Localization, Dynamics, and Protein Interactions Reveal Distinct Roles for ER and Golgi SNAREs

Rabbit antisera for msec13 (Tang et al., 1997), calnexin (Hammond and Helenius, 1994), rbet1 (Hay et al., 1997), syntaxin 5 (Hay et al., 1996), and a mouse monoclonal for GOS-28 (Subramaniam et al., 1996) have been previously described. A rabbit antibody to β-COP raised against the EAGE peptide (Duden et al., 1991) was a gift of Dr. Suzanne Pfeffer (Stanford University). Identical labeling patterns were obtained with antibodies raised against the same epitope, and were kindly forwarded by Dr. Irina Majoul and Dr. Jennifer Lippincott-Schwartz (National Institutes of Health). Anti-p58 was from Drs. Ralf Pettersson and Ulla Lahtinen (Ludwig Institute, Stockholm, Sweden). Immunofluorescence staining of syntaxin 5 was also carried out using a rabbit antiserum kindly provided by Dr. Bill Balch (Scripps Institute, La Jolla, CA). Rabbit antisera against mouse sec22b and rat membrin were raised by subcutaneous injection of bacterially expressed hexahistidine-tagged fusion proteins using Freund's adjuvant. The expression constructs included msec22b amino acids 2–195 and membrin amino acids 2–125. The rabbit antibodies were affinity-purified using bead-immobilized msec22b and membrin fusion proteins, respectively. The purified antibodies were eluted from affinity columns using 0.1 M glycine (pH 2.5), neutralized, and stored at 4°C in the presence of 0.05% azide. Mouse monoclonal antibodies against rbet1 were prepared by intraperitoneal injection of purified soluble bacterially expressed rbet1 (amino acids 2–95) produced as a glutathione-S-transferase (GST)– rbet1 fusion protein, and then thrombin-cleaved and separated from GST. Hybridoma production, ELISA screening, subcloning, expansion, and storage of cell lines were carried out by well-established methods (Harlow and Lane, 1988). Monoclonal antibodies 16G6 and 4E11 were affinity-purified using bead-immobilized recombinant rbet1. Routine Western immunoblotting experiments were carried out using ECL (Amersham Corp., Arlington Heights, IL) and autoradiography.

Subcellular localizations of the endogenous SNARE proteins were studied on ultrathin cryosections of PC12 and HepG2 cells (see Figs. 2–5) and exocrine pancreas, COS, and NRK cells (not shown). Cells and tissue were fixed in 2% formaldehyde and 0.2% glutaraldehyde in 0.1 M phosphate buffer, pH 7.4, and prepared for cryosectioning and double immunogold labeling as described (Slot et al., 1991) with the slight modification that sections were picked up in a 1:1 mixture of methyl cellulose and 2.3 M sucrose, which results in an improved ultrastructure (Liou et al., 1996). The best results were obtained with affinity-purified polyclonal rabbit antibodies against rsec22b, membrin, and syntaxin 5, and with culture medium from the mouse hybridoma 16G6 against rbet1, all at a dilution of 1:10. For GOS-28 (monoclonal HFD9), no specific labeling was obtained. Labeling density for a specific SNARE differed considerably between cell types. In PC12 cells, all SNARE proteins exhibited a high yield of labeling, and therefore all quantitative and double-labeling studies were carried out in these cells. For a quantitative analysis of the subcellular distributions of the ER/Golgi SNAREs (Table I), a high-quality section was selected at low magnification. Subsequently, the section was scanned systematically at 25,000×. All gold particles within a distance of 30 nm of a membrane were allocated to that type of membrane. About 90–98% of all gold particles were found over membranes of the ER, IC, and Golgi complex. The number of gold particles found over a specific compartment was then expressed as the percentage of total gold. For each SNARE protein, at least three EM grids were analyzed in this manner, and all quantitations were carried out in cells double-labeled for COPII. Since in ultrathin sections putative connections of COPII-coated buds with the ER are not always visible, for the quantitative analyses all COPII-coated membranes were considered part of the IC. The data in Table I summarizes the results of the different counting sessions, with about 150–200 gold particles counted per session.

In a separate experiment, the cis-trans distribution of membrin and syntaxin 5 over the Golgi complex was determined (Table II). About 25 Golgi complexes with a clear cis→ trans polarity were selected. The number of gold particles found over the individual cisternae, numbered G1→ G5 from cis to trans, was expressed as a percentage of the total number of gold particles present on the Golgi complexes.

To determine the relative enrichment of the ER/Golgi SNAREs on COPI- vs. COPII-coated membranes of the IC, double-labeling experiments for each SNARE with COPI or COPII were performed. In each counting session, at least 50 random IC membrane profiles that were positively stained for the SNARE protein were scored positive or negative for COPI or COPII. Tubular membrane profiles were scored positive for colocalization only if the SNARE gold particle was within a distance of 40 nm from the COPI or COPII gold particle. Quantitations of a specific combination of antibodies were carried out in at least four counting sessions, all on different EM grids. The number of SNARE-positive IC profiles that contained colocalizing COPI or COPII was expressed as the percentage of the total number of SNARE-positive IC profiles analyzed.

Cell lines were maintained in a 5% CO₂ incubator using routine media formulations. Before 15°C incubations, tissue culture medium was adjusted to 40 mM Hepes, pH 7.4. For 15°C incubations, cells cultured in microscope slide growth chambers (Nunc Inc., Naperville, IL) were placed below the water surface in a 15°C bath. Regardless of incubation temperature, cells were fixed with 4% paraformaldehyde at room temperature for 30 min. Immunostaining and fluorescence microscopy were carried out as described (Hay et al., 1996).

Two 15-cm plates of NRK cells growing at 37°C were buffered to pH 7.4 with 40 mM Hepes and then returned to 37°C or incubated below the surface of a 15°C water bath for 2 h. Cells were chilled on ice and then swollen in 10 mM ice-cold Hepes, pH 7.4, for 3 min. The Hepes was removed and replaced with 0.5 ml of homogenization buffer (Hay et al., 1996), and the cells were scraped from the plate with a no. 7 rubber stopper (sliced in half) and homogenized by five strokes in a Dounce homogenizer (Kontes Glass Co., Vineland, NJ). Nuclei and unbroken cells were removed by centrifugation at 1,000 g for 10 min. For the gradients in Fig. 8, A, D, and G, the postnuclear supernatant was adjusted to 0.75 M sucrose and used as the top layer of a 1.2-ml 0.75–1.75 M sucrose gradient in 20 mM Hepes, pH 7.2, 0.1 M KCl, 2 mM EGTA, 2 mM EDTA. The gradients were initially made with seven discrete layers, but diffusion during the overnight spin produced continuous density profiles. After an 18-h centrifugation at 135,000 g (r_av) in a TLS55 rotor (Beckman Instruments, Inc., Fullerton, CA), gradients were fractionated by hand from the top. Each fraction was supplemented with 2 μg BSA before precipitation with ice-cold methanol. The concentrated fractions were then electrophoresed, immunoblotted, and quantitated by autoradiography and densitometry. For gradients in Fig. 8, B, C, E, F, H, and I, 300 μl of postnuclear supernatant was adjusted to 600 μl of 0.75 M sucrose and layered on top of 700 μl of 1.125 M sucrose. Centrifugation and fractionation were as above.

Affinity-purified antibodies or control rabbit IgG were chemically conjugated to protein A- or G-Sepharose at a concentration of 2 mg antibody per ml packed Sepharose beads (Harlow and Lane, 1988). Rat liver microsomes were stripped with 1 M KCl and solubilized in 1% Triton X-100 as before (Hay et al., 1997). Liver extract was precleared by incubation overnight with protein A- and G-Sepharose to deplete it of protein A/G- and Sepharose-binding proteins. 400 μl of precleared extract was incubated with 10 μl 50% antibody-conjugated beads for 3 h at 4°C with agitation. Beads were then washed 3× with 400 μl of 20 mM Hepes (pH 7.2), 0.1 M KCl, 2 mM EGTA, 2 mM EDTA, 1% Triton X-100 plus protease inhibitors (see Hay et al., 1997), and once with the same buffer containing 0.2% Triton X-100. Immunoprecipitated proteins were eluted from the beads by incubation at room temperature with 1× SDS gel loading buffer lacking reducing reagents for 15 min. After removing the beads by centrifugation, the samples were adjusted to 5% 2-mercaptoethanol, boiled 5 min, and electrophoresed on SDS polyacrylamide gels. Electrophoresed proteins were transferred to nitrocellulose and immunoblotted as usual, except that ¹²⁵I-labeled secondary antibodies and Phosphor Imaging (Molecular Dynamics, Sunnyvale, CA) were used to quantify the results rather than ECL and autoradiography.

We estimate that the radioactive intensities of the bands detected by phosphor imaging in Fig. 9 are approximately indicative of the molar quantities of the proteins present. When scaled up, precipitations similar to that shown in the first column of Fig. 9 were electrophoresed and stained by Coomassie blue (Hay et al., 1997), and the relative intensities of the stained SNARE bands closely corresponded to the relative Western signals analyzed in Fig. 9.

The full-length or partial coding regions of rsly1, rsec22a, rsec22b, GOS-28, rbet1, membrin, syntaxin 5, and SNAP-25 were subcloned into the yeast expression vectors pGBT9 and pGAD424 (Clontech, Palo Alto, CA). The following vectors contain the following inserts: pGBT9/rsly1 (residues 15–648); pGBT9/rsec22a, /msec22b, /GOS-28, /rbet1, and /membrin and the corresponding pGAD424 constructs contain the complete coding regions of the proteins; pGAD424/syntaxin 5 and pGBT9/SNAP-25 also contain the complete rat coding sequence. Yeast strain HF7c (Clontech) was transformed with bait and prey vectors using the lithium acetate method (Schiestl and Gietz, 1989). Transformants were plated on selection plates lacking tryptophan, leucine, and histidine. After 5 d of incubation at 30°C, colonies were inoculated into supplemented minimal medium lacking tryptophan, leucine, and histidine, and were placed into a shaking incubator at 30°C for 44 h. Cultures were diluted 1:1 in fresh supplemented minimal medium lacking tryptophan, leucine, and histidine, and were incubated for 6 h while shaking at 30°C. β-galactosidase assays were performed on yeast extracts as described (Rose and Botstein, 1983).

Additional combinations of binding partners tested include pGBT9/ membrin in combination with pGAD424/membrin, msec22b, GOS-28, and rbet1. pGBT9/rsec22a as bait was tested in combination with pGAD424/msec22b, GOS-28, and rbet1. pGBT9/msec22b was tested for binding with msec22b, GOS-28, and rbet1. pGBT9/GOS-28 was tested for binding with pGAD424/GOS-28 and rbet1. Finally, pGBT9/rbet1 was tested with pGAD424/rbet1. None of these combinations resulted in sufficient binding to produce β-galactosidase activity above background.

Syntaxin 5 in vitro assembles with four other SNARE-like proteins: GOS-28, membrin, rsec22b, and rbet1. To better understand the subcellular organization of the ER and Golgi SNAREs, we took an immunoelectron microscopy approach. To enable immunogold labeling of the endogenous SNAREs, we produced new antisera to membrin, rsec22b, and rbet1. As shown in Fig. 1, anti-membrin and anti-rsec22b affinity-purified rabbit antibodies recognized only their specific target antigens in COS, HepG2, NRK, and PC12 cells. Two new mouse monoclonal antibodies to rbet1, 16G6, and 4E11 recognized rbet1 in a rat-specific fashion. Highly specific polyclonal antibodies to syntaxin 5 and rbet1 and a monoclonal antibody for GOS-28 have been described (Hay et al., 1996; Hay et al., 1997; Subramaniam et al., 1996).

The subcellular distributions of rsec22b, rbet1, membrin, and syntaxin 5 were established by immunogold labeling of ultrathin cryosections. All four SNARE proteins were detected in the distinct compartments of the early secretory pathway; i.e., the ER, IC, and Golgi. However, quantitative analysis of the labeling patterns obtained in PC12 cells revealed that their relative distribution over these compartments varied. Table I presents the fraction of the total immunoreactivity found in ER, IC, and Golgi. Most of the immunoreactivity for the four proteins investigated (between 65 and 81 percent) was found in membranes characteristic of the IC, whereas labeling over the ER was lower (between 7 and 12 percent). This distribution is consistent with the function of the intermediate compartment as the hub of vesicle trafficking in the early secretory pathway.

Subcellular localization of rsec22b was studied in detail in HepG2 (Fig. 2) and PC12 cells (Fig. 3,A), which gave identical labeling patterns. The majority of label was associated with the IC (Table I, Fig. 2,A). At the ultrastructural level, the IC consists of an elaborate network of vesicular and tubular membrane profiles that appear at distinct sites in the cell both near the Golgi and at the periphery. Many IC clusters, or VTCs, were found to be closely associated with RER cisternae exhibiting COPII-coated membrane buds representing export sites for anterograde cargo. Double-immunogold labeling revealed that rsec22b was often confined to these COPII-coated membranes of the ER and early IC (Fig. 2, A and B).

The labeling pattern obtained for rbet1 was very similar to that of rsec22b (Table I, Fig. 3). Simultaneous labeling of these two SNAREs within the same sections unequivocally localized them to the same membranes (Fig. 3,A). Membrin showed a more dispersed labeling over the IC than did rsec22b and rbet1, with label extending up to (and within, see below) the Golgi complex (Table I, Fig. 4, A and B). Syntaxin 5 showed a distribution pattern reminiscent of membrin, albeit with higher levels present in the Golgi (Table I, Fig. 5). Within the early membranes of the IC, syntaxin 5 colocalized with rbet1 and rsec22b (Fig. 5 A, inset) to the same membrane profiles.

While only a small percentage of rbet1 and rsec22b is found in the Golgi—6.3% and 8.7%, respectively—20% of membrin and 28% of syntaxin 5 are Golgi-localized. Within the Golgi, the rbet1 and rsec22b immunoreactivity, when present, was restricted to the cis-most cisterna (Figs. 2,C and 3,A). In contrast, both membrin and syntaxin 5 show significant immunoreactivity as far into the Golgi as the trans cisterna (Figs. 4,B and 5,B). A detailed analysis of the intraGolgi distribution of these latter two SNAREs revealed that membrin was mainly localized to the cis and medial cisternae, whereas equal amounts of syntaxin 5 were present in all but the trans-most cisternae (Table II).

To determine whether the ER/Golgi SNAREs displayed specificity for the characterized COPI and COPII vesicle pathways in the IC, we performed double-immunogold labelings with COPI (Figs. 2,B, 3,C, and 4 C; COPI is represented by β-COP) and COPII (Fig. 2, A and C, Fig. 3,B, and Fig. 5,A, COPII is represented by msec13). A quantitation of these analyses is presented in Table III. For this analysis we scored as positive only individual membrane profiles, including vesicles as well as larger tubular membranes, to which both sizes of gold particles colocalized. If a membrane was only partially coated, the structure was scored positive for colocalization only if the SNARE-representing gold particle was within a distance of 40 nm of the COP-representing gold particle. rbet1 and rsec22b were particularly enriched in COPII-coated structures (25% colocalization: Fig. 2, A and C, Fig. 3,B, Table III), consistent with their playing an important role on COPII-coated vesicles that are known to bud from the ER (Kuehn and Schekman, 1997), and are involved in the formation of peripheral VTCs (Rowe et al., 1998). All four SNAREs tested were present on COPI-coated membranes (Figs. 2,B, 3,C, and 4,C, Table III).

In summary, these data show that rsec22b and rbet1 have identical distribution patterns and localize preferentially to the early membranes of the IC, whereas membrin and syntaxin 5 have a more widespread distribution including the Golgi. All four SNAREs tested were detected in both COPII and COPI-coated membranes, consistent with the possibility that both of these vesicle types involve similar SNARE machinery.

Immunofluorescence subcellular localization of this set of proteins in NRK cells revealed that all five proteins most intensively stained the juxtanuclear Golgi area under control steady-state conditions (Fig. 6, first column; 37°C). This finding is consistent with the ultrastructural studies (Table I) since IC elements, which contain the majority of syntaxin 5, membrin, rsec22b, and rbet1, are most concentrated in the Golgi area and are in fact indistinguishable from the Golgi at the light level. GOS-28 appeared perhaps most tightly Golgi-localized, consistent with previous ultrastructural studies (Subramaniam et al., 1995; Nagahama et al., 1996). rsec22b and rbet1 occasionally displayed a distinctly spotty localization reminiscent of peripheral VTCs (Saraste and Svensson, 1991); however, most staining was juxtanuclear.

Colocalization of syntaxin 5, GOS-28, membrin, rsec22b, and rbet1 in the Golgi area under steady-state conditions provides no information on differences in their intracellular dynamics. For example, several of the proteins may operate itinerantly, traveling with cargo from the ER to Golgi and then recycling back for reuse while others may reside relatively statically on the Golgi or preGolgi elements. To compare intracellular dynamics of the proteins, normal rat kidney cells were incubated at 15°C, a temperature at which transport from the ER to Golgi is slowed dramatically, resulting in accumulation of vesicle components and cargo in peripheral VTCs (Saraste and Svensson, 1991; Matlin and Simons, 1983; Kuismanen and Saraste, 1989). As shown in Fig. 6, upon shift from 37°C to 15°C, the membrin distribution shifts to a much more peripheral spotty pattern. At 15°C, membrin is present in variously sized and shaped cytoplasmic foci, indicative of peripheral VTCs. Based on the observation that the membrin juxtanuclear staining deintensifies as the peripheral staining intensifies, it seems likely that retrograde transport continues at 15°C while anterograde movement is greatly slowed, resulting in a net shift of the steady-state localization from Golgi adjacent to peripheral VTCs. This redistribution also occurs in the presence of cyclohexamide (not shown), indicating that new membrin synthesis cannot account for intensifying peripheral sites. In the same set of cells, a resident Golgi protein, mannosidase II, remains statically in the Golgi as expected (not shown). Double-labeling between syntaxin 5 and rbet1 (Fig. 6, second and third rows) reveals that both of these proteins shift to peripheral VTCs when incubated at 15°C; however, syntaxin 5 appears to redistribute more slowly and less dramatically than rbet1. After 15 min at 15°C, syntaxin 5 exhibits a punctate staining pattern typical for peripheral clusters; however, these peripheral elements are relatively weak compared with the strong juxtanuclear staining. After 60 min, syntaxin 5 is more apparent in peripheral VTCs, but retains relatively strong juxtanuclear staining. The apparent partial commitment of syntaxin 5 to cycling VTCs may be a result of a pool of this protein that remains in the Golgi (see below and Figs. 8 and 9). A double-labeling between GOS-28 and rsec22b (Fig. 6, bottom two rows) reveals a dramatic difference in dynamics: throughout a 1-h 15°C incubation, GOS-28 remains statically in a relatively tight juxtanuclear pattern, whereas in the same cells, rsec22b rapidly concentrates in peripheral foci. These immunofluorescence images document a fundamental difference between the life cycles of GOS-28, on the one hand, which remains relatively statically localized, and rbet1, membrin, rsec22b, and syntaxin 5 on the other, which appear to be rapidly moving between peripheral and Golgi-centric sites.

Do the rapidly cycling proteins follow the same pathway? To look for potential differences in their recycling pathways, we performed double-label immunofluorescence comparisons among rbet1, rsec22b, and membrin, when arrested in peripheral VTCs. As shown in the top two panels of Fig. 7, the distributions of rbet1 and rsec22b at 15°C were very similar in double-labeled cells, with intense staining of VTCs. Significantly, in almost all instances both proteins were present in the same individual VTCs (Fig. 7, arrows), implying that rsec22b and rbet1 share a very similar recycling pathway. Notably, both proteins exhibited weak tubular/reticular ER staining (Fig. 7, arrowhead). Membrin and rbet1 antibodies also clearly labeled the same individual peripheral VTCs (Fig. 7, bottom two panels), indicating that these three dynamic SNAREs recycle together on the same VTCs. However, the membrin staining was very focused in the perinuclear area and peripheral VTCs, but was not apparent in the ER. Hence, unlike rsec22b and rbet1, membrin apparently does not cycle fully back through the ER. Alternatively, membrin may recycle through the ER very rapidly, making detection there difficult.

Since changes in antigen accessibility and concentration can potentially affect immunofluorescence and EM images, we sought to confirm independently the localization and dynamics of the SNARE proteins by biochemical fractionation of subcellular membranes on sucrose equilibrium gradients. The relative distributions of each SNARE on subcellular membranes was determined on 0.75–1.75 M sucrose gradients (Fig. 8, A, D, and G). A comparison of membrin with GOS-28, a protein limited to the Golgi (Subramaniam et al., 1995; Nagahama et al., 1996), demonstrates that these gradients do not resolve IC from Golgi since the bulk of both proteins substantially overlap in fractions 2, 3, and 4 (Fig. 8,A). Likewise, the majority of the IC marker protein p58 (not shown), syntaxin 5, rsec22b, and rbet1 also fall in fractions 2, 3, and 4 (Fig. 8, D and G). These gradients do, however, resolve IC/Golgi membranes of fractions 2, 3, and 4 from denser ER membranes of fractions 8, 9, and 10, where the ER protein calnexin peaks (Fig. 8,G). Hence, comparison of membrin with GOS-28 in Fig. 8,A reveals that very little of either of these SNAREs resides in the ER at steady state, consistent with the immunogold and immunofluorescence analyses (Table I, Figs. 6 and 7). As shown in Fig. 8,D, the 41-kD and 34-kD isoforms of syntaxin 5 distribute differently between Golgi/IC and ER membranes, as has been observed before (Hui et al., 1997), with the 41-kD isoform much more relatively enriched in the ER. rsec22b and rbet1 both exhibit a significant steady-state presence in the ER (Fig. 8,G), a distribution that is consistent with their apparent enrichment on COPII-coated vesicles that bud from the ER (Table III). The apparent discrepancy in the proportions of rsec22b, rbet1, and syntaxin 5 in the ER as determined by immunogold (Table I) and subcellular fractionation (Fig. 8) may be due to a lower immunogold labeling efficiency in the ER, perhaps due to protein interactions that limit antigen availability. It is also difficult to account for all of the gold particles that label the ER since this compartment is so broadly distributed throughout the cell.

Shallower sucrose gradients were able to resolve membranes of the Golgi and IC partially while pelleting ER. An analysis of membrin and GOS-28 on these gradients confirmed the dramatic difference in these SNAREs' dynamics observed by immunofluorescence. At 37°C, the two proteins resided on IC and Golgi membranes that substantially overlapped in their densities (Fig. 8,B). At 15°C, however, membrin and GOS-28 are noticeably better resolved (Fig. 8,C), presumably since membrin accumulates in peripheral VTCs, whereas GOS-28 remains statically in the Golgi (Fig. 6). We speculate that peripheral VTCs unable to migrate from the periphery at 15°C undergo shape, size, and/or content changes that reduce their density, resulting in a shift to lighter gradient fractions. Likewise, the Golgi at 15°C may be altered by the change in trafficking and shifts to fractions of slightly greater density. Other studies have identified accumulated VTCs in a light fraction approximately comigrating with TGN (J. Klumperman, A. Schweizer, H. Clausen, B.L. Tang, W. Hong, V. Oorschot, and H.-P. Hauri, manuscript submitted for publication). Interestingly, at 15°C, syntaxin 5 shifted to IC membranes of lesser density, as did membrin, but also retained a very significant presence on denser membranes containing GOS-28 (Fig. 8, E and F). This dual trafficking of syntaxin 5 would explain the protein's apparently slower and less complete recruitment to peripheral VTCs observed by immunofluorescence (Fig. 6). Perhaps a pool of Golgi syntaxin 5 is retained in the Golgi by its interactions with GOS-28 (see below). rbet1, rsec22b, and the IC marker p58 (called ERGIC-53 in primates) all displayed shifting distributions consistent with membrin at 37°C and 15°C (Fig. 8, H and I). Hence, the immunofluorescence (Figs. 6 and 7) and density gradient (Fig. 8) temperature shift experiments support itinerant cycling functions for rbet1, rsec22b, membrin, and syntaxin 5. In contrast, the relatively static localization of GOS-28 and a portion of syntaxin 5 is consistent with a function that is limited to the Golgi.

We previously found that two syntaxin 5–associated proteins—rbet1 and GOS-28—were not tightly associated with each other in detergent extracts (Hay et al., 1997), suggesting their presence in distinct mutually exclusive subcomplexes containing syntaxin 5 in common. To help clarify whether other exclusive interactions occur among this group of proteins, we performed immunoprecipitations with eight different polyclonal and monoclonal antisera against ER/Golgi SNAREs, and quantitated coprecipitating proteins by Western blotting using ¹²⁵I-labeled secondary reagents and phosphor imaging. As shown in Fig. 9 and Table IV, syntaxin 5, as expected, coprecipitated at least four ER/Golgi SNAREs, including GOS-28, rsec22b, membrin, and rbet1. All of these syntaxin 5–associated proteins reciprocally coprecipitated syntaxin 5, except for rsec22b, which unfortunately was primarily present as a partially degraded form that apparently does not participate in the protein–protein interactions. Hence, rsec22b precipitations did not provide information about protein complexes containing rsec22b.

The most striking feature of the pattern of coprecipitations was the relative isolation of GOS-28 from the other syntaxin 5–associated SNAREs. Although GOS-28 coprecipitated syntaxin 5 in near-stoichiometric amounts, other SNAREs failed to coprecipitate in significant amounts. A small portion of rsec22b coprecipitated as documented earlier (Hay et al., 1997), but this was only 11% of the amount of coprecipitating syntaxin 5. This apparent lack of association of GOS-28 with membrin, rsec22b, and rbet1 does not arise from a peculiarity of the anti-GOS-28 antibody since two different antibodies to membrin and three different antibodies to rbet1 coprecipitated insignificant or minor amounts of GOS-28 relative to other coprecipitating species. It appears that GOS-28 may be associated with syntaxin 5 in a complex lacking or excluding membrin, rsec22b, and rbet1. Most interestingly, this complex appears to involve primarily the 34-kD Golgi-centric isoform of syntaxin 5, as opposed to the 41-kD isoform or its major breakdown product of 31 kD. This is in contrast to rbet1 and membrin, which precipitate more balanced ratios of syntaxin 5 isoforms. These data suggest a functional difference among the syntaxin 5 isoforms; the 34-kD isoform may perform a distinct function in conjunction with GOS-28 in the Golgi.

Immunoprecipitation of rbet1 and membrin resulted in coprecipitation of stoichiometrically balanced ratios of associated SNAREs other than GOS-28, indicating that major exclusive relationships do not exist among syntaxin 5, rsec22b, membrin, and rbet1. These proteins may associate together simultaneously in a single quaternary complex.

Although four SNARE-like proteins coprecipitate with syntaxin 5 (Hay et al., 1997), the pattern of direct interactions among this group is unknown. We sought to determine which members of the complex may be included in the complex by virtue of direct interactions with syntaxin 5, as opposed to indirect interactions mediated through other proteins. The yeast two-hybrid system was used to detect and quantify interactions between syntaxin 5 (the 34-kD isoform) and seven potential syntaxin 5–binding proteins (Fig. 10). We used two-hybrid constructs containing the full-length SNAREs, including transmembrane domains. Two proteins presumably irrelevant to the syntaxin 5 complex, SNAP-25 and rsec22a, failed to stimulate β-galactosidase activity, consistent with the expectation that these proteins do not bind syntaxin 5. One of the proteins that coprecipitates with syntaxin 5, mouse sec22b, consistently stimulated a small amount of β-galactosidase activity, indicating that although this protein appears to bind syntaxin 5 weakly, other protein interactions may be critical for its inclusion in the syntaxin 5 complex. Four other members of the syntaxin 5 complex: rsly1 (Peterson et al., 1997; Dascher et al., 1996), GOS-28, membrin, and rbet1, all stimulated a relatively large amount of β-galactosidase activity, consistent with the possibility that each of these proteins can directly interact with syntaxin 5 independently of other protein interactions. Combinations of membrin, rsec22a, msec22b, GOS-28, and rbet1 (see Materials and Methods) were also tested for their ability to interact with each other in the two-hybrid system. No combinations were found to stimulate β-galactosidase activity above background levels. It is possible that pairwise interactions between these proteins are not of sufficient affinity to give a positive signal in the two-hybrid assay; however, the proteins may bind to each other in the context of higher order complexes. The two-hybrid studies delineating the pattern of interactions among this set of proteins limit the possible configurations of SNAREs that determine trafficking pathways in ER/Golgi transport, and help suggest supramolecular models of transport. However, since many of these proteins appear to interact directly with syntaxin 5, the number of possible distinct protein complexes is large, and more information is needed about precise individual complexes.

Here we document significant differences in the localization and dynamics among the set of ER/Golgi SNAREs that interact in vitro (Hay et al., 1997). As outlined below, the spatial and temporal differences begin to suggest modes of action for each of the proteins. Our localization data demonstrate that rsec22b and rbet1 are nearly identically distributed on membranes, and are particularly enriched in the COPII-coated membranes of ER exit sites and early IC elements, including vesicles and VTCs (Table I, Figs. 2 and 3). Our data implicate a role for these proteins on COPII vesicles that mobilize anterograde cargo from the ER, and fuse into nearby early peripheral VTCs (see our working model, Fig. 11). We often observe immunogold particles representing rsec22b and rbet1 on budding vesicles, much more often than for syntaxin 5 or membrin. rsec22b and rbet1 may mediate fusion of the transport vesicles with VTCs by virtue of SNARE complexes involving rsec22b, rbet1, membrin, and syntaxin 5 (primarily the 41-kD isoform; see model). A complex of these four proteins could involve interaction of four helical binding domains, one contributed by each protein. This interaction would be similar to the docking/fusion complex of the synapse in principle, where four helices interact, syntaxan and VAMP each contributing one helix, and SNAP-25 contributing two. rsec22b and rbet1 appear to reside on VTC elements as they cycle between peripheral and Golgi-adjacent locations, as evidenced by the dramatic redistribution of rsec22b and rbet1 caused by low-temperature incubations (Figs. 6, 7, and 8). Their presence on VTCs near the Golgi implicates them in a second round of membrane fusion: the fusion of mature VTCs to form nascent Golgi cisternae (Fig. 11). Although both rsec22b and rbet1 appear on Golgi-centric VTCs, they are uncommon in the Golgi (Table I) and are rarely found beyond the first full Golgi cisterna. Our localization data are consistent with rsec22b and rbet1 having the strongest retrieval determinants of all the SNAREs studied here, which would account for their relative abundance in peripheral locations including the ER and dramatic absence in the Golgi. We previously reported that myc-rbet1 targeted to Golgi and IC membranes in COS cells (Hay et al., 1996). Since Golgi cannot be distinguished from IC membranes at the light level, the relative distribution of the protein between Golgi and IC was not determined precisely (although we favored Golgi). The present EM study resolves that ambiguity, and indicates that very little rbet1 resides in the Golgi per se.

rsec22b and rbet1 were 2.5- to 3.5-fold more abundant on COPII-coated membrane profiles than were either syntaxin 5 or membrin (Table III). As discussed above, this fact, in conjunction with their presence in the ER (Fig. 8), implies that these two proteins function on ER-derived vesicles. However, none of the SNAREs were highly specific for COPI- vs. COPII-coated membranes. Hence, it appears that these two vesicle pathways involve highly overlapping SNARE machinery. It seems likely that all of the SNAREs examined in Table III recycle via COPI-coated membranes, and that this recycling may involve the same SNARE complexes as does anterograde transport (see model, Fig. 11). However, the interactions of SNAREs with coat protein machinery may be spatially regulated such that, for example, very early COPI vesicles would contain the majority of recycling rsec22b and rbet1, whereas later COPI vesicles in the Golgi would contain primarily syntaxin 5 and membrin. Such spatially regulated interactions could be one part of a means to prevent a seemingly random homogeneous distribution of ER/ Golgi SNAREs.

Membrin is significantly less abundant in the ER (Figs. 7 and 8), less enriched in early VTCs, more enriched in Golgi-adjacent VTCs, and extends significantly further in the secretory pathway than do rsec22b and rbet1, with representation throughout the cis and medial Golgi (Tables I and II). Although membrin does not appear enriched in COPII-coated membranes (Table III), it does share residence with rsec22b and rbet1 on the peripheral VTCs that accumulate at 15°C (Figs. 6 and 7). Hence, our data is consistent with a cycling role for membrin where it would participate in fusion events between VTCs and ER-derived vesicles, as well as in later fusion of VTCs to form nascent Golgi cisternae. Membrin's extension through the Golgi suggests a role in Golgi trafficking, perhaps in addition to its presumed role in ER/Golgi transport (see Fig. 11). In fact, a recent study found functional evidence for membrin's involvement in intraGolgi trafficking (Lowe et al., 1997). Although this study did not demonstrate a role for membrin in ER-to-Golgi trafficking, the presence of membrin on rapidly cycling VTCs and its participation in protein complexes containing rsec22b and rbet1 argue for a role in ER/Golgi trafficking. In addition, there is good evidence for involvement in ER/Golgi transport of BOS1p (Lian and Ferro-Novick, 1993), a yeast sequence homolog of membrin. Perhaps membrin's role at earlier steps is less susceptible to antibody inhibition. Although the antibody inhibition clearly demonstrated a role for membrin in Golgi trafficking, an anterograde vs. retrograde role cannot be distinguished due to the cycling nature of Golgi function.

Our data supports a dual role for syntaxin 5, like membrin, in ER/Golgi as well as intraGolgi transport. The 41-kD syntaxin 5, which is presumably more rigorously retrieved by virtue of its N-terminal COPI binding site (Hui et al., 1997), is more abundant in the ER and peripheral structures than the Golgi-centric form (Fig. 8). Our data is consistent with this isoform playing a role in the first fusion event, that between ER-derived rsec22b- and rbet1-containing transport vesicles and the early VTCs (see Fig. 11). This fusion event would presumably be directed by the complex we found between the 41-kD syntaxin 5, rsec22b, rbet1, and membrin (Fig. 9, Table IV). Since the 41-kD isoform exists in the ER as well (Table I, Fig. 8), it is also possible that it functions in the fusion of COPI-mediated retrograde vesicles with the ER, perhaps as a functional analog of the yeast ER SNARE UFE1p (Lewis and Pelham, 1996) that forms a presumed retrograde SNARE complex containing SEC22p and other proteins (Lewis et al., 1997). Our localization and dynamics data are in principle consistent with a recent study documenting a functional role for syntaxin 5 at an early preGolgi stage in transport (Rowe et al., 1998). However, our data do not favor a primary role for syntaxin 5 on ER-derived transport vesicles as this report suggested. Instead we find rsec22b and rbet1 much more prevalent on these structures (Table III), implying that syntaxin 5 may serve as the vesicle receptor on the early VTC membrane (see model, Fig. 11). Later, as the VTC migrates and matures, syntaxin 5 may participate in intra-VTC fusion that may form new Golgi cisternae, an event also perhaps involving the syntaxin 5/rsec22b/rbet1/membrin complex.

The 34-kD syntaxin 5 isoform, on the other hand, is more abundant in the Golgi and Golgi-adjacent VTCs (Fig. 8). In contrast to previous ultrastructural studies localizing syntaxin 5 to Golgi-adjacent VTCs and cis Golgi (Banfield et al., 1994), we find syntaxin 5 immunogold particles nearly equally distributed throughout the Golgi stacks (Table III). In support of a special role for the 34-kD isoform of syntaxin 5 in Golgi trafficking, we have documented a pool of 34-kD syntaxin 5 molecules that associates with GOS-28 in a complex that lacks or excludes the dynamic SNAREs membrin, rsec22b, and rbet1 (Fig. 9, Table IV). We speculate that this complex could serve to anchor a pool of the 34-kD syntaxin 5 in the Golgi, could regulate participation of syntaxin 5 in docking interactions with other SNAREs, or could provide a basis for intraGolgi vesicle/cisternae fusions. In each of these possible capacities, the syntaxin 5/GOS-28 complex would occur within the Golgi, since our studies indicate that GOS-28 and a portion of syntaxin 5 molecules do not travel on VTCs moving between peripheral and Golgi-adjacent locations (Figs. 6 and 8). It is tempting to speculate (see Fig. 11) that the syntaxin 5/GOS-28 complex represents a docking complex between Golgi-derived vesicles containing GOS-28 and Golgi cisternae containing 34-kD syntaxin 5. This possibility would be consistent with a previous report localizing recombinant GOS-28 to Golgi rim vesicles, and demonstrating an inhibitory effect of GOS-28 antisera on intraGolgi transport (Nagahama et al., 1996). It would also be consistent with the finding that overexpression of truncated syntaxin 5 weakly inhibited intraGolgi as well as ER-to-Golgi VSV G protein transport (Dascher et al., 1994). The data seems less consistent with the proposed role for GOS-28 in ER to Golgi transport (Subramaniam et al., 1996; Hay et al., 1997).

We (Hay et al., 1997; this report, Fig. 9) and others (Lowe et al., 1997) have observed coprecipitation of rsec22b with GOS-28 and coprecipitation of GOS-28 with rbet1, but these are on a molar basis minor compared with the major complexes described above. One of the anti-rbet1 antibodies, 4E11, coprecipitated significantly more GOS-28 than the other two rbet1 antisera; however, even in this most efficient precipitation the amount of associated GOS-28 was less than 19% of the associated rsec22b, membrin, or syntaxin 5. These relatively minor GOS-28-rbet1 interactions may arise from lateral associations of distinct SNARE complexes, or may represent intermediates in the transition between the major complexes. For example, although the function of GOS-28 with syntaxin 5 may not depend upon the presence of rbet1 (and vice versa), GOS-28 and rbet1 may not bind to syntaxin 5 in a strictly mutually exclusive manner, allowing some overlap between the major complexes to occur. Such overlap complexes could potentially serve a spatially defined regulatory purpose.

How does intraGolgi traffic work? Unfortunately, defining the distribution of SNAREs through the Golgi does not uniquely favor any one model of intraGolgi transport, including vesicle transport, cisternal maturation, and interstack tubules (Pelham, 1998; Mironov et al., 1998). For example, interstack vesicle transport throughout the Golgi could be mediated by syntaxin 5 (34 kD) and GOS-28 alone. There is no need for distinct t-SNAREs on each stack as long as spatial regulation or polarity is provided by other mechanisms such as microtubules or the Golgi spectrin/ankyrin cytoskeleton (Devarajan et al., 1997; Burkhardt et al., 1997). For completeness, the model shown in Fig. 11 incorporates nonmutually exclusive elements of cisternal maturation as well as interstack vesicle transport. It is possible that rsec22b, rbet1, syntaxin 5, membrin, GOS-28, and perhaps ykt6 (McNew et al., 1997) provide sufficient SNARE machinery to direct cargo from the ER all the way through to the TGN where cargo is sorted into one of several major pathways leading to destinations within the cell or without.

Though a complete picture of ER-to-Golgi and intraGolgi trafficking are not available, we now have enough information about the localization, dynamics, and interactions among ER/Golgi SNAREs to begin formulating useful models and specific testable hypotheses. Syntaxin 5 is likely to emerge from future studies as a common component of several spatially distinct membrane fusion complexes critical for early steps in the secretory pathway. Understanding the composition, regulation, and integration of these complexes with rabs, sec1 proteins, and membrane coat systems will provide the basis for a firm understanding of how cells mobilize and traffic their products.

We are indebted to Dr. Andy Bean (University of Texas-Houston Health Sciences Center) for technical training and advice on the yeast two-hybrid system, and for the gift of the pGAD-syntaxin 5 construct. We also wish to thank Dr. Keith Mostov (University of California-San Francisco) for the suggestion that a syntaxin 5-membrin-rsec22b-rbet1 complex could involve four SNARE binding domains, as does the synaptic ternary complex where SNAP-25 contributes two such domains.

IC

intermediate compartment

SNARE

soluble N-ethylmaleimide–sensitive factor attachment protein receptor

VTC

vesicular tubular cluster