The vast expansion in recent years of the cellular processes promoted by the endosomal sorting complex required for transport (ESCRT) machinery has reinforced its identity as a modular system that uses multiple adaptors to recruit the core membrane remodelling activity at different intracellular sites and facilitate membrane scission. Functional connections to processes such as the aurora B-dependent abscission checkpoint also highlight the importance of the spatiotemporal regulation of the ESCRT machinery. Here, we summarise the role of ESCRTs in viral budding, and what we have learned about the ESCRT pathway from studying this process. These advances are discussed in the context of areas of cell biology that have been transformed by research in the ESCRT field, including cytokinetic abscission, nuclear envelope resealing and plasma membrane repair.
Seminal work in yeast identified the endosomal sorting complex required for transport (ESCRT) through a series of studies of ∼40 vacuolar protein sorting (VPS) mutants that showed defective sorting of proteins to the vacuole . In a subset of these mutants, membrane proteins that were normally trafficked to the vacuole for degradation did not reach this compartment and accumulated in a perivacuolar structure termed the class E compartment [1–4]. These class E VPS genes were subsequently found to encode components of four ESCRT complexes, ESCRT-0, -I, -II and -III, that are necessary to recognise the ubiquitinated cargo and sort it into intralumenal vesicles (ILVs) at multivesicular bodies (MVBs) through inward invagination and budding of the limiting membrane away from the cytoplasm [5–10].
The ESCRT proteins are conserved from yeast to humans [8,11,12], although a gene expansion can be seen in higher eukaryotes. For example, ESCRT-III in humans possesses eight charged multivesicular body protein (CHMP) families, namely CHMP1–7 and human increased sodium tolerance 1 (hIST1). Additional genes for CHMP1 (CHMP1A and B), CHMP2 (CHMP2A and 2B) and CHMP4 (CHMP4A, 4B and 4C) exist, totalling 12 different ESCRT-III subunits identified so far. A key component required for ESCRT function throughout evolution is VPS4, an ATPase that maintains the activity of ESCRT-III by promoting its disassembly for recycling purposes [13–15].
In addition to their role in MVB formation, ESCRTs were subsequently found to be involved in topologically equivalent membrane remodelling events, notably facilitation of enveloped virus budding [16–19] and cytokinetic abscission [20,21]. Over the last decade, the number of discovered ESCRT-mediated processes has continued to expand to include plasma membrane repair [22,23], axonal pruning [24–26], nuclear envelope (NE) resealing [27,28] and defective nuclear pore complex (NPC) removal . In each of these cases, the ESCRT machinery must assemble on the cytosolic side of the limiting membrane, within membranous stalks, to stabilise negative membrane curvature and mediate budding and scission away from the cytosol.
This review will first summarise what is known about the role of ESCRTs in viral budding and how studies in this area have furthered our understanding of the ESCRT pathway. Focus will be paid to areas of cell biology that have been transformed by research in the ESCRT field, such as cytokinesis and NE resealing, together with spaciotemporal regulation of the ESCRTs and their coordination with the abscission checkpoint. Finally, the role of ESCRTs in plasma membrane repair and modes of ESCRT-III polymerisation that may be relevant for membrane remodelling will be discussed.
ESCRTs in viral budding
The pioneering work by Göttlinger et al.  showed that the p6 region of the main human immunodeficiency virus-1 (HIV-1) structural protein, Gag, is essential for viral release. The deletion of p6 resulted in tethering of nascent virions to the plasma membrane by thin membranous stalks, thus suggesting that membrane remodelling events required at the last step of viral assembly were impaired in these mutants. Subsequent work established that this phenotype could be largely attributed to a conserved N-terminal 7PTAP10 motif in p6 . Other conserved short amino acid motifs that were necessary for viral budding were soon identified, notably a YPxnL motif in equine infectious anaemia virus (EIAV) p9-Gag  and a PPxY motif in Rous sarcoma virus (RSV) p2b-Gag [33,34]. Given their activity in the final events of viral assembly, these motifs were termed late budding domains (L-domains). The short length and proline-rich nature of these motifs suggested that their role in viral budding could involve the recruitment of cellular proteins. This notion was further supported by observations that L-domains were interchangeable and position independent within retroviral Gag proteins [32,35,36].
The cellular pathway recruited by L-domains was revealed by the identification of the essential interaction between PT/SAP motifs in HIV-1 and Ebola virus with the ESCRT-I subunit tumour susceptibility gene 101 (Tsg101), via its ubiquitin enzyme variant (UEV) domain [16–19,37,38]. YPxnL motifs on the other hand were shown to recruit the ESCRT-associated protein ALG-2-interacting protein X (ALIX) via its V domain [11,12,39,40], whereas PPxY motifs promote viral budding by binding to members of the NEDD4 (neural precursor cell expressed developmentally down-regulated protein 4)-like HECT domain E3 ubiquitin ligases, notably WWP1 (WW domain-containing E3 ubiquitin protein ligase 1), WWP2 (WW domain-containing E3 ubiquitin protein ligase 2) and Itch via their WW domains [41,42]. Although functional redundancy between these ubiquitinases in viral budding was originally suggested, a recent study has shown a preference for NEDD4-1 and Itch ubiquitin ligases for Ebola virus budding , which encodes overlapping PTAP and PPEY motifs [18,44]. The striking discovery of a primordial ESCRT system in Archae of the genus Sulfolobus has revealed that the CHMP4 and VPS4 homologues are necessary for Sulphobus turreted icosahedral virus replication and budding [45–47]. These findings demonstrate an ancient, conserved role for ESCRTs in viral budding.
The widely conserved requirement in the release of enveloped viruses suggested a membrane remodelling function of the ESCRT machinery, in particular of ESCRT-III. Importantly, ALIX recruitment to YPxnL L-domains provides a direct link to ESCRT-III by directly binding to the CHMP4 proteins [11,12,40,48,49]. However, definitive pathways for ESCRT-III recruitment for the other two L-domains remain less clear. ESCRT-II has been suggested as a necessary activity for HIV-1 genomic RNA trafficking , and subsequent work has proposed a role for ESCRT-II as a potential bridging complex between ESCRT-I and ESCRT-III in PTAP-dependent budding [51,52]. However, these hypotheses remain controversial as other studies do not observe deleterious effects in viral assembly upon depletion of ESCRT-II [53,54]. Crucially, siRNA approaches and VPS4 dominant-negative expression have established the essential role of ESCRT-III in each case [17,37,55–57]. These findings together with the specificity found at the adaptor/L-domain level have given rise to the concept that the ESCRT pathway operates as a modular machinery that uses different adaptor proteins to recruit a core membrane remodelling activity at specific sites (Figure 1).
Detailed functional analysis of HIV-1, murine leukaemia virus (MLV) and EIAV assembly has shown that viral budding requires only a small subset of the 12 mammalian ESCRT-III subunits. In particular, a prominent role for CHMP2 and CHMP4 family members in L-domain activity has been demonstrated for HIV-1 and MLV [52,56]. Accordingly, an axis that involves Gag–ALIX–CHMP4B–CHMP2A–VPS4 interactions is required for EIAV release . The involvement of other ESCRT-III subunits in retroviral budding is less clear, as depletion of CHMP6, CHMP3 and the CHMP1A/B proteins only results in modest phenotypes at best [52,54,56,57]. These studies strongly suggest that the core membrane remodelling activity only requires the CHMP4–CHMP2 complex, while other ESCRT-III subunits may play accessory roles that are not required for retroviral release.
Many viruses such as HIV-1, Ebola virus and MLV encode multiple L-domains. In each case, one appears to show dominance whilst others seem to play subsidiary roles. For example, HIV-1 Gag has an ALIX-binding LYPxnL motif in p6, in addition to its PTAP motif [11,39,58]. The PTAP–Tsg101 interaction appears to be its dominant mode of ESCRT recruitment, but ALIX overexpression can rescue budding of PTAP mutant HIV-1 [40,49]. A similar phenomenon has been described for MLV, whose Gag encodes PPPY, PSAP and LYPxnL motifs, although the PPPY L-domain activity shows a clear dominance . Ebola VP40 contains a PTAPPEY sequence that binds Tsg101 and HECT ubiquitin ligases, and both activities are required for efficient viral release [18,41,60]. Similar to HIV-1 and MLV, a third L-domain in VP40 interacts with ALIX and plays an auxiliary function in viral release . Possession of multiple L-domains may have evolved to confer an evolutionary advantage to ensure viral propagation in the event of loss-of-function mutations of the dominant L-domain, albeit at reduced levels. Another advantage would be to broaden viral tropism such that alternative L-domains may exhibit differential dominance in multiple cell types, as has been shown for the LYPxnL motif of HIV-1 in T cells .
One intriguing question that still needs to be fully answered is the role of ubiquitin in L-domain activity. Evidence supporting a role here includes the enrichment of ubiquitin observed in many virus particles [63,64], the discovery of ubiquitin ligases as host factors for PPXY type L-domains [41,42,65] and the substitution of L-domain activity by attachment of ubiquitin to Gag [66,67]. Defining the mechanism and most relevant substrates for ubiquitination has proved challenging, perhaps suggesting considerable redundancy in substrate requirements. The importance of ubiquitination in viral budding is supported by the strong correlation between viral budding and Gag ubiquitination in those viruses that exhibit PPXY motifs [65,68]. Accordingly, mutation of ubiquitin acceptor sites in RSV Gag abolishes budding, particularly in the NC-p2 region that lies close to the L-domains . Importantly, L-domain-deficient EIAV and human T-lymphotropic virus can be rescued by fusion of ubiquitin to Gag . In the case of HIV-1, ubiquitination of residues in the vicinity of the PTAP motif increases Tsg101 binding . However, increased ubiquitination of Gag by introduction of a PPXY motif does not enhance budding in the HIV-1 context . In contrast, NEDD4-2 overexpression can rescue budding of an HIV-1 that lacks L-domains by conjugating K63-linked polyubiquitin chains, despite the lack of a PPXY motif . Subsequent work has shown that the truncated C2 domain of NEDD4-2 provides a natural Gag-targeting module, thus explaining an L-domain-independent recruitment of ubiquitin ligases to the sites of viral budding [72,73].
One possible scenario contemplates that Gag ubiquitination could serve as a docking site for ubiquitin-binding activities in ESCRT-I [Tsg101, UBAP1 (ubiquitin-associated protein 1)], ESCRT-II (Eap45) and ALIX [70,74–77]. Accordingly, Tsg101 depletion abolished the ability of NEDD4-2 to rescue an L-domain-deficient HIV-1 , and residues in the UEV domain of Tsg101 that bind ubiquitin are necessary for rescue of an L-domain-deficient EIAV . Moreover, mutation of the ubiquitin-binding sites in the V domain of ALIX impairs retroviral budding . A second scenario involving ubiquitination of trans-acting factors is also supported by data showing that NEDD4-2 overexpression induces the ubiquitination of ESCRT-I subunits . Conversely, the fusion of a deubiquitinating activity to either Tsg101 or ALIX inhibits HIV-1 budding . A more compelling example of the importance of trans-acting factor ubiquitination in viral budding has been provided by studies that used an engineered prototypic foamy virus Gag devoid of its single ubiquitination site, which was fully functional in viral budding . Ubiquitination-resistant foamy virus Gag was also capable of budding when the PSAP motif was replaced with a PPPY motif derived from MLV, and viral budding in this context was also enhanced by a catalytically active YFP-WWP1 in the absence of Gag ubiquitination [67,79]. Thus, a trans-acting factor must be the target of ubiquitination since Gag cannot be ubiquitinated in this context. Some plausible candidates for ubiquitinated trans-acting factors are the arrestin-related trafficking (ART) proteins. This family of proteins interacts both with HECT domain ubiquitin ligases (WWP1, WWP2, Itch and NEDD4) and ESCRT-associated proteins (Tsg101, ALIX and ubiquitin), therefore providing potential bridging interactions with the ESCRT machinery [41,80]. ARTs can be recruited to sites of viral budding and they reduce MLV budding when overexpressed, but more definitive evidence is needed to support their role in PPXY-dependent budding . One related hypothesis is that the identity of the ubiquitinated protein may not be critical for viral budding to proceed, as long as it is located in close proximity to Gag to allow the recruitment of ubiquitin-binding components in the ESCRT pathway .
In addition to the established role in viral particle release at the plasma membrane, ESCRTs have also been implicated in intracellular viral replication events that include both RNA and DNA viruses. The positive-strand RNA plant viruses including Tomato bushy stunt virus and Brome mosaic virus recruit ESCRT-III to facilitate inward invagination of peroxisome or endoplasmic reticulum (ER) membranes, respectively, to form replication compartments [81,82]. These compartments are thought both to provide a protective environment, away from viral RNA sensing host defence mechanisms, and to concentrate components necessary for viral replication. Unlike other ESCRT-mediated processes, no membrane scission is involved in the formation of the replication organelles, as this compartment remains attached to the membrane. How this incomplete budding event is controlled remains unknown but potential clues have been provided by the functional characterisation of ESCRT requirement in flavivirus replication. Propagation of dengue virus and Japanese encephalitis virus was inhibited by depletion of Tsg101 or ESCRT-III subunits. However, unlike other enveloped viruses, VPS4 is dispensable for flavivirus replication, perhaps suggesting that unknown viral mechanisms inhibit the recruitment of VPS4 in this context to prevent membrane scission by ESCRT-III .
Other uses of the ESCRT machinery in viral replication include herpesviruses. This family of viruses requires two separate envelopment stages for assembly and egress . Primary envelopment allows nucleocapsids to exit the nucleus through budding at the inner nuclear membrane (INM) and fusion of the resulting perinuclear virion with the outer nuclear membrane, subsequently releasing nucleocapsids into the cytoplasm. Secondary envelopment promotes budding of nucleocapsids, together with viral tegument proteins, into the lumen of cytoplasmic membrane compartments that contain the viral envelope proteins . Herpesviruses are likely to use the ESCRT machinery at multiple stages during assembly, as Epstein-Barr virus recruits ALIX through the viral protein BFRF1 for nuclear egress [85,86], whereas herpes simplex virus-1 (HSV-1) requires ESCRT-III for secondary envelopment at the trans-Golgi network/endosomal compartments . Although HSV-1 encodes potential L-domain motifs, the viral mechanisms that recruit ESCRT-III to promote secondary envelopment remain unclear .
ESCRTs in cytokinesis
Cytokinesis begins with the establishment of the spindle midzone, which derives from the metaphase spindle after mitosis. The central area of the spindle midzone, or central spindle, is formed by microtubules overlapping at their plus ends. The central spindle subsequently acts as a signalling hub from which signals to the cell cortex emanate to specify the central cleavage plane, where the cleavage furrow is constricted by an actomyosin ring [89,90]. Two important components of the central spindle orchestrate the cytokinetic process. The centralspindlin complex is composed of two mitotic kinesin-like protein 1 (MKLP1) kinesin-6 motor subunits and two cytokeratin-4 (CYK-4) Rho-family GTPase-activating subunits . The chromosomal passenger complex (CPC) is composed of Aurora B kinase, INCENP (inner centromere protein), Borealin and Survivin . Both centralspindlin and the CPC play roles in stabilising the central spindle and they are activated by dephosphorylation of cyclin-dependent kinase 1 (CDK1) substrate residues [93,94]. Aurora B-mediated phosphorylations of centralspindlin and other components of the spindle midzone are also important in central spindle maintenance . The guanine-nucleotide exchange factor ECT2 (epithelial cell-transforming sequence 2 oncogene) binds to the CYK-4 component of centralspindlin and thereby induces localised activation of the GTPase RhoA at the cell cortex in response to decreasing CDK1 activity . This localised RhoA activity determines the cleavage plane at which actomyosin ring formation occurs through nucleation of actin and myosin II filaments [97,98]. These filaments contract to eventually form a thin intercellular bridge connecting daughter cells known as the midbody, with a central electron dense region corresponding to the compacted microtubules of the central spindle known as the Flemming body [99,100]. Once the midbody is formed, the actomyosin ring is disassembled through inactivation of RhoA by PKCε , and a decrease in membrane associated PI(4,5)P2 levels that result from the action of Rab35 and p50RhoGAP, which are delivered to the midbody by RAB11 (Ras-related protein Rab-11A)/FIP3 (Rab11 family-interacting protein 3)-positive endosomes [102,103].
The final event in cytokinesis is termed abscission and requires the resolution of the midbody to facilitate the physical separation of the daughter cells [104,105]. The molecular mechanism underlying abscission has remained elusive until recently as the identity of the membrane remodelling activities involved in this process was not clear. An important clue to better understand abscission was the identification of centrosomal protein 55 (CEP55) as an essential adaptor protein required for midbody resolution [106,107]. In agreement with this role, CEP55 stays inhibited during the early stages of cytokinesis via its phosphorylation by Polo-like kinase 1 (PLK-1), which inhibits the interaction with MKLP-1 to prevent the premature recruitment of CEP55 to the midbody . Towards the end of anaphase, PLK-1 is targeted for proteasomal degradation as a consequence of its ubiquitination by the anaphase-promoting complex. As the level of PLK1 decreases, cytokinesis progresses and CEP55 is dephosphorylated to allow the interaction with MKLP-1 at the Flemming body .
Functional analogies with viral budding and the interaction with CEP55 were instrumental to uncover the essential role of the ESCRT machinery in cytokinetic abscission [20,21]. This discovery was quickly followed by the surprising identification of CHMP4 and VPS4 homologues that are involved in cell division in Archaea of the genus Sulfolobus, thus suggesting that abscission is the ancestral role for ESCRT proteins as these organisms lack endosomal systems [45,46]. In mammalian cells, CEP55 works as the adaptor protein, analogous to viral Gag proteins, that recruits the ESCRT machinery to the midbody by directly binding to Tsg101 and ALIX [20,21,109,110]. Like MVB formation and viral budding, ESCRT-III activity is required for abscission, as indicated by the requirement for the ALIX–CHMP4B interaction in this process [21,109]. More recent work has suggested that ALIX and the Tsg101/ESCRT-II axis constitute parallel arms that promote CHMP4B recruitment , although this model requires further validation. Abscission requires midbody maturation and thinning from an approximate diameter of 1.5–2 µm to ∼100 nm (Figure 2). This constriction can occur at either side of the Flemming body at an approximate distance of 1 µm and it always forms ∼10–20 min before abscission [112,113]. These secondary ingression zones coincide with the site of abscission and one model suggests that they are formed by fusion of Golgi and recycling endosomes with the membrane. This is thought to be facilitated by members of the exocyst complex and SNAREs [SNAP (soluble NSF attachment protein) receptor], which localise to the midbody via interaction with centriolin rings [114–116]. Alternatively, it has been proposed that ESCRT-III polymerisation itself drives secondary ingression , perhaps facilitated by hIST1 — an ESCRT-III subunit specific to cytokinesis that can direct polymerisation of filaments large enough to promote abscission [118–120].
Time-lapse experiments have shown that whilst MKLP-1 and CEP55 are present at the Flemming body from the time of its appearance, ESCRT proteins localise to the Flemming body at times closer to midbody resolution [112,113]. Importantly, monitoring of fluorescently tagged CHMP4B and CHMP4A shows that closer to the time of abscission, a separate pool of ESCRT-III appears specifically at the secondary ingression site at which abscission occurs [113,121]. Here, the membrane appears rippled containing 17 nm diameter filaments, which are likely to be those of ESCRT-III since they are not observed in CHMP2A-depleted cells . Interestingly, the appearance of this separate ring of ESCRT-III is preceded by an increase in the fluorescence of CHMP4B at the Flemming body, at the same side as the distal pool. VPS4 shows a similar localisation pattern, but appears at the secondary ingression just after CHMP4B, consistent with its role in ESCRT disassembly and recycling following abscission . This ESCRT-III localisation pattern is consistent with a model whereby nucleation of ESCRT-III polymerisation at the Flemming body first occurs followed by VPS4-mediated breakage and constriction of outer ESCRT-III spirals, thus propelling them away from the Flemming body to form the secondary ingression at an equilibrium position. VPS4-mediated disassembly would then mediate the final scission event . Other models have been proposed, such as continuous polymerisation and constriction of ESCRT-III spirals away from the Flemming body to deform the membrane and form the secondary ingression . Schiel et al. propose a model whereby ESCRT-III polymerisation instead stabilises a pre-formed secondary ingression formed by vesicle fusion, such as by FIP3-positive endosomes [103,122]. However, the importance of vesicle fusion remains controversial as the addition of vesicle fusion inhibitors has no effect on abscission .
Cytokinesis is characterised by a tight spatiotemporal regulation, perhaps explaining the requirement for most of the ESCRT-III subunits, in contrast with viral budding which only requires the core membrane remodelling subunits of ESCRT-III . Completion of cytokinesis requires a complex coordination with activities upstream of scission, such as furrow ingression and those involved in membrane binding, midbody stabilisation and microtubule disassembly. In this context, severing of microtubules derived from the spindle midzone has been shown to be a rate-limiting essential step in cytokinesis and its occurrence at secondary ingression sites correlates closely with completion of abscission [112,124]. The microtubule severing AAA-ATPase Spastin plays an essential role in this process, and when depleted, abscission is delayed [112,124]. All ESCRT-III subunits encode microtubule-interacting and trafficking (MIT) domain-interacting motifs (MIMs) that interact with MIT domain containing effector proteins, notably VPS4A/B and its effector LYST-interacting protein 5 (LIP5) [125,126]. In addition to VPS4, Spastin is another MIT domain-containing protein that is specifically recruited to sites of abscission via the MIMs of CHMP1B and hIST1 [124,127]. These interactions would be consistent with a role for Spastin in abscission, thus suggesting a mechanism that co-ordinates microtubule severing with ESCRT-III-mediated scission. An alternative model suggests that buckling of microtubules by spastin contributes to abscission, rather than microtubule severing. This result is based on the disorganisation of the central spindle microtubules in Spastin-depleted cells . However, subsequent work has shown that Spastin plays another key role during mitotic exit by coordinating NE sealing and spindle disassembly at NE–microtubule intersection sites, perhaps explaining the disorganised spindle in Spastin-depleted cells .
The ESCRT-III–Spastin interaction highlights an example of the adaptability of the ESCRT machinery that is required to facilitate cytokinetic abscission. A second adaptation of the ESCRT machinery is illustrated by MIT domain containing 1 (MITD1), which binds MIMs encoded by several ESCRT-III subunits, including CHMP1A, 1B, 2A and hIST1 [129,130]. Whilst the N-terminal MIT domain in MITD1 mediates the interactions with ESCRT-III, the C-terminal domain has a phospholipase D-like fold with a positively charged surface patch that interacts with PtdIns (phosphatidylinositol)-containing membranes. The phenotypic characterisation of MITD1-depleted cells suggests multiple roles in cytokinesis as evidenced by premature abscission, increased cortical blebbing and cytokinesis failure. These phenotypes are consistent with a role of MITD1 in maintaining the stability of the midbody and coordinating abscission with earlier cytokinetic events, perhaps by stabilising ESCRT-III filaments and regulating the actin cytoskeleton [129,131].
The different MIT domain-containing proteins involved in abscission could be brought into close proximity by interacting with the multiple MIMs that are present in the ESCRT filaments to couple activation of VPS4, membrane scission and microtubule severing (Figure 2). However, despite this progress in our understanding of abscission, some important questions remain unanswered. For example, the midbody persists from 80 min to several hours before abscission, depending on the cell type. The events that occur during this time remain largely unknown. A partial clue comes from the regulation of abscission by midbody tension , which can explain an ∼30 min delay in HeLa cells, but we still do not understand the events that precede abscission in these cells. It will also be important to understand how the recruitment of each of the ESCRT subunits is co-ordinated, and the nature of the signal that triggers the polymerisation of ESCRT-III in the final moments of abscission.
ESCRTs and the abscission checkpoint
One of the most striking adaptations of the ESCRT machinery for coordinating cytokinetic abscission comes from its interaction with the Aurora B-dependent abscission checkpoint, also known as NoCut [104,131]. This evolutionarily conserved control system ensures that abscission is delayed until anaphase chromatin bridges, caused by segregation errors, have been removed from the intercellular bridge connecting daughter cells [133–135]. Such segregation errors are often caused by defects in chromosome architecture, such as impaired decatenation of DNA and dicentric chromosomes [136–138]. Failure to restrain abscission until clearance from the midbody leads to cleavage furrow regression, tetraploidisation and DNA damage. NoCut was originally identified in yeast, in which the Aurora B homologue, Ipl1 kinase, was shown to inhibit septin-mediated abscission upon interaction with chromatin . In addition to chromatin bridges, the abscission checkpoint also delays midbody resolution in response to defective NPC assembly  and high levels of midbody tension that result from cell growth at low densities .
Aurora B plays a crucial role in the coordination of cytokinesis . One of its critical functions in this process is the phosphorylation of MKLP1 to stabilise the cleavage furrow, while the subsequent inactivation of Aurora B triggers abscission . Crucially, Aurora B activity is sustained in the presence of chromatin bridges to delay midbody resolution, although how this activity is regulated by chromatin is not known [133,135,141]. The signal by chromatin, however, is thought to be specific, since asbestos fibres within the intercellular bridge do not sustain the checkpoint . Intriguingly, the molecular origin of the chromatin bridges has an important effect on checkpoint induction. Whilst chromatin bridges induced by replication stress, condensation or decatenation defects sustain the abscission checkpoint, those due to the formation of dicentric chromosomes are not detected by NoCut, thus resulting in chromosome damage by the abscission machinery .
The physical connection between the abscission checkpoint and the abscission machinery was demonstrated by the functional interaction between regulatory subunits in the ESCRT pathway and components of the abscission checkpoint. More specifically, abscission delays induced by nucleopore disruption and chromatin bridges are abrogated by depletion of CHMP4C , a human ESCRT-III subunit that is closely related to the polymer forming CHMP4B. At a molecular level, CHMP4C engages the CPC by binding to Borealin and this interaction delays abscission via the Aurora B-dependent phosphorylation of CHMP4C at a unique insertion that is not present in CHMP4B [141,143]. Importantly, the interaction of ESCRT-III with the CPC is conserved from Drosophila to humans , and human polymorphisms in the CHMP4C gene are associated with increased risk for ovarian cancer , thus highlighting the importance of this damage preventing regulatory mechanism.
Abscission/NoCut Checkpoint Regulator (ANCHR) has been recently identified as another key component of the abscission checkpoint . ANCHR encodes two MIM sequences that interact with the MIT domain of VPS4 in an Aurora B-dependent manner. This interaction allows the formation of a ternary complex with CHMP4C that retains VPS4 at the Flemming body, thus inhibiting its localisation to the secondary ingressions that mark the abscission sites . Unc-51-like kinase 3 (ULK3) was subsequently identified as an essential regulator of the abscission checkpoint and its kinase activity is required for this function . ULK3 acts downstream from Aurora B to regulate abscission in a CHMP4C-dependent manner. Accordingly, ULK3 phosphorylates CHMP4C at sites distinct from those targeted by Aurora B. A subset of other ESCRT-III subunits is also phosphorylated by ULK3, although the preferred substrate is thought to be hIST1, which binds ULK3 via an especially strong MIT–MIM interaction. Critically, ULK3 phosphorylation of hIST1 is required to sustain the checkpoint in response to lagging chromosomes and nucleopore disruption. However, a mutated version of hIST1 that cannot be phosphorylated by ULK3 still supports abscission delays when the checkpoint is sustained by low cell tension, suggesting that some downstream factors involved in maintaining the checkpoint may differ depending on the stimuli . The exact mechanism by which the various phosphorylation events inhibit abscission is still unclear. One possible scenario is that the initial phosphorylation of CHMP4C by Aurora B could be subsequently ‘amplified’ to other ESCRT-III subunits by ULK3 (Figure 3). These phosphorylations could lock the ESCRT-III subunits such as hIST1 in their ‘closed’ inactive forms. It is also possible that the various phosphorylations may retain ESCRT-III and VPS4 at the Flemming body, thus preventing its polymerisation and scission activity at the secondary ingression sites. In agreement with this model, phosphomimetic mutations enhance the hIST1 interactions with VPS4 and LIP5 , a cofactor that promotes VPS4 oligomerisation and ATP hydrolysis [146–148].
One important question that remains open is the relevant subcellular location where CHMP4C inhibits abscission. One possibility is that CHMP4C might form an inhibitory complex in the cytoplasm, as suggested by the distinctive localisation of Aurora B to cytoplasmic foci in cells arrested by the abscission checkpoint . An independent line of evidence suggests that abscission regulation would require the midbody localisation of CHMP4C , which would be facilitated by ALIX and MKLP1 [111,149]. Once recruited to the midbody, CHMP4C is initially found in the midbody arms, whereas the phosphorylation of serine 210 by Aurora B allows its subsequent localisation to the Flemming body . Interestingly, the CHMP4C subset found at the midbody arms is phosphorylated at residues 214, 215 and 210, whereas the phosphorylation of residues 214 and 215 is lost in the transition to the Flemming body . In agreement with this notion, the mitotic phosphorylation of CHMP4C decreases around the time of abscission [140,141], suggesting that a phosphatase activity of unknown identity may be required to reverse the inhibitory activity of CHMP4C, thus allowing abscission in midbodies that are arrested by the abscission checkpoint.
ESCRTs at the nuclear envelope
The nuclear envelope (NE) is a double membrane structure continuous with the ER that acts as a barrier, establishing both nucleo-cytoplasmic compartmentalisation and protection of the genome from cytoplasmic nucleases . Whilst yeast undergoes a closed mitosis, mammalian cells undergo an open mitosis, in which the NE and the nuclear lamina that lines the inner nuclear membrane (INM) are disassembled, allowing chromosomes to access the mitotic spindle at prometaphase [150–153]. Dynein-mediated microtubule tearing breaks down the NE, co-ordinated by numerous kinases including regulators that are shared with cytokinesis, such as CDK1, Aurora B and PLK-1 . The broken down NE is incorporated into the mitotic ER, which remains intact, away from the spindle and segregating chromosomes [154–159]. Opposite events occur during NE reformation, during which Aurora B and PLK-1 relocate to the central spindle and midbody, to control cytokinesis, and CDK1 is inactivated . The NE is reformed from the ER by attachment of ER tubules to decondensing chromosomes . These tubules flatten and enclose chromatin, leaving discontinuities of the double membrane primarily at sites of mitotic spindle attachment [157,160–162]. Annular fusion must therefore take place to seal such holes in close coordination with microtubule severing. Two ground breaking studies have recently established the essential role of the ESCRT machinery in this process.
As observed for other functions of the ESCRT pathway, NE resealing requires the core membrane remodelling activity of ESCRT-III [27,28]. Accordingly, knockdown of ESCRT-III subunits disrupts the integrity of the NE and this phenotype correlates with increased DNA damage. This function is further supported by the localisation of CHMP2A and CHMP4B to the reforming NE into nucleo-cytoplasmic channels that are topologically equivalent to other membrane tethers resolved by the ESCRT pathway [27,28]. In a striking parallel with cytokinesis, hIST1 is required in NE reformation to recruit spastin and promote the disassembly of spindle microtubules . As observed in cytokinesis, microtubule severing is a rate-limiting step of the NE resealing process, as CHMP4B–eGFP persisted at unsealed holes upon inhibition of spastin activity . In contrast with cytokinesis, CHMP1B depletion did not perturb spastin localisation to the nucleus, suggesting differential routes of recruitment in these two processes .
The two most likely candidates identified so far as adaptor proteins responsible for ESCRT-III recruitment to the NE are ubiquitin fusion degradation protein 1 (UFD1) and CHMP7. UFD1 is an adaptor protein of the p97/UFD1/NPL4 AAA-ATPase that has previously been implicated in NE resealing [163,164]. Prior to this, a related complex containing p97 and p47 is involved in ER membrane recruitment to chromatin . P97/UFD1/NPL4 is also known to facilitate removal of Aurora B from chromatin to allow chromosome decondensation [163,164]. UFD1 was shown to be necessary for ESCRT-III recruitment to the reforming NE, and this function correlates with its interaction with CHMP2A . CHMP7 is essential for ESCRT-III recruitment to the NE, and this function correlates with a potential interaction with CHMP4B . Closer examination of its structure reveals a C-terminus similar to CHMP6 that is necessary for binding to CHMP4B, whereas the extended N-terminus contains tandem winged-helix domains that resemble the ESCRT-II subunit ELL-associated protein of ∼20 kDa (EAP20) [165,166]. The first of these tandem domains contains an extended loop between the β2–β3 hairpin that is important for the specific localisation of CHMP7 at the ER membrane . The continuity of the ER with the NE suggests a model whereby CHMP7 provides a platform that orchestrates the recruitment of ESCRT-III to the reforming NE.
The emerging role of the ESCRT pathway in establishing NE integrity has been extended beyond the mitotic nuclear regeneration. NE breakdown is thought to be a rare event outside mitosis , but recent work has shown that NE rupture occurs during migration of cells of the immune system, and similar nuclear ruptures have been observed in cancer cells during metastasis [168,169]. Studies monitoring cell migration through confined spaces have shown that NE blebbing occurs at the leading end of the cell in response to increased nuclear hydrostatic pressure, and the NE is forced through ruptures in the nuclear lamina . These blebs eventually burst resulting in nucleo-cytoplasmic mixing and DNA damage. The rapid localisation of ESCRT-III to the site of NE rupture is associated with a repair function that results in nucleo-cytoplasmic re-compartmentalisation and reduced DNA damage. Following the repair by ESCRT-III, the accumulation of lamin A forms ‘lamin scars’ that increase the local resistance of the NE to protect against subsequent rupture at the same sites .
The quality control of nuclear pore complexes (NPCs) has been identified as another function of the ESCRT pathway to maintain nucleo-cytoplasmic compartmentalisation . NPCs are central to maintaining nuclear identity by controlling entry and exit of proteins and RNA, and consist of ∼600–700 individual subunits from a repertoire of ∼30 different nucleoporins (Nups) [171–173]. Epistasis screens in yeast have uncovered a role for ESCRT-III in the clearance of defectively assembled early NPC intermediates [29,174]. Since yeast undergoes a closed mitosis, involving asymmetric division in which the NE is not broken down and reassembled, such a quality control mechanism is thought to be essential to re-establish a fully functional NE [29,170,175]. As yet, there is no evidence for an equivalent role for ESCRT-III in mammals, perhaps due to less demand as a consequence of the open mitosis in these cells.
The initial clue to uncovering the role of ESCRTs in NPC surveillance was provided by the conserved genetic interaction in fission and budding yeast between Vps4 and the transmembrane nucleoporin POM152 . Additional genetic interactions were subsequently observed between POM152, Nup170 and the ESCRT-III subunits sucrose nonfermenting protein 7 (Snf7) (CHMP4), Vps24 (CHMP3) and Vps2 (CHMP2) . Snf7 and, more recently, Chm7 (CHMP7) , were also found to interact with the INM LEM (LAP2, emerin, MAN1) domain proteins Heh1 and Heh2, both of which interact with Nup170, perhaps during NPC assembly . Chm7 was shown to be recruited to sites of incomplete NPC assembly , and its recruitment was shown to require Heh1/2 . This strongly suggests that Heh1/2 acts as putative adaptors for ESCRT-III recruitment to the NPC . In agreement with this model, deletion of Heh2, ESCRT-III or Vps4 leads to clustering of malformed NPCs in a structure termed the storage of improperly assembled NPCs (SINCs) . Conversely, Chm7 deletion prevents SINC formation suggesting a role in defective NPC clearance . This storage structure forms at a single location at the NE and is therefore not inherited by the daughter cells, thus representing an additional level of quality control of the NE in yeast [29,170]. Importantly, Chm7 is also required for sealing of the NE over the top of malformed NPCs . This demonstrates a conserved mode of action in both NE resealing/repair and defective NPC removal through sealing of the NE to maintain nucleo-cytoplasmic compartmentalisation.
How ESCRT-III and Vps4 could also potentially direct removal of defective NPCs is not clear. The catalytic activity of Vps4 has been suggested to disassemble the defective intermediates, either directly or through the removal of NPC-bound ESCRT-III [29,170,178]. An alternative model would involve the ESCRT-III-dependent budding of vesicles that contain the defective NPCs into the intermembrane space, a process that may resemble the nuclear egress of Epstein–Barr virus [85,86].
ESCRTs at the plasma membrane
A direct role of the ESCRT pathway in remodelling the plasma membrane in diverse functional contexts has emerged in the last few years. One of these processes bears striking resemblance to viral budding, namely the formation of arrestin domain containing 1 (ARRDC1)-mediated microvesicles (ARMMs) [179,180]. A central PSAP motif in ARRDC1 mediates interaction with Tsg101, as in viral budding, whereas localisation to the plasma membrane is directed via an N-terminal arrestin domain. Furthermore, two C-terminal PPXY motifs mediate interaction with WWP2 ubiquitin ligase, which ubiquitinates ARRDC1 and confers optimal ARMM release . ARMM formation is likely to require the core ESCRT machinery since vesicle release is inhibited by catalytically inactive VPS4. However, the subset of ESCRT-III units required for this process needs to be determined . Whilst the function of ARMM release remains uncharacterised, ARRDC1 can be transferred from donor to recipient cells, suggesting a role for ARMM contents or membrane-bound proteins in intercellular communication . This transfer of material between cells also raises the question of how the released vesicles fuse with the target cells. It is tempting to speculate that cellular fusogens may resemble the fusion activity of viral envelope proteins, to deliver the microvesicle content into the target cells. ESCRTs have also been implicated in shedding of ubiquitinated T-cell receptor (TCR) containing microvesicles from T cells at immunological synapses, although the involvement of ARRDC1 in this budding event is not clear. This mechanism facilitates the interaction of TCRs with major histocompatibility complex-bound antigens on antigen-presenting cells and promotes signal transduction .
Another ESCRT-mediated event at the plasma membrane is the microvesicle shedding induced by depletion of the lipid flippase TAT-5 . Loss of this P4-ATPase disrupts phosphatidylethanolamine symmetry leading to accumulation of this lipid on the exterior side of the plasma membrane (PM). This lipid asymmetry in turn triggers loss of cell adhesion and increased vesicle shedding, leading to a thickened appearance of the membrane, which contains components of the ESCRT machinery . Although the relevant adaptor protein is unknown, ESCRT-0 and ESCRT-I are thought to be necessary for membrane thickening, but the role of ESCRT-III in ectosome release needs to be established .
In addition to formation and shedding of microvesicles, ESCRTs have been implicated in PM repair by an analogous process that involves pinching out of the damaged regions of membrane followed by scission and shedding . The plasma membrane can be damaged by exposure to bacterial pore forming toxins [183,184] and mechanical stress, such as that seen in muscle tissue [185,186]. Experimental approaches to induce membrane wounding include laser ablation, detergents and micropipettes. All membrane repair mechanisms described so far share a dependence of calcium influx into the cytoplasm following wounding [186,187]. Such mechanisms include clotting, patching and endocytosis or exocytosis of damaged regions [188–191]. Likewise, the rapid recruitment of ESCRT-III to sites of wounding is calcium-dependent [22,23]. Whilst ESCRT-0 and ESCRT-II are not involved in PM repair, Tsg101, ALIX, CHMP3, 2A, 2B, 1A and importantly 4B have all been shown to localise to the PM upon wounding, and wound closure correlates with maximum CHMP4B levels at the repair site [22,23]. ALIX is thought to play an important role in wound repair by bridging the calcium-sensing protein apoptosis-linked gene-2 (ALG-2) with ESCRT-III at the PM, perhaps nucleating ESCRT-III polymers via its interaction with CHMP4B [22,23]. ALG-2 therefore appears to act as the adaptor for ESCRT recruitment to the PM. This model is consistent with the sequential recruitment of ALG-2, ALIX, ESCRT-III and VPS4 to sites of wounding, as shown by confocal microscopy and TIRF imaging [22,23].
Finally, ESCRTs also play indirect roles at the plasma membrane, for example, by promoting the release of exosomes . This function requires binding of the intracellular adaptor protein syntenin and ALIX, and this interaction promotes the release of exosomes as ILVs into multivesicular bodies by remodelling the endosomal membranes [193,194]. These late endosomal compartments subsequently fuse with the plasma membrane in an ESCRT-independent manner, thus releasing the exosomes to the extracellular medium.
ESCRT-III and membrane remodelling
In contrast with ESCRT-0, -I and -II, ESCRT-III does not form a stable complex, instead forming a transient assembly at sites of membrane remodelling to direct scission . The domain organisation and structure of CHMP proteins has been well characterised [120,196], and studying MVB formation and viral budding has helped define the core CHMP proteins required for all ESCRT-mediated processes. Despite this progress, a defined, unified mechanism for membrane sculpting and severing by ESCRT-III remains to be established. Likewise, functional studies in cytokinesis indicate that each of the CHMP proteins may play differential roles and cannot compensate for one another [117,131,197]. However, precise roles for each of the CHMPs in ESCRT-III polymerisation and membrane remodelling remain to be defined.
Crystal structures for CHMP1B, 3, 4 and hIST1 have revealed that all CHMPs share a similar structure [120,196,198–200]. In their ‘closed’ soluble autoinhibited conformation, they consist of a highly structured N-terminal 4-helix bundle and an unstructured C-terminus, containing another two helices [120,196,199,201,202]. In their ‘open’ polymerisation competent forms, the second and third helices from the N-terminal bundle form a single elongated helix which forms a positively charged hairpin, together with helix one , which binds acidic lipids to promote membrane binding [120,196]. Interactions between ESCRT-III subunits are also mediated by this hairpin, particularly at the tip [120,196,199]. Other regions involved in intersubunit interactions remain to be confirmed but are likely to include other surfaces of the first four helices that together comprise the core structure [56,198,200] and the fifth helix [120,196,201,203,204]. The fifth helix is, however, more commonly associated with an autoinhibitory function, maintaining CHMPs in their ‘closed’ monomeric conformation in the cytoplasm when not in use. The fifth helix folds back on the four helix bundle forming numerous contacts with the helices [120,196,201,202,204,205]. MIM domains are present in the C-terminus of all CHMPs except CHMP3 [199,202,206]. MIT domain-containing proteins can stabilise the open conformation of the CHMPs by binding to the conserved C-terminal MIM .
Studies in yeast initially showed the highly ordered assembly of CHMPs in a sequence that starts with Vps20 (CHMP6) and follows with Snf7 (CHMP4), Vps24 (CHMP3) and Vps2 (CHMP2), which recruits Vps4 to promote ESCRT-III disassembly [9,195]. According to this model, Vps20 would act as the nucleator for initial polymerisation of Snf7, which is thought to be the main constituent of the ESCRT-III filaments [204,207]. These polymers would be capped by the Vps24/Vps2 dimer that in turn recruits Vps4 for filament disassembly [125,195,208]. The specific roles of these core subunits are thought to be conserved in mammals [209,210], although additional proteins, such as ALIX and CHMP7, may also have nucleating activity in higher eukaryotes [28,40,165,211]. Nucleation of ESCRT-III involves transition of CHMPs from ‘closed’ to ‘open’ forms as a result of structural reorganisation upon binding to ESCRT-II/ALIX [195,207,212–214]. Membrane curvature has also been shown to promote nucleation [213,215]. One interesting feature observed in some of the nucleating factors is their ability to dimerise. For example, ESCRT-II contains two EAP20 subunits which are each able to bind separate CHMP6 molecules and therefore nucleate two ESCRT-III filaments [207,212,214,216,217]. ALIX can also dimerise through its V domain and bind two CHMP4 subunits . Such pairing of ESCRT-III filaments is likely to account for some of the diversity in filament width that is observed in vitro and in vivo [112,200,203,218,219]. In fact, ALIX has been shown to bundle pairs of 3 nm CHMP4B filaments in vitro [205,211], whereas the spiral filaments formed in vivo when CHMP4B is overexpressed have an approximate width of 6 nm , perhaps suggesting that bundling of two ESCRT-III filaments could underlie the formation of these structures. Similar observations have been made in the context of yeast Snf7, whereby bundling produces proto-filaments wider than 9 nm .
Electron microscopy has shown a variety of structures, both in vitro and in vivo for various ESCRT-III filaments, including spirals, tubes, coils, cones and domes [200,204,208,218–220]. The overexpression of CHMP4A in COS-7 cells forms flat spirals, which form tubes projecting from the plasma membrane when a VPS4 dominant-negative mutant is expressed, consistent with membrane budding away from the cytoplasm . Yeast Snf7 forms a variety of filaments, rings and sheets in vitro [205,211]. A recent study showed that, when incubated with membranes, Snf7 forms spirals in which the inner ring is over bent, hence storing less energy than outer rings which are underbent . Such spirals showed spring-like buckling activity when they eventually covered the entire membrane, pushing ESCRT-III outwards, suggesting that lateral compression causes the spirals to deform into tubes, releasing stored elastic energy to remodel membranes . Like CHMP4, CHMP2A also forms spirals when the C-terminal autoinhibitory region is removed [208,219], whereas CHMP2B forms tubes of varying diameters when overexpressed . Importantly, coexpression of CHMP2A and CHMP3 forms heteropolymeric tubes that end in dome-shaped caps [203,208,223]. Another fascinating study has recently determined a high-resolution hIST1–CHMP1B structure by cryo-electron microscopy . The tubular structure of this straight heteropolymer revealed an outer surface of ‘closed’ hIST1 that is negatively charged, whilst CHMP1B assumed an ‘open’ state in the inner surface that forms a cationic interior . Remarkably, the inner surface was able to bind membranes and promote scission of membrane tubules with normal topology, i.e. scission from the outside of the membrane rather than from the inside, as commonly associated with the ESCRT machinery . This study raises the question of whether closed CHMP conformations always correspond to autoinhibited incompetent forms, or just membrane-binding forms. Secondly, it raises the surprising possibility that CHMPs are indeed capable of directing membrane scission in diverse topological contexts, a model supported by increased endosomal tubulation in hIST1-depleted cells .
Various models for ESCRT-III action have been proposed based on the structures observed so far . These models take also into account the importance of CHMP4 and CHMP2/3 subunits, energetic considerations and localisation of subunits in relation to cargo in MVB formation, as well as observations in the context of viral budding. The ‘whorl’ model of ESCRT-III assembly was proposed to accommodate the initial role of ESCRT-I and -II in initial membrane stabilisation at bud necks of ILVs . ESCRT-I and -II form a crescent-shaped supercomplex that is believed to facilitate initial membrane bending that aids ESCRT-III nucleation. Pairs of ESCRT-III filaments are proposed to form spokes that extend away from each ESCRT-I–ESCRT-II complex, meeting at the site of scission . The natural propensity for filaments to bend would allow the ESCRT-III filaments to form a ‘whorl’ shaped structure that would constrict the membrane towards the site of scission. VPS4 has been suggested to play an organisational role in this context by clustering the ends of the spokes at the site of scission . Dome models have gained more plausibility given the strong experimental evidence that supports dome-shaped caps at the ends of tubular structures formed by CHMP2/CHMP3 [208,227]. Crucially, this model has the potential to explain how ESCRT-III could remodel membrane necks of different diameters ranging from 50 nm, as required for viral budding, to 1 µm in cytokinesis. Dome models involve binding of membrane to the outside of dome-shaped caps of tubules, which pull opposing membranes together to constrict bud necks and drive scission [208,225,227,228]. Whether membrane scission requires VPS4 in this context is unclear at present [200,228]. It is also unclear whether CHMP2 and 3 are incorporated into growing CHMP4 filaments or whether they form a separate dome-shaped cap on the end of the tubes [112,204,208,229]. This aspect could have profound implications in this model as a switch from CHMP4 homopolymerisation to heteropolymerisation of CHMP2/3 would be required to achieve membrane scission.
In addition to ‘whorl’ and ‘dome’ models, inverse buckling of ESCRT-III cones/domes has been suggested to drive membrane scission [221,230]. In this model, over bent filaments at the tip of the cones are severed from upstream ESCRTs, such as ESCRT-I, ESCRT-II and ALIX, by the action of VPS4, causing stored energy to be released by inverse buckling, back into 2D spirals [225,230]. Release of energy stored in the bent membranes would simultaneously drive scission. This model would require outward growth of filaments from the narrow tip towards the wider end of the cone, consistent with tapering of the cone towards HIV-1 Gag seen by deep-etch electron microscopy (DEEM) . However, it would be difficult to apply this model to cytokinesis, where ESCRT-I, ESCRT-II and ALIX localise to the Flemming body and have not been observed at the abscission sites [112,113]. The orientation of ESCRT-I and -II complex-bound cargo in relation to the ESCRT-III cones also makes this model attractive in terms of providing an explanation for how MVB sorted cargo is sorted into ILVs whilst ESCRT-III monomers are recycled to the cytoplasm [225,231].
Whilst the role of VPS4 in disassembly and recycling of ESCRT-III is well established, it is still unclear whether or not it plays a direct role in membrane deformation and/or the actual membrane scission event. Neither Vps4 nor Vps2 appears to be required for scission of ILVs in yeast, but both are required for subsequent rounds of scission, arguing for an ESCRT-III recycling role for Vps4 [229,232]. Furthermore, the hIST1–CHMP1B helical tubes discussed above could be formed in the absence of VPS4, suggesting a dispensable role for polymer assembly, at least in vitro . Despite this evidence, the role of VPS4 remains an open question, and a more direct role in membrane remodelling is possible, given its localisation to sites of membrane scission such as the stalks of budding HIV-1 virions [233,234] and abscission sites during cytokinesis prior to scission . One simple model that could unify several observations is the sequential removal of ESCRT-III subunits by VPS4, which could drive membrane constriction perhaps by constricting the ESCRT-III filaments up to a point where spontaneous membrane scission can occur [227,235].
As summarised in this section, there has been spectacular progress over the last few years in our understanding of the molecular mechanisms employed by ESCRT-III to promote membrane scission. However, the mechanism of ESCRT-III polymerisation should be treated with caution at this stage due to the overexpressed nature of some of the experimental systems, as well as the limitation posed by the lack of the entire set of ESCRT-III subunits that are found in physiological conditions. In these respects, greater effort should be devoted to better understanding the structure of the ESCRT-III filaments assembled under strict physiological conditions. Cytokinetic abscission perhaps represents an interesting area that could benefit from such work, given the large size of the polymeric ESCRT-III structures that are proposed to promote the physical separation of daughter cells.
The co-option of the ESCRT machinery by an increasing number of intracellular processes, together with the identification of various adaptors for recruitment of ESCRT-III to specific sites, has reinforced the modular nature of the ESCRT machinery. The control in cytokinesis by the abscission checkpoint, as well as the role of ESCRTs in multiple events at the NE, has demonstrated the interplay and precise spatiotemporal coordination of the ESCRT machinery with cellular processes that are involved in quality control.
The binding partners of the different L-domains and the essential ESCRT-III subunits required for viral budding are well established. However, the biggest remaining question in this field is exactly how ESCRT-III is recruited to the sites of viral budding, and how membrane remodelling is achieved in this context. Whilst ALIX provides direct bridging of viral structural proteins with ESCRT-III, we still lack a clear consensus regarding ESCRT-III recruitment in PTAP and PPXY-dependent budding. Questions that remain unanswered in cytokinetic abscission include how ESCRT-III is recruited to the sites of abscission that are adjacent to the Flemming body, and how the membrane tube of ∼1 µm is severed, given that the diameters resolved in other ESCRT-mediated processes are much smaller, in the range of 50–100 nm. Why the midbody persists for so long before abscission occurs, and what triggers the final cut is also unclear. Intriguing connections between the nucleus and the midbody are suggested by the delays in abscission that are induced upon disruption of the nucleopores, and the nature of this signal needs to be uncovered. This connection is further highlighted by the recent discoveries that demonstrate the essential role of the ESCRT machinery in establishing and maintaining the integrity of the NE, suggesting that a tight coordination of the ESCRT complexes is needed at multiple steps during the final events of cell division. In fact, this coordination might be required in other cellular compartments, as the localisation of CHMP7 in interphase strongly suggests potential roles of the ESCRT pathway at the ER.
The variety of structures seen for homo and heteropolymers of CHMP proteins in vitro and in vivo has provided the basis to better understand the dynamic conformation of ESCRT-III and the mechanism that promotes membrane scission. However, a better characterisation of the polymers formed in vivo is required to understand membrane remodelling by ESCRT-III; it is even possible that different structures might be adopted by ESCRT-III at different sites of action. Further insights into the mechanism of ESCRT-III action is likely to come from higher resolution microscopy techniques, such as the DEEM employed so far, and newly developed super-resolution microscopy will be better suited to address these issues.
In terms of the clinical implications of ESCRTs, their role in viral maturation and budding presents potential avenues for therapeutic intervention in HIV-1, Ebola, Hepatitis C and other emerging viral infections. Changes in expression levels of several ESCRT proteins have also been linked to various forms of cancer, and the connection between polymorphisms in human CHMP4C and increased risk for ovarian cancer is particularly exciting. Other relevant connections with human disease that need to be further explored include the link between CHMP2B and neurodegenerative diseases such as frontotemporal dementia and Alzheimer's disease [236–238].Figure 1,Figure 2,Figure 3
Site specific adaptors and pathways for ESCRT-III recruitment.
Localisation and action of the ESCRT machinery during cytokinetic abscission.
The ESCRT machinery in the NoCut Aurora B-dependent abscission checkpoint.
ALG-2-interacting protein X
Abscission/NoCut Checkpoint Regulator
arrestin domain containing 1
cyclin-dependent kinase 1
centrosomal protein of 55 kDa
charged multivesicular body protein
chromosomal passenger complex
deep-etch electron microscopy
ELL-associated protein of 20 kDa
ELL-associated protein of 45 kDa
equine infectious anaemia virus
endosomal sorting complex required for transport
homologous to the E6-AP carboxyl terminus
human Increased sodium tolerance 1
human immunodeficiency virus
herpes simplex virus-1
inner nuclear membrane
LYST-interacting protein 5
major histocompatibility complex
MIT domain-interacting motif
microtubule-interacting and trafficking
MIT domain containing 1
mitotic kinesin-like protein 1
murine leukaemia virus
neural precursor cell expressed developmentally down-regulated protein 4
nuclear pore complex
Polo-like kinase 1
Rous sarcoma virus
storage of improperly assembled NPCs
sucrose nonfermenting protein 7
total internal reflection fluorescence microscopy
tumour susceptibility gene 101
ubiquitin enzyme variant
ubiquitin fusion degradation protein 1
Unc-51-like kinase 3
vacuole protein sorting
WW domain containing E3 ubiquitin protein ligase 1
WW domain containing E3 ubiquitin protein ligase 2.
Work in J.M.-S.'s laboratory is funded by the Wellcome Trust [WT093056MA].
We thank Monica Agromayor for critical reading of the manuscript and Anna Caballe for help with the figures.
The Authors declare that there are no competing interests associated with the manuscript.
Juan Martin-Serrano was awarded the Biochemical Society’s GlaxoSmithKline Award in 2014; this review is based on the Award Lecture.