Dissecting the N-Ethylmaleimide-sensitive Factor

N-Ethylmaleimide-sensitive factor (NSF) is a homo-hexameric member of the AAA+ (ATPases associated with various cellular activities plus) family. It plays an essential role in most intracellular membrane trafficking through its binding to and disassembly of soluble NSF attachment protein (SNAP) receptor (SNARE) complexes. Each NSF protomer contains an N-terminal domain (NSF-N) and two AAA domains, a catalytic NSF-D1 and a structural NSF-D2. This study presents detailed mutagenesis analyses of NSF-N and NSF-D1, dissecting their roles in ATP hydrolysis, SNAP·SNARE binding, and complex disassembly. Our results show that a positively charged surface on NSF-N, bounded by Arg67 and Lys105, and the conserved residues in the central pore of NSF-D1 (Tyr296 and Gly298) are involved in SNAP·SNARE binding but not basal ATP hydrolysis. Mutagenesis of Sensor 1 (Thr373–Arg375), Sensor 2 (Glu440–Glu442), and Arginine Fingers (Arg385 and Arg388) in NSF-D1 shows that each region plays a discrete role. Sensor 1 is important for basal ATPase activity and nucleotide binding. Sensor 2 plays a role in ATP- and SNAP-dependent SNARE complex binding and disassembly but does so in cis and not through inter-protomer interactions. Arginine Fingers are important for SNAP·SNARE complex-stimulated ATPase activity and complex disassembly. Mutants at these residues have a dominant-negative phenotype in cells, suggesting that Arginine Fingers function in trans via inter-protomer interactions. Taken together, these data establish functional roles for many of the structural elements of the N domain and of the D1 ATP-binding site of NSF.

N-Ethylmaleimide-sensitive factor (NSF) 2 is required for most membrane fusion events in a cell (1,2). Conditional mutations in the yeast orthologue and the neuro-specific form in fly (Sec18p and comatose, respectively) disrupt constitutive secretion (3) or synaptic transmission (4), respectively, and overexpression of a dominant-negative mutant in mammalian cells is cytotoxic (5). NSF is thought to disassemble SNAP receptor (SNARE) complexes so that they can be recycled for subsequent membrane fusion events. SNAREs are the minimal machinery for membrane fusion (6). They form a four-helix SNARE complex that spans the apposing membranes and mediates fusion (7). After fusion, NSF uses the energy from ATP hydrolysis to disassemble spent SNARE complexes for recycling. NSF binds to the SNARE complexes by interacting with an adaptor protein called soluble NSF attachment protein (␣-SNAP) (8). Previous studies (9 -12) have suggested that three ␣-SNAPs coat the length of the SNARE bundle and position NSF at the membrane-distal end of the SNARE complex. This stimulates the ATPase activity of NSF and the subsequent hydrolysis-dependent conformational changes somehow drive unwinding of the SNARE complex. Despite this general outline, little is known about the structural changes that NSF undergoes during its ATP hydrolysis cycle and how the chemical energy is converted into the mechanical energy required for SNARE recycling.
NSF is a member of the AAA ϩ (ATPases associated with various cellular activities plus) family. Each protomer of the homo-hexamer contains one N-terminal domain (NSF-N), followed by two conserved, Walker-type, nucleotide-binding domains (termed NSF-D1 and NSF-D2). NSF-N contains a double -␤ barrel (13,14) and is required for SNAP⅐SNARE binding (15). Positively charged residues on its surface are proposed to interact with the negatively charged C terminus of ␣-SNAP (8,16); however, it is not clear how these residues contribute to SNAP⅐SNARE complex binding. NSF-D1 accounts for the majority of the ATP hydrolysis; NSF-D2 is required for oligomerization (15). Each nucleotide-binding domain is divided into two subdomains as follows: a ␤-sheet core at the base of the nucleotide-binding pocket, and an ␣-helical domain that partially caps the pocket (Fig. 1A) (12). In the hexamer, the nucleotide-binding pocket is located at the interface between adjacent protomers. Several conserved motifs contribute to the nucleotide-binding pocket, including the classical Walker A and B motifs (Fig. 1). The conserved lysines in Walker A boxes are crucial for ATP binding (17). The aspartates in the DEXX sequences of Walker B boxes are thought to coordinate Mg 2ϩ ions, which are required for ATP hydrolysis. The glutamates are proposed to activate water during the hydrolysis reaction. The second region of homology is another sequence motif that is highly conserved in AAA ϩ proteins (12). Sensor 1 is at the N terminus of the second region of homology and often contains a threonine/asparagine pair (18). At its C terminus are two arginine residues termed Arginine Fingers (19). These residues are thought to be critical for nucleotide hydrolysis similar to Arginine Fingers of GTPase-activating proteins (19). The ␣-helical subdomain of the nucleotide-binding domains contains a motif, called Sensor 2, which is composed of residues that are positioned adjacent to the ATP-binding site. This motif often contains a conserved arginine (e.g. GAR in ClpA, ClpB (20), and Hsp104 (21) and DLR in Cdc6p (22)) that interacts with the ␥-phosphate of ATP. This is absent in NSF-D1 (ELE) and is replaced with a lysine in NSF-D2 (GIK). The importance of Sensor 1 and Sensor 2 for ATP turnover and protein function has been evaluated in a number of AAA ϩ proteins (23), but only limited data are available for their importance in NSF (24).
The six protomers of NSF form two stacked homo-hexameric rings (9). Based on the structure of the NSF-D2, there is a central pore that could be contiguous with a predicted pore in NSF-D1 (11,25,26). This pore is thought to be important for substrate binding and processing in other AAA ϩ proteins (12). Three conserved residues YVG (aromatic, hydrophobic, and glycine) are involved in substrate engagement in ClpB and p97 (27,28). Currently, it is unclear whether the pore region of NSF-D1 is catalytically important to NSF function; however, the pore could be an important element for NSF-mediated SNARE disassembly.
To better understand how NSF binds to and disassembles SNARE complexes, it is necessary to investigate the roles of NSF-N and of the conserved motifs in NSF-D1. Studies of NSF have focused on Walker A and B motifs, showing that they are required for ATP binding and hydrolysis, respectively. In this study, we examine the roles of NSF-N, and Sensor 1, Sensor 2, Arginine Fingers, and the central pore motifs of NSF-D1. Using site-directed mutagenesis, we begin to assign specific functions to the various subdomains of NSF by monitoring four key activities as follows: SNAP⅐SNARE binding, complex disassembly, basal ATPase activity, and SNAP⅐SNARE-stimulated ATPase activity. We also test the dominant-negative effects of these mutants in vivo to assess whether mutant protomers can inactivate a hexamer. Our results reveal that NSF-N and the pore of NSF-D1 are important for SNAP⅐SNARE binding. Also, we show that Sensor 1 and Sensor 2 motifs are involved in ATP binding and hydrolysis, respectively. Finally, our data show that Walker B motif and Arginine Fingers are critical for ATP hydrolysis by NSF hexamers and are likely to facilitate communication between protomers.

EXPERIMENTAL PROCEDURES
Plasmids-Plasmids encoding His 6 -NSF and His 6 -␣-SNAP in the pQE9 expression vector were described previously (17,29). The cytoplasmic domain of human VAMP-8 (amino acids 1-73) was cloned into the BamHI and HindIII restriction sites in the pGEX-KG vector, resulting in a construct with N-terminal glutathione S-transferase (GST) tag. Human SNAP-23 (amino acids 1-211) and the cytoplasmic domain of human syntaxin 2 (amino acids 1-251) were cloned into the BamHI/ HindIII and NdeI/XhoI restriction sites, respectively, in the pRSFDuet-1 vector (EMD Bioscience, Madison, WI). This generated SNAP-23 with an N-terminal His 6 tag and syntaxin 2 with a C-terminal S tag. Green fluorescence protein (GFP)tagged wild-type NSF and the ATP hydrolysis-deficient E329Q mutant (NSF-GFP and E329Q-GFP) in pcDNA4/TO vector were kindly provided by Dr. Phyllis I. Hanson (Washington University, St. Louis, MO) and used for mammalian expression of NSF.
Mutagenesis and Protein Expression-Site-specific mutants were created with the QuickChange site-directed mutagenesis kit (Stratagene, La Jolla, CA) using wild-type His 6 -NSF or NSF-GFP as templates. All of the mutations were confirmed by dideoxynucleotide sequencing.
For mammalian expression, constructs containing wild-type NSF or mutants with C-terminal GFP fusion were transiently transfected into HeLa cells with FuGENE 6 (Roche Applied Science) according to the manufacturer's recommendations. Assays were performed 48 h post-transfection for viability measurements or as indicated for immunofluorescence microscopy. HeLa cells were maintained in Dulbecco's modified Eagle's medium (high glucose) containing 10% fetal bovine serum (FBS) (Invitrogen), 100 units/ml penicillin, 100 g/ml streptomycin, and 0.292 mg/ml glutamine at 37°C and 5% CO 2 .
For SNAP⅐SNARE disassembly assays, the binding reactions were performed with 0.5 mM ATP and 2 mM EDTA. The bound complexes were incubated in 0.5 ml of binding buffer containing 5 mM ATP and 10 mM MgCl 2 for 30 min at 4°C. The bound proteins were eluted with SDS-PAGE sample buffer, and the released proteins were recovered with trichloroacetic acid precipitation. The bound and released proteins were analyzed by Western blotting for His 6 tags with India-His-HRP (Pierce) and quantified by densitometry.
ATPase Activity Assay of NSF-Basal ATPase activity was measured as described previously with minor modifications (15). Basal ATPase assays were carried out using 5 g of NSF or mutants in ATPase assay buffer (25 mM Tris-HCl, pH 7.4, 100 mM KCl, 2 mM MgCl 2 , 0.5 mM 1,4-dithiothreitol, 1% glycerol, 5 mM ATP) supplemented with 10 Ci of [␣-32 P]ATP (MP Biomedicals, Santa Ana, CA). After 1 h at 25°C, the reactions were stopped by spotting samples (2 l of each reaction) onto a polyethyleneimine thin layer plate (Selecto Scientific, Suwanee, GA). AMP, ADP, and ATP were separated by ascending chromatography in 0.7 M LiCl and 1 M acetic acid. The radiolabel in each spot was quantified by using a Typhoon 9400 Imager (Amersham Biosciences), and ATPase activity was calculated by determining the percent of [␣-32 P]ADP produced relative to the total nucleotide present. To measure the SNAP⅐SNAREstimulated ATPase activity, SNAP⅐SNARE complexes were assembled for 2 h at 4°C and then incubated with NSF or mutants. The [␣-32 P]ADP produced was quantified as above. The fold increase in ATPase activity was calculated by taking the ratio of ADP produced with and without SNAP⅐SNARE complexes. In all cases, the ATPase activities of NSF and mutants were sensitive to N-ethylmaleimide (NEM) (Sigma) as reported previously (15), and the reactions were done in duplicate.
Viability Assay-HeLa cells were washed once with Dulbecco's modified Eagle's medium (low glucose). Propidium iodide (0.5 ml) (Invitrogen) was added for 30 min at 37°C. The cells were washed twice with Dulbecco's modified Eagle's medium (low glucose), and the stained cells were viewed using an Eclipse TS100 inverted microscope (Nikon, Japan). The percentage of cell death among transfectants was calculated by counting the dead cells (red) with green signal (NSF-or mutant-GFP-positive) divided by total GFP-positive cells (n Ͼ 200). Transfection efficiency was determined by dividing the number of GFP-positive cells by the total number of cells in a given field.
Immunofluorescence Microscopy-HeLa cells were seeded into sterile glass coverslips in 6-well plates and transfected for 15 or 27 h. Cells were washed once with phosphate-buffered saline (PBS), fixed with 3.7% (v/v) formaldehyde for 15 min at 25°C, then quenched with 50 mM NH 4 Cl for 5 min. Cells were rinsed with 10% FBS/PBS and permeabilized with 0.1% Triton X-100 for 5 min at 25°C. Rabbit anti-Giantin IgG (Covance, Princeton, NJ) (1:1000 dilution) in 10% FBS/PBS was then added and incubated for 1 h at 25°C. After washing with 10% FBS/PBS, cells were incubated with Texas Red-conjugated antirabbit IgG (Vector Laboratories, Burlingame, CA) (1:200 dilution) in 10% FBS/PBS for 1 h at 25°C. Coverslips were mounted with VECTASHIELD with 4Ј,6-diamidino-2-phenylindole (Vector Laboratories, Burlingame, CA) and examined with an E-600 epifluorescence microscope (Nikon, Japan). To determine the effects on Golgi morphology, the area of the Giantinpositive Golgi was measured and normalized to the area of the 4Ј,6-diamidino-2-phenylindole-positive nuclei. For this, image analysis was done using Image-Pro Plus 5.0 software (Media Cybernetics, Bethesda).
Immunoprecipitation Assay-HeLa cells were solubilized on ice for 30 min with lysis buffer as described (5). The lysate was cleared by centrifugation. Rabbit anti-GFP polyclonal antibodies (Invitrogen) or rabbit preimmune sera were preincubated with protein G-Sepharose Fast Flow (Amersham Biosciences) for 2 h at 4°C and then incubated with the clarified lysates for another 2 h at 4°C. The beads were recovered and washed three times with lysis buffer. The bound proteins were then eluted with SDS-PAGE sample buffer, boiled, and analyzed by Western blotting. The anti-NSF monoclonal antibody, 2E5 (30), was used to detect both endogenous NSF and exogenous NSF-GFP or mutants. The immuno-decorated proteins were quantified by densitometry, and the ratios of NSF-GFP to NSF were calculated. The correlation between NSF-GFP expression or transfection efficiency and cell death was analyzed using the correlation function (Pearson product-moment correlation coefficient) of GraphPad Prism 4.0 software (GraphPad Software Inc., La Jolla, CA).

Rationale for Site-directed Mutagenesis of NSF-D1-Previous
data suggest that NSF-D1 is the major catalytic domain for NSF (17). Therefore, defining the elements in NSF-D1 that affect ATPase activity should help to elucidate the molecular mechanism of the NSF protein. We introduced mutations into the conserved motifs that were predicted to be adjacent to the nucleotide-binding pocket. Based on the structure of NSF-D2, Sensor 1 and Sensor 2 contribute several residues to the nucleotide-binding pocket ( Fig. 1) (25,26). Sensor 1 in NSF-D1 contains three conserved amino acids (Thr 373 , Asn 374 , and Arg 375 ) (light blue in Fig. 1B) that are proximal to the ␥-phosphate of ATP. Since there are no conserved amino acids in Sensor 2 of NSF-D1, three corresponding residues (Glu 440 , Leu 441 , and Glu 442 ) (dark green in Fig. 1B) were chosen for mutagenesis because they are predicted, by sequence alignments, to be most proximal to the ATP-binding site. Two conserved arginines (Arg 385 and Arg 388 ), called Arginine Fingers (dark blue in Fig.  1B) (19), are proposed to polarize the ␥-phosphate in trans to facilitate hydrolysis. One early hypothesis posited that the N terminus of the SNARE complex contacts the central pore of NSF during complex disassembly (1). To address this point, single or double mutations of two conserved residues in NSF-D1 (Tyr 296 and Gly 298 ) were tested (black in Fig. 1A). All residues discussed above were mutated to either alanines or alternatively charged residues by site-directed mutagenesis. In all cases, the resulting recombinant mutant proteins were hexameric (data not shown) as determined using previously described sizing chromatography (17).
Basal ATPase Activity of NSF-Given the focus on the conserved motifs in NSF-D1, the first assay examined the effects of the mutations on basal ATPase activity. ATPase activity was determined in the presence or absence of NEM; NEM insensitive activity was considered background and subtracted. Under the conditions used, wild-type NSF produced ADP at a rate of 0.64 mol/mg/h, which is comparable with the published values (16,31).
Mutations in Sensor 1 showed the greatest negative effects on basal ATPase activity (Fig. 2). Mutation of the first conserved threonine (Thr 373 ) to alanine reduced the activity by ϳ50%. This is consistent with a previous report that mutation of the corresponding residue in Sec18p (T394P) eliminates ATPase activity (24). A more dramatic effect was observed when the second conserved amino acid, Asn 374 was mutated. The asparagine to alanine mutant (N374A) showed an 85% decrease in ATPase activity; the N374D mutation caused a 60% decrease. These effects on basal ATPase activities are consistent with the known importance of Sensor 1 in ATP binding and/or hydrolysis in other AAA ϩ proteins (12, 23, 32, 33).
Mutations in other regions showed variable but less remarkable effects on basal ATPase activity (Fig. 2).
SNAP-dependent SNARE Binding-A critical aspect of NSF function is its ability to bind SNARE complexes via its adaptor protein, ␣-SNAP. This activity also requires that NSF be in its ATP-bound state (15). To assay for this activity, equal amounts of wild-type or mutant NSF were preincubated with SNARE complexes, with or without ␣-SNAP, in the presence of a nonhydrolyzable ATP analogue (AMP-PNP). The resulting complexes were recovered with glutathione-agarose beads, and the extent of bound NSF or mutants was determined by Western blotting.
Most of the mutations in Sensor 1 negatively affected binding (Fig. 3). The T373A and N374A mutants were unable to bind above background; the N374D mutant retained only ϳ20% of binding activity compared with wild-type NSF. This is perhaps caused by the inability of the mutants to bind nucleotide, which is consistent with their lack of ATPase activity (Fig. 2). Interestingly, binding of the R375A mutant was partially reduced, but when the arginine was mutated to glutamate, binding was ablated. The variants in Sensor 2 had mixed effects on SNAP⅐SNARE binding. The E440R mutation severely affected binding although the E440A mutation was without effect. The L441A mutation partially affected binding, and the mutations at Glu 442 either enhanced or had a partial effect on binding. Mutations of Arginine Fingers (R385A and R388A) displayed only ϳ30% of wild-type binding activity.
In some AAA ϩ proteins, the presence of a conserved YVG motif, in one of the loops that make up the central pore, is essential for substrate binding and/or translocation (12). In NSF, only NSF-D1 contains this conserved motif. To determine its role in SNAP⅐SNARE binding, mutations of Tyr 296 and Gly 298 (Y296A/Y296F, G298A, or double mutations Y296A/ G298A and Y296F/G298A) were generated. The basal ATPase activities of these mutants were similar to that of wild-type NSF (Fig. 2); however, SNAP⅐SNARE binding was reduced in all the mutants (Fig. 3). This suggests that this region of NSF-D1 may be important for protein substrate binding (27,28,34).
SNAP⅐SNARE-stimulated ATP Hydrolysis-Mutants that showed approximately normal levels of basal ATPase activity and partial SNAP⅐SNARE binding, e.g. those in Sensor 2 and of Arginine Fingers, were analyzed further to determine whether they were defective in other aspects of NSF function. Two assays were used to investigate the effects of the mutations on  SNAP⅐SNARE complex disassembly and SNAP⅐SNARE-stimulated ATPase activity. In Fig. 4, NSF or mutants were prebound to SNAP⅐SNARE complexes under conditions where ATP hydrolysis was limited (ATP/EDTA). For wild-type NSF, greater than 60% of the initially bound protein was released upon return to hydrolytic conditions. The Sensor 1 mutant R375A was unable to release into the supernatant. The E440A and L441A mutants, in Sensor 2, were also not released into the supernatant, although a little of the E442R mutant was detected. The E442A mutant behaved like the wild-type protein. Both Arginine Finger mutants (R385A and R388A) showed a similar defect in release. Mutations of Tyr 296 had mixed effects; the Y296A mutant was defective in release, but the Y296F mutants showed close to wild-type activity. No protein was observed in the supernatant when the G298A mutant was tested. It should be noted that this assay measured the release as a percentage of prebound NSF or mutants and that assay conditions were set so that even poorly binding mutants could be analyzed for their release activity.  . ATPase-dependent release of NSF and NSF-D1 domain mutants from SNAP⅐SNARE complexes. A, recombinant SNARE complexes (GST-VAMP-8⅐His 6 -SNAP-23⅐syntaxin 2), His 6 -NSF, or mutants and His 6 -␣-SNAP were incubated in the presence of 0.5 mM ATP and 2 mM EDTA. The bound proteins were recovered with glutathione-agarose beads, and the buffer containing 5 mM ATP and 10 mM MgCl 2 was added. The reactions were incubated, and the proteins released into the supernatant were precipitated with trichloroacetic acid. The bound (p) and released (s) proteins were analyzed by Western blotting as in Fig. 3. The Western blots were quantified by densitometry (B). ATP-dependent NSF release was calculated as follows: released (s) protein divided by total protein (s/(p ϩ s)). Values have been normalized to that of the wild-type (WT) protein. These data are representative of two to three individual experiments, and the error bars represent the range.
We next assayed the SNAP⅐SNARE-stimulated ATPase activities of NSF and several mutants. Consistently, inclusion of SNAP⅐SNARE complexes led to an ϳ18-fold increase in ATPase activity for wild-type NSF (Fig. 5). No ADP was generated when SNAP⅐SNARE complexes were incubated alone (data not shown). Mutations (E442A and Y296F) that had no effect on NSF release (Fig. 4) showed similar stimulation of ATPase activities (Fig. 5). None of the mutations with impaired release (Fig. 4) showed an increased ATPase activity in the presence of SNAP⅐SNARE complexes (Fig. 5). Mutations of Sensor 2 and of Arginine Fingers affected SNAP⅐SNARE-stimulated ATPase activities, which most likely accounts for their negative effects on NSF⅐SNAP⅐SNARE complex disassembly.
Motifs Critical for NSF Function in Vivo-Previous in vitro studies showed that all six protomers must have functional ATPase activity for NSF to be active in inter-cisternal Golgi transport (17). This explains the dominant-negative and cytotoxic effects of the E329Q mutant when expressed in U2OS cells, assuming that the mutant and wild-type protomers, when co-expressed, randomly associate to form mixed hexamers (5). Simplistically, a dominant-negative phenotype of mixed hexamers could indicate that coordinated, inter-protomer communications are necessary for NSF function. To monitor whether the mutants from our study also have such dominantnegative effects, wild-type or mutant NSF expression constructs with C-terminal GFP tags were transiently transfected into HeLa cells, and then Golgi complex morphology and cell viability were examined. Specific focus was placed on mutants that were unable to disassemble NSF⅐SNAP⅐SNARE complexes or that failed to show SNAP⅐SNARE-stimulated ATPase activity (see Figs. 4 and 5). As shown by Dalal et al. (5), expression of the E329Q mutant led to a remarkable disruption of the Golgi complex (Fig. 6A). Expression of the Arginine Finger mutant, R385A, had a similar effect on Golgi complex structure. Within 15 h post-transfection, cells expressing R385A developed a swollen and fragmented Golgi complex that started to disperse into the cell periphery. The effects of expressing the E329Q and R385A mutants were more obvious when the dispersal of the Golgi complex was quantified (Fig. 6B), and the viability of cells expressing the mutants was measured (Fig. 6C). At 15 h posttransfection, there was a dramatic increase in the ratio of Golgi complex to nuclear area; at 48 h post-transfection, 60% of the cells expressing either mutant were propidium iodide-positive. Expression of the L441A mutant caused a subtle increase in the size of the Golgi complex (Fig. 6, A and B) but had no effect on cell viability (Fig. 6C). Expression of another Sensor 2 mutant (E440A) had little effect on Golgi complex morphology (supplemental Fig. 1 and Fig. 6B) or cell viability (Fig. 6C), despite their phenotypes in the assays mentioned above (see Figs. 4 and 5). Expression of mutants in the pore region (Y296F, G298A, and double point mutation Y296F/G298A) also had limited effects. Aberrant Golgi ribbons were observed when G298A and Y296F/G298A mutants were expressed (Fig. 6, A and B), and there was a subtle increase in cell viability (Fig. 6C). Expression of two mutants of NSF-N (R67A and Y83F, see results below) had almost no effect (supplemental Fig. 1 and Fig. 6, B and C). To confirm the in vivo formation of mixed hexamers containing both mutant and endogenous wild-type NSF protomers, extracts from transfected HeLa cells were subjected to immunoprecipitation using an anti-GFP antibody. In all cases, both endogenous NSF and exogenous NSF (or mutant)-GFP were co-immunoprecipitated, demonstrating the formation of mixed hexamers (Fig. 7). The ratios of the NSF (or mutant)-GFP to NSF in the "input" samples and the transfection efficiencies were measured in each case to determine whether cell toxicity and Golgi dispersal correlated with expression or transfection levels (supplemental Fig. 2). From this analysis, no correlation could be detected between expression of NSF (or mutant)-GFP or transfection efficiency and cell death (Pearson product-moment correlation coefficient (r): Ϫ0.6 Ϫ 0.1; data not shown). This would suggest that the phenotypes are not solely due to the levels of mutant proteins in the cells.
Important Residues of NSF-N-NSF-N is composed of two subdomains, N A and N B , joined by a linker or hinge region (Fig.  8A). The surface of NSF-N has several positively charged amino acids at the apex of the cleft formed by N A and N B , which could play a role in binding to the negatively charged C terminus of ␣-SNAP (13,14). To characterize this surface, five positively charged residues (Arg 10 , Lys 68 , Lys 104 , Lys 105 , and Lys 143 ) and three acidic residues (Asp 14 , Glu 15 , and Asp 142 ) were mutated (Table 1 and Fig. 8). Two residues (Tyr 83 and Cys 91 ) ( Table 1 and Fig. 8) in the hinge region at the base of the cleft were also mutated because this region has been shown to be a substratebinding site for the related AAA ϩ protein, p97 (35)(36)(37), and these residues have been reported to be the sites of post-translational modification of NSF (38,39). The SNAP⅐SNARE binding and basal ATPase activities were measured for most of the mutants, and the results were summarized in Table 1. As expected, none of the mutations had a drastic effect on the ATPase activity of NSF, although the D14R mutant did show an increase. Mutations of the positively charged residues (Arg 10 , Arg 67 , Lys 68 , Lys 104 , Lys 105 , and Lys 143 ) either abolished or severely affected binding to SNAP⅐SNARE complexes. The E15A, E15R, D142A, and D142R mutants were also defective in binding. The D14A and D14R mutants showed close to wildtype binding levels. Mutation of the residues (Tyr 83 and Cys 91 ) in the hinge region showed mixed effects with Y83E having the greatest negative effect and Y83F having only a partial effect. These results suggest that the surface located on the top of interface made by N A and N B is important for SNAP⅐SNARE complex binding (Fig. 8).

DISCUSSION
In this study, we have used site-directed mutagenesis and a battery of assays to understand the mechanism of the general secretory protein NSF. For its cellular function, NSF must bind to SNAP⅐SNARE complexes and use ATP hydrolysis to disassemble them. The conformational changes in NSF induced by ATP binding and hydrolysis are critical for its function. The data presented here show that the specific elements of NSF-N and NSF-D1 domains are required for ATP hydrolysis, for SNAP⅐SNARE complex binding and disassembly, and for activity in vivo (summarized in Table 2). Each element of the nucleotide-binding site of NSF-D1, Sensor 1, Sensor 2, Arginine Fingers, as well as the central pore plays a key role. The charged surface formed by the cleft between the subdomains of NSF-N plays a role in SNAP⅐SNARE binding. These findings may be applicable to other AAA ϩ proteins that use ATP binding/hydrolysis and adaptor proteins or accessory domains to carry out their cellular functions.
Sensor 1 and 2-The crystal structures of the processivity clamp loader ␥-complex RuvB or its eukaryotic homologue RuvBL1 have a threonine or asparagine residue in Sensor 1 that forms a hydrogen bond with the ␥-phosphate of ATP (33,40,41). This may allow discrimination between bound ATP and ADP. A similar arrangement is found in other AAA ϩ proteins FIGURE 6. Effects of NSF mutants in vivo. Golgi morphology in transfected HeLa cells is shown. Expression constructs encoding wild-type (WT) or mutant NSF-GFP were transiently transfected into HeLa cells for the indicated times. A, immunofluorescence analysis showed the localization of wild-type or mutant NSF-GFP and the Golgi complex were visualized with anti-Giantin antibody. The images were merged with NSF-GFP (green) and Golgi complex (red), and the nuclei stained by 4Ј,6-diamidino-2-phenylindole (DAPI) (blue). B, quantification of Golgi complex. The areas of the Golgi complex and nuclei shown in A were measured and their ratios were calculated (n Ն10 cells). Error bars represent standard deviations. C, viability of transfected HeLa cells. Cells, transfected for 48 h, were stained with propidium iodide (red) and counted. The dead cells (red) with green signal (NSF-GFP) were divided by total green cells (n Ͼ200). These data are representative of two individual experiments, and the error bars represent the range. (42,43). Mutagenesis of DnaA has shown a role for Sensor 1 in high affinity nucleotide binding (32). Sensor 2 is also proposed to allow discrimination between the nucleotide-bound and unbound states. A conserved arginine in Sensor 2 of some AAA ϩ proteins forms a salt bridge with the ␤-phosphate of ATP and ADP (33,40,41). Mutation of Arg 826 in Sensor 2 of Hsp104 equally weakens its affinity for ATP and ADP (21). However, the roles of Sensor 1 and Sensor 2 may extend further than just passive nucleotide-sensing. Several studies demonstrate that the two Sensors are involved in catalysis but not nucleotide binding (44 -50).

Dissecting the N-Ethylmaleimide-sensitive Factor
In the structure of NSF-D2, Ser 655 (which corresponds to one of the conserved asparagines in Sensor 1 of NSF-D1) is involved in a hydrogen bond network that positions a water molecule as a potential nucleophile for ATP hydrolysis (26). Although NSF-D2 has limited ATPase activity (17), such an orientation for an asparagine in NSF-D1 suggests that Sensor 1 could have a catalytic role in NSF. In our study, the T373A, N374A, and N374D mutations in NSF-D1 diminished the intrinsic ATPase activity of NSF, consistent with a role of Sensor 1 in nucleotide binding and/or hydrolysis. The mutations also abolished ATPdependent binding to SNAP⅐SNARE complexes ( Fig. 2 and Fig.  3). This phenotype is similar to that of a mutant in the Walker A motif (K266A), which is defective in nucleotide binding (17). It is thought that K266A does not bind SNAP⅐SNARE complexes because it cannot attain the ATP-dependent conformation required. Based on the comparison with K266A, the phenotypes of the T373A, N374A, and N374D mutants are consistent with these residues being required more for ATP binding than hydrolysis. Mutations of Arg 375 had little effect on basal ATPase activity (Fig. 2), but a negative charge at that position (R375E) did affect SNAP⅐SNARE complex binding (Fig. 3). Removal of the positive charge (R375A) eliminated ATPasedriven complex disassembly (Fig. 4). These phenotypes suggest that Arg 375 could play a role in mediating the nucleotide-dependent conformational changes required for both SNARE FIGURE 7. Mixed hexamer formation in HeLa cells. Forty eight hours posttransfection, immunoprecipitations from cellular extracts were carried out using rabbit anti-GFP antibody. Rabbit preimmune sera were used as a negative control. Samples immunoprecipitated with GFP antibodies were blotted with monoclonal antibody (2E5) for NSF. The endogenous NSF and exogenous NSF-GFP are indicated with arrows. complex binding and disassembly. The results presented here suggest that Sensor 1 in NSF-D1 is important for nucleotide binding and inducing a SNAP⅐SNARE-binding competent state. However, it should be noted that mutation of the first threonine in Sensor 1 (T394P) of Sec18p-D1 has a phenotype similar to that of NSF E329Q, which can bind but not hydrolyze ATP (24).
By contrast, most of the Sensor 2 mutations had only limited effects on basal ATPase activity, suggesting that Sensor 2 plays only a limited role in nucleotide binding (Fig. 2). The E440R (but not E440A), L441A, and the E442R (but not E442A) mutations did affect SNAP-dependent SNARE complex binding (Fig. 3). Of the mutations tested, only the E442A mutant retained wild-type activity in the ATPase-dependent release assay (Fig. 4) and the SNAP⅐SNARE-stimulated ATPase assay (Fig. 5). Given the differential effects of the E440R and E440A mutations, it would appear that the presence of a positive charge at that position is detrimental to ATP-dependent, SNAP⅐SNARE complex binding, but the presence of a negative charge is not essential. Comparison of E442A and E442R suggests that a positive charge negatively affects SNAP⅐SNARE complex-stimulated ATPase activity. Taken together it appears that the elements of Sensor 2 are not essential for nucleotide binding or basal hydrolysis (Fig. 2), but they do play a role either in how the ATPase of the NSF is stimulated by the SNAP⅐SNARE complex or in how ATP hydrolysis is coupled to complex disassembly. The fact that mixed hexamers, which include Sensor 2 mutants, did not have the dominant-negative effects in vivo (Fig. 6, B and C, and Table 2), indicates that Sensor 2 may act in cis and is not involved in inter-protomer communication during ATP hydrolysis.
Arginine Fingers-Arginine Fingers were first noted in the structure of the Ras/Ras-guanosine, triphosphatase-activating protein (Ras-GAP) complex (19). The positively charged guanidinium group of Arg 789 interacts with the fluoride of the GDP-AlFx in the nucleotide-binding site, and the main chain carbonyl oxygen forms a hydrogen bond with a glutamine residue (Gln 61 ) of the switch II region of Ras. This arginine is thought to participate in catalysis by stabilizing the transition state during GTP hydrolysis. A similar interaction was found in p97-D1, where Arg 359 from a neighboring protomer interacts with the ␥-phosphate of ATP, and Arg 362 forms a salt bridge with Glu 305 in the Walker B motif on an adjacent subunit (51). The significance of Arginine Fingers in the AAA ϩ family was first noted for FtsH; Arg 312 and Arg 315 are crucial for ATP hydrolysis and protease activity (52). Subsequently, the importance of Arginine Fingers has been shown for many AAA ϩ proteins (42,44,50,(53)(54)(55)(56)(57). The role of Arginine Fingers in NSF has been unclear given our previous report that mutations of Arg 385 or Arg 388 did not affect basal or stimulated ATPase activity (16). These mutants were re-examined in this study. Consistent with previous studies (16), basal ATPase activities 100 ND 83 Ϯ 6 a SNAP⅐SNARE complex binding and basal ATPase activity of wild-type NSF were set as 100%. The values for the mutants were normalized to the wild-type NSF. b ND means not determined. Ϫ, no effect). b ND means not determined.
were close to that of wild-type NSF for both R385A and R388A (Fig. 2) suggesting that the arginines are not required for basal hydrolysis. The R385A and R388A mutations did affect SNAP⅐SNARE complex binding (Fig. 3), but both mutations eliminated ATPase-dependent release (Fig. 4) and blunted SNAP⅐SNARE complex-stimulated ATPase activity in vitro (Fig. 5). The latter was a deviation from our previous report (16), perhaps because of our improved preparations of SNARE complexes. These data indicate that Arginine Fingers play a role in inducing a SNAP⅐SNARE-binding competent state and in mediating ATP hydrolysis when NSF is engaged with SNAP⅐SNARE complexes. Given their proposed position in NSF-D1 (Fig. 1B), Arginine Fingers are likely to function in trans on the adjacent protomer. Consistently, when expressed in HeLa cells, the R385A mutant had a dominant-negative effect as did the Walker B motif mutant E329Q (Fig. 6). Of the mutants tested, only E329Q and R385A caused appreciable cytotoxicity and alterations in Golgi complex morphology.
Inter-protomer Interactions-AAA ϩ family proteins generally form ring-shaped hexamers, and ATP binding appears to be important for oligomerization of some family members, e.g. yeast Vps4, mammalian SKD1/VPS4B, and bacterial ClpB (42,55,56,58). ATP binding could induce conformational changes that promote oligomerization. Alternatively, oligomerization could bring the residues from neighboring protomers in close contact to the bound nucleotide thus promoting hydrolysis and/or stabilizing the oligomer (these are not mutually exclusive). Arginine Fingers from a neighboring protomer are an example of these types of residues. They can serve as transacting elements and directly or indirectly interact with the ␥-phosphate. Mutations in Arginine Fingers abolish hexamerization in ClpB (56) and decrease cooperative helicase activity in MCM (50). In p97, the R362E mutation partially disrupts the hexamer, and R359E/R362E double mutants are predominantly monomers (59). A hexamerization defect was not observed in the NSF-D1 mutants reported here. NSF-D2 is required for oligomerization (15) but has no conserved Arginine Fingers nor does it catalyze significant ATP hydrolysis.
Only the E329Q and R385A mutants were toxic when expressed in HeLa cells (Fig. 6C). Given that both mutants did form mixed hexamers (Fig. 7), it seems plausible that inclusion of a mutant protomer "poisons" the hexamer, thus leading to cytotoxicity. This would imply some concerted mechanism to interconnect protomers. The Arginine Fingers of NSF could connect two adjacent protomers by interacting with the ␥-phosphate of ATP and/or participating in electrostatic interactions with Glu 329 (Fig. 1B), similar to what is proposed for p97 and the processivity clamp loader ␥-complex (41,51). An interaction between Arginine Fingers and the Walker B motif is also suggested in the MCM protein, where ATPase activity of the R473A (Arginine Finger) is rescued by co-incubation with the D404A (Walker B) mutant. A scenario specific for NSF would have ␣-SNAP stimulating the ATPase activity by positioning of the Arginine Fingers in the ATP-binding pocket of an adjacent subunit. In this configuration, they interact with the glutamate in Walker B motif, stabilizing the transition state and facilitating ATP hydrolysis. Other interactions with Arginine Fingers are also possible, because the mixture of Arginine Fingers and Walker A mutants partially restores the ATPase activity of RuvB, p97, and MCM (50,59,60). Further structural studies of the conformational changes of NSF during ATP hydrolysis, however, are required to decipher how these residues might affect nucleotide binding and hydrolysis in the hexameric configuration.
NSF-N Domain-NSF-N is essential for protein substrate binding for both NSF (2), and its close relatives, p97, CDC48A, VAT, and PEX1 (49,(61)(62)(63). For p97, it has been proposed that the N domain contains the binding sites for adaptor proteins, and the binding sites for denatured proteins are in its D2 pore region (28). For NSF, the conserved groove 3 of NSF-N, located on the top part of the cleft between N A and N B subdomains, is predicted (based on sequence conservation and charge) to be a binding site for ␣-SNAP (14). A surface with similar properties is present on the N domain of Sec18p (64). Two of the mutations in groove 3, R67E in NSF and G89D in Sec18p, inhibit the interaction of NSF and ␣-SNAP or Sec18p and Sec17p (yeast homologue of ␣-SNAP), respectively (16,65). Our studies further support the role of this surface and suggest that the upper portion of the cleft between N A and N B is important for SNAP⅐SNARE binding (Fig. 8A and Table 1). It should be noted that these data imply the importance of this cleft but do not conclusively identify it as the binding site. It is quite possible that this charged surface controls the position of NSF-N as proposed for p97-N by Brunger and DeLaBarre (66).
Two residues, Tyr 83 and Cys 91 , are reported to regulate NSF activities by phosphorylation and nitrosylation, respectively (38,39). Our characterization of the Y83E, Y83F, and C91A mutants were consistent with the published data. The phospho-mimetic mutant Y83E eliminated the binding to SNAP⅐SNARE complexes, whereas Y83F did not, supporting the hypothesis that phosphorylation of NSF can be used to control its activities. It is possible that addition of a negative charge at Tyr 83 disrupts an interface between the subdomains, thus perturbing ␣-SNAP binding. In vivo expression of Y83F caused no adverse effect on the HeLa cells, a result that differs from previous reports (38). This deviation is hard to explain but may be rooted in the different cell types used. However, the lack of effect when the R67A and Y83F mutants were expressed in HeLa cells is in good agreement with the conclusion that NSF hexamers do not need a full complement of functional N domains for activity (15). This conclusion is also consistent with the "3-in, 3-out" model for NSF-N function, where it is proposed only three N domain-containing protomers are needed to interact with SNAP⅐SNARE complexes (13).
Central Pore-Several AAA ϩ proteins, (e.g. Hsp104, ClpB, ClpA, ClpX, FtsH, HslUV, and Lon) share a common threading mechanism for interacting with protein substrates (12,67). The conserved YVG motif forms a hydrophobic patch, which sits above the pore in the hexamer and is responsible for substrate binding and/or translocation into the pore. The central pores in NSF or p97 have been less studied because it is thought that their narrow diameter makes a threading-through mechanism unlikely (12,68,69). However, it is possible that the central pore plays other roles. Mutagenesis studies of p97 and VAT suggest that the hydrophobic patch has a similar function to that of the AAA ϩ proteases. The aromatic residue is thought to be involved in substrate binding and unfolding (28,49,70). Single particle analysis of the NSF⅐SNAP⅐SNARE complex suggests NSF is arranged as a double-ring structure with a central hole. In NSF-D2, this hole tapers from 3-5 nm at the top (which is proposed to face NSF-D1) to 1.8 nm at the bottom (10,11). The free SNAP⅐SNARE complex is 7-8 nm in diameter and 14 -15 nm in length. In the NSF⅐SNAP⅐SNARE complex, NSF-D1 ring diameter increases (from 12 to 13.5 nm) and the SNAP⅐SNARE complex shortens (13 nm) (9 -11, 25, 26). This change in the dimensions of the components is the only evidence that indicates that the SNAP⅐SNARE complex rod could partially insert into the NSF-D1 pore. Our studies show that the hydrophobic patch (Tyr 296 to Gly 298 ) in one of the loops that protrudes into the central pore of NSF-D1 might mediate an interaction with the SNAP⅐SNARE complexes. The mutants in this motif (Y296A, Y296F, G298A, Y296A, G298A, Y296F, and G298A) affected SNAP⅐SNARE binding, although to different extents (Fig. 3). The Y296A (but not Y296F) and the G298A mutants were defective on SNAP⅐SNARE complex disassembly (Fig. 4), suggesting the aromatic residue and glycine residue in the NSF-D1 pore may assist complex dissociation. Although no cell death was seen when pore mutants were expressed in HeLa cells, G298A and Y296F/G298A mutants had moderate effects on Golgi complex morphology (Fig. 6, A and B). This would suggest that all six loops of the pore are important for optimal NSF function but are not essential.
Summary-Using mutagenesis and a series of functional assays, we dissected the roles of specific elements of NSF. We identified two possible regions of NSF that affect binding to SNAP⅐SNARE complexes, the positively charged surface on NSF-N and the central pore region of NSF-D1. We also characterized most of the important motifs in the catalytic NSF-D1. Combined with our previously published studies of Walker A and Walker B motifs, we have a clearer view of NSF function. The NSF⅐SNAP⅐SNARE binding and disassembly process can be divided into four steps as follows: 1) ATP binding; 2) SNAP⅐SNARE binding; 3) stimulated ATP hydrolysis; and 4) SNAP⅐SNARE complex disassembly. Walker A and Sensor 1 motifs work in step 1; NSF-N and the central pore work in step 2; Sensor 2, Walker B, central pore, and Arginine Finger motifs work in step 3 and/or step 4. The data presented here will be useful in designing future experiments because it describes NSF mutants that are defective at specific steps in the NSF reaction sequence. These mutants will be valuable for future studies that attempt to correlate conformational changes in NSF with specific stages of its reaction cycle.