In eukaryotic cells, delivery of transmembrane proteins into the lumen of the lysosome for degradation is mediated by the multivesicular body pathway. The function of the ESCRT protein complexes is required for both the formation of multivesicular body lumenal vesicles and the sorting of endosomal cargo proteins into these vesicles. Recent studies have identified additional factors that seem to function as an upstream cargo retention system feeding into the ESCRT machinery, given new insights into the dynamic structure of multivesicular bodies, and identified a potential mechanism for multivesicular body vesicle formation.
Eukaryotic cells maintain and adjust their cell surface protein composition in part by removing transmembrane proteins from the plasma membrane by endocytosis and delivering them to the lysosome for degradation (Figure 1). The topologic problem of degrading transmembrane proteins in the lumen of the lysosome is solved by the formation of endosomal structures called MultiVesicular Bodies (MVBs). MVBs are part of the endosomal system of eukaryotic cells and are formed by invagination and budding of vesicles from the limiting membrane of endosomes into the lumen of the compartment. During this process, endosomal transmembrane proteins destined for degradation are sorted into the forming vesicles and delivered to the lumen of the lysosome, while others are maintained on the limiting endosomal membrane. Proteins retained on the limiting membrane are either recycled to the trans-Golgi network (TGN) and plasma membrane or are delivered to the limiting membrane of the lysosome. Thus, the MVB pathway also plays an important role in regulating the composition and function of the lysosomal compartment. The sorting of cargoes into MVB vesicles is a highly regulated process and a variety of studies indicate that monoubiquitination serves as a signal that directs protein cargo into the MVB pathway (reviewed in 1–3).
Both MVB vesicle formation and sorting of ubiquitinated MVB cargo depends on the function of a group of at least 18 conserved proteins that were originally identified in the yeast Saccharomyces cerevisiae as class E Vps (Vacuolar Protein Sorting) proteins (4,5; reviewed in (3)). Deletion of each of the class E VPS genes in yeast results in the mislocalization of transmembrane proteins to the limiting membrane of the vacuole and the accumulation of endosomal cargoes in large aberrant structures adjacent to the vacuole, called the ‘class E compartment’ (6,7). The class E Vps proteins are highly conserved in eukaryotic cells and the phenotypes associated with loss-of-function of class E proteins in mammalian cells are analogous to the trafficking phenotypes in yeast class E mutants, suggesting that the roles of the class E proteins in MVB sorting are conserved from yeast to mammals (8–19). For each known component of the yeast MVB sorting machinery, one or more mammalian homologs have been identified (Table 1). The fact that mammalian cells express several homologs of some of the yeast class E proteins might reflect a higher complexity in the regulation of the mammalian MVB sorting machinery.
Table 1. Class E Vps proteins discussed in this review
VPS37A, B, C, D
CHMP4A, B, C
UIM, FYVE, VHS
UIM, VHS, SH3
The majority of the class E Vps proteins are constituents of three separate heteromeric protein complexes called ESCRT-I, ESCRT-II and ESCRT-III (Endosomal Sorting Complex Required for Transport) (20–22). These protein complexes are transiently recruited from the cytoplasm to the endosomal membrane where they function sequentially in the sorting of transmembrane proteins into the MVB pathway and in the formation of MVB vesicles. Based on biochemical analysis of the ESCRT complexes in yeast, a model for the function of the class E proteins in this pathway has been proposed (22) (Figure 2A). Initially, the ESCRT-I protein complex binds to ubiquitinated endosomal cargo and in some way activates ESCRT-II. ESCRT-II in turn initiates the oligomerization of at least four small coiled-coil proteins, resulting in the formation of a large endosome-associated structure, the ESCRT-III complex. ESCRT-III seems to function in the concentration of MVB cargo (22). In addition, ESCRT-III also recruits the deubiquitinating enzyme Doa4, which removes the ubiquitin tag from the cargo protein prior to sorting into the MVB vesicles (23). Finally, after protein sorting has been completed, a multimeric AAA-type ATPase, Vps4, binds to ESCRT-III and disassembles the ESCRT-III complex in an ATP-dependent manner (24,25). The Vps4-dependent dissociation of the ESCRT machinery is currently the final distinguishable step of cargo sorting and is a prerequisite for vesicle formation.
Recent studies have described additional factors that function up- and down-stream of the ESCRT complexes. Furthermore, large-scale protein interaction analyses have uncovered potential new interactions among the class E Vps proteins. This review will incorporate these findings into the current working model for MVB function and discuss several new studies that suggest novel mechanisms and components involved in the formation of MVBs themselves.
Cargo Recognition and Sorting
It has been shown that ubiquitin serves as a signal for entry into the MVB pathway, and it has been proposed that ESCRT-I serves as a sorting receptor that recognizes ubiquitin-tagged proteins at the endosome and initiates their proper sorting via the ESCRT machinery (20). However, recent studies have identified several additional factors that seem to function in the recognition of ubiquitinated cargo which appear to regulate ubiquitin-dependent sorting and concentration of cargo in concert with the ESCRT machinery and, surprisingly, clathrin.
Vps27 and its mammalian homolog HRS are particularly interesting as factors which may coordinate several sequential steps in the MVB pathway. The Vps27/HRS proteins contain two Ubiquitin Interaction Motifs (UIMs) that bind ubiquitinated proteins and are important for the sorting of MVB cargo (17,19,26–28). Vps27/HRS forms a protein complex with another ubiquitin-binding protein, Hse1/STAM, which also has been implicated in sorting of ubiquitinated endosomal cargo (17,19). Vps27/HRS is localized to early and late endosomal compartments and its proper localization depends on its FYVE domain, which binds to the endosome-specific lipid phosphatidylinositol-3-phosphate [PtdIns(3)P] (29–31). Interestingly, ESCRT-I localization to the MVB membrane is dependent upon its interaction with Vps27/HRS (18,32). Furthermore, Vps27/HRS contains a clathrin-binding motif and electron microscopy has demonstrated colocalization of HRS with endosomal clathrin coats (28,33,34). Unlike clathrin cages on plasma membrane and TGN, the endosomal clathrin coats are flat and do not seem to be involved in vesicle formation. Clathrin cages are known to concentrate adaptor proteins which function in the sorting of proteins exiting a donor compartment. Therefore, it has been proposed that on early endosomes, Vps27/HRS might act as an adaptor molecule that binds to both monoubiquitinated transmembrane proteins and ESCRT-I and acts to concentrate and localize cargo at the MVB via its interaction with the endosomal clathrin coat. According to this model, the HRS/clathrin coat retains ubiquitinated cargo in a subdomain of the early endosome and prevents recycling of cargo to the plasma membrane. Furthermore, Vps27/HRS recruits ESCRT-I to the endosome, which initiates the formation of MVBs and the sorting of the cargo into the luminal vesicles.
Another group of proteins that functions upstream of the ESCRTs are the GGA (Golgi-associated, γ-adaptin homologs, Arf-binding) proteins. These proteins have previously been shown to function as adaptors for the Golgi-to-endosome transport (reviewed in (35)). At the TGN the GGA proteins interact with both cargo and clathrin and thus are involved in the formation of Golgi-derived clathrin-coated vesicles. However, recent studies have now demonstrated that GGAs also bind ubiquitin and are necessary for biosynthetic and endocytic transport of ubiquitinated cargo into the MVB pathway. In yeast, the transport of newly synthesized general amino-acid permease 1 (Gap1) is regulated by the nutrient conditions. Under conditions where Gap1 is not required at the cell surface, the protein is delivered directly from the TGN to the endosomal/lysosomal system for degradation. This redirection at the TGN requires ubiquitination of Gap1 as well as the function of the GGA proteins (36–39). In mammalian cells one of the GGA proteins, GGA3, has been shown to be important for the endocytic trafficking and degradation of ubiquitinated cell-surface receptors (40). These observations suggest that the GGAs may function in a manner similar to Vps27/HRS for sorting and retaining ubiquitinated cargo in endosomal clathrin coats.
The endosomal clathrin coat with its associated ubiquitin binding adaptors seems to function as an upstream cargo concentration system that feeds into the ESCRT machinery (Figure 2A). It is not known, however, whether the endosomal clathrin coat is an integral part of the ESCRT machinery. The only known direct connection between the clathrin coat and the ESCRTs is Vps27/HRS, which plays an essential role in the localization of ESCRT-I. Vps27/HRS could function both in a clathrin-dependent and independent manner and thus could recruit the ESCRT machinery even in the absence of an endosomal clathrin coat. Furthermore, it is not clear if all ubiquitinated cargoes at the endosome interact with the clathrin coat or the clathrin coat plays a more specific role for cargoes of the endocytic pathway. This question is not easy to answer, as mutations that block the formation of the clathrin coat could broadly affect the endosomal system and thus might indirectly impair the sorting function of the ESCRT machinery.
Ubiquitin and the Sorting Machinery
A growing number of ubiquitin-binding domains have been identified which function in the endosomal protein sorting system. A recent addition is the NZF (Npl4 Zinc Finger) domain, which has been shown to generally act as a protein-interaction domain. The yeast ESCRT-II subunit Vps36 contains two NZF domains, one of which has been found to bind to ubiquitin and ubiquitinated proteins (41). Surprisingly, the NZF motifs are not conserved in the mammalian Vps36 homolog EAP45, which might suggest the presence of an additional protein in the mammalian ESCRT system that performs the function of the Vps36 NZF domains.
Detailed analysis of the interaction between ubiquitin and the different ubiquitin-binding domains of the endosomal sorting machinery have shown that all these interactions are weak (μm range) and occupy a similar surface of the ubiquitin molecule (28,42–44). The latter observation implies that ubiquitinated cargo is bound only by one ubiquitin-binding protein at a time. Therefore, to allow sequential interactions between cargo and the different ubiquitin-binding proteins, the cargo proteins must be ‘handed over’ from the upstream sorting machinery (Vps27/HRS, clathrin coat) to ESCRT-I and ESCRT-II.
Mutations in the NZF domain of yeast Vps36 that impair ubiquitin binding show phenotypes similar to those observed in class E VPS deletion strains, suggesting that ubiquitin binding is essential for ESCRT-II function (41). Similarly, the ubiquitin binding activity of the ESCRT-I subunit Vps23 has been shown to be necessary for MVB formation (20). This is in contrast to Vps27/HRS where ubiquitin binding mutations only affect sorting of ubiquitinated cargo but not the formation of MVBs (17,26). One possible explanation for these different phenotypes of ubiquitin-binding mutants is that in the case of ESCRT-I and ESCRT–II the interaction with ubiquitinated cargo not only serves to sort cargo but also has a regulatory function in activating downstream events, such as the formation of ESCRT-III. Another possibility is that, in contrast to Vps27/HRS, ESCRT-I and ESCRT-II might interact not only with ubiquitinated cargo but also with ubiquitinated proteins of the MVB sorting machinery itself, promoting the formation of a functional sorting complex. This model is supported by the observation that several proteins of the sorting machinery are monoubiquitinated (see Table 1). However, a study on TSG101 has shown that ubiquitination of this mammalian ESCRT-I subunit results in decreased MVB sorting and the solubilization of TSG101 from the endosomal membrane, suggesting that in case of ESCRT-I at least, ubiquitination plays a negative regulatory role (45).
Protein Interaction Network of MVB Sorting
As the ubiquitin-mediated interactions in the MVB pathway point out, the interactions between the sorting machinery and cargoes are most likely very complex. To get a deeper insight into the protein interaction network of the MVB sorting machinery, several yeast two-hybrid analyses have been performed. Two of these studies analyzed the interactions among the mammalian class E Vps proteins, and one study determined interactions within the yeast ESCRT system (46–48) (Figure 2B). The results from these studies support the proposed subunit composition of the three ESCRT complexes, as well as the predicted interactions between the ESCRT complexes that have been observed by biochemical assays (20–22). Additionally, two-hybrid analysis of the mammalian class E Vps proteins suggests a direct interaction between ESCRT-I and ESCRT–II, an interaction that had not been observed before using biochemical methods but which has been predicted based on genetic evidence (48). The two-hybrid analysis confirmed the predicted interaction between Vps4 and its substrate ESCRT-III. Furthermore, Vps4 was found to interact with the two class E proteins Fti1/CHMP1 and Vta1/SBP1. These two proteins together with Vps60, a Vta1 interacting protein, might function in the regulation of the Vps4 disassembly activity.
Although these two-hybrid studies give valuable insights into the network of interactions between the proteins of the MVB machinery, they cannot answer two important questions regarding the organization of the system: where do the proteins interact and when does this interaction occur? The MVB sorting machinery is a dynamic system where new protein interactions are continuously formed and broken. A good example is the class E Vps protein Bro1/ALIX. This protein binds to ESCRT-III and functions in the recruitment of the deubiquitinating enzyme Doa4 (49,50). In addition, two-hybrid analysis suggested that Bro1/ALIX also interacts with ESCRT-I and Vps4 (46–48). Based on these observations it has been speculated that Bro1/ALIX might act as a linker between the different ESCRT complexes. However, the same data would support a model in which Bro1 interacts transiently and in a consecutive fashion first with ESCRT-I, then with ESCRT-III and finally with Vps4. This model is supported by biochemical studies that found no stable association of Bro1 with either Vps4 or ESCRT-I (49) (M. Babst, unpublished results). Furthermore, the transient interaction between Bro1 and ESCRT-I (or Vps4) might not occur on the endosomal membrane but in the cytosol. Now that these studies have provided information on which proteins are able to interact with each another, it will be necessary to characterize these protein interactions in detail using functional, biochemical and cell biological methods to fully understand the architecture of the class E Vps proteins within the MVB sorting machinery.
MVB Vesicle Formation and Dynamics
The formation of MVB vesicles is unique in that it is directed toward the lumen of the compartment, rather than the cytosol. Thus, it is likely that the mechanism of ESCRT machinery function is different from the well-studied clathrin- and COP-dependent budding events at the endoplasmic reticulum, Golgi and plasma membrane. It is not easy to imagine how the ESCRT proteins could be directly involved in the invagination of the endosomal membrane without getting trapped in the lumen of the forming vesicle and ultimately degraded in lysosomes/vacuoles. This one-way use of the ESCRT machinery is unlikely since studies have shown that the ESCRT proteins have a slow turnover (M. Babst, unpublished results) and are recycled from the endosomal membrane by Vps4 (22,24). Therefore, it is possible that lipids, not proteins, are the key players in the membrane invagination step. In support of this idea, certain lipids, similar to protein cargoes, also seem to be concentrated and sorted into the lumenal vesicles of the MVB. One lipid which has been shown to be delivered via the MVB sorting pathway into the vacuolar lumen for degradation is PtdIns(3)P (51), the lipid that serves as an endosomal marker to properly localize the ESCRT machinery. In mammalian cells, the phospholipid lysobisphosphatidic acid (LBPA) has been found exclusively on late endosomes and, particularly, concentrated in the lumenal vesicles of MVBs (52). Endocytosed antibodies that specifically bind to LBPA have been shown to interfere with MVB sorting, suggesting that LBPA has an important role in the formation of MVBs (52,53). Furthermore, it has been demonstrated in a pure in vitro system that liposomes containing LBPA spontaneously form structures that resemble MVBs (54). The formation of these MVB-like structures is dependent on a pH gradient across the membrane of the liposome, similar to the pH gradient observed across endosomal membranes of living cells. The same study identified the class E Vps protein Bro1/ALIX as an LBPA-binding protein that was able to inhibit the formation of multivesicular liposomes. Based on these observations, a model for the MVB vesicle formation has been put forward in which the class E Vps proteins, in particular Bro1/ALIX, regulate an LBPA- and pH-driven invagination and budding event of the limiting endosomal membrane. However, the mechanism behind the LBPA-dependent invagination is not known. In the in vitro system, LBPA was incorporated in both leaflets of the membrane bilayer and therefore differences in lipid composition between the inner and outer leaflet cannot account for the LBPA-induced invagination. It has been proposed that the low pH in the lumen of the liposome might result in local assembly of LBPA on the inner leaflet, thereby causing a local asymmetry in the membrane bilayer that induces the necessary membrane curvature (54).
Data from yeast studies are not consistent with the proposed model of LBPA-dependent vesicle formation. First, deletion of the vacuolar ATPase, the protein that is responsible for the formation of a pH gradient across endosomal and vacuolar membranes, does not block the formation of MVB vesicles in yeast (7). Second, the MVB sorting phenotype of a Bro1 deletion strain can be suppressed by overexpression of the deubiquitinating enzyme Doa4, which suggests that the main function of Bro1 is the regulation not of vesicle invagination but of cargo (and/or machinery) deubiquitination (50). Finally, no evidence has been reported yet that the lipid LBPA is produced in yeast. It is possible that slight differences between the yeast and the mammalian MVB sorting systems exist. However, based on the high degree of conservation between the yeast and mammalian ESCRTs, it is unlikely that the two systems differ in key components of the vesicle formation machinery. Aside from the question of whether the ‘LBPA model’ describes accurately the in vivo MVB vesicle formation, the in vitro studies have clearly demonstrated that the proper lipid composition can drive invagination and lumenal vesicle formation. Therefore, local asymmetry in lipid composition between the two membrane bilayers remains a viable model able to describe the mechanism behind the formation of MVB vesicles (Figure 2A).
The topology of the MVB budding event mimics that of viral particle formation, which in the case of HIV-1 and other retroviruses has also been shown to depend on the function of the mammalian class E proteins (55). Although the different viruses show differences in the interaction with the ESCRT machinery, they all seem to recruit at least ESCRT-II, ESCRT-III and Vps4 to the site of virus particle formation. Mutations that inhibit the function of these class E Vps proteins block the virus formation at a late stage, resulting in the accumulation of immature virus particles that remain connected with the plasma membrane (56). This observation suggests that HIV-1 and other retroviruses require the late acting class E proteins for the budding and release of viral particles. Viruses are masters in bypassing regulatory mechanisms and appropriating cellular machinery for their replication. The fact that virus budding requires ESCRT-II, ESCRT-III and Vps4 suggests that these protein complexes have an essential function in final membrane closure and the budding step in the formation of both MVB vesicles and viral particles.
One final interesting point regarding MVB vesicle formation comes from studies in antigen-presenting cells of the immune system that suggest that MVBs might be much more dynamic systems than previously thought. These antigen-presenting cells form specialized MVBs that store MHC class II molecules in lumenal vesicles (57,58). Upon loading of the MHC class II complexes with antigen, the lumenal membranes seem to fuse back with the limiting membrane. The resulting endosomal compartment fuses with the plasma membrane where the active MHC class II molecules are presented. Using electron tomography, a recent study has now demonstrated that the internal membranes containing the MHC class II molecules are indeed vesicles disconnected from the limiting membrane (59). This observation implies that in order to relocate the MHC class II molecules to the plasma membrane, a fusion has to occur between the lumenal vesicles and the limiting membrane of the MVB. Although the MHC class II storage compartments are specialized MVBs, the data demonstrates that fusion of MVB vesicles with the limiting membrane is possible and it might occur in all MVBs. The mechanism of this fusion event remains unknown, as the SNARE-dependent fusion machinery is not present in the endosomal lumen. Furthermore, it is not clear if the fusion of the vesicles with the limiting membrane has a function in the MVB sorting pathway or in endosomal maturation.
Thanks to the large-scale two-hybrid screens and genome-wide analysis of yeast mutants it is likely that most key factors of the MVB sorting machinery have been identified. A lot of information has been collected about the interactions among these factors. The challenge now will be to make the step from this one-dimensional information to the four-dimensional model that can explain the dynamic structure of the MVB sorting system. The proposed models for the vesicle formation and the function of the ESCRTs are the first steps towards this goal.
I would like to thank Tamara Darsow and David J. Katzmann for insightful discussion and critical reading of the manuscript.