ESCRT ubiquitin-binding domains function cooperatively during MVB cargo sorting

Many integral membrane proteins targeted for lysosomal degradation are ubiquitinated and sorted into vesicles that bud from the limiting membrane into the lumen of endosomes during the biogenesis of multivesicular bodies (MVBs; Piper and Katzmann, 2007). Cargo selection during MVB sorting is dependent on several endosomal protein complexes known as endosomal sorting complexes required for transport (ESCRT). Several ESCRT components contain ubiquitin (Ub)-binding domains (UBDs) that may act as receptors for ubiquitinated membrane proteins. For instance, ESCRT-0, comprised of yeast Vps27 and Hse1 or mammalian Hrs and STAM (signal transducing adaptor molecule), contains Ub-interacting motifs (UIMs) required for MVB sorting of ubiquitinated cargo (Bilodeau et al., 2002; Shih et al., 2002; Piper and Katzmann, 2007). The Ub E2 variant (UEV) domain in Vps23/TSG101, an ESCRT-I component in yeast/mammalian cells, and the GLUE (GRAM-like Ub binding in EAP45) or Npl4 zinc finger (NZF) domains of Vps36, an ESCRT-II component, also contain UBDs that have been implicated in MVB cargo selection (Katzmann et al., 2001; Alam et al., 2004; Slagsvold et al., 2005). This had led to a model in which Ub cargo is passed sequentially from ESCRT-0 to ESCRT-I and to ESCRT-II for final deposit into the forming lumenal vesicles (Katzmann et al., 2002; Gruenberg and Stenmark, 2004; Hurley, 2008). Previous studies have indicated the importance of these UBDs during MVB sorting; however, recent high resolution structures that detail Ub interactions have enabled a more precise approach to dissect the contributions of these UBDs to MVB sorting (Alam et al., 2004, 2006; Sundquist et al., 2004; Teo et al., 2004; Hirano et al., 2006). Surprisingly, we found that mutant forms of Vps23 and Vps36 unable to bind Ub had no defects in MVB sorting, even in combination. This led us to search for additional UBDs within ESCRT-I and resulted in the discovery of a novel UBD within Mvb12. The UBD of Mvb12 works in concert with the UBDs of the ESCRTs to promote cargo recognition and MVB sorting.

Cocrystal structures of Ub with the UEV domains of Vps23 and TSG101 were used to guide mutagenesis experiments rendering Vps23 unable to bind Ub (Sundquist et al., 2004; Teo et al. 2004). The structures show that the UEV domains bind to Ub via a β hairpin tongue and a lip (Fig. 1 A). The structure of the tongue interface is highly conserved and includes residues F₅₂-G₅₇ of Vps23 and Y₄₂-G₄₇ of TSG101. The lip of Vps23 contacts Ub with residues Q₁₁₇ and W₁₂₅, which is different from TSG101 (Pornillos et al., 2002a,b). Two triple-mutant alleles of VPS23 were generated: vps23^ΔUb1 (F₅₂A, Q₁₁₇A, and W₁₂₅A) and vps23^ΔUb2 (F₅₂A, M₁₀₇A, and W₁₂₅A). The corresponding UEV domains were expressed as V5 epitope–tagged proteins in bacteria and used for binding assays with GST-Ub. The wild-type UEV domain of Vps23 bound Ub, whereas no binding was observed for the mutant UEV domains (Fig. 1 B). However, all UEV domains bound to the C terminus of Vps27, which contains two PSDP motifs (P₄₄₈SDP and P₅₂₄SDP) that directly interact with Vps23 (Bilodeau et al., 2003). TSG101 has a similar interaction with PTAP motifs of viral gag proteins and Hrs (Pornillos et al., 2002a,b). Although the structural basis of the TSG101–PTAP interaction is known, it is not known for the Vps23–UEV–PSDP interaction, and the PTAP-binding pocket of TSG101 is not conserved in Vps23. Regardless, this demonstrated that the Vps23 UEV mutants are specifically defective for Ub binding and did not compromise expression of corresponding full-length Vps23 in vivo (Fig. 1 C).

Mutagenesis of the ESCRT-II subunit Vps36 was also performed. The Ub-binding activity of EAP45 (mammalian Vps36) is found within the N-terminal GLUE domain (Alam et al., 2006; Hirano et al., 2006). However, the GLUE domain of yeast Vps36 lacks these key elements of Ub binding. Instead, the yeast Vps36 GLUE domain has an insertion of two NZF domains, the second of which (NZF2) binds to Ub (Fig. 1 D; Alam et al., 2004). Ub binding was abolished by mutating T₁₈₇F₁₈₈ within the Vps36 NZF2 domain and T₁₃F₁₄ in the Npl4 NZF domain (Alam et al., 2004). Fig. 1 E shows the nuclear magnetic resonance (NMR) structure of Npl4-NZF-Ub, a modeled structure of the Vps36-NZF-Ub, and the predicted position of Vps36 residues T₁₈₇F₁₈₈ (Wang et al., 2003). We altered Vps36 T₁₈₇F₁₈₈ to G₁₈₇A₁₈₈ to make the Vps36^ΔUb mutant. Ub binding was assessed in the context of the entire GLUE domain rather than just isolated NZF domains to ensure that Ub-binding activity was abolished. Mutation of Vps36 NZF2 was sufficient to abolish Ub binding by the GLUE domain. GST fusions of wild-type and Vps36^ΔUb GLUE domains were assayed for binding to ¹⁵N-labeled Ub by NMR heteronuclear single quantum coherence (HSQC) experiments (Fig. 1, E and F). The wild-type Vps36 GLUE gave distinct chemical shift changes in Ub. In contrast, the Vps36^ΔUb GLUE domain, at a sixfold higher concentration, showed no chemical shift changes in Ub. V5 epitope–tagged wild-type and ΔUb GLUE domains of Vps36 were also used in binding assays with Ub-GST (Fig. 1 G). The wild-type Vps36 GLUE domain bound to Ub, but the Vps36^ΔUb GLUE domain did not. The ΔUb mutation was then placed into the context of full-length Vps36 to create vps36^ΔUb. The mutations in Vps36 did not appear to have a deleterious effect on expression level because HA epitope–tagged alleles of VPS36 and vps36^ΔUb expressed from the endogenous VPS36 promoter showed similar levels (Fig. 1 H).

To assess the role of Ub binding by Vps23 and Vps36, yeast strains were constructed in which the mutant alleles were integrated in place of the endogenous genes. We found that the inability of Vps23 or Vps36 to bind Ub had no effect on the sorting of GFP-Cps1, a model MVB cargo (Odorizzi et al., 1998), into the vacuolar lumen (Fig. 2 A). No difference was observed between the vps23^ΔUb1 or vps23^ΔUb2 alleles (unpublished data). Analysis of additional mutant vps23 alleles with mutations in other residues that line the Ub interface (S₅₆, G₅₇, F₁₀₅, and N₁₂₃) also did not display any sorting defects (Fig. S1). Likewise, a double mutant containing vps23^ΔUb1 and vps36^ΔUb showed normal sorting of GFP-Cps1 (Fig. 2 B). These results were confirmed using vps23Δ and vps36Δ-null mutants transformed with wild-type and ΔUb alleles of VPS23 and VPS36 housed on low copy plasmids (unpublished data). Lastly, sorting of other MVB cargoes, including Ste3-GFP (Urbanowski and Piper, 2001), Fur4-GFP (Blondel et al., 2004), and Sna3-GFP (Reggiori and Pelham, 2001) was normal in the vps23^ΔUb1 and vps36^ΔUb alleles (unpublished data).

These findings differ from previous studies that found severe sorting defects upon mutation of the ESCRT-I UEV domain and the ESCRT-II NZF domain (Katzmann et al., 2001; Alam et al., 2004). Prior analysis of the UEV domain centered on a single mutant (M₈₅T) identified by a loss of function mutagenesis screen. M₈₅ faces the interior of Vps23 and not the Ub-binding interface (Fig. 1 A). Thus, mutation of this residue may indirectly impact the ability of the UEV domain to bind to Ub or other factors. The NZF mutation used in this study was also distinct from previously described mutations that displayed MVB sorting defects (Alam et al., 2004). In both cases, mutations used in this study may maintain the overall integrity of Vps23 and Vps36 (based on in vivo function), although specifically ablating Ub binding (based on in vitro assays). Additionally, our functional analysis was performed with untagged, integrated alleles of VPS23 and VPS36. The ability of the vps23^ΔUb vps36^ΔUb double mutant to successfully sort cargo into the MVB interior indicated that there might be additional UBDs within the ESCRT machinery.

A potential candidate for a protein with a novel UBD was Mvb12, a recently identified subunit of ESCRT-I. Previous studies showed that deletion of MVB12 caused differential defects in MVB sorting; Ste3-GFP and Sna3-GFP display sorting defects, whereas GFP-Cps1, Ste2-GFP, and Ste3-GFP–Ub (in which Ub is fused in frame to the C terminus) sorting is largely unaffected (Chu et al., 2006; Curtiss et al., 2007; Oestreich et al., 2007). This indicates that Mvb12 plays a specific role in recognizing a subset of MVB cargoes. A role for Mvb12 cargo recognition was tested with strains that combined vps23^ΔUb1, vps36^Δub, and mvb12Δ to probe for synthetic defects in MVB sorting. Combining these mutations caused dramatic sorting defects in GFP-Cps1 and Ste3-GFP (Fig. 2 B). The defect in sorting of GFP-Cps1 is not the result of a detectable loss of Cps1 ubiquitination (Fig. 2 C) but rather indicates a defect in recognition of Ub cargos. This was supported by examining the sorting of Ste3-GFP–Ub, which was normal in mvb12Δ-null and vps23^ΔUb1 vps36^ΔUb double-mutant cells but defective in vps23^ΔUb1 vps36^ΔUb mvb12Δ triple-mutant cells (Fig. 2 B).

The genetic interactions between Mvb12 and ESCRT-I and -II UBDs prompted us to determine whether Mvb12 could bind Ub. Recombinant V5 epitope–tagged full-length Mvb12 from Saccharomyces cerevisiae and the related budding yeast Saccharomyces kluyveri were used in binding assays with Ub-GST. Both proteins bound specifically to Ub-GST (Fig. 3 A). Deletion analysis of Mvb12 mapped the Ub binding to the C terminus (W₅₀-S₁₀₁; unpublished data). This interaction was confirmed by NMR HSQC experiments of ¹⁵N-labeled Ub combined with the C terminus of Mvb12. Mvb12 induced chemical shift perturbations in the spectrum of ¹⁵N-labeled Ub that mapped to a surface of Ub that included residues R₄₂, V₇₀, and L₈ (Fig. 3 E). This overlaps with the surface engaged by the Vps23 UEV domain, the Vps36 NZF domain, and the UIM domains of Vps27 and Hse1, indicating that simultaneous binding of these components to a single Ub molecule is unlikely (Bilodeau et al., 2003; Alam et al., 2004; Teo et al., 2004).

The C-terminal fragment of Mvb12 was then mutated to produce a series of recombinant proteins that were subjected to Ub-GST–binding experiments. Mutations that disrupted Ub binding were localized to residues F₇₁-G₈₄ as shown by two mutants with residues 71–76 and 79–84 replaced with alanines (Fig. 3 B). A recent crystal structure has shown that two patches of Mvb12, F₇₁D₇₂, and W₇₄Y₇₅ pack against a hydrophobic segment of Vps37 (Kostelansky et al., 2007). However, these residues did not seem critical for Ub binding, as mutation of W₇₄Y to R₇₄S, and F₇₁D to S₇₁N did not block binding to Ub-GST (Fig. 3 B). In contrast, mutation of residues C terminal to this region abolished Ub binding. Interestingly, in the context of the entire ESCRT-I, the UBD of Mvb12 would be oriented near the UEV domain of Vps23, suggesting that both of these UBDs are close in native ESCRT-I.

We initially assessed the function of wild-type and the two Mvb12 Ub-binding mutants as HA epitope–tagged proteins expressed from low copy plasmids containing the native MVB12 promoter. All three proteins expressed at comparable levels (Fig. 3 D), and all coprecipitated with Vps23 and Vps28, demonstrating that they properly assembled with ESCRT-I (Fig. 4 A). Gel filtration showed that tandem affinity purification (TAP)–tagged Mvb12^ΔUb1 (Y₇₉-G₈₅>A₇₉AAAAA) supported assembly into a large complex with an apparent molecular mass of 350 kD, as has been previously observed for ESCRT-I (Fig. S2). Furthermore, GFP-tagged Mvb12^ΔUb1 localized to endosomal compartments similar to GFP-tagged wild-type Mvb12 (Fig. S2). These data indicate that these Mvb12 mutants were able to assemble into the ESCRT-I complex. This allowed us to ascribe phenotypes to a deficiency in Ub binding of Mvb12 rather than loss of ESCRT-I association.

Cells expressing mvb12^ΔUb1 or mvb12^ΔUb2 (K₈₂KPPG>D₈₂DGGT) as their sole source of MVB12 showed no sorting defects for Ste3-GFP or Sna3-GFP (Fig. 4 B). Agreeing with previous studies, Ste3-GFP and to large extent Sna3-GFP showed defective sorting in mvb12Δ-null cells (Fig. 4 C; Chu et al., 2006; Curtiss et al., 2007; Oestreich et al., 2007). Other MVB cargo proteins such as GFP-Cps1 and Ste3-GFP–Ub were normal (unpublished data). The same results were obtained with wild-type or chromosomally integrated mvb12^ΔUb mutants (unpublished data).

Together, these data indicate that the mutations used to ablate Ub binding did not indiscriminately compromise Mvb12 function. They also implied that the function of Ub binding by Mvb12 alone is not critical for MVB sorting. We next assessed the role of the Mvb12 UBD in cells lacking the UBDs of Vps23 and Vps36 (Fig. 4 C). The vps23^ΔUb1 vps36^ΔUb mvb12Δ triple mutant was cotransformed with low copy plasmids expressing wild-type MVB12, mvb12^ΔUb1, or mvb12^ΔUb2 and GFP-tagged MVB cargos. We found that wild-type MVB12 conferred proper sorting of all cargos, whereas both mvb12^ΔUb alleles showed defective sorting of GFP-Cps1 when combined with vps23^ΔUb and vps36^ΔUb. These data show that the UBDs of Mvb12, Vps23, and Vps36 work jointly to efficiently sort MVB cargo because sorting was defective only when mutations were combined (Fig. 2 B and Fig. 4 C).

In contrast to the sorting of GFP-Cps1, other MVB cargos were sorted normally in the vps23^ΔUb vps36^ΔUb mvb12^ΔUb triple mutants. These cargos included Sna3-GFP and Ste3-GFP (Fig. 4 C). This demonstrated that some degree of Ub-dependent MVB sorting was still intact and that there might be other ESCRT components that contribute to MVB cargo recognition; one candidate is Vps27–Hse1 (ESCRT-0). The Vps27–Hse1 complex contains three UIMs and binds the Vps23 UEV domain via its two PSDP motifs within the C terminus (P₄₄₈SDP and P₅₂₄SDP). Although deletion of VPS27 altogether blocks ESCRT-I localization to endosomes (Katzmann et al., 2003), little phenotype is observed when the protein–protein interface between ESCRT-0 and ESCRT-I is disrupted by mutation of the two PSDP motifs in Vps27 (Vps27^ΔVps23, AA₄₄₈ADP and P₅₂₄AAA; Bilodeau et al., 2003). Localization of Vps23-GFP to endosomes was unperturbed as was recruitment of mCherry-tagged ESCRT-III subunit Vps20 (Fig. S1). To investigate whether the interaction with ESCRT-0 helped maintain sorting of MVB cargos in the vps23^ΔUb1 vps36^ΔUb mvb12^ΔUb strain, these mutations were combined with the vps27^ΔVps23 mutation. The resulting quadruple mutant displayed very significant sorting defects (Fig. 5 A). Sorting of all MVB cargoes was defective in vps23^ΔUb1 vps36^ΔUb vps27^ΔVps23 mvb12^ΔUb mutant cells. This included Ste-GFP, Ste3-GFP–Ub, and Sna3-GFP, which otherwise sorted normally in vps23^ΔUb1 vps36^ΔUb mvb12^ΔUb mutant cells (Fig. 4 C) or vps23^ΔUb vps36^ΔUb vps27^ΔVps23 cells (Fig. 5 A). These experiments clearly showed a critical contribution of the Mvb12 UBD because dramatic sorting defects were observed when the mvb12^ΔUb vps23^ΔUb vps36^ΔUb and vps27^ΔVps23 alleles were combined. In spite of missorting MVB cargoes, the quadruple mutant cells were still capable of a residual level of sorting lipids into the interior lumenal membranes, indicated by sorting of the fluorescent lipid marker NBD-PC into the vacuole lumen. Unlike ESCRT-null mutants in which NBP-PC accumulates exclusively in endosomes (Bilodeau et al., 2002), NDB-PC accumulated within the vacuoles of vps23^ΔUb vps36^ΔUb vps27^ΔVps23 mvb12^ΔUb cells (Fig. 5 B). Disruption of the Vps27–Vps23 interaction combined with mutation of the UBDs within ESCRT-I and -II revealed a separation of cargo sorting and MVB biogenesis.

Our data show that Ub binding by Mvb12 is functionally important and contributes to the overall process of protein sorting into MVBs. Moreover, models proposed previously that require ubiquitinated cargo to be sequentially recognized by the UIMs of ESCRT-0, the Vps23 UEV domain of ESCRT-I, and the Vps36 component of ESCRT-II need to be amended. The available data support several functional models (Fig. 5 C) that can account for the multiple UBDs with the ESCRT network (Fig. 5 D). The one we favor is that ESCRTs function together as a large supercomplex that contains multiple UBDs, each of which can recognize and sort partially overlapping sets of ubiquitinated cargo. Multiple UBDs within this complex would enable recognition of Ub attached to a wide variety of membrane proteins or allow for multiple cargo molecules per supercomplex. If this model were correct, then the observation that sorting of some cargo still occurs in the absence of Vps23, Vps36, and Mvb12 UBDs implies that there remain other UBDs within the ESCRT machinery. One possibility is that remaining UBDs are provided by components such as Vps27–Hse1 because disruption of the Vps27–Vps23 interface (the vps27^ΔVps23 mutation) caused a dramatic synthetic phenotype when combined with UBD mutations in Vps23, Vps36, and Mvb12. Alternatively, the UBDs of ESCRT-I and -II may perform other functions besides cargo recognition. In particular, coupled monoubiquitination of the ESCRT machinery itself, which has been observed in animal cells but not yet yeast, could cause UBDs to mediate a series of intra- and intermolecular interactions that could regulate assembly or efficiency of MVB sorting (Haglund and Stenmark, 2006; Hoeller et al., 2006). Indeed, the loss of UBDs in combination with the synthetic mutations that led to profound MVB-sorting defects also caused defects in vacuolar protein sorting (e.g., carboxypeptidase Y [CPY] secretion; Fig. S2), suggesting that UBDs may contribute to the general functions of the ESCRTs in a way that remains to be determined.

Chemicals, antibodies, and, growth methods were used as previously described (Bilodeau et al., 2003; Ren et al., 2007). S. cerevisiae strains used in this study are listed in Table I. Plasmids used in this study are listed in Table II. VPS36 was disrupted by inserting the HIS3 gene in place of codons 101–539 of VPS36. Wild-type or vps36^ΔUb (T₁₈₇F/G₁₈₈A) cloned into the integrating TRP1 plasmid pRS304 (pPL2646 and pPL2645, respectively) was linearized with ClaI and integrated at the vps36Δ∷HIS3 locus to make PLY3394 and PLY3395. The vps23Δ∷Kanr, mvb12Δ∷Kanr, and vps27Δ∷Kanr disruptions were made by amplifying the respective loci from the BY4742 deletion collection (Winzeler et al., 1999) and transforming the PCR product into the appropriate strain. Vps27^ΔVps23 was made by integrating linearized integration plasmid pLP2251 digested with BlgII into the URA3 loci of strains containing the vps27Δ∷LEU2 disruption. The vps23^ΔUb alleles were made by linearizing integrating pPL2940 into the VPS23 loci of PLY3405 and PLY3407 and looping out the insertion of the URA3 gene by selection of 5-FOA.

Unmarked disruption of MVB12 was performed by replacing the MVB12 ORF with LEU2 and looping out with Cre recombinase as previously described (Gueldener et al., 2002).The vps27Δ∷LEU2 disruption was made by replacing codons 177–622 with LEU2. MVB12 or mvb12 alleles were at the URA3 locus by transforming yeast with pPL3780, pPL3781, or pPL3784 linearized with StuI. HA-tagged Vps36 was made by integrating a triple HA epitope at the Vps36 C terminus made by homologous recombination with HA-containing host plasmid. Vps20-mCherry was made by amplifying the Vps20 ORF and 500 bp of the promoter from the genome and homologous recombining it with an mCherry-containing host plasmid.

HA epitope–tagged alleles of MVB12 were made by inserting a single HA tag at the C terminus of MVB12 followed by the PHO8 3′ untranslated region. Expression was driven by the MVB12 promoter (500 bp region upstream of the MVB12 ORF). Untagged MVB12 alleles were made similarly but lacked the C-terminal tag. TAP- and GFP-tagged MVB12 alleles were made as previously described (Oestreich et al., 2007).

GST-fusion proteins were isolated from bacteria using glutathione-Sepharose beads as previously described (Smith and Johnson, 1988). For pull-down assays, 250 µg of each isolated GST fusion protein was bound to 50 µl glutathione-Sepharose in PBS by rotation for 30 min at 25°C. Bound GST or GST fusion proteins were pelleted and washed three times with PBS. Cell lysate was added to each protein/bead complex and incubated for 2 h at room temperature. Unbound proteins were removed from beads using three washes of cold PBS eluted with 50 µl 50 mM glutathione in PBS, pH 7.2, and the bound bead fraction was analyzed by SDS-PAGE and immunoblotting.

Cells containing GFP-expressing plasmids were grown in synthetic dextrose minimal media to mid-log phase, resuspended in 0.2% NaN3 and 100 mM Tris, pH 8.0, and viewed using a 100× 1.4 NA aperture on a microscope (BX-60; Olympus) equipped with fluorescein isothiocyanate filters and Nomarski/differential interference contrast (DIC) optics at ambient temperature. Images were captured with a charge-coupled device camera (ORCA; Hamamatsu Photonics) with IPLABS software (BD) as previously described (Urbanowski and Piper, 1999). Image processing was performed in PhotoShop (CS; Adobe), where the original 12-bit images were converted to 8-bit images and the brightest pixel was set to 255. NBD-PC (1-myristoyl-2-[6-[(7-nitro-2-1,3-benzoxadiazol-4-yl)amino]hexanoyl]-sn-glycero-3-phosphocholine; Avanti Polar Lipids, Inc.) was used as previously described (Bilodeau et al., 2002).

Recombinant proteins mixed with ¹⁵N-labeled Ub were prepared as previously described (Bilodeau et al., 2004) or purchased from Spectra Stable Isotopes. [¹⁵N]Ub HSQC spectra were collected on a 500-MHz spectrometer (Avance II US2; Bruker) and analyzed with SPARKY (T.D. Goddard and D.G. Kneller, SPARKY 3, University of California, San Francisco, San Francisco, CA) as previously described (Bilodeau et al., 2004). Chemical shift differences were calculated using the formula (0.2 δN² + δH²)^1/2. Binding experiments with Vps36 were performed in 50 mM NaPO₄, pH 7.2; binding experiments with Mvb12 were performed in 50 mM imidazole and 100 mM NaCl, pH 7.2. A spectrum of unbound ¹⁵N-labeled Ub in the same buffer was used as the reference.

Metabolic [³⁵S]methionine labeling and immunoprecipitation of CPY were performed as described previously (Cooper and Stevens, 1996). Preparation of yeast lysates for gel filtration, immunoblotting, and CPY pulse-chase analysis were performed as described previously (Oestreich et al., 2007). Protein models were drawn with the PyMOL molecular graphics system (http://pymol.sourceforge.net). A model for the Vps36 NZF domain was computed using the Wurst protein server (Torda et al., 2004).

Fig. S1 shows the localization of GFP-Cps1 in strains with a range of mutations in Vps23, the structure of Vps23 UEV domain in complex with Ub highlighting mutated residues, the localization of Vps23-GFP, and Vps20-mCherry in strains containing vps27^Δvps23 mutations. Fig. S2 shows the localization of Mvb12-GFP, gel filtration experiments performed on wild-type and mutant Mvb12, and the results of CPY pulse-chase analysis of all of the strains used in this study.