ATP-dependent force generation and membrane scission by ESCRT-III and Vps4

See allHide authors and affiliations

Science  21 Dec 2018:
Vol. 362, Issue 6421, pp. 1423-1428
DOI: 10.1126/science.aat1839

Membrane scission by ESCRTs

The ESCRT protein complexes are essential for cell division, the release of HIV from infected cells via budding, and other cell processes involving the scission of narrow membrane necks from their inner surface. The unusual inside-directed membrane cutting has made it hard to recapitulate this reaction and understand its mechanism. Schöneberg et al. encapsulated ESCRTs inside lipid vesicles and used optical tweezers to pull out membrane nanotubes. In the presence of adenosine triphosphate, clusters of ESCRTs generated force and constricted the nanotube, eventually severing it. This approach provides a window into the molecular mechanisms involved in the activities of ESCRTs.

Science, this issue p. 1423


The endosomal sorting complexes required for transport (ESCRTs) catalyze reverse-topology scission from the inner face of membrane necks in HIV budding, multivesicular endosome biogenesis, cytokinesis, and other pathways. We encapsulated ESCRT-III subunits Snf7, Vps24, and Vps2 and the AAA+ ATPase (adenosine triphosphatase) Vps4 in giant vesicles from which membrane nanotubes reflecting the correct topology of scission could be pulled. Upon ATP release by photo-uncaging, this system generated forces within the nanotubes that led to membrane scission in a manner dependent upon Vps4 catalytic activity and Vps4 coupling to the ESCRT-III proteins. Imaging of scission revealed Snf7 and Vps4 puncta within nanotubes whose presence followed ATP release, correlated with force generation and nanotube constriction, and preceded scission. These observations directly verify long-standing predictions that ATP-hydrolyzing assemblies of ESCRT-III and Vps4 sever membranes.

Cellular membranes are constantly remodeled in the course of vesicular trafficking, cell division, the egress of HIV, and many other processes. Membranes can bud and be severed either toward or away from the cytosol. The latter is referred to as reverse-topology scission and is catalyzed by the endosomal sorting complexes required for transport (ESCRT) machinery, a set of ~18 proteins in yeast and ~28 in mammals (14). The core machinery of membrane scission by the ESCRTs consists of the ESCRT-III protein family. The most important components for membrane scission are Snf7, Vps24, and Vps2 (5, 6). When recruited to membranes, ESCRT-III proteins assemble into flat spiral disks (79), helical tubes (7, 10, 11), or conical funnels (1113). ESCRT filaments have a preferred curvature (8, 9, 14). When they are bent to curvatures of higher or lower values, ESCRT filaments act as springs that restore their own shape to the preferred value (9, 15, 16). This spring-like behavior has led to the prediction that ESCRTs exert measurable forces upon membranes, which we set out to test.

The AAA+ ATPase Vps4 (17) is intimately associated with the ESCRT machinery and is essential for the membrane scission cycle. Vps4 is recruited to scission sites by Vps2 (18, 19). Vps2 is thought to have a capping role whereby it inhibits Snf7 polymerization (6). By recycling Vps2 (20), Vps4 promotes Snf7 polymerization. Thus, Vps4 is critical for the recycling of ESCRT-III and the replenishment of the soluble cytoplasmic pool. Early attempts at in vitro reconstitution of ESCRT-mediated budding and scission using giant unilamellar vesicles (GUVs) suggested that the process was independent of Vps4 and ATP (21, 22), except for the final postscission recycling step. Cell imaging studies, however, showed that Vps4 localization peaked prior to scission in HIV-1 budding and cytokinesis (2328), consistent with its direct role in scission upon ATP hydrolysis. A second goal of this study was to determine if Vps4 and ATP hydrolysis are directly involved in membrane scission, as opposed to mere recycling.

We encapsulated into GUVs the minimal ESCRT-III-Vps4 module containing yeast Snf7, Vps24, Vps2, and Vps4 (referred to here as the module) (fig. S1) in a mixture of palmitoyloleoyl-phosphatidylcholine, palmitoyloleoyl-phosphatidyl-serine, and biotinyl–phosphatidylethanolamine (80:20:0.1) at near-physiological ionic strength (~150 mM NaCl) (Fig. 1, D to G). We used optical tweezers to pull nanotubes extending between the surface of a GUV held by suction on an aspiration pipette and the surface of a streptavidin-coated polystyrene bead held by an optical trap (Fig. 1, A to C). To fuel the AAA+ ATPase Vps4, we also encapsulated the caged ATP analog P3-(1-(2-nitrophenyl)ethyl)ester–ATP (NPE-ATP). An optical fiber was used to illuminate with UV one GUV at a time, so that experiments could be carried out sequentially on individual GUVs in the same microfluidic observation chamber. In control experiments, where all components were included except for ATP, UV illumination led to no change in the force exerted on the bead (Figs. 1H and 2B). In similar control experiments omitting only Vps4, UV illumination resulted in a slight drop in the pulling force (Figs. 1I and 2C), attributed to the generation of two product molecules upon NPE-ATP uncaging. Thus, in the absence of ESCRT activity, the membrane nanotube was stable. When ATP was uncaged in the presence of the complete ESCRT module, a large rise in retraction force was indeed observed (Figs. 1J and 2I and movie S1). Over ~2 to 10 min (Fig. 1J), the force exceeded the trap maximum of ~65 pN and pulled the bead out of the laser trap (movie S1). This showed that in the presence of ATP, the ESCRT module can exert forces on membranes.

Fig. 1 ESCRT-III exerts an ATP-dependent force on membrane tubes.

(A) Schematic of the experiment: a membrane tube (middle) is pulled out of a micropipette-aspirated GUV with a functionalized bead held in an optical trap, creating a reverse-curvature topology. Components of the ESCRT machinery and caged ATP are encapsulated in the lumen of the GUV. An optical fiber delivers UV light to uncage the ATP inside the vesicle and start the reaction. (B) Aspiration pipette, optical fiber, and UV light cone (blue) inside the microfluidic chamber for our experiments. (C) GUV (center) aspirated by the micropipette (left) and a tube-pulling bead (right). (D to G) Labeling different components of the ESCRT module [membrane label in (D), Snf7 in (E), Vps4 in (F); merged image in (G); 2 μM for all components] revealed a uniform distribution of proteins in the lumen of GUVs. (H to J) Force profiles over time detected by optical tweezers on a membrane tube pulled from a GUV that encapsulates components of the ESCRT module. (H) Control experiment on a full ESCRT module but with ATP omitted. No change in force could be measured. (I) Control experiment on a full ESCRT module but with Vps4 omitted. Apart from a minute dip in the force profile, which was due to small changes in osmolarity upon ATP uncaging, no effects were detected. (J) ATP uncaging (dashed line) in the presence of a full ESCRT module leads to a rise in force exerted on the tube, which is connected to the bead held by the optical trap. A large rise in force can overcome the trapping strength and pull the bead out of the trap.

Fig. 2 Molecular determinants of force generation.

(A) Interaction network of the ESCRT module. ESCRT proteins (space-filling structures and dashed lines) interact with the membrane (gray, bottom) as well as with each other. Key components (ATP), catalytic sites, and interacting motifs are highlighted in colors. (B to K) Individual components of the module contributed differently to the force exerted on membrane tubes: ATP and Vps4 were essential for force generation (B and C), and so was ATP hydrolysis (J, catalytically dead Vps4EQ mutant) and critical protein domains (K, Vps2 ΔMIM1). Only the full module (I), or pairings of Vps2 with Vps24 and Snf7 with Vps2 (G and H), led to significant force generation. Our data are consistent with the known interactions underpinning the activity of the ESCRT machinery.

We sought to determine which components of the ESCRT module were required for force generation (Fig. 2A). With Vps2 or Vps24 as the only ESCRT-III subunits, essentially no force was generated (Fig. 2, D and E). In the presence of Snf7, omission of Vps2 or Vps24 led to little or no force generation (Fig. 2, F and H), consistent with the role of Vps2 in coupling of ATP hydrolysis by Vps4 to ESCRT-III remodeling and a role of Vps24 in copolymerizing with Vps2. When both Vps2 and Vps24 were present, but Snf7 was missing, a force rise of up to 12 pN was produced, consistent with the ability of Vps24 and Vps2 to co-polymerize (10, 20) (Fig. 2G). The inactivated mutant E233Q protein (Glu233→Gln) of Vps4 (17) failed to generate force (Fig. 2J). Deletion of the Vps4-coupling MIM1 motif of Vps2 (Fig. 2K) (18, 19), which is essential for biological function, abrogated force production. Thus, the ability of the ESCRT module to exert forces on nanotubes, in an ATP-dependent manner (Fig. 2, B and C), correlates closely with the presence of all the components that are crucial for ESCRT-mediated membrane scission and their individual integrity (Fig. 2I).

We integrated a confocal microscope with optical tweezing capability to image membrane nanotubes pulled from GUVs containing fluorophore-labeled ESCRTs (Fig. 3, A to E, and fig. S2). By pulling on bare membranes, we obtained the bending modulus κ and standardized the calculation of the membrane nanotube radius (fig. S3). To maximize the signal in these experiments, Snf7 was labeled with the photo-stable dye Lumidyne-550 and imaged with a resonant scanner and a gallium arsenide phosphide (GaAsP) detector. We quantitated Snf7, Vps4, and membrane intensity using Gaussian fitting to the diffraction-limited tube profile (fig. S4).

Fig. 3 Confocal imaging of ATP-dependent membrane tube scission by ESCRTs.

(A to E) Micrograph of the experimental setup: the GUV (center, yellow) encapsulating the ESCRT module and caged ATP was aspirated by a micropipette (left). A membrane nanotube was pulled from the GUV using a bead (right) in an optical trap. Membrane, Snf7, and Vps4 are labeled with different fluorophores [composite (A) of individual channels shown in (B to E)]. Scale bar: 2 μm. (F and G) Progression of two representative membrane scission events (2 of 17). UV illumination at t = 0 s. Snf7 (green) and Vps4 (cyan) puncta became visible at the vesicle-tube junction and migrated into the tube. Scission happened at ~410 s and ~650 s, respectively. Scale bar: 1.5 μm. (H and I) Tube remnants on the bead identified scission in the middle of the tube. [Brightness adjusted compared to (F) and (G).] (J) Quantification of tube scission. (K) Protein copy numbers in puncta.

We monitored the response of a total of 46 nanotubes pulled from ESCRT module-filled GUVs following UV illumination. The nanotubes manifested a force increase and accumulation of Snf7 and Vps4 (Fig. 3, F to I, and movie S2). Of the 46 trials, 38 (83%) led to scission (Fig. 3J) as judged by simultaneous disappearance of the tube, sudden decrease in the force to zero, and appearance of membrane-, Snf7-, and Vps4-containing remnants on the trapped bead (Fig. 3, H and I). Tube lifetimes were widely distributed (Fig. 3J), with a mean lifetime of 425 s before scission. The distributions of lifetimes were similar at the superphysiological concentration of 2 μM, which was used to facilitate visualization, and the near-physiological concentration of 200 nM (Fig. 3J). One to two diffraction-limited puncta of Snf7 intensity appeared at >7 SD in 11 of 17 events analyzed. Typically, the puncta nucleated at the tube-vesicle junction (Fig. 3, F and G, and movie S2). Subsequently, the puncta sometimes moved or disappeared within the tubes. Snf7 puncta were essentially always colocalized with Vps4 (Fig. 3, F and G). At 2 μM, puncta contained in the range of 100 to 600 copies of Snf7 (Fig. 3K), which could exceed the minimum needed for scission yet is also roughly consistent with estimates in yeast cells (29).

To understand the relationship between the observed force (Fig. 4, A and B) and bulk and microscopic properties of the system, we quantitated the nanotube radius over time. Nanotubes used in this study were typically of a radius (r) of ~20 nm prior to ATP release (Fig. 4, C and D). The tubes began to narrow almost immediately following ATP release. Narrowing was associated with an increase in the amount of Snf7 and Vps4 seen in the tubes (Fig. 4, E and F), while essentially no change was seen in the intensity of the GUV membrane or the amount of Snf7 and Vps4 associated with the GUV (Fig. 4, G and H). This behavior was consistent over 17 traces (Fig. 4, K to N). Final values of r reached 5 to 10 nm by the time of scission. The uncertainty in the final values of r was substantial, because the membrane fluorescence signal was weak when r was <10 nm. In the case of constant membrane tension σ and bending modulus κ, the initial and final force and radius are related to each other by the equations f/fo = (r/ro + ro/r)/2 and Embedded Image. Figure 4, A to D, shows the values of f (Fig. 4, A and B) and r (Fig. 4, C and D) computed from the experimental values of r and f, respectively, with no adjustable parameters. The correlation between the measured and computed values was generally excellent (Fig. 4O). Consistent with an apparently constant κ, there was no accumulation of Snf7 on the GUV membrane over the course of the scission events (Fig. 4I). Consistent with a constant σ, the length of the membrane tongue in the pipette was typically constant (Fig. 4J). The formation of ESCRT puncta imposes nanotube radii smaller than the equilibrium radius ro, which is associated with an increased pulling force, f > fo. These findings are consistent with force generation by punctate microscopic assemblies of ESCRTs within the nanotubes, leading to the constriction of membrane tubes followed by scission.

Fig. 4 ESCRT membrane scission is mediated by tube constriction.

(A and B) Force profiles of the same events shown in Fig. 3 were normalized to their respective baseline tube holding force, f0. ATP released by UV uncaging led to a rise in force and eventually scission (indicated by the sharp drop of force at the end of the trace, 2 of 17). (C and D) Simultaneous observation of the tube radius revealed a continuous radius decrease. Force and radius plots are overlaid with predictions from the Helfrich model for a tube of unchanged bending rigidity under constant tension (dashed orange lines) (see methods description in the supplementary material). (E and F) Normalized and bleaching-corrected fluorescence intensity profiles of the scission events for the tube in the three fluorescence channels (Snf7, green; Vps4, cyan; membrane, red). SDs are shown as shaded areas. After ATP uncaging, the tube membrane fluorescence intensity decreased, while the tube Snf7 and Vps4 intensities increased. (G and H) The GUV membrane fluorescence intensity profile over the time course of the scission event was essentially constant and showed no correlation with the force (Snf7, green; Vps4, cyan; membrane, red). SDs are shown as shaded areas. (I) Fluorescence intensity profiles of the GUV membrane at the start and at the end of the scission trace. No increase in protein fluorescence on the membrane was detected. (J) GUV aspiration projection at the beginning, middle, and end of the scission trace, showing essentially no change. (K to N) Averages (N = 17) ± SD of the force, radius, and tube fluorescence showed a consistent force rise, radius decrease, and fluorescence increase in the tube. (O) Correlation (r) between experiments and prediction for radius and force (N = 17).

It has been inferred that the core ESCRT-III proteins Snf7, Vps24, and Vps2, together with Vps4, comprise the minimal ATP-dependent scission machinery (1, 4, 30, 31). Here, we directly confirmed this idea by visualizing scission in a minimal system that replicated a wide range of biologically validated structure–function relationships.

The most notable finding from the reconstituted system is that the core ESCRT-III proteins and Vps4 together exert an ATP-dependent axial force on the nanotube before severing. It was previously proposed (8) and then demonstrated (9) that Snf7 filaments have a preferred curvature and may exert forces when bent above or below their preferred value. It had also been hypothesized that breakage or remodeling of ESCRT filaments by Vps4 could contribute to force generation (4, 15, 20, 32). Our observations now provide experimental confirmation that ESCRTs indeed can generate force from within a narrow membrane tube, and our results show that this force contributes to membrane constriction and is correlated with reverse-topology membrane scission.

ESCRTs (7, 10, 11) and the “normal topology” scission factor dynamin (33) have been visualized as cylindrical membrane coats, and Snf7 has also been seen in the form of large spirals of hundreds to thousands of copies (9). In the case of dynamin, extended coating of the tube is not needed, and one or a few rings appear capable of mediating scission (34). Our measurements of scission by diffraction-limited puncta of Snf7 are consistent with imaging in cells (29) and with scission by rings or cones of molecular dimensions. The mechanical nature of constriction and force generation remains to be elucidated through structural approaches.

Supplementary Materials

Materials and Methods

Figs. S1 to S4

References (3537)

Movies S1 and S2

  • Department of Chemistry, University of Puget Sound, Tacoma, WA, USA.

  • § Department of Medical Biochemistry and Biophysics, Wallenberg Centre for Molecular Medicine, Umeå University, Umeå, Sweden.

  • || Department of Molecular Biology and Genetics, Johns Hopkins University School of Medicine, Baltimore, MD, USA.

References and Notes

Acknowledgments: We thank J.-Y. Lee, H. Aaron, S. Ruzin, and D. Schichnes for assistance with imaging; M. Vahey, D. Fletcher, and P. Lishko for advice on the aspiration pipette setup; A. Lee for assistance with the optical trap force calibration; and C. Glick for advice with the microfluidics. Funding: Research was supported by a Marie Skłodowska-Curie postdoctoral fellowship smStruct (J. S.), an NSF predoctoral fellowship (M.R.P.), NIH grant R56AI127809 and R01GM032543 (C.B.), DOE grant DE-AC02-05CH11231 (C.B.), the Howard Hughes Medical Institute (C.B.), NIH grant R01AI112442 (J.H.H.), the Max Planck Society (A.H.B. and G.H.), and the German Research Foundation CRC 807 (A.H.B. and G.H.). Author contributions: Conceptualization, J.S., M.R., D.H.G, S.Y., M.R.P., A.H.B., C.B., G.H., and J.H.H; methodology, J.S., S.Y., M.R., D.H.G, A.H.B., M.R.P., I.-H.L., and L.-A.C.; Confleezers, J.S., S.Y., and I.-H.L., software, J.S.; formal analysis, J.S. and A.H.B.; investigation, J.S., S.Y., M.R., M.R.P., A.H.B., and D.H.G; resources, J.S., S.Y., M.R., M.R.P., I.-H.L., L.-A.C., and X.R.; data curation, J.S; writing of original draft, J.S. and J.H.H; manuscript review and editing, J.S., S.Y., M.R., M.R.P., I.-H.L, L.-A.C, A.H.B., D.H.G., X.R., G.H., C.B., and J.H.H ; visualization, J.S.; supervision, G. H, C.B., and J.H.H. Competing interests: The authors declare no competing interests. Data and materials availability: All data are available in the main text or the supplementary materials. Code is available at
View Abstract

Stay Connected to Science

Navigate This Article