The ESCRT pathway is best known for facilitating membrane remodeling and fission for processes such as the endosomal intralumenal vesicle (ILV) formation, enveloped virus budding, nuclear envelope closure, and cytokinetic abscission (1–3). In these reactions, the ESCRT machinery assembles on the interior of a negatively-curved, cytoplasm-filled membrane neck, and pulls the membrane toward itself to the fission point. These fission reactions are topologically distinct from reactions in which cytoplasmic BAR domain-containing proteins and dynamin-family GTPases assemble around and constrict positively-curved membrane tubules.

ESCRT components are recruited to different membranes by site-specific adaptors that ultimately recruit ESCRT-III subunits and their binding partners, including VPS4-family ATPases. ESCRT-III assemblies promote membrane constriction and fission, possibly in concert with VPS4. Humans express 12 related ESCRT-III proteins, called CHarged Multivesicular body Proteins (CHMPs) 1A-7 and Increased Sodium Tolerance 1 (IST1) (1–3). Crystal structures of CHMP3 and IST1 show a common structure in which the first two helices form a long hairpin, the shorter helices 3 and 4 pack against the open end of the hairpin, and helix 5 folds back and packs against the closed end of the helical hairpin (4–6). This “closed” conformation appears to auto-inhibit ESCRT-III membrane binding and oligomerization (4, 7, 8). ESCRT-III subunits can also adopt a second, more extended “open” conformation that has been characterized biochemically, but not visualized in molecular detail. The open conformation appears to be the active, polymerization-competent state because mutations or solution conditions that favor this conformation typically promote polymerization (1–3). Many ESCRT-III subunits form spiraling homo- and heteromeric filaments, both in vitro (4, 9–15) and in cells (10, 16–19), but the structural basis for filament assembly is unclear.

We used cryo-EM to determine the molecular structure of a helical copolymer comprising human IST1 and CHMP1B. Full-length IST1 and CHMP1B spontaneously co-assembled under low ionic strength conditions into well-ordered helical tubes. Helical order was further enhanced by using a truncated IST1 construct that spanned residues 1–189, hereafter termed IST1_NTD, and by including small, acidic unilamellar vesicles (SUVs) to nucleate polymer formation. The resulting IST1_NTD-CHMP1B tubes were long, straight, and 24 nm in diameter (Fig. 1A). The 3D structure of IST1_NTD-CHMP1B assemblies was determined to a resolution of ~4 Å by real space helical reconstruction (Methods and figs. S1–4). Each tube comprised a right-handed one-start helical filament that packed with an inter-filament spacing of 5.1 nm/turn (Fig. 1B–D, Movie S1). Each filament was double-stranded, with distinct inner and outer strands (at 7.7 and 10.2 nm radii, respectively). Segmented densities from subunits in the outer strand corresponded well to the crystal structure of IST1_NTD in its closed conformation (PDB: 3FRR) (5), with only minor refinements required to optimize the position of helix A (Fig. 1E, G, and Movie S2). By contrast, the CHMP1B subunits of the inner strand adopted a very different, open conformation. These subunits were almost entirely α-helical and side chain densities were clearly evident in the EM density (Fig. 1B, D, F, Movie S3). The open CHMP1B conformation resembled an arm, with helices 1–3 forming the upper arm and biceps, helix 4 and helix A forming the forearm, and helix 5 forming the hand. Joints between helices 3 and 4 and between helices A and 5 correspond to the elbow and wrist, respectively (Fig. 1F).

High ionic strength conditions (4, 8) typically favor the monomeric, closed ESCRT-III subunit conformation (4, 8, 20, 21), and CHMP1B also remained monomeric under high ionic strength conditions (fig. S5). Lowering the ionic strength triggered coassembly of IST1 and CHMP1B, implying that CHMP1B subunits are captured as they open. To visualize this conformational change, we generated a structure-based homology model for the CHMP1B closed state (see Methods). In comparison to the modeled closed state, the helix 5 hand is displaced by ~100 Å when CHMP1B opens. This global reorganization requires only three local rearrangements: the elbow angle between helices 3 and 4 must change, and the loops that connect helix 2/3 and helix 4/A must become helical to create the longer, continuous helices that extend the upper arm to the elbow and create the forearm in the open state (Figs. 1F–G, 2A and Movie S4).

In the filament, the open CHMP1B conformation is stabilized by extensive intersubunit interactions along the inner strand (Fig. 2B). Each CHMP1B molecule interacts with four other CHMP1B subunits that pack together and cross the forearm of the original subunit. In addition, the helix 5 hand grasps the shoulder of the hairpin four subunits away, making a domain-swapped contact that is analogous to the intrasubunit interaction between the hairpin and helix 5 in the closed ESCRT-III conformations (Fig. 2A–B). Opening and assembly reduces the total solvent accessible surface area of CHMP1B from ~10,720Ų to ~6350Ų. The IST1_NTD-CHMP1B assembly is further stabilized by three additional types of interactions, which differ completely from crystallized contacts for soluble IST1-CHMP1B heterodimers (Figs. 2C–F, figs. S6–7 (6)).

A final notable feature of the IST1-CHMP1B tube is the remarkably cationic interior. This surface is created by a series of conserved basic residues from helix 1 of CHMP1B (Fig. 2D, F), which forms the principal membrane-binding site in other ESCRT-III proteins (5, 22). Its position inside the IST1-CHMP1B copolymer was unexpected because well characterized ESCRT-mediated membrane remodeling events require that membranes interact with the exterior surface of coiled ESCRT-III filaments, e.g., as in the neck of a nascent ILV or viruses. The functional roles of IST1 and CHMP1B in such canonical ESCRT activities have been enigmatic, however, because neither protein is required for ILV biogenesis (23–26) or virus budding (27). Moreover, IST1 functions in the resolution of endosomal tubules that project into the cytoplasm and recycle cargoes back to the plasma membrane (28). Looking in cells, we found that moderately overexpressed CHMP1B polymerized into elongated cytoplasmic structures (fig. S8). Deep-etch EM of the plasma membranes revealed filaments that were similar to those of CHMP4A and other well studied ESCRT-III proteins (17), except that CHMP1B filaments coated tubules ~35–60 nm in diameter that extended into rather than away from the cytoplasm (Fig. 3A–D) (17, 18). Replicas of cells transfected and immunolabeled for CHMP1B alone, or with IST1, also revealed immunodecorated organelles and tubules that were not attached to the plasma membrane but again had recognizable striations (Fig. 3E–K). The resemblance of these organelles to early endosomes (29) including the occasional presence of clathrin-coated buds (Fig. 3J–K) indicates that CHMP1B and IST1 can coat and potentially remodel endosomal tubules.

To determine whether distinct membrane remodeling topologies are intrinsic properties of different ESCRT-III filaments, we compared the structures induced by spirals of CHMP1B to those induced by the prototypical ESCRT-III protein CHMP4A. In earlier studies, CHMP4 proteins only deformed the membrane when bound to ATPase-deficient VPS4B (17) or an activated CHMP2A mutant (18). Here we found that deleting C-terminal sequences yielded a mutant CHMP4A that formed tight, membrane-deforming spirals (Fig. 4A). The deformations induced by CHMP4A spirals were directed away from the cytoplasm, as expected (17, 18). Comparable views of cells transfected with full-length and C-terminal truncations of CHMP1B confirmed that CHMP4A and CHMP1B induced cellular membranes to tubulate in opposite directions (Fig. 4B, fig. S9).

We also tested how CHMP1B polymers bind and remodel liposomal membranes in vitro. Under physiological solution conditions, CHMP1B formed single and double-stranded one-start helices and spirals around membrane tubules, with interstrand spacing of 4.7 ± 0.1 nm (Fig. 4C, figs. S10–S11). Adding IST1 to CHMP1B-induced membrane tubules generated copolymeric helices that were structural analogs of the membrane-free IST1-CHMP1B assemblies described above, as judged by their interstrand spacing (5.1 ± 0.1 nm vs. 5.2 ± 0.1 nm) and by the similarities of 2D class averages of the two assemblies (Fig. 4D, figs. S10–S11). The positively-curved membrane tubules within the protein coats could be visualized with negative staining of the CHMP1B and IST1-CHMP1B assemblies (Fig. 4C–D, fig. S10), in cryo-EM images, and in 2D class averages of the molecules along the tangential surface of the bilayer (fig. S11). Thus, CHMP1B and IST1 form external coats on positively-curved membranes in vitro and in cells.

Our analyses provide insight into how ESCRT-III helices can assemble structures of decreasing diameter, as might be required to draw membranes together to the fission point. In addition to uniform 24 nm helices, we frequently observed conical helical assemblies of membrane-associated CHMP1B and IST1_NTD-CHMP1B (Fig. 4C–D, figs. S10–11), as well as conical IST1-CHMP1B filaments assembled in the absence of nucleating vesicles (fig. S12). One class of membrane-free IST1_NTD-CHMP1B cones was sufficiently common to be reconstructed at low resolution, which confirmed that the cones are composed of the same double-stranded filaments as those seen in the IST1_NTD-CHMP1B helices (Fig. 4D, figs. S1, S11–12, Movie S5–6). Within the conical spiral, both the degree of filament curvature and the lateral interactions between adjacent filaments varied continuously. Small changes in filament curvature are likely accommodated by altering the angles of the “elbow” and “wrist” joints. Larger changes could, in principle, be accommodated (or even driven) by ratcheting the buttressing intersubunit interactions made by the “forearm” and “hand”. For example, changing the connectivity from i+4 to i+5 would tend to straighten the CHMP1B filament. The required continuum of differing lateral interactions in the IST1_NTD-CHMP1B spirals is apparently accommodated by small shifts in ionic inter-strand interactions, which is plausible because the basic charges are distributed almost uniformly along one edge of the double strand (Fig. 2C–D). Wider IST1-CHMP1B spirals will tend to propagate toward their preferred 24 nm diameter. At this diameter, the narrow lumen (~10 nm) would force internal opposing lipid bilayers (each ~4.7 nm wide) toward hemi-fission (30, 31), potentially providing a driving force for membrane constriction and fission. Finally, we note that other ESCRT-III subunits, such as CHMP4A, tubulate membranes in the opposite direction to that seen with CHMP1B. Such stabilization of negative, rather than positive, membrane curvature could be achieved simply by altering the intrinsic degree and direction of filament curvature, while retaining an analogous membrane-binding surface and subunit connectivity.