Endosomal Sorting Complex Required for Transport (ESCRT) Complexes Induce Phase-separated Microdomains in Supported Lipid Bilayers*

Background: Endosomal sorting complex required for transport (ESCRT) complexes sort ubiquitinated membrane proteins into vesicles that bud away from the cytosol. Results: Human ESCRT-II assembles into cholesterol-dependent clusters of 10–100 molecules that bind ubiquitin and the ESCRT-III subunit VPS20 and exclude an Ld-phase-specific dye. Conclusion: ESCRT proteins induce lipid phase separation in vitro. Significance: Lipid phase separation by ESCRTs may aid membrane budding. The endosomal sorting complex required for transport (ESCRT) system traffics ubiquitinated cargo to lysosomes via an unusual membrane budding reaction that is directed away from the cytosol. Here, we show that human ESCRT-II self-assembles into clusters of 10–100 molecules on supported lipid bilayers. The ESCRT-II clusters are functional in that they bind to ubiquitin and the ESCRT-III subunit VPS20 at nanomolar concentrations on membranes with the same stoichiometries observed in solution and in crystals. The clusters only form when cholesterol is included in the lipid mixture at >10 mol %. The clusters induce the formation of ordered membrane domains that exclude the dye 1,1′-dioctadecyl-3,3,3′,3′-tetramethylindodicarbo-cyanine perchlorate. These results show that ESCRT complexes are capable of inducing lateral lipid phase separation under conditions where the lipids themselves do not spontaneously phase-separate. This property could facilitate ESCRT-mediated membrane budding.

The ESCRT 5 complexes are conserved from yeast to humans and play key roles in receptor down-regulation and lysosome biogenesis (1)(2)(3), the budding of HIV-1 and other membraneenveloped viruses, and cytokinesis (4), among other cellular pathways. ESCRTs function in receptor down-regulation and endosome biogenesis by sorting ubiquitinated membrane proteins into nascent membrane buds that evaginate away from the cytosol, and a parallel pathway is directly involved in budding from the plasma membrane (5). The topology of the ESCRT membrane budding reaction is opposite to that of the well known coated vesicle budding pathways (6,7). The ESCRT budding topology is, however, the same as that of several lipiddriven pathways (8). These include virus-and toxin-driven clustering of the lipids GM1 and Gb3 (9), ceramide-dependent exosome budding (10), and lysobisphosphatidic acid-dependent budding into late endosomes (11). The key lipid requirement in the ESCRT-driven MVB biogenesis pathway is for phosphatidylinositol 3-phosphate (PI(3)P). PI(3)P is essential for the progression of MVB biogenesis because of its key role in the recruitment of ESCRT-0 and ESCRT-II to endosome membranes.
In vitro reconstitution of multivesicular body (MVB) biogenesis by the ESCRTs in a PI(3)P-containing giant unilamellar vesicle (GUV) model system suggested the following division of labor. ESCRT-I and -II form inward buds (12) by self-assembling together at the neck of the bud. Subsequently, ESCRT-III polymerizes on the membrane (13,14) into structures that drive the final scission of the membrane bud (15,16).
A structural model for an ESCRT-I-II pore assembly at the bud neck was proposed on the basis of a low resolution structure of the supercomplex (17,18), but the pathway for nucleation of the putative pore assembly is not clear. One prediction of the pore model is that ϳ6 -10 ESCRT-II complexes should line the neck of the pore of the 25 nm buds in yeast MVBs, and therefore, twice that number should assemble when 50-nm buds are formed in human cells. However, the size of ESCRT clusters on membranes has not been measured directly.
The energetic driving force for bud formation and the factors controlling the size of the membrane bud are important questions in the ESCRT field. A computational analysis of this pathway suggested that the neck-based budding mechanism would be strongly favored if it were coupled to the existence of line tension. 6 The magnitude of the line tension is a key element in determining the size of the bud in this model. 6 Line tension would occur only if a distinct lipid domain were present in the bud. 6 A key prediction of this model, which we sought to test, was that at least some of the ESCRT complexes should be capable of inducing lipid phase separation in membranes of the same compositions that give rise to membrane buds in GUVs (12). As a model system, we chose human ESCRT-II, whose interactions with ESCRT-III, ubiquitin, and membranes are particularly well characterized. Here, we show that ESCRT-II selfassembles on supported lipid bilayers (SLBs). The SLB format was selected for this study because of its advantages as a substrate for quantitative analysis by total internal reflection fluorescence microscopy (TIRFM). These experiments allow the size of ESCRT clusters and the stoichiometry of their interactions to be directly measured, which has not been possible in other modalities used in this field. The ESCRT-II assemblies have the expected ability to bind membrane-tethered ubiquitin and the most upstream of the ESCRT-III subunits, VPS20. These binding events occur via the known structural interfaces, demonstrating that these assemblies recapitulate functional interactions. TIRFM analysis in the SLB format allowed us to analyze the stoichiometry of these interactions on membranes with excellent precision. The self-assembly of ESCRT-II on SLBs requires the presence of cholesterol and leads to the induction of a liquid ordered (L o )-like domain. This provides the first direct evidence that ESCRT complexes are capable of inducing lipid phase separation, which in turn could give rise to line tension and so enhance budding and scission.

EXPERIMENTAL PROCEDURES
Protein Expression and Purification-Human ESCRT-III subunit VPS20 and its mutants were expressed and purified as described previously (19). Briefly, N-terminal His 6 -maltosebinding protein-TEV cleavage site fusion proteins were expressed in Escherichia coli BL21 Star at 30°C overnight after induction with 0.5 mM isopropyl thiogalactoside at an absorbance (A 600 nm ) of 0.8. The proteins were affinity-purified using Ni-NTA resin (Qiagen). VPS20 was cleaved using TEV protease and further purified by size exclusion chromatography on a Superdex 200 column (GE Healthcare). Full-length human ESCRT-II was purified from E. coli BL21 Star as described previously (20). Briefly, all three subunits were co-expressed from pST39 vector with a His 6 tag and TEV cleavage site fused to the VPS22 subunit. The protein was affinity-purified using Ni-NTA resin (Qiagen) followed by size exclusion chromatography on Superdex 200. Yeast ESCRT-II was purified as described previously (12,30). A fusion protein of GFP and the FYVE domain from EEA1 with N-terminal His 6 tag (His 6 -GFP-FYVE), a ubiquitin construct fused at its C terminus to Cys-Ala-His 6 , and its I44D mutant, were expressed using leaky expression in E. coli from the first cassette of pST39 at 30°C overnight. The protein was affinity-purified using Ni-NTA resin followed by size exclusion chromatography on Superdex 75 (for ubiqui-tin) or on Superdex 200 (for GFP-FYVE). Proteins were flashfrozen in liquid nitrogen and stored at Ϫ80°C until use.
Labeling by Fluorescent Dyes-The engineered Cys mutant proteins (VPS20-C109S/S119C; ubiquitin-Cys-Ala-His 6 , ubiquitin-I44D-Cys-Ala-His 6 ) were incubated overnight at 4°C with a 20-fold molar excess of Alexa488, Atto488, Atto647, or rhodamine maleimide derivatives. Labeled maltose-binding protein VPS20 was diluted to 1 M and cleaved by TEV protease. After cleavage, the protease was removed using Ni-NTA resin (Qiagen). The labeling efficiency estimated spectroscopically was ϳ100%. Wild-type human and yeast ESCRT-II, which contain multiple native Cys, were labeled by the same procedure, except with a 2.5-fold molar excess of dye. For all samples, unreacted dye was removed by gel filtration chromatography on Superdex 75 or 200. The labeling efficiency estimated spectroscopically was 200% (Alexa488) and 180% (Atto647), respectively, and more than 90% of fluorescence signal was found to be in the VPS25 subunit as judged by SDS-polyacrylamide gel followed by analysis using Typhoon laser gel scanner (GE Healthcare). Labeled proteins were flash-frozen in liquid N 2 and stored at Ϫ80°C until use.
Preparation of Liposomes-The following lipids were purchased from Avanti Polar Lipids: 1-palmitoyl-2-oleoyl-snglycero-3-phospho-L-serine (POPS), 1-palmitoyl-2-oleoyl-snglycero-3-phosphocholine (POPC), cholesterol, 1,2-dipalmitoyl-sn-glycero-3-phosphoethanolamine-N-lissamine rhodamine B sulfonyl. Dipalmitoyl phosphatidylinositol 3-phosphate (DPPI(3)P) was purchased from Echelon, and dioleoyl phosphatidylinositol 3-phosphate was purchased from Avanti Polar Lipids. One mg of lipid mixture containing POPC (62 mol %), POPS (10 mol %), cholesterol (25 mol %), and PI(3)P (3 mol %) was dried onto the borosilicate glass tube wall with a nitrogen stream. DPPI(3)P was used in all studies except where noted otherwise. For mixtures at a lower mole fraction of cholesterol, the cholesterol was replaced with POPC on an equimolar basis. The tube was dried for 3 h in a vacuum oven without heating and hydrated with buffer (20 mM Tris, pH 7.4, 150 mM NaCl, 3 mM ␤-mercaptoethanol, 0.16% sodium ascorbate) overnight at 4°C. The tube was vortexed and sonicated in a cup horn sonicator (Misonix) with room temperature water circulation until the small unilamellar vesicle (SUV) solution became clear (ϳ50 watts for 10 min). The SUV suspension was freshly prepared on the day of the experiment. MgCl 2 was added to the SUV suspension to a final concentration of 5 mM just before loading into the flow cell. The flow cell was slowly loaded with the SUV suspension, incubated at room temperature for approximately 1 h, and washed with buffer.
Flow Cell, Microscope, Camera, Illumination, and Data Processing-Flow cell preparation, microscopy setup, and protein surface concentration measurements were as described previously (21). The 25-m deep flow cells with inlet and outlet were assembled on fused silica slides. The slides were washed with Hellmanex II soap, rinsed extensively with deionized water, and air-dried for at least 2 h. A total internal reflection fluorescence inverted microscope with Nikon Plan Apo VC ϫ100 NA ϭ 1.40 oil immersion objective, DualView (Photometrics) adapter with dcxr630 dichroic, and an Andor iXonϩ DU-897E camera were used for imaging. The spatial resolution of the camera was 6 pixels/m. The presented images are 200 ϫ 200 pixels or 33 ϫ 33 m. The s-polarized evanescent wave was created near the surface of flow cells by 488 and 633 nm laser beams normally incident to 70°fused silica dove prism. The illumination laser lines were blocked between the objective and camera with Semrock LineStop filters (A Ͼ6). The laser power was controlled with calibrated attenuators. The camera signal was normalized according to the power and camera settings (reproducibility of the data were verified for different camera and power settings). Measured fluorescence intensity was proportional to the volume concentration of fluorescent molecules multiplied by the penetration depth of the evanescent wave (128 nm for 488-nm beam) plus the surface concentration of fluorescent molecules. The volume contribution to the fluorescence intensity signal was much smaller compared with the contribution of the surface for all ESCRT proteins at the concentration range used. The surface density of fluorescently labeled proteins was estimated by comparison between the fluorescent signals from the protein and a calibration measurement with fluorophore standard in the solution. The fluorophore standard had no detectable affinity to the lipid bilayer. The upper estimate of error for concentration measurements is 5 and 3% for fluorescence intensity measurements, and 2% is the day to day measured variance in laser excitation power. Taken together, we estimate that the number of bound ESCRT molecules has a combined error of ϳ14%. Metamorph 7 software was used for camera control and image acquisition. ImageJ was used for image analysis.

ESCRT-II Forms Clusters on the Supported Lipid Bilayer-
We visualized the assembly of fluorescently labeled full-length recombinant human ESCRT-II complex on SLBs by TIRFM. ESCRT-II formed relatively small clusters on the SLB using a simple model membrane composition capable of supporting budding in GUV experiments. Conditions were kept near physiological with pH 7.4, and an ESCRT-II concentration at the low nanomolar level (Fig. 1A). The number of molecules was estimated from the intensity of the fluorescence. We standardized the amount of fluorescent signal per dye using free dye in solution. The number of dye molecules per cluster was computed, and from this the number of ESCRT-II complexes was obtained. The modal ESCRT-II cluster contains ϳ50 copies of the complex (Fig. 1B). We cannot rule out that this number may be underestimated due to self-quenching effects. A range of sizes from ϳ10 to Ͼ100 is observed.
To probe whether there were pre-existing PI(3)P clusters or if generic PI(3)P binding proteins could induce clustering, the FYVE domain of EEA1 was flowed over the SLB. EEA1-FYVE bound but did not form clusters (Fig. 1B). Therefore, we ruled out the presence of pre-existing and nonspecific PI(3)P clustering, at least on the size scale observable by optical microscopy. Furthermore, the ESCRT-II sample in solution consists of nonaggregated singlet complexes as determined by size exclusion chromatography (data not shown). Because ESCRT-II behaves differently in this system from the generic PI(3)P-binding control, the cluster formation is best explained by the formation of specific lateral homotypic interactions on SLBs.
To compare these results to previous in vitro analysis of yeast ESCRTs on GUVs, the experiment was repeated with fluorescently labeled yeast ESCRT-II. Yeast ESCRT-II clusters do seem to be smaller and more numerous than human ESCRT-II. This probably explains why they appeared in the previous confocal study by Wollert and Hurley (12) as a continuum. We further examined whether human ESCRT-II formed clusters on GUVs, and we found that it does (Fig. 1E). The observation that human ESCRT-II clusters are clearly visible on confocal imaging of GUVs but the yeast clusters are not is explained by the finding that the human clusters are larger and but less numerous and hence better resolved from one another.
ESCRT-II Clusters Bind Cargo-To determine whether the ESCRT-II clusters corresponded to physiologically functional entities, we tested whether the clusters were capable of binding model cargo. To create model cargo tethered to the lipid bilayer, ubiquitin was fused to a C-terminal Cys and a His 6 tag (ubiquitin-Cys-Ala-His 6 ). The single Cys was used to covalently attach the fluorescent dye Alexa488, and the labeled ubiquitin was tethered to the lipid bilayer via its His 6 tag and the Ni 2ϩbinding lipid DOGS-NTA. When tagged ubiquitin and ESCRT-II were passed together through the flow cell, ESCRT-II-ubiquitin co-clusters formed (Fig. 2, A-C). The ratio and absolute number of molecules were estimated from the fluores-cence intensities, and the ratio was determined to be 1:0.9 Ϯ 0.2 ESCRT-II/ubiquitin. The ubiquitin mutant I44D that does not bind ESCRTs was used as a control for specificity (Fig. 2, D-F), demonstrating that the ESCRT-II clusters bind model cargo in a structurally specific manner.
ESCRT-II Clusters Recruit the ESCRT-III Subunit VPS20-To further test the physiological relevance of the observed ESCRT-II clusters, we probed whether they bound the immediate downstream ESCRT-III protein VPS20. We used twocolor TIRFM to monitor the formation of clusters of fluorescently labeled ESCRT-II and VPS20 that were added to the flow cell together at 5 nM concentration (Fig. 3, A-C). VPS20 bound preferentially to the SLBs at the positions of the ESCRT-II clusters, without observably perturbing them. The VPS20 ⌬ESCRT-II mutant lacking the ability to bind ESCRT-II was added to the flow cell in a separate experiment. VPS20 ⌬ESCRT-II was able to form clusters of its own, presumably because it is still able to bind to acidic lipids such as PI(3)P, and it retains its ability to self-interact on membranes. However, essentially no co-localization with the ESCRT-II clusters was seen on the SLBs (Fig. 3, D-F), even when the concentration of the mutant was raised to 30 nM (shown in Fig. 3, D-F). Thus, the interaction occurs via the crystallographically defined ESCRT-II/VPS20 interface.
We sought to determine whether VPS20 and ESCRT-II form assemblies that interact on the molecular size scale, as opposed to large low density clusters. A single cluster FRET experiment was performed. The single cluster FRET experiment takes advantage of the fact that ϳ90% the Alexa488 label in ESCRT-II was located in the VPS25 subunit, and therefore within 1-2 Förster radii of the VPS20-binding site. ESCRT-II was allowed to form clusters and then rhodamine-labeled VPS20 was flowed into the chamber. Energy transfer was observed as a function of time by monitoring the intensity of fluorescence in both donor and acceptor channels (Fig. 4, A and B).
The absolute number of molecules of ESCRT-II and VPS20 was estimated from the intensity of fluorescence and the calculated labeling efficiencies. The observed average ESCRT-II: VPS20 stoichiometry of 1:2.1 Ϯ 0.3 (data not shown) is close to the 1:2 stoichiometry suggested by the crystal structure. From  the VPS20 mutant experiments and the single cluster FRET, we conclude that clustered ESCRT-II binds directly to downstream ESCRT-III member VPS20.
Clustering of ESCRT-II Is Cholesterol-dependent-The clustering of ESCRT-II binding on SLBs was assessed as a function of mol % cholesterol by TIRFM. We observed no complex clustering in SLBs containing 0 -10% cholesterol (Fig. 5, A and B). At 15 mol % cholesterol, clustering was observed (Fig. 5C), but its character was more diffuse than at 25 mol % (Fig. 5D). Taken together with the observation that the EEA1 FYVE domain does not cluster on cholesterol-containing SLBs, these results are interpreted to mean that the clustering of ESCRT-II is cholesterol-dependent and depends on specific interactions among ESCRT-II complexes and specific lipid surface area. Because clustering is cholesterol-dependent, we considered whether clustering might also depend on the presence of saturated acyl chains in the lipids. We also sought to explore whether a PI(3)P with a tail composition that is thought to be more abundant in cells would support clustering. We repeated the clustering experiment with DOPI(3)P and found that the cluster formation was robust and similar to DPPI(3)P-dependent clustering (Fig. 5E).
ESCRT-II Induces Phase Separation in the SLB-Because clustering of ESCRT-II is cholesterol-dependent (Fig. 5), phase separation was assessed in the model lipid system. The phasesensitive fluorescent dye DiD (Molecular Probes) preferentially localizes to the liquid disordered phase and was used to monitor the phase state of the lipid bilayer. First, we determined if the DiD dye was distributed evenly and was free to move laterally within the SLB. The latter was addressed using a fluorescence recovery after photobleaching experiment. A circular region of the SLB was bleached, and fluorescence was completely recovered within ϳ100 s. The equal distribution and quick recovery of the DiD dye demonstrate that the lipid bilayer had no phase separation prior to ESCRT assembly, consistent with the absence of clustering of the FYVE domain construct. Unlabeled ESCRT-II and VPS20-Atto488 were added to the SLB that contained DiD (Fig. 6, A-C). The ESCRT clusters that formed were found to be depleted of DiD relative to the rest of the bilayer (Fig. 5B). The lipids associated with ESCRT-II clusters are thus in a more ordered environment than their surroundings.

DISCUSSION
The energetics and geometry of vesicle budding by ESCRT-I and -II dictate that multiple copies of these complexes must assemble to drive budding. A structural model for the bud neck assembly in yeast MVB biogenesis predicted that at least six copies each of ESCRT-I and -II would be present. In human cells, MVB buds are larger than in yeast, and the corresponding assembly would thus be predicted to have at least ϳ12 copies. The main importance of the cluster sizes seen in this study is that they are in the range needed, as estimated from structural modeling and on energetic grounds, to drive vesicle budding.
One of the key findings in this study is that the self-assembly of human ESCRT-II on SLBs is coupled to lipid phase separation. The self-assembly process requires the presence of cholesterol, indicating that lateral interactions between multiple ESCRT-II complexes are, on their own, insufficient to drive protein clustering. Furthermore, the lipid domains observed in this study are not spontaneously formed, as demonstrated by the uniform distribution of the EEA1-FYVE construct. The lipid composition used in this study is distant on the phase diagram from the expected point of spontaneous lipid demixing (22). We attribute the clustering observed in this study to a combination of weak lateral interactions between ESCRT-II complexes, preferential interactions between ESCRT-II complexes and phosphoinositides, and the ability of the phosphoinositide/cholesterol/POPC/POPS mixture to phase-separate in response to protein binding.
ESCRT-II clusters are functionally competent to bind stoichiometrically to ubiquitin at nanomolar concentrations. The finding that monoubiquitin is competent to direct endosomal protein trafficking in yeast (23) was initially paradoxical given that most ubiquitin binding domains of the trafficking machinery bind to monoubiquitin with K d values of 100 M or higher (24 -26). Membrane tethering of ubiquitin, via its covalent attachment to integral membrane proteins, increases the local concentration of ubiquitin moieties on endosomal membranes. Co-clustering of 65 nM membrane-tethered ubiquitin with yeast ESCRT-0, -I, and -II was observed on GUVs, suggesting that membrane tethering provided enough of an increment in affinity to resolve the apparent paradox. The GUV experiment did not, however, address the stoichiometry of the interaction. Here, we find that at 100 nM bulk concentration, tethered ubiquitin binds with 1:1 stoichiometry to human ESCRT-II. Human ESCRT-II has only one known ubiquitin-binding site, with its GLUE domain (27,28). These results show that human ESCRT-II contains a single ubiquitin-binding site in a functionally relevant membrane setting.
One of the fundamental and conserved functions of ESCRT-II is to couple the upstream ubiquitin-binding and membrane-budding machinery to the downstream ESCRT-III membrane scission machinery. This occurs through the binding of the two Vps25 subunits of the Y-shaped ESCRT-II complex (29,30) to the Vps20 initiator subunit of ESCRT-III (29,31). Both copies of the Vps25 subunit, and both Vps20 binding events, are required for function (30,32). Yeast Vps20 undergoes an activating conformational change upon binding to Vps25 (33), which then triggers the recruitment of Snf7 and more downstream ESCRT-III subunits, leading to membrane scission (15). The interacting fragments of human VPS25 and VPS20 have been co-crystallized (19). The interacting residues shown to be conserved and essential for ESCRT function in yeast and in the in vitro GUV assays (19). Here, we have been able to demonstrate co-clustering of VPS20 with ESCRT-II on SLBs at physiologically relevant nanomolar concentrations. The interaction occurs through the structurally characterized interface, because mutations designed on the basis of the crystal structure prevent the recruitment of VPS20 to the ESCRT-II clusters. The stoichiometry of VPS20 binding to ESCRT-II was previously inferred from crystal structures and mutational impairments of function. Atomic force microscopy data were consistent with these findings but did not directly differentiate between the two interacting partners (34). Here, we have found by direct quantitation of protein levels on SLBs that VPS20 binds ESCRT-II on membranes with a 2:1 stoichiometry.
The fundamental finding in this study is that ESCRT complexes can induce lipid phase separation coupled with their selfassembly. There have been suggestions based both on mechanistic grounds and on observations in cells that motivated these experiments. The other well characterized examples of coatless budding reactions appear to involve lipid phase separation. The effect of this phase separation is to induce line tension, which at sufficiently high values can drive budding and scission on its own (35,36). In our current view of ESCRT-mediated budding, line tension effects are not postulated to be large enough to drive budding and scission on their own. Indeed, unregulated lipid-driven budding would be undesirable for the cell and, in practice, appears to be shut down in the absence of the recruitment of ESCRTs by ubiquitinated cargo (37). We postulate that lipid phase separation is one facet of a more complex mechanism driven by the ESCRT-I, -II, and -III complexes (18).
There are physiological grounds for expecting that the ESCRTs might have a direct or indirect role in the budding of cholesterol-enriched membranes into MVBs. Impairment of various ESCRT components in cells (38,39,40) blocks endosomal cholesterol transport. It is particularly striking that ESCRT-0 depletion in HeLa cells appears to phenocopy depletion of the lysosomal cholesterol efflux proteins NPC1 and NPC2. Yet ESCRT-0 depletion does not affect the trafficking of these latter proteins (40), suggestive of a direct effect on cholesterol internalization in the endosome. The intralumenal vesicles of MVBs in human B lymphocytes are among the most cholesterol-rich internal membranes in these cells (41). The . ESCRT-II clustering is cholesterol-dependent. ESCRT-II was injected on membranes where cholesterol was replaced by POPC in the standard 25% cholesterol mixture to yield the indicated mole fractions of cholesterol. We observed no clustering at low cholesterol concentrations (A and B; 0 or 10%), and more diffuse clustering on the membrane containing 15% cholesterol (C) was compared with discrete clusters seen on the membrane with 25% cholesterol (D). E, dioleoyl PI(3)P supports clustering in combination with 25% cholesterol. Scale bar, 10 m. ESCRTs contain no known cholesterol binding domains, and there have been no reports that they bind to cholesterol. It seems more likely to us that the ESCRTs directly bind to phosphoinositides, which in turn forms clusters with cholesterol, as opposed to a direct interaction. This effect is observed even with DOPI(3)P, which contains unsaturated acyl chains and is thought to be more representative of cellular phosphoinositide pools.
The interplay between cholesterol and phosphoinositides in the clustering of proteins in model membranes has been explored in several other systems, and at least three different themes emerge. (i) The transmembrane domain-containing SNARE syntaxin-1A can cluster in the presence of either cholesterol (42) or an unsaturated PI(4,5)P 2 (43), but the roles of these lipids in any given clustering event appear to be antagonistic (44). (ii) The dipalmitoylated polybasic protein GAP43 binds to cholesterol-rich domains via its lipid anchor and can overcome the energy barrier to bring unsaturated PI(4,5)P 2 into these domains via strong electrostatic interactions with the inositide headgroup (45). (iii) The peripheral membrane protein Annexin A2, which has neither a lipid modification nor a transmembrane segment, binds electrostatically to PI(4,5)P 2 . Saturated (dipalmitoyl) PI(4,5)P 2 couples annexin clustering to cholesterol microdomain formation (46). The clustering phenomenon visualized here is most similar to case iii, although in this study, the clustering effect is seen even with an unsaturated phosphoinositide. It is intriguing that VPS20 is constitutively myristoylated in cells (31,47) and that this modification is required for function yet is dispensable for high affinity membrane binding. It is tempting to speculate that a VPS20 myristoylation in cells might synergize with and strengthen the clustering and phase separation effects described here.
In conclusion, the data presented here demonstrate that ESCRTs can provoke phase separation in model membranes in the course of cluster formation. The monitoring of ESCRT clusters by TIRFM on SLBs provides a tool to directly quantitate the interactions and stoichiometries of ESCRTs with each other and ubiquitin in the membrane setting. The early results for ESCRT-II cluster size and VPS20 and ubiquitin interactions are consistent with predictions from crystallographic, solution biochemical, and computational analyses. The mutual consistency of this growing body of data is encouraging with respect to current models of ESCRT mechanism in cargo sorting, membrane budding, and membrane scission.