Three αSNAP and 10 ATP Molecules Are Used in SNARE Complex Disassembly by N-ethylmaleimide-sensitive Factor (NSF)*

Background: The ATPase NSF works with the adaptor protein αSNAP to disassemble SNARE protein complexes involved in membrane fusion. Results: Kinetic assays demonstrate that three αSNAP and 10 ATP molecules are required to disassemble a SNARE complex. Conclusion: The energy requirements and functional stoichiometry of αSNAP in SNARE complex disassembly are established. Significance: The results suggest models of NSF action. The fusion of intracellular membranes is driven by the formation of a highly stable four-helix bundle of SNARE proteins embedded in the vesicle and target membranes. N-Ethylmaleimide sensitive factor recycles SNAREs after fusion by binding to the SNARE complex through an adaptor protein, αSNAP, and using the energy of ATP hydrolysis to disassemble the complex. Although only a single molecule of αSNAP binds to a soluble form of the SNARE complex, we find that three molecules of αSNAP are used for SNARE complex disassembly. We describe an engineered αSNAP trimer that supports more efficient SNARE complex disassembly than monomeric αSNAP. Using the trimerized αSNAP, we find that N-ethylmaleimide-sensitive factor hydrolyzes 10 ATP molecules on average to disassemble a single SNARE complex.

The eukaryotic cell moves cargo between membrane-bound compartments using vesicles that bud from one membrane and move to, dock, and finally fuse with a target membrane to effect delivery of luminal cargo and integral membrane components. Soluble NSF 5 attachment protein receptors (SNAREs) are membrane-associated proteins present on both vesicle (v-SNARE) and target (t-SNARE) membranes that are central to the process of membrane fusion (1,2). After a vesicle docks with its target membrane, the cytoplasmic regions of SNAREs associate into a coiled-coil of four ␣-helices, designated the SNARE complex (SC), that is thought to provide the free energy required for bilayer fusion (3). Force measurements of SNARE complex disassembly, using the neuronal t-SNAREs Syx1a (syntaxin 1a) and SNAP-25 (synaptosome-associated protein of 25 kDa), and the v-SNARE VAMP2 (vesicle-associated membrane protein 2)/synaptobrevin lacking their transmembrane anchors, indicate that the complex is stabilized by ϳ39 kcal mol Ϫ1 relative to the free SNARE proteins (4).
After bilayer fusion, N-ethylmaleimide-sensitive factor (NSF; EC: 3.6.4.6) disassembles the SNARE complex in order to recycle the SNARE proteins for further rounds of fusion (5)(6)(7)(8) (Fig.  1A). NSF utilizes the energy from ATP hydrolysis to dissociate the SNARE complex in the postfusion membrane, and it also disassembles potentially non-productive t-SNARE complexes to "prime" them for bilayer fusion (9). NSF is a hexameric ATPase that is a member of the protein family of ATPases associated with various cellular activities (AAAϩ) (10). These enzymes are generally involved in unfolding, disassembly, or remodeling of proteins and their complexes, and are typically pentamers or hexamers in their active state. The NSF monomer comprises a substrate-binding N-terminal domain (N-domain), followed by two AAAϩ ATPase domains, designated D1 and D2. Experiments in which either D1 or D2 is deleted, and others in which key catalytic residues are mutated in the active sites, suggest that D1 is the more active ATPase and indicate that its activity is essential for SNARE complex disassembly. Although D2 has little or no ATPase activity as an isolated domain, ATP binding to D2 is essential for hexamer formation and SNARE disassembly (11).
NSF does not directly disassemble SNARE complexes, but instead requires adaptor proteins known as soluble NSF attachment proteins (SNAPs) that bind to both the SNARE complex and NSF (Fig. 1B). ␣SNAP binds to NSF in an ATP-dependent manner and accelerates the ATPase activity of NSF (12)(13)(14). ␣SNAP is ubiquitously expressed and binds to all ternary SNARE complexes as well as binary complexes containing only t-SNAREs (15). The crystal structure of the yeast ␣SNAP homolog shows that the protein is an ␣-helical solenoid that is capped by a C-terminal globular helical domain (16). Mutations of positively charged residues along the solenoid disrupt bind-ing to the SNARE complex (17), and electron microscopy shows that binding of ␣SNAP to the rodlike SNARE complex increases the width but not the length of the rod, suggesting that the ␣SNAP solenoid binds along the length of the SNARE complex (18) (Fig. 1A). In the absence of NSF, ␣SNAP binds to the neuronal SNARE complex lacking transmembrane anchors in a 1:1 stoichiometry (15).
The NSF⅐␣SNAP⅐SC assembly, also known as the "20 S complex" (Fig. 1A), has been studied biochemically and structurally. The molecular mass of the 20 S complex obtained by electron microscopy and quantitative amino acid analysis of NSF and ␣SNAP bands isolated by SDS-PAGE of purified 20 S complex suggested the presence of three ␣SNAP molecules (19). Cryoelectron microscopy (cryo-EM) reconstruction of the 20 S complex is also consistent with the presence of three ␣SNAP molecules bound to the SNARE complex (20,21). In qualitative pull-down assays, a trimeric ␣SNAP produced by adding an N-terminal coiled-coil domain was shown to bind NSF in the absence of SNARE complex, whereas monomeric ␣SNAP bound NSF only in the presence of the SNARE complex (19), but the effect of trimerization on NSF ATPase or SNARE complex disassembly activities was not tested. The NSF N-domain interacts with the C-terminal domain of ␣SNAP (19,22), and ␣SNAP contains a membrane-associated loop near its N terminus that probably orients the C terminus of ␣SNAP toward the cytosol (23) (Fig. 1A).
Here we have addressed two key mechanistic aspects of NSFmediated SNARE disassembly. First, understanding the role of ␣SNAP as the adaptor between NSF and the SNARE complex has been complicated by the lack of a clear understanding of the functional stoichiometry of the 20 S complex. Second, the number of ATPs needed to disassemble a SNARE complex, which sets boundaries on potential models and efficiencies of action, is not known. We addressed both of these unknowns using equilibrium binding and kinetic analysis of SNARE disassembly and ATPase activity. Although only one ␣SNAP appears to bind to the SNARE complex in solution (15), we determined that three molecules of ␣SNAP are required for full activity of NSF on its substrate. We describe an engineered ␣SNAP trimer that enhances the dissociation of the SNARE complex by NSF, beyond that expected by simply providing multiple copies of ␣SNAP in a single molecule. Using pre-steady-state kinetics with the ␣SNAP trimer, we found that 10 ATP molecules are required to disassemble a SNARE complex.

EXPERIMENTAL PROCEDURES
Molecular Biology-DNA sequences encoding the proteins of interest were cloned into plasmids for expression in Escherichia coli. NSF from Cricetulus longicaudatus with an N-terminal His 6 tag was cloned into a pQE9 plasmid and transformed into E. coli M15 cells containing the pREP4 repressor plasmid (Qiagen Inc., Valencia, CA). The DNA sequence for Bos taurus ␣SNAP was cloned into a pGEX vector, yielding a construct with a cleavable N-terminal glutathione S-transferase (GST) tag.
Briefly, assembly primers 1-4 were used in a PCR with Platinum Taq (Invitrogen) and a T anneal ϭ 50°C. The crude PCR product was further amplified using flanking primers 1 and 2 in a PCR with T anneal ϭ 51°C. After cleanup, the assembly PCR target was utilized in a modified site-directed mutagenesis reaction with GST-␣SNAP in a pGEX vector to insert the trimerization domain in frame at the 5Ј-end of the ␣SNAP sequence.
GST-␣SNAP was expressed from a pGEX vector in E. coli BL21 RIL CodonPlus cells and purified via GST-Sepharose (Sigma-Aldrich) affinity chromatography. The GST tag was removed on column using thrombin (Sigma-Aldrich), and the cleaved protein was purified by size exclusion chromatography (S200 26/60) and anion exchange (MonoQ) (23). Fractions containing pure ␣SNAP were pooled and concentrated using Ultracel 10K centrifugal filters. Concentrated ␣SNAP was flash-frozen in liquid nitrogen and stored at Ϫ80°C. ␣SNAP containing the "foldon" domain was purified as a cleavable GST fusion protein in the same method as for unmodified ␣SNAP.
The syntaxin 1a SNARE domain (residues 188 -267), a SNAP-25 mutant in which the four cysteines normally palmitoylated were mutated to alanine, and the S79C mutant of VAMP2 lacking the C-terminal transmembrane anchor were expressed in E. coli BL21 RIL CodonPlus cells and purified as described previously (28). The SC was formed by incubating these individual SNARE proteins overnight in 20 mM Tris-HCl, pH 7.4, 100 mM NaCl, and 1 mM dithiothreitol, followed by purification by anion exchange chromatography (MonoQ) (17,29).
Purification of Fluorescently Labeled VAMP2 and SNARE Complex-Purified VAMP2(S79C) was labeled with Alexa Fluor 488-maleimide (Invitrogen) according to the manufacturer's directions. This labeled VAMP2 (VAMP2-A488) was separated from unlabeled VAMP2 by cation exchange chromatography (MonoS). Fractions containing pure VAMP2-A488 were pooled, concentrated, and flash-frozen in liquid nitrogen prior to storage at Ϫ80°C. The Alexa Fluor 488-labeled SNARE complex was assembled from purified syntaxin 1a SNARE domain, SNAP-25, and VAMP2-A488, followed by purification from free SNARE proteins as described previously (29). In some of the experiments measuring binding to ␣SNAP or ␣SNAP 3 , the purified SNARE complex was further purified using size exclusion chromatography. However, this extra purification step produced no significant difference in these binding assays and was not included in the purification of SNARE complexes used in the disassembly or ATP hydrolysis assays.
Nucleotide Binding Assay-Radioactively labeled nucleotides, [␣-32 P]ATP (specific activity, 3000 Ci/mmol), [␥-32 P]ATP (SA, 10 Ci/mmol) (PerkinElmer Life Sciences) were incubated with NSF at 4°C or 30°C in binding buffer (50 mM HEPES, pH 7.4, 300 mM potassium chloride, 10% (w/v) glycerol, 12 mM MgCl 2 , 2 mM dithiothreitol). Protein⅐nucleotide complexes were separated from free nucleotide using Zeba spin columns (Thermo Fisher Scientific) equilibrated with binding buffer and spun at 1500 ϫ g for 90 s. NSF concentration was varied at each nucleotide concentration, and the slope of the nucleotide bound versus NSF concentration was used to determine the amount of nucleotide bound specifically to NSF at each concentration.
ATP Hydrolysis Assay-All ATP hydrolysis assays were performed at 30°C. The reaction mixture contained 50 mM NaHEPES, pH 7.4, 120 mM potassium glutamate, 20 mM potassium acetate, and varying concentrations of [␥-32 P]ATP (10 Ci/mmol) diluted in unlabeled ATP to final concentrations of 300 -500 M. The reaction mixture was prepared on ice and incubated at 30°C for 5 min prior to the addition of Mg 2ϩ (MgCl 2 or Mg(CH 3 COO) 2 ) or NSF to start the reaction. Reaction aliquots (25-50 l) were removed and quenched using trichloroacetic acid, and released 32 P i was separated by molybdate-P i extraction and measured by liquid scintillation counting (30). ATP hydrolysis measurements used in steady-state analysis were confined to the first 4 -8 min of the reaction to obtain linear initial rates (of the 40 independent experiments, 35 had R 2 Ն 0.95, and 5 had 0.85 Ͻ R 2 Ͻ 0.95) free from substrate depletion and product inhibition. Linear rates were measured in duplicate, and the uncertainty reported the error in ATP hydrolysis as propagated from each of the regression lines establishing linear rates.
Fluorescence Anisotropy Measurements-Fluorescence anisotropy was used to measure the formation and dissolution of protein complexes containing VAMP2 labeled with the fluorophore Alexa Fluor 488 (VAMP2-A488). All measurements were performed in black, flat-bottomed 96-well microplates (Costar 3915, Corning Inc.), and readings were taken in a Synergy 4 Hybrid microplate reader (BioTek, Winooski, VT). Microplates were prewarmed to 30°C for 10 min prior to the start of the experiment. Time courses of 60 -120 min were recorded at 30°C, with the addition of Mg 2ϩ to start the reaction. Fluorescence anisotropy r was calculated as follows, where the subscripts indicate intensity measurements parallel and perpendicular to the incident polarization. Once thawed, A488-labeled SNARE complex gives reproducible anisotropy readings for 4 -5 days when stored at 4°C. All experiments where anisotropy values were directly compared were performed within 24 h.
␣SNAP⅐SNARE Complex Binding Assay-SC containing VAMP2-A488 at 25-1000 nM was incubated with varying concentrations of ␣SNAP or ␣SNAP 3 (25-5000 nM) in 50 mM NaHEPES, pH 7.4, 120 mM potassium glutamate, 20 mM potassium acetate at room temperature prior to incubation at 30°C. Complex formation at equilibrium was measured by an increase in anisotropy over a 60-min time course. Bovine serum albumin (Sigma-Aldrich) was added in place of ␣SNAP in parallel experiments as a control for nonspecific binding. ATP and Mg 2ϩ were found to have no effect on ␣SNAP-SC binding. Equilibrium binding experiments were performed in triplicate, and the error bars represent S.E. SNARE Complex Disassembly Assay-SC (25-1000 nM) containing VAMP2-A488 was incubated in 50 mM NaHEPES, pH 7.4, 120 mM potassium glutamate, 20 mM potassium acetate with ␣SNAP, NSF, and 400 -500 M ATP. The plate was prewarmed to 30°C for 10 min prior to the addition of Mg 2ϩ at a final concentration of 5 mM to start the disassembly reaction. Disassembly of the SC by NSF (1-10 nM) results in a decrease in fluorescence anisotropy that reaches a minimum value identical to free VAMP-A488. The assembled neuronal SNARE complex is resistant to SDS and runs as a single band on SDS-PAGE (31), so disassembly of SNARE complex was confirmed by SDS-PAGE. SC disassembly was monitored over a 60 -120-min time course. Initial rates of disassembly were determined from the initial linear phase (R 2 Ͼ 0.90) of the reaction, when higher SC concentrations were used. Controls lacking NSF, ATP, or Mg 2ϩ showed no disassembly. The SC concentration was maintained below 1 M to minimize the contribution of reassembly and to keep within the detection range of the fluorescence plate reader. Pre-steady-state reactions included an excess of NSF (2 M) and low concentrations of ␣SNAP (50 -100 nM). Steadystate measurements to determine K m and k cat for SC disassembly used NSF at 2 nM and SC at 50 -250 nM, with ␣SNAP varied from 25 to 2000 nM. SC concentrations above these values in steady-state conditions resulted in significant SC reassembly and therefore were avoided. Disassembly rate was plotted versus ␣SNAP concentration, and the data were fit to a hyperbola using the equation ). Steady-state SC disassembly experiments were performed in triplicate, and error bars depict S.E.; the titration disassembly experiments were performed in duplicate. There was no difference in k cat or K m ␣SNAP over the range of SC concentrations examined. Kinetic Modeling-KinTek Explorer was used to simulate the effect of different ␣SNAP/SC stoichiometries on the kinetics of SC disassembly by NSF with monomeric ␣SNAP, using the schemes shown in Fig. 1, B and C. Table 1 summarizes the K d values used in the model shown in Fig. 1C. The K d values for dissociation of the ␣SNAP⅐SC, measured in this work (K 1-4 ), were taken to be independent of the number of ␣SNAP molecules bound to the SC. Likewise, the K d values for dissociation of the (␣SNAP) n ⅐SC from NSF were set to the value of K m for ␣SNAP stimulation of SC disassembly and were assumed to be the same regardless of the value of n. The K d for the binding of ␣SNAP to NSF was taken as the K m for ␣SNAP stimulation of the NSF ATPase activity and was assumed to be independent of the number of ␣SNAP molecules bound to NSF. The remaining K d values, which describe the formation of intermediate ␣SNAP i ⅐SC⅐NSF complexes, were calculated from the above constants via thermodynamic cycles.
The forward rate k on was set at 100 nM Ϫ1 ⅐min Ϫ1 , and the reverse rate constant was calculated as k on ϫ K d . Modeling with these rate constants scaled up or down 1000-fold produced the same general shape of the curve (data not shown). To compensate for the fact that the relative but not absolute values of the rate constants were known, the reaction rates were normalized to the maximum rate of reaction for that model. Because Kin-Tek Explorer only permits the modeling of elementary reaction steps, we were unable to model the dissociation of the (␣SNAP) n ⅐SC⅐NSF complex into free NSF, ␣SNAP, Syx1a, SNAP-25, and VAMP2 as a single step. Thus, the catalytic step was defined as the conversion of the (␣SNAP) n ⅐SC⅐NSF complex into an (␣SNAP) n ⅐product⅐NSF, with a rate of formation equal to k cat . The (␣SNAP) n ⅐product⅐NSF complex then disso-ciated into (␣SNAP) n ⅐product ϩ NSF and then into n ␣SNAP ϩ product (in n elementary reaction steps). The rate constants for these subsequent dissociations were set at 10 10 min Ϫ1 to make them virtually instantaneous.
Disassembly reaction rates were calculated over the ␣SNAP concentration range 0 -7100 nM in 100 nM increments, with SC and NSF concentrations fixed at their experimental concentrations of 1000 nM. In all cases, the time for the modeling was fixed at 0.1 min (6 s) to ensure that the product versus time curves used to obtain the rates remained in the linear range.
The same procedures were followed for modeling trimerized ␣SNAP (␣SNAP 3 ) as for monomeric ␣SNAP, except that the reaction rates were calculated in the range 0 -1710 nM ␣SNAP 3 in 30 nM increments. The K d values used for modeling SC dissociation by trimeric ␣SNAP 3 (Fig. 1D) are summarized in Table 2. ␣SNAP 3 was treated as a single molecular species, so the experimental concentrations and K d values, which were calculated on a per monomer basis, were divided by 3 for the modeling. The SC and NSF concentrations were fixed at their experimental concentrations of 280 nM.
Calculation of 7 S Complex Concentration-The predicted concentration of ␣SNAP 3 ⅐SC complex (7 S complex) in the presteady-state experiment was calculated from the input concentrations of SC and ␣SNAP 3 using the measured dissociation constant K d ␣SNAP⅐SC ϭ 100 nM and the general form of the binding equation, where [a] and [b] are the input protein concentrations of SC and ␣SNAP 3.

RESULTS
Multiple ␣SNAP Molecules Are Required for SNARE Complex Disassembly-Recent biophysical data show that ␣SNAP forms a 1:1 complex with a variety of SNARE complexes lacking membrane anchors (15) (Fig. 2A), a surprising result given prior biochemical and structural data indicating that the 20 S complex contains three copies of ␣SNAP (17)(18)(19)(20)(21). We incubated a minimal neuronal SC, consisting of the SNARE domain of syntaxin 1a, SNAP-25, and the S79C mutant of VAMP2 labeled with Alexa Fluor 488, with ␣SNAP. Binding was measured by the increase in fluorescence anisotropy due to the decreased tumbling time of the ␣SNAP⅐SC relative to that of the SC (Fig.  2A). The ␣SNAP⅐SC equilibrium binding data can be fit to a single-site, 1:1 binding model with K d ␣SNAP ϭ 450 Ϯ 50 nM ( Fig.  2A). This value agrees closely with the reported value of K d ϭ 470 nM by biolayer interferometry (15). A model in which three ␣SNAP molecules bind to a single SC with three distinct binding sites and binding affinities did not provide a significantly better fit to the binding data despite the additional fitting parameters. These data are consistent with the observation that ␣SNAP binds to the SNARE complex lacking transmembrane FIGURE 1. NSF uses the energy of ATP hydrolysis and the ␣SNAP adaptor to disassemble the SNARE complex. A, NSF binds to the membrane-bound SC via ␣SNAP, which is probably membrane-associated. Using the energy from ATP hydrolysis, the SC is disassembled into its component proteins, and NSF and ␣SNAP are available for subsequent rounds of disassembly. B, simplified model for the dissociation of the SC by NSF bound to 3 ␣SNAP, in which either ␣SNAP binds SC before NSF (red pathway) or SC binds to ␣SNAP⅐NSF to form the 20 S complex (blue pathway in B and C). C, kinetic modeling scheme for the association of n ␣SNAP molecules with SC and disassembly of the SC by NSF. The red branch represents binding of n ␣SNAP molecules to the SC to form the complete NSF substrate before interaction with NSF; the blue branch corresponds to n ␣SNAP molecules binding to NSF before association with the SC. Potential intermediate, non-productive assemblies lacking sufficient ␣SNAP molecules to support SC disassembly are also shown. For a given model that uses n ␣SNAP molecules, the 20 S complex will be (␣SNAP) n ⅐SC⅐NSF, which disassembles into n ␣SNAPs, the individual SNARE proteins, and free NSF (curved green arrows). D, kinetic modeling scheme for the association of n ␣SNAP 3 molecules with SC and disassembly of the SC by NSF. The red branch represents binding of n ␣SNAP 3 molecules to the SC to form the complete NSF substrate before interaction with NSF; the blue branch corresponds to n ␣SNAP 3 molecules binding to NSF before association with the SC. Potential intermediate, non-productive assemblies lacking sufficient ␣SNAP 3 molecules to support SC disassembly are also shown. For a given model that uses n ␣SNAP 3 molecules, the 20 S complex will be (␣SNAP 3 ) n ⅐SC⅐NSF, which disassembles into n ␣SNAP 3 s, the individual SNARE proteins, and free NSF (curved green arrows).
anchors with a 1:1 stoichiometry (15) but do not in themselves rule out an ␣SNAP/SC stoichiometry of Ͼ1:1.
We next assessed the functional stoichiometry of ␣SNAP in SC disassembly by NSF by examining the dependence of the rate of SC disassembly on ␣SNAP concentration. We found, as has been noted by others (32), that the presence of Cl Ϫ diminishes NSF activity. Although NSF activity can be measured under very low ionic strength conditions to circumvent this problem (32), we found that a buffer containing carboxylates as the predominant anions, which more closely mimics physiological conditions, supports NSF activity (23). Disassembly of the SC in this buffer was measured by the decrease in anisotropy of labeled VAMP2 (VAMP2-A488) under steady-state conditions (Fig. 2, B and C). Analysis of initial rates of SC disassembly, measured by the dissociation of VAMP2-A488 from the SC as a function of ␣SNAP concentration, gave K m ␣SNAP ϭ 1.2 Ϯ 0.2 M and k cat disassembly ϭ 1.9 Ϯ 0.1 min Ϫ1 at 30°C (Fig. 2C and Table 3). We also assessed the association of ␣SNAP with NSF under these conditions by determining its K m for NSF-mediated ATP hydrolysis, which was 4.7 M (Fig. 2D and Table 1).
To determine the number of ␣SNAPs involved in SNARE complex disassembly, we performed a titration experiment by fixing NSF and SNARE complex at high concentrations and measuring initial SNARE disassembly rates over a range of ␣SNAP concentrations from substoichiometric to excess (Fig.  2E). At a given NSF concentration, the disassembly rate will be proportional to the amount of the ␣SNAP⅐SC, which is the NSF substrate. Considering the red pathway in Fig. 1B, if binding of ␣SNAP to the SNARE complex is faster than SNARE complex disassembly (i.e. the equilibrium for ␣SNAP⅐SNARE complex formation is established faster than the disassembly reaction), then with a fixed concentration of SNARE complex, the rate of disassembly will be a function of the concentration of input ␣SNAP. In the alternative case that ␣SNAP binds first to NSF (blue pathway in Fig. 1B), the rate of disassembly will also be a function of the concentration of input ␣SNAP if the equilibrium for SNARE complex binding to the SNAP⅐NSF complex is faster than disassembly. Thus, the number of ␣SNAP molecules required for disassembly can be obtained by comparing the observed disassembly rate with that calculated from models using the equilibrium binding constants for formation of ␣SNAP⅐SNARE and ␣SNAP⅐NSF complexes, K d ␣SNAP , and the input concentrations of ␣SNAP and SNARE complex for different stoichiometries of the ␣SNAP⅐SNARE complex interaction.
To assess whether the time for equilibration of the ␣SNAP-SC binding reaction is faster than SC disassembly (i.e. the condition needed for the titration experiment), we followed the association of ␣SNAP with the SC by fluorescence anisotropy (Fig. 2F). The t1 ⁄ 2 values estimated from progress curves of disassembly and ␣SNAP-SC association measured at the same SC concentration indicate that the latter is ϳ40 times faster (Fig. 2, B and  F). Therefore, the disassembly rate should be a function of the ␣SNAP⅐SC concentration based on the amount of input ␣SNAP and SC.
The stoichiometry of ␣SNAP in the (␣SNAP) n ⅐SC⅐NSF complex was determined by modeling the reaction with n ϭ 1, 2, 3, and 4 (red pathway in Fig. 1C) and comparing the predictions from each of these models to the experimental data (Fig. 2E). The experimental data do not follow a model with 1:1 stoichiometry and are best described by a 3:1 ␣SNAP/SC stoichiometry.
These models assumed that ␣SNAP binds SC before NSF (red pathway in Fig. 1, B and C). However, ␣SNAP stimulates NSF ATPase activity in the absence of SNARE complex (Fig.  2D), indicating that it can bind to NSF by itself. Therefore, it was also necessary to consider a second pathway in which SC binds to ␣SNAP⅐NSF to form the 20 S complex (blue pathway in Fig. 1, B and C); as we did for binding to free ␣SNAP, we assumed that the binding of SC to NSF-bound ␣SNAP reaches equilibrium faster than the rate of 20 S disassembly. Formally, there can be intermediates of the form ␣SNAP i ⅐SC⅐NSF (i Ͻ n) (Fig. 1C). We generated a model with such intermediates in the 3:1 ␣SNAP/SC case, which shows a decline in rate at high ␣SNAP concentrations due to sequestration of ␣SNAP in (␣SNAP) n ⅐SC and (␣SNAP) n ⅐NSF complexes rather than (␣SNAP) n ⅐SC⅐NSF complexes by mass action. The absence of a decline in the observed disassembly rate at high ␣SNAP concentrations suggests that there is no significant accumulation of these intermediate species, although the predicted decline could be due to misestimation of the K d values describing their formation. Importantly, the data cannot distinguish whether ␣SNAP binds NSF or SC first (i.e. blue versus red pathways in Fig. 1B).
Engineering a Trimeric ␣SNAP-Prior studies have shown that fusion of a trimeric coiled-coil sequence to the N terminus of ␣SNAP qualitatively enhances binding to NSF (19). Given the results of the titration experiment suggesting the involvement of three ␣SNAP molecules in SNARE complex disassembly, we hypothesized that an ␣SNAP homotrimer would support more efficient disassembly as well. We fused the bacteriophage T4 fibritin "foldon" domain, which assembles into a very stable ␤-propeller with an estimated K d in the low picomolar range (24,25), to the N terminus of ␣SNAP. Structural data indicate that ␣SNAP binds to the outside of the SNARE coiled-coil complex (21), so a (Gly-Ser) 6 linker was inserted between the foldon and the N terminus of ␣SNAP to allow each ␣SNAP in the trimer to simultaneously contact the SNARE complex. In this orientation, the foldon mimics tethering of individual ␣SNAP molecules to the membrane through the N-terminal loop (Fig.  3A) (23). Analysis of the foldon-␣SNAP fusion protein by size exclusion chromatography (Fig. 3B) and characterization by native gel electrophoresis (data not shown) confirmed that it assembles into a stable trimer with no apparent dissociation or aggregation into higher oligomers. We refer to this fusion protein as ␣SNAP 3 . ␣SNAP 3 binds to the SNARE complex with K d ϭ 100 Ϯ 4 nM (Fig. 3C), 4.5-fold more strongly than monomeric ␣SNAP, which is consistent with the presence of three copies of ␣SNAP in the oligomer but provides no indication of any more than a single site on the SNARE complex that is stably occupied with a bound ␣SNAP molecule. Steady-state measurements of SC disassembly in the presence ␣SNAP 3 gave k cat disassembly ϭ 1.6 Ϯ 0.1 min Ϫ1 and K m ␣SNAP ϭ 84 Ϯ 10 nM (Fig. 3D and Table  3). As with the ␣SNAP monomer, we measured the association of ␣SNAP 3 with NSF in the absence of SC by determining its K m for NSF-mediated ATP hydrolysis, which was 390 nM (Fig. 3E and Table 2).
A titration experiment similar to that described above is consistent with ␣SNAP 3 acting unimolecularly in SC disassembly (Fig. 3F). The similarity in k cat disassembly between ␣SNAP and ␣SNAP 3 (Table 3) suggests that the reaction is proceeding through the same rate-limiting step. However, the 14-fold decrease in K m ␣SNAP (Table 3) is significantly larger than the  (Table 2). E, the ATPase activity of NSF was measured as a function of ␣SNAP 3 concentration. The data are fit to the equation, ϭ V max /(1 ϩ (K m /[␣SNAP])) (solid black line) to give K m ␣SNAP3 for NSF ATPase activity. F, SC disassembly rates (y axis) were measured with SC and NSF at fixed concentrations (280 nM) over a range of ␣SNAP 3 concentrations (320 -1710 nM). The lines correspond to models with different ␣SNAP 3 stoichiometry, as indicated in the key. The steeper rise for the branched curve as compared with the unbranched curve results from the normalization of the calculated data to the maximum rate. Because the branched curve has a lower absolute maximum rate than the unbranched curve, it reaches its maximum more quickly. Descriptions of the models used can be found under "Experimental Procedures." Error bars, S.E.
3-fold decrease that would be expected from the presence of three copies of ␣SNAP in the trimer. The finding that monomeric ␣SNAP is required at a greater than 1:1 stoichiometry for SC disassembly, whereas trimeric ␣SNAP appears to act in a unimolecular manner, provides quantitative evidence that more than a single ␣SNAP participates in SC disassembly.
Nucleotide Binding and Hydrolysis Activities-The binding of nucleotides to NSF was measured by incubating purified NSF with [␣-32 P]ATP at 4°C or 30°C and measuring the radioactivity of the resulting protein⅐nucleotide complex (Fig. 4). The data indicate that 11.5 Ϯ 0.8 molecules of ATP bound to NSF with a K d app of 20 Ϯ 4.8 M. When the experiment was repeated with [␥-32 P]ATP, no radioactivity was detected at 30°C. These data indicate that NSF binds to 12 molecules of ATP, which are hydrolyzed to ADP with the release of inorganic phosphate to produce an ADP-bound form at steady state, and suggest that all 12 nucleotide-binding sites are active.
We examined the effects of ␣SNAP and ␣SNAP 3 on the ATP hydrolysis activity of NSF by steady-state kinetics. NSF alone hydrolyzes ATP with a k cat ATP ϭ 5.7 Ϯ 0.2 min Ϫ1 and a K m ATP ϭ 20 Ϯ 3.6 M (Table 4). There was no evidence of cooperative or distinct ATPase activities in the hydrolysis kinetics, despite the presence of 12 ATPase sites in the NSF hexamer (i.e. the data are well described by a single K m ATP value). In the presence of ␣SNAP, k cat ATP increases 2-3-fold, consistent with previous reports (9,(11)(12)(13) (Table 4). ␣SNAP 3 increases the NSF ATPase rate ϳ8-fold relative to NSF alone (Table 4). In the presence of ␣SNAP 3 ⅐SC, NSF ATPase activity is stimulated fur-ther relative to ␣SNAP 3 , and is 13-fold higher than that of NSF alone (Table 4).
Direct Measurement of ATP Utilization in SNARE Complex Disassembly-To determine directly the amount of ATP needed to disassemble the neuronal SNARE complex, we sought to measure SC disassembly under pre-steady-state conditions (i.e. with NSF in excess over its ␣SNAP⅐SC substrate). Under these conditions, the observed rate is that of the actual complex disassembly independent of enzyme concentration. By comparing this rate with the rate of ATP hydrolysis in the 20 S complex under the same conditions, the number of ATPs consumed in a single disassembly reaction can be obtained.
NSF is well behaved up to concentrations of about 5 M in the buffers used in our assays. To have NSF in sufficient excess for pre-steady-state measurements, we chose to have the ␣SNAP⅐SC substrate at concentrations in the range 10 -100 nM. A low concentration of SC (Ͻ1 M) is also desirable to avoid reassociation of SNAREs during the experiment. However, the affinity of ␣SNAP for the SC (K d ϭ 450 nM) indicated that it would be difficult to prepare a significant amount of ␣SNAP⅐SC (i.e. the NSF substrate) at these concentrations. We therefore exploited ␣SNAP 3 , which binds sufficiently strongly to carry out presteady-state measurements of SC disassembly. Moreover, given the apparent need for more than one ␣SNAP in SC disassembly, using the unimolecular ␣SNAP 3 (Fig. 3E) avoids the potential complication of having only one or two ␣SNAP molecules bound in the NSF⅐␣SNAP⅐SC complex. SC was combined with a limiting concentration of ␣SNAP 3 and an excess of NSF (Fig.  5A). The reaction was monitored for SNARE disassembly and for ATP hydrolysis, and the initial rates were determined by a linear fit to the early time points in the SC disassembly reaction (Fig. 5, B and C).
With NSF in excess, three different NSF-containing species contribute to the observed ATP hydrolysis (Fig. 5A Fig. 5A, iii). Because NSF is present in excess (ϳ20-fold) the amount of NSF involved in disassembly represents a small fraction of the total enzyme present. The hydrolysis mediated by the 20 S complex is of interest, because only this species leads to SNARE complex disassembly. To eliminate the contribution of NSF⅐␣SNAP 3 to the observed ATP hydrolysis, ATPase activity was determined as a function of increasing concentrations of SC, thereby saturating and removing free ␣SNAP 3. At nearly saturating SNARE complex concentrations (Table 5 and Fig. 5, D and E), the observed ATP hydrolysis from the remaining ϳ5% NSF⅐␣SNAP 3 is negligible, because the ATPase activity of this species is only about

activation of NSF-mediated ATP hydrolysis
NSF ATP hydrolysis rates were measured as a function of increasing ATP concentration (0 -600 M) and fit to the Michaelis-Menten equation using nonlinear regression with ͓NSF hexamer͔ ϭ 4 -8 nM, ͓␣SNAP͔ ϭ 4.5 M, ͓␣SNAP 3 ͔ ϭ 1.5 M, and ͓␣SNAP 3 ⅐SC͔ ϭ 2 M. At these concentrations, which were limited by protein solubility, ␣SNAP and ␣SNAP 3 in the absence of SC do not fully saturate NSF. Specifically, there is predicted to be 50% free NSF and 50% NSF⅐␣SNAP in the experiment for NSF⅐␣SNAP and 19% free NSF and 81% NSF⅐␣SNAP 3 in the experiment for NSF⅐␣SNAP 3 (calculated using K m for ␣SNAP and ␣SNAP 3 stimulation of ATPase activity; Fig.  2, D and E, respectively), and the k cat values have been corrected for these fractions bound using the ATP hydrolysis rate for NSF alone for the ␣SNAP-free fraction in this  half of the NSF⅐␣SNAP 3 ⅐SC complex ( Table 4). The contribution of free NSF to ATP hydrolysis was determined in parallel control reactions in which ␣SNAP 3 was omitted (Fig. 5C); the SNARE complex does not interact with NSF directly and therefore does not affect hydrolysis activity. The rate of ATP hydrolysis in the 20 S complex was then obtained by subtracting the rate of ATP hydrolysis of free NSF (the ␣SNAP-free control; Fig. 5C) from the rate of ATP hydrolysis in the SC disassembly reaction measured at nearly saturating SC concentrations ( Table 5); note that the amount of NSF involved in disassembly is negligible compared with total enzyme concentration (

DISCUSSION
The SNARE complex that is disassembled by NSF and ␣SNAP is a parallel four-helix bundle composed of four distinct helices. Recent biophysical analysis has shown that ␣SNAP binds to purified, non-membrane-anchored SNARE complexes in a 1:1 stoichiometry (15). Our ␣SNAP⅐SC binding data (Fig.  2B) are consistent with these findings while not ruling out the presence of much weaker binding sites. Indeed, all published structural analyses, including chemical cross-linking, mass spectrometry, and electron microscopy, suggest that three copies of ␣SNAP are present in the 20 S complex (17)(18)(19)(20)(21). It is possible that, although there is only one detectable binding site for ␣SNAP on the SC, there are much weaker binding sites that are not detected in the assays used here or used previously (15). In any case, the kinetic titration analyses presented here (Figs. 2E and 3E) represent the first functional analysis of ␣SNAP stoichiometry in NSF-mediated SNARE complex disassembly and suggest that multiple ␣SNAP molecules are required (Fig.  2E). Consistent with this analysis, a trimeric ␣SNAP acts unimolecularly in disassembly (Fig. 3F).
The observation that only a single ␣SNAP appears to bind to the SNARE complex in solution suggests that there is one preferred binding site (Fig. 6). It is possible that other binding sites are present but too weak to be detected in solution state binding assays. ␣SNAP binds to different ternary SNARE complexes as well as binary t-SNARE complexes, which may indicate that the preferred site is a surface created by the association of the t-SNARE components (15). Although cryo-EM analysis of the 20 S complex led to a model in which each of the three ␣SNAP molecules contacts the SNARE complex, the analysis required 3-fold averaging of the images and therefore cannot discern the inherent asymmetry of the SNARE complex or determine potential differences in its interactions with ␣SNAP (21).
Our design of the trimerized ␣SNAP 3 was inspired by the observation that a 20-fold lower ␣SNAP concentration was needed to support the NSF-mediated disassembly of liposomeanchored SNARE complexes versus soluble complexes, such as those used in this paper (23). In that work, it was suggested that the enhanced effect of lipid-associated ␣SNAP arises from an increase in local concentration due to restriction to the twodimensional bilayer, or perhaps that the lipid association produces a conformational change in ␣SNAP. This observation raises the possibility that fewer molecules of ␣SNAP might be needed for SC disassembly on membranes (i.e. that the functional stoichiometry of ␣SNAP in SC disassembly differs from that found here). Nevertheless, as noted above, the 14-fold lower K m ␣SNAP for disassembly mediated by ␣SNAP 3 than by monomeric ␣SNAP strongly argues that more than one ␣SNAP molecule is utilized in disassembly. If the reduced concentration of ␣SNAP required to disassemble lipid-anchored SC were to arise from a change in stoichiometry, then trimerized ␣SNAP 3 would be predicted to produce only a 3-fold change in K m ␣SNAP . Moreover, the similarity in k cat disassembly between ␣SNAP and ␣SNAP 3 (Table 3) suggests that the reaction is proceeding through the same rate-limiting step. Given that the soluble NSF enzyme disassembles both membrane-bound and soluble SNARE complexes, the most parsimonious interpretation of these data is that the effect of the interaction of ␣SNAP with the lipid bilayer is to increase local concentration rather than to fundamentally alter the mechanism of NSF-mediated disassembly of the 20 S complex, in particular the stoichiometry of ␣SNAP in this reaction.
␣SNAP interacts with NSF independent of the SNARE complex, as shown by the enhancement of ATP hydrolysis activity  (Table 4) (12)(13)(14). ATP hydrolysis is further enhanced when the SNARE complex is present, giving a 13-fold increase in hydrolysis versus free NSF. If multiple ␣SNAP molecules are required to bind to and stabilize a conformation of NSF that efficiently hydrolyzes ATP, the SNARE complex may help to organize ␣SNAP and NSF molecules into a productive arrangement, given the suggestion from the cryo-EM data that the N-terminal portion of the SNARE complex interacts with the NSF N-domains in the 20 S complex as well as with ␣SNAP. The enhanced SNARE disassembly and ATPase efficiency of the unimolecular ␣SNAP 3 may be due to the reduced probability of forming non-productive NSF⅐␣SNAP complexes (i.e. those with only one or two ␣SNAPs bound) versus monomeric ␣SNAP. An alternative model is that ␣SNAP serves as a processivity factor that must remain associated with NSF and the SNARE complex during each round of ATP hydrolysis.
The nature of the force transduction needed to disassemble the SNARE complex and the role of multiple ␣SNAP molecules in this process are not known. The ␣SNAP C-terminal domain binds to the NSF N-domain (19,22). The cryo-EM structure of the ADP⅐AlF x -bound 20 S complex indicates that the N-terminal region of ␣SNAP interacts with the C-terminal (membraneproximal) surface of the SNARE complex, and the ␣SNAP C-terminal domain binds to the NSF N-domain, which in turn contacts the SNARE complex (21) (Fig. 6). Structural analysis of isolated NSF reveals that in the ATP-bound state, the NSF N-domains lie above the plane of the D1 ring, where they interact with the ␣SNAP⅐SNARE complex (21,33), whereas in the ADP-bound state, the N-domains lie near the periphery of the D1 ring (Fig. 6, vi). The nucleotide-dependent downward movement of the NSF N-domains may represent the power stroke that pulls ␣SNAP and one or more components of the FIGURE 6. Model for SNARE complex disassembly by NSF. ␣SNAP molecules (blue cylinders) associated with the membrane using hydrophobic loops (red circles) bind the SC (multicolored cylinder) and NSF to form the 20 S complex (i). The primary binding site on the SC for ␣SNAP is shown as a yellow patch, and the gray bar indicates the membrane. Outward and downward movement of the N-domains upon ATP hydrolysis (power stroke) may release ␣SNAP transiently (ii), and rotation of the NSF hexamer positions a different ␣SNAP at the primary binding site (iii). ATP hydrolysis and rotation steps are repeated (iv and v), and disassembled SNARE proteins and ␣SNAP remain membrane-associated (vi). Displacement of ADP for ATP resets the N-domains into an "up" conformation, where a new ␣SNAP⅐SC complex can be disassembled (vii). If ␣SNAP were always bound to both the NSF N-domain and to the SC, then as the SC is unwound and NSF translocates, the NSF N-domain would tilt more and more with respect to the SC, shortening the lever arm and potentially reducing the force that can be applied to the complex. See "Discussion" for details.