Characterization of protein interactions in the structures

Our cryo-EM work represents the experimentally determined atomic model of the ATP-bound Vps4^E233Q hexamer with or without the cofactor Vta1, allowing visualization of specific interactions governing Vps4 oligomerization in its active state. Since the structure of the Vps4^E233Q hexamer present in the Vps4^E233Q–Vta1 complex is very similar with that in the Vta1-free Vps4^E233Q hexamer, we next only investigated the interface in the Vta1-free Vps4^E233Q hexamer in detail.

When one Vps4^E233Q molecule from each pair of the adjacent subunits aligned with each other, the other one is also well aligned, except for the pairs involving subunit A (Fig. 2a). This is expected as the interface areas between subunit A and its neighbouring subunits are much smaller as described above (Supplementary Fig. 2a). Therefore, we mainly focused on the analysis of the four similar interfaces and discussed interface between subunits C and D as a representative.

Open in a separate window

Figure 2

Characterizations of protein interactions in the Vta1-free and the Vta1-bound Vps4^E233Q hexamers.

(a) Comparison of six pairs of the adjacent Vps4^E233Q subunits. The colouring scheme is the same as in Fig. 1c,e. When one Vps4^E233Q molecule (shown in ribbon representation) from each pair of the adjacent subunits aligned with each other, the other one (shown in cartoon representation) is also well aligned, except for the pairs involving subunit A (red arrow). (b) A focused view of the domains that mediate the interactions between the Vps4^E233Q monomers (subunits C and D). Three boxed intermonomer interfaces are detailed in c–e. (c–e) The close-up views of the interfaces between the large ATPase domains (c), the large and small ATPase domains including the β-domain (d) and the C-terminal helixes (e). Potential residues participated in the interactions are represented by spheres. Hydrophobic, aromatic, positively charged and negatively charged residues are coloured grey, pink, blue and red, respectively. The orientations of all close-up views are optimized to show the interactions. (f–h) SEC analyses of the oligomeric states of the Vps4^E233Q mutants targeting the interface between the large ATPase domains (f), the large and small ATPase domains including the β-domain (g) and the C-terminal helixes (h). (i) A focused view of the interactions between the Vta1 CTD dimer (subunits I and J) and the adjacent Vps4^E233Q subunit pair (subunits C and D). Two different contact modes are highlighted by purple and blue dashed circles, respectively, and detailed in j,k. (j,k) The close-up views of the interactions between the Vps4^E233Q subunit D and the CTD dimer (j), and the Vps4^E233Q subunit C and the CTD dimer (k). (l,m) Assessments of the interactions by SEC between Vta1 and Vps4 mutants, E233Q and E233QR325A, at high concentration (100μM) in the absence of ATP (l), and at low concentration (5μM) in the presence of ATP (m). (n) Assessment of the stabilities of the Vta1-free Vps4^E233Q hexamer and the Vta1-bound Vps4^E233Q hexamer formed by different Vps4 mutants with or without apyrase treatment by SEC (See Methods for details).

The inter-subunit interactions occur both between the large ATPase domains and between the large ATPase and the small ATPase domains including the β-domain (Fig. 2b). The large ATPase domain interaction contributed by helix α3 and α4 from one subunit, and the loops between strand β2 and helix α3 and between strand β3 and helix α4 from neighbouring subunit, largely composed of polar residues, indicating that electrostatic or hydrogen-bonding interactions dominate this interface (Fig. 2b,c). The interaction between the large and small ATPase domains including the β-domain involves the E147-R163 region comprising the C-terminal half of helix α1 and the following loop from one subunit, and the C-terminal half of helix α8 and the Trp388-Glu398 region of β-domain from the neighbouring subunit (Fig. 2b,d). Atomic interactions at this interface are versatile, involving polar, hydrophobic and aromatic residues, suggesting a combination of hydrogen bonds, salt bridges and van der Waals contacts (Fig. 2b,d). Additional contact occurs between residues in the C-terminal helix of one subunit and the corresponding amino acids in the linker region between helix α9 and C-terminal helix of the neighbouring subunit, exemplified by potential hydrogen bonds or salt bridges between Asp431 and Arg414 (Fig. 2b,e). To validate the above interfaces and residues involved in oligomerization of Vps4, we constructed Vps4 mutants based on the E233Q mutant: D201A, K215A, R241A and E247 for the large ATPase domain interactions (Fig. 2c); E147A, L151A, K154AF155A, L158AF159A, R163A, I351AR352A, W388A, L396A and E398A for interaction between the large and small ATPase domains including the β-domain (Fig. 2d); and R414A and D431A for the C-terminal helix interactions (Fig. 2e). All of these mutants can be purified but eluted on SEC with eliminated or substantially reduced amount of hexamers in the presence of ATP (Fig. 2f–h). Importantly, all of these residues are highly conserved in eukaryotes, suggesting a conserved mechanism of Vps4 oligomerization (Supplementary Fig. 7a). Previous studies based on the Vps4 hexamer models derived from the p97 hexameric structure and the crystal packing interactions in Vps4 crystal forms have discovered some residues, L151 and R352, essential for the oligomerization^8,12,21. These interactions are also observed in our structure and are all at the interface between the large and small ATPase domains (Fig. 2b,d,g). Moreover, our structure discovered that the interactions between the large ATPase domains and between the C-terminal helices are also important for the hexamer assembly (Fig. 2b,c,e,f,h).

To further evaluate the functional relevance of the ATP-bound Vps4^E233Q hexamer structure, we mutated subunit interface residues that substantially affect the Vps4^E233Q assembly and compared the ATPase activities of these mutants with the wild-type Vps4. As shown in Supplementary Fig. 8, all of the mutants showed severely decreased ATPase activities to 20% of the wild-type activity and most of them reduced to ~10%, suggesting that the key residues identified in affecting Vps4^E233Q assembly based on ATP-bound Vps4^E233Q hexamer structure are also important for the ATPase activity of the wild-type Vps4 protein. These results confirmed the functional relevance of the subunit interface discovered by the ATP-bound Vps4^E233Q hexamer structure.

We next investigated the structure of the Vta1-bound Vps4^E233Q complex to understand the mechanism of Vta1-promoting activity on Vps4. Vta1 CTD dimer bridges between two adjacent Vps4^E233Q subunits in two different modes (Fig. 2i). One mode is guided by the interactions of the β-domain and helixes α7 and α9 of the small ATPase domain from one Vps4^E233Q subunit with the regions around the tip of α8/α9 helix hairpin from the outer CTD of one CTD dimer and the N-terminal half part of the helix α8 from the inner CTD of the same dimer, respectively (Fig. 2i,j). This interaction is consistent with the crystal structures of the Vta1 VSL dimer in complex with a Vps4 fragment¹⁴ and the Vta1 CTD dimer¹⁹, and a model generated by aligning these two crystal structures²⁶ as reported previously. The other interaction mode, completely different from that observed before, occurs between the helix α7 of the neighbouring Vps4^E233Q subunit and the α8/α9 helix hairpin tip of only the inner CTD, exemplified by potential polar contacts between Arg325 of Vps4, and Glu311 and Asp312 of Vta1 (Fig. 2i,k). The inner CTD binds to both of the adjacent Vps4^E233Q subunits simultaneously and hence is more stable than the outer CTD that binds to only one Vps4^E233Q subunit (Figs 1e and ​and2i).2i). Also, the tip of α8/α9 helix hairpin from each CTD in one CTD dimer makes contribution to both of the binding modes by interacting with different regions of the adjacent Vps4^E233Q subunits (Fig. 2i–k).

To corroborate the structural observations, we generated Vps4 mutant E233QR325A and examined its ability to associate with Vta1. Similar to the E233Q mutant, the E233QR325A mutant ran as a dimer without ATP and a hexamer in the presence of ATP on SEC, indicating this mutation does not affect the oligomerization ability of Vps4 (Supplementary Fig. 9). In the absence of ATP, both Vps4 mutant constructs can bind to Vta1 via the β-domain only at high concentration due to the weak interaction between Vps4 β-domain and Vta1 VSL as manifested by the relatively high Kd value of 85μM (Fig. 2l)¹⁴. When ATP is present, Vps4^E233Q assembles into hexamer forming an additional site for Vta1 to bind as revealed by our structure, and therefore should enhance the binding of Vta1. Indeed, in the presence of ATP, Vta1 can form complex with Vps4^E233Q hexamer at a concentration as low as 5μM of Vps4 (Fig. 2m). In contrast, when R325A is introduced to the additional binding site, Vta1 high-binding affinity to Vps4^E233Q hexamer is lost (Fig. 2m). Together, these results demonstrated that the interaction between the helix α7 from Vps4^E233Q and the tip of α8/α9 helix hairpin from Vta1 is indispensable for efficient binding of Vta1 to the ATP-bound Vps4^E233Q hexamer.

We next examined the Vta1 effect on the stability of ATP-induced Vps4^E233Q hexamer. We collected the SEC fractions of Vta1-free and Vta1-bound Vps4^E233Q hexamers in the presence of ATP and reapplied them to SEC in the absence of ATP. Interestingly, we found that Vta1-free hexamers nearly all disassembled into dimers, whereas the Vta1-bound hexamers remained intact (Fig. 2n). As a control, treatment with apyrase to deplete ATP from the samples before loading to SEC led to full disassembly of both Vta1-free and Vta1-bound hexamers, indicating that ATP still binds to the Vta1-bound hexamer even in a solution lacking ATP once the complex forms (Fig. 2n). These experimental findings strongly argue that Vta1 binding may stabilize the conformation of the ATP-bound Vps4 hexamer, thus facilitating the bound ATP for subsequent hydrolysis. Accordingly, mutation construct E233QR325A of Vps4 hexamer at high concentration incubated with Vta1 disassembled immediately in a solution that lacked ATP (Fig. 2n). Notably, the residue 325 of Vps4 is highly conserved as a positively charged amino acid in eukaryotes, but diverse in crenarchaeota in which the Vta1 is absent, further indicating the importance of this residue for the function of Vta1 (Supplementary Fig. 7b).

The finding of our structure of the Vps4^E233Q–Vta1 complex is the identification of the second site, R325 of Vps4^E233Q, for the Vta1 binding. To examine whether this site is also important for the Vta1-promoting activity on wild-type Vps4, we measured the ATPase activity of wild-type and R325A mutant Vps4 in the presence of Vta1. As shown in Supplementary Fig. 8, Vta1 was unable to stimulate the ATPase activity of the R325A mutant, whereas the ATPase activity of the wild-type Vps4 was increased by twofold in the presence of Vta1. As a control, the ATPase activities of the wild-type and the R325A mutant Vps4 proteins alone were similar. These results confirmed the physiological relevance of the second binding site at the Vps4–Vta1 interface even though our structures were obtained using the E233Q mutant Vps4.

Conformational changes and ATPase active centres

In our structure, we can clearly visualize ATP at the nucleotide-binding pocket of Vps4^E233Q protein (Supplementary Fig. 4). This ATP-bound structure of Vps4^E233Q provides us the opportunity to compare with the other nucleotide-bound structures to understand the mechanism of the ATP-dependent oligomerization of Vps4. In comparison with the ATPγS-bound Vps4^E233Q (PDB ID: 3EIH)¹², the ATP-bound Vps4^E233Q in the hexamer (taking subunit D as representative) has conserved regions contributing to ATP binding including Walker A loop (Phe174-Tyr181), the two negatively charged residues in Walker B (Asp232, and Glu233 that is mutated to Gln233 in our structure), sensor 1 (Asn277) and M307 that can be well aligned with a r.m.s.d. value of 0.884Å over 12 Cα atoms (Fig. 3a–d). On the other hand, significant differences are observed in the position of the large ATPase domain, especially the surface-exposed helixes including helixes α3, α4 and α5 that are far away from the ATP-binding pocket of the subunit itself, exhibiting a significant movement between these two structures (Fig. 3a,b). Consequently, this change does not affect the conformation of the ATP-binding pocket of the subunit D (Fig. 3c), but instead leads to a relatively compact ATP centre of the neighbouring subunit E by positioning Arg289, Asn261 and Asn265 residues into close proximity to the ATP of subunit E (Fig. 3d), enabling comprehensive close contacts with the ATP to not only allow efficient assembly of the hexamer but also result in elevated ATPase activity. Indeed, mutation of Arg289 to alanine or mutation of both Asn261 and Asn265 to alanines led to reduced ATP-dependent oligomerization activity and the ATPase activity (Fig. 3e,f). Specially, the mutation of Arg289, the highly conserved arginine finger, nearly failed assembly into the hexamer and retained only 7% of the wild-type ATPase activity (Fig. 3e,f). This observation is in agreement with the previous studies of the functions of the arginine finger. The arginine finger residues have been studied extensively and implicated in ATP hydrolysis and oligomer assembly^27,28. As for the Vps4 protein, mutations of the arginine fingers of mouse Vps4B/SKD1 (mmVps4) (R290R291)¹³ and Vps4 from Sulfolobus solfataricus (SsoVps4) (R262R263)²¹ impair hexamerization, and arginine finger mutations of SsoVps4²¹ and Vps4 from M. sedula (MsVps4) (R259R260)²² abolish ATPase activity.

Open in a separate window

Figure 3

ATPase active centres.

(a) Comparison of subunit D with the crystal structure of the ATPγS-bound Vps4^E233Q (PDB ID: 3EIH)¹². The crystal structure is coloured grey, and the subunit D is coloured based on the distances between aligned Cα atom pairs with colouring scheme showing on the right. Helixes α3, α4 and α5 are labelled and marked by black arrows. (b) A focused view of the adjacent Vps4^E233Q subunits D and E with each subunit superimposed with the crystal structure of the ATPγS-bound Vps4^E233Q. Two ATPase active centres are boxed and detailed in (c,d). ATP molecules are shown in stick representation, and the map density shown in mesh. (c,d) Close-up views of the ATPase active centre in subunit D (c) and subunit E (d). Residues contributing to ATP binding are shown in sphere representation. Large shifts of R289, N261 and N265 in subunit D compared with those in the crystal structure are shown by red dashed boxes in (d). The orientations of all close-up views are optimized to show the positions of the residues contributing to ATP binding. All alignments were done using the residues in Walker A loop (Phe174-Tyr181), Walker B (Asp232-Gln233), sensor 1 (Asn277) and M307 as a reference. (e) SEC analyses of the oligomeric states of the Vps4 mutants. (f) The ATPase activity assays. Data are represented as the average of three independent experiments. Error bars represent s.d. (g) Comparison of the ATPase centres of Vta1-free and Vta1-bound Vps4^E233Q hexamers. The colouring scheme is the same as in Fig. 1c,e. The ATP molecules in the Vta1-bound Vps4^E233Q hexamers are displayed in stick model. The R289 residues (red arrows) in Vta1-free and Vta1-bound Vps4^E233Q hexamers are coloured by pink and red, respectively. All alignments were performed using the subunits that the arginine fingers act on in trans.

Since Vta1 can increase the ATPase activity of Vps4 in vitro, we next examined whether there is any difference in the conformation of the ATPase centre between Vta1-free and Vta1-bound Vps4^E233Q hexamers. Comparisons of the ATPase active centres of these two hexamers indicate that in three ATP pockets located at the interface between subunits C and D, D and E, and E and F, the R289 residues of the Vta1-bound Vps4^E233Q hexamer clearly shift slightly towards the γ-phosphate of ATP compared with those of the Vta1-free Vps4^E233Q hexamer (Fig. 3g). As studied above, the arginine finger R289 is essential for the hexamer assembly and the ATP hydrolysis. In the AAA+ proteins, the arginine finger promotes oligomerization and hydrolysis by interaction with the γ-phosphate of ATP^27,28. Thus, the relatively short distance between the R289 and the γ-phosphate of ATP in the Vta1-bound Vps4^E233Q hexamer may facilitate their interactions to help hexamerization and accelerate the ATP hydrolysis.

Vps4^E233Q hexamer and Vps4^E233Q–Vtal complex assembly for EM

Assembly of Vps4 oligomer was completed through incubation of the purified E233Q protein with 1mM ATP at 4°C for 1h. The mixture was then subjected to a Superdex 200 10/300GL column (GE Healthcare) in the assembly buffer containing 20mM HEPES at pH 7.5, 100mM NaCl, 1mM dithiothreitol, 2mM MgCl₂ and 1mM ATP. For the reconstitution of the complex between Vps4^E233Q hexamer and Vta1, E233Q protein was mixed with a 1.5- to 2-fold molar excess of Vta1 and then incubated with 1mM ATP at 4°C for 1h. The mixture was then subjected to a Superose 6 10/300GL column (GE Healthcare) in the assembly buffer.

SEC with MALS

SEC–MALS experiments were performed using an Agilent 1200 high-performance liquid chromatography system (Agilent Technologies, Santa Clara, CA, USA) coupled to a Wyatt DAWN HELEOS-II MALS instrument and a Wyatt Optilab rEX differential refractometer (Wyatt, Santa Barbara, CA, USA). For chromatographic separation, 200μl of 5mgml⁻¹ (50μM) Vps4^E233Q protein preincubated with 1mM ATP at 4°C for 1h was injected onto a Superdex 200 10/300 GL column (GE Healthcare) at a flow rate of 0.3mlmin⁻¹ in the assembly buffer. The outputs were analysed by the ASTRA 6.0 software (Wyatt). MALS signals, combined with the protein concentration determined by the refractive index, were used to calculate the molecular mass.

SEC analysis of oligomeric states and protein interactions

To examine the oligomeric states of Vps4 mutants, each mutant protein at 100μM was incubated with 1mM ATP at 4°C for 1h, and then loaded onto a Superdex 200 10/300 GL column (GE Healthcare) preequilibrated with the assembly buffer. To examine the interactions between Vps4 and Vta1 in the absence of ATP, Vps4 mutants at 100μM were incubated with Vta1 at a molar ratio of 1:1.5 in the absence of ATP at 4°C for 1h before being loaded onto the Superdex 200 10/300 GL column (GE Healthcare) preequilibrated with the FPLC buffer. To examine the interactions between Vps4 and Vta1 in the presence of ATP, Vps4 mutants at 5μM were incubated with Vta1 at a molar ratio of 1:2 in the presence of 1mM ATP at 4°C for 1hour before being loaded onto the Superose 6 10/300 GL column (GE Healthcare) preequilibrated with the assembly buffer.

Stability assay

The formed Vta1-free Vps4^E233Q hexamers and the Vta1-bound Vps4^E233Q hexamers eluted from SEC in the assembly buffer were individually reloaded onto the Superose 6 10/300 GL column (GE Healthcare) in the FPLC buffer. For the apyrase treatment, the 10Uml⁻¹ apyrase (Sigma-Aldrich) was added to the samples and incubated at 4°C overnight before SEC.

ATPase activity assay

ATPase activity assays measure released phosphate based on the formation of coloured complex between malachite green reagent and free phosphate according to the manufacturer’s instructions using the QuantichromeTM ATPase/GTPase Assay Kit (BioAssay Systems). Briefly, 1μM of wild-type or mutant Vps4 proteins in the absence or presence of Vta1 in the ATPase assay buffer containing 20mM HEPES at pH 7.5, 100mM NaCl, 2mM MgCl₂ and 5mM ATP were incubated at 37°C for various times (0–4min), and stopped by adding reagent from the kit. The solution was incubated at room temperature for 30min, and immediately the absorbance at 620nm was detected using a plate reader (Enspire, Perkin Elmer). The released phosphate was calculated based on the absorbance standard curve established using KH₂PO₄ standards.

ATP/ADP quantification assay

The 50μM Vps4^E233Q and 100μM Vta1 protein were incubated for 1h on ice in the assembly buffer, and then loaded onto the Superose 6 10/300GL column (GE Healthcare) in the FPLC buffer. The eluted Vps4^E233Q–Vta1 complex was used to examine the ATP and ADP levels by the bioluminescent detection-based method (Bioassay System EnzyLight ADP Assay Kit, EADP-100) according to the manufacturer’s instructions. The luminescence values were detected using a plate reader (Enspire Multimode Plate Reader, Perkin Elmer). The nucleotide levels were calculated based on the luminescence standard curves established using ATP and ADP standards. Three independent experiments were carried out on at least two different batches of purified Vps4^E233Q and Vta1 proteins and each experiment was carried out in at least triplicate.

Stoichiometry determination

We used the fluorescence assay-based method to evaluate the stoichiometry of the ATP-bound Vps4^E233Q–Vta1 complex. We constructed the single-cysteine mutants (containing only one cysteine residue) of Vps4 (Vps4^E233QC376S) and Vta1 (Vta1^C124S154S) through site-directed mutagenesis and purified them as for other mutant proteins. The concentrations of all proteins were measured by Pierce BCA Protein Assay Kit (Thermo). Vps4 and Vta1 proteins were concentrated to 4 and 10mgml⁻¹ in the labelling buffer containing 50mM HEPES, 100mM NaCl and 1mM TCEP (tris(2-carboxyethyl)phosphine), pH 7.4, respectively. Cy5 maleimide and Cy3 maleimide (GE Healthcare) were added into the Vps4 and Vta1 protein solutions, respectively, at the protein/fluorophore ratio of 1:4. Labelling reactions were performed on ice overnight, and terminated by 3mM β-ME. The free dyes were removed by using the NAP-5 columns. Then, we generated the ATP-bound Vps4^E233Q–Vta1 complex using these labelled proteins by incubating them at the molar ratio of 1:2 in the presence of ATP on ice for 1h and then applying the mixtures to SEC in the assembly buffer to separate the complex. The fluorescence intensities of Cy5 and Cy3 of the complex were measured using a plate reader (Enspire Multimode Plate Reader, Perkin Elmer). Then, the protein concentrations of Vps4 and Vta1 in the complex were calculated according to the fluorescence intensities and the standard curves of the protein concentration versus the fluorescence intensity. Three independent experiments were carried out and each experiment was carried out in at least triplicate.

Cryo-EM sample preparation and data collection

The assembled ATP-bound Vps4^E233Q hexamer eluted from SEC was concentrated to ~1mgml⁻¹, and incubated with 0.02% glutaraldehyde for ~5min on ice. Then, aliquots of 4μl of this sample were applied to the glow-discharged Quantifoil R1.2/1.3 holey carbon grids (Quantifoil Micro Tools GmbH), blotted for 2s and plunge-frozen by using FEI Vitrobot Mark IV. The sample of the ATP-bound Vps4^E233Q–Vta1 complex was prepared same as the Vps4^E233Q hexamer, except without glutaraldehyde treatment. Grids were examined using an FEI Titan Krios operated at 300kV, and images were recorded using a K2 Summit direct electron detector (Gatan) in superresolution mode, at a nominal magnification of 22,500, and with the defocus ranging from −1.5 to −3.0μm. Images were collected under low-dose condition in a semiautomatic manner using UCSF-Image4 (ref. ⁴¹). The dose rate on the camera was set to be 8.2 electrons per pixel per second. For each micrograph stack, a total of 32 frames were collected with an exposure time of 8s, leading to a total accumulated dose of 50 electrons per Ų on the specimen.

Image processing

The collected original micrograph stacks were aligned and summed using whole-image motion correction⁴¹, and binned twofold, resulting in a pixel size of 1.30654Å per pixel. The defocus values of micrographs were estimated by CTFFIND3 (ref. ⁴²). The particles were picked by EMAN2 (ref. ⁴³) and RELION 1.3 (refs ^{44, 45}). All 2D and 3D classifications and refinements were performed using RELION 1.3 (refs ^{44, 45}).

A total of 31,883 and 66,905 particles were first boxed from 100 micrographs of the ATP-bound Vps4^E233Q hexamer and 120 micrographs of the ATP-bound Vps4^E233Q–Vta1 complex using e2boxer.py in EMAN2 (ref. ⁴³), respectively. Then, the boxed particles were extracted and reference-free 2D class averaging was performed. The generated 2D class averages were used as the templates for the subsequent autopicking of particles using RELION 1.3 (refs ^{44, 45}). A total of 467,563 particles of the ATP-bound Vps4^E233Q hexamer and 1,644,872 particles of the ATP-bound Vps4^E233Q–Vta1 complex were automatically picked out from 1,042 and 1,263 micrographs, respectively, and 2D classified using RELION 1.3 (refs ^{44, 45}). Then 411,144 Vps4^E233Q particles and 157,109 Vps4^E233Q–Vta1 particles were empirically selected from good classes of 2D classifications. We first performed 3D auto-refine using all 411,144 Vps4^E233Q particles and all 157,109 Vps4^E233Q–Vta1 particles with a disk-shaped volume generated from EMAN2 (ref. ⁴³) used as an initial model, respectively. The obtained reconstructions low-pass filtered to 60Å were then used as the references of the 3D classifications. All of the 411,144 Vps4^E233Q particles were divided into eight classes. The most homogeneous class of 106,918 particles was subjected to auto-refinement without any symmetry imposed and resulted in a reconstruction at an overall resolution of 4.1 and 3.9Å after applying a soft mask around the rigid part (excluding the subunit A) based on the gold-standard Fourier shell correlation (FSC) 0.143 criterion⁴⁶. All of the 157,109 Vps4^E233Q–Vta1 particles were divided into six classes. Auto-refinement of the best class of the 106,106 particles without any symmetry imposed resulted in a reconstruction of 4.3 and 4.2Å after applying a soft mask around the rigid part (excluding the subunit A and all bound Vta1 proteins). The handedness of the maps were determined and corrected by comparison with crystal structure of the ATPγS-bound Vps4^E233Q (PDB ID: 3EIH)¹². Additional cycles of 3D classifications and refinements did not further improve the overall resolution of the maps. Local resolution was estimated using ResMap⁴⁷.

Model building

For building the model of the ATP-bound Vps4^E233Q hexamer, the crystal structure of the ATPγS-bound Vps4^E233Q (PDB ID: 3EIH, chain A)¹² was docked into the density map using Chimera⁴⁸. The structure was manually adjusted based on the density map in Coot⁴⁹. Next, the model was refined by Phenix⁵⁰ in real space with stereochemical and secondary structure restraints, and further refined by REFMAC in Fourier space^51,52. The model of the ATP-bound Vps4^E233Q–Vta1 complex was built in the same way as the model of the ATP-bound Vps4^E233Q hexamer using the available crystal structures of the ATPγS-bound Vps4^E233Q (PDB ID: 3EIH)¹² and the Vta1 CTD dimers (PDB ID: 2RKL)¹⁹ as the starting models to refine. The atomic models were cross-validated according to previously described procedures⁵³. Briefly, the coordinates of the final model were randomly displaced by 0.5Å using the Phenix (PDB Tools) to remove potential model bias. The displaced model was then refined against one of the half maps (produced from a half set of all particles during refinement by RELION). FSC curves were calculated between the resulting model and half1 map (model versus half1 map, FSC_work, that is, used for refinement), the resulting model and half2 map (model versus half2 map, FSC_free, that is, not used for refinement) and the resulting model and the final density map (model versus summer map) from all particles. The lack of significant separation between work and free FSC curves suggested that the models were not overfitted.

MolProbity⁵⁴ (http://molprobity.biochem.duke.edu/) and EMRinger⁵⁵ were used to evaluate the final models, and final statistics of the model was provided in the Supplementary Table 1. Interface areas were calculated by PISA⁵⁶. Pymol (http://www.pymol.org/) and Chimera⁴⁸ were used for structural analysis and figure preparation.

ESCRT-III filament reconstitution

The ESCRT-III filament was reconstituted by incubating Snf7^R52E, Vps24 and Vps2 in the concentrations of 50, 25 and 25μM, respectively, at 30°C for 30min (ref. ⁵). Then, the formed filaments were centrifuged at 50,000r.p.m. for 30min in a TLA-100 ultracentrifuge rotor (Beckman Coulter). Pellet was resuspended in the buffer containing 50mM HEPES at pH 7.4, 100mM NaCl and 1mM TCEP. All the filaments were verified by SDS–PAGE and negative staining.

ESCRT-III filament disassembly assay

The reconstituted ESCRT-III filaments were incubated with 1μM Vps4 in the buffer containing 20mM HEPES at pH 7.4, 2mM ATP, 2mM MgCl₂, 1mM TCEP at 25°C for 5min (Supplementary Fig. 10a) or indicated time points (Supplementary Fig. 10b) and then 50mM EDTA were added to stop the reactions. The samples were centrifuged at 50,000r.p.m. for 30min in a TLA-100 rotor, and the supernatant and pellet were analysed by SDS–PAGE. The intensities of bands of Snf7 were quantified by ImageJ. Each experiment was carried out independently for three times.

Vps4 protein and ESCRT-III filament labelling

Vps4 proteins were concentrated to 2mgml⁻¹ in labelling buffer containing 50mM HEPES, 100mM NaCl and 1mM TCEP, pH 7.4. Cy5 maleimide (GE Healthcare) was added into the protein solution at protein/fluorophore ratio of 1:4. Labelling reactions were performed on ice overnight, and terminated by 3mM β-ME. The free dyes were removed by using a NAP-5 column. The labelling efficiency was 48%.

The reconstituted filaments were used for labelling with the Cy3 dye and biotin. Filaments, Cyanine3 NHS (Lumiprobe) and NHS-biotin (Thermo Scientific) were incubated at a molar ratio of 10:10:1 (Snf7^R52E monomer:Cyanine3 NHS:NHS-biotin) in the labelling buffer on ice for 2h. The free dyes and biotins were removed by centrifugation at 50,000r.p.m. for 30min. The labelled filaments were precipitated and resuspended in 50mM HEPES at pH 7.4, 100mM NaCl. Because NHS esters (N-hydroxysuccinimide esters) target and crosslink with primary amines on the proteins and reconstituted filaments are formed by tens of or hundreds of monomers, each filament was expected to be labelled with tens of Cyanine3 and several biotin. Although it is hard to estimate the labelling efficiency of filaments, they can be specifically attached to our coverslip via streptavidin coating and exhibit an order of magnitude stronger fluorescence intensity than single Cyanine3 fluorophore. Therefore, in our single-molecule assay, decreasing of Cyanine3 (Cy3) fluorescence signals was used to quantify dissembling of the reconstituted filaments.

Bulk experiments and single-molecule experiments demonstrated that Cy5-labelled Vps4 exhibits similar filament disassembly activities as unlabelled Vps4 and Cy3/biotin-labelled filaments could be disassembled with similar efficiency as unlabelled filaments by Vps4 (Supplementary Fig. 10 and Supplementary Table 2).

Acquisition and analysis of single-molecule data

Filaments were specifically attached to microscope flow cells formed by polyethylene glycol-passivated slides, decorated with biotin-polyethylene glycol and streptavidin⁵⁷. Single-molecule experiments were performed at 23°C in the buffer containing 50mM HEPES at pH 7.4, 100mM NaCl with an oxygen scavenging system, containing 3mgml⁻¹ glucose, 100μgml⁻¹ glucose oxidase (Sigma-Aldrich), 40μgml⁻¹ catalase (Roche), 1mM cyclooctatetraene (Sigma-Aldrich), 1mM 4-nitrobenzylalcohol (Sigma-Aldrich), 1.5mM 6-hydroxy-2,5,7,8-tetramethyl-chromane-2-carboxylic acid (Trolox, Sigma-Aldrich) added from a concentrated DMSO stock solution and 1% bovine serum albumin (Amresco) to reduce nonspecific binding. After filament immobilization, 100nM of Vps4 labelled with Cy5 alone or together with 500nM Vta1 was injected with 1mM ATP and 1mM Mg²⁺. After incubation for 30s, 100nM unlabelled Vps4 alone or together with 500nM Vta1 was injected along with 1mM ATP and 1mM Mg²⁺ to wash out labelled Vps4 in the microscope flow cell, and further incubated for a total time of 10min. Movies were recorded immediately after labelled Vps4 injection and continued for 10min. Fluorescence images of filaments were taken before Vps4 injection and after Vps4 incubation (Fig. 4a). Under our experimental conditions, rate of Cy5 photobleaching was 0.00071±0.00005, s⁻¹ (average±s.e.m.)

Single-molecule spectroscopic microscopy was performed on a home-built objective-type total internal reflection fluorescence microscope, based on a Nikon Eclipse Ti-E with an EMCCD camera (Andor iXon Ultra 897), and solid-state 488, 532 and 640nm excitation laser (Coherent Inc., OBIS Smart Lasers). To monitor fluorescence intensities of filaments and Vps4, alternating laser excitation²⁹ between two lasers excitation was achieved by synchronizing laser on–off switch with EMCCD frames. Fluorescence emission from the probes was collected by the microscope and spectrally separated by interference dichroic (T635lpxr, Chroma) and bandpass filters, ET585/65m (Chroma, Cy3) and ET700/75m (Chroma, Cy5), in a Dual-View spectral splitter (Photometrics, Inc., Tucson, AZ, USA). All movies were collected using Cell Vision software (Beijing Coolight Technology). Single-molecule events longer than three frames were picked by home-written Matlab scripts.

We thank Jianlin Lei, Yanji Xu and Xiaomei Li for the EM support, T. Yang and Y. Wang for computer support, J. Xie for her excellent technical support and Jiao Lin for computation support. We thank the Tsinghua University Cryo-EM Facility of China National Centre for Protein Sciences (Beijing) and the ‘Explorer 100’ cluster system of the Tsinghua National Laboratory for Information Science and Technology for providing EM and computation resources. This work was supported by funds from the Ministry of Science and Technology of China (2016YFA0501101) and the National Natural Science Foundation of China (31230016 and 31370717 to S.-F.S., 31400632 and 31670746 to S.S. and 31570754 to C.C.).