The endosomal sorting complex required for transport (ESCRT)-III is associated with a multitude of cellular processes involving membrane remodeling and abscission. The exact composition of ESCRT-III and the contribution of individual ESCRT-III family members to these diverse functions is unclear. Most of the currently available information about ESCRT-III was obtained with tagged, largely non-functional proteins, which may not correctly reflect the in vivo situation. Here, we performed a comprehensive biochemical analysis of ESCRT-III localization and composition in yeast under purely native conditions. Most of our findings are in line with the current concepts about ESCRT-III, but some findings are unexpected and call for adjustments to the model. In particular, our data suggest that the distinction between bona fide ESCRT-III components and ESCRT-III associated proteins is not justified. We detected a single complex containing all ESCRT-III members (except of Chm7) with Did2 as its main component. The classical core components were present in equimolar amounts. Our analysis of the impact of single deletions on the composition of ESCRT-III confirmed the central role of Snf7 for ESCRT-III assembly. For the other ESCRT-III family members predictions could be made about their role in ESCRT-III assembly. Furthermore, our cell fractionation points to a role of Vps20 at the endoplasmic reticulum.
Endosomal sorting complex required for transport (ESCRT)-III proteins are required for a multitude of processes occurring on cellular membranes. They are involved in sorting of proteins to vacuoles/lysosomes for degradation, in cytokinesis, retrovirus budding from the cell surface, membrane repair at the plasma membrane or the nucleus, resealing of the nuclear membrane during mitosis, autophagy, maturation of RNA viruses at the endoplasmic reticulum (ER) membrane, transport of viruses across the nuclear membrane and neuron pruning [1,2]. These processes are accompanied by membrane deformations and membrane abscission events.
ESCRT proteins were first identified in the yeast Saccharomyces cerevisiae in genetic screens for mutants blocked in protein transport to the vacuole [3,4]. These mutants were grouped according to their vacuolar morphology . In one class of mutants (‘class E mutants'), endocytic cargo proteins accumulated in a dot-like structure close to the vacuole (‘class E compartment'). These structures consist of stacked flattened membrane cisternae presumably corresponding to distorted endosomes [6,7]. The mutants are defective in the formation of intraluminal vesicles (ILVs) at the endosomal membranes. Endocytic cargo proteins destined for degradation in the vacuole are incorporated into the forming vesicle, which exhibits an inverse topology, i.e. the membrane bends away from the cytosol. Cargo proteins are usually marked by ubiquitination for incorporation into the vesicles. The resulting ILV containing endosome is called multivesicular body (MVB).
The class E proteins were grouped into several protein complexes called ESCRT-0, -I, -II -III and Vps4, which are thought to act sequentially [8–12]. The chain of events starts with cargo recognition by ESCRT-0, which then binds ESCRT-I. ESCRT-II bridges ESCRT-I and ESCRT-III. Finally, ESCRT-III recruits Vps4, which disassembles the ESCRT complexes and resets the membrane for a new round of ILV formation.
Here, we focus on ESCRT-III, which in combination with the AAA-ATPase Vps4 seems to be the main factor in membrane remodeling. ESCRT-III, a hallmark of eukaryotic cells , is composed of small hydrophilic proteins, which form a protein family with eight members in yeast and twelve in mammalian cells. There are mammalian counterparts (called CHMPs) for each of the yeast proteins, but the mammalian family is expanded through the formation of isoforms. The ESCRT-III proteins seem to share the same 3D-structure [14,15] and have a propensity to form filaments [14,16–20].
The conventional view is that yeast ESCRT-III consists of four proteins: Snf7/CHMP4, Vps2/CHMP2, Vps20/CHMP6 and Vps24/CHMP3. The other members of the protein family (Did2/CHMP1, Ist1/IST1, Mos10 (Vps60)/CHMP5 and Chm7/CHMP7) are called ESCRT-III like or ESCRT-III associated proteins. The prevailing view is that the ESCRT-III complex is assembled sequentially . First, two molecules of Vps20 are recruited to the endosomal membrane by interaction with the two Vps25 subunits of ESCRT-II. Snf7 is thereby converted from an inactive closed conformation to a polymerization competent open conformation. An Snf7 filament is formed. Filamentation is curbed by Vps2 and Vps24, which recruit Vps4 (maybe in conjunction with the ESCRT-III like proteins Did2, Ist1 and Mos10), which ultimately disassembles ESCRT-III.
This view has been challenged by recent results [22,23]. In two studies, the association of ESCRT-III proteins with the midbody during cytokinesis and with ILV budding sites during MVB formation was investigated. Contrary to the sequential model, there seems to be a continuous, stochastic exchange of ESCRT-III proteins and Vps4 with these structures. Thus, ESCRT-III appears to be far more dynamic than originally thought. Interestingly, Vps4 does not seem to be required to sustain ESCRT-III dynamics, but rather seems to be necessary for membrane invagination and ILV abscission.
Here, we examined the localization and composition of ESCRT-III in yeast under purely native conditions. In previous studies, information about the localization and composition of ESCRT-III was mainly obtained with tagged proteins that are largely non-functional. Thus, it is not clear whether these data correctly reflect the in vivo situation. We also examined the impact of deletion of ESCRT-III proteins on the composition of ESCRT-III. Most of our findings are in line with the current way of thinking, but some are unexpected and call for adjustments to the model.
Purification of 6His-tagged ESCRT-III proteins from Escherichia coli
An overnight culture of the E. coli strain BL21 (DE3) transformed with plasmids expressing 6His-ESCRT-III protein fusions (listed in Supplementary Table S1) was diluted 1 : 20 into LB-medium (0.5% yeast extract, 1% tryptone, 1% NaCl) with 15 µg/ml kanamycin and grown at 37°C for 1.5 h. Expression of the fusion proteins was induced by addition of 1 mM isopropyl-β-d-thiogalactopyranoside (IPTG). After 2 h at 37°C, cells were harvested, washed with 1 ml of lysis buffer (300 mM NaCl, 50 mM NaH2PO4, 10 mM imidazole, pH 8) and resuspended in 1 ml lysis buffer. Cells were lysed by sonication (3 × 10 s, 0.9 s pulse, 0.1 s pause, 70% amplitude) and cell debris was removed by centrifugation at 10,000g for 10 min at 4°C. The cleared lysate was mixed with 100 µl of a 50% suspension of Ni-NTA beads (Qiagen, Hilden, Germany). After 2 h at 4°C on a rocker, the beads were washed three times with 1 ml of lysis buffer. The bound proteins were eluted from the beads with 1× SDS–PAGE sample buffer at 95°C for 5 min. The purified proteins were separated on an SDS–PAGE gel and stained with Coomassie blue (Supplementary Figure S3B).
Rabbits were immunized with 6His-tagged ESCRT-III proteins purified from E. coli with Ni-NTA beads by a scale-up of the procedure described above. For the purification of 6His-Vps24, a codon optimized variant expressed from plasmid scVPS24 was used . Immunizations were performed by Seramun Diagnostica (Heidesee, Germany). The specificity of the antisera was tested by Western blotting by comparing the signals obtained from a wild-type strain with the signals from the corresponding single ESCRT-III deletion strain (Supplementary Figure S1).
Sucrose density gradient centrifugation
Yeast cells were grown overnight to exponential phase at 30°C in YPD medium (1% yeast extract, 2% peptone and 2% glucose). Fifty OD of cells was harvested by centrifugation at 4000g for 5 min, washed in 5 ml cold 10 mM NaN3 and resuspended in 100 µl of STE10 lysis buffer (10% w/w sucrose, 10 mM Tris–Cl pH 7.6, 1 mM EDTA, 1 mM DTT, +proteinase inhibitors). The cells were lysed by shaking with glass beads for 5 min at 4°C. After addition of 500 µl of lysis buffer, the cell extract was centrifuged at 500g for 10 min to remove cell debris. Four hundred microliters of the cleared cell extract was loaded onto a 5 ml sucrose gradient.
The sucrose gradients were prepared as follows: 1.7 ml each of 20, 36 and 53% w/w sucrose solutions (in 10 mM Tris–Cl pH 7.6, 1 mM EDTA, 1 mM DTT) were carefully layered on top of each other in a 5 ml ultracentrifuge tube. The tubes were sealed and turned in to a horizontal position. After 3 h of diffusion, the tubes were again turned into an upright position and were now ready to use. Centrifugation was performed at 100,000g (37 000 rpm) for 16 h at 4°C in a Sorvall Discovery M120SE centrifuge with S52-ST rotor. Eighteen 280 µl fractions were collected from the top of the gradient, diluted with one volume of 2× SDS–PAGE sample buffer, and denatured at 50°C for 15 min.
Yeast cells were grown overnight to exponential phase at 30°C in YPD medium (1% yeast extract, 2% peptone and 2% glucose). 320 OD of cells (OD600 ≤ 1) were harvested by centrifugation at 4000g for 10 min, washed once in 15 ml of ice-cold PBS (150 mM NaCl, 10 mM NaH2PO4, pH 7) and resuspended in 1.6 ml PBS with protease inhibitors. For cell lysis, the cells were agitated with 11 g of glass beads (0.4–0.6 mm diameter) in a FastPrep bead mill cell disrupter (Fisher Scientific, Schwerte, Germany) two times 1 min at 4 m/s in the cold room. The lysate was diluted with 8 ml PBS (+protease inhibitors) and the liquid was removed from the glass beads. The cross-linker 3,3′-dithiodipropionic acid di(N-hydroxysuccinimide ester) (DSP) was added to a concentration of 2.5 mM, and the samples were incubated for 30 min at RT on a rocker. The reaction was quenched with 65 mM Tris–Cl pH 8. In addition, 1.25% of Triton X-100 was added to solubilize the membranes. After 15 min at RT on a rocker, 100 µl were collected as the input sample, which served as a reference for the determination of the total protein content. After removal of cell debris by a 10 min 500 g spin, the supernatant was divided into eight aliquots. To each aliquot, 10 µl of anti-ESCRT-III antisera was added. One sample, to which no antibody was added, served as a negative control. Samples were incubated for 1 h at 4°C on a rocker. Then protein A sepharose beads were added to 3% and incubation was continued for 1 h. Then the beads were washed three times with PBS. Centrifugation was carried out at the slowest centrifugation speed possible (100 rpm in a tabletop centrifuge) for 1 min, to avoid unspecific sedimentation of ESCRT-III complexes. The bound proteins were eluted from the beads by heating with 1x SDS-sample buffer at 50°C for 15 min and were then separated on an SDS–PAGE gel and examined by Western blotting with anti-ESCRT-III antibodies.
Determination of co-immunoprecipitation (co-IP) efficiency
In this study, eight different strains were analyzed (the wild-type, and the single ESCRT-III deletion strains, except Δchm7, Supplementary Table S2). With each cell extract, seven immunoprecipitations (IPs) were performed. The immunoprecipitated proteins from each set were loaded onto an SDS–PAGE gel together with the input sample and the no-antibody control. An example of the Western blot analysis is shown in Supplementary Figure S5.
The intensities of the Western blot signals were quantified with the program ImageJ. From the comparison of input and IP signal intensities, the percentage of protein recovered in the immunoprecipitates was determined. Primary IP efficiencies were in the range of 10–30%, co-IP efficiencies in the range of 1–7%. Occasionally, a weak signal was observed in the no-antibody control due to non-specific precipitation of ESCRT-III complexes. This signal was subtracted from the measured co-IP signals. The experiments were performed in triplicate. The average values for the co-IP efficiencies (with their standard deviations) are listed in Supplementary Table S4.
We also noted that under native conditions, some of the ESCRT-III antisera displayed a certain degree of cross-reactivity against other ESCRT-III proteins. Some sera (e.g. anti-Did2, anti-Snf7, anti-Vps2) were quite specific, while others showed substantial cross-reactivity. An estimate of the cross-reactivity was obtained from the single ESCRT-III deletion strains. Any signals obtained from IPs with antibodies directed against the protein that is not present in the cell extract must be derived from a primary IP with cross-reacting antibodies. The cross-reactivities are listed in Supplementary Table S4. The values for the cross-reactivities were subtracted from the measured co-IP efficiencies.
Finally, the co-IP efficiencies were normalized to 100% based on the primary IP efficiencies of the different ESCRT-III proteins (i.e. we assume that the immunoprecipitated fraction is a representative sample of the whole protein present in the cell extract). If, for instance, the primary IP efficiency for Did2 is 27%, and the anti-Did2 antiserum co-precipitates 3.4% of Snf7, then the normalized co-IP efficiency for Snf7 is 12.6%.
In the last step of our calculations, the co-IP efficiencies were converted into molecule numbers by making use of our determined molecule numbers per cell (Table 1, e.g. 12.6% co-IP corresponds to 8800 × 0.126 = 1108 molecules of Snf7 per cell in a complex).
Average of four independent determinations.
SD, standard deviation.
Intracellular distribution of ESCRT-III proteins
ESCRT-III proteins are involved in a multitude of cellular functions [1,2]. In yeast, however, only one function, the role in sorting of cargo proteins in the endocytic pathway, has been clearly established. To get a general idea about additional ESCRT-III functions, we looked at the intracellular distribution of ESCRT-III proteins. In previous studies, the localization of ESCRT-III proteins was examined by fluorescence microscopy with fluorescent protein fusions. The problem with this approach is that fusion of GFP to the C-terminus of ESCRT-III proteins interferes with their function . In case of Snf7, for instance, this gives rise to a dominant negative protein. Insertion of a linker between the C-terminus of Snf7 and GFP somewhat ameliorated this problem by converting the dominant negative protein into a recessive, yet still non-functional mutant protein. Since most of the available information about ESCRT-III proteins in yeast was obtained with non-functional proteins, it is not clear whether it correctly reflects the in vivo situation. Here, we studied the localization of native, untagged proteins by the use of specific antibodies. This is also the first comprehensive study examining all eight ESCRT-III proteins in yeast.
To determine the intracellular localization of ESCRT-III proteins, yeast cell extracts were fractionated on sucrose density gradients. The distribution of the ESCRT-III proteins was compared with the distribution of organellar marker proteins (Figure 1 and Supplementary Figure S2). The ER marker Dpm1 had its main peak in fraction 5 and the second peak in fraction 12 (Figure 1A). Single peaks were obtained for the vacuolar marker alkaline phosphatase (ALP) (fraction 9) (Figure 1B), the endosomal marker Pep12 (fraction 6/7) (Figure 1C) and the plasma membrane marker Pma1 (fraction 15) (Figure 1D).
Fractionation of marker proteins on sucrose gradients.
For the ESCRT-III proteins, three types of fractionation profiles were observed (Figure 2 and Supplementary Figure S2). The first class is represented by Chm7 (Figure 2A) and Vps20 (Figure 2B), which had their main peaks in fractions 4/5 pointing to an ER localization, in line with previous results . Vps20 had an additional peak in fraction 9, which could correspond to a vacuolar pool of Vps20. The second class consists of Did2 (Figure 2C), Ist1 (Figure 2D), Vps2 (Figure 2E) and Vps24 (Figure 2F), which fractionated in the middle of the gradient with a similar profile. For Did2 and Vps24 we observed two peaks in fractions 6 and 9, in line with an endosomal and vacuolar localization of the proteins. The Ist1 distribution looked similar to the Did2 and Vps24 distribution, but was broader and extended more towards the denser fractions of the gradient. Vps2 was more restricted to the endosomal peak in fraction 7, only a hint of a shoulder was seen in fraction 9. The third class is formed by Snf7 (Figure 2G) and Mos10 (Figure 2H). These proteins were mostly soluble (fractions 1–3). For Mos10 no membrane-associated fraction was obvious, for Snf7 a shoulder in fraction 6 was visible, i.e. a small portion of the protein seemed to be associated with endosomes. This is in line with previous fractionation experiments, showing that only ∼20% of the protein is associated with membranes at a given time . Remarkably, despite this low Snf7 membrane pool, the second class of ESCRT-III proteins (Did2, Ist1, Vps2, Vps24) was almost exclusively membrane associated, suggesting that their stable association with the membrane does not require the constant presence of Snf7. This is an unexpected finding since Snf7 is considered to be the main and central component of ESCRT-III . On top of the described localizations, no indication of a plasma membrane pool of ESCRT-III proteins was obtained in our experiments.
Fractionation of ESCRT-III proteins on sucrose gradients.
Determination of ESCRT-III protein copies per cell
There are estimates about the stoichiometry of the ESCRT-III subunits in the complex, but again mostly obtained with tagged proteins. Due to potential tagging artifacts, this may not be a true representation of the natural situation. Also, the ESCRT-III family as a whole has not been examined so far. To be able to determine the composition of native, undisturbed ESCRT-III, we first needed to know the copy number of the individual ESCRT-III proteins in a cell. The copy number was determined by comparing the intensity of Western blot signals from a defined amount of yeast cell extract to a calibration curve with ESCRT-III protein standards. The ESCRT-III standard proteins were purified from E. coli by Ni-NTA affinity chromatography. The protein concentrations of the purified proteins in turn were obtained from a BSA standard curve. Representative gels are shown in Supplementary Figure S3 and Supplementary Figure S4. The calculation of the copy number is detailed in the legend to Supplementary Figure S4. The results from four independent measurements are summarized in Table 1.
Unexpectedly, the highest copy numbers per cell of all ESCRT-III proteins were measured for Did2 (12 700) and Chm7 (10 300). These numbers, however, should be treated with caution, since due to the variability of the data, some of the differences may not be significant (Supplementary Table S3). For instance, the probability that Did2 and Snf7 are part of the same distribution (i.e. have the same copy number) is 0.42. Traditionally, only four proteins are considered to be part of ESCRT-III in yeast (Snf7, Vps2, Vps20 and Vps24). We think that this view is not justified, especially with respect to Did2, which seems to be an integral component of the ESCRT-III complex (see also results below). In the traditional model of ESCRT-III, Snf7 is thought to be by far the most abundant subunit . From this, one would expect that Snf7 also outnumbers the other ESCRT-III members on a cellular level. However, our data do not support this notion. With 8800 copies, Snf7 is present in about equal numbers in the cell as Vps2 (7700 copies) and Vps24 (6800 copies). Ist1 (5000 copies) and Mos10 (3000 copies) are present in somewhat lower amounts in the cell. The least abundant ESCRT-III protein is Vps20 with 1900 copies.
Isolation of ESCRT-III complexes by immunoprecipitation
ESCRT-III has been implicated in an array of cellular functions. It is not clear, how it performs all these functions. One idea would be that ‘ESCRT-III' is a fixed entity that is recruited to various sites be specific adapter proteins to perform its single function. Alternatively, there could be several different complexes with varying composition. In the latter case, antibodies against different ESCRT-III proteins may pull out ESCRT-III complexes with distinct composition. We, therefore, performed a comprehensive IP analysis by using antibodies against all eight ESCRT-III members as a bait. This analysis was conducted with a wild-type yeast strain and with all single ESCRT-III deletion mutants, to see how the loss of individual ESCRT-III components affects the composition of the complexes. During the analysis, it turned out that Chm7 antibodies barely precipitated any other ESCRT-III protein. Although it is a highly abundant ESCRT-III protein in yeast, it does not seem to be active under standard growth conditions (exponential phase, rich medium, ambient temperature). The condition under which Chm7 springs into action remains to be identified. Chm7 is, therefore excluded from the following analysis. The whole analysis was performed in triplicate. There was substantial variability in the co-IP efficiencies (Supplementary Table S4). Therefore, the molecule numbers presented should be taken with a grain of salt, and we should focus on the broad picture and not so much on the precise numbers. Nevertheless, despite the variability in the data set remarkably consistent co-IP patterns were observed.
The results for the wild-type strain are presented in Figure 3. The overarching conclusion from this set of experiments is that we are detecting one type of complex that contains all seven ESCRT-III proteins tested in varying amounts. The complexes pulled out by Did2, Snf7, Vps20 antibodies showed all in all a similar composition. On a whole cell basis, roughly equimolar amounts of the classical core components Snf7, Vps2 and Vps24 are present in these complexes (in total about 600–800 molecules per cell). Did2 was 2–3-fold more abundant than the other core components and is thus the main component of the precipitated complexes. Vps20 is usually present in very low numbers in the complexes (approximately 50 molecules per cell). This is consistent with the current idea that a single Vps20 molecule primes the polymerization of an Snf7 filament and with the previously published ratio of Snf7:Vps20 in the complex of about 10 . The Mos10 protein also occurs in the complexes in low numbers of about 50–100 molecules per cell. The Mos10-IP, however, looks basically like the Snf7-IP, with the notable exception that no Snf7 could be detected in the Mos10 precipitates. A possible interpretation is that Snf7 and Mos10 bind mutually exclusive to the other core components. With the Ist1 and Vps2 antibodies basically the same ratios of precipitated proteins were obtained, yet the number of precipitated proteins was ∼2–3-fold higher. Perhaps, the epitopes on the ESCRT-III complex are better accessible to these antibodies. For Snf7 in the Vps2-IP, this would mean that ∼33% of total Snf7 is present in the ESCRT-III complex. This is in line with our previous differential centrifugation experiments, where ∼40% of Snf7 was membrane associated . The Vps24-IP showed an altered pattern compared with the other IPs. Curiously, no Did2 could be detected in these complexes. The reason for this finding is not clear.
ESCRT-III composition in a wild-type strain.
Impact of single ESCRT-III deletions on complex composition
We then examined the impact of single ESCRT-III deletions on the composition of ESCRT-III. Deletion of SNF7 completely wiped out ESCRT-III complex formation. No co-IP whatsoever could be detected (Supplementary Table S4). This highlights the central importance of Snf7 for our ESCRT-III complex. Similarly, VPS2 deletion virtually abrogated ESCRT-III complex formation (Supplementary Table S4). The only significant interaction detected was a 1 : 1 complex between Snf7 and Vps20 with about 500 molecules each per cell. This places Vps2 early in the assembly pathway of ESCRT-III. The Snf7/Vps20 complex could constitute the first step in ESCRT-III formation. Curiously, the number of Vps20 molecules in the 1 : 1 complex with Snf7 is much higher than in the fully formed complex. A possible interpretation of this finding is that the main part of ESCRT-III is released from Vps20 at some point in the assembly pathway or functional cycle of ESCRT-III.
The IST1 deletion did not have much impact on ESCRT-III composition. Overall the co-IP patterns were very similar to wild-type (Figure 4) Curiously, the Vps2-IP now looked like the other IPs. In contrast with the wild-type strain, the co-IP efficiencies were not increased compared with the other IPs.
ESCRT-III composition in an IST1 deletion strain.
The deletion of DID2 prevented binding of Ist1 to the complex (Figure 5). No ESCRT-III proteins could be immunoprecipitated with Ist1 antibodies, nor was there any Ist1 detectable in the immunoprecipitates of the other antibodies. This shows that Did2 recruits Ist1 to the ESCRT-III complex, in line with earlier findings [25–28]. Another effect of the deletion was a strong increase in the amount of Vps2 present in the complex in the Snf7, Mos10 and Vps24 IPs. This suggests that Did2 counteracts Vps2, maybe by preventing further elongation of Vps2 filaments. Also, we noted a higher amount of Mos10 in the complex compared with wild-type, especially in the Vps2-IP.
ESCRT-III composition in an DID2 deletion strain.
No gross defects in ESCRT-III composition were observed in the MOS10 deletion strain (Figure 6). Similar to the situation with the Ist1 and Vps2 IPs in the wild-type strain, we observed a tendency to higher molecule numbers. In the Snf7 and Vps2 IPs, the molecule numbers were about twice as high as in wild-type. In the Did2 and Vps20 IPs only the Snf7 levels were elevated and in the Vps24-IP the Vps2 level was higher than in wild-type. As already discussed, this could reflect differences in the structure of the complex, which could impact the accessibility of the epitopes recognized by the antibodies.
ESCRT-III composition in an MOS1O deletion strain.
In the VPS24 deletion, only low numbers of ESCRT-III proteins could be precipitated by most antisera (Figure 7). The Snf7-IP, however, was a notable exception. About 1000 molecules of Vps20 per cell were precipitated along with nearly wild-type numbers of Did2 and Vps2. This points to the existence of an ‘early' complex containing Did2, Snf7, Vps2 and Vps20, which lacks Ist1 and Mos10. Similar to the situation in the VPS2 deletion, the number of Vps20 molecules is quite high in this complex. If our assumption is correct that Vps20 is released from the rest of the complex at some point in the assembly pathway, then the presence of Vps24 would be essential for this release step.
ESCRT-III composition in an VPS24 deletion strain.
There is a lack of reciprocity in the co-IP efficiencies of the different antisera. While the Snf7 antiserum shows high co-IP efficiencies with Did2, Vps2 and Vps20, the other sera precipitate much smaller numbers of these proteins. Again, we suppose that this is due to differences in the accessibility of the antigens by the antibodies. Some of the proteins may be hidden inside the complex and may thus not be accessible to the corresponding antibody.
The effect of the VPS24 deletion is clearly less severe than the effect of the VPS2 deletion. This places Vps24 downstream of Vps2 in the assembly pathway of ESCRT-III. Thus, Vps24 depends on the presence of Vps2 and not the other way around, as suggested previously .
Is there evidence for the existence of additional ESCRT-III complexes? The VPS20 deletion could provide a hint on the existence of such complexes. According to the current model of the endosomal/vacuolar complex, Vps20 is the first protein in line in the ESCRT-III assembly pathway . Thus, its deletion should completely prevent complex formation. In line with this notion, no co-IPs could be detected with the Ist1 and Mos10 antibodies in the VPS20 deletion strain (Figure 8). Nonetheless, some co-IPs were observed with the other antibodies. For instance, a substantial number of Vps2 molecules was precipitated by anti-Snf7 and low molecule numbers of other ESCRT-IIIs were also detected in the other IPs. This could point to the existence of (an)other ESCRT-III complex(es) not involved in MVB sorting.
ESCRT-III composition in an VPS20 deletion strain.
Here we examined for the first time the composition of ESCRT-III in yeast under completely native conditions. In previous studies, non-functional tagged proteins were used to analyze the composition and dynamics of ESCRT-III, which may not properly reflect the normal situation [21,22]. This is also the first comprehensive study covering the whole ESCRT-III protein family. In the following, we summarize our findings about yeast ESCRT-III. Most of the findings agree with the current way of thinking, but some are also unexpected and need to be integrated into the model about ESCRT-III.
A prominent role for Did2
The ESCRT-III family in yeast consists of eight proteins. Traditionally, only four of them (Snf7, Vps2, Vps20 and Vps24) are considered part of ESCRT-III, the others are called ESCRT-III associated or ESCRT-III like. From our data, this distinction does not appear to be justified. In our co-IP experiments, we basically pulled out a single complex containing all ESCRT-III proteins except Chm7. No matter which antibody was used as a handle, essentially the same complex was retrieved. Unexpectedly, Did2 turned out to be the most abundant subunit of the complex. The other ‘core components' Snf7, Vps2 and Vps24 were present in roughly equimolar amounts. This is in contrast with earlier findings, where Snf7 was found to be present in 3–5-fold excess over Vps24 [21,22]. But again, these results were obtained with a non-functional Snf7-GFP variant and may reflect a kinetic block in the assembly of fully functional ESCRT-III. From in vitro experiments, evidence has been presented that Snf7 and Vps2/Vps24 form side-by-side filaments , which restricts Snf7 polymerization. It is conceivable that the GFP portion attached to Snf7 could interfere with the formation of these side-by-side filaments resulting in excessive Snf7 polymerization.
The distinction between core components and associated components is based on phenotypic differences observed with deletion mutants. It has been claimed that the effect of DID2 and MOS10 deletion on MVB sorting is less severe than the effect of the deletion of core components [25–27]. In our hands, this difference does not exist. The effect of DID2 or MOS10 deletion on vacuolar degradation of the endocytic cargo protein Ste6 or on vacuolar sorting of Cps1 is indistinguishable from the effect of core component mutants . Did2 and Mos10 should, therefore, be considered bona fide members of ESCRT-III. Did2 seems to restrict assembly of Vps2, as we observed a higher amount of Vps2 in the complex in the DID2 deletion mutant. In line with previous results, we could clearly show that Did2 recruits Ist1 to the complex. Ist1 is part of the ESCRT-III complex, but its deletion does not affect MVB sorting. Thus, in this case, the term ‘ESCRT-III associated’ seems to be appropriate.
Localization of ESCRT-III and role of Snf7
The localization of ESCRT-III proteins to endosomes is well established. We were interested to see, whether ESCRT-III proteins can also be detected at other cellular compartments, which would point to other roles of ESCRT-III in yeast beyond MVB sorting. But it appears that we mainly detected the well-known endosomal/vacuolar complex. Did2, Vps2, Vps24 and Ist1 fractionated in two peaks on the sucrose gradients matching closely the endosomal and vacuolar marker proteins. Recently, it has been demonstrated that the ESCRT-system can also assemble directly on the vacuolar membrane . Thus, the two peaks that we observed most likely correspond to endosomes and the vacuole.
Surprisingly, Snf7 (and also Mos10) showed a different fractionation profile. Most of Snf7 was soluble and only a small fraction was membrane associated. This cannot be easily reconciled with the sequential assembly model of ESCRT-III. Because, according to this model, Vps20 recruits Snf7, which then recruits Vps2/Vps24 and the other ‘ESCRT-III associated proteins'. In this model, Snf7 is the central component of ESCRT-III and it is difficult to conceive how ‘the late components’ could stably exist at the membrane without Snf7. But, this appears to be the case. This suggests that the association of Vps20 and Snf7 with the other components is either transient or at least unstable. This view is supported by earlier findings. Despite the purported ratio of Snf7:Vps24 of about 3–5 in the complex [21,22], there is also evidence from another study that more Vps24 than Snf7 is associated with the membrane . Rue et al.  present at least indirect evidence that Vps2/Vps24 may associate with endosomes independently of Snf7/Vps20. Also, it was reported that Did2 and Ist1 associate with endosomes independently of Snf7 [7,25,27]. Still, Snf7 seems to be an essential component of ESCRT-III, at least at some point in the assembly of the complex, since we could not detect any co-IP in an SNF7 deletion strain. Also, one should keep in mind that most of the conclusions about endosomal localization of ESCRT-III proteins were reached with GFP fusions in a Δvps4 background. Maybe it is not always appropriate to equate ‘class E compartment’ localization in a VPS4 deletion strain with endosomal localization.
Interesting in this context is that based on sequence conservation, the two classes of ESCRT-III proteins with the different fractionation patterns belong to two subgroups. Apparently, the last common eukaryotic ancestor (LCEA) already had two copies of ESCRT-III proteins, from which these two classes, the ‘Snf7 class' (Snf7, Mos10, Vps20) and the ‘Vps2 class' (Did2, Vps2, Vps24) originated . Thus, the proteins from these two classes could perform distinct roles within ESCRT-III.
ER role of ESCRT-III
We found that Vps20 largely cofractionates with the ER. An ER localization of Vps20 is in line with previous observations, where we detected Vps20-mCherry and Vps2-mCherry in proliferated ER membrane structures called ‘karmellae' . Here, Vps20 had a fractionation profile very similar to Chm7. For Chm7, we had already suggested that it performs a role at the ER . Its mammalian homologue CHMP7 plays a role in nuclear envelope resealing during mitosis [30,31]. Evidence for a function at the nuclear envelope of the yeast counterparts has also been presented [32,33]. Whatever the function in yeast, Chm7 does not appear to be active under the conditions tested, since we were not able to detect co-IP with other ESCRT-III proteins. This is probably also true for a putative Vps20 complex at the ER, because we could not detect a significant accumulation of other ESCRT-III proteins in the ER fraction. The ER function of Vps20 remains to be elucidated.
Most of our findings are consistent with the current model of ESCRT-III, but some of the findings also call for an extension of the model. Recently, the prevailing sequential assembly model of ESCRT-III has been challenged [22,23]. It appears that ESCRT-III is far more dynamic than originally thought. All ESCRT-III components seem to be present simultaneously at the site of ILV formation and are not recruited in a stepwise manner. Filaments continuously assemble and disassemble without the assistance of Vps4. The role of Vps4 is the most controversial issue. Rather than being involved in the final disassembly of ESCRT-III, Vps4 seems to function mainly in membrane invagination and abscission. The role of ESCRT-III could be to ‘prepare' the membrane for Vps4 action. Our findings nicely fit in into this new concept of ESCRT-III.
endosomal sorting complex required for transport-III
last common eukaryotic ancestor
C.H. and L.M. performed the sucrose gradient centrifugation and the immunoprecipitations, T.B. purified the ESCRT-III proteins for immunizations and R.K. designed the study, quantified the ESCRT-III proteins and wrote the manuscript.
This work was supported by Deutsche Forschungsgemeinschaft (DFG) Grant KO-963/8-1.
We like to thank Carsten Sachse for sending us the codon optimized scVPS24 variant.
The Authors declare that there are no competing interests associated with the manuscript.
These authors contributed equally to this work.
Present address: Fachbereich Biologie, Molekulare Genetik, Universität Konstanz, 78464 Konstanz, Germany