Interaction of the Mammalian Endosomal Sorting Complex Required for Transport (ESCRT) III Protein hSnf7-1 with Itself, Membranes, and the AAA+ ATPase SKD1*

SKD1/VPS4B is an AAA+ (ATPase associated with a variety of cellular activities) protein involved in multivesicular body (MVB) biogenesis. In this study, we show that the impairment in MVB biogenesis caused by the ATP hydrolysis-deficient mutant SKD1(E235Q) is accompanied by assembly of a large detergent-insoluble protein complex that includes normally soluble endogenous components of mammalian endosomal sorting complex required for transport (ESCRT) I and ESCRT-III complexes. Membrane-bound ESCRT-III complex has been proposed to be the substrate that recruits SKD1 to nascent MVBs. To explore this relationship, we studied interactions among the human ESCRT-III components hSnf7-1 and hVps24, membranes, and SKD1. We found that a significant portion of overexpressed hSnf7-1 associated with membranes where it formed a large protein complex that recruited SKD1 and perturbed normal MVB biogenesis. Overexpressed hVps24 also associated with membranes and perturbed endosome structure but only when fused to green fluorescent protein. Domain analysis revealed that the basic N-terminal half of hSnf7-1 localized to membranes and formed detergent-resistant polymers, some of which looked like filopodia extending into the lumen of swollen endosomes or out from the plasma membrane. The C-terminal acidic half of hSnf7-1 did not associate with membranes and was required for interaction of hSnf7-1 with SKD1. Together with earlier studies, our work suggests that a variety of ESCRT-III-containing polymers can assemble on membranes and recruit SKD1 during formation of the MVB.

Cells deliver endocytosed soluble and membrane-bound cargo to late endosomes en route to lysosomes. Newly synthesized lysosomal proteins also travel through the late endosome to reach the lysosome (1). A critical feature of late endosomal function is the ability to invaginate membrane and selected proteins into the interior or lumen, giving late endosomes their alternate name of multivesicular body (MVB) 1 (2,3). Mem-brane proteins internalized into the MVB include receptors destined for down-regulation (for example, stimulated growth factor receptors), lysosomal enzymes, and characteristic proteins such as the tetraspanin CD63 (3)(4)(5). In certain specialized cells (particularly in the immune system) internal vesicles are released from the cell as exosomes following fusion of the endosome with plasma membrane (6,7).
A network of proteins first identified in Saccharomyces cerevisiae as the class E subset of vacuolar protein sorting (vps) mutants (8) are required for internalizing proteins and membrane into the lumen of the endosome (4). Dysfunction or deletion of these proteins in yeast leads to overgrowth of a prevacuolar endosome known as the "class E compartment" and causes proteins destined for the vacuolar interior to accumulate in the class E compartment and on the limiting membrane of the vacuole. There are at least 18 class E proteins known in yeast (4,9,10). Interestingly all are soluble and cycle on and off the endosomal membrane. Studies of homologues of class E proteins in mammalian cells indicate that the function of these proteins is conserved from yeast to mammals (for reviews, see Refs. 11 and 12).
Elegant studies carried out over the last few years have established that class E proteins assemble into functional complexes. These include the S. cerevisiae endosomal sorting complex required for transport (ESCRT) complexes I, II, and III (13)(14)(15) as well as a Vps27-Hse1 complex (16). Current models suggest that these complexes act sequentially to select and move cargo into the lumen of the MVB, although the actual reactions that promote vesicle budding and release into the endosomal lumen have not been defined. The proposed MVB pathway begins when the Vps27-Hse1 complex binds phosphatidylinositol 3-phosphate and ubiquitin-conjugated proteins on the endosomal membrane (using FYVE and ubiquitininteracting motif domains). Vps27 then recruits ESCRT-I (Vps23, Vps28, and Vps37) via interactions between it and Vps23 (17). This complex is next thought to engage the ESCRT-II complex (Vps22, Vps25, and Vps36) (14,18,19), which in turn recruits components of the ESCRT-III complex (Vps20, Snf7, Vps2, and Vps24) (15). Equivalent functional complexes have been defined so far for the mammalian homologues of Vps27-Hse1 (Hrs-STAM) and ESCRT-I (Tsg101, hVps28, and hVps37) (20 -22).
One class E protein, Vps4, is an AAA ϩ (ATPase associated with a variety of cellular activities) ATPase (23,24). It is not a part of the ESCRT machinery, but interfering with its activity causes ESCRT components to accumulate in a large detergentresistant complex on the class E compartment membrane (14,15,25,26). This suggests that Vps4 is a critical regulator of proteins in the MVB pathway. Its function depends on its ability to bind and hydrolyze ATP (25) with class E defects appearing both in yeast that lack Vps4 and in yeast that express ATP hydrolysis-deficient Vps4 mutants (23,27).
Mammalian homologues of Vps4, VPS4A and SKD1 (suppressor of K ϩ transport growth defect 1, also known as VPS4B), appear to play a comparable role to their yeast counterpart in regulating MVB biogenesis. Expressing ATP hydrolysis-deficient mutants of either isoform in cultured cells causes the mammalian equivalent of the class E phenotype: vacuolation of endosomes and impairment of late endosomal trafficking (28 -31). We will use VPS4 to refer generally to mammalian isoforms and VPS4A or SKD1 to refer to a specific one. Interestingly expressing ATP hydrolysis-deficient VPS4 also blocks budding of enveloped viruses such as human immunodeficiency virus, type 1, suggesting that the machinery for generating vesicles inside the MVB is generally involved in budding and fission of vesicles leaving the cytosol (32,33). How VPS4 and other class E proteins cooperate with each other and cargo proteins to promote this budding and fission, however, is far from clear. Based on what is known about other AAA ϩ ATPases (34), it has been proposed that VPS4 disassembles a stable protein complex of membrane-bound ESCRT proteins. While all class E proteins are candidate substrates, it seems unlikely that VPS4 acts directly on so many structurally diverse proteins. Instead VPS4 may modify one or a few components that hold the ESCRT machinery together.
Several observations suggest that components of the ESCRT-III complex are the most likely substrates for VPS4. In yeast, inactive Vps4 fails to bind membranes in cells that lack subunits of the ESCRT-III complex (including Vps2 and Vps24) (15). Comprehensive yeast two-hybrid studies of interactions among all of the class E proteins in both yeast and mammalian systems revealed interactions between Vps4 and individual components of the ESCRT-III complex or structurally related coiled-coil-containing proteins, while no interactions were detected between Vps4 and other class E proteins (33,(35)(36)(37)(38)(39)(40).
To further define the nature of interactions between mammalian ESCRT-III proteins and VPS4 and to more generally gain insight into the role of these proteins in formation of the MVB, we characterized human homologues of two ESCRT-III components, hSnf7-1 and hVps24, and their interactions with the VPS4 isoform SKD1. We found that many of the properties attributed to a heteropolymeric ESCRT-III complex in yeast (membrane binding, polymer assembly, and interaction with VPS4) could be recapitulated in mammalian cells by simple overexpression of hSnf7-1. Domain analysis allowed us to assign membrane binding and polymerization to the N-terminal basic half of the protein and interaction with SKD1 to the acidic C-terminal half. We propose that the properties and interactions of hSnf7-1 shed light on the likely organization of ESCRT-III polymers in the MVB pathway.

MATERIALS AND METHODS
Plasmids and Mutagenesis-To express SKD1 in Escherichia coli, mouse SKD1 cDNA (a kind gift from Dr. Carol Vandenberg (University of California)) was inserted into pHO4d (41) between NcoI and EcoRI sites with amino acids PNSG between the C terminus of SKD1 and the His 6 -Myc tag. The E235Q mutation was generated by QuikChange TM site-directed mutagenesis (Stratagene). To express SKD1 wild type or E235Q in mammalian cells, the corresponding cDNA was inserted into pEGFP-N1 (Clontech) between XhoI and BamHI sites. The resulting proteins have amino acids ADPPVAT between SKD1 and EGFP. To create tetracycline-regulated constructs, the entire SKD1-green fluorescent protein (GFP) fusion was amplified and inserted into pcDNA4/TO (Invitrogen) using BamHI and XhoI sites.
To clone human homologues of Snf7 and Vps24, we amplified proteins identified in a data base search from human melanoma cDNA (a gift from Dr. Helen Piwnica-Worms, Washington University). hVps24 was amplified using primers complementary to neuroendocrine differentiation factor (GenBank TM accession number AF219226), while human Snf7 was amplified with primers complementary to HSPC134 (GenBank TM accession number AF161483). Our Snf7 clone differed by three amino acids from HSPC134 and is identical to the more recently described hSnf7-1 (GenBank TM accession number AAQ91193). DNA fragments were inserted into pGEX4T-1 (Amersham Biosciences, GST tag) and pET28a (Novagen (Madison, WI), His 6 and T7 tags) using BamHI and XhoI sites for expression in E. coli. For expression in mammalian cells, hSnf7-1 and hVps24 were inserted into three vectors. Untagged constructs were generated by cloning into pcDNA3.1(ϩ) (Invitrogen) between BamHI and XhoI sites. Myc-tagged constructs were generated by cloning into pcDNA3.1/Myc-His(Ϫ)A (Invitrogen) between XbaI and HindIII sites with a linker of KLGP between hSnf7-1/hVps24 and the Myc epitope. GFP-tagged constructs were created by cloning into pEGFP-N1 (Clontech) between NheI and HindIII sites with a linker of KLRILQSTVPRARDPPVAT between hSnf7-1 or hVps24 and EGFP.
To generate hSnf7-1 fragments, DNA corresponding to amino acids 1-222 (hSnf7 full length (FL)), 1-116 (hSnf7-N), or 117-222 (hSnf7-C) was amplified and inserted into pGEX4T-1 and pET28a between BamHI and XhoI sites for expression in E. coli and into FLAG-pcDNA3.1(ϩ) (a gift from Dr. Kenneth Johnson, Washington University) between BamHI and XhoI sites for expression in mammalian cells. The cloning resulted in a linker of GS between the N-terminal FLAG epitope and the hSnf7-1 protein. Sequences of all constructs were confirmed by sequencing using ABI big dye reagents at the Nucleic Acid Chemistry Laboratory (Washington University).
RNA Blot Hybridization-DNA encoding hVps24 or hSnf7-1 was excised from pGEX4T-1 constructs with BamHI and XhoI for use as a hybridization probe. DNA fragments were labeled by random priming (RediPrime TM II kit (Amersham Biosciences)) according to the manufacturer's recommendations. Human multiple tissue Northern blots (Clontech) were hybridized with these probes in ExpressHyb solution at 68°C for 1 h according to the manufacturer's recommendations and visualized by autoradiography.
Cell Culture and Transfection-T-REx TM HEK293 cells (Invitrogen) expressing tetracycline repressor were grown in Dulbecco's modified Eagle's medium supplemented with 10% tetracycline-free fetal bovine serum (Hyclone), 2 mM L-glutamine, and 5 g/ml blasticidin. COS-7 cells were maintained in Dulbecco's modified Eagle's medium supplemented with 10% fetal bovine serum and 2 mM L-glutamine. All cells were grown in a 5% CO 2 incubator at 37°C.
Cells were transfected using Lipofectamine (Invitrogen) according to the manufacturer's recommendations. Transiently transfected COS-7 cells were analyzed 18 -24 h following transfection. Tetracycline-inducible SKD1 stable cell lines were generated essentially as described previously (42). Cell lines were maintained in T-REx HEK293 medium containing 125 g/ml Zeocin. To induce protein expression, tetracycline (0.5 g/ml) was added for the indicated length of time. If not otherwise specified, standard conditions for inducing SKD1 expression were 4 -5 h of tetracycline.
Immunostaining and Microscopy-For imaging, cells were plated onto plain (COS-7 cells) or collagen-coated (HEK293 T-REx cells) glass coverslips. Cells were fixed in phosphate-buffered saline (140 mM NaCl, 15 mM phosphate, pH 7.4) containing 4% paraformaldehyde. GFP fusion proteins were imaged directly. For antibody staining, fixed cells were permeabilized in phosphate-buffered saline containing 0.2% Triton X-100, blocked with 5% goat serum, and incubated with primary and secondary antibodies. Cells were examined using a Zeiss Axioplan2 microscope coupled to a Radiance plus confocal laser system (Bio-Rad) with 488 and 543 nm laser lines. Images were merged and assembled using MetaMorph® Imaging System 6 (Universal Imaging Corp.) and Adobe Photoshop 7 (Adobe Systems, San Jose, CA).
Cell Fractionation-For the fractionation shown in Fig. 1C, SKD1(E235Q)-GFP stable cells induced for 4 h were lysed with a ball-bearing homogenizer in 20 mM HEPES, pH 7.5, 1 mM EDTA, 1 mM EGTA, 5 mM ATP, 0.25 M sucrose, 1 mM phenylmethylsulfonyl fluoride, and Complete TM protease inhibitor (Roche Applied Science). Lysates were cleared of nuclei by centrifugation at 500 ϫ g for 3 min, layered on top of a continuous sucrose gradient (0.25-1.75 M in the same buffer), and centrifuged at 100,000 ϫ g for 18 h in a Beckman SW41 rotor. Fractions were collected, separated by SDS-PAGE, and analyzed by Western blotting with the indicated antibodies.
For Fig. 2, SKD1(E235Q)-GFP stable cells induced for 4 h were lysed with a ball-bearing homogenizer in 20 mM HEPES, pH 7.5, 150 mM NaCl, 2 mM MgCl 2 , 5 mM ATP, and 1 mM DTT. Cell lysates were precleared (1000 ϫ g for 5 min) to remove nuclei and unlysed cells and then centrifuged at 100,000 ϫ g for 30 min to collect membranes. To test solubilization conditions, P100 membranes were resuspended in homogenization buffer containing the indicated reagents for 30 min on ice or at 30°C and then repelleted. For Fig. 4A, cell lysates were precleared as above and centrifuged at 20,000 ϫ g for 10 min to yield supernatant and pellet.
For Figs. 5C and 7C, transfected COS-7 cells were lysed with a ball-bearing homogenizer in 20 mM HEPES, pH 7.5, 1 mM EDTA, 1 mM EGTA, 0.25 M sucrose, 1 mM phenylmethylsulfonyl fluoride, and Complete protease inhibitor. Lysates were cleared of nuclei (500 ϫ g, 3 min), layered on top of a continuous sucrose gradient (0.25-1.75 M), and processed as above. For Fig. 5D, lysates were centrifuged at 20,000 ϫ g, and resulting pellets were resuspended in buffer containing 1% Triton X-100 for 30 min prior to a second centrifugation at 20,000 ϫ g.
Magnetic Immunoisolations and Pull-down Assays-SKD1(E235Q)-GFP-expressing cells (induced for 5 h) were resuspended in 20 mM HEPES, pH 7.5, 150 mM NaCl, 2 mM MgCl 2 , 5 mM ATP, 1 mM DTT, 1 mM phenylmethylsulfonyl fluoride, Complete EDTA-free protease inhibitor (Roche Applied Science), and 1% Triton X-100. Cell lysates were sonicated (three bursts using a microtip at output 2, 50% duty cycle) and incubated with Protein G Dynabeads® (Dynal Inc.) prebound to the indicated antibodies for 1 h at 4°C. After washing, proteins bound to the beads were eluted in SDS sample buffer and analyzed by Western blotting with the indicated antibodies.
For the pull-down assays in Fig. 5E, glutathione-Sepharose bound to the indicated protein was incubated with His 6 -tagged hSnf7-1 for 2 h at 4°C in 20 mM Tris, pH 7.4, 100 mM NaCl, 1 mM DTT, 0.2% Triton X-100, and 1 mg/ml bovine serum albumin. After washing, bound proteins were eluted in SDS sample buffer and analyzed by Western blotting using antibody against the T7 tag present in His 6 -hSnf7-1.
Phosphoinositide Binding Assays-PIP Strips TM were purchased from Echelon Biosciences (Salt Lake City, UT) and probed with 0.5 g/ml full-length, N, or C His 6 -hSnf7 proteins according to the manufacturer's recommendations. Bound proteins were detected with T7 tag antibody.

A System for Controlled Expression of SKD1(E235Q)-To
better define the membranes and proteins that SKD1/VPS4B interacts with in the mammalian endosomal pathway, we generated HEK293 cell lines stably expressing GFP-tagged wildtype or ATP hydrolysis-deficient SKD1 under the control of a tetracycline-regulated promoter. Adding tetracycline rapidly induced expression of SKD1-GFP in both wild-type and E235Q mutant cell lines (Fig. 1A). By fluorescence microscopy, we saw that SKD1(E235Q) cells developed first small and later large (up to 2-4 m in diameter) vacuoles rimmed by SKD1(E235Q)-GFP (Fig. 1B, top). We refer to these vacuoles as "EQ compartments." Induced wild-type SKD1-GFP remained cytoplasmic, and these cells did not form vacuoles. Within 4 h of adding tetracycline, all SKD1(E235Q)-GFP-expressing cells contained sizable EQ compartments, and we therefore carried out further studies of how SKD1(E235Q) affects endosomal membranes and associated proteins at this time point. Studies with other AAA ϩ ATPases suggest that this hydrolysis-deficient, ATPbound mutant is likely to bind tightly to and "trap" normal substrates of SKD1 (24,42).

SKD1(E235Q) Binds to Late Endosomes and Causes
Ubiquitinated Proteins to Accumulate-Studies of transiently transfected ATP hydrolysis-deficient SKD1 and VPS4A have shown that these mutants induce and accumulate on enlarged endosomal compartments that contain proteins characteristic of both early and late endosomes (28 -31). To determine whether a particular subset of endosomes are first to recruit SKD1(E235Q), we immunostained EQ compartments in SKD1(E235Q)-GFP-expressing cells induced for 4 h with markers of early and late endosomes as well as other organelles. We found that EQ compartments contain a number of proteins characteristic of late endosomes and lysosomes, including CD63 and LAMP-2 (Fig. 1B). In contrast, the early endosomeassociated protein EEA1 was not present on EQ compartments. EQ compartments also did not contain proteins of other major organelles including the Golgi apparatus. Localization of SKD1(E235Q) to endosomal membranes was further confirmed by subcellular fractionation studies described below (see Fig.  1C).
CD63 is a late endosomal tetraspanin normally found primarily on luminal membranes of the MVB (31, 44 -46). Its rapid accumulation on the limiting membrane of EQ compartments (Fig. 1B) suggests not only that it is late endosomes that first recruit SKD1 but also that SKD1(E235Q) interferes with entry of membrane proteins into the lumen of these organelles. Based on this as well as other recent work on VPS4 (30, 31), we expected that internalization of cargo into the MVB might be generally impaired by SKD1(E235Q). We took advantage of the known link between ubiquitination, deubiquitination, and entry of many proteins into the MVB (47-50) and asked whether expression of SKD1(E235Q) affected ubiquitin homeostasis. Using an antibody specific for ubiquitin-conjugated proteins, we found that EQ compartments had high concentrations of ubiquitinated proteins on their surface (Fig. 1B). A recent study showing that transiently transfected mutant VPS4 causes ubiquitinated CXCR4 chemokine receptors to accumulate supports the idea that at least some of these ubiquitinated proteins correspond to cargo trapped in the limiting membrane of the EQ compartment (51). SKD1(E235Q) Binding to Membranes Is Saturable-Although SKD1(E235Q) was enriched on the rim of the swollen endosomes that we refer to as EQ compartments, its binding to membranes was clearly saturable because much of it remained soluble, particularly as levels of SKD1(E235Q) increased. This could be seen as increased diffuse cytoplasmic fluorescence (compare early and late time points in Fig. 1B, top) and more distinctly in the bimodal distribution of protein in cell homogenates separated on a linear sucrose gradient (Fig. 1C). The two peaks of SKD1(E235Q) represent cytosolic (left) and mem-brane-associated (right) protein. The membrane-associated SKD1 cofractionates with the late endosome/lysosome marker LAMP-2 and is clearly separated from the Golgi/intermediate compartment marker ␤-COP. That SKD1(E235Q) was associated with membranes was confirmed by loading cell homogenate in 52% sucrose on the bottom of a gradient and floating the membranes up into an overlaid 35% sucrose cushion (not shown). Inducing SKD1(E235Q) expression for longer time (overnight) resulted in a higher proportion of soluble protein around similarly stained enlarged vacuoles (not shown). These observations indicate that SKD1(E235Q) binds to specific membranes through receptors and/or cofactors that are present in limited supply.

SKD1(E235Q) Induces Formation of a Large Protein
Complex-To identify protein(s) on the membrane that interact with SKD1(E235Q), we separated membrane-bound from soluble SKD1(E235Q) by centrifugation and tried to define conditions with which to solubilize the membrane-bound SKD1 and associated proteins. Comparison of SKD1(E235Q) from induced HEK293 homogenates with SKD1(E235Q) produced in E. coli showed that the mutant enzyme was pelleted only in the HEK cell lysates, confirming that it pellets because it is associated with something other than itself (i.e. membranes and/or proteins). We tested the effects of different detergents on pelleted SKD1(E235Q) always in the presence of ATP to stabilize interactions between SKD1 and its putative substrates. Adding 1% Triton X-100 did not change the behavior of SKD1(E235Q) (Fig.  2B). This was true at both 0 and 30°C, ruling out association with lipids that are poorly soluble in cold Triton X-100 (i.e. components of lipid rafts). On the other hand, adding 1% cholate (Fig. 2B) solubilized SKD1(E235Q) but also disrupted interactions between it and potentially interacting proteins detected by cross-linking extracts with bis(sulfosuccinimidyl) suberate (not shown). Additional treatments either failed to solubilize SKD1(E235Q) or disrupted protein cross-linking. We therefore decided to look for SKD1-interacting proteins in the Triton X-100-insoluble material.
Interaction of SKD1(E235Q) with Endogenous ESCRT Proteins-As mentioned in the Introduction, studies in both yeast and mammalian systems suggest that components of the ES-CRT-III complex are the best candidates for VPS4 substrates among the class E proteins. There are 10 human proteins with homology to ESCRT-III components that we will call ESCRT-III proteins. They are named either as human homologues of particular yeast proteins or as charged multivesicular body proteins (CHMPs) (Fig. 3A) (38, 52). There is currently no consensus as to which of these (or all) interact with VPS4 because different studies report specific interactions between VPS4 and varying subsets of ESCRT-III proteins (for a summary, see Ref. 35). We cloned and studied two human ESCRT-III proteins, hSnf7-1/CHMP4A and hVps24/CHMP3, as representatives of the two subcomplexes proposed to form the core heterotetrameric ESCRT-III complex in yeast (15). Both hSnf7-1 and hVps24 show about 50% similarity to their yeast homologues and have a conserved bipartite structure in which the N-terminal half of the protein is basic and the C-terminal half is acidic (see Fig. 7A). Both also contain stretches of sequence predicted to form coiledcoils. In addition, hSnf7-1 has conserved proline-containing motifs ( 192 PSVP and 200 PXXP) near its C terminus. Northern blot analysis demonstrated that both hSnf7-1 and hVps24 are widely expressed with particular concentrations in kidney, liver, and skeletal and cardiac muscle (Fig. 3B). This is similar to the distribution of SKD1/VPS4B (53,54), consistent with the idea that these proteins function in a common pathway. The presence of three bands representing hVps24 on the Northern blot may indicate that its transcripts are alternatively spliced.  2. SKD1(E235Q)-GFP is part of a large Triton X-100-insoluble complex. A, postnuclear supernatant of HEK293 cell lines expressing wild-type (WT) or SKD1(E235Q)-GFP (E235Q) was separated by centrifugation at 100,000 ϫ g into soluble (S100) and pelletable (P100) fractions. Samples were analyzed by Western blotting using an antibody against SKD1. For comparison, recombinant SKD1(E235Q) purified from E. coli was treated in the same way. B, P100 fractions from SKD1(E235Q)-expressing cells were resuspended in buffer containing the indicated detergent for 30 min. After centrifugation resulting supernatants (S) and pellets (P) were analyzed by Western blotting for SKD1. TX100, Triton X-100. FIG. 3. hSnf7-1 and hVps24 as representative components of the mammalian ESCRT-III complex. A, phylogenetic tree of human proteins related to yeast ESCRT-III components. Two protein names for each are shown: one is based on the name of the related yeast protein, while the other is a separate nomenclature used for this family of proteins in higher eukaryotes (38). The tree was constructed using the neighbor-joining algorithm (60) in Vector NTI 7 (Invitrogen) with default settings. GenBank TM accession numbers for sequences used are: hSnf7-1, AAQ91193; hSnf7-2, AAQ91194; hSnf7-3, AAQ91195; hVps20, AAQ91196; hVps24, AAF26737; hVps2-1, NP_940818; hVps2-2, AAD34079; hVps46-1, AAG01448; hVps46-2, NP_065145; and hVps60, AAH16698. B, Northern blot showing hSnf7-1 and hVps24 expression in adult human tissues. C, Western blots (WB) showing that affinitypurified antibodies against hSnf7-1 or hVps24 specifically recognize the appropriate Myc-tagged or untagged proteins in lysates of overexpressing COS-7 cells.
To study native hSnf7-1 and hVps24, we generated rabbit polyclonal antibodies using recombinant proteins expressed in E. coli. The antibodies showed specific reactivity with their respective proteins overexpressed in COS-7 cells (Fig. 3C) and also recognized endogenous proteins in HEK293 cells (Fig. 4A). On SDS-polyacrylamide gels, bacterially produced, transfected, and endogenous hSnf7-1 and hVps24 all exhibited the same apparent mobility, slower than predicted based on mass (37-39 versus 25 kDa) and similar to the yeast ESCRT-III proteins (15,25).
To determine whether these proteins are part of the protein complex associated with SKD1(E235Q) on the membrane, we lysed uninduced and induced SKD1(E235Q)-expressing cells and separated the homogenates into soluble and membraneassociated fractions. hSnf7-1 and hVps24 were soluble in uninduced cells (Fig. 4A). However, inducing SKD1(E235Q) expression shifted all of the hSnf7-1 and about 50% of the hVps24 from supernatant to pellet. The ESCRT-I protein Tsg101 also shifted into the pellet, consistent with the results of an earlier study (33). The distribution of unrelated proteins (including ␣-SNAP and EEA1, both partially membrane-bound) was not changed (not shown). These results suggest that SKD1(E235Q) associates with and stabilizes several endogenous mammalian class E proteins on membranes.
To ask more directly whether hSnf7-1 and hVps24 interact with SKD1(E235Q), we immunoisolated SKD1(E235Q)-GFP together with associated proteins. Because of the large size of the SKD1(E235Q)-containing complex, we sonicated Triton X-100-treated cell homogenates and used magnetic beads to collect SKD1(E235Q)-GFP. The same components that were shifted into the pelletable fraction by SKD1(E235Q) expression (Fig. 4A) co-immunoprecipitated with SKD1(E235Q)-GFP, while they did not with a control antibody (Fig. 4B). The unrelated protein ␣-SNAP did not co-immunoprecipitate with SKD1(E235Q)-GFP, indicating that the detected interaction of SKD1(E235Q) with hSnf7-1, hVps24, and Tsg101 is specific. While we cannot say whether SKD1 binds directly or indirectly to each of these proteins, their association with the SKD1(E235Q)-induced protein complex supports a common function in the MVB pathway.
Membrane Binding and Polymerization of hSnf7-1 and hVps24 -The above results are consistent with the idea that ESCRT-III proteins are at least in part responsible for recruiting SKD1 to the membrane and may be its substrates in the MVB pathway. Because ESCRT-III proteins, like all class E proteins identified to date, are primarily soluble under normal conditions, we need to better understand how their association with membranes is controlled to define how they might recruit SKD1 to the membrane. In addition, we need to understand how their assembly into an ESCRT-III complex is controlled.
We began by examining the distribution of overexpressed hSnf7-1 and hVps24 in transiently transfected COS-7 cells, a The distribution of the tagged proteins across the gradient was detected by Western blotting. The peak of membraneassociated hSnf7-1 comigrated with LAMP-2 (not shown). D, COS-7 cells expressing hSnf7-1 or hVps24 were lysed and centrifuged at 20,000 ϫ g for 20 min to collect pelletable protein. Pelleted protein was resuspended in buffer containing 1% Triton X-100, incubated for 30 min, and recentrifuged to yield Triton X-100 (TX100) supernatant (S) and pellet (P). E, interaction of purified hSnf7-1 proteins with each other. His 6 -tagged hSnf7-1 was incubated with glutathione beads prebound to GST-hSnf7-1-FL (FL), GST-hSnf7-C (C), or nothing (bead). Bound His 6 -tagged hSnf7-1 was visualized by Western blotting. system in which the concentration of transfected protein greatly exceeds that of potential endogenous binding partners. Both proteins with GFP fused to their C terminus accumulated on the limiting membranes of enlarged vacuoles. hSnf7-1-GFP was localized primarily on these vacuolar membranes (Fig. 5A,  top), while hVps24-GFP was both on the membranes and diffusely distributed throughout the cell (Fig. 5A, bottom). Similar behavior of other ESCRT-III proteins tagged with fluorescent proteins (GFP and red fluorescent protein) has been noted by others with the conclusion that adding fluorescent protein tags to ESCRT-III components generates dominant inhibitors for this pathway (33,38,40,52). To determine what role the GFP tag plays in these effects, we replaced GFP with C-terminal Myc or N-terminal FLAG epitope tags. In transfected cultures expressing approximately equivalent levels of protein, hVps24-Myc was diffusely distributed throughout the cell (and there were no enlarged vacuoles), but hSnf7-1-Myc still induced and accumulated on enlarged vacuoles (Fig. 5A). FLAG-tagged proteins behaved similarly (data not shown). To exclude any effects of epitope tags, we expressed untagged hSnf7-1 and hVps24. Again hVps24 was diffusely distributed, while hSnf7-1 induced vacuolation and bound to membranes (Fig. 5A). To confirm these results biochemically, we lysed cells expressing hVps24 or hSnf7-1 and fractionated the homogenates on sucrose density gradients (Fig. 5C). hVps24 remained entirely at the top of the gradient with soluble proteins, while hSnf7-1 was partially associated with membranes. Interestingly, although in most cells hSnf7-1 localized primarily to internal membranes (Fig. 5A), we also found cells in which it was present on the plasma membrane (Fig. 5B). Explanations for this may relate in part to differing levels of protein expression in each cell but may also include variations in the signaling activity of individual cells. We conclude that a simple increase in the concentration of hSnf7-1 in COS-7 cells is enough to promote association of a significant portion of the protein with membranes. For hVps24, this is only true when the protein is attached to GFP.
ESCRT-III proteins in yeast have been proposed to assemble into a heterotetrameric ESCRT-III complex on endosomal membranes (15). The complex is large and cannot be solubilized by Triton X-100. To explore the possibility that hSnf7-1 in transfected COS-7 cells might form a homopolymeric ESCRT-III complex, we carried out fractionation and detergent solubilization studies. hSnf7-1 that pelleted at 20,000 ϫ g could not be solubilized by Triton X-100 (Fig. 5D). This indicates that an ESCRT-III-like complex can be formed by a single ESCRT-III protein presumably by binding to itself. In support of this possibility, yeast two-hybrid studies have indicated that most Snf7 isoforms bind to themselves (33,35,39). To determine directly whether hSnf7-1 binds to itself, we purified recombinant proteins from E. coli and carried out pull-down assays. GST-tagged hSnf7-1 (FL) bound His 6 -tagged hSnf7-1, while control protein and beads did not (Fig. 5E). Interestingly the extent of hSnf7-1 binding to itself was increased by expressing GST-and His 6 -tagged hSnf7-1 together in E. coli (not shown).
Overexpressed hSnf7-1 Impairs Endosomal Function-Because the vacuoles induced by overexpressing all versions of hSnf7-1 and GFP-tagged hVps24 were similar to those formed by SKD1(E235Q), we wondered whether they were derived from common endosomal membranes. We immunostained COS-7 cells transfected with hSnf7-1-Myc for CD63 and found that, similar to SKD1(E235Q)-expressing cells, most of the large vacuoles had CD63 on their periphery (Fig. 6A, top row). This shows that late endosomal membranes recruit hSnf7-1. To determine whether the endosomes swollen by hSnf7-1 overexpression were impaired in cargo processing, we again stained cells with a ubiquitin conjugate-specific antibody and found that vacuoles induced by hSnf7-1 were rimmed with ubiquitinconjugated proteins (Fig. 6A, bottom row). The tight colocalization of hSnf7-1 with these membrane-associated ubiquitin conjugates indicates that hSnf7-1 is likely to be intimately connected with components that recognize ubiquitin signals in the MVB pathway.
Overexpressed hSnf7-1 Recruits SKD1-Do SKD1 and hSnf7-1 interact directly? The spontaneous assembly of hSnf7-1 homopolymers on membranes gives us an opportunity to ask whether SKD1 and this individual ESCRT-III protein interact. In cells cotransfected with hSnf7-1 and SKD1(E235Q)-GFP, hSnf7-1 colocalized with SKD1(E235Q) on the limiting membranes of endosomes and along the plasma membrane (Fig. 6B, top row). Because these could be binding to the same membranes for different reasons, we also looked for recruitment of the normally soluble wild-type SKD1 by hSnf7-1 on the membrane. Interestingly hSnf7-1 on enlarged endosomes retained some cotransfected wild-type SKD1-GFP (Fig.  6B, bottom row) demonstrating that membrane-bound hSnf7-1 polymer recruits SKD1.
hSnf7-1 N-terminal Fragment Binds Membranes and Inhibits Endosomal Function-What causes hSnf7-1 to associate with membranes? One possibility is that hSnf7-1 could be post-translationally modified with a hydrophobic moiety such as the myristoylation involved in bringing Vps20 (CHMP6 in mammalian systems) to the membrane (15,55). There are, however, no consensus sequences for lipid modification in hSnf7-1. Another, more likely, possibility is that hSnf7-1 associates with membranes by binding directly to specific phospholipids. The N-terminal half of rat Vps24 has recently been reported to bind phosphatidylinositol 3,5-bisphosphate via a basic motif that is conserved in most ESCRT-III proteins including hSnf7-1 (56). Interactions between hSnf7-1 and other lipids are also possible. Finally, association of hSnf7-1 with the membrane could be mediated by binding to a small number of proteins that nucleate its assembly into a large membraneanchored protein complex.
To define elements important for recruiting hSnf7-1 to membranes, we prepared hSnf7-1 protein fragments. Based on the charge distribution of the protein (Fig. 7A), we divided it into two pieces: an N-terminal basic fragment (hSnf7-N, residues 1-116) and a C-terminal acidic and proline-rich fragment (hSnf7-C, residues 117-222). Following expression in COS-7 cells, we observed that hSnf7-N was entirely bound to membranes, while hSnf7-C remained cytosolic (Fig. 7, B and C). hSnf7-N, like full-length hSnf7-1, perturbed endosomal membrane trafficking as seen by the formation of enlarged vacuolar structures that accumulated ubiquitin-conjugated proteins on their surface (see Fig. 8).
To begin to understand why hSnf7-N bound so efficiently to membranes, we tested whether it and full-length hSnf7-1 interact with specific phospholipids in vitro (Fig. 7, D and E). Purified hSnf7-N and full-length hSnf7-1 bound to phosphoinositides in a blot overlay experiment, while hSnf7-C did not bind to any phospholipids tested. hSnf7-N bound to both monoand diphosphorylated phosphoinositides including most prominently phosphatidylinositol 3-phosphate and phosphatidylinositol 3,5-bisphosphate, while full-length hSnf7-1 bound best to phosphatidylinositol 3-phosphate. These interactions are similar to, although less specific than, those previously reported for rat Vps24 (56). However, we were unable to reverse the interaction of either full-length hSnf7-1 or hSnf-N with membranes by treating transfected cells with the phosphatidylinositol 3-kinase inhibitor wortmannin, indicating that phosphatidylinositol 3-phosphate-containing phosphoinositides may not be necessary for keeping these proteins bound to membranes (not shown).
hSnf7-1 N-terminal Fragment Forms Polymers and May Affect Membrane Structure-The above results show that the N-terminal half of hSnf7-1 promotes membrane binding. Further examination of cells expressing this fragment revealed additional features of the behavior of FLAG-tagged hSnf7-N. For one, hSnf7-N was present not only on enlarged endosomes (Fig. 7B) but also (and often more prominently) on the plasma membrane (Fig. 8, A-C). It was sometimes diffusely distributed across the membrane (Fig. 8A), sometimes concentrated in round structures in the plane of the membrane (Fig. 8B), and sometimes present in filopodia-like structures that protruded FIG. 7. The N-terminal half of hSnf7-1 associates with membranes and binds phosphoinositides. A, stick diagram of hSnf7-1 showing its bipartite charge distribution and predicted coiledcoil motifs (black boxes, predicted by COILS (61)). aa, amino acid. B, cellular distribution of hSnf7-1 N and C fragments. COS-7 cells shown were transfected with FLAG-tagged hSnf7-N or hSnf7-C and immunostained with FLAG antibody. The scale bar represents 10 m. C, distribution of hSnf7-N in homogenates of transfected COS-7 cells separated on a linear 0.25-1.75 M sucrose gradient. D, purified hSnf7-1-FL, -N, and -C proteins used for PIP blot overlay assays separated by SDS-PAGE and visualized by staining with Coomassie Blue. E, PIP Strips TM showing phosphoinositide binding by purified hSnf7-1-FL, -N, and -C. Bound hSnf7-1 was detected by Western blotting. Lipids are spotted as follows: 1, lysophosphatidic acid; 2, lysophosphocholine; 3, phosphatidylinositol; 4, phosphatidylinositol 3-phosphate; 5, phosphatidylinositol 4-phosphate; 6, phosphatidylinositol 5-phosphate; 7, phosphatidylethanolamine; 8, phosphatidylcholine; 9, sphingosine-1-phosphate; 10, phosphatidylinositol 3,4-bisphosphate; 11, phosphatidylinositol 3,5-bisphosphate; 12, phosphatidylinositol 4,5-bisphosphate; 13, phosphatidylinositol 3,4,5-trisphosphate; 14, phosphatidic acid; 15, phosphatidylserine; 16, blank. out from the surface of the plasma membrane (Fig. 8C). We wondered whether the uneven distribution of this fragment might indicate that it too assembles into higher order polymers. Indeed we found that adding Triton X-100 to transfected cells before fixation did not significantly affect the distribution of hSnf7-N (Fig. 8D), suggesting that N-N interactions are likely to be responsible for forming Triton X-100-resistant hSnf7-1 polymers. Additional evidence that N-N interactions are responsible for hSnf7-1 polymerization came from the tight colocalization seen between coexpressed full-length and hSnf7-N (Fig. 8, G-I), while there was no colocalization between the coexpressed full length and hSnf7-C (Fig. 8, J-L).
One of the most striking features of cells expressing hSnf7-N was the presence of filopodia-like structures both along the cell periphery and on internal membranes (Fig. 8, C, E, and F). These can be seen to extend out from the cell (Fig. 8C) or into intracellular vacuoles (themselves rimmed by ubiquitin-conjugated proteins) (Fig. 8, E and F). That membranes are bent out from the cell or into vacuoles by hSnf7-N raises the possibility that such distortion might be involved in internalizing membrane into the MVB.
hSnf7-1 N-terminal Fragment Does Not Colocalize with SKD1-We showed above that hSnf7-1 colocalizes with coexpressed wild-type and mutant SKD1 (Fig. 6B). We wondered whether hSnf7-N would also recruit SKD1. In cells co-expressing hSnf7-N and SKD1(E235Q), there was little or no colocalization of the two (Fig. 8, M-O). This demonstrates that the C-terminal half of hSnf7-1 is necessary for binding to SKD1. This is consistent with a recent study in which the interaction between SKD1 and one of the other ESCRT-III components, mVps2/CHMP2A, was shown not to require a predicted coiledcoil domain in the N-terminal half of mVps2/CHMP2A (57).

DISCUSSION
While ESCRT-III proteins and the AAA ϩ ATPase SKD1/ VPS4B have been implicated in MVB biogenesis in both yeast and mammalian systems, much remains to be learned about the relationships between these proteins and their effects on endosomal membranes. In this investigation, we studied the interaction of two human ESCRT-III proteins, hSnf7-1 and hVps24, with membranes and SKD1. Based on our analysis of hSnf7-1, we propose that the properties of a single ESCRT-III protein, the ability to bind membranes and homopolymerize via an N-terminal basic half and to bind SKD1 and other factors via a C-terminal acidic half, reveal the underlying organization of proteins in the ESCRT-III complex (Fig. 9). Although studies in yeast have suggested a model in which the ESCRT-III complex contains several (possibly four) proteins, this complex differs from the more clearly defined ESCRT-I and ESCRT-II complexes because there is as of yet no defined size or precise stoichiometry among its components (15,25,35). Our analysis of the large detergent-resistant polymers formed by hSnf7-1 alone suggests that there may not be an obligate stoichiometry among components of an ESCRT-III complex. Instead the precise composition of ESCRT-III complexes may depend on regulated exposure of membrane-binding motifs in different ES-CRT-III proteins.
Assembly of ESCRT-III Polymers on Membranes-We found that overexpressed hSnf7-1/CHMP4A spontaneously formed membrane-associated, detergent-insoluble polymers under conditions in which its concentration greatly exceeded that of potential endogenous binding partners (i.e. other components of the ESCRT-III complex, ESCRT-I and ESCRT-II complexes, endogenous VPS4, and others). These polymers appeared therefore to be composed primarily of hSnf7-1. Binding between differently tagged forms of purified hSnf7-1 expressed in E. coli supports the proposal that such polymers assemble via direct interactions between hSnf7-1 subunits (Fig. 5). In further support of this, we have recently seen that hSnf7-1 forms homogeneous linear polymers on cell membranes that can be resolved by quick freeze, deep etch electron microscopy. 2 Overexpressed hSnf7-1, with or without attached epitope tags, inhibited MVB maturation as seen by swelling of endosomes (Figs. 5 and 6), accumulation of ubiquitinated proteins on these enlarged endosomes (Fig. 6), and inhibition in the rate at which internalized EGF was degraded. 3 These findings agree with other recent studies that have shown that overexpressing epitope-tagged ESCRT-III proteins perturbs the endosomal system (33,38,52,56,58) and interferes with enveloped virus budding (33,39,40). A common theme is that the increased steady-state association of ESCRT-III proteins with membranes caused by overexpression (and in many cases exacerbated by fused protein tags as discussed further below) is correlated with dysfunction in the MVB pathway.
Binding of hSnf7-1 to membranes and to itself is mediated by the N-terminal half of the protein, which contains a predicted coiled-coil motif and many basic residues (Figs. 7 and 8). Expressing this N-terminal fragment in transfected cells perturbed endosome structure, similar to what has previously been reported for the N-terminal fragment of rat Vps24 (56). Expressing the C-terminal half of hSnf7-1 had no effect. Deleterious effects of overexpressed ESCRT-III proteins are therefore likely to be mediated by their N-terminal domains and proteins or lipids with which these domains interact.
The N-terminal half of hSnf7-1 was often not uniformly distributed on membranes of transfected cells but was instead concentrated in spikelike structures and bright round patches (Fig. 8). The spikelike structures were present along the plasma membrane and less often on and pointing into enlarged internal vacuoles. Their shape suggests that this domain of hSnf7-1 could help bend membranes away from the cytoplasm. Membrane deformation is essential for invagination into the MVB, and how it occurs is not understood. If ESCRT-III polymers facilitate this membrane distortion, they might cooperate with lipids such as lysobisphosphatidic acid to promote MVB invagination (59). Further studies of how N-domains from hSnf7-1 and other ESCRT-III proteins bind and affect membranes and interact with each other are clearly required to explore these possibilities.
Regulating ESCRT-III Polymerization and Membrane Association-There are significant differences in the subcellular distribution of different ESCRT-III proteins studied here and elsewhere. At steady state, endogenous hSnf7-1 and hVps24 were largely soluble, consistent with the distribution of their homologues in yeast (15,25). Expressing SKD1(E235Q) shifted endogenous hSnf7-1 and hVps24 to the membrane (Fig. 4, also seen by microscopy 3 ). When overexpressed, hSnf7-1 was partially associated with membranes even in the absence of SKD1(E235Q) (Fig. 5). The N-terminal fragment of hSnf7-1 was entirely associated with membranes (Fig. 7). Overexpressed hVps24 was soluble except when fused to GFP when it became partially membrane-associated and perturbed endosomal structure (Fig. 5). In other recent studies, overexpressed ESCRT-III proteins have been found to behave similarly (33,38,40,58). Based on the fact that the N-terminal fragments of both hSnf7-1 (this study) and rat Vps24 (56) bind to membranes, we suggest that the differences in membrane association of ESCRT-III proteins in different situations reflect variations in exposure of the N-terminal domains of the proteins for interaction with membranes and with each other. This hypothesis requires further testing but is consistent with available data and provides an explanation for the variability in the behavior of these proteins in different systems.
How might binding of ESCRT-III proteins to membranes normally be controlled? If intramolecular contacts between Nand C-terminal halves of ESCRT-III proteins keep the N-terminal half from binding to membranes, then it seems likely that proteins that interact specifically with either half (and perhaps especially with the C-terminal half) might help "open" the protein and expose the membrane binding domain (Fig. 9). In addition, exposed N-terminal domains might propagate their open state by binding to N-terminal domains in other ESCRT-III proteins. Binding partners for individual ESCRT-III proteins include other ESCRT-III proteins, ESCRT-II proteins, the ESCRT-I proteins Vps28 and Vps37, Bro1/Alix/AIP1, Doa4, Rim20, Vta1/SBP1, and Vps4 (for a summary, see Ref. 35). Little is known about which parts of the ESCRT-III protein(s) are involved in these interactions. One well studied interaction that appears to involve the C-terminal half of an ESCRT-III protein is the binding between Snf7 and Bro1 (Alix or AIP1 in mammalian systems) (26,33,40,50). A single point mutation (L217A) near the C terminus of CHMP4B/hSnf7-2 abolished binding to AIP1, suggesting that this interaction involves the C-terminal half of hSnf7-2. Further evidence in support of this comes from the work of Strack et al. (40) who showed that mutant VPS4 prevents binding of AIP1 to hSnf7-1/CHMP4A. One explanation for this observation is that VPS4 and AIP1 compete for a common binding site in the C-terminal half of hSnf7-1. The conserved proline-containing motifs found in the C-terminal half of hSnf7-1 (and in different forms in other ESCRT-III proteins, see Fig. 9) are possible binding sites 2 P. I. Hanson, R. Roth, Y. Lin, and J. E. Heuser, manuscript in preparation. 3 Y. Lin and P. I. Hanson, unpublished observations.
FIG. 9. Model of interactions among hSnf7-1, SKD1, and membranes. hSnf7-1 assumes a proposed "closed" conformation in the cytosol, potentially stabilized by interactions between oppositely charged N-and C-terminal halves of the protein. When exposed by "opening" (or truncation as in this study), the N-terminal half of the protein binds both to membranes and to itself to form large membrane-anchored homopolymers that may distort the underlying membrane structure. hSnf7-1 polymers are likely to be dissociated by SKD1/VPS4 via interactions between this enzyme and the C-terminal half of hSnf7-1. Opening of hSnf7-1 may be promoted by proteins that bind to its C-terminal half. These could include putative upstream components of the ESCRT machinery, AIP1/Alix, and any factors that bind to conserved prolinecontaining motifs ( 192 PSVP 195 and 200 PXXP 203 in hSnf7-1). Other ES-CRT-III proteins are likely to undergo similar cycles of polymerization and disassembly and probably form heteropolymeric complexes with hSnf7-1.
for other proteins in the MVB pathway, and their role in regulating ESCRT-III assembly needs to be examined.
Role for SKD1 and ESCRT-III in MVB Biogenesis-Our data showing that SKD1(E235Q) rapidly blocks MVB maturation while trapping endogenous components of the ESCRT-I and -III complexes in a large membrane-anchored protein complex are consistent with current models that suggest that normal function and interactions among these proteins are required for successful invagination into the MVB (4,31). How does SKD1 promote MVB maturation, and in particular what are its substrates? hSnf7-1 directly recruits SKD1 to the membrane (Fig.  6). This agrees with other recent reports that show that several ESCRT-III proteins colocalize with SKD1(E235Q) on membranes (57,58) and with results from protein-protein interaction studies that indicate that VPS4 interacts with a number of different ESCRT-III proteins (for a summary, see Ref. 35). It seems likely that several, and possibly all, ESCRT-III proteins can, under the right circumstances, recruit VPS4.
The presumed function of VPS4 toward ESCRT-III-containing protein complexes is to disassemble them, releasing individual ESCRT-III proteins from the membrane. Our domain analysis shows that SKD1 interacts with the C-terminal half of the ESCRT-III protein (hSnf7-1), suggesting that this half of the protein provides a "handle" that SKD1 might pull on to disassemble an ESCRT-III complex. This is at present only a model because we were unable to see disassembly of polymerized hSnf7-1 by wild-type SKD1 in transfected cells. 3 Possible explanations for this include that the ratio between coexpressed SKD1 and hSnf7-1 substrate was not correct (an important consideration given the requirement for Vps4 to oligomerize to be active (25)) or that SKD1 acts more efficiently on heteropolymeric than on homopolymeric ESCRT-III complexes. These possibilities will need to be explored as we move toward in vitro reconstitution of SKD1-catalyzed rearrangement or disassembly of ESCRT-III polymers.
In summary, we found that homopolymers of an individual ESCRT-III protein, hSnf7-1, have many of the properties presently associated with the heterotetrameric ESCRT-III complex in yeast. A better understanding of how these and other ESCRT-III polymers affect cellular membranes will be an important next step in defining the reactions responsible for invagination into the MVB.