Membrane Fission Reactions of the Mammalian ESCRT Pathway

Abstract

The endosomal sorting complexes required for transport (ESCRT) pathway was initially defined in yeast genetic screens that identified the factors necessary to sort membrane proteins into intraluminal endosomal vesicles. Subsequent studies have revealed that the mammalian ESCRT pathway also functions in a series of other key cellular processes, including formation of extracellular microvesicles, enveloped virus budding, and the abscission stage of cytokinesis. The core ESCRT machinery comprises Bro1 family proteins and ESCRT-I, ESCRT-II, ESCRT-III, and VPS4. Site-specific adaptors recruit these soluble factors to assemble on different cellular membranes, where they carry out membrane fission reactions. ESCRT-III proteins form filaments that draw membranes together from the cytoplasmic face, and mechanistic models have been advanced to explain how ESCRT-III filaments and the VPS4 ATPase can work together to catalyze membrane fission.

Overview

The mammalian endosomal sorting complexes required for transport (ESCRT) pathway comprises more than 30 proteins that catalyze a series of important cellular membrane fission events, including intraluminal vesicle formation at endosomes, virus and vesicle budding from the plasma membrane, and abscission of the intercellular bridge during cytokinesis (Figure 1). In each case, membrane fission is effected from within the cytoplasmic face of thin membrane tubules and therefore differs from vesiculation processes such as endocytosis, where fission is effected from the vesicle neck exterior. Thus, the ESCRT pathway is a mobile machinery that can be recruited to different cellular membranes to catalyze “reverse topology” fission events.

Figure 1.

Membrane fission reactions promoted by the mammalian ESCRT pathway. Adaptor complexes that direct the pathway to specific sites of action are shown schematically, and ESCRT-remodeled membranes are highlighted red. Circled numbers denote the panel in Figure 3 that shows the adaptor structure.

The existence of a reverse topology fission machinery was first implied by the observation that late endosomal compartments, termed multivesicular bodies (MVBs), contained intraluminal vesicles and that integral membrane receptors destined for lysosomal degradation trafficked through these vesicles (reviewed in Reference 1). Core ESCRT components were subsequently identified in seminal yeast genetic studies as factors required for MVB vesicle formation and vacuolar targeting of membrane proteins. The pathway was named when it was shown that ESCRT components could be subdivided into discrete complexes with distinct biochemical functions (1, 2). Yeast studies have continued to illuminate structural and mechanistic aspects of the conserved ESCRT pathway, but metazoan ESCRTs perform a series of additional functions that are best studied in their native contexts.

The ESCRT Machinery

The five different classes of proteins and complexes that constitute the core ESCRT machinery can be subdivided into (a) early acting factors that coordinate ESCRT assembly, membrane deformation, and cargo sorting (Bro1 family proteins, ESCRT-I and ESCRT-II complexes) as well as (b) late-acting factors that catalyze membrane fission and ESCRT disassembly (ESCRT-III and VPS4 complexes) (Figure 2 and see the online Supplemental Table 1 for a summary of ESCRT nomenclature). Follow the Supplemental Material link from the Annual Reviews home page at http://www.annualreviews.org. Regulation occurs at multiple steps to ensure that the ESCRT machinery assembles correctly and cooperatively. Control mechanisms include the use of (a) adaptors with specificity for different phosphoinositides and substrates (Figure 3), (b) sequential binding interactions (Figures 2 and 4), (c) specialized subunit isoforms, (d) conformational cycling between “inactive” (soluble) and “active” (assembled, membrane-associated) states, (e) avidity effects that help oligomeric ESCRT subunits to assemble in an “all-or-none” fashion, (f ) regulatory cofactors that promote or inhibit pathway activities, and (g) posttranslational modifications (particularly ubiquitylation) that identify protein cargoes and regulate ESCRT machinery activities (Figure 5).

Figure 2.

Core ESCRT components and their interactions, shown within a stylized bud neck. Lines denote regions of unknown structure, dashes denote linkers, cylinders denote helices, arrows denote protein-protein interactions, and circled numbers denote the panel in Figure 4 or 5 that shows the relevant interaction. Relevant Protein Data Bank identity numbers are shown in parentheses: human ALIX (20EV); yeast ESCRT-I core (2P22); human TSG101 UEV domain (1S1Q); yeast Vsp28p C-terminal domain (2J9U); MAPB domain (3TOW); SOUBA domain (4AE4), note that the MAPB and SOUBA domains would not be present simultaneously in any single ESCRT-I complex; ESCRT-II core (2ZME); EAP45 GLUE domain (2HTH); human CHMP3 (2GD5), with individual subunits modeled into a circular filament; VPS4 MIT (2JQ9); VPS4 core (1XWI), modeled as a hexamer based on the structure of the related AAA ATPase p97 (1E32); Vta1p C-terminal domain (2RKL); and LIP5:CHMP5 (2LXM). Abbreviations: CTD, C-terminal domain; MIT, microtubule interacting and transport; PRR, proline-rich region; Ub, ubiquitin.

Figure 3.

ESCRT adaptors. The functions and cellular locations of the different adaptors are shown in Figure 1. Here, each adaptor is shown as a cartoon representation, as a schematic model showing domain and motif maps, and, where possible, a composite of structurally characterized domains (ribbon diagrams) and intermolecular complexes (surface renderings). Circled numbers in panel a denote ubiquitin complexes that are shown in Figure 5. Relevant Protein Data Bank identity numbers are shown in parentheses. Panel a contains the HRS:STAM helical core (3F1I), HRS tandem VHS-FYVE domain (1DVP), HRS DUIM (2D3G), HRS PSAP:TSG101 UEV domain complex (3OBQ), STAM SH3 domain (1UJ0), and the STAM VHS domain (3LDZ). Panel b contains the syndecan peptide:syntenin PDZ2 domain complex (1YBO). Panel c contains HIV Gag Myr-MA (1UPH), CA (3H47), NC (1MFS), PTAP:TSG101 UEV domain complex (1M4Q), LYPX_nL:ALIX (2R05). Panel d contains the CEP55 coiled coil:ALIX proline-rich region (PRR) complex (3E1R). Abbreviation: CBB, clathrin-binding box.

Figure 4.

The structural basis for intercomplex interactions within the ESCRT pathway. The relevant interactions are as depicted in Figure 2, except that interacting domains from yeast Vps28p (ESCRT-I) and Vps36p (ESCRT-II) are shown in panel b because the divergent human ESCRT-I:ESCRT-II interaction site has not been structurally characterized and that interacting domains from yeast Vta1p and Vps4p are shown in panel d because the homologous human LIP5:VPS4 interaction has not been structurally characterized. Multiple ESCRT-III MIM:MIT domain interactions have been structurally characterized, and four distinct classes of interactions are shown [and see Figure 2 for the structure of the LIP5(MIT)₂:CHMP5 complex]. Relevant Protein Data Bank identity numbers are shown in parentheses. Abbreviations: MIT, microtubule interacting and transport; MIM, MIT interaction motif.

Figure 5.

Structurally characterized ubiquitin interactions within the ESCRT pathway. Ubiquitin (Ub)-binding domains from the HRS:STAM adaptor (ESCRT-0) (see Figure 3a) and from ESCRT-I and ESCRT-II (Figure 2) are shown in red, bound to ubiquitin in gray. The Ub I44 side chains are shown explicitly (stick) to emphasize that the very different ubiquitin-binding domains recognize the same Ub surface. Relevant Protein Data Bank identity numbers are provided within the figure.

ESCRT Pathway Recruitment

At least five different classes of adaptors recruit the ESCRT pathway to different sites of action (Figures 1 and 3):

1. HRS:STAM (ESCRT-0) and related adaptors bind the TSG101 subunit of ESCRT-I to help create MVB vesicles (3, 4).

2. Syntenins collaborate with syndecans to bind ALIX and create vesicles destined for extracellular release as exosomes (5, 6).

3. Arrestin domain-containing proteins, such as ARRDC1, link cargoes to ubiquitin (Ub) ligases and ESCRT machinery during the formation of shedding microvesicles (7–9).

4. Viral structural proteins like the HIV Gag protein employ late assembly domains to bind TSG101, ALIX, and/or NEDD4family Ub E3 ligases to promote virus budding from the plasma membrane (10, 11).

5. CEP55 recruits TSG101 and ALIX to intercellular bridges to facilitate abscission (12–14).

Early Acting ESCRT Factors

The early acting ESCRT complexes localize to bud necks where they help facilitate membrane curvature; integrate inputs from site-specific adaptors, membranes, and ubiquitylated protein cargoes; and ultimately recruit the downstream membrane fission machinery (15, 16). They are rich in ubiquitin-binding domains (UBDs), reflecting the prominent role of this posttranslational modification in ESCRT function (Figure 5). UBD-Ub interactions are generally weak and transient but can be enhanced by the concentrating effects of the two-dimensional membrane (17), by avidity and/or cooperativity, and by preferential binding to polymeric Lys63-linked Ub chains (18).

Bro1 proteins

Bro1 protein family members bind membrane-specific adaptors and recruit the downstream ESCRT-III subunits CHMP4 and CHMP5 (19, 20). Humans express at least five Bro1 domain-containing proteins of which ALIX (21) and HD-PTP (22) are the best characterized (see Supplemental Table 1). These two proteins appear to be paralogs, with HDPTP functioning primarily in MVB protein sorting (23) and ALIX functioning in MVB sorting (24), exosome biogenesis (5), virus budding (25), and abscission (12).

ALIX comprises Bro1, V, and proline-rich elements (Figure 2). Short amphipathic helices or β-hairpins from the ESCRT-III proteins bind the concave surface of the banana-shaped Bro1 domain (Figure 4a) (19, 20), whereas the basic convex surface may bind and induce (or sense) negative membrane curvature (26). The V domain has two extended helical “arms,” and the second arm binds both Lys63-linked polyubiquitin chains (27) and YPX_nL motifs found within viral and exosomal adaptors and MVB cargoes (Figure 3b,c) (5, 24, 25, 28, 29). The analogous region of HD-PTP binds the ESCRT-I subunit UBAP1, and this interaction helps define an endosome-specific Bro1-ESCRT-I pair (30).

The C-terminal proline-rich region (PRR) of ALIX folds back and autoinhibits ligand binding to the upstream domains (31, 32). The PRR contains binding epitopes for a number of other proteins, including CEP55 (12–14) and TSG101 (25, 33, 34), that presumably capture the “free” PRR conformation, thereby inducing (or sensing) ALIX activation. ALIX can also be phosphorylated, within both the Bro1 domain and the PRR, and both modifications likely play regulatory roles (35, 36). The fully assembled ALIX protein has been modeled as an antiparallel dimer with open, associated V domains. This observation implies that the Bro1 domains can bridge or nucleate two ESCRT-III filaments (37). ALIX also binds other important cellular components, including actin and endocytic, apoptotic, and calcium signaling machinery, although their ESCRT connections are not yet well understood (21, 38).

ESCRT-I

Like Bro1 proteins, ESCRT-I complexes link membrane-specific adaptors and MVB protein cargoes to downstream ESCRT machinery, and ALIX and ESCRT-I can often function interchangeably in supporting virus budding (11, 39). Humans express a complex array of heterotetrameric ESCRT-I complexes, each of which contains a single copy of the unique human TSG101 subunit and single copies of VPS28 (two variants, A and B), VPS37 (four isoforms, A-D), and MVB12 (at least three isoforms, MVB12A, MVB12B, and UBAP1) (30, 40, 41). Several ESCRT-I subunit isoforms have been shown to function in specific applications (30, 42, 43), but the full range of combinatorial complexity and functional diversity of mammalian ESCRT-I complexes remains to be explored.

The core of yeast ESCRT-I is composed of a “stalk” and a “headpiece” that together form a highly extended structure ∼18 nm long, with ligand-binding domains flexibly tethered at either end (Figure 2) (44–46). At one end, the N-terminal UEV domain and adjacent linker of TSG101 bind Ub (Figure 5c) (47, 48); the CEP55 adaptor (Figure 3e) (12–14); and P (T/S)AP peptide motifs within ALIX, the MVB HRS:STAM adaptor, the shedding vesicle ARRDC1 adaptor, and many viral protein adaptors (Figure 3a,c,d) (9, 33, 34, 49, 50). Terminal domains from some VPS37 and MVB12 subunits also have Ub-binding activities (51), including the solenoid UBA domain (SOUBA) of UBAP1, which can bind up to three Ub molecules simultaneously (Figure 2) (42). At the other end of the ESCRT-I core, MVB12A and -B subunits display MVB12-associated β-prism (MABP) domains, which bind membranes (Figure 2) (52). The C-terminal domain of VPS28 forms a four-helix bundle that binds ESCRT-II, albeit via structurally divergent interactions in yeast and mammals (Figure 4b) (44, 45, 53, 54).

ESCRT-II

ESCRT-II is a stable heterotetrameric complex that can bridge the ESCRT-I and ESCRT-III complexes (55). The complex contains single copies of EAP45 and EAP30, as well as two copies of EAP20. Both human and yeast ESCRT-II core complexes are shaped like the letter Y, with EAP20 subunits constituting the arms and EAP45 and EAP30 together constituting the base (Figure 2) (54, 56, 57). All three proteins contain tandem copies of the “winged-helix” fold, suggesting a common molecular ancestry. N-terminal elements from EAP45 and EAP30 are flexibly tethered to the base of the Y, and the EAP45 GLUE domain and adjacent linker bind ESCRT-I, phosphoinositides, and Ub (Figure 5d) (44, 53, 54, 58, 59). The two EAP20 subunits bind the ESCRT-III subunit, CHMP6 (Figure 4c) (53, 56, 60, 61); thus, ESCRT-II can nucleate the assembly of two ESCRT-III filaments (62). The yeast ESCRT-I:ESCRT-II supercomplex can adopt an ensemble of crescent shapes in solution and can stabilize the necks of vesicular membrane invaginations (15, 63).

ESCRT-II activity is essential for endosomal protein sorting and MVB vesicle formation in yeast (55), and appears to perform a similar role at the mammalian MVB (64). Indeed, overexpression of yeast ESCRT-II can bypass the ESCRT-I requirement, suggesting that ESCRT-II activation may be the most important ESCRT-I function during yeast MVB vesicle formation (55).

ESCRT-III Proteins

Humans express eight different subfamilies of ESCRT-III proteins. The conserved ESCRT-III core structure contains a long helical hairpin and two shorter helices that pack against the open end of the hairpin (Figure 2) (65– 67). In their soluble monomeric states, the C-terminal tails of ESCRT-III subunits fold back against the core and autoinhibit protein assembly (66, 68–70). The proteins then “open” when recruited to sites of membrane fission by the early acting factors ALIX and ESCRT-II, thereby freeing the core domains to bind membranes and polymerize. ESCRT-III polymerization also exposes arrays of C-terminal microtubule interacting and transport interaction motif (MIM) elements within their C-terminal tails that can bind microtubule interacting and transport (MIT)-domain containing proteins, including VPS4 (71–73), its activator LIP5 (74– 76), the microtubule severing enzyme spastin (77), and the deubiquitylating enzymes UBPY and AMSH (Figures 2 and 4e) (78, 79).

As discussed below, ESCRT-III proteins coassemble into membrane-bound filaments that play critical roles in membrane fission (80, 81). Despite their structural similarities, different ESCRT-III subunits clearly perform distinct roles, both in membrane fission and in cofactor recruitment. Studies of yeast MVB vesicle formation have revealed that the “core” ESCRT-III proteins are recruited sequentially in the order: CHMP6 (Vps20p), CHMP4 (Snf7p), CHMP3 (Vps24p), and CHMP2 (Did4p), followed by VPS4 (82–85). CHMP4 and CHMP2 subunits appear to constitute particularly important building blocks of ESCRT-III filaments because (a) they are required for all known ESCRT-dependent membrane fission processes (82, 85–87); (b) when overexpressed, both CHMP2 and CHMP4 subunits can form membrane-enclosed helical tubules that extrude from the plasma membrane (88, 89); and (c) CHMP4 subunits are present at higher stoichiometry than the other ESCRT-III proteins (84).

VPS4 ATPase Complexes

VPS4 ATPases power ESCRT-mediated membrane remodeling at the end of the catalytic cycle by using the energy of ATP hydrolysis to disassemble ESCRT-III filaments into their constituent subunits (Figure 2) (90, 91). There are also indications that VPS4 remodeling contributes directly to membrane fission during virus budding and cytokinesis (92–94), although perhaps not during MVB vesicle formation (85). Humans express two closely related VPS4 proteins (VPS4A and -4B) that can function redundantly in at least some contexts (72).

VPS4 contains an N-terminal MIT domain that binds ESCRT-III substrates, a mobile linker, a central ATPase cassette composed of large and small domains that resembles other AAA ATPases, a three-stranded antiparallel β-sheet inserted within the small ATPase domain, and a terminal helix that connects the two ATPase domains (Figure 2) (95, 96). The enzyme cycles through two states: a catalytically inactive, disassembled state and a catalytically active, higher-order assembly (97). Catalytically inactive Saccharomyces cerevisiae Vps4p can form double-ring structures in vitro (reviewed in Reference 90), but more recent analyses indicate that the active enzyme may be a hexamer (Figure 2) (N. Monroe, personal communication).

VPS4 assembly represents an important regulatory node within the ESCRT pathway. Enzyme recruitment, assembly, and activation are promoted by the high concentrations of MIM elements displayed by ESCRT-III filaments (71–73, 98, 99). In addition, two ESCRT-III-containing complexes, LIP5:CHMP5 and IST1:CHMP1B, appear to play particularly important roles in promoting VPS4 localization and activation (Figures 2 and 4d). IST1 and CHMP1B can copolymerize into helical tubes in vitro (J. McCullough & A. Frost, unpublished), and they play particularly important roles in helping to recruit VPS4 and other cofactors during abscission (66, 100–103).

LIP5 and CHMP5 form a soluble cytosolic complex that promotes VPS4 oligomerization and helps link the enzyme to the ESCRT-III lattice (74, 95, 98, 104). The first MIT module within the tandem MIT domain of LIP5 binds the tails of different ESCRT-III proteins, with a preference for CHMP1B, whereas the second MIT module binds tightly to CHMP5 (Figure 2) (74–76). The dimeric C-terminal VSL domain of LIP5 binds the C-terminal domains of VPS4 (Figure 4d) and stimulates VPS4 hexamerization and ATPase activity in vitro (95, 105, 106). This interaction may also link multiple VPS4 hexamers into an extended hexagonal array (106).

VPS4 enzymes apparently disassemble ESCRT-III filaments by first making MIT:MIM interactions with subunit tails (Figure 4e), then making a second contact near the center of the hexameric ring, and ultimately translocating the subunits up into (or through) the ring, thereby disrupting ESCRT-III lattice interactions (74, 90, 95, 99).

ESCRT:Membrane Interactions

In conjunction with the essential functions performed by ESCRT proteins, the lipid membrane also actively participates in ESCRT-mediated, “reverse-topology” membrane fission reactions by (a) targeting ESCRT factors to specific sites of action, (b) providing a two-dimensional surface that concentrates and organizes the ESCRT machinery and restricts MVB cargo diffusion, and (c) helping facilitate the fission reaction (17, 107). In the first case, different phosphoinositides direct the pathway to different cellular membranes. For example, the endosome-specific phospholipid PI(3)P targets the ESCRT pathway to endosomes for MVB vesicle formation through interactions with the HRS FYVE domain (108) and probably also with the GLUE domain of EAP45 (ESCRT-II) (59). By contrast, HIV-1 targets the ESCRT pathway to the cell periphery because the viral Gag adaptor binds specifically to PI(4,5)P₂ in the plasma membrane (109). PI(4,5)P₂ and PI(3)P also localize to the cleavage furrow (110), but their roles in targeting ESCRT factors to intercellular bridges remain to be determined. More generally, the membrane surface serves as an important organizational platform because all core ESCRT complexes except VPS4 also contain exposed basic/hydrophobic motifs that bind acidic lipids and help orient the complexes on the membrane surface. The topology of the membrane platform also appears to be important, as the dramatic membrane curvature generated prior to fission apparently helps recruit, retain, and organize ESCRT core complexes at the bud neck (15, 63, 111).

In addition to these targeting and organizational functions, specialized lipids such as cholesterol perform important functions in the membrane remodeling and fission reactions. Bilayers can segregate their lipids laterally to form cholesterol-rich microdomains called lipid rafts (112), and both MVB vesicles and HIV virions are cholesterol rich (16, 17, 113). Cholesterol is required for ESCRT-II self-assembly on supported lipid bilayers, and the resulting assemblies induce phase separation into liquid-ordered domains (114). This is potentially important because phase separation between liquid-ordered and -disordered regions can create line tension that promotes membrane scission (17). Two other specialized lipids have also been linked to ESCRT-related vesiculation processes. The phospholipid 2,2′-lysobisphosphatidic acid (LBPA) is enriched at late endosomes, where it can associate with ALIX (115). LBPA can drive spontaneous membrane deformation and intraluminal vesicle formation in vitro (115), apparently by adopting a cone shape under acidic conditions (116). The sphingolipid ceramide also adopts a cone shape and has been implicated in lipid microdomain assembly and exosome formation (117–119).

Biological Functions

Multivesicular Body Vesicle Formation

The ESCRT pathway plays a central role in targeting membrane proteins to the lysosome for degradation. Such proteins are first sorted into regions of the endosomal membrane that bud into the lumen (Figure 6a,b) and can then be degraded when the MVB fuses with the lysosome (Figure 1) (16, 120). Sorting into MVB vesicles requires (a) cargo clustering, (b) membrane invagination, (c) cargo transfer into nascent vesicles, and (d) vesicle neck fission, and ESCRT factors facilitate all of these steps (15, 81). MVB cargos are initially clustered on endosomal membranes by HRS:STAM (ESCRT-0) and related adaptors (e.g., GGA3 and TOM1L1). ESCRT-I:ESCRT-II supercomplexes promote membrane curvature and organize ESCRT assemblies at the vesicle neck, and late-acting ESCRT-III and VPS4 complexes catalyze membrane fission and factor recycling.

Figure 6.

(a-c) ESCRT pathway functions in multivesicular body (MVB) vesicle formation, (d-e) enveloped virus budding, and (f-h) the abscission stage of cytokinesis. (a) Slice from an electron microscopy (EM) tomogram showing a vesicle budding into an endosome (large arrow) at a site adjacent to the HRS:STAM:clathrin coat (arrowheads). The small arrow shows a gold particle used to label the endosome. (b) EM tomographic reconstruction of an MVB, pseudocolored to show the internal vesicles and limiting membrane (green). Panels a and b were reprinted with permission from Reference 193, copyright 2003, National Academy of Sciences, U.S.A. (c) Fluorescence micrographs showing that Snf7p (CHMP4, green) concentrates at the neck of an MVB-like vesicle budding into a giant unilamellar vesicle in a reconstituted system. Ubiquitin cargos (blue) and membranes (red) are shown to define the vesicle. Reprinted with permission from Reference 15. (d) Cryoelectron microscopy (cryo-EM) image of a budding HIV-1 viral particle. The virion is ∼100 nm in diameter, the Gag protein lattice is visible as a protein-dense layer inside the plasma membrane, and the open neck of the virion is the site where ESCRT-mediated membrane fission occurs. Reprinted with permission from Reference 109, copyright 2012, Cold Spring Harbor Laboratory Press. (e) Graph showing the time course of CHMP4B (ESCRT-III) recruitment (green) as HIV Gag molecules (red) assemble into a single budding virion on a HeLa cell plasma membrane. Note that Gag assembles gradually (over ∼7 min), whereas CHMP4B appears in a sharp burst immediately prior to virion release. Reprinted with permission from Reference 92. (f) Structured illumination microscopy fluorescence image of an intercellular bridge prior to abscission, showing TSG101 (orange) forming two rings on either side of the midbody (with an alternate view inset) and microtubules (white). Reprinted with permission from Reference 94 (g) Cryo-EM tomographic image of an intercellular bridge showing the midbody, microtubules, and 17-nm filaments within constriction zones that undergo microtubule severing and abscission. Panels g and h were modified and reprinted with permission from Reference 171. (h) Pseudocolored EM reconstructions showing an intercellular bridge late in cytokinesis. Microtubules are shown in red with balls denoting their ends, and 17-nm filaments are shown in green shades. Abscission occurs at the narrow, microtubule-free constriction zone to the right of the filaments.

The heterodimeric HRS:STAM adaptor is shown schematically in Figure 3a (121). Important motifs include an HRSPSAP motif that helps recruit TSG101(ESCRT-I) (50), an HRS clathrin-binding box that connects the adaptor to a flat clathrin coat (122, 123), a STAM SH3 domain that recruits deubiquitinating enzymes (124), and four different UBDs that help concentrate ubiquitylated cargoes (Figures 3a and 5a,b) (125). Ub is the major signal for cargo sorting, and Lys63-linked polyubiquitin chains serve as the sorting signals for at least some cargoes, such as the epidermal growth factor receptor (18, 126).

NEDD4 proteins (Rsp5p in yeast) are the most important class of ESCRT-associated Ub ligases. NEDD4 ligases can either bind directly to PPXY motifs within cargoes (127) or can be recruited by arrestin-related adaptor proteins (Figure 3c) (7). Ubiquitylation is a highly dynamic process, and Ub is ultimately removed from cargoes prior to MVB vesicle incorporation by the deubiquitylating ESCRT enzymes, UBPY or AMSH, which interact with both the HRS:STAM adaptor and ESCRT-III subunits (124, 128). Cargoes can also interact directly with early acting ESCRT factors, as when the G protein–coupled receptor PAR1 uses a YPX_nL motif to bind ALIX (24).

Intraluminal vesicles appear to form just beyond the edge of the HRS:STAM:clathrin coat (Figure 6a), which begs the question of how cargo proteins are transferred from the adapters into the nascent vesicles (15). One possibility is that the crescent-shaped ESCRT-I:ESCRT-II supercomplexes open and close to extract cargoes from the HRS:STAM:clathrin adaptor and transfer them into the nascent vesicles (63). Coassembly of ESCRT factors at the nascent bud neck would then provide a barrier against cargo diffusion from the vesicle prior to fission (Figure 6c).

Intriguing future issues surrounding MVB biology include (a) the relative importance of endosomal membranes and associated ESCRT protein networks as platforms for signaling (129) and small interfering RNA (siRNA) generation (130), and (b) the mechanistic basis for the remarkable observation that direct contacts between multivesicular endosomes and the endoplasmic reticulum stimulate ESCRT-mediated sorting of epidermal growth factor receptors into MVB vesicles (131).

Exosomes and Shedding Microvesicles

Cells can release different types of extracellular microvesicles, including two that have ESCRT connections: exosomes and shedding microvesicles (ectosomes). Exosomes likely constitute a major pathway for intercellular communication, with reported functions in antigen presentation and T-cell activation; immune tolerance; immune evasion; maternal immunosuppression and transplant tolerance; intercellular transport of proteins, DNA, and RNA; tissue repair; and neural communication (including roles in the spread of prion/plaque diseases) (reviewed in References 132 and 133). However, although exosome production and physiological responses have been demonstrated in tissue culture systems, exosomes are challenging to purify and assay, and clear examples of in vivo functions are rare.

ESCRT pathway involvement is expected because exosomes originate as MVB vesicles, and proteomic analyses of purified exosomes have identified subunits from all core ESCRT complexes (134, 135). Exosome production from cells requires the dimeric cytoplasmic protein syntenin, which acts as an adaptor that recruits ALIX to sites of intraluminal vesicle formation via interactions between the ALIX V domain and three syntenin LYPX_nL motifs (Figures 1 and 3b) (5). Syntenin binds the cytoplasmic tails of the transmembrane protein syndecan (136). Syndecan clustering stimulates exosome production and the process also requires ALIX, EAP30 (ESCRT-II), CHMP2A, CHMP4A/B/C, and VPS4 proteins, implying ESCRT dependence. MVB trafficking away from the lysosome and to the plasma membrane for fusion remains to be understood in detail but is reportedly regulated by Rab27A/B GTPases (132).

Shedding microvesicle formation is less well characterized but also appears to involve the ESCRT pathway. For example, ESCRT-I is required for early stages of shedding microvesicle formation in Caenorhabditis elegans (137) and the human arrestin domain-containing protein ARRDC1 uses a PSAP motif to recruit TSG101 (ESCRT-I) for shedding microvesicle release from the plasma membrane (Figure 3c) (9).

In summary, exosomes and shedding microvesicles have an exciting research future, particularly if they prove to be major pathways for intercellular communication. Future goals include developing robust functional assays, learning how specific proteins and RNA cargoes are packaged, determining how MVBs traffic and fuse with the plasma membrane, and understanding how target cells receive signals from microvesicles.

Enveloped Virus Budding

Viruses rely upon host cellular machinery for many functions, and many enveloped viruses usurp the ESCRT pathway to bud from cells. ESCRT-mediated virus release is best characterized for retroviruses, particularly HIV-1 (11, 39), but it is also well documented for arena-, filo-, paramyxo-, orbi-, and rhabdoviruses. There is also a recent report that Epstein-Barr virus, a herpesvirus, uses ESCRT machinery for nuclear egress (138; and see References 139 and 140). Many other classes of enveloped viruses have additional intriguing ESCRT links, although some encode their own membrane fission machinery, e.g., influenza A, (141), whereas others form external protein coats that apparently alleviate the need for ESCRT-catalyzed membrane fission, e.g., alphaviruses (142).

ESCRT-dependent viruses employ late assembly domains, often in multiple copies, to recruit early acting factors of the ESCRT pathway. The three best-characterized late assembly domains are P(T/S)AP (e.g., as in HIV-1 Gag), which binds directly to the UEV domain of TSG101(ESCRT-I) (Figure 3d); YPX_nL [e.g., equine infectious anemia virus (EIAV) Gag], which binds the V domain of ALIX; and PPXY [e.g., Rous sarcoma virus (RSV) Gag], which binds the WW domains of NEDD4 family Ub ligases (11, 39). All three of these viral late domains mimic interactions used during cellular MVB vesicle and shedding microvesicle formation (see above). Viruses can also connect to the ESCRT pathway in additional ways, and elucidating these connections will likely reveal new mechanisms by which cellular factors can link into the pathway (143–147). Ub contributes to the budding of many retroviruses, including HIV-1 where Lys63-linked chains have been implicated. Gag ubiquitylation likely contributes to ALIX and ESCRT-I recruitment in some cases, but ubiquitylation of host ESCRT machinery may also be important because budding of the prototypic foamy virus remains Ub dependent, even when the Gag protein is mutated so that it lacks any lysines (148–151).

Viruses apparently require only a subset of the downstream ESCRT machinery for efficient budding (39). This is presumably because (a) viral structural proteins, unlike MVB cargoes, are capable of targeting to specific membranes, self-assembling, and generating membrane curvature (Figure 6d), and because (b) viruses do not require all of the biochemical activities that are needed during abscission, such as microtubule severing.

The ESCRT protein network is least well understood in the PPXY-dependent viruses, but the arrestin-related trafficking proteins appear to participate (8), which is consistent with known connections between yeast arrestins and Rsp5p (the sole yeast NEDD4 homolog) (7). There is also a report that ESCRT-II is required for RSV budding (152).

Viruses like EIAV that bud via YPX_nL motifs use ALIX to recruit CHMP4 family members, particularly CHMP4B (19, 37, 86, 153, 154). ALIX builds up steadily throughout the entire EIAV assembly process (92), whereas ESCRT-III and VPS4 levels spike immediately prior to the membrane fission event (Figure 6e) (92, 93). This implies that physical mechanisms (such as threshold effects), generation of proper membrane curvature, or Ub transfer must induce ESCRT-III recruitment late in the process (92). ALIX functions are augmented by V domain Ub binding (27) and by additional unidentified Bro1 domain interactions (154–156).

HIV-1 budding requires CHMP4, CHMP2, and VPS4 (72, 86). These proteins can interact directly in vitro and apparently represent the minimal set of required late-acting ESCRT proteins, although CHMP3 and CHMP1 proteins can also contribute to budding efficiency (86, 157, 158). The protein-protein interactions that lead to ESCRT-I-mediated recruitment of ESCRT-III subunits are not yet fully defined, however, because several groups have reported that siRNA depletion of ESCRT-II and CHMP6 subunits does not significantly reduce virus budding efficiency (86, 159, 160). This result implies either that the ESCRT-I:ESCRT-II:CHMP6:CHMP4 relay system, which is used in MVB cargo sorting (1, 15), may be nonessential (or redundant) in virus budding or that siRNA-treated cells retain sufficient quantities of ESCRT-II and CHMP6 subunits to support budding. Conceivably, ESCRT-II functions may be most critical when bud neck deformation and ESCRT-III recruitment are difficult, as in MVB vesicle formation, but less important during virus budding and abscission when viral structural proteins or the contractile ring helps deform the membrane and stabilize the narrow membrane tubules. Alternatively, viral adaptors may recruit additional ESCRT-associated factors, such as ALIX and NEDD4 family Ub ligases, that provide additional links to ESCRT-III and reduce the ESCRT-II requirement. This issue requires further study, however, because ESCRT-II does play an essential role in recruiting ESCRT-III subunits in a reconstituted HIV-1 assembly system (158) and the complex also reportedly functions in HIV-1 RNA trafficking (161).

The position of the membrane fission site relative to the ESCRT assemblies is of considerable interest. If fission occurs on the virion-proximal side of the ESCRT-III assembly (i.e., as in traditional models for MVB vesicle formation), then there must be a mechanism for disrupting Gag:ALIX or ALIX:ESCRT-III linkages during budding. If fission occurs on the virion-distal side of the ESCRT-III assembly (i.e., as in abscission), then ESCRT factors should end up within the virion. Thus far, only ALIX has been shown to accumulate to appreciable levels within HIV-1 and EIAV virions (25, 33, 92), whereas the ESCRT-III and VPS4 proteins appear to be recycled back into the cytoplasm following membrane fission (92). However, the fate of ESCRT subunits during virus budding merits further study. There is also much to be learned about how (a) ESCRT-dependent budding functions can be negatively regulated or even bypassed in some contexts (162, 163); (b) the roles of candidate accessory factors such as actin, ALG-2 (164), sprouty 2 (165), and CC2D1 proteins (166, 167); and (c) the contribution of ESCRT pathway regulation to antiviral innate immunity (168).

The Abscission Stage of Cytokinesis

The process of cytokinesis separates daughter cells and completes cell division. Early in mammalian cell cytokinesis, the actomyosin contractile ring constricts the cleavage furrow into an intercellular bridge that contains antiparallel microtubule bundles and a midbody (Figure 6f–b). The intercellular bridge is subsequently resolved through the complex membrane fission process of abscission, which can occur on either (or both) side (s) of the midbody (reviewed in Reference 169). ALIX, ESCRT-I, both VPS4 isoforms, and nearly all of the different ESCRT-III proteins participate in abscission, likely catalyzing the membrane fission step (12, 13, 87, 94, 100, 101, 169–171). Cytokinesis appears to have been the original ancestral function of the ESCRT pathway because some crenarchaeal genera like Sulfolobus that diverged from eukaryotes several billion years ago use a minimal set of ESCRT factors for cell division but lack endosomes (172–174). Furthermore, although the ESCRT pathway is not essential for cell division in S. cerevisiae, ESCRT depletions synthetically enhance other cytokinesis defects (175).

The early acting ESCRT-I and ALIX proteins form large membrane-associated rings on either side of the mammalian midbody (Figure 6f). Both are recruited through direct interactions with the central hinge region of CEP55, which forms an unusual coiled coil that binds proline-rich motifs in ALIX and TSG101(Figure 3e) (12, 13, 77). ESCRT-III proteins then form more distal rings that appear to polymerize outward, apparently forming the large (∼17-nm) spiraling filaments visualized within intercellular bridges (Figure 6f–h) (94, 170, 171). These 17-nm filaments require CHMP2A to form and likely contain multiple ESCRT-III subunits, but the spatial organization of different subunits within the filaments remains to be determined. ESCRT-III factors, particularly IST1, then recruit the VPS4 enzymes prior to membrane fission (94, 100, 101). Membrane fission occurs ∼1 μm away from the midbody at constriction zones where the microtubules are cut (Figure 6g,h) (94, 171). ESCRT-III assemblies localize to both midbodies and cut sites but appear to be discontinuous (170).

ESCRT proteins also recruit a subset of the many other factors that localize to intercellular bridges (176). The complexity of abscission appears to explain why mammals have evolved so many different ESCRT-III subunits, and all proteins that contain MIT domains must be considered candidates for ESCRT-regulated roles in cytokinesis. For example, the ESCRT-III protein CHMP1B binds the MIT domains of the AAA ATPase spastin (Figure 4e), which severs microtubules immediately prior to abscission (77). Additional MIT domain-containing proteins recruited by ESCRT-III subunits to function in abscission include the phospholipase D–like protein, MITD1 (CHMP1A, CHMP1B, CHMP2A, and IST1) (177, 178), spartin (SPG20) (IST1) (179), and the protease calpain 7 (IST1:CHMP1B complexes) (180).

Multiple signaling and cell cycling pathways presumably help coordinate ESCRT activities during abscission, and their roles are now emerging. For example, Wnt5a signaling was recently shown to position ESCRT-III at functional sites within intercellular bridges (181). The role(s) of Ub in cytokinesis is (are) not yet well understood, but Ub concentrates at midbodies prior to abscission, and the Ub ligase BRUCE as well as the deubiquitylating enzymes UBPY and AMSH are required for abscission (182, 183). ESCRT-III proteins also help regulate abscission timing via the Aurora B-mediated abscission checkpoint (NoCut) (184). CHMP4C phosphorylation within a unique region that is absent in CHMP4A and CHMP4B promotes binding to the Borealin subunit of the chromosomal passenger complex, which activates the abscission checkpoint and thereby delays abscission until any lagging chromosomes have cleared the intercellular bridge. Thus, this system links proper chromosome segregation to ESCRT-III abscission activity via the phosphorylation state of CHMP4C.

In summary, the mammalian ESCRT pathway catalyzes a series of critical membrane fission reactions and is integrated into many important cellular processes resulting in rich biology and physiology: See the sidebar ESCRTs in Cell Biology, Development, and Disease and Table 1.

ESCRTs in Cell Biology, Development, and Disease.

The ESCRT pathway impacts developmental biology and physiology in important ways, reflecting the central roles of cellular ESCRT functions. Mammalian ESCRT factor knockouts are typically embryonically lethal (189), but partial loss of function mutations have been linked to a variety of different pathologies, including neurodegeneration and cancer, whose underlying mechanisms are often unclear (reviewed in References 130 and 189–192). ESCRT factors have also been implicated in other less well-characterized cellular functions, including autophagy, cell cycle regulation, RNA localization, and cell polarity and migration. In some cases, these functions do not obviously involve membranes and may instead reflect ways in which the ESCRT pathway is integrated with other cellular pathways. Table 1 highlights Such “noncanonical” ESCRT pathway functions and important ESCRT connections to developmental biology and disease.

Table 1. ESCRTs in cell biology, development, and disease.

Biological phenomenon	ESCRT connection	Key references
Autophagy
Protein aggregate clearance	CHMP2B	194
Autophagosome accumulation	CHMP2B	195
Endosomal microautophagy	TSG101, VPS4	196
Cell cycle regulation
Abscission checkpoint/chromosomal passenger complex	CHMP4C	184
Chromatin structure	CHMP1A	197
Centrosome/spindle maintenance	TSG101	198
	Arabidopsis Tsg101	199
	Schizosaccharomyces pombe, Vps22	200
	ESCRT-III, VPS4	87
	Drosophila CHMP5	201
RNA localization
Maternal RNA localization	Drosophila ESCRT-II	202
siRNA spreading	Drosophila Vps25	203
	Drosophila Hrs	203
	HRS, TSG101	204
	Drosophila Hrs, VPS36, Alix	205
Cell polarity and migration
Apicobasal polarity, cell migration	ESCRT pathway	206
Cancer
Tumorigenesis and metastatic potential	HRS	207
Benign brain tumors	HRS	208, 209
Ovarian cancer	TSG101	210, 211
Mammary cancer	TSG101	212
Gastrointestinal stromal tumors	TSG101	213
Papillary thyroid cancer	TSG101	214
Metastatic tumors	TSG101	215
Neoplastic transformation	Drosophila Tsg101	203, 216
Hepatocellular carcinoma	VPS37A	217
Ovarian cancer	VPS37A	218
Neoplastic transformation	Drosophila Vps25	219
Prostate cancer	CHMP3	220
Non-small cell lung cancer	CHMP3	221
Ductal pancreatic cancer	CHMP1A	222
Renal cell carcinoma	CHMP1A	223
Hyperactive epidermal growth factor receptor signaling	Drosophila Hrs	224
	TSG101	225
	CHMP6, VPS4A	226
Neurodegeneration/other diseases
Frontotemporal dementia	UBAP1	227
	CHMP2B	228
Amyotrophic lateral sclerosis	CHMP2B	229
Lewy body disease	CHMP2B	230, 231
Spongiform neurodegeneration	TSG101	232
Charcot-Marie-Tooth disease	TSG101	233
	HRS:STAM	234
Hereditary spastic paraplegia	CHMP1B	235
Autosomal dominant cataracts	CHMP4B	236

Models for Membrane Constriction and Fission

It is of fundamental importance to understand how the ESCRT machinery draws opposing membranes together and mediates fission. ESCRT-III proteins perform central roles in these processes, apparently by forming filaments within the necks of budding vesicles, viruses, and intercellular bridges (16, 80, 81, 169). Many ESCRT-III proteins can form filaments in vitro (37, 66, 91, 111, 185, 186) and in cells (88, 89, 157, 171), and these filaments often wrap up into helical tubes or spirals. The role of VPS4 in fission is less clear because the enzyme is not required for single rounds of vesicle formation in a reconstituted yeast MVB system (85), but appears immediately prior to virus budding (Figure 6e) (92, 93) and abscission (94), where it seems to be required for fission (92, 93).

ESCRT-III filaments likely function in pairs (or contain paired strands) because both known ESCRT-III-recruiting complexes, ESCRT-II and ALIX, contain two ESCRT-III-binding sites (37, 54, 56, 57, 62) and because yeast Snf7p (CHMP4) can form paired helical filaments on supported lipid monolayers (187). One recurring assembly is a 3–4-nm wide filament, which has been observed in several ESCRT-III assemblies in vitro, including helical CHMP2A:CHMP3 copolymers (157), CHMP4B and Snf7p filaments (37, 187), and lipidated IST1 tubes (J. McCullough & A. Frost unpublished findings), as well as in cellular CHMP2B and CHMP4A assemblies (88, 89). These filaments can further associate into single, paired, or multistart helical tubes with diameters of 40–400 nm. However, distinct ESCRT-III filaments and tubes of differing widths and morphologies have been reported in other cases (111, 171, 185, 186), and a clear consensus structure has therefore yet to emerge. In particular, the 17-nm ESCRT-III filaments that spiral toward the cut sites within the intercellular bridge (171) must either contain multiple copies of smaller ESCRT-III filaments or be different structures entirely.

Another key issue is how ESCRT-III filaments can promote membrane constriction and fission. As illustrated in Figure 7a, a leading model holds that ESCRT-III filaments spiral inward and promote membrane constriction by pulling the opposing membranes together over the resulting “domes” (80, 188). Spiraling would require ESCRT-III subunits to make nonidentical contacts between consecutive coils, which is unusual behavior because protein:protein interfaces typically do not accommodate the required continuum of different contacts. Nevertheless, ESCRT-III tubes can taper into terminal dome-like structures in vitro, demonstrating that such constricting spirals are physically possible (66, 91). Furthermore, cryoelectron microscopy tomograms of the ESCRT coats that mediate cleavage furrow ingression and cell division in Sulfolobus suggest an “hourglass” model in which two opposing domes of spiraling ESCRT-III filaments create a symmetric cleavage furrow, pulling the membranes on either side of the furrow down into a single fission point (G. Jensen, personal communication). Spiraling dome models are attractive because they nicely explain how ESCRT-III assemblies can mediate membrane constriction and fission across very different size scales, ranging from the necks of budding MVB vesicles (∼25 nm) and retroviruses (∼50 nm) to mammalian intercellular bridges and crenarchaeal cells (each ∼1 μm). In all cases, ESCRT-III assemblies could initially form rings that match the starting diameter and then spiral closed toward the fission point. In principle, VPS4 could help promote ESCRT-III filament constriction (188) or remodel the dome to promote the hemifission to fission transition (93).

Figure 7.

Models for ESCRT-mediated membrane fission reactions. (a) The “dome” model for multivesicular body (MVB) vesicle budding (188), (b) the “break and slide” model for intercellular bridge constriction and abscission (94, 170), and (c) the “whorl” model for MVB vesicle budding (63). ESCRT-I:ESCRT-II supercomplexes are shown in red; ESCRT-III filaments are shown in shades of green and yellow; VPS4 enzyme complexes are shown in purple; and red arrows denote motion. Panels in the left column show ESCRT factors assembling within membrane tubules (brown). Panels in the central column show models for membrane constriction. Panels in the right column show models for the membrane fission step (see text for details).

A variation on the dome model (Figure 7b) was proposed to explain the observation that ESCRT-III proteins form discontinuous structures within intercellular bridges, first appearing as wide rings that are proximal to the midbody and later concentrating at the distal abscission sites (94, 170). This “break and slide” model holds that (a) ESCRT-III filaments nucleate near the midbody and initially begin to polymerize as wide spirals; (b) these wide spirals are unstable owing to their expanded diameters and eventually break, apparently assisted by VPS4 activity; (c) the unanchored spirals then slide along the membrane and constrict into lower-energy helices that narrow the intercellular bridge to the point where a fission dome can form. This model nicely explains the observation that ESCRT-III components concentrate at two different sites, and the model is supported by computational analyses, but ESCRT-III filament sliding has yet to be visualized directly.

Finally, Hurley and colleagues have recently proposed another significant variation on the dome model, termed the “whorl” model (Figure 7c). They propose that the necks of MVB vesicles are ringed with 6–10 copies of the ESCRT-I:ESCRT-II supercomplex that align with their long axes parallel to the neck and that each nucleate two ESCRT-III filaments (63). The filaments then grow toward the center of the neck like spokes, forming a multistart whorl that constricts the membrane along one face. VPS4 assemblies could then help organize the central site, where the growing filaments ultimately meet. This model is consistent with a large body of biophysical and structural data on MVB vesicle formation (63) but is more difficult to envision in the cases of mammalian abscission and crenarchaeal cytokinesis because the whorling ESCRT-III filaments would need to extend ∼500 nm to meet at the center of the midbody. Hence, a hybrid model seems more attractive in those cases.

In summary, ESCRT-III filaments play central roles in membrane fission reactions, and it is critical to define (a) their molecular structure(s); (b) how the different ESCRT-III subunits copolymerize and why so many different subunits are required, particularly during cytokinesis; (c) the mechanism by which ESCRT-III filaments constrict membranes; and (d) whether VPS4 enzymes play an active role in this process or instead function primarily to disassemble the ESCRT-III filaments following fission.

Glossary

ESCRT

endosomal sorting complexes required for transport

Intercellular bridge

a thin membrane tubule ∼1 μm in diameter that connects daughter cells during cytokinesis

Multivesicular body (MVB)

a late endosome filled with small intraluminal vesicles

Phosphoinositides

phospholipids that help define different cellular membranes, including endosomes [PI(3)P] and plasma membranes [PI(4,5)P₂]

Exosomes

small extracellular vesicles (50–100 nm in diameter) that are released when multivesicular bodies fuse with the plasma membrane

Shedding microvesicles

small extracellular vesicles (50–80 nm in, diameter) that bud directly from the plasma embrane

Late assembly domains

short viral protein motifs that recruit ESCRT factors, including YPXnL:ALIX, P (T/S)AP:TSG101 (ESCRT-I), and PPXY:NEDD4 proteins

NEDD4 proteins

a family of nine human ubiquitin E3 ligases, many of which are associated with the ESCRT pathway

Ubiquitin-binding domains (UBDs)

these domains bind ubiquitin and are common in adaptors and early acting ESCRT factors

ESCRT-III proteins

a homologous family of 12 small filament-forming human proteins named CHMP (charged multivesicular body protein) or IST1

Microtubule interacting and transport interaction motifs (MIMs)

motifs within the C-terminal tails of ESCRT-III proteins that are bound by MIT domains

Microtubule interacting and transport (MIT)

domains found in more than a dozen human proteins that bind ESCRT-III filaments

AAA ATPases

ATPases associated with diverse cellular activities that form ring-like assemblies that remodel cellular macromolecules

Midbody

a protein-dense structure in the center of the intercellular bridge