To assess whether ESCRT-0 ubiquitination is required for MVB sorting, we generated a Vps27^K>R mutant lacking all 47 lysine residues in the full-length protein (Fig. 1 B). All of the Lys residues were altered because coupled ubiquitination can promiscuously use a variety of lysines (Polo et al., 2002; Miller et al., 2004; Woelk et al., 2006; Uchiki et al., 2009). Ubiquitination of Vps27^K>R was undetectable, even in cells lacking the deubiquitinating enzyme Ubp2, which antagonizes the HECT-type Ub ligase Rsp5 responsible for a wide variety of ubiquitination events within the endocytic pathway (Kee et al., 2005, 2006; Rotin and Kumar, 2009). In contrast, a sizable proportion of wild-type Vps27 was ubiquitinated in ubp2Δ cells because ~20% was found in the higher ~105-kD form (Fig. 1 C). We also found that both wild type and the Vps27^K>R protein were equally stable (Fig. S1 A).

To determine if loss of Vps27 ubiquitination impairs MVB sorting, we examined the localization of Ste3-GFP, a GFP-tagged G protein–coupled receptor normally targeted to MVBs. Cells lacking Vps27 accumulated Ste3-GFP on the vacuole limiting membrane and in endosomes. However, cells with either wild-type Vps27 or Vps27^K>R sorted Ste3-GFP into the vacuole interior (Fig. 1 D). Both Vps27 proteins functioned equally in cells also lacking Ubp2, which implies that although wild-type Vps27 is ubiquitinated to a greater extent in these cells, the MVB sorting process is normal. Under conditions where MVB sorting is less efficient, the Vps27^K>R mutant performed like wild type, which indicates that loss of Vps27 ubiquitination does not increase sorting efficiency (Fig. S1 B).

Yeast ESCRT-0 contains both Vps27 and the STAM homologue Hse1 (Bilodeau et al., 2002). STAM also undergoes ubiquitination (Row et al., 2006), raising the possibility that ubiquitination of Hse1 may regulate ESCRT-0 activity. To test this, we used the SEY6210 parental yeast strain, which requires Vps27 but not Hse1 for MVB sorting (unpublished data and Emr, S.D., personal communication). We found that Ste3-GFP sorting was normal in hse1Δ vps27^K>R cells where all the lysines of ESCRT-0 are eliminated (Fig. 1 D).

Hrs associates with HECT-type Ub ligases such as Nedd4 and AIP4, which promote ubiquitination of Hrs both in vivo and in vitro (Katz et al., 2002; Marchese et al., 2003; Woelk et al., 2006). The yeast homologue of Nedd4 is Rsp5, which interacts with ESCRT-0 (Ren et al., 2007), and is a prime candidate for the ligase that ubiquitinates Vps27. Vps27 ubiquitination was severely attenuated in rsp5-1 temperature-sensitive cells at 37°C to ~10%, the level observed in wild-type cells (loss of Ub-Vps27 was 91% ± 2.7% SD, n = 3; Fig. 2 A). These data indicated that Rsp5 is the major Ub ligase responsible for ESCRT-0 ubiquitination, which suggests that Rsp5 inhibition might be a means of assessing the role of ESCRT ubiquitination.

We next examined MVB sorting of Mup1-GFP in cells with compromised Rsp5 activity. We first examined sorting in cells lacking many of the adaptors Rsp5 requires to modify cargo (Nikko and Pelham, 2009), and in which we found ubiquitination of Mup1-GFP to be severely compromised (Fig. 2 B and S1 C). We also assessed sorting in rsp5-1 cells at 37°C (Fig. 2 C). Either method of inhibiting Rsp5 dramatically blocked sorting of Gap1-GFP and Mup1-GFP, which is consistent with previous studies demonstrating a requirement for Rsp5-mediated ubiquitination (Rotin and Kumar, 2009). However, simply adding Ub in frame restored sorting of Mup1-GFP, indicating that ESCRT-dependent MVB sorting operates normally as long as the requirement for cargo ubiquitination is satisfied.

Previous studies indicate that rsp5-1 cells might still harbor residual Rsp5 activity at 37°C (Kee et al., 2006), perhaps enough to mediate a low level of ESCRT ubiquitination. RSP5 is essential; however; its essential function can be bypassed by expressing a truncated version of the Mga2 transcription factor that undergoes Rsp5- and proteasome-dependent cleavage (Hoppe et al., 2000; Chellappa et al., 2001). We generated rsp5Δ cells, which grew slowly at 30°C and were inviable at 37°C (Fig. S1 D). Sorting of proteins such as Mup1-GFP, Ste3-GFP, and Gap1-GFP along the MVB pathway was completely blocked in rsp5Δ cells. In contrast, fusion of Ub onto the C terminus of either Ste3-GFP or Mup1-GFP restored sorting (Fig. 2 D), which indicates that ESCRT ubiquitination by Rsp5 is not required for MVB sorting.

Rsp5 favors formation of K63-linked polyubiquitin chains (Kee et al., 2005, 2006), and several studies suggest that K63-linked Ub chains, rather than a single mono-Ub, serve as a necessary signal for entry into the MVB sorting pathway (Lauwers et al., 2010). Indeed, cargo fused to a single Ub, which induces MVB sorting, might still undergo additional K63 ubiquitination (Reggiori and Pelham, 2001; Urbanowski and Piper, 2001). However, our data demonstrate that fusion of a single Ub to a cargo protein is sufficient for MVB sorting in cells where Rsp5-dependent K63 polyubiquitination is absent. We further tested the role of K63 Ub chains in cells that were entirely incapable of generating K63 polyubiquitin chains (Fig. 2 E). The sole source of Ub in these cells was Ub^K63R, which was a condition previously shown to cause severe MVB sorting defects for membrane proteins such as Gap1, Jen1, and Sit1 (Erpapazoglou et al., 2008; Lauwers et al., 2009; Paiva et al., 2009). Similarly, we found that Mup1-GFP was localized exclusively to the cell surface and intracellular puncta in Ub^K63R cells. However, Mup1-GFP-Ub^K63R, containing a single in-frame Ub also lacking K63, sorted normally to the vacuole interior. Collectively, these results indicate that a single Ub is sufficient for cargo to sort into the MVB lumen. They also imply that the sorting defects resulting from depriving cells of K63-linked Ub chains may arise from global changes that disrupt trafficking pathways upstream of the MVB or perturb the dynamic cycle of cargo ubiquitination and deubiquitination, from which Mup1-GFP-Ub is immune. In addition, fusion of Ub may route cargo more efficiently to endosomes, circumventing many of the obstacles that other cargos might encounter in Ub-K63–deficient cells during their endocytic itinerary (see supplemental material). Our data clearly agree with previous studies establishing some type of role for K63-linked chains in regulating MVB sorting (Erpapazoglou et al., 2008; Lauwers et al., 2009; Paiva et al., 2009). However, our finding that fusing Ub^K63R is sufficient for MVB sorting suggests that K63-linked chains need not be on cargo themselves. It still remains possible that the sorting of some cargo proteins benefits from the added avidity such chains have for the ESCRT Ub receptors. K63-linked chains may allow cargo to resist the effects of deubiquitinating enzymes as they travel through the endocytic pathway.

The preceding data indicate that neither removing ubiquitinatable lysine residues from ESCRT-0 nor eliminating the major Ub ligase responsible for the majority of ESCRT-0 ubiquitination critically interferes with ESCRT function. However, these approaches have caveats. First, although eliminating Rsp5 dramatically reduces ESCRT-0 ubiquitination, other Ub-ligases might still modify ESCRTs at undetectable levels. This is a distinct possibility because coupled ubiquitination in vitro can be mediated by many different ligases or by E2-conjugating enzymes alone (Hoeller et al., 2007; Uchiki et al., 2009). Second, although we eliminated all of the lysine residues within ESCRT-0, it is possible that another protein associated with the canonical ESCRT-0 may be a relevant Ub acceptor. Finally, Rsp5 is capable of ubiquitinating the N termini of proteins in vitro (Kim and Huibregtse, 2009), whereas other ligases can ubiquitinate cysteine, serine, and threonine residues (Cadwell and Coscoy, 2005; Wang et al., 2007). Thus, the prospect of removing all ubiquitinatable residues within ESCRT-0 and its associated proteins is daunting and leaves no way to control for structural perturbations caused by the mutations themselves. Therefore, we devised a different method to render proteins resistant to ubiquitination. Because in vivo removal of Ub is catalyzed by deubiquitinating peptidases (DUbs), we reasoned that it was possible to generate nonubiquitinatable proteins by fusing them to the catalytic domain of a DUb (Fig. 3 A). We chose three catalytic domains from cysteine-based DUbs that can be inactivated by mutating their catalytic cysteine (Reyes-Turcu et al., 2009): the C-terminal catalytic domain from yeast Ubp7, the UL36 Ub-specific protease domain from type I Herpes virus (Gredmark et al., 2007; Kattenhorn et al., 2005), and the M48 Ub-specific protease domain of murine cytomegalovirus (Schlieker et al., 2007).

When these catalytic domains were fused to GFP-tagged membrane proteins that normally undergo MVB sorting, their localization to the vacuole interior was completely blocked, with localization primarily at the cell surface (Fig. 3 A). The effect on Fur4 and Gap1 was consistent with the strong plasma membrane localization of Fur4 and Gap1 lacking their ubiquitination sites, as described in previous studies (Blondel et al., 2004; Lauwers et al., 2009). Fusion of a DUb domain did not adversely affect the normal localization of Vph1 to the vacuole limiting membrane. In addition, coexpressing DUb-tagged membrane proteins did not affect the sorting of other proteins that normally follow a Ub-dependent pathway to the vacuole interior (Fig. 3 B). As expected, the stability of Ste3-GFP fused to an active DUb was substantially higher than when fused to a catalytically inactive DUb (Figs. 3 C and S1 E). Similarly, fusion of Ste3 to active UL36 catalytic domain abolished detectable ubiquitination (Fig. 3 D). Despite its steady-state localization to the cell surface, we found that Ste3-GFP-DUb was still internalized (Fig. 3 E) because it accumulated in endosomal compartments of rcy1Δ mutants where recycling is blocked (Wiederkehr et al., 2000; Galan et al., 2001). However, Ste3-GFP-DUb did not accumulate in the late endosome–like class E compartments that accumulate in vps4Δ cells (Babst et al., 1997). These data imply that Ste3 internalizes in a Ub-independent manner and that without Ub, it quickly recycles to the cell surface via early endosomal compartments without reaching late endosomal compartments.

Equipped with a new method to immunize proteins against ubiquitination, we reexamined the role of ESCRT ubiquitination (Fig. 4). Fusing DUb catalytic domains to ESCRTs should not only prevent ubiquitination of the ESCRTs themselves, but if suitably robust, the DUb should also strip Ub from cargo proteins and block their entry into MVBs (Fig. 4 A). Moreover, if ESCRT ubiquitination is not required for the sorting process, cargo fused to Ub so that it cannot be removed by DUb activity should be able to sort normally. Fusion of UL36 catalytic domain to ESCRT-0 subunit Hse1 prevented accumulation of ubiquitinated Hse1 and Vps27 (Fig. 4 B). Expression of Hse1-DUb proteins in wild-type cells caused a dramatic block in the sorting of Ste3, Fur4, Gap1, and Mup1, all of which rely on Rsp5-dependent ubiquitination for sorting into the vacuole (Fig. 4 C). Importantly, both Ste3-GFP-Ub and Mup1-GFP-Ub, which contained a single “non-deubiquitinatable” Ub, were properly sorted to the vacuole interior. We also fused DUb catalytic domains to Vps23 and Mvb12, two ESCRT-I subunits thought to act as a downstream Ub cargo receptor and whose mammalian counterparts also undergo ubiquitination (Kim et al., 2007; Tsunematsu et al., 2010). The Vps23-DUb and Mvb12-DUb fusions also interrupted the Ub- and Rsp5-dependent sorting of membrane cargo to the MVB lumen yet did not block sorting of cargo with an in-frame fusion of Ub. All of the MVB sorting defects were abolished when the catalytic activity of the DUb domains was inactivated (Fig. 4 C), and no nonspecific defects in MVB sorting were observed when we expressed catalytic domains fused to His3, a cytosolic protein within the histidine biosynthesis pathway (Struhl and Davis, 1977). All of the DUb fusion proteins were expressed to comparable levels (Fig. S1, F and G), and none caused a vacuolar protein secretion (VPS) phenotype (Fig. S2 A), which is consistent with ESCRT function being maintained. We also examined the effect of fusing a DUb catalytic domain to Gga2, a Ub-binding clathrin adaptor protein thought to sort ubiquitinated proteins from the TGN to endosomes (Scott et al., 2004). GGA proteins may also contribute to Ub-dependent transport of proteins from the cell surface to endosomes in yeast; however, evidence for a clear role in this pathway is obscured because ESCRT-0 also operates in this pathway in parallel (Erpapazoglou et al., 2008; Deng et al., 2009; Lauwers et al., 2009). The dominant-acting Gga2-DUbs dramatically blocked sorting of Gap1, Mup1, and Fur4 but did not affect that of Ste3 to the vacuole, which indicates that yeast Gga proteins may also mediate Ub-specific sorting of a subset of cargos as they travel between the cell surface (Puertollano and Bonifacino, 2004).

The ESCRT-DUb fusion proteins could also functionally substitute for their wild-type counterparts in the MVB sorting process (Fig. 5, A–C). Neither cells with Hse1-UL36 as their sole source of Hse1 nor cells with Vps23-UL36 as their sole source of Vps23 sorted Ste3-GFP or Gap1-GFP into the vacuole lumen. However, both cell types were able to sort Ste3-GFP-Ub and Fur4-GFP-Ub fusion proteins. Mvb12-Ubp7 and Mvb12-UL36 were also able to functionally substitute for wild-type Mvb12 to promote the sorting of Ste3-GFP-Ub, but they effectively blocked sorting of Fur4-GFP (Fig. 5 A). Because loss of Mvb12 results in only mild MVB sorting defects (Curtiss et al., 2007), we assessed Mvb12-DUb function in cells that also had mutations in the UBDs of Vps23 and Vps36, which yield a severe synthetic class E VPS phenotype (Shields et al., 2009). These data indicate that the ESCRT-fused DUb domains are robust enough to deubiquitinate cargo and prevent ubiquitination of ESCRTs, yet are still able to sort Ub-tagged cargo.

We then evaluated whether fusion of Rsp5 to a DUb catalytic domain would create a dominant “reversal” enzyme that could deubiquitinate Rsp5 substrates in vivo. Expressing Rsp5 fused to the catalytic domain of Ubp7 effectively blocked sorting of many Rsp5 substrates (Fig. 5 D). No effect was observed with Rsp5 fused to an inactive catalytic domain (Fig. 5 D) expressed at comparable levels (Fig. S2 B). Expressing Rsp5-DUb also inhibited growth, which is consistent with the essential role of Rsp5 for viability (Fig. S2 C). Although Rsp5-DUb removed Ub from endogenous Rsp5 substrates and blocked their sorting, cargo such as Ste3-GFP-Ub and Mup1-GFP-Ub sorted into the vacuole normally, which supports the idea that their single in-frame Ub is sufficient for their sorting. Furthermore, these data indicate that any deubiquitination of ESCRTs by Rsp5-DUb does not hamper ESCRT sorting activity.