Structural Basis for Budding by the ESCRT-III Factor CHMP3

Developmental Cell 10, 821–830, June, 2006 ª2006 Elsevier Inc.

DOI 10.1016/j.devcel.2006.03.013

Structural Basis for Budding by the ESCRT-III Factor CHMP3 Tadeusz Muzioł,1,4 Estela Pineda-Molina,1,4 Raimond B. Ravelli,1 Alessia Zamborlini,2 Yoshiko Usami,2 Heinrich Go¨ttlinger,2 and Winfried Weissenhorn1,3,* 1 European Molecular Biology Laboratory 6 rue Jules Horowitz 38042 Grenoble France 2 Program in Gene Function and Expression Program in Molecular Medicine University of Massachusetts Medical School Worcester, Massachusetts 01605

Summary The vacuolar protein sorting machinery regulates multivesicular body biogenesis and is selectively recruited by enveloped viruses to support budding. Here we report the crystal structure of the human ESCRT-III pro˚ resolution. The core structure of tein CHMP3 at 2.8 A CHMP3 folds into a flat helical arrangement that assembles into a lattice, mainly via two different dimerization modes, and unilaterally exposes a highly basic surface. The C terminus, the target for Vps4-induced ESCRT disassembly, extends from the opposite side of the membrane targeting region. Mutations within the basic and dimerization regions hinder bilayer interaction in vivo and reverse the dominant-negative effect of a truncated CHMP3 fusion protein on HIV-1 budding. Thus, the final steps in the budding process may include CHMP protein polymerization and lattice formation on membranes by employing different bilayer-recognizing surfaces, a function shared by all CHMP family members. Introduction Complex and conserved protein machineries largely encoded by class E vacuolar protein sorting (VPS) genes regulate the sorting and trafficking of proteins into multivesicular bodies (MVB). This involves the delivery of cargo such as epidermal growth factor receptor (EGF) (Gordon et al., 1978) into small vesicles that are formed at the endosomal membrane and are released into the endosome, giving rise to MVB. MVB structures then fuse with late endosomes and lysosomes, thus delivering their cargo for degradation (Futter et al., 1996). Tagged by ubiquitin, protein cargo is first recognized by the HRS-STAM complex that recruits ESCRT-I to the endosomal membrane (Bache et al., 2003; Katzmann et al., 2001, 2003). This in turn activates assembly of ESCRT-II and -III complexes (Babst et al., 2002a, 2002b; Gruenberg and Stenmark, 2004; Katzmann et al., 2002), which have been implicated in protein sorting, vesicle formation, and budding of vesicles from the endosomal membrane (Babst et al., 2002a, 2002b; Gruenberg *Correspondence: [email protected] 3 Lab address: http://www.embl-grenoble.fr/ 4 These authors contributed equally to this work.

and Stenmark, 2004; Katzmann et al., 2002). ESCRT-III contains six members in yeast, forming three distinct subcomplexes, namely Snf7p/Vps20p (CHMP4/6), Vps2p/Vps24p (CHMP2/3), and Did2p/Vps60p (CHMP1/ 5) (Babst et al., 2002a). Similarly, human cells contain ten ESCRT-III proteins, the CHMP family (CHMP1–6; charged multivesicular body protein), with two isoforms of CHMP1 and 2 and three isoforms of CHMP4 (Howard et al., 2001; Martin-Serrano et al., 2003; von Schwedler et al., 2003). Although the exact sequence of recruitment is unknown, most molecular interactions between CHMP members have been described. CHMP6 is N-myristoylated, recruited by ESCRT-II, and interacts with CHMP4 members, which in turn bind CHMP3. CHMP3 recognizes the phospholipid PtdIns(3,5)P2 (Whitley et al., 2003), which has been implicated in vacuole size control in yeast (Rudge et al., 2004). Furthermore, CHMP3 interacts with CHMP2 members (Morita and Sundquist, 2004) and AMSH, thus linking CHMP3 to the HRS-STAM complex (McCullough et al., 2006). Based on the network of interactions between CHMP family members, it has been suggested that upon interaction with cellular membranes, single proteins and/or subcomplexes assemble into a distinct protein network that regulates late steps in budding (Babst et al., 2002a; Katzmann et al., 2002; Lin et al., 2005; Morita and Sundquist, 2004). Membrane targeting of CHMP proteins in turn recruits the AAA-type ATPase Vps4, which disassembles and recycles ESCRT complexes (Babst et al., 1998, 2002a; Fujita et al., 2003; Lin et al., 2005; Morita and Sundquist, 2004; Strack et al., 2003; von Schwedler et al., 2003; Yoshimori et al., 2000). MVB formation involves vesicle budding away from the cytosol and is thus morphologically similar to the budding mechanism of enveloped viruses. HIV-1, for example, enters the MVB pathway via Gag p6 late domain interactions with ESCRT-I Tsg101 and Alix/AIP1, which helps to recruit downstream factors such as ESCRT-III members and Vps4 (Garrus et al., 2001; Martin-Serrano et al., 2003; Strack et al., 2003; von Schwedler et al., 2003). Notably, expression of a number of ESCRT-III CHMP isoforms, including CHMP3, can exert a dominant-negative effect on retrovirus budding. Its induced phenotype is similar to the one observed for expression of inactive Vps4, indicating that ESCRT-III is absolutely required for retrovirus release and most likely acts upon the final steps of the budding process (Martin-Serrano et al., 2003; Strack et al., 2003; von Schwedler et al., 2003). Here we report the crystal structure of a CHMP protein family member and relate its structure to its proposed role in the membrane fission process that leads to release of endosomal vesicles and enveloped viruses from cellular membranes. Results CHMP3 Dimerizes upon Removal of a C-Terminal Region In Vitro Chemical crosslinking and gel filtration chromatography reveal that recombinant human full-length CHMP3

Developmental Cell 822

(residues 1–222) and an N- and C-terminal truncated form of CHMP3, containing residues 9–183 (CHMP3[9– 183]), are both monomeric in solution. No significant bands corresponding to potential dimeric forms of CHMP3, which should migrate at w60 kDa (CHMP3WT) and w40 kDa (CHMP3[9–183]), are observed upon crosslinking with ethylene glycol bis-succinimidylsuccinate (EGS) (Figure 1A, lanes 1–6). However, CHMP3, containing residues 9–164 (CHMP3[9–164]), lacking most of the acidic region (residues 141–222), forms dimers under the same crosslinking conditions as indicated by the appearance of a new band migrating at w35 kDa (the monomer migrates at w17 kDa) (Figure 1A, lanes 7–9). In addition, dynamic light scattering yielded hydrodynamic particle radius values of w2.5 nm for both CHMP3 and CHMP3(9–183), and w3.8 nm for dimeric CHMP3(9–164). These results indicate that CHMP3 is a monomer in solution and removal of part of the C terminus leads to spontaneous dimerization. The Crystal Structure of CHMP3 To gain insight into the structural requirements for CHMP protein function including membrane interaction, we crystallized CHMP3(9–183). The structure was determined from a selenomethionine-containing crystal using the single wavelength anomalous dispersion method and diffraction data to 2.8 A˚ resolution, which resulted in an interpretable electron density map (Figure 2A). The P1 crystal unit cell contains four monomers (Figure 2B) forming two identical antiparallel dimers, with dimensions of 88 3 43 3 35 A˚ (Figure 2C). The CHMP3(9–183) monomer folds into a helical arrangement, which is dominated by a 70 A˚ long helical hairpin that together with two short helices forms a four-helical bundle (Figure 2B). A fifth helical segment is connected to the core via a 20 residue long largely disordered linker and is positioned perpendicularly to the core (Figures 2C and 2D). This arrangement generates a flat molecule with the C-terminal ends emanating from one side of the structure (Figure 2D). Dimerization is mainly mediated by helix 2-helix 2 interactions, with some contributions from helices 1 and 5. Although the flat interface covers an extensive total surface of 3476 A˚2, approximately half of the residues at the interface are not in close enough contact (>3.5 A˚) to mediate direct interactions (Figure 3A). Dimerization involves mostly van der Waals contacts and a few polar interactions such as a salt bridge between E42 OE1 and R99* NH2 (*, second monomer) and hydrogen bonds between the carbonyl of Q38 and R99* NH1 as well as N85 OD1 and N85* ND2 (Figure 3C). To test whether this dimerization interface exists in solution, centrally located residues were mutated in CHMP3(9– 164) (mutant 1, M1, Y78 to R, A79 to E), as CHMP3(9– 183) does not dimerize in solution (Figure 1A). Purification of M1 on gel filtration reveals two overlapping peaks, a minor one at 10.1 ml and a major peak at 11.2 ml, respectively. Chemical crosslinking shows that the 11.2 ml peak is monomeric, as M1 cannot be efficiently crosslinked and remains monomeric (Figure 1B, lanes 4–6), while the 10.1 ml peak is efficiently crosslinked to a dimeric form migrating at w37 kDa (Figure 1B, lanes 1–3). Extension of the changes in the interface to generate a mutant M1.2 (Y78R, A79E, A82Y; Figure 3A) re-

Figure 1. Removal of C-Terminal Sequences Activates the Transition from a Monomeric to a Dimeric Form of CHMP3 (A) Chemical crosslinking of recombinant CHMP3 wild-type (lanes 1–3), CHMP3(9–183) (lanes 4–6), and CHMP3(9–164) (lanes 7–9) (no EGS, lanes 1, 4, and 7; 0.5 mM EGS, lanes 2, 5, and 8; 5 mM EGS, lanes 3, 6, and 9); complete dimerization is only observed in the case of CHMP3(9–164). Crosslinking of CHMP3 and CHMP3(9– 183) reveals faint bands corresponding to the size of dimers (*) and internal crosslinking of the monomeric form as evident from a second monomer band migrating slightly faster on SDS-PAGE. (B) Chemical crosslinking of the mutant M1 (Y78R, A79E) reveals monomers and dimers. Lanes 1–3 show crosslinking of a sample from the minor 10.1 ml gel filtration peak, and lanes 4–6 from the major 11.2 ml gel filtration peak. (C) Chemical crosslinking of mutant M2 (K54S, Q56N, V59E, V62D, L63D) shows monomers in the major 11.4 ml gel filtration peak (lanes 1–3) and dimer formation in the minor 10.4 ml gel filtration peak (lanes 4–6). (D) Chemical crosslinking of M3 (a combination of M1 and M2) reveals only monomeric CHMP3(9–164) (lanes 1–3). In (B)–(D), lanes 1 and 4, no EGS; lanes 2 and 5, 1 mM EGS; and lanes 3 and 6, 5 mM EGS.

vealed an identical monomer-dimer mixture as shown for M1 (data not shown). These results suggested that this dimerization interface (dimer 1) is used in vitro but mutation of central residues is not sufficient to completely prevent dimerization. Because the crystal packing presents a second dimerization mode via the tips of the helical hairpins (Figure 2E, dimer 2), we mutated residues within this interface (Figure 3B), which is formed by a small hydrophobic core (V59, V62, L63) that is capped by polar contacts on either side (E40 OE1-K54 NZ, Q56 OE1-E66 OE1) (Figures 3D). Purification of M2 (K54S, Q56N, V59E, V62D, L63E) by gel filtration revealed two overlapping peaks, a minor one eluting at w10.4 ml and a major one eluting at w11.4 ml, similar to the elution profile of M1. Chemical crosslinking of samples from both peaks identified that peak 1 contains a dimeric form of M2, as it can be crosslinked to a new

Crystal Structure of CHMP3 823

Figure 3. Close-Up of the CHMP3 Dimerization Interface

Figure 2. Crystal Structure of Dimeric Human CHMP3 (A) Stereo image of the experimental electron density map obtained from SAD phases and noncrystallographic symmetry averaging showing a central fragment of the long helical hairpin as an allatom model. (B) Crystal structure of CHMP3(9–183). Ribbon representation of the CHMP3(9–183) monomer that is dominated by a 70 A˚ long helical hairpin that together with the short helices 3 and 4 forms a four-helical bundle. A fifth helix, connected via a long disordered loop (residues w140–161), packs perpendicular to the helical bundle structure and its orientation is determined by dimerization contacts. Secondary structure elements and residue numbers are indicated. The disordered loop region connecting helices 4 and 5 is drawn as a dotted line. (C) Ribbon representation of the CHMP3(9–183) dimer (dimer 1) in yellow and orange. Antiparallel dimerization is mediated mainly by interactions via the long helical hairpin and additional contribution of helix 5. Dimerization generates a flat assembly of 88 3 43 3 35 A˚. (D) Ribbon diagram of dimer 1 viewed from the side. This shows the perpendicular packing of helix 5 against the tip of the helical hairpin and the orientation of the C termini toward one side of the flat molecule. (E) The crystal lattice reveals a dimerization mode via the tips of the helical hairpins (dimer 2). Such dimer formation in vitro was confirmed by mutagenesis analyses (M2).

(A) Surface representation of the CHMP3 monomer. Residues in close contact (<3.5 A˚) within the dimer 1 interface are labeled in yellow and those buried but further apart than 3.5 A˚ are shown in orange. (B) Surface representation of the CHMP3 monomer rotated by 180º along the y axis showing the dimer 2 interface. Same color coding as in (A). (C) Close-up of the dimer 1 interface of one half related by a 2-fold axis. The dimer interface occupies a total surface of 3476 A˚2. (D) Close-up of the dimer 2 interface. Residues in close contact (<3.5 A˚) are drawn with side chains and polar interactions are indicated by dashed lines in (C) and (D). The dimer interface occupies a total surface of 840 A˚2.

band migrating at w37 kDa on SDS-PAGE (Figure 1C, lanes 4–6), while the 11.4 ml peak contained only the monomeric form that did not crosslink (Figure 1C, lanes 1– 3). When we mutated both potential dimerization regions simultaneously, the resulting M3 mutant protein eluted as a single peak (11.9 ml) from a gel filtration column that remained monomeric upon chemical crosslinking (Figure 1D). This indicates that CHMP3(9–164) dimerizes in solution, most likely by employing both dimerization interfaces. It is thus conceivable that the crystallization conditions induced the formation of dimers by dislodging the C-terminal acidic region of monomeric CHMP3(9– 183) from its N-terminal core. This mechanism is

Developmental Cell 824

Figure 4. CHMP Sequence Conservation (A) Structure-based sequence alignment of human CHMP3 (GenBank accession number NM_016079), CHMP1A (AF281063), CHMP2A (NM_014453), CHMP4A (Q9BY43), CHMP5 (NM_016410), and CHMP6 (NM_024591). Residues implicated in CHMP3 dimerization are indicated by red asterisks (dimer 1). Residues involved in further observed dimer-dimer interactions are marked by green triangles (convex dimer; Figure 5D), blue triangles (concave dimer; Figure 5D) and black triangles (dimer 2; Figures 3D and 5D). The N- and C-terminal ends of the construct crystallized are indicated by arrows. Secondary structure elements are shown and the disordered region connecting helices 4 and 5 is drawn as a dashed line. (B) Surface representation of the proposed membrane targeting surface of homodimeric CHMP proteins composed of helices 1 and 2 (the remaining C-terminal part of the structure including residues 101–183, which probably does not contribute to the membrane binding surface, was omitted for clarity). CHMP protein sequences were modeled onto the CHMP3 structure based on the sequence alignment shown in (A). Basic residues are shown in blue, acidic residues in red, polar residues in yellow, and hydrophobic residues in green. One basic residue exposed in the center of helix 1 and the strictly conserved basic residue corresponding to CHMP3 K49 are indicated for orientation.

indirectly supported by the high flexibility at the C-terminal region of CHMP3, including helix 5 (B factors > 90 A˚2) in all four monomers of the P1 cell. CHMP Protein Sequence Conservation and Potential Heterodimerization Although CHMP proteins from the six subgroups display only limited sequence conservation (Figure 4A), the corresponding orthologs from different species are highly conserved, including human CHMP3 and yeast Vps24 (Whitley et al., 2003). Biochemical studies have shown

that yeast Vps24p and Vps2p (CHMP3/CHMP2) form a subcomplex (Babst et al., 2002a), a result consistent with the observed interaction of human CHMP3 and CHMP2A/B (Martin-Serrano et al., 2003; von Schwedler et al., 2003). Sequence comparisons reveal CHMP2A as the closest homolog of CHMP3 (with 23.9% identity, compared to 11%–17% with all other CHMPs). Because most of the residues involved in both dimerization modes are conserved between CHMP3 and CHMP2A (Figure 4), heterodimer formation (Babst et al., 2002a; Martin-Serrano et al., 2003; von Schwedler et al., 2003)

Crystal Structure of CHMP3 825

Figure 5. Electrostatic Potential Maps of CHMP3(9–183) Reveal a Flat Membrane Targeting Surface (Electrostatic Potential <220 kBT in Red and >+20 kBT in Blue) (A) One surface of the flat CHMP3 dimer 1 carries an extensive basic surface charge which is involved in bilayer interaction via electrostatic forces. Basic residues conserved among CHMP3 homologs are indicated. The total surface exposed for membrane interaction is w2700 A˚2 per monomer. Note that mutagenesis of residues R24, K25, R28, R32, and R35 abolishes membrane targeting and thus inhibits the dominant-negative effect of CHMP3 on HIV-1 budding. (B) Dimer 1 viewed from the side showing helix 5 and the C-terminal end (arrow) pointing away from the surface opposite the proposed membrane targeting surface (same orientation as in Figure 2D). (C) View of the surface opposite the membrane targeting surface, which is less positively charged. The direction of the C-terminal ends (arrow) indicates their optimal position to interact with the AAA-type ATPase Vps4. Thus, the Vps4 binding site accessibility on CHMP3 within a potential CHMP protein lattice allows disassembly (Babst et al., 1998). Residues indicating the C-terminal end and some basic residues are labeled (same orientation as in Figure 2C). (D) The CHMP3(9–183) dimer forms a lattice in the crystal which exposes the proposed membrane interaction surface on one side. Six dimers of the lattice are shown, which form three types of dimer-dimer interactions, exposing a convex (1), a concave (2), and a flat (3; or dimer 2) basic surface (see arrow).

could involve similar interactions as observed for the CHMP3 homodimerization in vitro (Figures 3C and 3D). Electrostatic Potential Map of CHMP3 Reveals a Membrane Targeting Surface The electrostatic potential map of CHMP3(9–183) reveals that one side of the flat molecule, composed of the long helical hairpin (helices 1 and 2), displays an extensive basic surface charge covering a total surface of w5500 A˚2, which we suggest as constituting the membrane interaction surface (Figure 5A). Notably, most of the exposed basic residues are conserved between human and yeast CHMP3 (Whitley et al., 2003), but not among mammalian CHMP family members (Figure 4A). The diversity of the membrane targeting properties of mammalian CHMP family members becomes obvious when the residues are mapped onto models of CHMP proteins based on the dimeric CHMP3 structure (Figure 2C) and the sequence alignment (Figure 4A). The differences in surface charge of the proposed membrane targeting side of CHMP1A, 2A, 4A, 5, and 6 imply indirectly that CHMP protein membrane interactions may have slightly different requirements for bilayer lipid contents (Figure 4B). In contrast, the opposite site of the molecule from which the C termini emanate (Figure 5B) displays a smaller amount of basic charges, of which much fewer are conserved (Figures 5C). This arrangement places the acidic C terminus in an ideal position to recruit the ATPase Vps4 (Figures 2D and 5C), which has been proposed to disassemble CHMP protein lattices on membranes initiated by interactions of the Vps4 MIT domain with the C-terminal acidic tails of CHMP proteins (Fujita et al., 2004; Lin et al., 2005; Scott et al., 2005a, 2005b). The extensive basic region becomes even more evident upon calculation of the electrostatic potential of the CHMP3 lattice, suggesting how CHMP3 may form a membrane-associated network (Figure 5D).

Effects of Basic Residues and of the Dimerization Interfaces on HIV-1 Budding A number of studies have shown that the expression of CHMP proteins either as fusion proteins or as C-terminally truncated forms exerts a dominant-negative effect on virus budding in HIV-1-producing cells, or induce an aberrant cellular class E compartment similar to that induced by Vps4 active site mutants (Babst et al., 1998; Lin et al., 2005; Martin-Serrano et al., 2003; Strack et al., 2003; von Schwedler et al., 2003; Whitley et al., 2003). Here we show that expression of CHMP3 residues 1–150 fused to GFP-PK in 293T cells cotransfected with HIV-1 proviral DNA has a dominant-negative effect and completely blocks HIV-1 virus production, as measured by the release of the HIV-1 CA protein (Figure 6A, WT, lane 2). The substitution of three basic residues in GFP-PK-CHMP3(1–150) (M4, R24S, K25A, R28S; Figure 5A) profoundly diminishes the dominant-negative effect, allowing HIV-1 virion production (Figure 6A, M4, lane 3). A less dramatic effect was observed when only two basic residues on the proposed membrane targeting surface were mutated (M5, R32A, R35S; Figure 5A), which reduced the dominant-negative effect only slightly (Figure 6A, M5, lane 4). However, the combination of the M4 and M5 mutations completely reversed the dominant-negative effect observed for wild-type GFP-PK-CHMP3(1–150), allowing wild-type levels of virion production (Figure 6A, M4+5, lane 5). The CHMP3 fusion proteins had no or only a moderate effect on the overall levels of cell-associated HIV-1 Gag; however, wild-type CHMP3(1–150), in particular, altered the ratio of CA to CA-p2, as expected for a late assembly defect (Figure 6B, lanes 1–5). Of note, each mutant CHMP3 fusion protein was expressed at least as well as the parental molecule (Figure 6C, lanes 1–5). Cellular distribution of wild-type GFP-PK-CHMP3(1–150) indicates predominantly plasma membrane and vesicular localization, while the mutant forms of GFP-PK-CHMP3(1–150) M4

Developmental Cell 826

and M4+5 show a predominant cytoplasmic localization (Figure 6D). These results are thus consistent with a model that implicates an extensive basic surface cluster of CHMP3 that is conserved from yeast to mammals (residues R24, K25, R28, R32, R35; Figure 5A) in lipid bilayer interaction, which is required for CHMP3 protein function in HIV-1 budding. The same HIV-1 budding assay was used to evaluate the dimerization mutants M1, M2, and M3 in the context of GFP-PK-CHMP(1–150) expression. This showed that the M1 and M1.2 mutations prevented virus production, like wild-type GFP-PK-CHMP3(1–150) (Figure 6A, WT, M1, M1.2, lanes 7–9), while the M2 mutation slightly reduced the dominant-negative effect of wild-type GFPPK-CHMP3(1–150), allowing some virus production (Figure 6A, WT and M2, lanes 12 and 13). In contrast, the double mutant M3 completely reversed the dominant-negative effect of wild-type GFP-PK-CHMP3(1– 150), allowing normal HIV-1 virion production (Figure 6A, M3, lane 10). These results suggest that changes on the helical hairpin surface (M2; dimer 2) are slightly more sensitive to HIV-1 budding inhibition than those within the M1 (dimer 1) interface; however, our results suggest that both cooperate to achieve a dominant-negative effect on HIV-1 budding. The expression of the mutant GFP-PK-CHMP3(1–150) had no or only a moderate effect on the overall levels of cell-associated HIV-1 Gag (Figure 6B, lanes 6–13), and each mutant GFP-PKCHMP3(1–150) was expressed at similar levels as wildtype GFP-PK-CHMP3(1–150) (Figure 6C, lanes 6–13). Finally, mutations within both dimerization interfaces altered the cellular localization of GFP-PK-CHMP3(1– 150). All three dimerization mutants, M1, M2, and M3, showed predominantly cytoplasmic localization as compared to wild-type CHMP3(1–150) (Figure 6D), implying that the dimer interfaces are required for efficient plasma membrane localization. Discussion

Figure 6. Both Basic Residues and the CHMP3 Dimerization Interfaces Are Required for the Dominant-Negative Effect of a Truncated CHMP3 Fusion Protein on HIV-1 Virus Particle Production (A) HIV-1 particle release upon expression of GFP-PK (control; lanes 1, 6, and 11), wild-type GFP-PK-CHMP3(1–150; lanes 2, 7, and 12), and GFP-PK-CHMP3(1–150) mutants: M4 (R24S, K25A, R28S), lane 3; M5 (R32A, R35S), lane 4; M4+5, lane 5; M1 (Y78R, A79E), lane 8; M1.2 (Y78R, A79E, A82Y), lane 9; M2 (K54S, Q56N, V59E, V62D, L63D), lane 13; M3 (combination of M1 and M2), lane 10. HIV-1 Gag proteins were detected by Western blot analysis with anti-CA serum. (B) Cellular levels of the HIV Gag proteins Pr55, p41, CA, and CA-P2 upon expression of GFP-PK (control) and GFP-PK-CHMP3(1–150) mutants as indicated in (A). (C) Expression levels of GFP-PK (control), wild-type GFP-PKCHMP3(1–150), and GFP-PK-CHMP3(1–150) mutants as indicated in (A) were detected by Western blot using an anti-GFP antibody.

CHMP proteins form ESCRT-III complexes and together with ESCRT-I and -II belong to the class E Vps protein machinery that has been implicated in MVB biogenesis and enveloped virus budding (Katzmann et al., 2002; Morita and Sundquist, 2004). They are targeted to cellular membranes, where they have been suggested to form a protein lattice which is required for late steps in budding (Babst et al., 2002a; Lin et al., 2005; Martin-Serrano et al., 2003; Strack et al., 2003; von Schwedler et al., 2003). Here we show that wild-type CHMP3 folds into a monomer and that removal of most of the C-terminal acidic domain is sufficient to induce dimerization in vitro, indicating that CHMP3 monomers are in a metastable conformation. Although we crystallized a truncated form of CHMP3 that is monomeric in solution, the crystal structure of CHMP3(9–183) reveals two potential dimerization interfaces from the crystal lattice packing. The dimer 1 interface is extensive and typical for a protein

(D) Confocal microscopy images showing plasma membrane association for expression of wild-type GFP-PK-CHMP3(1–150) and a predominantly cytoplasmic localization for expression of mutants M1, M2, M3, M4, and M4+5.

Crystal Structure of CHMP3 827

surface which can undergo multiple transient interactions (Nooren and Thornton, 2003). This may include a potential interaction with the C-terminal acidic region in the monomeric conformation, supporting the hypothesis of CHMP protein activation via the displacement of the C-terminal acidic tail from the N-terminal core structure (Lin et al., 2005). The second dimerization interface present in the crystal lattice (dimer 2) is less extensive. Our mutagenesis studies show that changes within both interfaces affect CHMP3 dimerization in vitro. Because we have no direct evidence for a role of CHMP3 dimers in vivo, we suggest that the interfaces identified as mediating CHMP3 dimerization in vitro may represent interaction surfaces for other CHMP proteins in vivo promoting CHMP protein lattice formation on membranes. Therefore, both interfaces may be used to form either CHMP3-CHMP2A or even CHMP4/6 and CHMP1/5 subcomplexes (Babst et al., 2002a; Martin-Serrano et al., 2003; von Schwedler et al., 2003) that exist already in solution and/or in a CHMP protein lattice on membranes. The structure suggests that CHMP3 interacts with membranes via its basic charged surface, which was confirmed by mutagenesis studies. Such protein-membrane interactions via electrostatic forces are favored by the relative high content of monovalent acidic phospholipids such as phosphatidylserine (w15%–30%) present in the inner leaflet of the plasma membrane (McLaughlin, 1989; Murray et al., 1999). In addition, CHMP3 K49, implicated in binding to PtdIns(3,5)P2 (Whitley et al., 2003), is exposed on this surface, which is further consistent with this side constituting the membrane interaction surface. The C-terminal ends, which are the target for the ATPase Vps4 (Fujita et al., 2004; Lin et al., 2005), emanate from the opposite side of the flat molecule allowing Vps4 access when bound to the membranes, as required for ESCRT complex disassembly (Babst et al., 1998, 2002a; Fujita et al., 2003; Lin et al., 2005; Martin-Serrano et al., 2003; Strack et al., 2003; von Schwedler et al., 2003). CHMP fusion proteins have been shown to exert a dominant-negative effect on retrovirus budding (Martin-Serrano et al., 2003; Strack et al., 2003; von Schwedler et al., 2003). Here we show that a C-terminal truncation of CHMP3 containing the core four-helical bundle fused to GFP-PK exerts a strong dominant-negative effect on HIV-1 budding, a process which requires CHMP3 membrane targeting, as shown by our mutagenesis results. Similarly, a CHMP3 fusion protein form containing only the long helical hairpin induces the typical aberrant class E phenotype and leads to the accumulation of truncated CHMP3 on endosomal membranes (Whitley et al., 2003; Lin et al., 2005). This implies that the helical hairpin, which contains the membrane targeting surface according to our mutagenesis analyses, is sufficient to participate in CHMP lattice formation. Because the target site for Vps4 is either deleted or not accessible in the CHMP fusion proteins (Scott et al., 2005a, 2005b), the CHMP proteins are not released from cellular membranes and a dominant-negative effect is detected. A similar phenotype on HIV-1 budding is observed upon expression of an inactive Vps4 ATPase (Garrus et al., 2001; Martin-Serrano et al., 2003; von Schwedler et al., 2003), which implies that Vps4 is either actively required for the budding process or the observed phenotype is a result of a block in CHMP protein recycling.

The HIV-1 budding assays presented here revealed further that both dimer interfaces are important for budding and cooperate in the inhibitory effect of GFP-PKCHMP3(1–150). The relocalization of the dimerization mutants from the plasma membrane to a predominant cytoplasmic localization also implies that the interfaces are required for efficient CHMP3 membrane targeting. However, because the single mutants (M1 and M2) are still dominant negative, the amount of M1 and M2 targeted to the membrane may still be sufficient for its function. Therefore, one possibility is that M1 and M2 can enter a CHMP protein network which cannot be extended due to the mutations and thus exert a dominantnegative effect. This last possibility would thus suggest that the double mutant M3 cannot enter the network and thus has no effect. Alternatively, the dominant-negative effect of M1 and M2 may arise from CHMP3 interaction with soluble subcomplexes. Both enveloped virus budding and MVB biogenesis involve budding away from the cytosol, which supplies all factors required for the process. This poses the question as to how the evolving lipid neck structure of a bud brings the membranes into close enough contact to overcome the energetic barrier for controlled membrane fission. The dominant-negative CHMP proteins and inactive Vps4 both block a late step in virus budding which is morphologically characterized by fully assembled viral particles that are connected to the plasma membrane via a w100–200 A˚ (in diameter) large membrane neck structure. This suggests that CHMP proteins either alone or together with Vps4 are essential components of the membrane fission machinery (Martin-Serrano et al., 2003; Strack et al., 2003; von Schwedler et al., 2003). Our structural and functional analyses of CHMP3 together with previous studies suggest the following possible scenarios. First, CHMP monomers become activated and polymerize potentially via the two dimerization interfaces over the neck opening in such a way that the protein lattice connects all membrane ends. Stepwise disassembly or removal of CHMP proteins from the lattice may then lead to the narrowing of the neck opening. Although the role of CHMP family members in virus budding seems to be somewhat redundant (Ward et al., 2005), the overall composition of a CHMP protein lattice may have a specific effect on the membrane structure. Second, CHMP protein function may be coupled to the lipid bilayer content (Matsuo et al., 2004), which is indirectly supported by the extensive lipid bilayer interaction surface of CHMP3 (w2250 A˚2 per monomer). A number of studies have shown that lipid bilayers can separate into different lipid phases in vitro, which in turn can induce spontaneous vesicle budding (Bacia et al., 2005; Baumgart et al., 2003; Julicher and Lipowsky, 1993; Roux et al., 2005). Therefore, CHMP-membrane interactions may influence lipid phase separation processes in vivo. Such a function may explain the necessity of six different CHMP family members in yeast or ten in mammals, as each has a different membrane interaction surface. In conclusion, our structural and functional data suggest that CHMP3 may be part of the membrane fission machinery that requires ESCRT-III lattice formation on cellular membranes. The structure will thus serve as a platform to better

Developmental Cell 828

understand the process of budding during MVB biogenesis and enveloped virus egress from the cell. Experimental Procedures Cloning, Protein Expression, and Purification cDNA fragments derived from human CHMP3 encoding residues 1– 222 (wild-type), 9–183 (CHMP3[9–183]), and 9–164 (CHMP3[9–164]) were subcloned into the expression vector pProExHTb (Invitrogen). Mutagenesis was performed using standard PCR procedures. The mutants have the following residues changed: M1, Y78 to R, A79 to E; M1.2, Y78 to R, A79 to E, A82 to Y; M2, K54 to S, Q56 to N, V59 to E, V62 to D, L63 to D; M3 is a combination of M1 and M2; M4, R24 to S, K25 to A, R28 to S; M5, R32 to A, R35 to S. Mutations were introduced into CHMP3(9–164) for bacterial expression and GFP-PK-CHMP3(1–150) for mammalian expression. All sequences were confirmed by DNA sequencing. All wild-type and mutant CHMP proteins were expressed and purified in the same way. Briefly, protein expression was performed in Escherichia coli strain BL21 codon plus (Invitrogen) and the incorporation of selenomethionine in the case of CHMP3(9–183) was performed using standard procedures. The bacterial cell lysate was applied onto an Ni2+ Sepharose (Amersham Pharmacia Biotech) column in buffer A (50 mM Tris [pH 8.8], 100 mM NaCl). The column was extensively washed with buffer A containing 1 M NaCl and 1 M KCl, followed by a wash step with buffer A containing 1% CHAPS. The column was subsequently equilibrated with buffer A and CHMP proteins were eluted by applying an imidazole gradient in buffer A. The His tag was removed from wild-type and all mutant CHMP3 proteins by TEV protease cleavage (1:200; w/w, 4ºC, overnight) and the tag and the TEV protease were removed on an Ni2+ Sepharose column. Wild-type and mutant CHMP3 proteins were further purified on a Superdex 75 column in either buffer B (20 mM bicine [pH 9.0], 100 mM NaCl), buffer C (20 mM HEPES [pH 8.5], 100 mM NaCl), or in the case of selenomethionine-substituted CHMP3(9–183), in buffer D (20 mM sodium cacodylate [pH 6.5], 100 mM NaCl, 2 mM b-mercaptoethanol). Chemical crosslinking of wild-type and mutant CHMP3 was performed at a protein concentration of 0.5 mg/ml in buffer C using ethylene glycol bis-succinimide at the concentrations indicated. The reaction was quenched with 100 mM Tris (pH 8.0). Crystallization, Diffraction Data Collection, and Structure Solution CHMP3(9–183) crystals (usually thin plates) were grown at 20ºC by vapor diffusion by mixing 1 ml of reservoir solution (8%–12% PEG 3350, 200 mM ammonium acetate) and 1 ml CHMP3(9–183) (w10 mg/ml in buffer D). A selenomethionine-substituted crystal was cryoprotected in the reservoir buffer supplemented with 25% glycerol, and then flash cooled in liquid nitrogen. A single wavelength anomalous dispersion (SAD) data set was collected at 100 K at beamline ID14-EH4 (European Synchrotron Radiation Facility, Grenoble) using an inverse beam strategy for collection of 360º at the peak wavelength of the selenium K edge (wavelength 0.97927 A˚). The data were processed using the program XDS (CCP4, 1994; Kabsch, 1988). The crystals belong to space group P1 with unit cell dimensions of a = 33.56 A˚, b = 72.53 A˚, c = 89.16 A˚, and a = 68.66º, b = 83.59º, g = 76.69º and diffract to a resolution of 2.8 A˚ (SAD data statistics: 94.7% completeness; total number of reflections 76,650; unique number of reflections 35,544; Rmerge = 0.096; Friedel pairs unmerged). Analysis of Matthews coefficients revealed that there were two to four possible molecules in the P1 cell. The positions of 22 selenium sites (15 expected per monomer) were found using the program SHELXD (Schneider and Sheldrick, 2002). Initial phases were calculated using SHELXE (Sheldrick, 2002) and eight additional sites were found by further refinement of the sites with the program SHARP, yielding good phasing statistics (FOMacen = 0.40; last shell 2.92–2.80 A˚ = 0.25; PPacen = 1.61; last shell 2.92–2.80 A˚ = 0.60) (de La Fortelle and Bricogne, 1997). The total of 30 sites was validated with SOLVE (Terwilliger and Berendzen, 1999) and density modification was performed with the program RESOLVE (Terwilliger, 2000), which revealed three noncrystallographic symmetry (NCS) operators. However, inspection of the solvent boundaries of the electron density map as well as the distribution of the anomalous peaks suggested the presence of four molecules in the P1 cell. The additional

NCS operator was found using the program PROFESSS (CCP4, 1994) and 4-fold averaging was performed using the program DM (CCP4, 1994). The resulting electron density map revealed the presence of helices and the program ARP/wARP (CCP4, 1994) was used to build 17 short helical fragments as a polyalanine model. Iterative model building and refinement was done using the programs COOT (Emsley and Cowtan, 2004), O (Jones et al., 1991), CNS (Brunger et al., 1998), and REFMAC (CCP4, 1994). Poorly defined regions such as loop connections and the position of helix 5 were confirmed by calculating simulated annealing omit maps. The final model contains four monomers (602 residues out of 716): chain A residues 12– 141 and 161–172, chain B residues 9–140 and 161–180, chain C residues 12–98, 102–140, and 161–176, and chain D residues 9–141 and 156–181. One chain per monomer (chains B and D) contains four extra residues at the N terminus derived from the vector, which make important crystal contacts between CHMP protein layers. The structure was refined to an Rfactor of 0.259 (total reflections, 16,944) and Rfree of 0.301 (5% test set reflections, 903) with an overall average B factor of 59.7 A˚2 and good stereochemistry (rmsd bond length = 0.017 A˚; rmsd bond angles = 1.51º); 85.6% of the residues are in most favored and 13.5% are in additionally allowed regions of a Ramachandran plot according to analysis with the program PROCHECK (CCP4, 1994). The coordinates have been deposited in the Protein Data Bank (PDB code: 2GD5). Structure Analysis Figures were generated with the programs ESPript (Gouet et al., 1999), PyMOL (http://www.pymol.org), and GRASP (Nicholls et al., 1991). Sequences were aligned using Clustalx (Thompson et al., 1997). Secondary structure elements were assigned using the program DSSP (Kabsch and Sander, 1983). Analysis of HIV-1 Virus Production A fragment encoding CHMP3 residues 1–150 was amplified by PCR and fused to a GFP-pyruvate kinase chimera (GFP-PK) provided by Warner Greene (University of California, San Francisco). To determine the effect of GFP-PK-CHMP3(1–150) on virus production, 293T cells (1.5 3 106) were seeded into T25 flasks and transfected 24 hr later using a calcium phosphate precipitation technique. The cultures were cotransfected with 1.5 mg of the full-length HIV-1 proviral construct HXBH10 and with 0.5 mg of either GFP-PK or wildtype and mutant versions of GFP-PK-CHMP3(1–150). Twenty-four hours posttransfection, virions released into the culture medium were pelleted through sucrose. HIV-1 Gag proteins in virion and cell lysates were detected by Western blotting with a rabbit antiHIV CA antiserum (Advanced Biotechnologies), and CHMP3 fusion proteins were detected with an anti-GFP monoclonal antibody (Molecular Probes). Expression of wild-type GFP-PK-CHMP3(1–150), and that of all mutants, was analyzed by confocal microscopy 24 hr posttransfection in 293T cells. Acknowledgments We thank all members of the EMBL/ESRF JSBG for access to the ESRF beamlines, Dr. E. Latz for help with confocal microscopy, Dr. W. Greene for providing the plasmid pGFP-PK, Drs. M. Spano and J. Marquez for crystallization screening, and J. McCarthy for excellent technical assistance. This work was supported by the EMBL (W.W.), Deutsche Forschungsgemeinschaft (SFB 593, Marburg) (W.W.), and NIH grant AI29873 (H.G.). E.P.-M. was supported by a long-term EMBO fellowship and a Marie Curie fellowship from the European Union. Received: September 1, 2005 Revised: February 20, 2006 Accepted: March 20, 2006 Published: June 5, 2006 References Babst, M., Wendland, B., Estepa, E.J., and Emr, S.D. (1998). The Vps4p AAA ATPase regulates membrane association of a Vps protein complex required for normal endosome function. EMBO J. 17, 2982–2993.

Crystal Structure of CHMP3 829

Babst, M., Katzmann, D.J., Estepa-Sabal, E.J., Meerloo, T., and Emr, S.D. (2002a). ESCRT-III: an endosome-associated heterooligomeric protein complex required for MVB sorting. Dev. Cell 3, 271–282.

Kabsch, W., and Sander, C. (1983). Dictionary of protein secondary structure: pattern recognition of hydrogen-bonded and geometrical features. Biopolymers 22, 2577–2637.

Babst, M., Katzmann, D.J., Snyder, W.B., Wendland, B., and Emr, S.D. (2002b). Endosome-associated complex, ESCRT-II, recruits transport machinery for protein sorting at the multivesicular body. Dev. Cell 3, 283–289.

Katzmann, D.J., Babst, M., and Emr, S.D. (2001). Ubiquitin-dependent sorting into the multivesicular body pathway requires the function of a conserved endosomal protein sorting complex, ESCRT-I. Cell 106, 145–155.

Bache, K.G., Brech, A., Mehlum, A., and Stenmark, H. (2003). Hrs regulates multivesicular body formation via ESCRT recruitment to endosomes. J. Cell Biol. 162, 435–442.

Katzmann, D.J., Odorizzi, G., and Emr, S.D. (2002). Receptor downregulation and multivesicular-body sorting. Nat. Rev. Mol. Cell Biol. 3, 893–905.

Bacia, K., Schwille, P., and Kurzchalia, T. (2005). Sterol structure determines the separation of phases and the curvature of the liquid-ordered phase in model membranes. Proc. Natl. Acad. Sci. USA 102, 3272–3277.

Katzmann, D.J., Stefan, C.J., Babst, M., and Emr, S.D. (2003). Vps27 recruits ESCRT machinery to endosomes during MVB sorting. J. Cell Biol. 162, 413–423.

Baumgart, T., Hess, S.T., and Webb, W.W. (2003). Imaging coexisting fluid domains in biomembrane models coupling curvature and line tension. Nature 425, 821–824. Brunger, A.T., Adams, P.D., Clore, G.M., DeLano, W.L., Gros, P., Grosse-Kunstleve, R.W., Jiang, J.S., Kuszewski, J., Nilges, M., Pannu, N.S., et al. (1998). Crystallography & NMR system: a new software suite for macromolecular structure determination. Acta Crystallogr. D Biol. Crystallogr. 54, 905–921. CCP4 (Collaborative Computational Project, Number 4) (1994). The CCP4 suite: programs for protein crystallography. Acta Crystallogr. D Biol. Crystallogr. 50, 157–163. de La Fortelle, E., and Bricogne, G. (1997). SHARP: A MaximumLikelihood Heavy-Atom Parameter Refinement Program for the MIR and MAD Methods, Volume 276 (Orlando, FL: Academic Press). Emsley, P., and Cowtan, K. (2004). Coot: model-building tools for molecular graphics. Acta Crystallogr. D Biol. Crystallogr. 60, 2126– 2132. Fujita, H., Yamanaka, M., Imamura, K., Tanaka, Y., Nara, A., Yoshimori, T., Yokota, S., and Himeno, M. (2003). A dominant negative form of the AAA ATPase SKD1/VPS4 impairs membrane trafficking out of endosomal/lysosomal compartments: class E vps phenotype in mammalian cells. J. Cell Sci. 116, 401–414. Fujita, H., Umezuki, Y., Imamura, K., Ishikawa, D., Uchimura, S., Nara, A., Yoshimori, T., Hayashizaki, Y., Kawai, J., Ishidoh, K., et al. (2004). Mammalian class E Vps proteins, SBP1 and mVps2/ CHMP2A, interact with and regulate the function of an AAA-ATPase SKD1/Vps4B. J. Cell Sci. 117, 2997–3009. Futter, C., Pearse, A., Hewlett, L., and Hopkins, C. (1996). Multivesicular endosomes containing internalized EGF-EGF receptor complexes mature and then fuse directly with lysosomes. J. Cell Biol. 132, 1011–1023. Garrus, J., von Schwedler, U., Pornillos, O., Morham, S., Zavitz, K., Wang, H., Wettstein, D., Stray, K., Cote, M., and Rich, R. (2001). Tsg101 and the vacuolar protein sorting pathway are essential for HIV-1 budding. Cell 107, 55–65. Gordon, P., Carpenter, J.L., Cohen, S., and Orci, L. (1978). Epidermal growth factor: morphological demonstration of binding, internalization, and lysosomal association in human fibroblasts. Proc. Natl. Acad. Sci. USA 75, 5025–5029. Gouet, P., Courcelle, E., Stuart, D., and Metoz, F. (1999). ESPript: analysis of multiple sequence alignments in PostScript. Bioinformatics 15, 305–308. Gruenberg, J., and Stenmark, H. (2004). The biogenesis of multivesicular endosomes. Nat. Rev. Mol. Cell Biol. 5, 317–323. Howard, T.L., Stauffer, D.R., Degnin, C.R., and Hollenberg, S.M. (2001). CHMP1 functions as a member of a newly defined family of vesicle trafficking proteins. J. Cell Sci. 114, 2395–2404. Jones, T., Zou, J., Cowan, S., and Kjeldgaard, M. (1991). Improved methods for building protein models in electron density maps and the location of errors in these models. Acta Crystallogr. A 47, 110– 119.

Lin, Y., Kimpler, L.A., Naismith, T.V., Lauer, J.M., and Hanson, P.I. (2005). Interaction of the mammalian endosomal sorting complex required for transport (ESCRT) III protein hSnf7-1 with itself, membranes, and the AAA+ ATPase SKD1. J. Biol. Chem. 280, 12799– 12809. Martin-Serrano, J., Yarovoy, A., Perez-Caballero, D., and Bieniasz, P.D. (2003). Divergent retroviral late-budding domains recruit vacuolar protein sorting factors by using alternative adaptor proteins. Proc. Natl. Acad. Sci. USA 100, 12414–12419. Matsuo, H., Chevallier, J., Mayran, N., Le Blanc, I., Ferguson, C., Faure, J., Blanc, N.S., Matile, S., Dubochet, J., Sadoul, R., et al. (2004). Role of LBPA and Alix in multivesicular liposome formation and endosome organization. Science 303, 531–534. McCullough, J., Row, P.E., Lorenzo, O., Doherty, M., Beynon, R., Clague, M.J., and Urbe, S. (2006). Activation of the endosome-associated ubiquitin isopeptidase AMSH by STAM, a component of the multivesicular body-sorting machinery. Curr. Biol. 16, 160–165. McLaughlin, S. (1989). The electrostatic properties of membranes. Annu. Rev. Biophys. Biophys. Chem. 18, 113–136. Morita, E., and Sundquist, W.I. (2004). Retrovirus budding. Annu. Rev. Cell Dev. Biol. 20, 395–425. Murray, D., Arbuzova, A., Hangyas-Mihalyne, G., Gambhir, A., BenTal, N., Honig, B., and McLaughlin, S. (1999). Electrostatic properties of membranes containing acidic lipids and adsorbed basic peptides: theory and experiment. Biophys. J. 77, 3176–3188. Nicholls, A., Sharp, K.A., and Honig, B. (1991). Protein folding and association: insights from the interfacial and thermodynamic properties of hydrocarbons. Proteins 11, 281–296. Nooren, I.M., and Thornton, J.M. (2003). Structural characterisation and functional significance of transient protein-protein interactions. J. Mol. Biol. 325, 991–1018. Roux, A., Cuvelier, D., Nassoy, P., Prost, J., Bassereau, P., and Goud, B. (2005). Role of curvature and phase transition in lipid sorting and fission of membrane tubules. EMBO J. 24, 1537–1545. Rudge, S.A., Anderson, D.M., and Emr, S.D. (2004). Vacuole size control: regulation of PtdIns(3,5)P-2 levels by the vacuole-associated Vac14-Fig4 complex, a PtdIns(3.5)P-2-specific phosphatase. Mol. Biol. Cell 15, 24–36. Schneider, T.R., and Sheldrick, G.M. (2002). Substructure solution with SHELXD. Acta Crystallogr. D Biol. Crystallogr. 58, 1772–1779. Scott, A., Chung, H.Y., Gonciarz-Swiatek, M., Hill, G.C., Whitby, F.G., Gaspar, J., Holton, J.M., Viswanathan, R., Ghaffarian, S., Hill, C.P., and Sundquist, W.I. (2005a). Structural and mechanistic studies of VPS4 proteins. EMBO J. 24, 3658–3669. Scott, A., Gaspar, J., Stuchell-Brereton, M.D., Alam, S.L., Skalicky, J.J., and Sundquist, W.I. (2005b). Structure and ESCRT-III protein interactions of the MIT domain of human VPS4A. Proc. Natl. Acad. Sci. USA 102, 13813–13818. Sheldrick, G.M. (2002). Macromolecular phasing with SHELXE. Z Kristallograph 217, 644–650.

Julicher, F., and Lipowsky, R. (1993). Domain-induced budding of vesicles. Phys. Rev. Lett. 70, 2964–2967.

Strack, B., Calistri, A., Popova, E., and Gottlinger, H. (2003). AIP1/ ALIX is a binding partner for HIV-1 p6 and EIAV p9 functioning in virus budding. Cell 114, 689–699.

Kabsch, W. (1988). Evaluation of single-crystal X-ray diffraction data from a position-sensitive detector. J. Appl. Crystallogr. 21, 916–924.

Terwilliger, T. (2000). Maximum-likelihood density modification. Acta Crystallogr. D Biol. Crystallogr. 56, 965–972.

Developmental Cell 830

Terwilliger, T.C., and Berendzen, J. (1999). Automated MAD and MIR structure solution. Acta Crystallogr. D Biol. Crystallogr. 55, 849–861. Thompson, J., Gibson, T., Plewniak, F., Jeanmougin, F., and Higgins, D. (1997). The CLUSTAL_X windows interface: flexible strategies for multiple sequence alignment aided by quality analysis tools. Nucleic Acids Res. 25, 4876–4882. von Schwedler, U., Stuchell, M., Muller, B., Ward, D., Chung, H., Morita, E., Wang, H., Davis, T., He, G., and Cimbora, D. (2003). The protein network of HIV budding. Cell 114, 701–713. Ward, D.M., Vaughn, M.B., Shiflett, S.L., White, P.L., Pollock, A.L., Hill, J., Schnegelberger, R., Sundquist, W.I., and Kaplan, J. (2005). The role of LIP5 and CHMP5 in multivesicular body formation and HIV-1 budding in mammalian cells. J. Biol. Chem. 280, 10548–10555. Whitley, P., Reaves, B.J., Hashimoto, M., Riley, A.M., Potter, B.V.L., and Holman, G.D. (2003). Identification of mammalian Vps24p as an effector of phosphatidylinositol 3,5-bisphosphate-dependent endosome compartmentalization. J. Biol. Chem. 278, 38786–38795. Yoshimori, T., Yamagata, F., Yamamoto, A., Mizushima, N., Kabeya, Y., Nara, A., Miwako, I., Ohashi, M., Ohsumi, M., and Ohsumi, Y. (2000). The mouse SKD1, a homologue of yeast Vps4p, is required for normal endosomal trafficking and morphology in mammalian cells. Mol. Biol. Cell 11, 747–763.

Accession Number The crystal structure of the human ESCRT-III protein CHMP3 has been deposited in the Protein Data Bank under ID code 2GD5.