 was constructed by inserting the 2.6-kbp SalI-SpeI fragment from pOH (21) into the SalI/SpeI sites of pRS314. pPB18 was constructed by inserting the 2.6-kbp SalI-HindIII fragment from pOH into the SalI/HindIII sites of pRS315. BY4743 and derived yeast strains were obtained from Invitrogen. All other yeast strains were derived from SEY6210 or from the meiotic progeny of a diploid formed from mating SEY6210 and SEY6211 (39) derivatives. Replacement of the RCY1 gene in strain SHY35 with the His3MX6 cassette was achieved by using a PCR-based method (40) resulting in strain SNY140. The same method was used to replace the APM3 gene in strain SNY17 with His3MX6 resulting in strain SNY106. A centromeric plasmid carrying the pep12-49 temperature-sensitive allele, pRS414-pep12-49, was obtained from B. Horazdovsky (Mayo Clinic, Rochester, MN). The 1.2-kbp ClaI-SmaI fragment from this plasmid was inserted into pRS306 digested with KpnI, blunted with Klenow polymerase, and then cut with ClaI. The resulting plasmid, pSH24, was digested with KpnI and was transformed into SNY36-9A. Uraϩ transformants were grown nonselectively and then propagated in the presence of 5-fluoroorotic acid to select for UraϪ loop-outs. These were screened for temperature-sensitive CPY secretion phenotype giving rise to strain SNY156. To generate an apm3⌬ pep12-49 double mutant, strains SNY106 and SNY156 were mated, and the resulting diploid was sporulated and dissected giving rise to strain PBY36-1B.
Pulse-Chase Immunoprecipitation-The procedure for immunoprecipitation of Och1-HA from [ 35 S]methionine/cysteine-labeled cells was performed as described previously (41). Likewise, the immunoprecipitation of invertase from the cell-associated and media fractions was performed as described previously (33). Radioactively labeled proteins were quantified from gels using a PhosphorImager system (Fuji Photo Film Co., Tokyo, Japan). The half-time of Och1-HA turnover was determined by first calculating the percentage of protein remaining at a given time point relative to that present at the 0-min time point (defined as 100%). Linear regression analysis was then carried out on plots of the log of the percentage protein remaining as a function of time.
Subcellular Fractionation-The fractionation of organelles by differential centrifugation was carried out by incubating 10 A 600 units of cells, grown to log phase in selective minimal media, in 2 ml of 0.1 M Tris, pH 9.4, 10 mM dithiothreitol, 10 mM NaN 3 for 10 min at room temperature. The cells were pelleted and spheroplasted in a 2-ml solution containing 50 mM Tris, pH 7.5, 2 mM MgCl 2 , 1.4 M sorbitol, 10 mM NaN 3 , and 25 g/ml oxalyticase at 30°C. The spheroplasts were pelleted, washed once with 1.2 M sorbitol, and osmotically lysed in 1 ml of an ice-cold solution containing 25 mM sodium phosphate, pH 7.4, 200 mM mannitol, 1 mM EDTA, and protease inhibitors. Unlysed cells were pelleted at 500 ϫ g, and the supernatant was then centrifuged at 12,300 ϫ g for 12 min resulting in pellet (P12) and supernatant (S12) fractions. The S12 fraction was centrifuged at 150,000 ϫ g for 60 min to generate P150 and S150 fractions. The S150 fractions were trichloroacetic acid-precipitated, and pellets were washed with acetone. The S150 trichloroacetic acid pellets as well as the P12 and P150 membrane pellets were resuspended in 8 M urea, 5% SDS, 5% 2-mercaptoethanol, 50 mM Tris, pH 6.8, and equivalent percentages were subjected to SDS-PAGE followed by blotting to nitrocellulose. The blots were probed with the indicated primary antibodies followed by incubation with alkaline phosphatase-conjugated anti-rabbit or anti-mouse secondary antibodies and chemiluminescent detection using the Lumi-Phos substrate (Pierce). The blots were imaged using a Fuji LAS-1000 CCD camera and ImageReader LAS-1000 1.2 software (Fuji Photo Film Co., Tokyo, Japan). Images were further adjusted and formatted using Adobe Photoshop 7.
To fractionate organelles on sucrose density gradients, 450 A 600 units of cells were spheroplasted and washed using conditions described above and scaled up accordingly. The spheroplasts were resuspended in 5 ml of ice-cold lysis buffer (200 mM sorbitol, 50 mM potassium phosphate, pH 7.5, 1 mM EDTA) containing freshly added protease inhibitors and were Dounce-homogenized. The lysate was centrifuged at 1000 ϫ g for 10 min to generate S1 and P1 fractions. 2.0 ml of the supernatant (S1) fraction was then layered on top of a sucrose step gradient made in 10 mM HEPES-KOH, pH 7.6, 1 mM EDTA. The % sucrose and volume of each step were as follows from bottom to top: 1 ml 60%, 1 ml 45%, 1.5 ml 41.1%, 2 ml 36%, 2 ml 30.9%, 1.5 ml 27%, and 1 ml 22%. The gradients were centrifuged at 118,000 ϫ g for 17 h at 4°C. 15 fractions of 0.8 ml each were removed from the top and were sub-jected to immunoblot analysis as described above, and the captured images were quantified using the Fuji ImageGauge 3.3 software.
Fluorescence Microscopy-The procedures for preparation of fixed spheroplasted yeast cells and attachment to microscope slides were described previously (42). All secondary antibodies were diluted 1:500 before use. Simultaneous detection of Och1-HA and c-myc-Snc1p was achieved by incubating with each of the following reagents followed by extensive washing: (a) rabbit anti-HA and mouse anti-c-Myc, (b) biotinconjugated donkey anti-mouse IgG (H ϩ L), (c) Alexa488-conjugated goat anti-rabbit IgG (H ϩ L) and Texas Red-streptavidin. Sole detection of Och1-HA, MnnI-HA, and A-ALP was carried out by using the following incubations: (a) rabbit primary antibody against HA or ALP, (b) biotin-conjugated goat anti-rabbit IgG (HϩL), and (c) FITC-streptavidin. Yeast cells were photographed by using an Olympus BX-60 fluorescence microscope (Olympus, Lake Success, NY) equipped with a Hamamatsu C4742-95 digital camera (Hamamatsu Corp., Bridgewater, NJ). Images were initially captured using Openlab 3.1.7 software (Improvision Inc., Lexington, MA) and were processed into figures using Adobe Photoshop 7.0.

Invertase Is Secreted in an Underglycosylated Form in a cog3
Allele That Exhibits Little or No Block in Secretion-Previously, two classes of temperature-sensitive COG3/GRD20 alleles were isolated as exemplified by cog3-201 and cog3-202 (previously called grd20-1 and grd20-2, respectively) (33). At the nonpermissive temperature the cog3-201 allele caused a growth defect, a delay in invertase secretion presumably because of ER-to-Golgi trafficking defects, glycosylation defects, and mislocalization of the late Golgi resident protein Kex2p to the vacuole. Similarly, cog3-202 strains exhibited Kex2p localization defects; however, this class of mutants exhibited near normal growth at the nonpermissive temperature. In this study, we wished to focus on the role of COG3 in trafficking events distal to the ER-to-Golgi trafficking step. We hypothesized that the cog3-202 allele may be nearly normal for ER-to-Golgi trafficking while exhibiting the other defects and thus would be a useful tool for this analysis.
We therefore analyzed the rate of invertase secretion and glycosylation in wild type, cog3-201, and cog3-202 cells at the nonpermissive temperature by immunoprecipitating invertase from internal and external (secreted) fractions after a 10-min pulse and 0-and 30-min chase (Fig. 1). As observed previously, the cog3-201 mutant exhibited both severe glycosylation defects and a pronounced delay in secretion. The cog3-202 mutant also exhibited a severe glycosylation defect, but invertase in FIG. 1. The cog3-202 strain is highly defective for Golgi-specific invertase glycosylation while secreting invertase at a nearly normal rate. Strains SBY4-10A/pSB18/pSB29, SBY4-10A/ pSN335/pSB29, and SBY4-10A/pSN336/pSB29 were propagated at 23°C, spheroplasted, shifted to 36°C for 15 min, radioactively pulsed for 10 min, and chased for 0 and 30 min. Invertase immunoprecipitated from intracellular (I) and extracellular (E) samples was then analyzed by SDS-PAGE and autoradiography. The positions of the ER-glycosylated (core) and Golgi-modified forms (outer chain) are indicated. The percent invertase secretion at each time point as determined by Phos-phorImager analysis is indicated below the E lanes. The data shown are representative of that obtained from three independent experiments. this strain was secreted at near wild type levels at the two time points. These results are consistent with the near normal rates of trafficking of CPY and Kex2p through the early secretory pathway in cog3-202 mutants (33). Both cog3 alleles appear to exhibit a defect in elaboration of the core-glycosylated form of invertase that is dependent on Golgi-localized mannosyltransferases (43). Although glycosylation defects in cog3 mutants have been noted previously (31,33), the data in Fig. 1 rules out that a block in ER-to-Golgi transport causes the glycosylation defect in cog3-202 cells and suggests instead that a defect in the Golgi glycosylation machinery is the root cause.
Multiple Golgi Membrane Proteins Are Mislocalized in cog3-202 Mutant Cells-The dramatic glycosylation defects and relatively normal anterograde transport through the secretory pathway in cog3-202 cells suggested that Cog3p is required to maintain the activity of one or more glycosyltransferases in the Golgi apparatus. We therefore analyzed the localization in wild type and cog3-202 cells of two enzymes involved in Golgi glycosylation, ␣1,6-mannosyltransferase (Och1p) and ␣1,3-mannosyltransferase (Mnn1p). Both proteins were tagged with the HA epitope to facilitate detection. Och1-HA has been shown to exhibit localization and cycling patterns that are indistinguishable from wild type Och1p (21). Likewise, MnnI-HA exhibits localization similar to that of wild type Mnn1p (16). Localization of Och1-HA and MnnI-HA by immunofluorescence microscopy in wild type cells revealed cytoplasmic, punctate staining patterns (Fig. 2) as observed previously for Och1-HA and MnnI-HA (16,44). However, in cog3-202 cells at the nonpermissive temperature Och1-HA and MnnI-HA exhibited a fine punctate/diffuse staining pattern suggesting that these proteins localized to structures that were smaller and more numerous than in wild type cells.
Similar results were observed for A-ALP (Fig. 2), a model late Golgi membrane protein known to cycle between the Golgi and endosomal system (41). Thus a loss of Cog3p function affected the localization of multiple Golgi membrane proteins whose localization is maintained by cycles of transport either within the Golgi or between the Golgi and anterograde organelles.
In an attempt to identify the organelle to which Och1p was mislocalized, we performed cell fractionation experiments on wild type and cog3-202 cells that had been shifted from the permissive (23°C) to the nonpermissive temperature (36°C) for 80 min. We initially employed a differential centrifugation approach in which cell lysates were centrifuged at 12,300 ϫ g to FIG. 3. Subcellular fractionation of COG3 and cog3-202 cells by differential centrifugation. Strains SBY4-10A/pSB18/pOH and SBY4-10A/pSN336/pOH were propagated at 23°C for several doublings and then were shifted to 36°C for 80 min before being harvested, lysed, and subjected to centrifugation at 12,300 ϫ g to generate pellet (P12) and supernatant (S12) fractions. The S12 fraction was then centrifuged at 150,000 ϫ g to generate P150 and S150 fractions. The three fractions were run on SDS-PAGE; gels were transferred to nitrocellulose, and blots were probed using antibodies to detect the indicated proteins. The percentage of Och1-HA in the three fractions from each strain was quantified from two independent experiments, and the average is indicated below the Och1-HA blot.
generate a pellet (P12) and supernatant (S12) fraction followed by centrifugation of the S12 fraction at 150,000 ϫ g to generate P150 and S150 fractions. In wild type cells under these conditions most of the early Golgi protein Och1-HA was found in the P12 fraction (60%) with some also present in the P150 fraction (36%; Fig. 3). In contrast, the late Golgi/endosomal resident proteins Kex2p and Vps10p were predominantly found in the P150 fraction with minor amounts in the P12 fraction. The vacuolar membrane marker Vph1p and ER marker Dpm1p were mainly found in the P12 fraction. Most markers exhibited similar fractionation patterns in the cog3-202 cells except for Och1-HA which exhibited a partial shift from the P12 fraction (39%) to the P150 fraction (51%) consistent with it being mislocalized. The level of Kex2p was markedly reduced in the cog3-202 fractions (Fig. 3) consistent with mislocalization of Kex2p to the vacuole in cog3 mutants where it is rapidly degraded (33).
Sucrose density gradient fractionation was used to analyze further mislocalization of Och1-HA in cog3-202 cells. Lysates were loaded at the top of sucrose step gradients and centrifuged to equilibrium, and fractions were analyzed for the presence of Och1-HA and several organelle marker proteins (Fig. 4). Och1-HA, Kex2p, and (to a lesser degree) the early endosomal marker Tlg1p (45) shifted to membranes that were less dense in cog3-202 cells as compared with wild type cells. Vph1p and Dpm1p exhibited little or no change in cog3-202 compared with wild type. The region of the gradient that Och1-HA and Kex2p shifted to in cog3-202 cells clearly does not coincide with either vacuolar (Vph1p) or ER membranes (Dpm1p) and thus must represent some other organelle. Alternatively, this could represent a change in density of an organelle such as the late Golgi that Och1p and Kex2p normally populate as part of their cycling itinerary.
Retrograde Trafficking of Och1p to the cis-Golgi Depends on Sed5p and Sft1p-The COG complex has been proposed to act as a tethering complex at the cis-Golgi for retrograde vesicular trafficking from later regions of the Golgi and possibly the endosomal system. If this were the case, a loss of Cog3p func-tion would presumably cause Och1p to accumulate in such vesicles. Little is known of the mechanism of Och1p retrograde trafficking; however, the cis-Golgi t-SNARE Sed5p and v-SNARE Sft1p have been shown to mediate retrograde trafficking of other cargo to the cis-Golgi (22,46). In sed5-1 and sft1-15 temperature-sensitive mutants, the Golgi membrane proteins Sft2p, Anp1p, and Mnn1p rapidly disperse into vesicle-like structures. In order to test the idea that the COG complex collaborates with these SNAREs in retrograde traffic of Och1p, we first assessed whether inactivation of Sed5p and Sft1p affected cis-Golgi localization of Och1-HA.
Och1-HA localization was assessed in wild type, sed5-1, and sft1-15 cells by immunofluorescence microscopy. After shifting to the nonpermissive temperature for 30 min, the localization pattern of Och1-HA in both the sed5-1 and sft1-15 strains dramatically changed to a diffuse cytosolic pattern suggestive of vesicular localization (Fig. 5). An ER staining pattern typically appears as a ring around the nuclear DNA that can be visualized by 4Ј,6-diamidino-2-phenylindole staining. The Och1-HA pattern in the sed5-1 and sft1-15 strains was clearly distinct from an ER pattern ( Fig. 5 and data not shown). Given the known role of Sed5p and Sft1p as SNAREs involved in intra-Golgi retrograde vesicular traffic, these data suggest that Och1-HA is carried by retrograde vesicles originating from the later regions of the Golgi, and/or from compartments distal to the Golgi, that then fuse with the cis-Golgi. Och1-HA in sed5-1 and sft1-15 mutants exhibited a staining pattern similar to that of cog3-202 cells (Fig. 2), suggesting that in all three mutants Och1-HA might be accumulating in the same retrograde vesicles.
The Compartment That Och1-HA Is Mislocalized to in sed5-1 Cells Is Distinct from That of cog3-202 Cells-To test whether the COG complex functions in Och1p trafficking strictly as a cis-Golgi vesicle tether, we asked whether the Och1-HA-containing membranes in sed5-1 cells had the same density as those in cog3-202 cells. Most surprisingly, both Och1-HA and Kex2p-containing membranes had a higher density in sed5-1 cells than in wild type (Fig. 6). This is opposite of the effect Och1p Trafficking Requires the COG Complex observed in cog3-202 cells compared with wild type (Fig. 4) and indicates that Och1-HA is not accumulating in retrograde vesicles in cog3-202 cells. It should be noted that Sed5p also mediates fusion of ER-derived vesicles with the cis-Golgi (47); thus newly made Och1-HA would be expected to accumulate in ER-derived vesicles given the ϳ60-min half-life of Och1-HA (see below). Nevertheless, accumulation in retrograde vesicles would also be expected, and absolutely no shift of Och1-HA to lighter density fractions in sed5-1 cells was observed. Thus, these data are not consistent with the COG complex acting strictly as a membrane tether.
Och1p Does Not Normally Cycle to and from Endosomes-To examine further the role of Cog3p in trafficking of Och1p, we performed a series of experiments to ascertain the trafficking pathways of Och1p. Och1p is known to rapidly cycle at least as far as the late Golgi and then is transported back to the cis-Golgi (21), although the compartment that Och1p is retrieved from is not known. Membrane proteins of the late Golgi maintain their localization by cycling between the Golgi and endosomes (48 -51). To address whether Och1-HA cycles to and from an early endosomal compartment, we looked at its local-ization and trafficking in rcy1⌬ and cog8⌬ mutants. rcy1⌬ mutants exhibit defective trafficking through an early endosomal organelle leading to a defect in a post-internalization step of endocytosis and in membrane recycling from an early endosome back to the plasma membrane and/or Golgi (52). Cog8p is a member of the COG complex, but unlike cog3 and cog2 mutants, a cog8 knockout does not give a strong growth defect. In cog8⌬ mutants, as well as rcy1⌬ mutants, the v-SNARE Snc1p fails to properly cycle from the early endocytic pathway back to the Golgi (34,53) from where it is then normally transported to the plasma membrane. Och1-HA exhibits normal punctate staining in cog8⌬ and rcy1⌬ mutants that is indistinguishable from wild type cells (Fig. 7). Although Snc1p exhibits a clear plasma membrane staining pattern in wild type cells, in rcy1⌬ and cog8⌬ mutants it instead accumulates in cytoplasmic structures that are presumably early endosomes or vesicles. The structures in rcy1⌬ and cog8⌬ strains that Snc1p accumulates in do not colocalize with Och1-HA-containing compartments (Fig. 7), further supporting the idea that Och1-HA does not follow the early endosomal cycling pathway used by Snc1p. We next assessed whether the rate of turnover of Och1-HA was affected in rcy1⌬ or cog8⌬ mutants. Och1-HA was immunoprecipitated from cultures radioactively pulsed for 10 min and chased for various times. In wild type cells Och1-HA was turned over with a half-time of 61 min and was stabilized to 95 min in a pep4 prb1 prc1 triple mutant strain lacking vacuolar hydrolases (Fig. 8A) indicating that a pool of Och1-HA is turned over in the vacuole. Turnover of Och1-HA in the cog3-202 mutant was delayed presumably due to its accumulation in a low density organelle (Figs. 2 and 4). Och1-HA turnover kinetics were unchanged in rcy1⌬ (Fig. 8B) and cog8⌬ mutants (data not shown) compared with wild type, again consistent with the idea that Och1p does not visit the early endosome. We assessed whether Och1-HA is retrieved from the prevacuolar/endosomal compartment (PVC) by measuring its rate of turnover in a vps5⌬ mutant that lacks function of the retromer complex required for PVC-to-late Golgi transport (54 -56). If the retromer was required for returning Och1-HA back to the Golgi, then Och1-HA should be turned over more rapidly in a strain lacking retromer function. However, in the vps5⌬ mutant the rate of turnover was unchanged compared with wild type (Fig. 8B) strongly suggesting that Och1-HA is not retrieved from the PVC. Consistent with the above conclusions, mutations in the GARP-VFT complex that is required for fusion of endosome-derived vesicles with the late Golgi do not appear to affect localization of Och1p (Ref. 44; data not shown). Taken together, these data are most consistent with a model in which Och1p cycles between the cis-Golgi and the late Golgi in wild type cells.
Trafficking of Och1p to the Vacuole for Degradation Requires Both the CPY Pathway and the AP-3-mediated Pathway-Although it appears that Och1-HA does not maintain its Golgi localization by retromer-mediated retrieval from the PVC, Och1-HA is clearly transported to the vacuole for its degradation because it is partially stabilized in a strain lacking vacuolar hydrolases. If Och1-HA is transported from the Golgi to the vacuole via the PVC (CPY pathway), it should be stabilized by a mutation in the PVC-localized t-SNARE Pep12p. Most surprisingly, a strain carrying the pep12-49 temperature-sensitive allele at the nonpermissive temperature consistently exhibited a similar turnover rate as wild type (Fig. 9). An alternative Golgi-to-vacuole pathway that bypasses the PVC is mediated by the AP-3 adaptor complex (57)(58)(59). To assess whether Och1-HA reaches the vacuole via the AP-3 pathway, we measured its rate of turnover in an apm3⌬ strain lacking the AP-3 subunit. Again we observed no difference in the rate

FIG. 5. Loss of function of intra-Golgi trafficking SNAREs
Sed5p and Sft1p causes Och1-HA to be rapidly mislocalized to vesicle-like structures. Strains SEY6210/pOH, DB51/pOH, and DB115/pCEN-sft1-15/pSN390 were propagated at 23°C for several doublings. Cultures were either shifted to 37°C for 30 min or were left at 23°C, as indicated, before being fixed and spheroplasted. Och1-HA was then detected, and the cells were viewed as described in the legend to Fig. 2. WT, wild type; DIC, differential interference contrast. of turnover in the apm3 cells compared with wild type. However, in a pep12-49 apm3⌬ double mutant we saw a marked stabilization of Och1-HA similar to that observed in the pep4 prb1 prc1 triple mutant. Taken together, these results suggest that Och1-HA can readily utilize both the AP-3 and the CPY pathways for its degradation in the vacuole.

DISCUSSION
This is the first study to address directly the basis for the glycosylation defect in cells lacking function of the COG complex. We show that the ␣1,6-mannosyltransferase Och1p is mislocalized to lighter density membranes in the cog3-202 mutant strain at the nonpermissive temperature. Because subsequent oligosaccharide modifications rely on the addition of ␣1,6-mannose, the dramatic underglycosylation observed in cog3 mutants must be due, at least in part, to the absence of Och1p activity in the cis-Golgi. Our results complement previous studies documenting severe glycosylation defects in a number of cog mutants (31, 33, 60). In addition to Och1p, we also observed mislocalization of Mnn1p (Fig. 2) and an Mnt1-ALP fusion 2 in cog3-202 cells. Thus the severe glycosylation defect in cog3 mutants of N-glycosylated proteins is probably because of a lack of proper localization of several different glycosylation enzymes. Most interestingly, a number of cog mutants have been shown to be defective in O-glycosylation of HSP150 (31) suggesting one or more enzymes necessary for O-glycosylation are also affected. However, cells lacking Cog8p function did not appear to mislocalize Och1p (Fig. 7) consistent with a previous observation that cog5, cog6, cog7, and cog8 mutants exhibit very moderate growth and glycosylation defects compared with the other cog mutants (34). These observations are consistent with the existence of distinct COG complex "lobes" with specialized functions (34,61,62).
The idea that the COG complex functions on the cis-Golgi as a tethering factor was suggested by genetic interactions between COG3 and COG2/SEC35 and with genes encoding tethering factors (28,29). Furthermore, in vitro assays of ER-to-Golgi transport suggested a requirement for the COG complex in this process. Homology has been observed between two COG subunits (Cog3p and Cog8p) and components of the "exocyst complex" (Sec5p and Exo70), which functions as a tethering factor for fusion of secretory vesicles with the plasma membrane (34,63). Finally, a marked reduction in the rate of secretion of secretory proteins from some but not all cog mutants has been noted (30,(32)(33)(34)60) consistent with the idea that the COG complex could function as a tethering factor in anterograde trafficking between the ER and Golgi. In vitro assays showed that in the absence of Cog2p and Cog3p function there was a reduced rate of acquisition of ␣1,6-mannose residues on a cargo protein as it was transported to the cis-Golgi (28,29). Our results indicate that ␣1,6-mannosyltransferase activity is depleted from the cis-Golgi in cog3 cells, and most likely this is the case for cog2 cells also. Thus it is possible that reductions in ␣1,6-mannose addition observed in such assays could reflect defects in retrograde transport of ␣1,6-mannosyltransferase activity rather than defects in ER-to-Golgi transport. For this reason use of a read-out system not reliant on mannosyltransferases might be warranted. A more convincing case in these studies was made by in vitro vesicle diffusion assays that do not rely on ␣1,6-mannosyltransferase activity. This type of assay demonstrated a role for Cog2p and Cog3p in consumption of COPII-derived vesicles (28,29). Thus it seems clear that the COG complex is required for ER-to-Golgi transport, but the issue of whether the COG complex plays a direct role in ER-to-Golgi trafficking is still up for debate.
Our observation that the ER-to-Golgi trafficking block can be uncoupled from other Golgi-specific phenotypes in the cog3-202 mutant could be interpreted to mean that the COG complex has a direct role in both ER-to-Golgi and in retrograde transport to 2 S. F. Nothwehr, unpublished data.
FIG. 6. Och1-HA and Kex2p shift to higher density membranes upon loss of Sed5p function. Strains SEY6210/pOH and DB51/pOH were grown for several doublings at 23°C before being shifted to 37°C and grown for an additional 80 min. The cells were then analyzed by sucrose density gradient as described in the legend to Fig. 4. Data for the COG3 and cog3-202 strains are indicated by the solid and dotted lines, respectively.
FIG. 7. Och1-HA localization is unaffected in mutant strains defective for trafficking through the early endosomal system. Strains BY4743, BY4743 cog8⌬, and BY4743 rcy1⌬ each carrying plasmids pPB18 and pMyc-Snc1 were fixed and spheroplasted, and the cells were costained for Och1-HA and c-Myc tagged Snc1p as described under "Materials and Methods." The cells were viewed by epifluorescence using filters specific for each fluorochrome. The right-hand column shows an overlay of the images from the left-hand and middle column. Note the absence of any extensive colocalization between the two proteins in any of the three strains. WT, wild type. the cis-Golgi. In this view, the cog3-202 mutation would render the COG complex defective for the retrograde function while still retaining the anterograde function. Alternatively, cog3-202 could simply be a partially defective allele, and the threshold of Cog3p activity leading to the loss of the retrograde function could be lower than that leading to loss of the anterograde activity. However, the two classes of cog3 alleles that were identified were highly distinct; thus we feel that the latter possibility is unlikely.
However, a fundamentally different role for the COG complex is suggested by two recent studies (31, 60) that detect genetic and physical interactions between COG subunits and the COPI vesicle coat. These observations raise the possibility that the COG complex might be involved in sorting of cargo such as Och1p into intra-Golgi retrograde vesicles. A role for COG in COPI cargo sorting would fit with localization studies that demonstrate that although the COG complex is primarily localized to the cis-Golgi, it can be detected in other regions of the Golgi, including the late Golgi (30,33,64). In this context it is also interesting to note that Cog2p and Cog3p have been implicated in cargo sorting at the vesicle budding stage at the ER (65).
What does the mislocalization of Och1p in cog3-202 say about the role of the COG complex in trafficking of Och1p? We were surprised to find that in cog3-202 cells Och1-HA and Kex2p were mislocalized to membranes that had a lighter density than wild type, and in sed5-1 cells they were mislocalized to membranes with a higher density. Given the fact that Sed5p functions as a cis-Golgi t-SNARE for both anterograde and retrograde trafficking (22,46), Och1-HA is most likely trapped in retrograde vesicles in sed5-1 and is trapped in some other type of compartment in cog3-202 cells. An alternative possibility that Och1p is trapped in the ER in sed5-1 cells seems highly unlikely. First, immunofluorescence localization did not indicate any Och1p in the ER in sed5-1 cells (Fig. 5), and second, Och1p does not cycle between the Golgi and ER as part of its itinerary (24). If the COG complex were acting solely as a cis-Golgi tethering complex, it seems likely that Och1-HA would accumulate in the same type of vesicle in cog3-202 cells as in sed5-1 cells. Thus our data appear to contradict the idea that the COG complex acts solely as a cis-Golgi tethering complex. Instead our data are consistent with the model in which COG functions with COP1 in the loading of Och1p into intra-Golgi retrograde vesicles. A lack of such sorting would likely cause Och1p to be mislocalized to a post-Golgi compartment such as the plasma membrane, endosomal system, or vacuole. Our data indicate that in cog3-202 cells Och1p localizes to an internal compartment that is distinct from the ER and vacuole. Mislocalization of Och1p to this compartment causes slower delivery to the vacuole than in wild type cells. Some overlap in the density of Och1-HA and the early endosomal marker Tlg1p was observed suggesting that Och1-HA may be mislocalized to the endosomal system in such a manner as to not be rapidly transported to the vacuole by default. Retention in an early endosomal organelle has been shown for a Pep12p mutant lacking a sorting signal for direct trans-Golgi network-to-PVC trafficking (66). In summary, we propose that the COG complex is required for sorting of Golgi membrane proteins into vesicles that form from Golgi compartments for transport back to the early Golgi. It should be noted, however, that the two models for COG complex function may not be mutually exclusive, and it is possible that the COG complex may have both a sorting role and a tethering role at the cis-Golgi.
Our analysis of Och1p trafficking strongly suggests that Och1p does not normally reach the endosomal system in wild type cells. Mutations known to affect trafficking of cargo proteins through early and late endosomes did not alter the kinetics of Och1p degradation (Fig. 8B). Furthermore, Och1p did not FIG. 9. Och1-HA can use either the CPY pathway or the AP-3 pathway to reach the vacuole for its degradation. Strains SHY35, SNY156, SNY106, and PBY36-1B carrying pOH were propagated at 23°C for several doublings before being preshifted to 37°C for 15 min, pulsed for 10 min, and then chased for the indicated times. Analysis of Och1-HA degradation was then performed as described in the legend to Fig. 7. WT, wild type.
Och1p Trafficking Requires the COG Complex colocalize with Snc1p in mutants known to accumulate Snc1p in the early endosomal structures (Fig. 7). Och1p is known to reach a compartment containing the Kex2p endoprotease (21), a protein that is primarily localized to the late Golgi but also cycles between the late Golgi and endosomes (48). Taken together these results suggest that Och1p utilizes a cycling itinerary between the late Golgi and the cis-Golgi.
Och1p was partially stabilized in a mutant lacking vacuolar proteases indicating that Och1p is degraded by being shunted from the Golgi to the vacuole. The fact that Och1p was not completely stabilized by a loss of vacuolar protease function indicated that it uses another pathway for degradation. The vacuole-independent degradation of Och1p persisted when ERto-Golgi trafficking was blocked in a sec18-ts mutant, 3 suggesting that the ER-associated degradation pathway (67) degrades a pool of Och1p. Most surprisingly, the rate of vacuolar degradation of Och1p was unchanged when trafficking to the vacuole via the PVC was blocked by a pep12-ts mutation and also when the AP-3 pathway was blocked using an apm3⌬ mutation. However, when these two mutations were combined, the rate of degradation was slowed to a rate comparable with that of a strain lacking vacuolar proteases. The effect was observed shortly after shifting the pep12-ts apm3⌬ to the nonpermissive temperature, suggesting that the effect is directly due to blocking the two known routes of vacuolar delivery rather than an indirect effect. We conclude that Och1p is fully capable of using either pathway.
How is Och1p diverted from its normal intra-Golgi cycling pathway into a late Golgi-to-vacuole pathway? The two proteins known to use the AP-3 pathway, ALP and Vam3p, contain dileucine-like signals in their cytosolic domains that appear to mediate entry into this pathway (68,69). Most interestingly, the 16-amino acid cytosolic domain of Och1p contains a dileucine-like sequence (Leu-Ile), but mutation of these residues to alanines did not affect the rate of degradation in a pep12-ts strain, 3 indicating that they are not required for entry into the AP-3 pathway. There is evidence that sorting of proteins from the Golgi into the CPY pathway can be mediated by attachment of ubiquitin by the late Golgi-localized ubiquitin ligase Tul1p as part of a quality control mechanism (70). Thus it is possible that misfolded Och1p could be recognized in the late Golgi by such a quality control mechanism that would then initiate the delivery of Och1-HA into the late Golgi. Alternatively, entry of Och1-HA into these pathways could be a random process reflecting a less-than-perfect late Golgi-to-cis-Golgi retrieval process. The idea that Och1p slowly "leaks" into the CPY and AP-3 pathways, rather than being sorted by a receptor, would be consistent with an apparent lack of a preference for one pathway over the other. Further experiments will be needed to determine whether vacuolar degradation of Och1p reflects inefficient retrieval from the late Golgi or a quality control process.