Phosphatidic acid induces conformational changes in Sec18 protomers that prevent SNARE priming

Eukaryotic cell homeostasis requires transfer of cellular components among organelles and relies on membrane fusion catalyzed by SNARE proteins. Inactive SNARE bundles are reactivated by hexameric N-ethylmaleimide-sensitive factor, vesicle-fusing ATPase (Sec18/NSF)-driven disassembly that enables a new round of membrane fusion. We previously found that phosphatidic acid (PA) binds Sec18 and thereby sequesters it from SNAREs and that PA dephosphorylation dissociates Sec18 from the membrane, allowing it to engage SNARE complexes. We now report that PA also induces conformational changes in Sec18 protomers and that hexameric Sec18 cannot bind PA membranes. Molecular dynamics (MD) analyses revealed that the D1 and D2 domains of Sec18 contain PA-binding sites and that the residues needed for PA binding are masked in hexameric Sec18. Importantly, these simulations also disclosed that a major conformational change occurs in the linker region between the D1 and D2 domains, which is distinct from the conformational changes that occur in hexameric Sec18 during SNARE priming. Together, these findings indicate that PA regulates Sec18 function by altering its architecture and stabilizing membrane-bound Sec18 protomers.

PA production, however, is not always a positive signal for fusion. In the case of Sec18, increased PA levels lead to reduced priming activity likely due to a decrease in recruitment to cis-SNAREs (13). On yeast vacuoles, PA is converted to diacylglycerol (DAG) by the PA phosphatase Pah1, an ortholog of mammalian Lipin1. In the absence of Pah1 activity, PA levels remain intact and sequester Sec18 from cis-SNARE complexes to prevent priming and arrest the fusion pathway, whereas deleting the other PA yeast phosphatases had no effect on fusion (17). DAG can be converted to PA through the action of the DAG kinase Dgk1, whose inactivation leads to elevated DAG concentrations that enhance fusion through modulating the activity of the Rab GTPase Ypt7 (27). Thus, the interconversion of PA and DAG serves as a regulatory switch to control vacuole fusion.
Here we asked what effects PA binding has on the overall architectural dynamics of Sec18 that could lead to a decrease in its priming activity. To do so, we measured binding of monomeric and hexameric Sec18 to different forms of PA. We report that monomeric Sec18 has significantly stronger binding than the hexameric form to all forms of PA. We probed changes to the architecture of Sec18 when bound to short-chain PA and found that the protein exists in a significantly different conformation in its PA-bound state, without significant changes to its secondary structure. To study the mechanism of Sec18 binding to PA, molecular dynamics simulations were performed using the mammalian version of Sec18, namely NSF. NSF was used as it has high identity to Sec18 and has more structural information available at the Protein Data Bank (PDB ID 3J94) (28). The molecular dynamic simulations performed suggest NSF binds to PA at regions of the protein that are only exposed in the monomeric state of the protein. Taken together, we propose that PA regulates the priming activity of NSF/Sec18 by limiting the formation of its active hexamer.

Sec18 monomer binds to PA with higher affinity than the hexameric form
Our previous work showed that Sec18 preferentially bound to liposomes containing phosphatidylcholine (PC), phosphatidylethanolamine (PE), and PA relative to those composed of only PC and PE, or ones where PA was replaced with DAG or phosphatidylserine (PS) (13). This was in keeping with older findings showing that mammalian NSF bound to resin-linked PA (12). Here our studies were extended to further define how Sec18 binds to PA. To start we used microscale thermophoresis (MST) to acquire binding affinities to dioctanoyl-PA (C8-PA), which prevents Sec18 from binding cis-SNARE complexes, consequently precluding priming from occurring (13). We used both monomeric and hexameric Sec18 with a range of C8-PA. The C-terminal His 8 tag of Sec18 was labeled with Ni-NTA Atto 488. As shown in Fig. 1B, monomeric Sec18 (mSec18) bound to C8-PA with a K D of 1.4 Ϯ 0.68 M, whereas the hexameric form (hSec18) had a K D of 29 Ϯ 8.6 M. This suggested that either hSec18 has residues occluded for PA binding or is in a suboptimal conformation to efficiently bind C8-PA. It is possible that a small soluble C8-PA could access a binding site on Sec18 that is obscured in the hexamer, whereas membranous long-chain PA is unable to reach PA-binding regions on Sec18 hexamers, especially regions contained in the hexamerization interface of Sec18 holoenzyme.
Due to the difference in binding affinities to C8-PA, we next asked if limiting the mobility of PA to two dimensions would show a similar disparity between the monomer and hexamer. To this aim we used extruded 0.8-m diameter liposomes to approximate the diameter of yeast vacuoles. We found that mSec18 bound 0.8 M liposomes containing 10% long-chain PA (80% POPC, 10% POPE) with a K D of 29 Ϯ 20 M, whereas hSec18 bound these liposomes poorly with a K D of 423 Ϯ 215 M (Fig. 1, C and D). We next used PC liposomes as a negative control and found that hSec18 bound as poorly as it did to PA liposomes, whereas mSec18 lacked any detectable interactions above the background noise of the system. These findings further establish that hexameric Sec18 lacks the ability to bind PA, potentially by masking a binding site or by restricting conformational changes needed to bind PA.

Sec18 binding to PA is specific
Our previous work showed that Sec18 binds to vacuoles in a PA-dependent manner (13). However, others have shown that proteins with PA binding capabilities sometimes associate with other anionic lipids including phosphoinositides and phosphatidylserine (PS) in a nonspecific, charge-dependent manner (29). We previously observed that Sec18 does not associate with PS in the membrane, however, we wanted to additionally verify that its membrane association was not due to highly anionic phosphoinositides. Here we show that Sec18 does not readily bind to C8-PI(4,5)P 2 (K D Ն 350 M) (Fig. 1C). This is significant because it was previously shown that PI(4,5)P 2 is required for SNARE priming to occur at the vacuole (30). Our results show that Sec18 association with lipids at the vacuole membrane is PA specific.

ATP blocks PA binding by Sec18
Sec18/NSF, like many other AAA ϩ proteins contains two nucleotide-binding domains (NBD) each residing in a one of the domains that make up the rings of the hexameric protein.
The D1 ring of Sec18 hydrolyzes ATP to generate the mechanical force needed to disrupt cis-SNARE bundles, whereas the D2 ring binds ATP to stabilize the hexameric form of the protein. This is reflected in the different affinities for ATP found between the two NBDs. In NSF the D1 NBD binds ATP with a K D of 15-20 M, whereas the D2 NBD binds with a K D of 30 -40 nM (31). Here we asked if ATP binding would affect PA binding. We added 1 mM ATP-Mg 2ϩ to binding assays with Sec18 and 0.8-m PA liposomes. We found that ATP-Mg 2ϩ reduced mSec18 binding to PA liposomes, whereas not affecting the already poor binding by hSec18 (Fig. 1, D and E). The effect of ATP on mSec18 was possibly due to hexamerization of the protein and reduced masking of the PA-binding site. Interestingly,

Phosphatidic acid and Sec18 architecture
ATP had no effect on the interaction of mSec18 and C8-PA (Fig. 1F). The difference could be due to the close apposition of mSec18 to the membrane that affects how ATP alters PA binding, which is not recapitulated in solution.

PA blocks Sec18 binding to Sec17-SNARE complexes
Previously we showed that C8-PA inhibits priming activity on isolated vacuoles and by preventing its association with SNAREs. To see the direct effect on protein complex formation we used surface plasmon resonance (SPR). Sec18 was linked to Ni-NTA sensor chips through its C-terminal His 8 tag. For these experiments we used pre-assembled soluble SNARE bundles in which the transmembrane domains were deleted from Vam3, Vti1, and Nyv1. The soluble SNAREs were mixed with the naturally soluble SNARE Vam7 to allow for 4-helical bundle formation as described previously (32,33). Soluble SNARE bundles were mixed with recombinant Sec17 and flowed over Ni-NTA-bound Sec18. This approach showed 100 response units. However, when C8-PA was added in the mixtures with the SNAREs and Sec17, we observed a 50% reduction in Figure 1. Sec18 hexamer and monomer binding affinity for PA. A, schematic of Sec18 monomer and hexamer. Shown is the priming reaction where monomeric Sec18 (mSec18) is initially bound to the membrane after which it is released upon Pah1 activity. Released mSec18 then forms hexameric Sec18 (hSec18) and is recruited to cis-SNARE bundles decorated with Sec17. Sec18 hydrolyzes ATP to dissociate the SNAREs into individual proteins leading the release of Sec17 and the soluble SNARE Vam7. B, C8-PA MST measurements were performed using purified Sec18 monomer and hexamer labeled with Ni-NTA Atto 488 dye at 90% LED and High MST using NT.115 labeled thermophoresis. Binding affinity was measured using thermophoresis at 15 s by mixing separate reactions of half 100 nM Atto 488-labeled Sec18 monomer and half 1:1 titrations of C8-PA with the highest concentration (370 M) according to the GraphPad Sigmoidal 4PL curve. C, MST measurements of monomeric Sec18 with C8-PA versus C8-PI(4,5)P 2 . D, the effect of 1 mM ATP-Mg 2ϩ Sec18 binding to liposomes. 800-nm diameter extruded PA liposomes (10% PA, 70% PC, and 20% PE) and PC liposomes (80% PC, 20% PE) were incubated with Sec18 monomer or hexamer in the presence or absence of ATP. E, the quantitation of multiple experiments run in panel D. Asterisk (*), no measureable PC-liposome binding was detected above the background noise of the system. F, SPR of ATP competition of soluble C8-PA binding to Sec18. G, SPR of Sec18 -His 8 bound to CM5 sensor chips. Soluble SNARE complexes (167 nM) (lacking transmembrane domains) decorated with 500 nM Sec17 at a 3:1 ratio were flowed over the bound Sec18 in the presence or absence of 1.25 M C8-PA. n Ն 3 for all data shown.

Phosphatidic acid and Sec18 architecture
response units (Fig. 1G). Because SPR response units are based on mass, we conclude that PA blocked the assembly of the larger SNARE-Sec17-sSNARE complex.

Both D1 and D2 domains bind Sec18
Based on structural predictions for PA binding (see below), we hypothesized that Sec18 binds PA using both its D1 and D2 domains. To test this, we compared full-length Sec18 with domains or protein truncations lacking the N-terminal domain (D1-D2) or the D2 domain (N-D1) ( Fig. 2A). Using SPR, GST-Sec18 constructs were flowed over nanodiscs (ND) containing long-chain PC and PE (80:20) or long-chain PC, PE, and PA (80:15:5). The ND were linked to the Ni-NTA chip through the His tags of the membrane scaffold proteins. First we tested the known PA-binding domain DEP of the murine protein Dvl2 (34). GST-DEP bound to PA-ND with a K D of 18.9 Ϯ 2 M (Fig.  2B). Next we tested binding of full-length GST-Sec18 to PA-ND and found that it bound with a K D of 2.7 Ϯ 2 M, illustrating that Sec18 interacted with PA as well or better than the bona fide PA-binding domain DEP (Fig. 2C). To verify that the GST tag did not alter the binding of Sec18 to PA, we linked Sec18 -His 8 to the sensor chip and flowed C8-PA. By this approach we measured a K D of 1.4 Ϯ 0.46 M, illustrating that the GST tag had no effect on PA binding (Fig. 2D). In parallel we tested the D1-D2 construct, lacking the N-terminal domain, and found that it bound PA with nearly identical affinity (K D , 1.2 Ϯ 0.4 M) to full-length Sec18. Because the Sec18 N-domain has a polybasic surface at its Sec17binding site, we tested if it also bound to long-chain PA. We found that the N-domain bound poorly to PA-ND with a K D of 31.8 Ϯ 3.7 M and to PC-ND with a K D of 11 Ϯ 1.6 M (Fig. 2, E and F). These data suggest that the N-domain has no lipid-binding specificity and that the effect of PA on Sec18 is limited to the D1-D2 domains. Next, we used MST to verify our SPR data and to further examine PA binding by various constructs. In Fig. 2G, we show that full-length Sec18 bound C8-PA with a K D of 0.927 Ϯ 0.161 M, which was similar to the value we observed with SPR. We next tested the individual domains. The individual D1 and D2 domains bound PA with K D values 1.5 Ϯ 0.35 and 2.96 Ϯ 0.55 M, respectively, which is near to what we saw with the D1-D2 construct. In comparison, the N-D1 construct bound PA poorly relative to the D1 construct. This suggested that the N-domain could sterically interfere with PA binding or perhaps induce a suboptimal lipidbinding conformation in D1 that does not occur in the full-length protein.

Phosphatidic acid and Sec18 architecture Phosphatidic acid alters the conformation of Sec18
Our data thus far suggests that Sec18 undergoes conformational changes that allow mSec18 to bind PA, whereas hSec18 lacks the ability to bind the lipid. To further probe for conformational changes to Sec18 we tested whether PA significantly alters binding of 8-anilino-1-naphthalenesulfonic acid (ANS) to mSec18. ANS is a dye that has been extensively used to test lipid-binding proteins because it associates with solution-exposed hydrophobic motifs (35,36). Binding of ANS to a protein results in an increase in fluorescence yield and a blue-shifted emission. Because we have previously seen PA binding to mSec18 we expected ANS to also bind the protein in our assay. As expected, we observed ANS binding to mSec18 in a dose-dependent fashion (Fig. 3, A and B). We next wanted to test for any conformational changes upon PA binding that altered ANS binding to Sec18. To do this, we titrated increasing amounts of C8-PA into our assay and measured changes in the ANS fluorescence spectra. Because C8-PA is partially hydrophobic, ANS was first incubated with each lipid concentration to obtain a background spectrum before protein was then added to the assay, and fluorescence was again measured. The difference spectra from these measurements shows that addition of C8-PA increases the binding of ANS to Sec18 (Fig. 3, C and F).
To confirm that the changes in ANS fluorescence were specific to PA binding, we tested the addition of DAG, the product of Pah1 activity on PA. No change in ANS fluorescence was detected in the presence of C8-DAG, which is consistent with the inability of Sec18 to bind to DAG (Fig. 3, D and F). We also tested the anionic lipid phosphatidylserine (PS). Similar to what we observed with DAG, the addition of C8-PS had no effect on ANS fluorescence (Fig. 3, E and F). Together these data suggest that C8-PA binding to Sec18 results in a conformational change in the protein that exposes additional hydrophobic pockets to solution. Such a change may account for the differences previously seen in Sec18 priming activity and cis-SNARE association (13).
To further probe for conformational changes to Sec18 induced by PA we utilized a limited proteolysis assay. Proteins can exhibit differences in their proteolytic cleavage profiles when bound to a ligand that significantly changes their overall architecture (35). Because we observed an increase in solutionexposed regions of Sec18 in the presence of PA, i.e. increased ANS fluorescence, we expected to also see an increased sensitivity to protease degradation in the same conditions. To measure this, mSec18 was incubated with increasing concentrations of trypsin with and without C8-PA addition. As expected, mSec18 sensitivity to trypsin degradation increased in the presence of C8-PA, whereas the presence of DAG had no effect (Fig.  4, A and B).
Additionally, we performed the same limited proteolysis assay using thrombin in place of trypsin. Thrombin displays much higher specificity than trypsin and should only cleave proteins at specific recognition sites. Incubation of Sec18 with thrombin alone showed no proteolytic degradation of the protein, indicating that no recognition sites were accessible to the protease. However, upon addition of C8-PA thrombin was able to cleave Sec18 (Fig. 4, C and D). Once again, inclusion of C8-DAG did not show a similar effect indicating once again that the observed conformation change was PA specific. Finally, we titrated C8-PA into a thrombin cleavage assay keeping the concentration of the protease constant. Cleavage of Sec18 by thrombin showed dose dependence for C8-PA (not shown). These data illustrate that C8-PA binding to Sec18 alters the conformation of the protein allowing for the exposure of an otherwise shielded thrombin-recognition site. Sec18 has one predicted thrombin-recognition site (after Arg-638), which is

Phosphatidic acid and Sec18 architecture
located in the D2 domain of the protein (exPASy). The D2 domain is responsible for the multimerization of Sec18 to its active hexamer when it is in a nucleotide-bound state (37,38). This further suggests that PA alters the conformation of the Sec18 D2 domain, or potentially the conformation of D2 with respect to D1 allowing binding to PA. Changes to the D2 domain structure could alter nucleotide binding or disrupt key interactions between protomers thereby decreasing Sec18 hexamer formation. Sec18 is known to associate with cis-SNAREs in its active hexameric form, so inhibition of hexamer formation could decrease its ability to properly recruit to inactive SNARE complexes. This idea is consistent with previous observations that showed increased PA at the vacuole led to decreased recruitment of Sec18 to cis-SNARE complexes (13).

Phosphatidic acid has no significant effect on the secondary structure of Sec18
Because we observed significant changes in the conformation of Sec18 upon binding to C8-PA we next wanted to monitor changes in the secondary structure of the protein when bound to the lipid. To do this we observed the ␣-helix and ␤-sheet content of Sec18 in the presence of PA using circular dichroism (CD). CD spectra of mSec18 were obtained in the absence and presence of C8-PA to determine if the protein's secondary structure was significantly affected by binding the lipid. The CD spectrum obtained for mSec18 alone showed that the protein was well folded (Fig. 4E). Upon addition of C8-PA, no significant changes were seen in the . C8-PA alters the proteolytic cleavage profile of Sec18, but does not affect secondary protein structure. Sec18 -His 8 was incubated with C8-PA, C8-DAG, or alone before incubation with increasing concentrations of trypsin (A) or thrombin (C). Densitometry values of the un-cleaved band were measured for each concentration and normalized against the input lane for trypsin (B) and thrombin (D). Deg., degradation products. E, CD spectra were measured (260 to 200 nm, 50 nm min Ϫ1 ) for Sec18 -His 8 in the presence and absence of C8-PA (100 M). F, intrinsic tryptophan fluorescence was measured with Sec18 -His 8 (500 nM) incubated with increasing concentrations of C8-PA and fluorescence spectra were measured (excitation 295, emission 300 -400 nm). The fluorescence (emission 333 nm) for each concentration tested was normalized against the no lipid control and is shown. SDS (0.25%) was used as a control for changing tryptophan fluorescence upon loss of secondary structure. G, differential scanning fluorimetry first derivative melting curves were measured (SYPRO orange: excitation 490, emission 560 nm) for increasing concentrations of C8-PA. n ϭ 3 for all data shown. M r , relative molecular mass markers. *, p Ͻ 0.05; **, p Ͻ 0.01; ***, p Ͻ 0.001.

Phosphatidic acid and Sec18 architecture
spectrum suggesting the lipid binding does not alter secondary structure features within the protein.
To rule out any denaturation caused by binding of C8-PA to Sec18, intrinsic tryptophan fluorescence was measured with and without lipid addition. Sec18 contains three tryptophan residues (Trp-88, Trp-91, and Trp-632) in its N and D2 domains. Upon denaturation of Sec18 with SDS, Trp fluorescence was red-shifted and showed decreased intensity (Fig. 4F). Upon incubation with C8-PA, no shift or intensity change was observed. This suggests that PA binding to Sec18 did not lead to denaturation, i.e. causing a conformational change large enough to alter the local environment of any of the Trp residues found in the protein.
Finally, to test whether binding PA altered the thermal stability of Sec18, we used differential scanning fluorimetry (39). Sec18 was labeled with SYPRO orange dye, incubated with different concentrations of C8-PA in separate wells, and equilibrated prior to starting a melting curve. Fluorescence was scanned across a temperature gradient of 20 to 95°C and the first derivative of the fluorescence data were used to determine the T m for each condition. Differential scanning fluorimetry has the ability to show multiple melting transitions (40,41). Our data show that mSec18 has three melting transitions. The first mSec18 transition (T M1 ) occurred at ϳ45°C, whereas T M2 and T M3 were at 60 and 64°C, respectively (Fig. 4G). The addition of C8-PA had no effect on T M2 and T M3 , as the curves overlapped with the that of apo-Sec18. That said, C8-PA has a striking effect at T M1 where we observed a dose-dependent increase in fluorescence. This likely mirrors the conformational changes seen with limited proteolysis and ANS fluorescence. Taken together these observations lead us to conclude that PA binding to Sec18 induces a significant change to the architecture of the protein but does not denature the protein nonspecifically.

NSF D1-D2 undergoes large conformational change during transition between hexameric and monomeric forms
To examine the Sec18 conformational changes we observed previously at a more detailed level, atomic molecular dynamics (MD) simulations were performed using NSF, the mammalian homolog of Sec18. The NSF D1-D2 monomer extracted from the cryo-EM structure of an ATP-bound NSF complex (PDB 3J94) after removing bound ATPs was equilibrated with restraints for 20 ns and then relaxed for 200 ns (Video S1). Based on the overall ␣ carbon (C␣) r.m.s. deviation, the monomer undergoes conformational changes up to 15 Å apart from the form originally adopted in the hexamer (Fig. 5A). Calculation on the secondary structure components showed that only the modeled loop region from residue 458 to 478 transitioned from the helix during the relaxation to turn and coil (data not shown). This is expected as the loop was poorly resolved in cryo-EM and was only stabilized by interactions with the N-domain in the template crystal structure (the N-D1 domain of p97) used in homology modeling. The stable secondary structure observed in D1 and D2 domains indicated that the large deviation did not come from secondary structural changes, further verifying CD experiments. Instead, we observed that the conformational change was accompanied by an opening-up process of D1 and D2 domains during the relaxation. The observation was in agreement with the hypothesis that NSF hexamerization might require certain conformations of D1-D2 monomer and that the conformation required could be further stabilized at the hexamer interface. Fig. 5B shows the equilibration of individual protomers in a NSF hexamer. These data indicate that the hexamer is more stable (Video S2).

Residues of NSF shown to bind to C8-PA are not available for PA binding when it is in the hexameric form
Computational flooding studies were performed for both monomeric and hexameric forms of the NSF D1-D2 domains, based on the structural information of mammalian NSF (Fig. 6A). The binding affinity of each amino acid residue was analyzed using percent bound as determined by the fraction of time when PA lipids were found in proximity (minimal distance between phosphate group atoms of PA and the amino acid residue below H-bond distance). Residues determined to have highest percent bound were determined for both monomeric and hexameric forms of D1-D2. Our flooding simulations of NSF hexamer showed that residues having the highest percent bound PA in the monomer (Fig. 6B) were shielded to block lipid binding in the hexameric NSF D1-D2 construct (Fig. 6C). This suggests that PA binding specificity lies somewhere within the hexameric interface.

Binding prediction and clustering analysis of PA-binding regions of NSF
Ensemble molecular docking of C8-PA to NSF monomer was performed using the aforementioned D1-D2 equilibrium simulation (42). Snapshots from the equilibrium trajectory were utilized for molecular docking every 100 ps to fully sample conformational dynamics. The resulting docked C8-PA poses were clustered (43) and an average Autodock Vina scoring function scores were determined for the clusters of monomeric NSF. These scoring function results show some correlation with the MST binding measurements of mSec18 and hSec18 to C8-PA (44) (Fig. 7, A and E). To further verify the cluster analysis, a SiteMap analysis was performed and the top three site scores were chosen (Fig. 7D). Both Fig. 6, A and D, indicate affinity to two main regions of Sec18, mainly in the interface of D1 and D2, as well as a significant affinity to the D2 ATP-binding site (Fig.  7, B and C). Furthermore, to fully evaluate the potential PAbinding sites of Sec18, the hexameric form was also surveyed, using the same methodology as the monomer. The results showed a preferential docking to D1 (green) for the hexamer (Fig. 7, E and F).

Molecular dynamics of NSF-bound PA may indicate a hinge mechanism
Molecular dynamics simulations were selectively performed on clusters 3 and 4 (purple and orange), which are found in the interface between D1 and D2, not overlapping with the D2 ATP-binding site. From each of the two clusters a top pose was filtered using the highest Autodock Vina scoring function score for simulation (Videos S3 and S4). These simulations indicate flexibility between the D1-D2 interface (Fig. 8). These simulations show either an opening or closing of D2 with respect to D1 as demonstrated by overlapping trials. These results further support the mechanism as described in Fig. 9.

Discussion
Membrane fusion is a necessary process for all eukaryotes, and Sec18/NSF is the only known protein responsible for uti-lizing energy from ATP to prime SNAREs (3,4,8). To achieve compartmental specificity, unique SNARE combinations are utilized by defined organelles as well as smaller transport vesi-

Phosphatidic acid and Sec18 architecture
cles budding from such organelles (45). Each organelle varies in both size and function, and must contain its own unique combination of protein and lipid factors to allow for specificity in trafficking and membrane-fusion events. Regulation of Sec18/ NSF is of special significance due its direct role in maintenance of fusion and compartmentalization throughout the eukaryotic cell. Therefore, it is important to understand the role that regulatory factors have on ubiquitous fusion machinery such as Sec18/NSF to adequately model how specificity and efficiency are balanced and maintained at different locations in the cell.

Phosphatidic acid and Sec18 architecture
Protein function can be regulated directly through posttranslational modifications or through their interactions with other molecules, including lipids. The vacuole fusion pathway is regulated at various stages by distinct lipids such as phosphoinositides, ergosterol, DAG, and PA (13, 17, 27, 30, 39, 46 -52). The priming stage requires the presence of ergosterol, PI(4,5)P2, as well as the conversion of PA to DAG (13,17,30,49).
Previously we found that vacuolar PA sequestered Sec18 from cis-SNAREs and that the PA phosphatase Pah1/Lipin1 was required to convert PA to DAG to allow Sec18 dissociation from the membrane and recruitment to SNARE complexes (13). We should also reiterate that the other yeast PA phosphatases have no effect on vacuole fusion, illustrating that this reg-ulation is specific for Pah1 function. Although PA turnover is needed for priming, the presence of the lipid is also required downstream for mechanisms that remain to be characterized. Deletion of PAH1 or the DAG kinase DGK1 alters the balance of PA and DAG on vacuole to dramatically affect membrane fusion (17,27). We thus postulate that enzymatic changes that alter PA levels can in turn shift the equilibrium of Sec18 from a lipid-bound to a SNARE-associated state. Such changes would likely have significant effects on SNARE disassembly and the overall progression of the membrane fusion cascade. That said, any membrane where PA is lacking would be expected to have unfettered access Sec18 to SNAREs.
In this study we demonstrated that Sec18 directly binds PA with high affinity on par with a known PA-binding domain.  Fig. 6A including ATP aligned into ATP-binding sites of D1 and D2 domains. D, Schrodinger generated image as in Fig. 6B including ATP aligned as in Fig. 6C. E, VMD generated image of NSF hexamer as in Fig. 6A including autodock generated cluster analysis. F, Schrodinger generated image of NSF hexamer as in Fig. 6B including a sitemap generated binding site analysis.

Phosphatidic acid and Sec18 architecture
Moreover, only monomeric Sec18 could bind both long-chain PA in membranes and soluble C8-PA, whereas hexameric was only able to bind C8-PA. This signifies that C8-PA could access PA-binding residues that are blocked in the hexamer to prevent membrane association. Our findings indicate that Sec18 may exist in both a monomeric lipid-bound pool and SNARE-bound hexamers. Because ATP is required for Sec18 hexamerization, we tested PA liposome binding in the presence of ATP. PAliposome binding by Sec18 was blocked by ATP. We posit that micromolar ATP concentrations may shift the monomeric pool of Sec18 used to a hexameric pool decreasing its affinity for PA. It is worth noting that ATP concentrations in the cytoplasm have been measured in the low millimolar range, which would likely promote spontaneous hexamer formation and maintenance at the vacuole membrane (53). This would suggest Sec18 almost always exists as an active hexamer under physiological conditions. However, we have previously observed disruption of Sec18 activity upon a shift to high levels of PA at the vacuole even in the presence of millimolar concentrations of ATP (13,17). Additionally, Sec18 association with the vacuole membrane is maintained throughout the fusion cycle, even after it hydrolyzes ATP and allows for the release of Sec17 from the membrane (3). Sec18 binding to the vacuole is also required for membrane association of the fusion factor LMA1, even in the absence of ATP (54). These findings agree with our results and suggest Sec18 may associate with the vacuole membrane in a nucleotide-free state. It is, however, unclear whether some unknown interaction between Sec18 and a fully functional vacuole would limit, or regulate, ATP binding prior to its release upon Pah1 activity. Given our results and these previous findings, we hypothesize that PA at the vacuole membrane stabilizes a nucleotide-free, monomeric form of Sec18 before its recruitment to cis-SNAREs. Our evidence for this, however, is indirect and future work should aim to investigate the presence of a monomeric form of Sec18 in vivo.
During priming, Sec17/␣-SNAP is recognized by Sec18/NSF in an ATP-bound state at D1 before subsequent ATPase activity occurs. We think that Sec18 exists in both lipid-bound and SNARE-bound states and that the presence of ATP at the D1 NBD may determine the state in which the protein primarily exists. Membrane PA may prevent the association of ATP with the D1 NBD locking the protein in an inactive lipid-bound state preventing recruitment to inactive SNARE complexes. This is in line with our data in this study and with observations from previous work (13).
The fact that Sec18 monomer binding to PA liposomes was inhibited at a saturating ATP concentration for the D1 NBD could indicate that the PA-binding site for Sec18 lies near the D1 ATP-binding site. Alternatively, it is possible the conformation of Sec18 in its ATP-bound state shields the protein's unique PA-binding site. The idea that Sec18 binding to PA may not specifically depend on the D1 ATP-binding site was supported by computational flooding experiments performed on both hexamer and monomer in the presence and absence of ATP. Flooding experiments allowed for C8-PA to equilibrate  Binding PA causes a conformational change in the protomers between the D1 and D2 domains to stabilize membrane association. The PA phosphatase Pah1 hydrolyzes PA to diacylglycerol leading to a conformational reversion that is compatible to dissociation from the membrane and incorporation into a Sec18 holoenzyme.

Phosphatidic acid and Sec18 architecture
with NSF monomer, and binding was measured using the length of time a PA molecule resided near a given residue of NSF. Many of the long-term amino acid residues sharing the longest contact time to PA were predictably basic residues, especially lysine and arginine. However, dramatic differences in these residues were not noticed between the ATP and non-ATP simulations. Furthermore, many of the residues with longer PA-binding time were not of importance for PA binding in the hexamer simulation. This result is in corroboration with the high binding affinity of Sec18 monomers to PA liposomes versus the hexameric form. This further indicates that the Sec18 monomer and hexamer are differentially regulated. Furthermore, it suggests that PA may influence the formation of the activehexamerbycontrollingtheavailabilityofitsinactivemonomer at membranes.
We propose that Sec18/NSF PA regulation is achieved by sequestration of its protomers by PA to block the formation of the active hexamer to prevent unchecked priming. Upon binding PA, Sec18/NSF undergoes a significant conformational change that coincides with a reduction in its SNARE priming activity. Thus, PA sequestration can negatively regulate SNARE priming (55). Additionally, it is possible that PA at the site of priming could have a globally positive influence on priming by increasing the local concentration near the site of action. Previous work has shown that PA is necessary for vacuole fusion to occur and is required for Sec18 association with the membrane (56). Although this appears to somewhat contradict our findings presented here, we do not believe this is the case. Unregulated Sec18 activity has previously been shown to be detrimental to vacuole fusion (57)(58)(59)(60). Factors such as the PA phosphatase Pah1/Lipin, could thus serve to activate Sec18/NSF only once its activity was required (13,17). In this way, PA could serve as a temporal regulator of SNARE priming activity and the membrane fusion process as a whole. It is also likely that PA membrane concentrations could differentially affect organelles with distinct lipid content. Depending on the concentration and localization of PA at a given membrane Sec18 sequestration by PA could either play a larger or less prominent role in regulating the priming of SNAREs.
Based on our computational studies, it appears that there are numerous candidate residues that might contribute to Sec18 PA binding. Membrane simulations have been performed (data not shown); however, due to the size and flexibility of Sec18 monomer, long-time scales in the microsecond range may be required to show final binding events sequestering Sec18 to a PA containing membrane. We plan to further probe this binding event using membrane simulations at longer time scales to capture the exact binding event of Sec18 to a PA membrane, and to specifically identify the numerous residues that may be involved.
Because NSF binds PA, we assume that the results from yeast Sec18 will translate to the mammalian system. Our computational results using NSF indicate a conformational change in a novel region of NSF that would not be expected merely from the function of D1 ATPase activity on the N-domain. Our model suggests that PA may influence the ability for NSF to both localize and polymerize to form an active hexameric priming complex. This priming complex could then be differentially regulated at a given organelle utilizing PA as a regulatory lipid, as indicated by our studies of Pah1 at the vacuole (17).

Plasmid construction
Plasmid for expression of Sec18 -His 8 was created by amplification of SEC18 by PCR from genomic DNA of yeast strain DKY6281 using primers containing NdeI and XhoI restriction cut sites (forward: 5Ј-ACGTACGTCATATGTTCAAGATAC-CTGGTTTTGG-3Ј, reverse: 5Ј-ATCGAATGCTCGAGTG-CGGATTGGGTCATCAACT-3Ј). The PCR product was inserted into pET42a using NdeI and XhoI in-frame with a C-terminal His 8 tag sequence under control of a T7 promoter to create pSec18H8.

Nanodisc preparation
Lipid composition of PA nanodiscs consisting of 3.023 mol of POPC diC 16 , 0.098 mol of PA diC 16 , and 0.78 mol of POPE, and PC nanodiscs consisting of 3.121 mol of POPC diC 16 and 0.78 mol of POPE were combined, dried, and des-iccated overnight. Lipids were then dissolved in 20 mM sodium deoxycholate in TBS (50 mM Tris-HCl, pH 7.4, 150 mM NaCl, and 0.02% NaN 3 ) and sonicated. MSP1D1 membrane scaffold protein (MSP) was then added in a ratio of 70:1 lipid to protein and detergent was removed with Bio-Beads SM-2 (Bio-Rad). Nanodiscs were isolated using size exclusion chromatograph and quantified using a NanoDrop and the extinction coefficient of 21,000 liter mol Ϫ1 cm Ϫ1 for MSP1D1 (24.66 kDa), and the resultant mg/ml divided by two because there are two MSP proteins per nanodisc (67).

Surface plasmon resonance
SPR measurements were performed on a Biacore T200 instrument equipped with an Ni-NTA chip (68). Approximately 2000 response units of 5% PA nanodiscs were immobilized noncovalently using 1 M NiSO 4 flowed at 10 l/s followed by a blank buffer injection of HEPES, pH 7.4, 150 mM NaCl (HBS buffer). Injections were performed in HBS buffer at a flow rate of 30 l/min with an association time of 90 s, dissociation time of 300 s, and binding was measured in relative response units as described (67). Regeneration with EDTA was performed at flow rate 30 l/s for 120 s using 100 nM EDTA buffer. Proteins were injected using 1:1 dilutions from the highest concentration and steady state was obtained using GE BIAcore T200 evaluation software version 3.0 (BIAevaluate). Proteins were injected using 1:1 dilutions for the Sec18 monomer

Microscale thermophoresis
Thermophoresis measurements were performed using a Monolith NT.115-labeled thermophoresis machine (69). Sec18 -His 8 was labeled with Ni-NTA Atto 488 according to the manufacturer's protocol mixing 200 nM protein with 100 nM dye and allowing to sit at room temperature for 30 min followed by centrifugation. M.O. Control software was used for operation of MST. Target protein concentrations were 50 nM for all His tag-labeled proteins Sec18 monomer, Sec18 hexamer, PA nanodiscs, and PC nanodiscs. LED excitation power was set to 90% and MST set to high allowing 3 s prior to MST to check for initial fluorescence differences, 25 s for thermophoresis, and 3 s for regeneration after MST off. Analysis was performed using M.O. Affinity Analysis Software as the difference between the initial fluorescence measure in the first 5 s as compared with thermophoresis at 15 s. All measurements were performed in PBS buffer (137 mM NaCl, 2.7 mM KCl, 8 mM Na 2 HPO 4 , and 2 mM KH 2 PO 4 , pH 7.4) without Tween and binding affinity was generated using GraphPad Sigmoidal 4PL fit from points exported from M.O. Affinity Analysis software using K D Model with target concentration fixed at 50 nM generating bound, unbound, and fraction bound for export to GraphPad to estimate the final K D .

Phosphatidic acid and Sec18 architecture Limited proteolysis
Cleavage reactions were carried out in proteolysis buffer (20 mM HEPES, pH 7.2, 150 mM NaCl, 2 mM ATP, 2 mM MgCl 2 ). Sec18 -His 8 (2 M) was added to proteolysis buffer and incubated with the indicated lipid concentration on ice for 5 min. Trypsin or thrombin diluted in 1ϫ HBS was added to assay tubes at the indicated concentrations and incubated at 25°C for 30 min. Cleavage reactions were stopped with the addition of SDS sample buffer containing 1 mM PMSF. Samples were resolved with SDS-PAGE and gels were stained using Coomassie Blue. Gels were destained with methanol/acetic acid solution (50%/7%) and imaged using a ChemiDoc MP Imaging System (Bio-Rad).

Tryptophan fluorescence spectroscopy
Sec18 -His 8 (500 nM) was incubated with the indicated concentrations of C8-PA in fluorescence assay buffer (20 mM HEPES, pH 7.2, 150 mM NaCl, 1 mM MgCl 2 , 1 mM ATP). Lipid dilutions were first prepared in assay buffer and measured for background fluorescence before Sec18 -His 8 was added and incubated at 25°C. Intrinsic tryptophan fluorescence measurements were made using a fluorimeter with Peltier temperature control (Agilent Technologies). Samples were excited at 295 nm and the emission spectra were collected from 300 to 400 nm. Samples were measured in a 100-l cuvette (Starna Cells). Initial background fluorescence spectra for each lipid concentration were subtracted from final measurements.

1,8-ANS fluorescence spectroscopy
ANS binding experiments were carried out in fluorescence assay buffer with 5 M ANS (Cayman Chemical). Initial spectra were taken without Sec18 -His 8 to measure any background fluorescence from buffer or added lipids (excitation 350 nm, emission 390 -620 nm). Sec18 -His 8 diluted in assay conditions was then added to the assay to the indicated concentration and incubated at 25°C for 5 min before spectra were obtained. Initial background fluorescence spectra for each lipid concentration were subtracted from final measurements.

Circular dichroism
Monomeric Sec18 -His 8 purified in phosphate buffer was incubated with and without C8-PA to equilibrium (25°C, 15 min). Protein concentration used was 5 M and lipid concentration used was 100 M. CD was measured using a spectropolarimeter (JASCO). All spectra were recorded from 260 to 200 nm at 50 nm min Ϫ1 and measurements were taken in a 1-mm path length cuvette.

Differential scanning fluorimetry
Sec18 (2.75 mg/ml) was diluted to a final concentration of 0.11 mg/ml in phosphate buffer containing 1 mM ATP, 1 mM MgCl 2 , and 4ϫ SYPRO orange dye. Next, 22.5 l of this mix was added to a white hard-shell 96-well PCR plate (Bio-Rad), which contained 2.5 l of serial dilutions of C8-PA in phosphate buffer. The plates were then sealed with Microseal "B" film (Bio-Rad), and samples were allowed to equilibrate at room temperature for 30 min before beginning the assay. Melting curves were performed using a Bio-Rad CFX Connect real-time detection system. The melt curve protocol was 25°C for 3 min followed by a 25-90°C gradient with 0.5°C increments. Each temperature was held for 10 s and the fluorescence intensity was measured (excitation ϭ 490 nm, emission ϭ 560 nm). The first derivative of the fluorescence readings was used to determine the melting temperature(s) for each condition.

Preparation of D1-D2 monomer and hexamer models
The D1-D2 monomer model (residues 215-737) was derived from an Cryo-EM structure of ATP-bound NSF complex (PDB 3J94, chain A) (8). Missing residues (335-346, 458 -478 in PDB 3J94 (chain A)) were built via homology modeling using the crystal structure of the homologous N-D1 domain of p97 (PDB 1E32) as a template by MODELLER 9.19 (70). The complete D1-D2 hexamer model was prepared (71) using the same PDB 3J94 as the monomer. Missing loops in each monomer were modeled in CHARMM GUI to ensure that no clashes or topological errors exist in the complex structure. cis-Peptide bonds in both monomer and hexamer structures were examined and fixed manually using Cispeptide plugin in VMD (72). A further refinement of loops built in the hexamer was performed via MDFF (73).

Equilibrium MD simulations of D1-D2 monomer and D1-D2 hexamer
The MD simulations were performed with NAMD 2.12 (74) using CHARMM36m force field (75). Langevin dynamics and Langevin piston Nose-Hoover methods (76,77) were used to maintain constant temperature at 310.15 K and pressure at 1 atm. The long-range electrostatic forces were evaluated using the particle mesh Ewald method (78, 79) with a 1-Å grid spacing. The van der Waals interactions were calculated with a cutoff of 12 Å and a force-based switching scheme after 10 Å. Integration time step was set at 2 fs with SETTLE algorithm (80) applied. VMD 1.9.3 was used for MD trajectory visualization and analysis (81). The D1-D2 monomer model was first equilibrated for 20 ns with harmonic restraints (0.05 kcal/mol/Å2) on protein C␣ atoms except modeled loops, then followed by 200-ns equilibration without restraints. Furthermore, the D1-D2 hexamer was modeled as for D1-D2 monomer, and simulated for 50 ns without restraints, this was utilized for ensemble docking of PA.

PA lipids flooding simulations of D1-D2 monomer and D1-D2 hexamer
To demonstrate that hexamerization of sec18 monomer shields PA-binding sites, three independent PA lipids flooding simulations were carried out for: 1) D1-D2 monomer in the absence of ATP molecules; 2) D1-D2 monomer in contacts with four ATP molecules taken from the hexamer structure; and 3) D1-D2 hexamer with all bound ATP molecules. Flooding simulations were prepared by first placing the protein in a PA lipids grid with a grid spacing of 25 Å, where PA lipids were modeled with truncated acyl chains as described in the highly mobile membrane-mimetic model (84,85) to reassemble the C8-PA experiments as well as avoid micelle formation, followed by solvation and ionization to a NaCl concentration of 150 mM Phosphatidic acid and Sec18 architecture using the SOLVATE and AUTOIONIZE plugins within VMD (81). The two final D1-D2 monomer systems contained 61 PA lipids in a simulation box of size 95 Å ϫ 94 Å ϫ 120 Å, whereas the D1-D2 hexamer system contained 223 PA lipids in a simulation box of size 188 Å ϫ 187 Å ϫ 133 Å, resulting in a similar PA lipids concentration (ϳ120 mM) in all three systems. D1-D2 monomer systems were simulated for 350 ns each and the hexamer system was simulated for 166 ns, recorded every 20 ps each. Harmonic restraints (0.1 kcal/mol/Å 2 ) were applied on protein C␣ atoms except for modeled loops as well as the ATP molecules throughout the simulation to prevent conformational changes of the protein or the disassociation of ATP. The PA binding affinity of each amino acid residue was evaluated by calculating the fraction the time that any PA phosphate group atom can be found within 3 Å of this residue (hydrogen atoms not included).

Binding site probing of NSF for PA
To characterize C8-PA and D1-D2 monomer interactions, molecular ensemble docking of PA was done on D1-D2 monomer using AutoDock Vina (42). The previously mentioned equilibrium simulation of D1-D2 was used to fully sample the dynamics of D1-D2 for molecular docking, where snapshots were taken every 1000 ps of the 200-ns trajectory. For each snapshot, an 80 Å ϫ 94 Å ϫ 108 Å grid box was used to fully sample the entire structure. Each snapshot was docked with an exhaustiveness of 10, yielding a total of 2000 PA-docked poses, with the affinities of each pose obtained from the resultant log files. These poses where then clustered using a hybrid K-centers and K-medoids clustering algorithm using the root mean square deviation (r.m.s. deviation) method (43,82) for which three main clusters where identified. These clusters where then compared with SiteMap (83). Schrodinger SiteMap was used on equilibrated D1-D2 NSF monomer indicating top potential ligand-binding regions of the NSF D1-D2 monomer including shallow binding sites. The same protocol of ensemble docking of PA was done on the 50-ns simulation of hexameric D1-D2, with snapshots taken every 100 ps. A grid box size of 135 Å ϫ 135 Å ϫ 135 Å with a search exhaustiveness of 10 was used, yielding a total of 5000 PA docked poses that were then clustered using the same methodology as the monomer.

MD simulations of top poses from ensemble scoring function
To further probe the effect of C8-PA on D1-D2 conformation, MD simulations were performed as for equilibrium simulation using both NAMD and CHARMM36m force field. Poses from each cluster provided by the ensemble docking from AutoDock Vina with highest scoring function score were selected for simulation. Monomer MD simulations of 100 ns were performed for each of the top poses from each cluster in the same solvent as for flooding containing 150 mM NaCl in water. VMD was used to visualize results and create figures.

Data analysis and statistics
Results are expressed as the mean Ϯ S.E. Experimental replicates (n) are defined as the number of separate experiments with different batches of protein, liposomes, and nanodiscs. Where appropriate, significant differences were calculated using two-tailed unpaired t-tests. p values Յ0.05 were considered significant.