The ESCRT-III Isoforms CHMP2A And CHMP2B Display Different Effects On Membranes Upon Polymerization

ESCRT-III proteins are involved in many membrane remodeling processes including multivesicular body biogenesis as first discovered in yeast. In humans, CHMP2 exists as two potential isoforms, CHMP2A and CHMP2B, but their physical characteristics have not been compared yet. Here, we use a combination of technics on biomimetic systems and purified proteins to study their affinity and effects on membranes. We establish that CHMP2B binding is enhanced in the presence of PI(4,5)P2 lipids. In contrast, CHMP2A does not display lipid specificity and requires CHMP3 for binding significantly to membranes. On the micrometer scale and at moderate bulk concentrations, CHMP2B forms a reticular structure on membranes whereas CHMP2A (+CHMP3) binds homogeneously. Eventually, CHMP2A and CHMP2B unexpectedly induce different mechanical effects to membranes: CHMP2B strongly rigidifies them while CHMP2A (+CHMP3) has no significant effect. Altogether, we conclude that CHMP2B and CHMP2A cannot be considered as isoforms and might thus contribute differently to membrane remodeling processes.

The ESCRT-III (Endosomal Sorting Complex Required for Transport) complex is involved in a variety of cellular contexts 1 such as biogenesis of multi-vesicular bodies (MVB) 2 , plasma membrane wound repair 3 , neuron pruning 4 , dendritic spine formation 5 , nuclear envelope repair or nuclear envelope sealing during telophase 6,7 abscission at a late step of cytokinesis 8,9 , and budding and release of some enveloped viruses from the plasma membrane of infected cells 10 .
In yeast, the sequence of recruitment of ESCRT-III proteins during MVB formation is Vps20 -Snf7 -Vps24 -Vps2, forming a core complex 12 . Their human homologs are respectively CHMP6 -CHMP4 (A, B, C) -CHMP3 -CHMP2 (A, B). Both CHMP2A and CHMP2B present a high sequence homology with the yeast protein Vps2 and have therefore been considered isoforms. Indeed, CHMP2B appears to be a relatively recent acquisition in the evolution of the ESCRT-III complex resulting from a Vps2 gene duplication event 11 (Supplementary Fig. 1-B). Together, CHMP2A and CHMP2B act in most ESCRT-catalyzed membrane remodeling processes, except in MVB formation 5 , where CHMP2A but not CHMP2B is required and neuronal pruning which requires CHMP2B but not CHMP2A. Yet so far, the dual roles of CHMP2A and CHMP2B in the same process remain elusive 6,[13][14][15][16][17] (Supplementary Fig. 1-C).
ESCRT-III proteins cycle between an inactive cytosolic state [18][19][20] and an activated lumenal state [21][22][23] leading to filamentous polymers forming spirals or helical tubular structures in vitro and in vivo 19,[24][25][26][27][28][29][30][31][32][33][34][35][36][37] . Purified recombinant CHMP2A can coil up into flat snail-like structures 38 or form helical tubular polymers with CHMP3 in the absence of membrane 25 . On the other hand, overexpressed CHMP2B in cells 32 leads to the formation of tubular helical structures, but in vitro assembly of recombinant CHMP2B has never been visualized, neither alone nor together with CHMP3. SiRNA knockdown of individual ESCRT-III proteins demonstrated a minimal requirement of one CHMP4 and one CHMP2 member for HIV-1 release 14 . Also CHMP3 acts synergistically with CHMP2A but not CHMP2B 39 , indicating a distinct role for CHMP2B independently of CHMP3. In contrast, both CHMP2A and CHMP2B are important for cytokinesis 40 . So far, CHMP2A and CHMP2B have been considered as functional homologs, but practically no study has questioned yet whether CHMP2A and CHMP2B behave similarly upon binding to membranes to validate this hypothesis.
Here we have investigated in vitro the functional homology of CHMP2A and CHMP2B in the ESCRT machinery, using biomimetic membrane systems with purified CHMP proteins. We have compared their protein-membrane binding and their mechanical effects on membrane by confocal microscopy, Flow cytometry (FACS), quartz crystal micro-balance with dissipation monitoring (QCM-D) and high-speed atomic force microscopy (HS-AFM). We compare the binding affinities of the proteins to membranes with different charged lipid compositions and investigated the role of CHMP3 on the polymerization of CHMP2A and CHMP2B, respectively.
We confirm that CHMP3 works synergistically with CHMP2A for enhancing their mutual binding towards membranes, but reduces the binding of CHMP2B. We establish that CHMP2B binding is enhanced in the presence of PI(4,5)P2 lipids forming a protein network on the membrane surface, whereas CHMP2A+CHMP3 interact via electrostatics with no phosphoinositide binding specificity and bind homogeneously onto membranes. Moreover, we study the mechanical properties of membranes coated with these different ESCRT assemblies. We show by micropipette aspiration, osmotic shock and HS-AFM deformation that CHMP2A and CHMP2B have opposite mechanical effects on the membrane. While CHMP2B highly rigidifies membranes, CHMP2A+CHMP3 have almost no effect on it. Altogether, our study demonstrates that CHMP2A and CHMP2B cannot be considered as functional homologs. Thus, the observed opposite mechanical properties are likely important for understanding the mechanics of membrane remodeling and membrane scission.

CHMP2B and CHMP2A display different membrane binding characteristics
Phosphoinositides constitute a minority of the phospholipid family with a concentration lower than 1% in cell membranes. Nevertheless, PIP lipids play an essential role for signaling in cells. PI(3)P is the main phosphatidyl inositide present in the endosomal compartments of the MVB pathway where the ESCRTs were first identified, and this lipid has been used in purified systems to reconstitute MVB formation using yeast proteins 41 . However, ESCRT-III-mediated processes also occur on membranes enriched in PI (4,5)P2, notably at the plasma membrane, including for instance HIV egress, plasma membrane repair and cytokinesis events, or at the nuclear envelope 42,43 . We have first compared the interactions of CHMP2A and CHMP2B with membranes containing different phosphatidyl inositides. To improve protein stability and avoid protein aggregation, CHMP2A was expressed and purified with an MBP tag.
A previous in vitro study 44 has shown that the interaction of CHMP2B proteins with membrane is significantly enhanced in the presence of PI(4,5)P2 lipids in comparison with DOPS or PI(3,5)P2-membranes. Thus, we compared the preferential binding of CHMP2A versus CHMP2B on GUVS doped with 10% PI(4,5)P2 using confocal imaging.
10% PI(4,5)P2-GUV (see composition 1 in the Methods section) were incubated for 30 min with CHMP2A or CHMP2B proteins at a concentration of 500 nM in the Protein Binding buffer (BP buffer), which has been optimized to ensure the highest protein density on the GUV membrane ( Supplementary Fig. 2-A). It has been shown that the displacement or truncation of the C-terminal region of CHMP proteins results in the activation of the protein required for polymerization on membranes 19,21,22,44 . Thereupon, the constitutively active version of CHMP2A and CHMP2B, i.e. their C-terminal truncated mutants, respectively CHMP2A-ΔC and CHMP2B-ΔC, were used in the following experiments.
While CHMP2B shows a homogenous binding to the GUV in these conditions ( Fig. 1-A, first panel), the interaction of MBP-CHMP2A-ΔC is rather weak under the same conditions ( Fig.   1-A, third panel). MBP cleavage increases the interaction but also induces aggregation of CHMP2A-ΔC in solution and on the membrane ( Supplementary Fig. 2

-B).
Previous experiments have shown that in solution, combinations of CHMP2A-ΔC and CHMP3-ΔC as well as CHMP2A-ΔC and CHMP3-FL co-polymerize to form tubular helical structures more efficiently than combinations of CHMP2A-FL and CHMP3-FL or CHMP2A-FL and CHMP3-ΔC 25 . We have thus tested the effect of CHMP3-FL on the polymerization of CHMP2A-ΔC. In all the following, MBP-CHMP2A-ΔC, CHMP2B-ΔC and CHMP3-FL, will be referred to CHMP2A, CHMP2B and CHMP3, respectively.
After incubation of 10% PI(4,5)P2-GUVs with CHMP2A (or CHMP2B) + CHMP3 (500 nM and 2 μM respectively in BP buffer), we found that CHMP2A strongly binds to GUVs in the presence of CHMP3 ( Fig. 1-A fourth panel). The quantification of the fluorescence intensity (see details in Methods section) of CHMP2A on GUVs by confocal microscopy shows that the binding of CHMP2A to the membrane is increased by a factor of at least 2.5 in the presence of CHMP3 ( Fig. 1-B), even when the MBP tag is present, justifying that we keep the tag for the rest of our experiments. Surprisingly, when CHMP3 is incubated with CHMP2B, the binding of CHMP2B is no longer homogenous and appears as patches on the GUV colocalizing with CHMP3 ( Fig.1-A   second panel). The relative amount of CHMP2B on the membrane is decreased by a factor of 2 as compared to the relative CHMP2B amount measured in the absence of CHMP3 ( Fig. 1-B).
To quantify the amount of protein bound to the GUV membrane with higher statistics, we have used Flow cytometry (FACS) 45 and 2% PI(4,5)P2-GUVs incubated with a combination of CHMP2A, CHMP2B with and without CHMP3, respectively at 500 nM for both CHMP2A and CHMP2B proteins and 2 M for CHMP3 for 30 min. The fluorescence intensity of the membrane and of the proteins is proportional to the amount of fluorophores in the membrane and proteins bound to it or present in the detection zone, respectively. From the plot of the protein intensity versus lipid signal for all recorded events, we could isolate the signals corresponding to CHMP proteins bound to GUVs and plot the corresponding histogram of these intensities for the different conditions. The median value of this histogram is related to the average density of proteins bound to GUVs. When CHMP2A or CHMP3 are incubated alone with the 2% PI(4,5)P2-GUV suspension, an extremely weak signal is detected by FACS, but as previously observed by confocal microscopy binding increases significantly by almost 100 times-when both proteins are incubated together, in comparison to CHMP2A alone ( Fig. 1-C). On the contrary, the presence of CHMP3 decreases the binding efficiency of CHMP2B -by approximately 150 % ( Fig. 1-C). We conclude that CHMP2A and CHMP3 synergize in binding to membranes, while CHMP3 acts as a negative regulator for CHMP2B membrane binding in vitro. considering that DOPS has a net charge of -1, PI(4,5)P2 (or PI(3,5)P2) of -3 at pH 7.5, and PI(4)P (or PI(3)P) of -2 46 , we found that CHMP2A+CHMP3 has almost no preference for any phosphoinositide (Fig. 1-D). Within the error bars, the binding efficiency are more or less equal for all the PIP species tested (i.e. PI(4,5)P2, PI(3,5)P2, PI(3)P and PI(4)P), and lower than to DOPS alone.
In contrast, we found that CHMP2B has a stronger affinity for PI(4,5)P2 than CHMP2A+CHMP3 ( Supplementary Fig. 2-D). After charge normalization of the binding density of the PIP species and renormalization by the binding to DOPS lipids, we did not measure a significantly higher binding efficiency of CHMP2B for PI(4,5)P2 membranes than for pure DOPS membranes ( Fig. 1-D, p-value = 0.04), nevertheless much stronger than CHMP2A+CHMP3 ( Fig.   1-D). Moreover, the binding of CHMP2B is almost doubled on PI(4,5)P2 membranes than on PI(3,5)P2. In contrast, we did not observe such a preference for the full length protein CHMP2B-FL ( Fig. 1-D).  Fig. 2-E). The amount of proteins adsorbed to the bilayer increased by 50 % when the amount of DOPS was increased from 30 % to 40 % ( Fig. 1-E). Indeed, increasing the number of negatively charged lipids in the membrane increases the amount of proteins adsorbed on it. This implies that electrostatic interactions play a key role in mediating the interaction between the proteins and the membrane in agreement with the exposure of basic surfaces in ESCRT-III polymers 35,50 .
Furthermore, in order to discriminate between the specific affinity for PI(4,5)P2 lipids and electrostatic interactions, we prepared SLBs with a constant total net charge with either 40 % DOPS or 10 % DOPS + 10 % PI(4,5)P2, the total net charge of these SLBs being equivalent. We observe that CHMP2B density is approximately 60 % higher when PI(4,5)P2 lipids are present in comparison with SLBs made of DOPS only ( Fig. 1-E). Compared to experiments on GUVs ( Fig. 1-D), this higher enhancement is probably due to an effective higher PI(4,5)P2 fraction in the SLBs as compared to the GUVs. Moreover, when PI(4,5)P2 lipids are replaced by the same fraction of PI (3,4,5)P3, the amount of protein bound to the SLB decreases significantly and becomes almost equal to the amount of proteins bound to SLB with 30 % DOPS only, although PI(3,4,5)P3 lipids have a higher negative net charge (-4) as compared to PI(4,5)P2 lipids (-3) 46,51 . Altogether, these experiments further support that CHMP2B preferentially interacts with PI(4,5)P2 lipids.
Globally, our results show that while CHMP2B is capable of binding to membrane alone, the binding of CHMP2A to membranes is greatly enhanced by CHMP3 (Figs. 1B, 1C). Additionally, CHMP3 has an opposite effect on CHMP2B and it reduces the membrane its association (Figs. 1B, 1C). Moreover, we found that the binding of CHMP2A+CHMP3 does not depend on the PIP species present in the membrane composition, in strong opposition with the enhanced binding of CHMP2B in the presence of PI(4,5)P2 lipids. This non-specificity of CHMP2A (+CHMP3) proteins to any of the PIP species including PI(4,5)P2 is in agreement with their presence in most cellular processes involving the ESCRT-III complex 1 , contrary to CHMP2B which is only required for processes occurring at the plasma and nuclear membranes that are enriched in PI(4,5)P2 lipids 3, 52, 53 .

CHMP2A and CHMP2B exhibit different supra-molecular assemblies on membranes
Previous studies have shown that cellular overexpression of CHMP2B leads to helical scaffolds deforming the plasma membrane into long rigid tubes protruding out of the cell 32 .
Similarly, CHMP2A + CHMP3 co-assemble in bulk into helical tubes in vitro 25,39 . Yet, the way CHMP2B or CHMP2A in synergy with CHMP3 assemble in vitro onto membranes has not been addressed. The characterization of the effect of these proteins on deformable model membranes is crucial to understand their mechanical properties and their functioning. Hence we studied the supramolecular assemblies of CHMP2B versus CHMP2A+CHMP3 on 10%PI(4,5)P2-GUVs by spinning disk confocal microscopy.
Above 500 nM protein bulk concentration, CHMP2B proteins fully cover the surface of GUVs with no observable distinctive structure, i.e. no inward or outward tubulation (Fig. 2-A first panel). At optical resolution, CHMP2B proteins appear homogeneously distributed on the surface of the vesicles, besides some protein-lipid patches. At bulk concentration lower than 500 nM, CHMP2B proteins form a peculiar reticular-like network wrapping around the whole vesicle ( Fig. 2-A second panel). All these observations were performed after 15 min GUV incubation in the protein solution. It suggests that, at high bulk concentration, a reticulum-like network forms transiently, becoming denser with time and leading to an apparent continuous coverage at optical resolution. This CHMP2B network colocalizes with PI(4,5)P2 lipids (Fig. 2-B), indicating that CHMP2B recruits negatively charged PI(4,5)P2 lipids, further confirming the specific interaction between CHMP2B and PI(4,5)P2 lipids.
In contrast, the assembly of CHMP2A+CHMP3 appears very uniform optically, devoid of any visible network, independently of the incubation time and protein concentration ( Fig. 2-C and Supplementary Fig. 2-F). In some vesicles (approx. 10 %), we observed CHMP2A (+ nonlabeled CHMP3)-containing short out-ward protrusions ( Fig. 2-C, and zoom-in). These protrusions were however rarely visible on most of the vesicles. We conclude that CHMP2B and CHMP2A-CHMP3 do not tubulate GUV membranes in this concentration range.
We next investigated whether these proteins perturb the mechanical properties of the membranes.

CHMP2A and CHMP2B have opposite mechanical effects on model membranes
To study the mechanical effect of CHMP2B and CHMP2A+CHMP3 on membranes, we first used the micropipette aspiration technique developed by E. Evans 54 , to measure the elasticity of 10 % PI(4,5)P2-GUV (lipid composition 1) coated with CHMP2A or CHMP2B in the presence or not of CHMP3 .
In the absence of CHMP proteins, micropipette aspiration of PI(4,5)P2-GUVs easily induced the formation of a characteristic tongue inside the micropipette (Fig. 3

-A-first panel).
In contrast, PI(4,5)P2-GUVs incubated with a CHMP2B concentration leading to full coverage could not be aspirated and deformed even at high tensions ( Fig. 3-A -second panel) (up to 10 -3 N.m -1 ). However, during aspiration at high tension, in approximately 20 % of the aspirated GUVs ( Fig. 3-B), an occasional rupture of CHMP2B protein coat could be observed, allowing the formation of a short tongue devoid of proteins inside the micropipette (Fig. 3

-A-third panel).
This observation indicates that CHMP2B polymer itself cannot be aspirated or deformed and behaves as a solid shell. Surprisingly, the subsequent CHMP3 incubation with GUVs with preformed CHMP2B polymers on their surface resulted in the softening of the CHMP2B shell, which allowed aspiration of the GUV (Fig. 3-A-fourth panel). The quantification of the percentage of aspirated vesicles at a tension of approximately 10 -3 N.m -1 clearly indicates that while less than 20 % of the CHMP2B-coated GUVs could be aspirated in the absence of CHMP3, generally upon shell rupture, almost 100 % of the CHMP2B-coated GUVs could be aspirated when CHMP3 proteins were added (Fig. 3-B). Thus the data suggests that CHMP2B polymers form a rigid shell around the vesicle that cannot be deformed by aspiration even at tensions as high as a few 10 -3 N.m -1 unless CHMP3 is present. The presence of CHMP3 softens this rigid shell allowing its deformation by the micropipette.
In contrast, PI(4,5)P2-GUVs co-incubated with CHMP2A+CHMP3 could be easily deformed during aspiration with an increase of the tongue length inside the micropipette as a response to the aspiration increase (Fig. 3-A -fifth panel). Fig. 3-C shows the variation of the membrane tension as a function of the fractional excess area, Δα, for two representative experiments. The stretching modulus, χ, calculated from the slope of all the curves for both conditions (see Method section), Fig. 3-D, is within the errors bars not affected by the presence of CHMP2A+CHMP3 on the membrane. It is found to be equal to χ = 10 ± 1 mN.m -1 (N = 30 GUVs) for CHMP2A + CHMP3 covered GUVs, similar to the protein-free GUVs, χ = 25 ± 5 mN.m -1 (N = 20 GUVs). Note that the value of the stretching modulus for the bare lipid membrane is lower than values reported for dioleoyl, DO, chains, in the presence of cholesterol 55 , probably because of the absence of a pre-stretching step in our experiments, as usually performed to suppress any pre-existing uncontrolled excess area 56 . Here, pre-incubation of the GUVs with proteins prevented any pre-stretching of the GUVs in order to limit the contact between pipette and protein-coated GUV and thus adhesion. Nevertheless, our aim was not to measure the absolute stretching modulus of the membranes coated by ESCRTs but to perform measurements relatively to bare lipid membranes in the same experimental conditions. Moreover, the stretching modulus of membrane covered with CHMP2B+CHMP3 could not be measured as the tongue covered with these proteins systematically adhered to the pipette, thus impeded any mechanical measurement. We can however conclude that CHMP2B strongly rigidifies membranes, whereas CHMP2A+CHMP3 membrane interaction does not alter membrane elastic properties.
We next applied different mechanical constraints onto CHMP2B-covered GUVs to test their resistance to mechanical deformations. Spherical GUVs change shape when they are deflated upon an hyperosmotic shock since the surface/volume ratio increases and even becomes unstable when the osmotic shock is too strong 57 . We thus studied the effect of an hyperosmotic shock on 10% PI(4,5)P2-GUVs fully-covered with CHMP2B by increasing the osmolarity in the external medium by salt or sugar addition. We carefully checked that the buffer change did not induce unbinding of the ESCRT-III proteins from the membrane. An osmotic shock equal to 150 % (osmolarity of the external medium ≥ 190 mOsm.L -1 ) transforms spherical protein-free GUVs into elliptical vesicles (Fig. 4-A) with an average eccentricity index equal to 0.72 ± 0.11 (Fig. 4-B) (note that the eccentricity index is the ratio between-foci distance and the major axis-length of an ellipse. It ranges between 0 (for a circle) and 1 (for a linear segment)). At higher osmotic shock, GUVs were completely destabilized in the absence of proteins. On the other hand, CHMP2B-covered GUVs better preserved their spherical shape for the same 150 % osmotic shock (Fig. 4-A) with an average eccentricity index equal to 0.35 ± 0.03 ( Fig. 4-B).
Moreover, in contrast with bare membranes, vesicles covered with CHMP2B proteins could even stand a 300 % osmotic shock in a solution at 500 mOsm.L -1 , showing again that CHMP2B polymer assembly on GUVs' surface preserves vesicles from deformation by forming a rigid shell.
We next aimed to determine the effect of CHMP2B on the mechanical properties of membranes at the nano-scale, mimicking the relevant cellular dimensions. To do this we applied a High Speed-AFM imaging based deformation approach using Small Unilamellar Vesicles (SUVs) with a typical diameter between 20-80 nm. First of all, a difference in surface roughness is observed between the two kinds of vesicles (with and without CHMP2B). Bare SUVs show a smooth surface and CHMP2B-coated vesicles possess a rougher surface, indicating the presence of the protein on the outside of the vesicles ( Fig. 4-C first panel). Next, we increased the imaging force and it can be observed that the vesicles are progressively more deformed. The deformation of the SUVs is measured by recording the height change. (Fig. 4-C second and third panels). To assure that the vesicles had undergone elastic deformation, even at the maximum applied imaging force, the imaging force was reduced again to the minimum value at the end of the experiments. Only vesicles that bounced back to more than 90% of the initial height were considered for the analysis and typically the vesicles did recover their shape and size (movie S1).
After assuring that the vesicles behave elastically, we deduce the relative stiffness krel of the vesicles from the inverse of the slope of height change versus force increment ( Fig. 4-D). In Supplementary Fig.3 we show all the data points on both types of vesicles and the transformation from absolute height to relative height. It can be observed that there is a clear difference between the two kinds of vesicles. For bare SUVs we find krel ≈ 3.7 ± 0.1 (N=31, mean ± SD) and for CHMP2B-coated vesicles we find a 2-fold stiffening with krel ≈6.7 ± 0.1 (N=32, mean ± SD) revealing that CHMP2B also stiffens membranes at physiologically relevant length scales.

DISCUSSION
The objective of our study was to compare the membrane binding properties of ESCRT-III proteins CHMP2A and CHMP2B in vitro in order to determine their capacity to substitute each other during membrane remodeling processes.
First, we show that CHMP2A binding is strongly enhanced in the presence of CHMP3 to bind membranes, in agreement with previous in vivo and in vitro studies 25,26,39,58 , whereas CHMP2B interacts with membrane independently of CHMP3. This is in agreement with the synergy exerted by CHMP3 in the presence of CHMP2A and with the absence of synergy exerted by CHMP3 in the presence of CHMP2B on HIV-1 budding 25,26,39,58 . In fact, we show further that CHMP3 acts as a negative regulator of CHMP2B for membrane interaction and polymerization, as indicated by the reduced binding of CHMP2B to GUV membranes in the presence of CHMP3.
Second, we confirm that CHMP2B displays a stronger binding for PI(4,5)P2 containing membranes as compared to other phosphoinositides and DOPS lipids 44 . In contrast CHMP2A and CHMP3 require only negatively charged membranes for binding with no preference for specific lipid head groups. The binding affinity with PI(4,5)P2 lipids is in agreement with the spontaneous localization of CHMP2B to the plasma membrane enriched in PI(4,5)P2 42 upon VPS4 knockdown 32 . In this context, all ESCRT-driven remodeling processes that involve CHMP2B, take place at PI(4,5)P2-containing membranes such as HIV1 budding , plasma membrane repair, cytokinesis nuclear envelope reformation and dendritic spine formation 8,43 . Thus our in vitro data suggest that CHMP2B recruitment to membranes may be regulated by PI(4,5)P2 and thus PIP signaling.
While CHMP2A and CHMP3 assemble homogenously on the GUV membrane, CHMP2B forms a striking reticulum-like structure at the GUV surface at low density. The network colocalizes with PI(4,5)P2 indicating clustering of PI(4,5)P2 upon CHMP2B network formation.
This network formation leads to a strong mechanical stiffening of the membrane. The CHMP2B coat behaves as a rigid shell that can be occasionally fractured upon strong micropipette aspiration. The effect of CHMP3 on CHMP2B membrane binding/polymerization influences also the stiffness of the membrane by softening it compared to CHMP2B only coated membranes. At smaller scale, on SUVs, this stiffening is also observed. In contrast, the mechanics of GUVs coated with CHMP2A + CHMP3 is almost unchanged. Previous experiments performed on yeast ESCRT-III proteins reported a plastic deformation of membrane coated with Snf7 30 . As a consequence, the mode of action of the ESCRT-III may be regulated by the balance of stiffening and elastic behavior.
In general, in the concentration range explored in our study, with both CHMP2A (+CHMP3) and CHMP2B, we did not observe spontaneous GUV membrane tubulation.
Tubulation depends on protein spontaneous curvature, surface fraction, membrane tension, protein-protein interactions and protein assembly stiffness 59 . Considering the propensity of the ESCRT-III proteins to form spiral or helical polymers in solution, we could have expected that they might also induce membrane deformation upon polymerization on a lipid membrane. One possible explanation is that we have not included CHMP4, an ESCRT-III member essential for all ESCRT-catalyzed processes 8 . Although CHMP4 assembles on flat membranes 30,60 , it seems to prefer negative membrane curvature for interaction 61 . Thus, the CHMP2B and CHMP2A+CHMP3 membrane binding observed here has produced assemblies that are different from ESCRT-III assemblies observed in vitro 62,63 lacking spontaneous curvature and/or being too elastic to deform membranes.
The differences observed between CHMP2A and CHMP2B with regard to their membrane interaction and their capacity to affect membrane rigidity, indicate that both isoforms exert different functions that require different mechanical properties during ESCRT-catalyzed membrane remodeling processes. As an example, the CHMP4B isoform is likely present in the ESCRT-III spirals formed at the mid-body during cytokinesis 64,65 whereas CHMP4C is implicated in abscission control 15 . The increased rigidity imposed by the CHMP2B network might be important for dendritic spine maintenance 66 where it might limit protein diffusion, in agreement with experiments showing that CHMP2B forms a diffusion barrier at membrane necks reconstituted in vitro 44 . It might also significantly contribute to the mechanical property of the ESCRT-III spirals at the cytokinetic bridge that become very loose when CHMP2B is depleted 65 .
Interestingly, CHMP2B function might be modulated by CHMP3, which limits CHMP2Bmembrane interaction and softens the CHMP2B assembly. This indicates that in vivo CHMP3 either limits CHMP2B polymerization or/and copolymerizes with CHMP2B into a structure with different mechanical properties, in agreement with observations of copolymerization of ESCRT-III proteins in solution 60 .
We thus propose that CHMP3 could play a key regulatory role in the sequence of recruitment of CHMP2B and CHMP2A and in their respective stoichiometry on the membranes during ESCRT-III function. In late steps of cytokinesis, pulling forces exerted by daughter cells on the intercellular bridge appear to regulate abscission, allowing daughter cells to remain connected until they have settled in their final locations. Moreover, counter-intuitively, a release of tension conducts membrane scission 67 . Thus, membrane unaltered softness may be important at the very last stage of the membrane scission event carried out by the ESCRT-III complex, whereas a rigid structure would oppose this process. However, a certain degree of membrane rigidity might help the constriction process prior to scission, but at this stage, it is difficult to conclude on this aspect.
In summary our data provides evidence that CHMP2A and CHMP2B polymerize differently on membranes and thereby impose different mechanical properties on the membrane structure. Our data thus strongly argue against a sole redundancy of the CHMP2A and CHMP2B proteins and indicate that different isoforms exert complementary functions within the ESCRT-III system.

GUV preparation for confocal, Spinning Disk and FACS experiments.
GUVs were prepared by spontaneous swelling on polyvinyl alcohol (PVA)-based gels 68  To further characterize and compare the interaction of CHMP proteins on GUVs, we measured total intensity of the protein on the vesicle and normalized this value by the GUV area.
Image acquisition for protein quantification was performed using a confocal microscope composed of an inverted microscope (Eclipse TE2000 from Nikon), two objectives (60x water immersion and 100x oil immersion), a C1 confocal head from Nikon, three lasers (λ=488 nm, λ=561 nm and λ = 633 nm). One confocal plane image was taken for each set tension.

QCM-D experiments.
Supported lipid bilayers (SLB) were generated with or without PIPs lipids.
In the absence of PI(4,5)P2, SLB made of 30% and 40% DOPS-SUV composition were produced with a buffer containing Ca 2+ (150 mM NaCl, 10 mM Tris (at pH 7.5) + 2 mM Ca 2+ ) 41 . After SLB formation, the bilayer was rinsed with the same buffer but supplemented with EDTA (150 mM NaCl, 10 mM Tris pH 7.5, 10 mM EDTA) to remove Ca 2+ excess. SLBs were also produced in the