The Escrt Complexes: Structure and Mechanism of a Membrane-Trafficking Network

Abstract

The ESCRT complexes and associated proteins comprise a major pathway for the lysosomal degradation of transmembrane proteins, and are critical for receptor downregulation, budding of the HIV virus, and other normal and pathological cell processes. The ESCRT system is conserved from yeast to man. The ESCRT complexes form a network that recruits monoubiquitinated proteins and drives their internalization into lumenal vesicles within a type of endosome known as a multivesicular body. The structures and interactions of many of the components have been determined over the past three years, revealing mechanisms for membrane and cargo recruitment and for complex assembly.

Mini-glossary

Ubiquitin: A highly conserved 76 amino-acid protein that can be covalently attached to itself or other proteins through an isopeptide bond with its terminal carboxylate.

Endosome: An intracellular vesicle involved in transfer of proteins and lipids between different organelles.

Multivesicular body: An endosome containing internal vesicles.

Class E VPS: Phenotype manifesting alterations in mutlivesicular bodies.

Proteasome: Large multiprotein complex that proteolyses polyubiquitinated proteins, typically soluble proteins or membrane proteins that are substrates of ER associated degradation.

Summary of main points:

1. The yeast class E VPS genes and their human homologs code for a network of proteins that comprise the ESCRT complexes and associated proteins

2. The Vps27/Hse1, ESCRT-I, ESCRT-II, and ESCRT-III complexes are recruited to endosomal membranes in an ordered manner. Recruitment is initiated by the lipid PI(3)P and the presence of monoubiquitinated transmembrane proteins.

3. The role of the ESCRT network is to 1) recruit and cluster monoubiquitinated cargo, and 2) drive the formation, invagination, and fission of cargo-containing vesicles into the lumen of the multivesicular body.

4. The structural basis for PI(3)P recruitment via FYVE domains and ubiquitin recruitment via UIM, UEV, and NZF domains has been determined.

5. Vps27/Hse1, ESCRT-I and ESCRT-II are soluble protein complexes. The structure of the ESCRT-II core complex has been determined.

6. The ESCRT-III complex is thought to be a tightly membrane-associated array. In concert with the ATPase Vps4, ESCRT-III may be involved in mechanical deformation of the membrane to drive the invagination and fission of intralumenal vesicles into MVBs.

7. The ESCRT network is essential for budding of the HIV virus and structural studies of these complexes present potential targets for development of new anti-HIV therapeutics.

8. The mechanism of invagination is essentially unknown and presents a major challenge for mechanistic studies.

INTRODUCTION

Targeted degradation is a fundamental mechanism of protein regulation and quality control (39, 44, 80). Soluble proteins are marked for degradation by the proteasome by their modification with polymeric chains of ubiquitin, a highly conserved 76-amino acid protein. These polyubiquitin chains are added by a series of enzymes known as E1, E2, and E3. Polyubiquitin chains target proteins so modified to the regulatory subunit of the proteasome, and thence to degradation by the catalytic subunit. This entire process is tightly regulated and highly specific, as errors can lead to inappropriate degradation or failure to degrade appropriate substrates resulting in severe consequences for the cell.

A strikingly different, yet equally elaborate process, governs the degradation of many, if not most, transmembrane proteins. These proteins are covalently modified by a single ubiquitin moiety, as opposed to a polyubiquitin chain (36, 40, 54, 87, 98). Monoubiquitinated membrane proteins are recognized by a series of receptors that contain specific monoubiquitin-binding domains. These receptors target monoubiquitinated membrane proteins through a series of trafficking steps that ultimately deliver them to their destruction in the lysosome. Unlike the proteasome, the lysosome is a membrane-delimited organelle. Lysosomal proteases and lipases efficiently degrade small internal vesicles loaded with membrane proteins. This delivery system has other functions in addition to protein degradation. Many lysosomal and vacuolar hydrolases arrive via this pathway, and the pathway also is used for antigen presentation. Much attention has focused on the pathway because it is co-opted by HIV and other enveloped retroviruses in order to bud from cells (34, 66, 91, 101, 111).

The MVB Sorting Pathway

The ubiquitin-dependent downregulation of activated signaling receptors at the lysosome requires sorting of the receptors at the endosome into a unique class of vesicles that invaginate into the interior of the endosome. The endosomal compartment containing these vesicles is referred to as the multivesicular body or MVB. Pioneering electron microscopy studies showed that ferritin-conjugated EGF bound to the EFGR is sorted into the lumenal vesicles of MVBs en route to the lysosome (35, 37). Fusion of the MVB with the lysosomal membrane results in delivery of the lumenal MVB vesicles and their contents into the lysosome where the vesicles and the transmembrane receptor are degraded. This unique mechanism enables cells to degrade entire transmembrane proteins as well as lipids in the MVB vesicles. Membrane proteins that are excluded from the inner MVB vesicles remain within the limiting membrane of the MVB. Following fusion with the lysosome, these proteins are transferred to the limiting membrane of the lysosome. Recent studies have demonstrated that ubiquitin serves as a signal for efficient sorting into the MVB transport pathway (6, 53, 88, 108).

Studies in mammalian cells have revealed critical roles for MVBs in such seemingly distinct processes as growth factor receptor downregulation, antigen presentation and retroviral budding. However, the simple yeast Saccharomyces cerevisiae has served as an important model system for the discovery of the molecular machinery essential for MVB sorting (53). An unexpectedly large number of protein complexes have been identified that directly bind to ubiquitin-modified cargo and also appear to direct the complex process of receptor sorting and MVB vesicle formation (see Table 1). The conservation of these components in other organisms including humans highlights the importance of this transport route in all eukaryotic cells.

Table 1.

Class E Vps proteins and Complexes

Complex	Yeast protein	Human protein	Motifs	Binds to
Vps27-Hse1 complex	Vps27	HRS	UIM, FYVE, VHS	Ubiquitin, PI(3)P, ESCRT-II (Vps23)
	Hse1	STAM1, STAM2	UIM, VHS, SH3	Ubiquitin
ESCRT-I complex	Vps23	TSG101	UEV	Ubiquitin, Vps27
	Vps28	VPS28
	Vps37	VPS37A, B, C, D	Coiled-coil
ESCRT-II complex	Vps22	EAP30	Coiled-coil
	Vps25	EAP25		ESCRT-III (Vps20)
	Vps36	EAP45	GLUE, NZF	Ubiquitin
ESCRT-III complex	Vps2/Did4	CHMP2A, B	Charged, Coiled-coil
	Vps20	CHMP6	Charged, Coiled-coil	ESCRT-II (Vps25)
	Vps24	CHMP3	Charged, Coiled-coil
	Snf7/Vps32	CHMP4A, B, C	Charged, Coiled-coil
Vps4 complex	Vps4	VPS4A, B	AAA ATPase, Coiled-coil	ESCRT-III
Other MVB Proteins	Bro1/Vps31	ALIX/AIP1	Charged, Coiled-coil	LBPA, Doa4, ESCRT-III
	Vps60/Mos10	CHMP5	Charged, Coiled-coil	ESCRT-III
	Fti1/Did2	CHMP1A, B	Charged, Coiled-coil	ESCRT-III
	Vta1	LIP5		Vps4
	Vps44/Nhx1	SLC9A6	Sodium/Proton Exchanger

Class E Vps Proteins

Genetic studies in yeast have identified more than 60 gene products involved in Vacuolar Protein Sorting (Vps). These genes encode transport components that function at distinct stages of protein traffic between the Golgi complex and the vacuole. A subset of the Vps proteins, the class E Vps proteins, has been shown to function in the MVB sorting pathway (17, 53, 76). Presently, a total of 17 class E VPS genes have been identified (Table 1, Fig. 1). Class E vps mutants accumulate endosomal membranes and exhibit defects in the formation of MVB vesicles. The characterization of these proteins has resulted in the identification of three high molecular weight cytoplasmic protein complexes that function in the MVB sorting pathway. These complexes are referred to as ESCRT (Endosomal Sorting Complex Required for Transport) complexes I, II and III (reviewed in (41, 54, 73)). The ESCRT machinery is required for the formation of MVBs. This process occurs at the late endosomal compartment, where the limiting membrane invaginates and buds small vesicles into its lumen, giving rise to the characteristic morphology of numerous intralumenal vesicles within a larger membrane-enclosed endosome (28, 31, 76).

Figure 1.

Domain structure and interactions in the ESCRT network. Protein:protein interactions within the network are indicated by solid blue lines. Interactions with lipids, ubiquitin moieties, and other proteins are shown with black arrows. For simplicity, only two of the ESCRT-III subunits, Vps20 and Snf7, are shown. The GLUE domain of human Vps36 binds to PIP₃; the lipid specificity of the GLUE domain of yeast Vps36 has not been characterized.

The yeast class E protein Vps27 and its mammalian homologue (yeast nomenclature will be used throughout for simplicity; see Table 1) are also required for protein sorting in the MVB pathway. Vps27 forms a complex in the cytoplasm with the Hse1 protein (5, 10, 12, 57, 84) and directly binds to monoubiquitinated cargo (10, 12, 84). ESCRT-I (Vps23, Vps28, and Vps37) also interacts with ubiquitinated cargo (14, 32, 53). Genetic studies indicate that ESCRT-II (Vps22, Vps25, and Vps36) acts downstream of ESCRT-I, however, the mechanism of this interaction is not yet known (6). ESCRT-III is composed of two major functional subcomplexes (Vps20/Snf7 and Vps2/Vps24 in yeast) that localize to endosomal membranes. ESCRT-III components fail to localize to endosomes in cells lacking ESCRT-II, suggesting a role for ESCRT-II in the recruitment/assembly of ESCRT-III at the endosome (6). Together, the ESCRT complexes appear to function in both cargo sorting and MVB vesicle formation. The genetic and biochemical data argue for an ordered reaction sequence; Vps27 recruits ESCRT-I, which in turn recruits ESCRT-II, which then recruits ESCRT-III to the endosome.

In addition to the complexes described above, there are several other important players. Vps4 is an AAA-type ATPase (8, 9). It has a critical role in the MVB sorting pathway by catalyzing the dissociation of all three ESCRT complexes from the endosome (6, 9, 53). Inactivation of Vps4 results in the accumulation of the ESCRT machinery on the surface the endosome (class E compartment). Purified Vps4 assembles into a homo-oligomer when loaded with ATP and is recruited to the endosome via the ESCRT-III complex (6, 9). The enzyme Doa4 (2, 6) deubiquitinates MVB cargoes prior to sorting into MVB vesicles also is recruited to the ESCRT-III complex. Doa4 is recruited via an interaction with Bro1 (56, 63, 77), another class E Vps protein (Vps31).

THE VPS27/HSE1 COMPLEX

The Vps27/Hse1 complex binds Ub via UIM motifs in both subunits (12, 14, 52, 72, 81, 84, 97). Vps27 possesses a FYVE domain (for Fab1, YGL023, Vps27, and EEA1), that binds to the endosomal lipid phosphatidylinositol 3-phosphate (PI(3)P (20, 33, 69, 78, 90, 100)}. Endosomal PI(3)P recruits the Vps27/Hse1 complex to the endosome. Vps27 is a docking site for the ESCRT-I complex, and thereby initiates the MVB sorting reaction at the limiting membrane of the endosome. ESCRT-I physically interacts with membrane-bound Vps27/HRS through a motif in the COOH-terminal portion of Vps27 (4, 10, 13, 23, 55, 83). Therefore, the Vps27 protein appears to direct the compartment-specific activation of MVB sorting and Vps27 function is regulated by specific interactions with both PI(3)P and ubiquitinated cargo at the late endosome. Hse1 and STAM contain an SH3 domain. Vps27, Hse1, and their mammalian homologs contain predicted helical regions that are essential for formation of the complex (86).

The extreme C-termini of Vps27 contains the sequences LIEF or LIEL, which bind to a groove in the clathrin amino-terminal β-propeller domain (106). The human homologue of Vps27 has been shown to bind to clathrin via this motif (85). Planar clathrin lattices have been seen on endosomes containing Hrs (89). Clathrin may serve to cluster Vps27 and cargo at sites that will later invaginate to form the MVB vesicles. The human homologues of Vps27 and Hse1 are phosphorylated on Tyr residues, but the functional significance of this is unknown (57).

Vps27 FYVE domain and membrane localization

The FYVE domain that is responsible for localizing the Vps27/Hse1 complex is a compact 60 amino acid Zn²⁺-finger that selectively recognizes PI(3)P in preference to all other phosphoinositides. The crystal structure of the Vps27 FYVE domain (69) showed how such a small domain could selectively bind membrane-embedded PI(3)P (Fig. 2). A basic RHHCR motif on the first β strand provides all but one of the basic ligands for the acidic PI(3)P headgroup, as shown by the structure of the complex of the EEA1 FYVE domain with Ins(1,3)P₂ (27); reviewed in (70). The His residues with their relatively short side-chains are proximal to the 3-phosphate group, while the Arg residues are more distal, allowing room for their longer side-chains to reach the headgroup.

Figure 2.

Structure of Vps27. Structures are shown where available for Vps27 proteins (FYVE (69), UIM1-ubiquitin complex (103), UIM1-UIM2 (103)), otherwise modeled on the basis of the closely related structure of the Hrs-VHS domain (64). Linker regions between the domains were generated arbitrarily and are shown only to indicate the length of the segment. The C-terminal putative Hse1-binding domain and the extended region of Vps27 are not shown. Ubiquitin is shown fused via an isopeptide bond between Gly-76 and Lys-8 of the prototypical cargo pro-carboxypeptidase S, and the transmembrane helix of pro-CPS was modeled as an ideal helix. Membrane docking of Vps27 is based on the computationally predicted optimal docking mode for the FYVE domain.

The FYVE domain, like most other membrane-lipid targeting domains (25, 46, 47, 61), is anchored to cell membranes by a combination of specific lipid binding and by less specific electrostatic and hydrophobic forces (26, 27, 58-60, 69). The Vps27 FYVE domain has low (K_d = 90 μM) affinity for the soluble inositol (1,3)-bisphosphate, which corresponds to the headgroup of PI(3)P (15). The affinity rises to K_d = 30 nM when PI(3)P is presented in a realistic model of an endosomal membrane (15). This 3000-fold gain in affinity is explained by a hydrophobic protrusion (turret loop) of the FYVE domain that inserts into the hydrophobic core of the membrane. The depth and angle of membrane penetration dictate how the FYVE domain will be oriented relative to the bilayer, and what other surfaces are available for interactions. These parameters have been analyzed by spectroscopy (59), computational simulations (26), and by structure-based modeling (27, 64, 69). While there is no consensus model for a single membranebinding geometry for all FYVE domains, a consistent general picture has emerged for the best studied FYVE domains, including that of Vps27. For all PI(3)P binding FYVE domains, the turret loop is buried in the membrane. The Vps27 FYVE domain is best described as binding PI(3)P in a “side-on” orientation (26, 69), while the EEA1 FYVE domain is tilted with its PI(3)P binding face close to the membrane in a partially “faceon” manner (26, 27). Modeling suggests that differences in the distribution of charged residues on the FYVE domain surface control these differences. Given their variability, it seems likely that FYVE domains can wobble in situ, and are not likely to be rigidly constrained by membrane forces alone. Such flexibility might make FYVE domain proteins well-adapted to function in dynamic protein networks, in which complexes are rapidly formed and broken down, and the context of the domain’s interactions is subject to constant change.

VHS domains in the Vps27/Hse1 complex

Both subunits of the complex have N-terminal Vps27/Hrs/STAM (VHS) domains whose function is intriguing but still unknown. VHS domains are octahelical bundles (64, 68, 71, 95, 118) that are present only at the extreme N-termini of the proteins that contain them. Although the VHS domains of the GGA trafficking adaptors bind directly to acidic cluster-dileucine motifs in cargo proteins in a groove between two helices of the VHS domain (68, 71, 95, 118), critical residues in this groove are altered in the VHS domains of the Vps27/Hse1 complex, and the ligands for these VHS domains are unknown. Some non-ubiquitinated G-protein coupled receptors are trafficked by the human homologue of Vps27 (43), and its VHS domain is one candidate for recognizing such cargo (38).

UIMs and recognition of ubiquitinated cargo

UIMs are short helical motifs first identified on the basis of homology to the ubiquitin-recognition sequence in the proteasome subunit S5a (45). Hse1 has one UIM, and Vps27 has two. These UIMs bind monoubiquitin with low affinity, with K_d values in the range of 200 μM -2 mM (30). The second UIM of Vps27 has a ~10-fold lower affinity for ubiquitin than the first (30). The structure of the second Vps27 UIM has been determined alone (30), showing that the UIM comprises a single helix. The structure of a Vps27 tandem UIM construct has been determined in complex with ubiquitin by NMR (103). Only the first UIM (UIM-1) was found to participate in complex formation in the structure, consistent with its higher affinity. The second UIM was found to flop about freely in solution. The 28-residues linker between the two UIMs is completely disordered and the two UIMs appear to sample all accessible conformational space relative to each other.

The Vps27 UIM-1 binds to the Leu-8, Ile-44, Val-70 hydrophobic patch on the ubiquitin surface. This is the same surface that is recognized by all other ubiquitin binding domains characterized to date. The UIM-1 contains a single 6-residue hydrophobic strip that interacts with ubiquitin. The signature C-terminal Ser of the UIM forms a hydrogen bond with the main-chain of ubiquitin, and conserved N-terminal Glu residues form salt bridges with ubiquitin Arg-42 and Arg-72. The 400 Ų surface area buried in the complex is smaller than that seen in most protein:protein complex, but is typical of what we have come to expect for low affinity ubiquitin binding domain complexes.

SH3 domain of Hse1 and STAM

The SH3 domain of STAM recruits the deubiquitinating enzyme UBPY (49). The biological rationale for a deubiquitination event at an early stage of entry into the ESCRT pathway is unclear, since the ESCRT proteins interact with ubiquitinated cargo. It seems possible that the STAM/UBPY interaction functions as an off switch to direct cargo out of the pathway or to inactivate components of the ESCRT machinery that are ubiquitinated. The SH3 domain of STAM recognizes a non-canonical SH3 binding motif within UBPY of the form PX(V/I)(D/N)RXXKP (48) with a relatively low affinity of 27 μM, and the structure of this complex has been determined. The SH3 domain of Hse1 has been proposed to recruit Upb7 based on a large-scale proteomic study of yeast SH3 domains (107), although Ubp7 does not contain the PX(V/I)(D/N)RXXKP motif, and the interaction has yet to be confirmed.

THE ESCRT-I COMPLEX

The ESCRT-I complex directly binds to monoubiquitinated protein cargo through its UEV domain, a catalytically inactive variant of an ubiquitin conjugating enzyme (53). ESCRT-I interacts with Vps27 and a number of other cellular proteins, including the mammalian counterpart of Bro1 (101, 111) and the ubiquitin ligase Tal (3) via their P(S/T)XP sequences, as described above. Intense interest has centered on this motif since the discovery that HIV and certain other enveloped retroviruses contain this motif in their envelope proteins and use it to hijack the MVB sorting machinery to bud from host cells (24, 32, 65, 74, 109). These P(S/T)XP sequences also bind to the UEV domain. Tsg101 and Vps23 contain a C-terminal “steadiness box” (29) that is important for stability in vivo. The functions of the other domains and motifs within the ESCRT-I complex have yet to be worked out.

The UEV domain of Vps23

The cargo-recruitment end of the ESCRT-I complex is the UEV domain, which is responsible for binding both to monoubiquitin moieties and to P(S/T)XP motifs (Fig. 3). The structures of the UEV domains of Vps23 (105) and Tsg101 (102) have been determined in complex with ubiquitin. In both structures, two different regions of the UEV domains contact ubiquitin. The β1-β2 “tongue” contacts the Ile 44 hydrophobic patch of ubiquitin, the same region involved in contacts with the Vps27 UIM and with other monoubiquitin binding domains. Extending the oral analogy, the loop between the α3 and α3’ helices forms a “lip” that contacts a hydrophilic site centered on Gln-62 of ubiquitin (102, 105). In contrast to the Ile-44 site, the Gln-62 site does not participate in most other known Ub/Ub-binding domain interactions. Even though the UEV domain was discovered as a catalytically inactive homolog of ubiquitin conjugating enzyme, ubiquitin binds to the UEV domain in a completely different manner.

Figure 3.

Structure of ESCRT-I. Structures are shown for the UEV domain of human Vps23 in complex with ubiquitin (102) and the HIV-1 p6 PTAP-containing peptide (82). The remainder of the ESCRT-I structure is not yet available.

The structure of the Tsg101 UEV domain in complex with the PTAP peptide of HIV-1 p6 shows how P(S/T)XP sequences are recognized by the ESCRT-I complex (82). The second Pro of the P(S/T)XP is a particularly critical element. The XP sequence, together with the first C-terminal flanking residues, form one turn of a type II polyproline helix. This conformation is also seen in canonical SH3 and WW (tryptophan-tryptophan) domain complexes with Pro-based peptide motifs (117). The UEV domain recognizes the second Pro by using an aromatic pocket reminiscent of Pro-recognition pockets in SH3 and WW domains. The first Pro is in an extended conformation and binds in a shallow pocket. The Thr hydroxyl appears to hydrogen bond to both main-chain and side-chain residues, although this was not fully defined in the NMR structure. The concept of using peptidomimetics directed at this site to block HIV release has attracted considerable interest.

THE ESCRT-II COMPLEX

ESCRT-II forms a nexus between ubiquitinated cargo, the endosomal membrane, and the ESCRT-I and III complexes (7). Its structural organization is currently the best understood of the three ESCRT complexes (Fig. 4). The complex contains two Vps25 subunits and one each of Vps22 and Vps36 (42, 104). The N-terminal two-thirds of Vps36 contains a series of domains (GLUE (99) and NZF (1)) that interact with membranes and cargo, and perhaps have other functions. Vps22, Vps25, and the C-terminal third of Vps36 form a tightly organized core that contains two binding sites for the ESCRT-III subunit Vps20 (104).

Figure 4.

Structure of ESCRT-II. The structure of the ESCRT-II core (42) is shown docked to a membrane on the basis the interaction between its Vps25 subunit and the membrane-bound ESCRT-III subunit Vps20 (104) (myristoyl group on Vps20 not shown for simplicity). The uncomplexed NZF1 domain and the NZF2-ubiquitin complex are modeled on the basis of the Npl4-NZF structure and ubiquitin complex (1). The GLUE domain (a variant of the GRAM domain, which is in turn a variant of the PH domain) is modeled and docked to the membrane on the basis of the GRAM domain of the lipid phosphatase MTMR2 (11). The GLUE domain of human Vps36 binds to PI(3,4,5)P₃ and ubiquitin at sites that have yet to be determined (99), and are not shown. The binding properties of the yeast Vps36 GLUE domain have yet to be reported. The dashed line between the GLUE domain and Vps36 core winged helix (WH) region indicates residues 290-395 of Vps36, whose structure is unknown.

The ESCRT-II core

The ESCRT-II core is shaped like the letter “Y”, with one Vps25 subunit forming the stalk. One of the branches is formed by the second Vps25 subunit, and the other branch is formed by the subcomplex consisting of Vps22 and the C-terminal third of Vps36 (42, 104). Although the three subunits do not have primary sequence homology to each other, each consists of two repeats of a winged helix (WH) domain. WH domains are typically found in DNA binding proteins, where the “wing” (a loop between two of the β strands) contacts the nucleic acid. Vps22 and Vps36 formed a tightly bound subcomplex. The second WH domains (WH2) of Vps22 and Vps36 bind to the Vps25 subunits through closely equivalent interactions. Vps22 and Vps36 present an aromatic cage to the Vps25 subunit. The N-terminus of Vps25 contains two repeats of the sequence PPXY. Sequences of this type are better known for binding to WW (a polyproline-rich peptide motif binding domain that is unrelated to the WH domain, despite the similar abbreviation) domains in a conformation in which the Tyr side-chain is exposed and presented to the WW domain. In Vps25, the Tyr side-chain is buried, and the diPro sequence interacts with the aromatic cages in the WH2 motifs of Vps22 and Vps36. The Tyr of the PPXY motif of Vps25 is buried even in the isolated subunit unbound to the rest of the complex (113), suggesting it is unlikely to become exposed in a conformation available for WW domain binding. All of the intersubunit contacts are required for the complex to mediate CPS sorting in yeast (42).

Vps36 NZF domains

Yeast Vps36, but not its human counterpart, contains two Npl4 zinc finger (NZF) motifs in the region N-terminal to the core WH domains. The Npl4 NZF domain was first shown to bind ubiquitin, and this interaction has been characterized structurally (1). The NZF domain binds to the same Ile-44 patch on ubiquitin as the UIM and UEV domains. Thus, none of these domains can bind to ubiquitin simultaneously. The second NZF domain of Vps36 binds to ubiquitin, and the ubiquitin-binding site on the NZF domain is important for the sorting function of Vps36. The function of the first NZF domain is unknown.

Vps36 GLUE domain

The absence of NZF or other known ubiquitin binding domains in human Vps36 and the other human ESCRT-II subunits led to a puzzle: how could the human ESCRT-II complex function in sorting ubiquitinated cargo without a ubiquitin binding domain. This seeming paradox was resolved with the discovery of the GLUE (Gram-like ubiquitin) domain near the N-terminus of the human protein (99). Vps36 also contains a GLUE domain, which is split into two segments by the first NZF domain. GLUE domains bind not only ubiquitin, but the lipid PIP₃ as well (99). The structure of the GLUE domain has not been determined, but it appears to be similar to the phospholipid-binding pleckstrin homology (PH) domain based on distant sequence similarity.

THE ESCRT-III COMPLEX

The ESCRT-III complex consists of several subunits that are homologous to each other, highly charged, and contain predicted coiled coil regions (6). Four subunits, Vps2, Vps20, Vps24, and Snf7 are essential for sorting function in yeast (6) (Figure 4). Two other proteins, Vps60 and Did2, share homology to the other ESCRT-III subunits, but are less critical for function (6). Isolated ESCRT-III subunits are cytoplasmic in yeast (6), and soluble, at least at low concentrations, in vitro (56, 62). One of the subunits, Vps20, is myristoylated, but in isolation is nevertheless cytosolic. Thus the myristoyl group alone seems insufficient to drive it to the membrane. Upon association, the subunits form a large and tightly membrane-bound assembly of indeterminate stoichiometry (Fig. 5). It is currently thought that this assembly comprises an oligomer of ESCRT-III complexes on the membrane surface.

Figure 5.

Structure of ESCRT-III. The structure of ESCRT-III is unknown. This cartoon shows a tripartite variant of the bipartite model proposed by Hanson and colleagues (62). The C-terminal anionic regions of ESCRT-III subunits is roughly twice as large as the N-terminal basic region, and appears to have functions both in oligomerization of ESCRT-III on membranes and in binding to other proteins, such as Vps4. Black arrows indicate directions of in which the oligomeric array can grow.

The major link between the ESCRT-III complex and upstream complexes occurs via ESCRT-II in yeast (16). This interaction is conserved in humans (111). However, in the human counterpart of Bro1 bridges Vps23 of ESCRT-I with Snf7 of ESCRT-III and provides a second link. The Vps20 subunit of ESCRT-III (16, 111) directly binds to Vps25 of yeast and human ESCRT-II (104, 116). The Vps2/Vps24 subcomplex of ESCRT-III recruits Vps4 in yeast (6). However, Snf7 and other ESCRT-III proteins can also bind to Vps4 (16, 111), suggesting the ESCRT-III/Vps4 interaction is not stringently specific. Vps60 binds to the class E Vps protein Vta1 in yeast (16, 96) and humans (112). Human vps24 has been proposed to bind to PI(3,5)P₂ (114). In summary, all ESCRT-III subunits have a conserved primary structure and interact with each other in a similar manner, while the individual subunits retains specific interactions of their own.

The N-terminal one-third or so of ESCRT-III subunits are highly basic, while the C-terminal two-thirds are acidic. Both regions contain predicted coiled-coils. The simple model for the monomeric form of the subunit is that the basic and acidic regions form an antiparallel coiled-coil pair with each other, stabilized by electrostatic interactions between the two halves. Experimental evidence for this model comes from a comparative analysis of human Snf7 and Vps24 and their fragments (62). The basic N-terminal regions are found to bind tightly to membranes even in isolation (62). The C-terminal regions, in contrast, bind to human Vps4 (62, 94). These is also evidence that the N-terminal basic regions can interact with other proteins (116), and not just membranes. These data suggest that monomeric ESCRT-III subunits are in a closed conformation in which the N- and C-terminal portions are tightly associated with each other, and are not available for interactions with membranes or other proteins.

What initiates formation of the insoluble ESCRT-III complex from its soluble subunits? In a working model, interactions with other proteins compete with the internal N/C interaction and liberate the N-terminal portion for interactions with membranes. This triggers the membrane localization of the subunit. In a working model, the membranebound form of the ESCRT-III subunit is in a more open conformation, more available for interactions with other subunits, and thus disposed to form a polymeric assembly on the membrane (Fig. 5). The formation of the ESCRT-III complex in independent of Vps4 and ATP (6), and ESCRT-III-like aggregates of isolated recombinant subunits have been observed to form spontaneously (62). It seems likely that in vivo formation of the ESCRT-III complex follows the binding of isolated ESCRT-III subunits to ESCRT-II, Bro1, or other factors thay allosterically promote the complexation-compentent conformation. These considerations point to factors such as ESCRT-II and Bro1 as the likely initiators of ESCRT-III complexation.

The membrane association of ESCRT-III is essentially irreversible in the absence of an energy input: Vps4 must hydrolyze ATP to solubilize the assembly, once formed. In this respect, ESCRT-III proteins are analogous to the SNAREs of membrane fusion, which must be separated by the action of NSF once tight complexes have been formed. In contrast to ESCRT-III subunits, however, SNAREs contain transmembrane regions and are permanently tethered to membranes (18), while ESCRT-III proteins cycle on and off membranes.

BRO1

Bro1 is a monomeric protein that is intimately associated with the ESCRT complexes (Figure 5). Bro1 is involved in deubiquitination of the general amino acid permease Gap1 (75) and is required for trafficking of carboxypeptidase S via the MVB pathway (77). Bro1 recruits Doa4, which deubiquitinates MVB cargo proteins (63). Bro1 is recruited to MVBs through its interactions with the ESCRT-III subunit Snf7, also known as Vps32 (16, 77).

The human homolog of Bro1 interacts with the human counterpart of Snf7 (50, 51, 101, 111), hence this key interaction is conserved from yeast to man. It contains a C-terminal Pro-rich region that interacts with the endocytic proteins SETA (22, 92), endophilins (21), and the human Vps23 subunit of ESCRT-I (101, 111). In contrast, Bro1 lacks a P(S/T)XP sequence and does not interact strongly with Vps23 (16). The human homologue of Bro1 is a key player in retroviral budding, because it interacts with HIV-1 and other retroviral proteins containing the sequence motif YPXL (34, 66, 101, 111). Little is know about the function of YPXL motif host proteins that bind to Bro1 homologs, although one protein, PacC, has been described in Aspergillus (110).

Bro1 domain

Bro1 and several other late endosomal proteins share a conserved N-terminal Bro1 domain. The Bro1 domain consists of roughly 370 residues and has a complex structure that is built around a core helical solenoid similar to TPR domains (56) (Fig. 6). The Bro1 domain is necessary and sufficient for binding to the ESCRT-III subunit Snf7 and for the recruitment of Bro1 to late endosomes (56). Snf7 binds to a conserved hydrophobic patch on Bro1 that is required for protein complex formation and for the protein sorting function of Bro1 (56). The Bro1 domain of the Bro1 protein does not contain the binding site for Doa4 (56). A second conserved hydrophobic patch at one tip of the domain is not critical for sorting, and its function has yet to be determined. The structure resembles a boomerang with its concave face filled in, and it is tempting to speculate that the convex face could mediate the putative ability of the human Bro1 counterpart to sense negative curvature in invaginating lumenal vesicles (67); however, this idea has not been tested for the Bro1 domain.

Figure 6.

Structure of Bro1. The structure of the Bro1 domain (56), which comprises the N-terminal half of Bro1, is shown docked to a negatively curved membrane. The interaction of this domain with negatively curved membranes is suggested by the shape of the structure and by the properties of the full-length human Bro1 homologue; however, this interaction and docking mode are speculative and have yet to be directly tested.

VPS4

Vps4 is responsible for the ATP-dependent disassembly of the ESCRT complexes (8, 9). Vps4 is a homo-oligomer in yeast, and consists of an N-terminal MIT domain and a central AAA ATPase domain (Fig. 7). Two Vps4 isoforms in human, Vps4A and Vps4B, can hetero-oligomerize with each other. AAA ATPases are ubiquitous disassembly machines that are ring hexamers and function in a wide range of cell processes (19). The structure of the AAA domain of Vps4 has been determined (93) and is similar in outline to the structures of other AAA ATPases such as NSF and p97. The N-terminal MIT domain (first referred to as an ESP domain (79)) of Vps4 binds to the human ortholog of the ESCRT-III-like protein Did2, and to other ESCRT-III subunits as well (94, 115). The structure of a MIT domain from human Vps4 has been determined and shown to be a three-helix bundle (94) that contains the equivalent of 1.5 TPR repeats, reminiscent of the Bro1 domain. The structure of the Vps4 monomer has been used to model a hexamer that assembles in the presence of ATP. This hexamer contains a central pore. In a working mechanism for ESCRT-III disassembly, supported by mutational analysis of the modeled pore, individual membrane-bound ESCRT-III subunits are fed through the pore into solution and so converted to their monomeric, soluble state.

Figure 7.

Structure of Vps4. The modeled structure of the Vps4 AAA domain hexamer was modeled (93) is shown together with the corresponding six copies of the MIT domain (94). For simplicity, a single subunit of the ESCRT-III complex is shown engaged to a single MIT domain. The ESCRT-III subunit is thought to be fed through the central pore in the hexamer (not shown in this view) as ATP is hydrolyzed by the AAA domain.

CONCLUSIONS AND FUTURE PERSPECTIVES

In gross terms, the ESCRT systems has two major functions: the recruitment of cargo to MVBs, followed by its internalization via the invagination of lumenal vesicles (Fig. 8). Much has been learned in the past five years about the structures, interactions, and ordered assembly of the ESCRT machinery, and the mechanism of cargo recruitment. The presence of ubiquitin binding domains in Vps27/Hse1, ESCRT-I, and ESCRT-II has led to the suggestion that there is a serial hand-off of ubiquitinated cargo from one complex to the next. Favoring this model, structural analysis shows that the ubiquitin binding domains interact with just one region of ubiquitin, the Ile-44 patch. This offers an elegant mechanism to prevent more than one complex from interacting with ubiquitin at a given time. However, no direct evidence of hand-off is available. In an alternative model, the presence of multiple ubiquitin binding sites in the Vps27/ESCRT-I/ESCRT-II “complex of complexes” offers an attractive mechanism for receptor clustering by simultaneous binding. The presence of two ubiquitin binding domains in Vps27 and ESCRT-II seems more consistent with a clustering model than a hand-off model, since there would be little reason to orchestrate hand-off within a single complex. The two models are not mutually exclusive, since hand-off might be operative at one stage in the pathway and clustering at another.

Figure 8.

The ESCRT complexes in MVB Sorting - The Vps27 protein complex initiates the MVB sorting process. It is targeted to endosomal membranes via its FYVE domain that binds PI(3)P, and its UIM domains which bind ubiquitinated MVB cargo such as carboxypeptidase S (CPS). Vps27 subsequently recruits and activates the ESCRT-I complex via the P(S/T)XP motif in the C-terminal domain of Vps27 that interacts with the UEV domain of Vps23 in ESCRT-I. Ubiquitinated cargo is recognized by ESCRT-I (via the UEV domain of Vps23) and by ESCRT-II (via the NZF domain in Vps36). ESCRT-III is required for concentration of cargoes into MVB vesicles and coordinates the association of accessory factors such as Bro1 and the Doa4 deubiquitinating enzyme that removes ubiquitin from cargo. The AAA-type ATPase Vps4 plays a critical role in catalyzing the dissociation of the ESCRT complexes. Together, these proteins appear to direct MVB vesicle formation, cargo sorting into MVB vesicles and vesicle fission. See text for further details.

Despite many advances in other aspects of ESCRT function, the fundamental mechanism of vesicle budding remains a matter of conjecture. MVBs from mammalian cells have been reported to be rich in the phospholipid 2,2’ lysobisphosphatidic acid (2,2’ LBPA), which has been suggested to be a key driving force for membrane deformation leading to vesicle invagination (67). On the other hand, no evidence has surfaced for the presence of 2,2’ LBPA in yeast, and no Vps proteins have emerged as candidate regulators of LBPA levels. Given the high conservation of the protein machinery involved, it would be surprising if completely different mechanisms were responsible for the invagination of lumenal vesicles in yeast and mammals. It seems likely that changes in vesicular pH and/or ionic strength are important. The Na⁺/H⁺ antiporter Vps44 is the only transmembrane protein among the class E vps proteins, and remains the most mysterious in terms of its role in budding. It will be important to determine whether Vps44 drives pH or ionic strength changes in the lumen of the MVB and/or the budding vesicles, and if it does, what proteins act as the downstream effectors of these changes.

One of the major motivations for studying the ESCRT pathway is the discovery that this pathway is central to HIV budding, and presents an unprecedented number of new potential therapeutic targets. A number of proteins in the pathway are clearly essential for budding, and are incorporated directly into HIV virions (101, 111). The usefulness of pathway members as targets will depend on the relative sensitivity of virus and host to inhibition of the ESCRT pathway. The locus of HIV budding through this pathway remains controversial. The mechanism by which nascent virions avoid the fate of most other MVB cargodestruction in the lysosome- remains unknown, and its identification would present an exceptionally interesting target.

The emerging picture of ESCRT assembly on endosomal membranes suggests the formation of arrays of inexact stoichiometry. Dynamic protein networks that assemble on membranes are central to signal transduction and subcellular trafficking. Their complexity, kinetic fragility, membrane localization, and inexact stoichiometry present great challenges to obtaining a precise structural and mechanistic understanding. The payoff will be equally great, perhaps providing a roadmap for analysis of many analogous membrane-bound signaling and trafficking systems.

List of acronyms and abbreviations

AAA

ATPases associated with diverse cellular activities

AIP1

ALG-2 interacting protein 1, synonym for Alix

Bro

BCK1-like resistance to osmotic shock

CC

coiled-coil

CHMP

chromatin modifying protein (human orthologs of ESCRT-III subunits)

CPS

carboxypeptidase S

Doa

degradation of alpha2

DUB

de-ubiquitinating enzyme

EAP

ELL associated protein (human orthologs of ESCRT-II subunits)

EEA1

early endosomal antigen 1

EGF

epidermal growth factor receptor

EGFR

EGF receptor

ESCRT

endosomal sorting complex required for transport

FYVE

Fab1, YOTB, Vac1, EEA1

Gap

general amino acid permease

GAT

GGA and TOM

GLUE

GRAM-like ubiquitin binding in EAP45

GRAM

glucosyltransferases, Rab-like GTPase activators and myotubularins

HIV

human immunodeficiency virus

Hrs

hepatocyte growth factor receptor substrate

Hse

has symptoms of class E mutants; resembles Hrs, STAM, East

LBPA

lysobisphosphatidic acid

MIT

microtubule interacting and trafficking

MVB

multivesicular body

NZF

Npl4 zinc finger

PI

phosphatidylinostiol

PI(3)P

phosphatidylinositol 3-phosphate

PIP₃

phosphadtidylinositol (3,4,5)-trisphosphate

SB

steadiness box

SH3

src homology-3

Snf

sucrose non-fermenting

STAM

signal transducing adaptor molecule

TOM

target of myb1

TPR

tetratricopeptide repeat

Tsg101

tumor suppressor gene 101, the human ortholog of vps23

Vps

vacuolar protein sorting

Ub

ubiquitin

UBP

ubiquitin isopeptidase

UEV

unusual E2 variant

UIM

ubiquitin interacting motif

VHS

Vps27, Hrs, STAM

WH

winged helix

WW

tryptophan-tryptophan