Molecular Mass, Stoichiometry, and Assembly of 20 S Particles*

N -Ethylmaleimide-sensitive factor (NSF), soluble NSF attachment proteins (SNAPs), and SNAP receptor (neu-ronal SNARE) complexes form 20 S particles with a mass of 788 6 122 kDa as judged by scanning transmission electron microscopy. A single NSF hexamer and three a SNAP monomers reside within a 20 S particle as determined by quantitative amino acid analysis. In order to study the binding of a SNAP and NSF in solution, to define their binding domains, and to specify the role of oligomerization in their interaction, we fused domains of a SNAP and NSF to oligomerization modules derived from thrombospondin-1, a trimer, and cartilage oligomeric matrix protein, a pentamer, respectively. Binding studies with these fusion proteins reproduced the interaction of a SNAP and NSF N domains in the absence of the hexamerization domain of NSF (D2). Trimeric a SNAP (or its C-terminal half) is sufficient to recruit NSF even in the absence of SNARE complexes. Further-more, pentameric NSF N domains are able to bind a SNAP in complex with SNAREs, whereas monomeric N domains do not. Our results demonstrate that the oligomerization of both NSF N domains and a SNAP pro-vides a critical driving force for their interaction and the assembly of 20 S particles.

Transport within the eukaryotic cell distributes cargo molecules to the proper membrane-bound compartments and thus maintains the distinct protein and lipid composition of organelles. Along the secretory pathway, vesicular carriers mediate each transport step. Continuous budding and fusion events ensure the bi-directional transport of cargo molecules and the preservation of the donor and acceptor compartment structure. Vesicular transport is conserved from cells as diverse as unicellular yeast to neurons of vertebrate organisms (1,2).
Two classes of proteins form a core machinery involved in membrane fusion reactions. The first class of proteins, referred to as SNAREs 1 (3), specifies the targeting of transport vesicles and underlies their subsequent fusion with acceptor compartments (4). SNAREs are a family of membrane receptors that are enriched either on vesicles (v-SNAREs) or target membranes (t-SNAREs). v-and t-SNAREs form heteromeric complexes that pair a vesicle and target membrane for fusion.
Cognate v-and t-SNARE reconstituted in separate liposome populations are sufficient to induce lipid bilayer fusion and content mixing (4 -9). SNAREs were first identified at the neuronal synapse, and three proteins, vesicle-associated membrane protein/synaptobrevin (on the synaptic vesicle) (10,11), syntaxin 1A/1B (12), and SNAP-25 (13) (both on the neuronal plasma membrane) are the founding members of the family. These three proteins assemble in a trimeric complex that assumes a parallel four-stranded helical bundle arrangement with all membrane-spanning domains at one end of the complex (14 -19).
The second class of proteins includes the soluble proteins NSF (20) and SNAPs (21). SNAPs exist in three isoforms; ␣SNAP and ␥SNAP are expressed ubiquitously, whereas ␤SNAP expression is restricted to the brain (21)(22)(23). SNAPs bind to SNARE complexes and in turn enable NSF to bind, yielding 20 S particles, named for their characteristic sedimentation coefficient of 20 Svedberg units (24). Mutational analysis of ␣SNAP has shown that both its N-terminal and C-terminal regions are required for the interaction with SNAREs (25)(26)(27). The crystal structure of the yeast homologue of ␣SNAP, Sec17p (28), revealed a flat, sheet-like structure consisting of numerous ␣-helical repeats with a pronounced left-handed twist; the N and C termini reside at opposite ends of the structure (29). Electron microscopy indicated that ␣SNAP binds to the lateral side of the SNARE complex rod in a sheath-like manner with its N terminus in close proximity to the SNARE membrane anchors. NSF binds the SNAP⅐SNARE complex at the end opposite to the SNARE membrane anchors (16).
NSF and SNAPs cooperate to disassemble 20 S particles through the ATP hydrolysis activity of NSF (24,30), yielding free v-and t-SNAREs for subsequent membrane trafficking reactions. SNAPs may transfer the energy generated by ATP hydrolysis to dissociate SNARE complexes (for a review see Ref. 31). NSF is a homohexameric ATPase (32), and each NSF subunit is composed of three domains, an N-terminal N domain and two ATP binding domains termed D1 and D2 (33). The ATPase activity of the D1 but not of the D2 domain is required for intra-Golgi protein transport and in vitro SNARE complex disassembly (34,35). Hexamerization of NSF is mediated by the D2 domain (36,37). The N domain is required for the binding of NSF to the SNAP⅐SNARE complex (35). Crystallographic studies revealed that individual N domains consist of two subdomains with an intervening groove. Both the side chains that line the groove between the N subdomains (38) and a flexible loop connecting the subdomains (39) have been suggested to serve as a SNAP-binding site. However, the nature and the stoichiometry of the NSF-SNAP interaction remain largely unknown.
To investigate the NSF-SNAP interaction further, we determined the stoichiometry of these proteins in 20 S particles and the overall mass of 20 S particles. Our results indicate that three ␣SNAP monomers are present in a 20 S particle, and we hypothesize that this oligomerization of ␣SNAP by SNARE complexes is sufficient to recruit a single NSF hexamer into a 20 S particle. These data led us to develop a binding assay to examine the requirement for oligomerization in the interaction between ␣SNAP and NSF. NSF and SNAP (or domains thereof) were expressed as fusion proteins with synthetic oligomerization domains derived from heptad repeat sequences that form oligomeric coiled-coils. Heptad repeats share the sequence periodicity (abcdefg) n with hydrophobic amino acids at positions a and d, and n indicates the number of repeats (40). These sequences were initially identified in tropomyosin where they mediate homodimerization (41). Heptad repeat domains have subsequently been shown to direct trimerization, tetramerization, and pentamerization of polypeptides (42,43). The identity of the amino acid residues in positions a and d can determine the degree of homo-oligomerization (42), and cysteines can stabilize oligomers by forming interhelical disulfide bridges (44). We prepared fusion proteins consisting of the following: (i) the N domain of NSF and the pentamerization domain of the cartilage oligomeric matrix protein, while omitting the D1 and the D2 (oligomerization) domains of NSF; and (ii) ␣SNAP and the trimerization domain of thrombospondin. In the latter case, the trimerization domain replaces the function of the SNARE complexes that act as an oligomerization domain by providing three ␣SNAP-binding sites (25,45). Binding studies with these fusion proteins demonstrated that oligomerization of the N domain of NSF and trimerization of ␣SNAP are a driving force behind their interaction.

EXPERIMENTAL PROCEDURES
Scanning Transmission Electron Microscopy-For scanning transmission electron microscopy, 20 S particle preparations (16) were stabilized with 0.05% (v/v) glutaraldehyde and applied to thin carbon films, supported by a fenestrated carbon layer that covered the goldcoated copper grid. The grids were washed in 6 drops of quartz doubledistilled water and freeze-dried at Ϫ80°C overnight in the STEM HB-5 microscope. Digital images were subsequently recorded from the unstained samples at a nominal magnification of ϫ 200,000, using doses of 200 -400 electrons/nm 2 and an accelerating voltage of 80 kV. The mass evaluation was performed with the specialized program package IMP-SYS run on a VAX computer system (46). Individual particles visually discernible on the images were selected by enclosing them in circles. For each of these defined areas, the total number of electrons elastically scattered by the particle and its supporting carbon film was calculated, and the signal from an equivalent area of carbon film alone was subtracted. For thin samples, this signal is directly proportional to the mass. Details of the instrument's calibration and the technique of STEM mass measurement may be found in Refs. 46 and 47. The beam-induced mass loss was determined experimentally by repeatedly scanning the same grid region and monitoring the change in sample mass (46). The mass analysis data sets were corrected accordingly.
Quantitative Amino Acid Analysis-Three different 20 S particle purification procedures were utilized in the quantitative amino acid analysis. In the first procedure (Table I, experiment 1), 20 S particles were assembled on immobilized GST-␣SNAP and purified by velocity centrifugation as described (16). Gradient fractions containing 20 S particles were mixed with HPC-1 (anti-syntaxin) antibodies covalently crosslinked to protein G-Sepharose beads and incubated at 4°C overnight with gentle shaking. The beads were washed three times in buffer A (25 mM HEPES-NaOH, pH 7.0, 100 mM NaCl, 10% glycerol, 0.05% Triton X-100, 2 mM EDTA, and 0.5 mM ATP, and 1 mM dithiothreitol), and bound proteins were eluted with 100 mM glycine, pH 2.7, 0.1% Triton X-100, precipitated with trichloroacetic acid, washed twice with acetone, and boiled in Laemmli buffer. In the second 20 S particle purification procedure (Table I, experiment 2), glycerol gradient fractions containing 20 S particles (assembled on GST-␣SNAP) were concentrated and loaded on a TosoHaas TSK500SW XL size exclusion column attached to an HPLC system. The column was run in buffer A, and the fractions containing 20 S particles were collected and precipitated for electrophoresis as above. In the third 20 S particles purification procedure, 20 S particles were assembled in solution by mixing 3.0 mg of bovine brain extract, 120 g of ␣SNAP, and 500 g of His 6 -NSF-myc in buffer A (final volume 450 l) and sedimented into glycerol gradients (16). Fractions containing 20 S particles were immunoprecipitated with HPC-1 antibodies and processed as above. All immunoprecipitates and column fractions were resolved by electrophoresis on High-Tris-urea gels as described (16) and stained with Coomassie Blue R-250. Gel slices containing ␣SNAP or NSF were forwarded to the W. M. Keck Facility (Yale University) for quantitative amino acid analysis using tritiated norleucine as an internal standard. Polypeptides embedded in the gel slices were completely acid-hydrolyzed into their constituent amino acids. The hydrolysis products were separated by ion exchange chromatography, stained with ninhydrin, and quantitated spectroscopically. For each gel slice containing NSF or ␣SNAP, a control gel slice (matching the sample slice in weight and migration position) from an adjacent empty lane of the gel was utilized for background subtraction. In each experiment, the amount (in nanomoles) of 12 amino acids and amino acid derivatives (Asx, Thr, Ser, Glx, Ala, Val, Ile, Leu, Tyr, Phe, His, and Lys) was obtained for His 6 -NSF-myc and ␣SNAP. The amount of each amino acid or amino acid derivative was divided by the frequency of the corresponding residue(s) in His 6 -NSF-myc or ␣SNAP, respectively. The resulting values for His 6 -NSF-myc were divided by the corresponding values for ␣SNAP to obtain a NSF:␣SNAP ratio for each of the 14 amino acids or amino acid derivatives measured. For each experiment, the NSF to ␣SNAP molar ratio listed in Table I represents the average of these 14 ratios.
Plasmids for the Expression of Multimerization Domains-The Escherichia coli expression plasmid pHIN-PEP is based on the Pab fusion gene described in Ref. 48 and was generated by ligating oligonucleotides that encode the sequence outlined in Fig. 2A into the plasmid pQE9 (Qiagen) linearized with BamHI and HindIII. The position of a Myc epitope and relevant restriction sites are outlined in Fig. 2B. To obtain pTRI-HIN, a cDNA coding for amino acids 265-314 of thrombospondin (GenBank TM accession number M25631) was generated by PCR using TSP1 as template. The cDNA encoded, in sequential order, a 5Ј KpnI site, amino acids 265-314 of thrombospondin, the amino acids CDACG, a camel IgG hinge (HIN), a Myc epitope, a stop codon, and 3Ј NotI and HindIII sites. The amplification product was digested with appropriate restriction enzymes and ligated into pHIN-PEP linearized with KpnI and HindIII. To generate pHIN-TRI, a cDNA encoding the trimerization motif of thrombospondin and the flanking elements was amplified by PCR as outlined in Fig. 2B with a 5Ј BglII and a 3Ј HindIII site. The PCR product was ligated into pQE9 linearized with BamHI and HindIII. The DNA sequences of the inserts described are deposited in GenBank TM .
Plasmids for the Expression of Monomeric and Multimeric ␣SNAP and NSF-To obtain an expression vector for ␣SNAP trimerized at its C terminus (␣SNAP-tri), a cDNA encoding ␣SNAP was synthesized by PCR with a 5Ј BamHI and a 3Ј EcoRI site. The PCR product was ligated into pHIN-TRI linearized with BamHI and MunI. For expression vectors encoding ␣SNAP (tri-␣SNAP) and ␣SNAP (residues 157-295) trimerized at the N terminus (tri-␣SNAP-(157-295)), DNA fragments encoding full-length ␣SNAP and the amino acids 157-295 were prepared and ligated as outlined for ␣SNAP-tri. A expression vector for the N domain of NSF (NSF-(1-207)) was obtained by amplifying DNA coding for the amino acids 1-207 by PCR with a BamHI site at the 5Ј end and a stop codon TGA followed by a HindIII site at the 3Ј end. The fragment was ligated into pQE9 linearized with the same enzymes. To obtain an expression vector for the pentamerized N domain of NSF (pent-NSF-(1-207)), the corresponding DNA fragment was synthesized by PCR with a BamHI site at the 5Ј end and a MunI site at the 3Ј end and ligated into pHIN-PEP linearized with the same enzymes. In all cases the restriction sites were added immediately before or after the indicated amino acid codons.
Coupling of Antibodies to Protein G-Sepharose-HPC-1 (49) and the anti-Myc antibody 9E10 (50) were covalently coupled to protein G-Sepharose at a concentration of 1.0 mg of antibody per ml of beads with dimethyl suberimidate dihydrochloride (Fluka) as described (51).
Analysis of the Interaction of NSF with Mutant ␣SNAP-2 g of His 6 -NSF were incubated with 300 ng of myc-␣SNAP and 250 g of bovine brain membrane extract, 300 ng of myc-␣SNAP, 500 ng of tri-␣SNAP, 300 ng of tri-␣SNAP-(157-295), 300 ng of trimerization domain, and 500 ng of ␣SNAP-tri, respectively, in 250 l of buffer C (150 mM KCl, 2 mM EDTA, 1 mM ATP, 0.5 mg/ml ovalbumin, 10% glycerol (w/v), 0.5% Triton X-100, and 25 mM HEPES, pH 7.4) at 4°C for 1 h. The samples were subjected to a clarifying spin at 14,000 ϫ g for 5 min, and the supernatants were incubated with 9E10 coupled to protein G-Sepharose for 30 min at 4°C. The samples were washed as above, and bound protein was eluted with 1% (w/v) Bigchap (Sigma) in 0.1 M glycine (pH 2.8). The eluates were precipitated with 10% trichloroacetic acid (w/v), and protein was dissolved in Laemmli buffer (52), boiled, subjected to SDS-PAGE, transferred to nitrocellulose, decorated with the indicated antibodies, and visualized with ECL (Amersham Pharmacia Biotech).
Analysis of the Interaction of SNARE-bound ␣SNAP with Mutant NSF-200 ng of ␣SNAP and 250 g of bovine brain extract (3) were incubated with one of the following: 2.5 g of His 6 -NSF, 600 ng of pent-NSF-(1-207), 300 ng of monomeric NSF-(1-207), or 300 ng of pentamerization domain for 5 min at 4°C in 300 l of buffer D (75 mM KCl, 2 mM EDTA, 1 mM ATP, 0.5 mg/ml ovalbumin, 10% glycerol (w/v), 0.5% Triton X-100, and 25 mM HEPES, pH 7.4). After a clarifying spin at 14,000 ϫ g for 5 min, the supernatant was added to 12.5 l of HPC-1 protein G-Sepharose and incubated at 4°C for 1 h. The beads were washed 3 times in buffer C and 2 times in buffer C containing 0.2% Triton X-100 and lacking ovalbumin. The bound proteins were dissolved in 50 l of Laemmli buffer. Western blot analysis was carried with the antibody 5F7. This monoclonal antibody was obtained by immunizing a Balb/c mouse with hexahistidine-␣SNAP followed by fusion with SP/2 myeloma cells (51). The antibody recognizes the hexahistidine motif of proteins expressed from the E. coli expression plasmid pQE9.

RESULTS
The Molar Ratio of ␣SNAP and NSF within 20 S Particles-20 S particles were isolated, stabilized with a low concentration of glutaraldehyde, and examined by scanning transmission electron microscopy (STEM) to determine their mass. 783 particles were selected from the digital images. Their mass values were corrected for beam-induced mass loss using the global scale factor 1.04 (see "Experimental Procedures"), pooled, and displayed in a histogram (Fig. 1). The data were fitted by two Gauss curves. The major peak is centered at 788 kDa (S.D. ϭ Ϯ122 kDa; n Ϸ 690). The accuracy of the result (S.E. ϭ Ϯ5 kDa where S.E. ϭ S.D./͌n ) is limited by the 5% uncertainty in the absolute calibration of the instrument. Accordingly, the mass measured is precise to Ϯ39 kDa. The other much broader Gauss curve implies that the higher mass values arise from a wide range of nonspecific aggregates.
To evaluate the NSF to ␣SNAP stoichiometry, 20 S particles were purified, and their constituents were separated electrophoretically on polyacrylamide gels. The bands corresponding to ␣SNAP and NSF were excised and hydrolyzed with acid, and their amino acid composition was quantitatively determined (see "Experimental Procedures"). Different 20 S particle purification procedures yielded nearly identical NSF to ␣SNAP molar ratios, regardless of whether the 20 S particles were assembled on GST-␣SNAP (Experiments 1 and 2) or in solution (Experiments 3 and 4), and further purified by immunoprecipitation with anti-syntaxin (HPC-1) antibodies (Experiments 1, 3, and 4) or by gel filtration chromatography (Experiment 2). The results of these four independent experiments are consistent with a single NSF hexamer and three copies of ␣SNAP in a 20 S particle. However, the molar ratio of NSF to ␣SNAP above 2 (average 2.3, see Table I) leaves the possibility that under the employed experimental conditions, a subpopulation of 20 S particles containing only two ␣SNAPs, was also isolated. It remains open whether such an arrangement represents functional 20 S complexes capable of SNARE complex dissociation. Nevertheless, previous binding studies of SNAPs to SNAREs (25,45) and the mass determination of the 20 S particle suggest a stoichiometry for NSF⅐SNAP⅐v-t-SNARE complex of 6:3:1. The calculated mass of a His 6 -NSF-myc hexamer (510 kDa), three ␣SNAPs (105 kDa), one SNARE complex consisting of syntaxin 1A (33 kDa), SNAP-25 (23 kDa), and synaptobrevin II/vesicle-associated membrane protein-2 (12 kDa), and a Triton X-100 micelle (81 kDa) is 764 kDa. This is in good agree- FIG. 1. Molecular mass of 20 S particles. A preparation of 20 S particles (16) was stabilized with glutaraldehyde and adsorbed to thin carbon films. Images were recorded at a magnification of 200,000 by scanning transmission electron microscopy. Individual particles were selected for mass analysis. The data have been corrected for beam induced mass loss (32). The histogram displays the relative molecular mass as a function of the number of particles. The data analysis (dotted lines) fits best with two populations of particles. One has an average mass of 788 Ϯ 122 kDa. The broad peak of high mass represents a minor population and cannot be interpreted. The sum of the two peaks is displayed to show the overlap with the histogram.  (46). Protein Interactions within 20 S Particles-The molar ratio of NSF and ␣SNAP in 20 S particles served as a guide for the design of oligomerization constructs. Bacterial expression plasmids were designed to allow trimerization of proteins of interest at their N or C terminus, respectively, or pentamerization at their C terminus. The pentamerization motif was chosen because we are not aware of a heptad repeat that can act as a hexamerization domain. The trimerization domain derived from thrombospondin was based on a mutational analysis that identified a 140-amino acid stretch sufficient for trimerization (53). By using coiled-coil prediction (54) and sequence comparison to the pentamerization domain of cartilage oligomeric matrix protein (55), we narrowed the 140amino-acid-long stretch to a polypeptide of 49 amino acids, which was used for the trimerization construct ( Fig. 2A). The pentamerization construct is derived from cartilage oligomeric matrix protein as described previously (48). The oligomerization domains were separated from their fusion partners by a flexible linker encoding a 26-amino-acid camel IgG hinge region (HIN) (Fig. 2B). For the binding experiments presented in this study, ␣SNAP and NSF were fused to multimerization domains as outlined in Fig. 3A.
The Oligomeric Status of the Fusion Proteins-The oligomerization status of all fusion proteins outlined in Fig. 3A was characterized by velocity sedimentation analysis (Fig.  3B). The pentameric N domain of NSF sedimented much more rapidly than the monomeric N domain indicating that the pent-NSF-(1-207) domain fusion protein behaves as a pentamer in solution. Tri-␣SNAP-(157-295) has a predicted molecular mass of 77 kDa. It sedimented faster than monomeric ␣SNAP (33 kDa) and co-sedimented with bovine serum albumin (68 kDa), suggesting that its oligomerization status is trimeric. Full-length ␣SNAP trimerized at the N or the C terminus has a predicted mass of 130 kDa and migrated in the glycerol gradient as a broad peak that could encompass trimers and possibly dimers. The shape and hydrodynamic volume of these oligomers will affect their sedimentation behavior. Interestingly, full-length Sec17p (the yeast ortholog of ␣SNAP) forms an asymmetric extended structure, in contrast to its more globular C-terminal domain, which could explain the aberrant migration of full-length tri-␣SNAP. We conclude that the domain of thrombospondin outlined in Fig. 2A is sufficient for trimerization of a heterologous protein and that the major form of the ␣SNAP fusion proteins is trimeric. Both the trimerization and the pentamerization domains remained in their oligomeric status in the presence of reducing agents such as 2 mM 2-mercaptoethanol.
The Trimerization of ␣SNAP Is Sufficient to Bind NSF-In the first set of binding experiments we sought to identify the ␣SNAP oligomerization required to bind NSF. In all cases, Myc-tagged ␣SNAP constructs were incubated with the indicated binding partners in solution. ␣SNAP was then immunoprecipitated with anti-Myc antibodies immobilized on pro-  Fig. 2. B, velocity gradient centrifugation of ␣SNAPand NSF-derived proteins. The proteins (a-f) were sedimented into 4 -20% glycerol gradients as described under "Experimental Procedures." 12 fractions were collected. The proteins were precipitated, subjected to SDS-PAGE, transferred to nitrocellulose, and analyzed with the antibody 5F7. This antibody recognizes the hexahistidine motif of the proteins analyzed. The molecular mass standards cytochrome c (12.5 kDa), bovine serum albumin (68 kDa), and aldolase (158 kDa) migrated at the indicated positions. tein G beads, and bound proteins were analyzed by immunoblotting and detected by chemiluminescence. Monomeric ␣SNAP was incubated with NSF in the presence and absence of neuronal SNAREs, whereas NSF only bound to ␣SNAP in the presence of SNAREs (Fig. 4A, lanes 1 and 2). In contrast, NSF bound to tri-␣SNAP (trimerized at its N terminus) in the absence of SNAREs (lane 3). As a control, ␣SNAP-tri (trimerized at its C terminus) did not bind NSF under these conditions (Fig. 4A, lane 6). This observation suggests that the C terminus of ␣SNAP might interact directly with NSF and that the trimerization domain might interfere with the SNAP-NSF N domain interaction. To test this proposal, we prepared a fusion protein consisting of amino acids 157-295 of ␣SNAP (tri-␣SNAP-(157-295)) fused at its N terminus to the trimerization domain, and we found that it can support NSF binding (Fig. 4, lane 4).
Pentameric, but Not Monomeric, NSF N Domains Bind to ␣SNAP-In a second set of binding experiments we characterized the ␣SNAP-binding properties of NSF N domains. ␣SNAP, neuronal SNAREs, and NSF-derived proteins were included in the binding reaction; protein complexes were immunoprecipitated with anti-syntaxin antibodies and analyzed as above. Full-length NSF bound to the complex of ␣SNAP and neuronal SNAREs (Fig. 4B, lane 1). The pentamerized N domain of NSF (pent-NSF-(1-207)) bound to the SNAP⅐SNARE complex (lane 3), whereas monomeric NSF N domains or the pentamerization domain alone did not (lanes 5 and 6). This result suggests that the D1 and D2 domains of NSF are not required for ␣SNAP binding and do not contribute to its SNAP/SNARE-binding surface. Although hexamerization of the N domain would be the preferred choice to examine the oligomerization requirements for N domain binding to SNAP⅐SNARE complexes, pentamerization of the NSF N domain is sufficient to bind the ␣SNAP⅐SNARE complex. However, there is a higher sensitivity toward the ionic strength. The results of the binding experiments with trimerized ␣SNAP and full-length NSF (Fig. 4A) were obtained at 150 mM KCl, whereas the binding experiments with pentamerized NSF N domains (Fig. 4B) were carried out in the presence of 75 mM KCl. In the presence of 150 mM KCl, the amount of bound pentamerized N domain of NSF decreased by 50%. 2 In conclusion, the binding experiments suggest that the driving force behind 20 S particle formation is the SNARE-induced trimerization of ␣SNAP that results in the recruitment of NSF. The domains of physical contact between NSF and ␣SNAP in 20 S particles include the N domain of NSF and the C-terminal half of ␣SNAP.

DISCUSSION
Assembly and disassembly of 20 S particles are crucial events in generating fusion-competent SNAREs. We have isolated 20 S particles, and subsequently determined their molecular mass by STEM. Based on quantitative amino acid analysis and morphological data of 20 S particles, we constructed oligomeric versions of ␣SNAP and the N-terminal domain of NSF that mimic the protein interactions in 20 S particle assembly. The study shows that trimerization of ␣SNAP is sufficient for NSF binding. A model for the interaction is proposed in Fig. 5. These results are also consistent with previous studies demonstrating that the neuronal SNARE complex binds three SNAP molecules (25) and that the yeast exocytic SNARE complex binds three copies of the SNAP homologue, Sec17p (45).
NSF is a hexameric ATPase with two ATP-binding sites per subunit. The ATPase activity of the D1 domain is sufficient to support vesicular transport (34,35). The hydrolysis of 6 ATP to ADP amounts to 6ϫ 32 kJ/mol, which should provide sufficient 2 C. Wimmer, unpublished observations.  2-6). Bound proteins were analyzed by Western blot and decorated with the anti-NSF antibody 6E6 (33) or with the anti-Myc antibody 9E10 (50). B, binding of truncated NSF oligomers to ␣SNAP in the presence of neuronal SNAREs. NSF and the the pentamerized N domain of NSF were incubated with SNAREs from bovine brain membrane extract in the presence (lanes 1 and 3) and in the absence of ␣SNAP (lanes 2 and 4). The monomeric N domain of NSF (lane 5) and the pentamerization domain (lane 6) were incubated with neuronal SNAREs and ␣SNAP. SNAREs and bound proteins were immunoprecipitated with the anti-syntaxin antibody HPC-1 and analyzed by immunoblotting with the 5F7 antibody. Aliquots of the input to the immunoprecipitation reactions are shown in lanes 7-12.

FIG. 5. Model of the protein interactions within 20 S particles.
SNAREs have a parallel orientation with their membrane anchors situated on the same side of the rod-like structure. A SNARE complex binds three ␣SNAPs, and this trimerization is sufficient to recruit a NSF hexamer. ␣SNAP binds to the N domain of NSF through its C-terminal half. free energy to dissociate 20 S particles resulting in the release of ␣SNAP and NSF and the disassembly of the SNARE complex. Few thermodynamic data about 20 S particle formation are available. However, it is known that neuronal SNAREs form the most stable SNARE complexes reported so far (56). Their thermal disassembly has been monitored by circular dichroism spectroscopy (57), and the analysis of the melting curve by an Arrhenius plot revealed a dissociation energy of 100 kJ/mol. This result suggests that at least 3 ATP are consumed to dissociate the SNARE complex. Additional ATP might be required for other disassembly coupled rearrangements in the 20 S particle, e.g. occurring at SNAP-SNARE interfaces. The requirement for several ATP to disassemble v-t-SNARE complexes also explains the need for an oligomeric ATPase (NSF), whose individual subunits likely hydrolyze ATP in a concerted manner transducing the generated force via SNAPs to the SNARE complex. Such a concerted ATP hydrolysis is required for NSF function, since NSF hexamers containing one or several subunits defective in ATP hydrolysis are inactive in vesicular transport (34). Further studies are necessary to determine if and which role SNAP trimerization plays in coordinating ATP hydrolysis.
A 20 S particle consists of elements belonging to different symmetry groups. NSF is a protein of 6-fold symmetry (14,16,58,59). The crystal structure of the yeast ␣SNAP homologue Sec17p does not reveal any symmetry within a Sec17p monomer (29). Therefore, the ␣SNAPs within a 20 S particle can at most contribute a 3-fold symmetry. The v-t-SNARE complex consists of four helices forming a parallel bundle. However, all of these helices have a different twist (18) implying that the substrate for the highly symmetrical NSF does not contain obvious repetitive elements. Since the 6-fold symmetry of NSF does not change upon ATP hydrolysis (14), it is tempting to speculate that the NSF-driven dissociation of SNARE complexes induces a transition state with 3-or 6-fold symmetry.
Two hexameric ATPases are known to be involved in the fusion of intracellular membranes, NSF and p97. NSF is important for the heterotypic fusion of transport vesicles with their target membrane along the exocytic pathway and in its retrograde direction (60). p97 is involved in the reformation of Golgi cisternae by catalyzing the homotypic fusion of mitotic Golgi fragments (61). Both proteins are members of the family of ATPases associated with diverse cellular activities (AAA proteins) (for review see Refs. 59 and 62). The activities of this family include protein degradation as metalloproteases or as part of the 26 S proteasome, peroxisome function, cell cycle regulation, and membrane fusion. NSF is conserved from yeast (Sec18p) to mammals (20), whereas p97 is conserved from halobacteria to mammals (59), suggesting that p97 is the more ancient protein presumably involved in cytokinesis and organelle biogenesis (62,63). Both proteins are structurally related, forming rings of 6-fold symmetry and of similar dimensions (14,16,58,59). Both NSF and p97 interact with adaptor proteins to promote membrane fusion. NSF binds to ␣SNAP and p97 binds to p47 in a stoichiometric ratio of one hexameric ATPase to 3 molecules of accessory protein (this study and see Ref. 58). Despite these similarities, NSF and p97 show some significant differences concerning their interaction with target proteins. ␣SNAP trimerization is mediated by its binding to v-t-SNARE complexes (25) (Fig. 4). In contrast, p47 by itself forms trimers, and no additional proteins are required for p97 binding (58). Furthermore, p97⅐p47⅐syntaxin5 complexes appear to have "two feet," probably representing SNAREs emanating from the p97⅐p47 complex in the presence of Mg/ATP as observed by electron microscopy (61). Under similar conditions NSF⅐SNAP⅐v-t-SNARE complexes would dissociate. To what extent these differences represent different reaction mechanisms remains to be determined.