Structural insight into the ESCRT-I/-II link and its role in MVB trafficking

Yeast ESCRT-I binds directly to ESCRT-II with nanomolar affinity

To determine the affinity of the ESCRT-I/-II interaction, we used an N-terminally FlAsH-tagged Vps28 C-terminus (residues 148–242) (FlAsH-Vps28-CT) (Adams et al, 2002). The isolated FlAsH-Vps28-CT, a small 12 kDa construct, has a fluorescence anisotropy of 0.19. Upon binding of FlAsH-Vps28-CT to the 140 kDa full-length ESCRT-II complex, the fluorescence anisotropy increases to 0.27. Direct titration with intact ESCRT-II shows that the isolated FlAsH-Vps28-CT binds ESCRT-II with a K_d of approximately 27 nM (Figure 1A). A similar K_d (57 nM) was determined for the direct titration of a Vps36 fragment containing the double NZF insertion (residues 110–205, encompassing both the N- and C-terminal NZF domains, Vps36-NZF-NC) (Figure 1A). This is in good agreement with an affinity of 10 nM for Vps28-CT binding to GST-Vps36-NZF-NC as determined using BIAcore analysis (data not shown). Therefore, the yeast-specific insertion present within Vps36 of ESCRT-II is sufficient for the high-affinity interaction we observe with the C-terminus of Vps28 in ESCRT-I.

To measure the affinity of intact ESCRT-I for intact ESCRT-II, we used a competition assay, where we monitored dissociation of FlAsH-Vps28-CT from ESCRT-II by titration of heterotrimeric ESCRT-I(Δ21-Vps37) (an engineered complex missing 21 residues in the N-terminus of the Vps37 subunit) (Teo et al, 2006). Competition experiments show that non-labelled ESCRT-I(Δ21-Vps37) and non-labelled Vps28-CT have similar K_d values (44 and 53 nM, respectively) for ESCRT-II (Figure 1B). In addition, a competition experiment showed that a non-labelled Vps36-NZF-NC also gives a similar K_d (32 nM) (data not shown). These results show that the yeast Vps28-CT in ESCRT-I and the yeast-specific insertion of Vps36 within ESCRT-II are both necessary and sufficient for high-affinity ESCRT-I and ESCRT-II complex formation in vitro.

Crystallisation of the yeast ESCRT-I/-II link

To structurally characterise the yeast ESCRT-I/-II ‘supercomplex', we focused on the minimal domains required for these two large complexes to form a stable interaction in vitro. We generated subcomplexes of Vps28-CT (residues 148–242) with various Vps36-GLUE constructs (Vps36-GLUE domain (residues 1–289), Vps36 yeast-specific insertion (residues 110–205) and a minimal Vps36-NZF-N domain (residues 110–151)) (Figure 1C). For crystallisation, all the subcomplexes tested could be copurified by gel filtration except for a Vps28-CT/minimal Vps36-NZF-N complex. The interaction between these two fragments could be easily detected by GST-pulldown analysis, but the two proteins did not copurify on gel filtration, and therefore were mixed for crystallisation trials.

Because initial co-crystallisation experiments using the Vps28-CT/minimal Vps36-NZF-N complex failed, we re-examined the domain boundaries for the Vps36-NZF-N that were originally derived from the structure of the homologous NZF domain from Np14 (Wang et al, 2003). As the minimal Vps36-NZF-N domain showed only low-affinity binding to ESCRT-I, in comparison to full-length ESCRT-II, we tested whether a C-terminally extended construct (residues 110–176, including conserved hydrophobic residues in the linker region between NZF-N and NZF-C) would bind with higher affinity. Coexpression of Vps28-CT with this extended Vps36-NZF-N resulted in a stable complex that could be copurified by gel filtration. The extended Vps36-NZF-N binds to FlAsH-Vps28-CT with a K_d of 61 nM, confirming that only the extended Vps36-NZF-N and not the Vps36-NZF-C domain interacts with ESCRT-I (Figure 1A).

This Vps28-CT/extended Vps36-NZF-N subcomplex crystallised readily in space group C222₁ with unit cell dimensions a=66.6 Å, b=100.0 Å and c=115.6 Å. The 2.0 Å resolution structure was determined with a single anomalous dispersion experiment using the intrinsic Zn²⁺ bound to the protein. The data collection, phasing and refinement statistics are given in Table I.

The structure of the yeast ESCRT-I/-II link

There are two copies of Vps28-CT/extended Vps36-NZF-N complex in the asymmetric unit. The entire Vps28-CT domain is a well-ordered four-helix bundle with an up-down-up-down topology with each helix packing in an antiparallel fashion with the helix immediately preceding it in the sequence (Figure 2). The first two helices have a similar length and tightly associate to form a straight helical hairpin connected by a short loop. The second two helices form a more splayed helical hairpin, where the shorter helix α3 packs against the longer helix α4 with an angle of −154°. The α3/α4 loop is the longest loop in the four-helix bundle and is well ordered, in part owing to hydrogen bonding with the adjacent α1/α2 loop. The two helical hairpins associate with each other around an extensive hydrophobic core. There are essentially no differences between the 2.0 Å resolution Vps28-CT and the 3.05 Å resolution structure that was recently reported for unliganded yeast Vps28-CT (Pineda-Molina et al, 2006). Sequence alignment of fungal and metazoan ESCRT-I Vps28-CT domains is shown in Figure 3A.

The Vps36-NZF-N domain binds to the α2/α3 face of the Vps28-CT helical bundle (Figure 2A). Residues 114–162 of the Vps36-NZF-N domain are ordered. The fold of the Vps36-NZF-N domain is a classical zinc-ribbon fold with a left-handed coordination of the Zn²⁺ ion between two loops containing CXXC motifs that are often referred to as rubredoxin knuckles (Krishna et al, 2003). The Vps36-NZF-N is a β-sandwich consisting of a two-stranded β-sheet and a three-stranded β-sheet. The three-stranded β-sheet (β1/β2/β5 sheet, lower sheet in Figure 2B) packs orthogonally against the two-stranded β-sheet (β3/β4 sheet, upper sheet). The β1/β2 and β3/β4 loops form the primary contact with the Vps28-CT four-helix bundle. The interaction between the β1/β2 and β3/β4 loops is stabilised by a tetrahedral coordination with a single Zn²⁺ ion involving two Cys residues from each of the loops. These Cys residues are part of two rubredoxin knuckles (120–123 and 143–146). Residues 110–151 of the Vps36-NZF-N domain have a core fold similar to the NZF domain from Npl4 (Wang et al, 2003); however, additional residues in the Vps36-NZF-N domain form a single-turn α-helix and an additional β-strand, which pairs in an antiparallel fashion to strand β1 making a three-stranded β-sheet to complete the overall fold of the Vps36-NZF-N domain. The resulting five-stranded β-sandwich has a fold that is similar to the fold of the iron-sulfur protein rubredoxin.

A schematic of the binding interface is shown in Figure 2B and detailed interactions are highlighted in Figure 2C. The primary interaction site is hydrophobic; the invariant Ile122^Vps36 and highly conserved Val148^Vps36 of the two hairpin loops in the Vps36-NZF-N domain slot into a hydrophobic pocket (Figure 2B and C) formed between the splayed helices α2/α3 on the Vps28-CT. Flanking this primary hydrophobic interface are additional polar and hydrophobic contacts. His178^Vps28 and Asn210^Vps28 form hydrogen bonds with the main-chain carbonyls of Cys146^Vps36 and Asn145^Vps36, respectively. On the other flanking side, Arg190^Vps28 forms a hydrogen bond with Asp151^Vps36. A secondary hydrophobic interface is formed by the highly conserved Leu154^Vps36 slotting into a small pocket lined by the aliphatic portions of the side chains of Arg190^Vps28 and Arg193^Vps28.

Given the 61 nM affinity for this interaction, the total surface area buried by complex formation is surprisingly low, 1008 Ų. To verify the functional importance of the interactions observed in the structure of the complex, we mutated a set of interacting residues in the Vps36-NZF-N domain and determined the affinity of the mutants for FlAsH-Vps28-CT (Table II). Mutations of residues within the central hydrophobic interface in Vps36-NZF-N have the greatest impact on binding to Vps28-CT. The I122D mutation reduced the binding over 8000-fold, and the double mutant I122D/V148D has no detectable binding to FlAsH-Vps28-CT. However, single mutations of flanking residues Asp151 or Leu154 resulted in Vps36 NZF-N domains that still retained moderate binding affinity (0.4 and 3.9 μM, respectively).

ESCRT-I/-II coupling is essential for efficient MVB cargo sorting

The functional effect of mutations that abolish the ESCRT-I/-II link in vitro was investigated in yeast using a GFP-CPS vacuolar-sorting assay. Both a double mutant (Vps36^I122D/V148D) and quadruple mutant (Vps36^{I122D/V148D/D151R/L154R}) when expressed in a yeast strain lacking Vps36 (vps36Δ) resulted in GFP-CPS localisation to the limiting membrane of the vacuole and to a class E compartment rather than localisation in the vacuolar lumen as observed for the wild-type Vps36 (Figure 2D). This class E phenotype observed in the ESCRT-I/-II link mutants is identical to the vps36Δ mutant. Therefore, the ESCRT-I/-II link is essential for efficient cargo sorting.

Structural alignment of Vps36-NZF-N and Npl4-NZF domains reveals a core fold with non-overlapping binding sites

NZF are a widespread family of Zn-binding modules, many of which function in recognising Ub using their rubredoxin knuckles. The NMR structure of the rat Npl4-NZF domain in a complex with Ub identified a critical Ub-binding motif, C-X₂-CTF-X₈-C-X-φ-C, where φ is a bulky hydrophobic residue (Alam et al, 2004). ESCRT-II Vps36 subunit contains two tandem NZF domains: Vps36-NZF-N, which binds to Vps28-CT with nanomolar affinity, and Vps36-NZF-C, which binds Ub with micromolar affinity (Alam et al, 2004). As the Vps36-NZF-N and Vps36-NZF-C domains appear to be highly specific for their respective binding partners despite having similar folds, we wanted to examine the structural basis for this selectivity.

Alignment and superimposition of the Vps36-NZF-N structure on the Npl4-NZF is shown in Figure 3B and C. The hydrophobic core formed by the packing of two orthogonal β-sheets and the conformation of the rubredoxin knuckles are essentially identical in both the Vps36-NZF-N and Npl4 NZF structures (Figure 3C). However, two major structural differences are apparent. Firstly, the NZF-N domain contains an additional α-helix (α1) and a β-strand (β5), with the strand completing the lower β1/β2/β5 sheet (Figure 3C). The extended C-terminus is essential for generating nanomolar binding affinity (mutation of Leu154^Vps36 in this region decreases the affinity more than 60-fold) (Table II) as opposed to the weaker interaction formed by Vps36-NZF-C (encompassing only the NZF core fold) for Ub (K_d 180 μM) (Alam et al, 2004). The second striking difference between the Vps36-NZF-N and other Npl4-family zinc-fingers is that the β2/β3 loop in Vps36-NZF-N is much longer (Figure 3C). It is unlikely that this extended loop in Vps36-NZF-N has a functional role, as it is distant from the Vps28-CT-binding site and not conserved among fungi (Figure 3B).

Mapping the binding interfaces from the Vps36-NZF-C/Ub and Vps36-NZF-N/Vps28-CT complexes on the Vps36-NZF-N domain shows clearly that the binding selectivity arises from two distinct binding sites on the same fold (Figure 3D). It is apparent from sequence alignment of Vps36-NZF-N and Vps36-NZF-C with the Npl4-NZF domain that each NZF domain in the yeast Vps36 subunit of ESCRT-II is adapted for only one of these two potential binding partners (Figure 3B). We have identified a critical motif in Vps36-NZF-N (Ile122/Val148/Leu154) that is required for Vps28-CT binding. The corresponding residues in the Saccharomyces cerevisiae Vps36-NZF-C domain are Ala185, His202 and Lys209. The presence of the charged His149 in the place of valine is a substitution similar to the V148D mutation that reduced the affinity of Vps36-NZF-N for Vps28-CT by more than 8000-fold (Table II), suggesting why Vps36-NZF-C does not bind to Vps28-CT. Conversely, Vps36-NZF-N lacks the critical Ub-binding motif, with the 187-TF-X₁₀-M-199 of Vps36-NZF-C aligning to 124-MV-X₁₀-N-145 in Vps36-NZF-N. The NZF Ub-binding interface has been shown to be sensitive to such minor changes, where mutation of residues 141-TF-142 in Vps36-NZF-C into 141-GS-142 completely abolishes Ub binding (Alam et al, 2004).

The effect of Vps36-NZF-N binding on the structure of Vps28-CT

The isolated yeast Vps28-CT crystallised readily in a high-salt crystallisation condition that dissociated the Vps28-CT/Vps36-NZF-NC complex. This enabled us to compare the unliganded Vps28-CT with the Vps36-bound form. Overall, the 2.0 Å resolution structure of the isolated Vps28-CT domain is essentially identical to Vps28-CT in a complex with Vps36-NZF-N (0.5 Å RMSD for 94 Cα atoms) (Figure 4A, left panel; see Table I for crystallographic data collection and refinement statistics). Conformational change is restricted to side chains involved in the Vps36-NZF-N binding interface. The orientation of Arg193^Vps28 changes and Arg190^Vps28 becomes ordered on Vps28-CT binding to the Vps36-NZF-N domain, and this minor conformational change acts to provide flanking interactions that are important for generating the nanomolar binding affinity. The rest of the binding interface is rigid and remains unchanged in the non-bound and bound crystal structures, suggesting that the high surface complementarity with the Vps36-NZF-N knuckles is important for generating the high-affinity interaction despite the low contact surface area.

Structural comparison of the isolated yeast and metazoan Vps28-CT

We also determined a 1.3 Å resolution structure of the Xenopus laevis Vps28-CT domain. The isolated X. laevis Vps28-CT domain forms a four-helix bundle that is very similar to the yeast domain (1.2 Å RMSD for 90 Cα atoms) (Figure 4A, right panel). Despite this structural conservation, the X. laevis Vps28-CT does not bind to the yeast ESCRT-II (data not shown). Examining the X. laevis Vps28-CT surface corresponding to the Vps36-NZF-N interface highlights that the central hydrophobic pocket in yeast Vps28 has been disrupted by the protrusion of two charged residues that are highly conserved in metazoans (Arg158 and Glu162 in X. laevis, corresponding to Ala182 and Ile186 in yeast Vps28) (Figure 3A). These charged residues would disrupt the hydrophobic patch essential for Vps36-NZF-N binding.

Interestingly, a highly conserved surface patch is seen on the opposite face of the Vps28-CT four-helix bundle as compared to the yeast Vps36-NZF-N-binding site (Figure 4B). Adjacent to this most conserved region is a solvent-exposed phenylalanine (Phe228) flanked on either side by acidic residues, Asp229 and Glu231. Recently, this region was implicated in binding the ESCRT-III subunit Vps20, as mutation of the yeast sequence 228-FDLE-231 to 228-ASLA-231 abolishes the interaction of Vps28-CT with yeast Vps20 in vitro (Pineda-Molina et al, 2006). Further investigation of the functional role of this larger conserved patch in ESCRT-I is required.

ESCRT-I binds directly to a new subunit, Mvb12, forming a stable 1:1:1:1 heterotetrameric complex

We have previously shown that yeast ESCRT-I(Δ21-Vps37), a heterotrimeric complex composed of Vps28, Vps23 and Δ21-Vps37, forms a 1:1:1 complex in vitro (Teo et al, 2006) (Δ21-Vps37 is a deletion variant of S. cerevisiae Vps37, missing residues 1–21 that are prone to proteolysis in Escherichia coli and which significantly promote ESCRT-I aggregation). However, a recent S. cerevisiae proteome-wide screen of protein complexes identified a new putative ESCRT-I-binding protein, YGR206W (systematic yeast ORF designation) (Krogan et al, 2006). This 12 kDa protein of unknown function localises to endosomes in vivo, making it an enticing target for further investigation (Huh et al, 2003) (http://yeastgfp.ucsf/getOrf.php?orf=YGR206W). Independently, YGR206W was recently identified as a component of ESCRT-I and renamed Mvb12 (multivesicular body sorting factor of 12 kDa), a nomenclature we have adopted (Chu et al, 2006).

Gel filtration analysis shows that purified recombinant Mvb12 binds directly to recombinant ESCRT-I(Δ21-Vps37) to form a stable heterotetrameric complex (Figure 5A). Similarly, Mvb12 (with an N-terminal MAH₆ tag) and full-length ESCRT-I (untagged) can be readily copurified when the four components are coexpressed in E. coli (Supplementary Figure S1). The most striking result of full-length ESCRT-I coexpression/purification in the presence of Mvb12 was that no aggregation of full-length ESCRT-I was observed (Figure 5B). This is in contrast to full-length ESCRT-I behaviour when purified without Mvb12, which elutes on gel filtration as an aggregate. Despite its small 12 kDa size, Mvb12 changes the apparent size of ESCRT-I(Δ21-Vps37) from 233 to 313 kDa as judged by gel filtration (Figure 5B).

To determine the stoichiometry of the ESCRT-I/Mvb12 complex, we carried out gel filtration and ultracentrifugation analyses. Gel filtration indicated that a single copy of Mvb12 binds to a single ESCRT-I(Δ21-Vps37) heterotrimer. An equimolar mixture of Mvb12 and ESCRT-I(Δ21-Vps37) yields a single species containing four components; however, even 50% molar excess of Mvb12 results in an additional peak containing free Mvb12 (Figure 5C). Furthermore, we used equilibrium analytical ultracentrifugation to determine accurately the size and stoichiometry of the ESCRT-I/Mvb12 complex. This gave an apparent molecular weight of 104 407±379 Da for the ESCRT-I/Mvb12 heterotetramer (data not shown). This is in agreement with the expected size of ESCRT-I/Mvb12 complex with a 1:1:1:1 stoichiometry (109 072 Da).

Mvb12 binds heterotrimeric ESCRT-I with nanomolar affinity

We measured the affinity of Mvb12 for ESCRT-I using a fluorescence anisotropy assay with an FlAsH-tagged Mvb12 (FlAsH-Mvb12). Direct titration of ESCRT-I(Δ21-Vps37) (the most intact soluble, non-aggregated complex that can be purified without Mvb12) into FlAsH-Mvb12 gave a K_d of 28 nM (Figure 6A). To map the Mvb12-binding site in the ESCRT-I complex, we carried out the fluorescence assay with various truncations of heterotrimeric ESCRT-I (Table II). These results indicate that the coiled-coil region of Vps23 (residues 252–304) represents the primary binding site for Mvb12. Deletion of this region from the ESCRT-I complex reduces the affinity for Mvb12 more than 30-fold. However, the 3 μM K_d of Mvb12 for the ESCRT-I core suggests that there are additional interactions with ESCRT-I outside the coiled-coil region of Vps23. In addition, we find that the C-terminal domain of Vps28 does not participate in binding Mvb12. A construct consisting of only Vps23 residues 200–304 encompassing the coiled-coil region required for high-affinity Mvb12 binding was not soluble when expressed in E. coli (data not shown), and therefore we could not measure its affinity for Mvb12 directly.

The large footprint of Mvb12 on ESCRT-I suggests an important structural role in stabilising the ESCRT-I complex. Indeed, coexpression of Mvb12 with ESCRT-I prevents aggregation (Figure 5B) and greatly enhances the homogeneity of purified recombinant ESCRT-I complexes as Mvb12 protects regions of ESCRT-I that are prone to proteolysis (data not shown).

Mvb12 does not regulate the ESCRT-I/-II supercomplex in vitro

Gel filtration analysis was carried out using purified recombinant ESCRT-I/Mvb12 and ESCRT-II. The addition of Mvb12 to ESCRT-I did not affect the formation of an ESCRT-I/-II supercomplex in vitro (Figure 6B). Using a competition fluorescence anisotropy assay similar to that previously outlined (Figure 1B), we measured the affinity of the heterotetrameric ESCRT-I(Δ21-Vps37)/Mvb12 complex for ESCRT-II. ESCRT-I(Δ21-Vps37) either with or without Mvb12 is equally efficient at competing with FlAsH-Vps28-CT for ESCRT-II (K_d=44 and 30 nM, respectively) (Figure 6C). Therefore, binding of Mvb12 does not affect ESCRT-I binding to ESCRT-II in a recombinant system in vitro. Full-length ESCRT-I/Mvb12 complex is also an equally efficient competitor (K_d=23 nM) of an FlAsH-Vps28-CT/ESCRT-II complex, demonstrating that the N-terminus of the ESCRT-I Vps37 subunit also does not affect the affinity of the ESCRT-I/-II supercomplex formation in vitro (data not shown). Using direct titration, a FlAsH-Mvb12/ESCRT-I complex binds to ESCRT-II with a similar affinity (K_d=34 nM) as determined using FlAsH-Vps28-CT (data not shown).

The role of the Vps28 (ESCRT-I) C-terminal domain

The Vps28 subunit of ESCRT-I is composed of two discrete domains, the N-terminal core domain, which is essential for the formation of a stable ESCRT-I complex, and the C-terminal domain, which is not necessary for the stability of the ESCRT-I complex. Deletion of the C-terminal domain in yeast in vivo results in a strong class E phenotype (Kostelansky et al, 2006), showing that this domain has an important function in the MVB pathway. The region of the Vps28 core immediately upstream of this domain folds into three small helices and packs differently in the two different crystal forms of the ESCRT-I core (Kostelansky et al, 2006; Teo et al, 2006). Therefore, it appears that the Vps28-CT domain is connected to the rigid ESCRT-I core by a flexible tether. The use of small flexibly connected domains (Vps28-CT and Vps36-NZF-N) in the ESCRT-I/-II interaction may be important to prevent steric clashes between the two large multi-subunit ESCRT-I and -II complexes.

Structural comparison of the yeast Vps28-CT and X. laevis Vps28-CT domains highlights that the rigid four-helix bundle is highly conserved in evolution, but despite this, the NZF-N-binding surface is not conserved. As both plants and metazoa lack an NZF insertion in their Vps36 subunits (ESCRT-II), it appears that the NZF is a relatively recent elaboration of the ancient ESCRT complexes. If metazoan ESCRT-I and -II do associate, it would involve a different mechanism. Surface analysis of the ESCRT-I Vps28-CT domain reveals a highly conserved patch centred on residues 160-FITxxDxx(K/R)L-169. Surface residues near this patch (228-FDxE-231) have been recently linked to binding the yeast Vps20 subunit of ESCRT-III in vitro (Pineda-Molina et al, 2006). Further experiments are necessary to explore the role of this conserved Vps28 patch in yeast and mammalian MVB pathways.

The role of the Vps36 (ESCRT-II) NZF insertion

The Vps36 subunit of ESCRT-II contains two tandem NZF domains, Vps36-NZF-N and Vps36-NZF-C, which enable ESCRT-II to bind simultaneously to the Vps28 subunit of ESCRT-I and Ub, respectively (Teo et al, 2006). The structure of the Vps36-NZF-N domain offers an insight into the molecular mechanisms of a non-Ub-binding class of NZF domains. Despite an identical conformation of the rubredoxin knuckles, NZF zinc-fingers that bind Ub or Vps28-CT use two different surfaces to achieve binding partner specificity. A simple predictive tool for identifying Ub-binding NZF domains was characterised by the highly conserved sequence motif 187-TF-188/φ-199 (where φ is a bulky hydrophobic residue; numbers corresponding to the yeast Vps36-NZF-C, which binds Ub) (Alam et al, 2004). Here, we show that the Vps36-NZF-N domain, which binds to Vps28-CT, uses a distinct binding surface characterised by the hydrophobic signature motif Ile-122/Val-148/Leu-154 in S. cerevisiae.

The binding interface in Vps36-NZF-N for Vps28-CT is more extensive than the Ub-binding interface in Npl4-NZF. This is due to a C-terminal extension in Vps36-NZF-N, which is important for achieving the nanomolar affinity of yeast ESCRT-I for ESCRT-II. It is likely that this extension also stabilises the overall NZF fold, as it forms the third strand of the three-stranded β-sheet in the Vps36-NZF-N domain. A C-terminal extension may be a feature that is shared by at least some other NZF domains, but in the subset of NZF domains that bind Ub, such an extension would not directly participate in binding because Ub interacts with the NZF surface distal to the extension.

Mvb12 and its role in the ESCRT-I/-II link

We have demonstrated that ESCRT-I, which was previously thought to be a heterotrimer, binds directly with nanomolar affinity to a new interacting protein, Mvb12, identified in a recent proteome-wide purification of yeast complexes (Krogan et al, 2006). Deletion of the gene for Mvb12 is associated with partial mis-sorting of carboxypeptidase-Y (Bonangelino et al, 2002) and CPS (Chu et al, 2006), suggesting Mvb12 has a positive role in the MVB pathway. Despite the similar nanomolar affinities of the Mvb12/ESCRT-I and ESCRT-I/-II interactions (K_d of 28 and 27 nM, respectively), only the Mvb12/ESCRT-I heterotetramer and not the ESCRT-I/-II supercomplex copurifies from wild-type yeast, implying that the ESCRT-I/-II interaction is regulated in yeast (Chu et al, 2006). However, an ESCRT-I/-II supercomplex readily copurifies from an Mvb12Δ yeast strain (Chu et al, 2006), suggesting that Mvb12 either directly or indirectly regulates an ESCRT-I/-II link in vivo. Furthermore, this study shows the formation of an Mvb12-dependent ESCRT-I oligomer in yeast. Our in vitro measurements with recombinant proteins demonstrate that Mvb12 does not result in ESCRT-I oligomerisation, nor does Mvb12 affect the ESCRT-I interaction with ESCRT-II, implying a more complex regulatory mechanism in yeast cytosol. This regulatory mechanism could involve either a covalent modification of ESCRT-I or ESCRT-II components or the recruitment of another component that prevents a constitutive ESCRT-I/-II supercomplex in yeast.

We find that Mvb12 considerably reduces proteolysis during purification of the ESCRT-I complex expressed in E. coli and also completely eliminates the aggregation of full-length ESCRT-I heterotetramer. This is likely to be a simple consequence of Mvb12 binding to a significant portion of the ESCRT-I complex, which encompasses the central coiled-coil region of Vps23 (residues 252–304) and the ESCRT-I core. Such a stabilising role of Mvb12 for ESCRT-I we observe in vitro may also be important for ESCRT-I in vivo.

The ESCRT interaction network

Structural information on the ESCRT complexes has offered glimpses into an extended network formed between ESCRT machinery, ubiquitinated cargo and endosomal membranes. The ESCRT complexes use multifunctional domains to efficiently sort mono-ubiquitinated cargo into the lysosomes. These multifunctional binding domains can be thought of as hubs or branches regulating a network of interactions (Praefcke et al, 2004). The GLUE domain of ESCRT-II is one such hub that can recognise endosomal membranes, ubiquitinated cargo and ESCRT-I simultaneously. The yeast Vps28-CT domain could act as a branch point by its ability to bind to both ESCRT-II and ESCRT-III complexes. Structural information on the extended networks formed by the ESCRT machinery with cargo on endosomes and how this might influence the morphology of the endosomes is needed to understand fully the MVB pathway. Critical questions remain to be answered. What are the molecular mechanisms for triggering lumenal vesicle formation at the start of MVB biogenesis? What is the functional significance of the differences between yeast and metazoan ESCRT complexes and their links? The discovery of a new component of ESCRT-I underscores that there is still much to be learned about the MVB pathway.

Plasmids

The S. cerevisiae Vps28 C-terminus (Vps28-CT, residues 148–242), Vps23 C-terminus (residues 252–385), Vps36-GLUE (residues 1–289), Vps36 yeast-specific insertion (residues 110–205), minimal Vps36-NZF-N (residues 110–151) and extended Vps36-NZF-N (residues 110–176) were PCR amplified from full-length clones described previously (Teo et al, 2006). Vps36 constructs and Vps28-CT were cloned into the first and second cassettes of the polycistronic, coexpression vector pOPC with an N-terminal MAH₆ tag on Vps36 or on Vps28-CT, as indicated in the figures, following the general cloning strategy described by Tan (2001). The Vps28-CT and the minimal Vps36-NZF-N were also expressed individually with an N-terminal MAH₆ tag using the pOPTH vector. All heterotrimeric ESCRT-I complexes were cloned into the pOPCH vector where Vps28 is in the first cassette (with an N-terminal MAH₆ tag), Vps23 in the second cassette and Vps37 in the third cassette. The S. cerevisiae Mvb12 (YGR206W) was amplified from genomic DNA (ATCC 9763D) and cloned into the pOPTH vector. Subcloning of His₆-tagged Mvb12 into the fourth cassette of the pOPC vector (Vps28, Vps23 and Vps37 in the first, second and third cassettes) generated coexpression vectors for the heterotetrameric complexes. Mutations in Vps36 were generated using the QuikChange Mutagenesis Kit (Stratagene). A new vector (pOPTHFlash) was constructed for expression of proteins in E. coli and subsequent labelling with Lumio Green reagent (FlAsH, Invitrogen). This vector introduces an MAHHHHHHSSGSENLYFQGSCCPGCCGHM tag (His-rsTEV-Flash tag) to the N-terminus of a protein. The N-terminal His-tag can be cleaved by the addition of TEV protease leaving a small 11 amino-acid tag for labelling with FlAsH (Adams et al, 2002). Similarly, another vector (pOPTHLFlash) was constructed to introduce a His₆-tagged lipoyl domain (followed by the TEV cleavage site and Flash tag) to the N-terminus of a protein (for expression of proteins poorly expressed in E. coli using pOPTHFlash). All constructs were verified by sequencing.

Protein expression and purification

See Supplementary data.

Fluorescence binding assays

In all direct binding assays, analyte protein was titrated into a cuvette containing 15 nM Vps28-CT or 25 nM full-length Mvb12 labelled at its N-terminus with Lumio Green (FlAsH-Vps28-CT/FlAsH-Mvb12) in 2.5 ml binding buffer (20 mM Tris (pH 7.4, 20°C), 100 mM NaCl, 10 mM ammonium sulphate and 10mM β-mercaptoethanol). In all competition assays, test proteins were titrated into a cuvette containing preincubated 15 nM FlAsH-Vps28-CT and 120 nM full-length yeast ESCRT-II. Fluorescence was measured using a Perkin-Elmer LS-55 spectrophotometer with λ_ex=490 nm and λ_em=530 nm. Excitation and emission slits were 15 and 20 nm, respectively. Anisotropy was measured with an integration time of 5 s. Protein was titrated into the cuvette with a Hamilton-MicroLab titrator, allowing 50 s stirring after each titration step and a wait of 10 s before the anisotropy was recorded. Titration of binding buffer alone in either a direct or competition binding assay results in no overall change in anisotropy. The K_d values were calculated from direct fitting of the curves obtained from direct binding assays and competition assays in the program Datafitter (DB Veprintsev). At least two independent experiments were conducted to determine K_d values.

Crystallisation, data collection and structure determination

See Supplementary data.

ESCRT-I analytical gel filtration and ultracentrifugation

See Supplementary data.

Microscopy

Living yeast cells expressing GFP-CPS were harvested at an OD₆₀₀ of 0.4–0.6 and labelled with FM4-64 for vacuolar membrane staining and resuspended in a medium for visualisation. Visualisation of cells was performed on a fluorescence microscope (Axiovert S1002TV; Carl Zeiss Microimaging Inc.) equipped with fluorescein isothiocyanate and rhodamine filters, captured with a digital camera (Photometrix) and deconvoluted using Delta Vision software (Applied Precision Inc.). Results presented were based on observations of >120 cells at 30°C for vps36Δ and Vps36^{I122D/V148D/D151R/L154R} and at 37°C for Vps36^I122D/V148D.

Accession numbers

The coordinates for the structures described in this work have been deposited in the Protein Data Bank with ID codes 2J9U (ScVps28-CT/Vps36-NZF-N), 2J9V (ScVps28-CT) and 2J9W (Xenopus Vps28-CT).