Characterization of a mammalian Golgi-localized protein complex, COG, that is required for normal Golgi morphology and function

The eukaryotic secretory pathway consists of a set of membrane-bound compartments through which proteins move in transport vesicles or larger transport intermediates (Rothman, 1994). Vesicular transport from one compartment to the next can be divided into four sequential phases: vesicle budding; movement to the next compartment; association of the vesicle with the acceptor compartment; and fusion of vesicle and acceptor membranes (Rothman, 1994; Mellman and Warren, 2000). For the Golgi apparatus, the transport structures may be the Golgi cisternae themselves, as some secretory cargo moves through the Golgi without exiting the stack in vesicles (Bonfanti et al., 1998; Mironov et al., 2001).

Many Golgi proteins, such as glycosylation enzymes, are maintained in a nonuniform steady-state distribution through the cisternae. These proteins must be transported to their sites of residence and then either be retained there or, if they move beyond them, retrieved by retrograde vesicular transport (Harris and Waters, 1996; Martinez-Menarguez et al., 2001). Thus, all transport steps appear to involve the movement of small vesicles (either carrying cargo in the anterograde or resident enzymes in the retrograde direction) between larger membrane compartments (Pelham and Rothman, 2000).

Studies in yeast and mammalian cells have led to the identification of several multisubunit peripheral membrane protein complexes that are thought to be involved in membrane trafficking and/or compartment function, including the Sec6/8 (Grindstaff et al., 1998; Guo et al., 1999), the TRAPP I (Sacher et al., 2001), and the Sec34/35 (VanRheenen et al., 1998) complexes. The yeast Sec34/35 complex has eight subunits (Whyte and Munro, 2001), is localized to the Golgi apparatus (Spelbrink and Nothwehr, 1999; Kim et al., 2001), and has been suggested to be involved in anterograde ER to Golgi (VanRheenen et al., 1998; Kim et al., 2001), retrograde intra-Golgi (Ram et al., 2002), and endosome to Golgi (Spelbrink and Nothwehr, 1999) traffic. A mammalian Sec34-containing complex has also been identified and implicated in Golgi protein trafficking (Suvorova et al., 2001).

Another complex, called the Golgi transport complex (GTC),^* was isolated because of its ability to stimulate an in vitro intra-Golgi transport assay (Walter et al., 1998). It is a hetero-oligomeric Golgi-localized complex of ∼800 kD, comprising proteins between 70 and 110 kD. The 90-kD subunit of GTC, GTC-90, was identified and a cDNA encoding the protein was cloned.

The ldlCp complex was discovered during the analysis of low-density lipoprotein receptor (LDLR)–defective CHO cell mutants, ldlB and ldlC (Krieger et al., 1981; Podos et al., 1994; Chatterton et al., 1999). The ldlB- and ldlC-null mutants, which are viable, exhibit defects in multiple Golgi-associated reactions that result in the abnormal processing of the LDLR and many other glycoconjugates (Kingsley et al., 1986). The defects are not due to drastic disruptions in secretion or endocytosis (Kingsley et al., 1986; Reddy and Krieger, 1989). The ldlCp protein (∼80 kD) is a component of a large peripheral Golgi complex whose size and Golgi-association are dependent on ldlBp (∼110 kD) (Podos et al., 1994; Chatterton et al., 1999), which suggested that they are components of the same complex.

Similarities in the sizes, locations, and subunits of the ldlCp complex and GTC suggested that they might be identical (Chatterton et al., 1999). We have explored this possibility through identification of all the subunits of purified GTC and examination of the physical association of the proteins in tissues and wild-type and mutant cultured cells. We have found that GTC and the ldlCp complexes are the same, and it appears to be to the mammalian homologue of the yeast Sec34/35 complex. This is consistent with the report of Whyte and Munro (2001)(which appeared during the preparation of this manuscript), which showed that the yeast homologue of GTC-90, Cod4p, is a subunit of the Sec34/35 complex. We suggest that this mammalian complex be called conserved oligomeric Golgi (COG) complex, and its subunits be designated Cog1–8. We have also examined by EM the structure of the COG complex and the effects of the null mutations in ldlB (Cog1) and ldlC (Cog2) on Golgi structure.

We modified the previously described (Walter et al., 1998) method of COG purification from bovine brain to improve the yield (Table I) and followed the purification by anti-Cog5 immunoblotting. During the purification, two peaks of Cog5 were resolved by hydroxyapatite chromatography (Fig. 1 A). Similar complexes were also seen in CHO cell lysates not subjected to the multiple purification steps (see Fig. 7 B). We purified the protein eluting in the second peak because it represented the larger complex (Fig. 1 B), here defined as COG. It is not clear if both complexes are present in vivo or if the small complex was generated during cell/tissue lysis, or if the peak 1 material, which is slightly smaller in size than COG (Fig. 1 B), shares COG components other than Cog5.

SDS-PAGE analysis of the fractions from the final purification step (Table I and Fig. 2) indicated that COG is composed of eight subunits. If one copy of each subunit were present in the complex, the calculated molecular weight would be 700 kD.

Purified COG subunits were excised from the SDS-PAGE gel individually (Cog1, -2, -6, and -8) or in pairs (Cog3/Cog5 and Cog4/Cog7) and sequenced by mass spectrometry. Eight protein components were identified (Table II and Fig. 2 B). All except Cog7 (see below) had been previously suggested to be components of various mammalian Golgi-associated complexes. The four individually excised proteins were identified: Cog1 as ldlBp; Cog2 as ldlCp; and Cog6 and -8 as the homologues of yeast Cod2p and Dor1p, respectively. One pair, Cog3/Cog5, was identified as the mammalian Sec34 and GTC-90; the other, Cog4/Cog7, was found to be the homologue of yeast Cod1p (Cog4) and a previously unidentified protein (Table II and Fig. 2 B). The individual bands in the Cog3/Cog5 doublet were identified as Sec34 and GTC-90 based on immunoblotting (unpublished data). The assignments of the bands in the Cog4/Cog7 doublet were based on the predicted molecular weights of their human homologues and will require verification. The presence of Cog1, -2, and -3 in addition to Cog5 in the purified material was confirmed by immunoblotting partially (Fig. 3) and completely (Fig. 2 A) purified fractions.

Although present in the COG complex, the Cog1, -2, and -3 subunits were not detected by immunoblotting in the smaller Cog5-containing complex (Fig. 3, fractions 30–32).

It is noteworthy that during the early stages of development of this new purification protocol, we did not include the hydroxyapatite purification step (unpublished data). This resulted in the copurification of COG and the Sec6/8 complex (Hsu et al., 1996), which suggests that they share many physicochemical characteristics. This may not have been fortuitous, as it has been proposed that the COG and the Sec6/8 complexes are related (Whyte and Munro, 2001).

The human Cog7 cDNA (GenBank/EMBL/DDBJ accession no. XM_041725) encodes a protein of 770 residues. Analysis of the human genome database (http://www.ncbi.nlm.nih.gov/genome/guide/human/) shows that the Cog7 gene is located on the short arm of chromosome 16 and contains 17 exons. The Paircoil algorithm (Berger et al., 1995) detected a potential coiled-coil domain between amino acids 70 and 100 of Cog7, however, the structure remains to be determined. Closely related homologues of Cog7 can be found in numerous vertebrates and invertebrates and several plants (Table III). However, homologues in Caenorhabditis elegans or Saccharomyces cerevisiae were not found. Thus, Cog7 joins Cog1 and Cog2 in having no obvious homologue in yeast. The functions of Cog1, -2, and -7 may be performed by other proteins in lower eukaryotes (see Discussion).

We used immunofluorescence microscopy to locate endogenous Cog1 in CHO cells and Cog7 in HeLa cells transfected with an expression vector encoding a hemagglutinin (HA) epitope–tagged Cog7. Fig. 4 A shows that both Cog1 (top) and Cog7 (bottom) colocalized significantly, but not completely, with the Golgi marker mannosidase II in a perinuclear distribution. Similar results have been observed for Cog2 and Cog5 (Podos et al., 1994; Walter et al., 1998). Thus, like other COG complex components, Cog1 and Cog7 are Golgi associated.

Fig. 4 B shows double staining experiments in HeLa cells using antibodies to HA–Cog7 and Cog1, -3, and -5. These experiments show a high degree of colocalization of these four proteins, providing further evidence that they are components of a single complex.

Copurification of proteins provides strong, but not unequivocal, evidence that the proteins are members of a complex. Therefore, we performed coimmunoprecipitation experiments. COG from rat liver cytosol was immunoprecipitated (Fig. 5 A) using an anti-Cog2 antipeptide antibody (Podos et al., 1994) in the absence (lane 2) or presence (lane 3) of the immunogenic peptide to block the precipitation and subjected to SDS-PAGE and immunoblotting with antibodies to COG subunits. All of the COG subunits examined coprecipitated with Cog2 (lane 2) in an anti-Cog2 antibody-specific (compare lanes 1 and 2) and immunogenic peptide–inhibitable (lane 3) fashion. In each precipitation, ∼10% or more of the total cytosolic content of each component (Fig. 5 A, lane 4) was recovered (lane 2). Similar results were observed using an anti-Cog1 monoclonal antibody to precipitate the complex from a partially purified sample of bovine brain cytosol (Fig. 5 B). As expected, the anti-Cog1 antibody did not precipitate Sec8, a component of the Sec6/8 complex (Hsu et al., 1996) (Fig. 5 B). Thus Cog1, -2, -3, and -5 are present in the same complex, and the GTC, ldlCp, and Sec34 complexes are the same entity.

The ultrastructure of purified COG was visualized by quick freeze/deep etch/rotary shadow EM (Heuser, 1983). The images in Fig. 6 are 3-D “anaglyphs,” which are best viewed with red/green stereo glasses (Heuser, 2000). Samples that were prefixed with glutaraldehyde (Fig. 6, fixed) were compared with unfixed COG (Fig. 6, unfixed); glutaraldehyde fixation was used to preserve structures that might disassemble on contact with mica (Heuser, 1989). However, images of fixed samples must be interpreted with caution, as fixation can introduce structural artifacts due to cross-linking. Most of the images of the fixed samples and many of the unfixed samples suggest that COG is bilobed. In some images, the lobes appeared to be interconnected by thin rods and/or globules. The fixed form was 37 ± 4 (n = 18) nm long. Thus, COG appears to be slightly larger than the size marker thyroglobulin (670 kD; Fig. 6, leftmost in bottom row). The bilobed appearance of COG looks substantially different than comparable images of the mammalian Sec6/8 complex (Fig. 6, first four images in fifth row). The Sec6/8 complex appears to have a large globular or brick-shaped “core” that is surrounded by short extensions (Hsu et al., 1998). Other molecules suitable for comparison in size and shape with COG are shown in the bottom panels of Fig. 6 and are described in its legend.

About a third of the images of unfixed COG (Fig. 6, unfixed) appeared similar to the fixed samples, whereas the others appeared somewhat heterogeneous and “splayed.” The splayed images tended to display two globules connected by disordered globular or rod-like connections. The splayed forms ranged up to 50–75 nm long. The overall size of this form was similar to that of a clathrin triskelion (Fig. 6, fourth image in row 5; Heuser and Kirchhausen, 1985). Presumably, such splayed forms represent COG complexes partially disassembled by adhesion with mica. Nevertheless, they illustrate that native COG seems to be a group of globular domains interconnected by flexible arms. Additional studies will be required to determine if the native COG within cells is a tightly compacted, bilobed, or a relatively splayed structure.

The Cog1 and Cog2 proteins were initially identified during the analysis of two classes of CHO cell mutants, ldlB and ldlC (Krieger et al., 1981; Kingsley et al., 1986; Podos et al., 1994; Chatterton et al., 1999). We used immunoblotting of whole cell lysates to determine the steady-state levels of COG subunits in wild-type and mutant cells, as well as stable transfectants in which the mutants' phenotypes were corrected by transfection with expression vectors encoding either Cog1 (ldlB[COG1]) or Cog2 (ldlC[COG2]). Fig. 7 A shows that, as previously described (Podos et al., 1994), Cog2 was not detected in the ldlC mutant (lane 4), whereas a somewhat greater than wild-type expression level of Cog2 was observed in ldlC[COG2] cells. Similar results were obtained for Cog1 expression in wild-type (lane 1), ldlB (lane 2), and ldlB[COG1] cells (lane 3), as expected from prior studies (Chatterton et al., 1999). In ldlB cells, the level of Cog5 was significantly lower (lane 2) than in the wild-type (lane 1) or transfectant (lane 3) controls, but the levels of Cog3 and Cog2 were not altered. In contrast, the levels in ldlC cells of Cog1, -3, and -5 were dramatically reduced (lane 4) compared with wild-type (lane 1) or transfected (lane 5) controls. The residual level of Cog5 in ldlC cells (lane 4) was similar to that in ldlB cells (lane 2). The absence of Cog1 (lane 2) or Cog2 (lane 4) did not alter the levels of components of two other complexes involved in secretion, β-COP and Sec8 (Fig. 7 A). Thus, the absence of some COG subunits affected the steady-state levels of other COG components, providing genetic evidence for their interactions. The different effects of the absence of Cog1 and Cog2 on the levels of other COG subunits suggest that Cog1 and Cog2 play distinct roles in the organization of the complex.

The absence of one or more subunits in a macromolecular assembly, such as COG, should alter the size of the residual complex. We size fractionated wild-type (Fig. 7 B, top), ldlB (middle), and ldlC (bottom) cytosols and examined the distribution of the Cog1, -2, -3, and -5 subunits by immunoblotting. In wild-type cytosol, Cog1, -2, -3, and -5 comigrated (Fig. 7 B, fractions 21 and 22), and we detected a distinct, smaller complex containing Cog5 but not the other immunodetectable COG components (fractions 23–26, compare with Fig. 3, peak 1). In ldlB (Cog1) cytosol, the other COG subunits were no longer present in the large complex (Fig. 7 B, fractions 21 and 22; Chatterton et al., 1999). Rather, their mobilities showed, in addition to the Cog5 present in the smaller complex (fractions 23–26), that they were components of an even smaller entity whose size was much larger than that expected for the individual subunits. In the ldlC (Cog2) mutant, where the steady-state levels of total cellular Cog3 and Cog5 were significantly reduced (Fig. 7 A), these subunits were no longer found in the large complex (Fig. 7 B, fractions 21 and 22). These data support the conclusion that Cog1, -2, -3, and -5 are components of one large complex. They also show that in CHO cells, as was the case for bovine brain cytosol, some Cog5 was present in a distinct complex that was smaller than COG and did not contain Cog1, -2, and -3. The chromatographic mobility of this smaller Cog5-containing complex was apparently unperturbed by the absence of Cog1 or Cog2.

The disruptions in the size and composition of COG in ldlB and ldlC mutants (Fig. 7) are associated with pleiotropic Golgi functional defects (Kingsley et al., 1986; Podos et al., 1994; Chatterton et al., 1999). Therefore, we examined, by transmission EM, the morphology of the Golgi apparatus in wild-type CHO, ldlB, and ldlC cells and their corresponding phenotypically corrected transfectants (ldlB [COG1] and ldlC[COG2]). Fig. 8 shows that the Golgi morphology in both ldlB and ldlC mutants was abnormal. At least half, and sometimes all, of the identifiable cisternae were dilated in the mutants. These dilated cisternae were not seen in control wild-type cells or transfectants, suggesting that they were related to the functional defects in ldlB and ldlC mutants. Additional studies will be required to determine if the dilated cisternae are located predominantly on the cis or trans sides of the Golgi stack.

We characterized the composition, structure, and function of a mammalian conserved oligomeric Golgi-localized protein complex, the COG complex. Purification of COG and identification of its subunits, Cog1–8, showed that some of its polypeptides are known components of the GTC, the ldlCp complex, and the Sec34 complex (Table II). Coimmunoprecipitation experiments showed that COG components (Cog1, -2, -3, and -5) are physically associated. Furthermore, the steady-state levels of some COG subunits (Cog1, -2, -3, and -5) were dramatically reduced in ldlB (Cog1) or ldlC (Cog2) mutants, establishing genetic associations between them. The loss of Cog1 or Cog2 in the mutants also led to a reduction in the sizes of the residual complexes. In addition, seven COG subunits (Cog1–5, -7, and -8) have been localized to the mammalian Golgi apparatus (Podos et al., 1994; Walter et al., 1998; Suvorova et al., 2001; Whyte and Munro, 2001; this study). Finally, independent analyses of the yeast Sec34/35 complex (Kim et al., 1999, 2001; VanRheenen et al., 1999; Whyte and Munro, 2001; Ram et al., 2002) show that it too is a hetero-octameric complex, and five of its subunits are close homologues of COG subunits (Cog3–6 and -8) (Table II). The remaining three subunits in the COG and Sec34/35 complex are also likely to be related functionally. Indeed, yeast Sec35p and Cog2 have limited sequence similarity (Whyte and Munro, 2001).

Quick freeze/deep etch/rotary shadow EM suggests that COG, at least after fixation with glutaraldehyde, is ∼37 nm long and divided into two almost equally sized lobes (designated A and B, see below). In unfixed specimens, we observed similar bilobed structures and, more frequently, splayed structures with multiple small globular domains connected by relatively flexible extensions (Fig. 6).

Insight into the relationship between the bilobed structure of COG and its subunit composition can be deduced from our studies in combination with studies of the yeast Sec34/35 complex (Fig. 9). During the COG purification, we noted that Cog5 was present both in COG (Fig. 1, peak 2) and in a smaller complex (Fig. 1, peak 1). The smaller complex may represent a fragment of the COG complex (Fig. 9, lobe B; see below), contain COG subunits in addition to Cog5 (see below), and thus correspond to one of the two lobes seen in the EM images. Because Cog1, -2, and -3 are not part of the Cog5-containing smaller complex, they presumably reside in the other lobe.

Division of COG into two approximately equal-sized lobes, each having four subunits, is consistent with the classification by Whyte and Munro (2001) of the yeast Sec34/35 complex's subunits into two groups of four, based on the severity of their mutant phenotypes. The first group, characterized by more severe phenotypes, contains Sec34p, Sec35p, Cod1p, and Cod3p, whereas the other group, distinguished by weaker phenotypes, comprises Dor1p, Cod2p, Cod4p, and Cod5p (Whyte and Munro, 2001). We suggest that the stronger and weaker phenotypes are the result of mutations in subunits residing in distinct lobes of the complex; mutations in lobe A produce stronger phenotypes (e.g., sec34), and those in lobe B (e.g., cod4) give weaker ones (Fig. 9).

With this assumption we can assign other subunits to the A and B lobes of COG. The weak phenotype of the cod4 mutant places Cog5/Cod4p in lobe B and consequently Cog1, -2, and -3, which are in the other lobe, in lobe A (Fig. 9). Moreover, we propose that the fourth protein in lobe A is Cog4/Cod1p because of the severe cod1 phenotype (Kim et al., 2001; Whyte and Munro, 2001; Ram et al., 2002). Finally, the remaining proteins, Cog6–8, must reside in lobe B (Fig. 9), consistent with the weaker phenotypes of cod2 and dor1 mutants (Whyte and Munro, 2001), the yeast counterparts of Cog6 and Cog8. The proposed composition of lobe B is consistent with it potentially representing the smaller Cog5-containing complex in bovine brain and CHO cell cytosols.

As a consequence of these assignments (Fig. 9 and Table IV), we suggest that Cog7 in lobe B is functionally analogous to yeast Cod5p. Lastly, we speculate that (a) Cog2 and yeast Sec35p are functional analogs because of their weak sequence similarities and assignments to the A lobes, and (b) the remaining unpaired subunits, Cog1 and yeast Cod3p in lobe A, are probably functional analogs. (Fig. 9 and Table IV). Supporting this, mammalian Cog2 and Cog3, or the corresponding yeast Sec35p and Sec34p, comigrate at a smaller size when Cog1/Cod3p mutant cytosol is gel filtered (Fig. 7; Ram et al., 2002). Further work will be required to rigorously test this model for COG.

Cog1 and Cog2 were originally identified in a genetic screen for mutants that block LDLR activity in CHO cells (Krieger et al., 1981). Analysis of the transport/secretion of the LDLR and vesicular stomatitis virus (VSV) in ldlB and ldlC mutants and in a revertant of ldlC cells bearing an extragenic suppressor of the LDLR-deficient phenotype (Reddy and Krieger, 1989) strongly suggested that neither membrane protein transport through the secretory pathway nor normal endocytic cycling are disrupted profoundly. The pleiotropic defects in Golgi–associated glycosylation reactions in these mutants resulted in abnormally glycosylated, and consequently unstable, LDLRs. The ldlB and ldlC cells have essentially identical defects in numerous medial- and trans-Golgi–associated reactions that affect virtually all protein- (N-linked and O-linked) and lipid-linked glycoconjugates (Kingsley et al., 1986). The pleiotropic nature of the defects suggested that the mutations might affect the regulation, compartmentalization, transport, or activity of several different Golgi enzymes and/or enzyme substrates, or the internal milieu of the Golgi apparatus. This correlates well with our finding in this study that Golgi cisternae are dilated in these mutants.

Another COG component, Cog5, was first identified (Walter et al., 1998) in an in vitro assay that measures glycosylation of a cargo protein as a consequence of a membrane fusion event (Balch et al., 1984). The assay is dependent on maintenance of an intraluminal environment that is compatible with glycosylation (Hiebsch and Wattenberg, 1992) and on vesicular transport factors, especially those involved in membrane docking or fusion (Block et al., 1988; Clary and Rothman, 1990; Waters et al., 1992; Legesse-Miller et al., 1998). Although this system was originally thought to measure anterograde vesicle traffic through the Golgi apparatus (Balch et al., 1984), it has been suggested (Love et al., 1998) that it may in fact measure retrograde traffic of the glycosyltransferase to the cargo and, as such, recapitulate retrograde intra-Golgi traffic. The fact that COG is active in this in vitro system suggests that it may act in docking and/or fusion of vesicles within the Golgi apparatus, possibly including retrograde glycosyltransferase-bearing vesicles.

Further insight into COG function can be derived from studies of its yeast homologue, the Sec34/35 complex. sec34 and sec35 exhibit slow growth, missorting of proteins in the secretory/vacuolar system, defective protein secretion, underglycosylation of proteins, accumulation of vesicles, and dilated membranous compartments (Wuestehube et al., 1996; Kim et al., 2001). Mutations in Sec34/35 complex subunits genetically interact with mutations in components involved in multiple steps of transport to and through the Golgi apparatus (VanRheenen et al., 1998; Kim et al., 1999, 2001; Spelbrink and Nothwehr, 1999; VanRheenen et al., 1999; Whyte and Munro, 2001; Ram et al., 2002). Of particular note is the recent observation that Sec34/35 complex components, which are located throughout the Golgi apparatus (Kim et al., 2001), show strong genetic interactions with subunits of the COPI vesicle coat (Kim et al., 2001; Ram et al., 2002), which is thought to function in retrograde traffic within the Golgi apparatus and from the Golgi apparatus to the ER (Letourneur et al., 1994). This prompted Ram et al. (2002) to suggest that the Sec34/35 complex may function in retrograde intra-Golgi traffic, a possibility we have independently considered for mammalian COG. Indeed, it is known that maintenance of Golgi enzymes requires their retrograde vesicular transport from distal compartments (Harris and Waters, 1996; Martinez-Menarguez et al., 2001). Thus, a defect in intra-Golgi retrograde traffic could lead to the glycosylation defects seen in mammalian and yeast COG mutants. Compromised COG activity might also contribute to an aberrant Golgi structure through mislocalization of crucial Golgi proteins. In addition, the compromised biosynthetic trafficking seen in some yeast Sec34/35 complex mutants (Wuestehube et al., 1996; Kim et al., 2001) may be a consequence of secondary anterograde trafficking defects arising from primary retrograde trafficking defects (Gaynor and Emr, 1997; Reilly et al., 2001).

Whyte and Munro (2001) suggested that the yeast Sec34/35 complex may be evolutionarily and functionally related to the Sec6/8 complex, also called the exocyst, which has been proposed to tether transport vesicles to the plasma membrane (Grindstaff et al., 1998; Guo et al., 1999). We have found that the mammalian COG and Sec6/8 complexes comigrate during numerous purification steps. Thus, they share many biochemical and biophysical properties, including hydrodynamic size and shape and overall surface charge/hydrophobicity (this study; unpublished data). Also, as noted previously (Walter et al., 1998), both complexes contain eight subunits with very similar molecular weights. However, their appearances in deep etch EM differ. Therefore, further work to elucidate the relationship of these complexes is required.

In summary, mammalian COG and its yeast homologue clearly play important roles in determining the structure and function of the Golgi apparatus and can influence intracellular membrane trafficking. However, their precise mechanism of action remains unknown. This is highlighted by the fact that the yeast Sec34/35 complex's subunits exhibit numerous genetic interactions both with cellular components proposed to be involved in tethering vesicles to the early Golgi apparatus (VanRheenen et al., 1998, 1999; Kim et al., 1999, 2001; Ram et al., 2002) and with the COPI vesicle coat (Kim et al., 2001; Ram et al., 2002). Genetic and biochemical studies raise the possibility that the COG and Sec34/35 complexes may be directly involved in vesicular membrane transport processes (e.g., vesicle budding, targeting, or fusion reactions). Alternatively, COG may indirectly influence trafficking as a consequence of a direct influence on the structure or activity of one or more compartments of the Golgi apparatus (e.g., potential scaffolding activity).

The characterization of the composition and structure of COG combined with the analysis of COG mutants provides a strong foundation for future studies of the function of this important molecular machine.

Antibodies used for immunoblotting (IB) and for immunofluorescence (IF) and their dilutions were as follows: rabbit polyclonal affinity-purified anti-Cog5 (anti–GTC-90) (IB, 1:500–1,000; IF, 1:100; Walter et al., 1998), rabbit polyclonal affinity-purified anti-Cog2 Cpep (anti-ldlCp Cpep) (IB, 1:2,000–3,000; Podos et al., 1994), rabbit polyclonal anti-Cog3 serum (anti-hSec34) (IB, 1:10,000; Suvorova et al., 2001), affinity-purified rabbit anti-Cog3 (anti-hSec34) (IB, 1:3,000; IF, 1:3,000; Suvorova et al., 2001), mouse monoclonal anti-Cog1 (anti-ldlBp) (IB, 1:250–600; IF, 1:100; BD Biosciences), rabbit polyclonal anti-Cog1 (anti-ldlBp) antiserum (IB, 1:3,000), affinity-purified rabbit polyclonal anti-Cog1 (anti-ldlBp) (IF, 1:150), mouse monoclonal anti-Sec8 (IB, 1:2,500; BD Biosciences), mouse monoclonal anti–β-COP (IB, 1:1,000; Sigma-Aldrich), rabbit anti–mannosidase II (IF, 1:2,000; a gift from Kelley Moremen, University of Georgia, Athens, GA), and mouse monoclonal anti–HA-11 (IF, 1:400; Covance).

Dialysis membranes (mol wt cutoff of 12–14 kD) were from BioDesign Inc. The ceramic hydroxyapatite column was from Bio-Rad Laboratories, and all other chromatography materials and secondary antibodies conjugated with HRP for immunoblotting and chemiluminescence were obtained from Amersham Pharmacia Biotech. Compactin was a gift from A. Endo (Tokyo Nodo University, Tokyo, Japan). Newborn calf lipoprotein-deficient serum (NCLPDS) and LDL were prepared as previously described (Krieger, 1983). Glutaraldehyde, osmium tetroxide, uranyl acetate, and eponate were obtained from Ted Pella Inc. All other chemicals were purchased from Sigma-Aldrich, unless otherwise noted.

All procedures were performed at 4°C. Solutions were buffered with 25 mM TrisCl, pH 8.0, 1 mM DTT (TD) and contained the indicated amounts of KCl (e.g., TD + 1 M KCl) and glycerol, unless otherwise noted. Bovine brain cytosol was ammonium sulfate precipitated as previously described (Waters et al., 1992). After resuspension of the precipitated proteins and adjustment of the salt concentration to 1M KCl, the sample was chromatographed on butyl-Sepharose. The COG complex was eluted with TD + 10 mM KCl, concentrated, and the concentration of KCl in the TD buffer was adjusted to 160 mM KCl by dialysis and the addition of salt. The sample was then applied to a MonoQ HR10/10 column, COG eluted in a KCl gradient and dialyzed overnight against 25 mM Hepes-KOH, pH 7.0, 100 mM KCl, 10% glycerol. After adding potassium phosphate (KP_i) to a final concentration of 5 mM, the dialyzed sample was chromatographed on a ceramic hydroxyapatite column. COG complex was eluted with a KP_i gradient and fractions were immediately dialyzed against TD + 100 mM KCl, 10% glycerol in a microdialyzer (GIBCO BRL). The second peak of Cog5 immunoreactivity (Fig. 1 A, peak 2) was then concentrated and loaded onto a Superose 6 HR10/30 sizing column. Fractions containing COG complex were pooled and chromatographed on a MiniQ column. An estimated 70 μg of COG complex eluted at ∼220 mM KCl in the gradient.

Analytical gel filtration of the two different Cog5-containing complexes (Fig. 1 B) was performed on a Superose 6 3.2/30 column.

Gel filtration chromatography of COG complex in cell lysates from CHO cell lines (Fig. 7 B) was performed as previously described (Chatterton et al., 1999).

COG complex (∼40 μg) was separated on a 1-mm-thick 7.5% SDS-polyacrylamide gel. The gel was stained with colloidal Coomassie blue (Invitrogen) and destained in water. Four individual bands and two doublets were excised and subjected to electrospray MS/MS peptide sequencing after in-gel trypsin digestion and HPLC separation (Yates et al., 1999) at the Harvard Microchemistry Facility.

To visualize single COG complex particles, ∼20 μg of purified COG complex (Fig. 2 A, fraction 18) was visualized on mica by quick freeze/deep etch/rotary shadowing microscopy as previously described (Heuser, 1983). For glutaraldehyde fixation, the 10 μg/ml COG solution was mixed with a small volume of 7% glutaraldehyde and incubated for 30 s before mixing with a suspension of mica flakes. Final anaglyph 3-D images were generated as described by Heuser (2000).

Cells were rapidly rinsed with PBS containing 1 mM MgCl₂ and 0.1 mM CaCl₂, fixed with 2% glutaraldehyde in 0.1 M sodium cacodylate-HCl, pH 7.2, for 60 min at 4°C, and postfixed with 2% osmium tetraoxide in 0.1 M sodium cacodylate-HCl for 60 min at 4°C. The cells were harvested by scraping from the plastic dish and centrifugation at 750 g for 5 min to form a small pellet. The pellet was then dehydrated in graded alcohol up to 70% and en block stained with 70% alcoholic uranyl acetate for 30 min at 4°C. The samples were then processed for epon embedding (Vasile et al., 1983). Thin epon sections were poststained with uranyl acetate and Reynold's lead citrate and viewed with a Philips 400 electron microscope operated at 80 kV.

For immunoprecipitation, 5 μg of monoclonal anti-Cog1 antibody was added to 50 μl of partially purified bovine brain COG complex (purified through the MonoQ step). For immunoprecipitation with affinity-purified anti-Cog2 antibody, 5 μg of antibody with or without antigenic peptide (see below) was added to 400 μg of rat liver cytosol. The samples were adjusted to a total of 500 μl with PBS with (for anti-Cog1) or without (for anti-Cog2) 5% (wt/vol) nonfat dried milk and incubated overnight at 4°C. Subsequently, the sample was incubated for 2 h at room temperature with 10 μl of a 50% (vol/vol) slurry of protein A–Sepharose CL-4B in PBS. The beads were then washed four times with 500 μl PBS with (for anti-Cog1) or without (for anti-Cog2) 5% (wt/vol) nonfat dried milk and once with 500 μl of PBS. Precipitated proteins were eluted in SDS sample buffer and analyzed by immunoblotting. For anti-Cog2 antigenic peptide competition, the anti-Cog2 antibody was preincubated for 1 h at room temperature with 10 μg of the COOH-terminal antigenic peptide (Cpep; Podos et al., 1994) in 100 μl of PBS before addition to the cytosol.

Full length Cog7 cDNA was cloned into the pHM6 vector (Roche) with the restriction enzymes HindIII and KpnI, resulting in the plasmid pHM6-Cog7 coding for an HA epitope–tagged Cog7.

Wild-type CHO, ldlB, ldlC, ldlB[COG1], and ldlC[COG2] cells were grown at 37°C in plastic culture dishes as previously described (Podos et al., 1994; Chatterton et al., 1999). Here we have renamed two stably transfected cell lines to conform to the nomenclature introduced in this paper (ldlBp is Cog1; ldlCp is Cog2) as follows: ldlB[LDLB] cells (Chatterton et al., 1999) are now called ldlB[COG1] cells and ldlC[LDLC] cells (Podos et al., 1994) are called ldlC[COG2] cells. HeLa cells for immunofluorescence were grown in standard medium on coverslips and transfected with the calcium phosphate method (Ausubel et al., 1995).

Cells grown on coverslips were processed at room temperature as follows. Cells were washed once with PBS and fixed by rinsing and incubating for 25 min with 2% paraformaldehyde (EM Sciences) in 0.1 M NaP_i, pH 7.2. The coverslips were then rinsed twice and incubated for 15 min in 0.1 M glycine in PBS. Cells were then blocked in P-B (0.1% saponin, 1% BSA, 2% normal goat serum [Chemicon]) for 30 min. The coverslips were washed with PBS followed by a 5-min treatment with 6 M urea in PBS, four 5-min washes in PBS, and an additional 45-min incubation in P-B. The treatment with urea was omitted in the case of double staining with anti–mannosidase II and anti-HA antibodies. The cells were then incubated for 1 h with primary antibodies and washed with P-B four times for 5 min each. Secondary antibodies (Alexa^®488 goat anti–rabbit IgG [H+L] conjugate and Alexa^®546 or Alexa^®568 goat anti–mouse IgG [H+L] conjugate [Molecular Probes]) diluted 1:750 in P-B were applied for 1 h, and the coverslips were washed with PBS eight times for 5 min and with water for 5 min. Images were obtained with a ZEISS LSM 510 laser confocal microscope.

For a more detailed description of several methods and the generation of the anti-Cog1 and anti-Cog2 polyclonal antibodies, see the supplemental materials located.

We are grateful to Sean Munro (Medical Research Council, Cambridge, UK), Rachna Ram (MIT), and Chris Kaiser (MIT) for communicating unpublished results and cooperating in naming the complex, Sandy Schmid (Scripps Research Institute, La Jolla, CA) for encouragement and support, and Sarah Richman (Princeton University) for the first hint that Sec34/35 and COG are homologous. We thank the Harvard Microchemistry Facility for expert mass spectrometry and Joe Goodhouse (Princeton University) and H.R. Horvitz (MIT) for help with confocal fluorescence microscopy. We are grateful to Kelley Moremen and Juan Bonifacino (National Institutes of Health, Bethesda, MD) for providing antibodies.

This work was supported by the National Institutes of Health (GM59280 to M.G. Waters, GM59115 to M. Krieger, and GM29647 to J.E. Heuser).