The progression of ESCRT (Endosomal Sorting Complexes Required for Transport) pathways, which mediate numerous cellular membrane fission events, is driven by the enzyme Vps4. Understanding of Vps4 mechanism is, therefore, of fundamental importance in its own right and, moreover, it is highly relevant to the understanding of many related AAA+ ATPases that function in multiple facets of cell biology. Vps4 unfolds its ESCRT-III protein substrates by translocating them through its central hexameric pore, thereby driving membrane fission and recycling of ESCRT-III subunits. This mini-review focuses on recent advances in Vps4 structure and mechanism, including ideas about how Vps4 translocates and unfolds ESCRT-III subunits. Related AAA+ ATPases that share structural features with Vps4 and likely utilize an equivalent mechanism are also discussed.
Endosomal Sorting Complex Required for Transport (ESCRT) pathways drive membrane fission in numerous cellular processes, including cytokinesis, multivesicular body formation, plasma membrane repair, neuron pruning, exovesicle shedding, nuclear pore complex quality control, nuclear envelope reassembly, nuclear envelope repair after rupture, unconventional protein secretion, endolysosomal repair, and virus budding [1–8]. These pathways progress when upstream ESCRT factors trigger assembly of ESCRT-III filaments at sites that are destined for membrane fission. In turn, the ESCRT-III filaments recruit Vps4, whose activity subsequently drives both membrane fission and recycling of ESCRT-III subunits to their soluble state. The mechanism of membrane fission is not entirely clear, but an attractive possibility is that formation of the ESCRT-III filaments stabilizes a metastable conformation of the lipid bilayer, and the subsequent removal of ESCRT-III introduces stress that resolves by membrane fission [2,6]. Regardless, it is well established that Vps4 is essential for this process in multiple ESCRT pathways. Thus, understanding of Vps4 structure and mechanism has been a high priority since the discovery of its importance in the ESCRT pathway that buds vesicles into multivesicular bodies .
Domain structure of Vps4
The large family of AAA+ ATPases has been classified into multiple clades , including the meiotic clade that includes Vps4 and the microtubule-severing enzymes spastin and katanin . Like many other AAA+ ATPases, Vps4 is active as a hexameric assembly that uses the energy of ATP hydrolysis to drive protein translocation through the central pore of the hexamer, thereby unfolding the substrate protein [12–14]. Also like many other AAA+ ATPases, Vps4 possesses an N-terminal domain that is connected to the ATPase cassette by a flexible linker (Figure 1). The N-terminal domain of Vps4 is a three-helix MIT (Microtubule Interacting and Trafficking) domain that binds to MIM (MIT Interacting Motif) sequences located in the C-terminal regions of ESCRT-III subunits [15–18]. These MIM sequences are inherently flexible and accessible in the ESCRT-III filament, and provide the initial interaction that recruits Vps4 for ESCRT-III processing and membrane fission [19–21].
Domain structure of Vps4.
The Vps4 ATPase cassette resembles that of other AAA+ ATPases, and comprises a large and small ATPase domain (Figure 1). The large domain contains the Walker A and B motifs that drive ATP binding and hydrolysis, respectively, as well as pore loops that engage substrate in the central pore of the hexamer . The small ATPase domain, which follows the large ATPase domain, comprises four helices, and is followed by a C-terminal helix that packs against the large domain. The β-domain emanates from an internal loop of the small ATPase domain to mediate interactions with the Vta1/LIP5 cofactor protein [22,23]. Vta1/LIP5 binds and stabilizes the assembled Vps4 through its dimeric, C-terminal VSL (Vta1–SBP1–LIP5) domain  and also displays flexibly connected N-terminal MIT domains that may further enhance association with ESCRT-III filaments [24–26].
Structure of the active Vps4 complex
For many years, the primary goal of Vps4 structural studies has been to visualize its active, assembled conformation. Early studies established that Vps4 can exist in unassembled (monomeric/dimeric) or assembled states, and that the assembled state is associated with sites of membrane fission and performs ATP binding and hydrolysis . The actual assembly state was the subject of considerable uncertainty, with confusion further worsened by three early EM structures that showed double-ring 12-mer and 14-mer structures that were in conflict with each other [27–29]. It is now apparent that each of those structures was misleading because subsequent biochemical studies have established that the active wild-type complex is, in fact, a hexamer that binds substrates in its central pore [30,31]. An X-ray crystal structure of a Vps4 hexamer  now also seems to have been misleading because it displays a high degree of symmetry, whereas three independently determined cryo-EM structures that were reported in 2017 are in good agreement with each other and seem to represent the active, more asymmetric conformation [33–35].
Despite the similarity of the three recently reported Vps4 cryo-EM structures [33–35], they have given rise to quite different mechanistic interpretations. Our preferred model is based on a cryo-EM structure of the yeast Saccharomyces cerevisiae Vps4 ATPase cassette in complex with an ESCRT-III-derived peptide, the Vta1 VSL domain, and ADP·BeFx (Figure 2). The primary advantage of this structure, which was initially determined at 4.3 Å resolution  and later reported at 3.2 Å resolution , is that it includes a bound ESCRT-III peptide, which defines the substrate-binding groove and constrains mechanistic models.
Structure of the active Vps4 hexamer complex.
Substrate peptide binds a helical array of Vps4 subunits
The cryo-EM structures reveal that the Vps4 hexamer comprises a helical assembly of five subunits (subunits A–E) and a sixth subunit (F) that, as discussed below, appears to be transitioning between the ends of the five-subunit helix. The five-subunit helix is right-handed with a translation of ∼6.3 Å and a rotation of 60° between subunits. This matches the symmetry of the bound ESCRT-III peptide, which adopts a β-strand conformation that spirals tightly around the helix axis to bind in a groove formed by the five helical Vps4 subunits . The concordance in symmetry between enzyme and bound substrate has important mechanistic implications because it means that successive substrate dipeptides bind in the same manner to successive Vps4 subunits, or more accurately, to successive interfaces between Vps4 subunits. Thus, the side chain of the first residue of each dipeptide binds into a ‘class I' binding pocket, while the side chain of the second residue of each dipeptide binds into a ‘class II' binding pocket, and these pockets are repeated at each of the four interfaces between the five helical Vps4 subunits (Figure 3).
Binding of ESCRT-III by Vps4.
Class I pockets are primarily formed by the tryptophan residue from pore loop 1 of adjacent Vps4 subunits, with the preceding lysine residue also contributing. The side chain of the first residue of the substrate dipeptide packs between the two tryptophans of adjacent subunits, which form a hydrophobic notch that appears capable of accommodating a wide variety of amino acid side chains. The binding of hydrophilic side chains is presumably accommodated by the highly solvated nature of the Vps4 pore, which will allow water molecules to satisfy the hydrogen-bonding needs of the ESCRT-III residues. The same principles are evident in the architecture of the class II binding sites that bind the second residue of each substrate dipeptide. In this case, the pore loop 1 methionine residues of adjacent Vps4 subunits form a hydrophobic notch, while pore loop 2 residues flank the binding pocket. Once again, the structure shows that a wide variety of amino acids can be accommodated, with the hydrogen-bonding groups of hydrophilic side chains presumably satisfied by interaction with water molecules from the solvent-exposed pore. Thus, Vps4 presents a helical array of equivalent dipeptide-binding sites that can accommodate the diverse amino acid residues of the peptide that is bound in the structure. This indicates that binding will be largely sequence-independent, as expected for translocation of the variable sequences of the ESCRT-III protein substrates.
Model of substrate translocation
The geometry of substrate binding suggests a ‘conveyer-belt’ model of substrate translocation in which the Vps4 hexamer propagates along the ESCRT-III substrate. In this model, Vps4 ‘walks' along the substrate through the action of individual Vps4 subunits peeling off the trailing end of the five-subunit helix (subunit E), passing through the ‘transitioning' conformation (subunit F), and joining the growing end of the helix (subunit A) to bind the next dipeptide in the ESCRT-III sequence (Figure 4). Consistent with this model, the large ATPase domain of subunit F is displaced from the helix axis, disengaged from the ESCRT-III peptide, and makes only minor contacts with the large domains of neighboring Vps4 subunits. Subunit F is maintained within the Vps4 assembly primarily through more peripheral contacts of the small ATPase domains. Advancement of the Vps4 hexamer along the extended ESCRT-III polypeptide is equivalent to conveying the ESCRT-III through the Vps4 hexamer pore, and will result in unfolding and disassembly of ESCRT-III. An attractive feature of the model is that each Vps4 subunit only needs to adopt one conformation as it and its associated ESCRT-III dipeptide pass through the pore from the A-subunit position to the E-subunit position. Thus, major conformational changes required of Vps4 are limited to those associated with transitioning from the E-subunit position at one end of the helix, through the F-subunit position and on to the A-subunit position at the other end of the helix.
Model of Vps4 translocation mechanism.
The Vps4 structure was determined in complex with the non-hydrolyzable nucleotide analog ADP·BeFx, which can resemble ATP, the hydrolysis transition state, ADP with dissociated phosphate, or ADP without phosphate . The resolution of the cryo-EM map is not sufficient to be definitive, but it appears to indicate that ADP·BeFx (analogous to ATP) is bound at the A–B, B–C, and C–D subunit interfaces, while ADP is bound at the D–E and E–F interfaces . The nucleotide state of the transitioning subunit F is not apparent due to the low local resolution in this region of the model, but the structure suggests that this site is open to allow nucleotide exchange, which may mean that ATP binding is favored in the cellular context where ATP concentration is typically 10-fold higher than ADP concentration. The E–F interface is also quite open. The most intriguing observation on nucleotide coordination is the apparent binding of ADP at the D–E interface, whereas the very similar A–B, B–C, and C–D interfaces all appear to bind ATP (ADP·BeFx). Despite a high level of similarity to the A–B, B–C, and C–D interfaces, there are some differences at the D–E interface that may reflect the distinct nucleotide states, including a relative ∼1 Å shift in the arginine finger residues of subunit E that co-ordinate the β- and γ-phosphates of ATP (ADP·BeFx).
These observations suggest that ATP is hydrolyzed at the D–E interface, which presumably destabilizes this interaction and promotes transition to the more open EF conformation. Why hydrolysis might preferentially occur at the D–E interface rather than at the similar A–B, B–C, and C–D interfaces is not apparent, but presumably results from conformational changes that result from the more open E–F interface and propagate to the active site at the D–E interface. This model explains why substrate translocation proceeds in the direction indicated in Figure 4. At the other end of the helix, we envision that the open F–A interface allows binding of ATP, which will promote transition to the AB conformation. In this manner, the Vps4 helix will grow at the A-end of the helix as ATP binds and co-ordinately shrink at the E-end of the helix as ATP is hydrolyzed. The direction of translocation indicated by this model is consistent with the proteasome, which has a related structure and is known to translocate substrates in the equivalent direction into the proteasome's proteolytic chamber .
Comparison with other AAA+ ATPases
Several recently reported structures of other AAA+ ATPases overlap with the structure of Vps4, which suggests that they may all utilize an equivalent mechanism of substrate translocation. In particular, multiple structures have now been reported with substrate bound in the central pore and with a superimposable helix of subunits [38–44] (Figure 5), which strongly suggests that common mechanisms of nucleotide binding/hydrolysis and substrate binding/translocation will be utilized. Moreover, several AAA+ ATPase structures that have been determined in the absence of engaged substrate also adopt closely similar structures [34,35,45]. In some cases, including katanin  and another structure of Vps4 , an alternative mechanism of substrate disassembly was proposed. Nevertheless, we favor the conveyor-belt model outlined above because it is consistent with the binding of substrate peptides seen in Vps4 and other AAA+ ATPases, including YME1 , HSP104 , VAT , TRIP13 , ClpB , and NSF . It should be noted that multiple structures of hexameric AAA+ ATPases that probably translocate protein substrates have been reported in distinctly different conformations, including p97  and ClpX . We favor the model that the mechanism of AAA+ ATPases requires some inherent flexibility, and that these alternative structures are favored in the biochemical/crystallographic environment of sample preparation, but that the active translocating state corresponds to the structure observed for Vps4 and other AAA+ ATPases with bound substrate.
Superposition of Vps4 and other AAA+ ATPase structures.
Of the AAA+ ATPases structures reported thus far, the mitochondrial quality control protein, YME1, merits discussion. The structure of inactive YME1, in which the catalytic Walker B glutamate was mutated to glutamine, was determined by cryo-EM in the presence of ATP and found to include peptides that fortuitously copurified after expression in E. coli and bind in the same manner as the ESCRT-III peptide binds to Vps4 . The YME1 and Vps4 structures are remarkably similar, but ATP is bound to all four of the subunit interfaces of the five-subunit YME1 helix, whereas the last of these positions appears to be occupied by ADP in the case of Vps4 (D–E interface; above). This led to a subtle difference in mechanistic emphasis, with ATP hydrolysis proposed to occur as the last helical subunit YME1 (subunit E in Vps4 nomenclature) transitions to the F conformation, while the Vps4 model is more consistent with ATP hydrolysis occurring before the closed D–E interface opens and the E subunit transitions to the F conformation. This distinction cannot be resolved from the currently available structures of an inactive mutant in the presence of ATP (YME1) and a wild-type protein in the presence of ADP·BeFx (Vps4). Interestingly, a recent report of structures of active proteasome with engaged substrate indicates that ATP hydrolysis occurs at the interface corresponding to D–E of Vps4 .
Vps4 appears to show the ESCRT-III peptide binding in an orientation that is consistent with translocation from the C-terminal to N-terminal direction. This is consistent with biochemical studies that show progressive unfolding toward the N-terminus , and with the biological function of engaging the C-terminal residues of ESCRT-III before unfolding the structured N-terminal regions and thereby disassembling ESCRT-III filaments . In contrast, other AAA+ ATPase complexes have been modeled with substrates bound N to C, C to N, or a mixture of both orientations, which is consistent with the view that different AAA+ ATPases can translocate substrates in either or both directions . These observations indicate important questions for future studies, including: (i) Are the peptide orientations assigned for Vps4 and the other AAA+ ATPase complex structures correct? This is pertinent because the resolution attained in the various cryo-EM structures is generally marginal for the assignment of orientation for an isolated peptide. (ii) Is the same orientation also adopted when other peptide sequences bind to Vps4? (iii) Is the assigned orientation for Vps4 favored by a main chain hydrogen bond, as proposed , and (iv) what are the factors that dictate substrate-binding orientation for the various AAA+ ATPases?
AAA+ ATPases that act on protein substrates have been proposed to utilize mechanisms that are concerted , stochastic  or sequential . Although the model that we favor for Vps4 is sequential, the extent to which variations may occur is unclear, especially given that ClpX, which has long served as a model for AAA+ ATPase studies, can accommodate inactive subunits  and operate with uneven step sizes . Interestingly, the structures of related AAA+ ATPases that perform work on nucleic acid substrates also appear to utilize mechanisms that are sequential. In particular, the E1 helicase, which translocates DNA, shows five subunits forming a right-handed helix, the sixth subunit appearing to transition from one end of the helix to the other, and with a strand of DNA bound to the pore . An analogous configuration occurs for the Rho RNA helicase, which has a different orientation of ATPase domains but nevertheless shows helical symmetry for five of the ATPase subunits and bound RNA substrate .
Although considerable progress has been made, numerous important questions about AAA+ ATPase mechanisms remain. These include the timing of ATP hydrolysis, the orientation of substrate binding and translocation, the extent to which substrate specificity is defined by binding preferences in the translocation pore, and whether or not the translocation pore can accommodate more than one substrate polypeptide chain at a time [54–56]. While we believe that structures of Vps4 and other AAA+ ATPase complexes mimic translocating states, the possibility that the Vps4 structure corresponds to the initiation of translocation should be considered because the bound peptide corresponds to a relatively tight-binding segment of an ESCRT-III subunit . On this note, and echoing studies of the proteasome , it will be of interest to determine the relative affinities of distinct amino acid sequences, and the impact of binding affinity upon substrate engagement and translocation. More generally, the extent to which other AAA+ ATPases follow the sequential, conveyer-belt mechanism proposed here is unclear, especially in light of the 6-fold rotationally symmetric structures of p97  and CDC48 . Finally, as noted above, our model is at odds with alternative interpretations of independent Vps4 and katanin structures. Resolving these questions will be important priorities for future studies.
ATPases associated with diverse cellular activities
Endosomal Sorting Complexes Required for Transport
MIT Interacting Motif
Microtubule Interacting and Trafficking
vacuolar protein sorting-associated protein 4
Work on Vps4 in the Hill laboratory is funded by the National Institutes of Health P50GM082545.
The Authors declare that there are no competing interests associated with the manuscript.