|Home | About | Journals | Submit | Contact Us | Français|
The ESCRT-I and ESCRT-II supercomplex induces membrane buds that invaginate into the lumen of endosomes, a process central to the lysosomal degradation of ubiquitinated membrane proteins. The solution conformation of the membrane-budding ESCRT-I-II supercomplex from yeast was refined against small-angle X-ray scattering (SAXS), single-molecule Förster resonance energy transfer (smFRET), and double electron-electron resonance (DEER) spectra. These refinements yielded an ensemble of 18 ESCRT-I-II supercomplex structures that range from compact to highly extended. The crescent shapes of the ESCRT-I-II supercomplex structures provide the basis for a detailed mechanistic model, in which ESCRT-I-II stabilizes membrane buds and coordinates cargo sorting by lining the pore of the nascent bud necks. The hybrid refinement used here is general and should be applicable to other dynamic multiprotein assmeblies.
Much of contemporary cell biology focuses on the visualization of complex processes at ever-increasing spatial and temporal resolution. At the same time, contemporary structural biology is concerned with understanding ever larger, more complex, and more dynamic systems. The biogenesis of multivesicular bodies (MVBs) by the ESCRT complexes is an example of a cell process that has been the target of intensive efforts using approaches from both fields (Hurley and Hanson, 2010; Williams and Urbé, 2007). MVBs are a key intermediate in the trafficking of ubiquitinated membrane proteins to their degradation in the lysosome. MVBs form by the budding of intralumenal vesicles (ILVs) into the lumen of the endosome. The budding occurs with the opposite topology relative to the cytosol as compared to coated vesicle budding events (Hurley and Hanson, 2010). In human cells, the ESCRTs are hijacked by HIV-1 and other enveloped viruses that bud away from the cytosol (Morita and Sundquist, 2004), and the ESCRTs are also required for cytokinesis (Carlton and Martin-Serrano, 2007).
The ESCRT complexes divide the tasks of MVB biogenesis as follows. ESCRT-0 is primarily responsible for clustering ubiquitinated cargo into microdomains and delivering it to the downstream ESCRTs (Raiborg and Stenmark, 2009; Wollert and Hurley, 2010). ESCRT-I and -II (Babst et al., 2002b; Katzmann et al., 2001) drive bud formation by forming an assembly at the bud neck and thereby stabilizing it (Wollert and Hurley, 2010). ESCRT-III is responsible for membrane scission (Hanson et al., 2008; Wollert et al., 2009). ESCRT-III forms spiral or dome-shaped filamentous structures that are thought to mediate membrane scission (Fabrikant et al., 2009; Hanson et al., 2008; Lata et al., 2008). The 1:1 complex of the ESCRT-I and -II complexes (hereafter referred to as the “supercomplex”; Gill et al., 2007) connects the ESCRT-0-cargo domains to ESCRT-III and is therefore the linchpin that coordinates the entire pathway.
The structure and function of the individual domains and the core assemblies of ESCRT-I and -II are now well understood. Yeast ESCRT-I consists of one copy each of four subunits, Vps23, Vps28, Vps37, and Mvb12 (Gill et al., 2007; Kostelansky et al., 2007). Vps23 contains an ubiquitin E2 variant (UEV) domain (Katzmann et al., 2001) at its N-terminus. The Vps23 UEV domain binds to (S/T)DP motifs of ESCRT-0 and ubiquitin at nonoverlapping sites (Ren and Hurley, 2011; Teo et al., 2004b). The UEV domain is linked to the core portion of ESCRT-I by a ~50 residue proline-rich region (PRR). The ESCRT-I core is 180 Å long and consists of a stalk and a headpiece region. Vps23 and Vps37 participate in both regions, Vps28 participates in the headpiece, and Mvb12 participates in the stalk. The core has no known binding partners and is thought to function at least in part as a rigid mechanical element that controls the spatial organization of the other domains. Vps28 contains at its C-terminus a four-helix bundle (CTD) that is the primary binding site for ESCRT-II (Gill et al., 2007; Kostelansky et al., 2006; Teo et al., 2006). A 30-residue flexible linker connects the Vps28 core and CTD regions. The N-terminus of Vps37 contains a putative basic helix that is involved in contacts with acidic membrane lipids (Kostelansky et al., 2007) and is flexibly attached to the stalk portion of Vps37. All of these flexibly attached regions are dynamic in solution with respect to the ESCRT-I core (Boura et al., 2011).
Yeast ESCRT-II consists of two copies of Vps25 and one each of Vps22 and Vps36, arranged in the shape of the capital letter “Y” (Hierro et al., 2004; Teo et al., 2004a). The Vps25 subunits make up the stem and one branch of the Y, whereas the core portions of Vps22 and Vps36 make up the second branch. The Vps25 subunits each bind one copy of the early-acting ESCRT-III subunit Vps20 at the tip most distal to the rest of the Y-shaped core (Im et al., 2009). The N-termini of Vps22 and Vps36 project away from one tip of the Y. At this tip, a basic helix of Vps22 is exposed and contributes to membrane binding (Im and Hurley, 2008). The N-terminus of yeast Vps36 is complex. Most of this region is a variant PH domain referred to as a “GLUE” domain, which is responsible for binding 3-phos-phoinositides (Slagsvold et al., 2005; Teo et al., 2006). Within an internal loop in the yeast GLUE domain, two Npl4-type zinc fingers (NZF1 and NZF2) are inserted. NZF1 binds to the C-terminal domain of Vps28 of ESCRT-I (Gill et al., 2007), whereas NZF2 binds to ubiquitin (Alam et al., 2004). Hydrodynamic analysis and modeling suggest that the GLUE domain is in a relatively compact conformation with respect to the core (Im and Hurley, 2008), yet the GLUE-core linker is accessible to proteolysis in solution (Hierro et al., 2004) and no crystals have been obtained for any ESCRT-II construct containing both the core and the GLUE domain. One of the objectives of this study was to obtain an improved structural model for the full-length ESCRT-II complex in solution, taking into account its flexible elements.
ESCRT-II assembles into a 1:1 complex with ESCRT-I with a Kd value of ~30–50 nM (Gill et al., 2007) via the interface between its Vps36 NZF1 domain and the CTD of ESCRT-I Vps28. Deformation of the membrane into buds is only carried out efficiently by ESCRT-I and -II in combination (Hurley and Hanson, 2010; Wollert and Hurley, 2010). Visualization of the structure of the ESCRT-I-II supercomplex is thus a prerequisite to understanding the deformation mechanism. Its molecular weight of 244 kDa puts the supercomplex at the lower margin of feasibility for current cryoEM analysis yet renders it too large for conventional solution NMR of uniformly labeled samples. However, the most significant obstacle to structural analysis of the intact ESCRT-I-II supercomplex is not its size but rather the way in which two ordered cores and several autonomously folded domains are tethered to one another by intrinsically disordered regions. The dynamics of the “spaghetti”-like regions seem to be important for their function but render them intractable to X-ray crystallography. The combination of ordered and disordered structural elements seems to be inherent in many dynamic cellular processes.
To address this class of problems, we developed a quantitative scheme for the multimodal refinement of simultaneously ordered and disordered structures against small-angle X-ray scattering (SAXS), single-molecule Förster resonance energy transfer (smFRET), and double electron-electron resonance (DEER) data, starting from known crystal structures of the ordered portions. SAXS provides constraints on the size and shape of the supercomplex, whereas DEER and smFRET yield distibutions of distances between pairs of electron spin and between fluorophore labels, respectively, at defined positions in the polypeptide sequence. Building on an earlier multimodal structural analysis of intact yeast ESCRT-I in solution using this approach (Boura et al., 2011), here we deduce the solution conformations of yeast ESCRT-II and the ESCRT-I-II supercomplex. This structure of the ESCRT-I-II supercomplex fills in a final missing link in the structural connectivity of the pathway, allowing us to propose a start-to-finish structural mechanism for MVB biogenesis.
To determine the overall conformation of ESCRT-II and the ESCRT-I-II supercomplex in solution, SAXS data were collected (Figures 1A and 1B). Data were collected at multiple concentrations, and no concentration-dependence was noted up to the highest concentrations tested (Figures S1A and S1B available online). The Kratky plot (Figure 1C) for ESCRT-II shows a modest but discernable rise beginning at q = 0.23 Å−1, consistent with the presence of a limited amount of nonglobular structure (Figure 1E). The ESCRT-I-II supercomplex manifested substantial nonglobular structure on the basis of the Kratky plot (Figure 1D) and was highly extended based on the experimental Dmax value of 375 Å (Figure 1F).
Domain-specific conformational distributions were probed by engineering Cys pairs such that one label was in the core of ESCRT-II and the other in the Vps36 GLUE domain (Figures 2A and 2B). The GLUE domain is the primary membrane-targeting domain in ESCRT-II (Teo et al., 2006). Cys residues were introduced in the most N-terminal ordered helix of the core portion of Vps22 (Cys34), the GLUE domain (Vps36 Cys96 and Cys282), and the first winged helix (WH) domain of the core of Vps36 (Cys417). The labeled pairs were analyzed to provide three independent measurements of the position and dynamics of the GLUE domain relative to the core of ESCRT-II (Figure 3A). The FRET efficiency distribution was broad for all three FRET pairs, demonstrating that the GLUE domain is in multiple conformations with respect to the core of the ESCRT-II complex.
In order to probe the conformations of the ESCRT-I-II supercomplex, the FRET efficiency distributions of the labeled ESCRT-II samples were measured in the presence of a ~104-fold excess of unlabeled ESCRT-I (Figure 3B). The distributions resembled those seen in the absence of ESCRT-I; thus, ESCRT-I does not substantially perturb the orientation of the GLUE domain. The conformation of ESCRT-I in the presence of ESCRT-II was examined by smFRET analysis of the Vps28 Cys65-Cys151 pair, which probes the distance between the core and Vps28 CTD (Figure 3C). The ESCRT-I Vps28 CTD contains the binding site for ESCRT-II. A small decrease in the peak height of the efficiency distribution was observed in the presence of ESCRT-II at the margin of significance. Thus, ESCRT-II does not change structure perceptibly when it binds ESCRT-I. In ESCRT-I, there is a small shift to slightly longer distances between the core and Vps28-CTD. Indeed, in one of the closed conformations obtained previously for ESCRT-I alone, the ESCRT-II binding site on the Vps28 CTD is sequestered against the ESCRT-I headpiece domain (Boura et al., 2011). This type of closed conformation would be precluded in the ESCRT-I-II supercomplex.
To explore the conformation of ESCRT-I in the presence of ESCRT-II, DEER spectra were collected for three dually spin-labeled samples of yeast ESCRT-I (Figures 4A–4C). These consist of the Vps28 Cys65-Cys151 sample described previously (Figure 4A), Vps23 Cys108-Cys256, which probes the ESCRT-I UEV domain-core distance (Figure 4B), and Vps37 Cys12- Vps23 Cys 223, which probes the Vps37 N-terminal helix-core distance (Figure 4C). The UEV domain binds to ubiquitin and ESCRT-0, whereas the Vps37 N-terminal helix is involved in membrane interactions. As compared to free ESCRT-I, the Vps37 Cys12- Vps23 Cys223 pair, remote from the ESCRT-II binding site, does not change (Figure 4F). The Vps28 Cys65-Cys151 and Vps23 Cys108-Cys256 pairs show a population loss in the short distance states when bound to ESCRT-II (Figures 4D and 4E). Interspin distances of around 2 nm contribute to the fast initial decay of the measured DEER signal. The DEER decay slows in the ESCRT-I-II supercomplex for Vps28 Cys65-Cys151 and Vps23 Cys108-Cys256 spin-label pairs. These data show that the closed population of ESCRT-I is somewhat reduced when bound to ESCRT-II in the mixture of open and closed states.
We performed unbiased replica exchange Monte Carlo (REMC) simulations of the ESCRT-II complex in solution and then refined the resulting simulation ensemble using the experimental SAXS and FRET data. In Figure S2, we compare the simulation data obtained before any refinement to match the experimental data for ESCRT-II. We find that for ESCRT-II, the simulation ensemble without refinement almost fully accounts for the SAXS data over the entire q range. The raw simulation ensemble also accounts well for two out of the three smFRET efficiency histograms, but overall the comparison to the smFRET data reveals detail that is not fully accounted for by the unrefined simulations. Whereas the smFRET histograms from the simulations do cover the full range of efficiencies observed in the experiments, the weights of structures with low smFRET efficiencies are overestimated somewhat for labels (Vps36 Cys96– Vps22 Cys34) and significantly for (Vps36 Cys96 – Vps36 Cys417). In effect, the refinement of the ESCRT-II structure eliminates conformations from the initial simulation ensemble, in which these two label pairs are distant, while further improving the agreement with the SAXS data.
In the refinement, a representative solution ensemble was determined based on a minimum-ensemble approach (Boura et al., 2011; Figure S2). An ensemble of 15 structures best represents the conformations of ESCRT-II (Figure 5). The Dmax values for various structures in the ESCRT-II ensemble range from 138 to 183Å, consistent with the maximum dimension in the P(r) distribution (Figure 1E). Only in the most open conformations of ESCRT-II does the GLUE domain fail to contact the core. Roughly 90% of ESCRT-II complexes have the GLUE domain contacting the core, although the contact modes vary. The yeast ESCRT-II GLUE-core interaction thus seems to be weak and dynamic, similar to the case for the human complex (Im and Hurley, 2008). These results are in good agreement with the broad FRET efficiency histograms and past observations that the GLUE-core linker is highly susceptible to proteolysis (Hierro et al., 2004; Teo et al., 2004a).
The REMC simulations described previously were extended to the ESCRT-I-II supercomplex with the addition of a model based on the crystal structures of yeast Vps23 UEV (Teo et al., 2004a), the yeast ESCRT-I core (Kostelansky et al., 2007), yeast Vps28 CTD (Gill et al., 2007), and the Vps37 N-terminal helix and linkers. For the ESCRT-I-II supercomplex, we again find good agreement of the unrefined simulation data with the SAXS measurements (Figure S3). However, somewhat larger deviations at small q values indicate that the simulation ensemble is overweighted in fully extended structures with a large radius of gyration. The agreement with the DEER measurements is excellent for one of the label pairs and good for the other two, capturing the plateau values but being off in the initial decay by about a factor of two. For the smFRET data, the efficiency distributions calculated from the unrefined simulation ensembles agree quite well with experiment for three of the four label pairs and in all cases cover the FRET efficiency range seen in experiment (with the possible exception of the highest efficiencies for Vps36 Cys96 – Vps36 Cys417). The refinement of the ESCRT-I-II supercomplex structural ensemble is thus again achieved largely by elimination of structures inconsistent with at least some of the pair distance measurements, while improving the agreement with the SAXS data through reweighting. The ESCRT-I-II SAXS, smFRET, and DEER data were jointly best represented by an ensemble of 18 structures (Figures 6 and S4). The smFRET efficiency distributions calculated for this ensemble fully account both for the measured data (Figure 3D). For cross-validation, we omitted the Vps28 Cys65-Cys151 pair from the refinement and again found excellent agreement between calculated and measured smFRET data (Figure 3E).
The conformations of the ESCRT-I and -II complexes individually are qualitatively similar to their conformations in isolation. ESCRT-I consists of a mixture of open and closed states, whereas ESCRT-II is primarily in a relatively compact state. In the supercomplex, ~8% of the population is in a closed conformation, in which the two complexes are folded back on one another, with a Dmax value of 208 Å, similar to that of ESCRT-I alone (Figure 6, upper left). The remainder of the population is in highly extended conformations, with Dmax values ranging up to 381 Å (Figure 6), consistent with the ab initio P(r) distribution (Figure 1F). Most of the extended conformations are partially folded back on themselves, though without any inter-complex interactions to tether them shut. Most of these conformations have the shape of a crescent, which is often diagnostic of proteins that target to membranes and induce curvature (Peter et al., 2004; Shibata et al., 2009).
The main function of the ESCRT-I-II supercomplex in MVB biogenesis is to stabilize the neck of nascent membrane buds, to which it localizes (Wollert and Hurley, 2010). On the basis of membrane elasticity theory, the calculated low energy geometry of the bud neck is a catenoid (Michalet et al., 1994). In this minimal surface, the negative curvature along the pore axis exactly cancels the positive curvature of the bud rim. The radii of curvature of the supercomplex conformations are roughly consistent with the bud neck widths of ~25 nm observed in yeast MVBs (Wemmer et al., 2011). We considered the implications of positioning the crescent-shaped conformations of the supercomplex in terms of two major classes of models. In the head-to-tail ring model the convex face docks onto the negatively curved inner surface of the pore, whereas in the spoke model their concave face docks onto the positively curved profile of the rim. The head-to-tail ring model has a topology analogous to the postulated mode of inverse BAR (I-BAR) domain function in promoting negative membrane curvature (Mattila et al., 2007). The “spoke” model is so-named for its analogy to spokes of the nuclear pore (Strambio-De-Castillia et al., 2010). The spoke model positions the ESCRT-III-binding tips of the ESCRT-II complex into the center of the pore (Figures 7A and and8),8), whereas the locus of ESCRT-0 binding on ESCRT-I projects outside the pore. Unlike the head-to-tail ring model, the spoke mode is consistent with the sequential function of ESCRT complexes as deduced from yeast genetics (Babst et al., 2002a, 2002b; Katzmann et al., 2001) and with the division of labor deduced from in vitro reconstitution (Wollert and Hurley, 2010).
ESCRT-II is a Y-shaped complex that binds the Vps20 subunit of ESCRT-III at the tips of its two Vps25 subunits (Hierro et al., 2004; Im et al., 2009; Teo et al., 2004a). Biological function in MVB biogenesis requires that both tips be functional (Hierro et al., 2004; Teis et al., 2010), but it has not been clear why two sites are needed instead of one. The spoke model suggests that roughly 6–10 copies of ESCRT-II could fit in the bud neck, which would template the initiation of twice that number of short ESCRT-III filaments that drive membrane scission (Figure 7B). The number and density of the filaments would be critical to provide enough binding affinity and induce the correct geometry for membrane scission. The number and density depend directly on the number of functional copies of ESCRT-II Vps25 lining the pore. The loss of function in single-Vps25 variants of ESCRT-II could thus be explained in terms of a halving of the number of ESCRT-III filaments. This model can be considered a variant of the dome model for ESCRT-III-mediated membrane scission (Fabrikant et al., 2009). Presumably, these short filaments would arrange themselves into whorls (Figure 7B) because other arrangements would leave large gaps between the filaments. Thus, one specific prediction of this model is that under appropriate conditions ESCRT-III whorls form in cells.
The ESCRT-I-II supercomplex, and the broader ESCRT pathway, is typical of most of the cell’s regulatory machinery in the way that it combines rigid and flexible elements. Although abundant and critically important in cell physiology, these systems have been intractable by conventional approaches. They represent a blind spot in structural-level understanding of cell function. Here, we have applied a series of techniques that, taken separately, provide low-resolution global information (SAXS) or high-resolution information limited to specific sites (DEER, smFRET). All of these methods are widely used independent of one another and are sometimes used together in a comparative and qualitative way. Here, we have integrated these methods in a joint structural refinement, in the context of a system where all of the component crystal structures are known. We find that a self-consistent solution is obtainable, despite the varied nature of the experimental conditions. Strikingly, a large number of conformations are required to fit the experimental data. We attribute this to the high information density in the smFRET histograms in particular, which provide a direct readout of the full conformational distribution for each domain pair tested. In a previous analysis of ESCRT-I alone, an ensemble of two structures was sufficient to fit SAXS data alone, while an ensemble of six was able to fit SAXS and DEER data together. In this study, refinement was carried out against SAXS, DEER, and smFRET data simultaneously. As compared to refinement against SAXS and DEER data sets, larger ensembles are needed principally to represent the detail-rich smFRET histograms. Because the robustness of this approach was unknown, and because the experimental data are relatively sparse, cross-validation was a key aspect of this study. We were encouraged by the excellent agreement between the observed and computed smFRET histogram for an ESCRT-I dye pair excluded from the refinement.
The alignment of the ESCRT-I-II supercomplex in the bud neck, in the context of the spoke model, allows us to rationalize the elongated and curved shapes of the supercomplex and present a structural model for a number of otherwise unexplained observations (Figure 8; Movie S1). The mechanism of cargo transfer from the ESCRT-0 coat into the nascent bud has been mysterious (Raiborg and Stenmark, 2009; Wollert and Hurley, 2010). The discovery that the ESCRT-I-II complex is in equilibrium between open and closed states suggests a mechanism for cargo transfer. If ESCRT-I were to transition from the open to closed states following release of ESCRT-0, this would drag cargo along the membrane into the neck of the nascent bud. This model also helps explain why the budding machinery evolved as such a complicated system, with two separate complexes, both having substantial intrinsic flexibility. A simpler and more rigid machinery would not permit such a large conformational transition, and would so be unable to move cargo as far as ~300 Å along the membrane.
Since the ESCRT-I-II supercomplex is the main organizer of the neck of the nascent ILV, the structural insights allowed us to propose a model that integrates a decades’ worth of genetic, biochemical, and structural data. The model plausibly accounts for ESCRT-dependent cargo sorting, vesicle budding, and vesicle scission (Figure 8; Movie S1). The model makes testable predictions, including the existence of ESCRT-III whorls.
The refined ensemble of these flexible and partially disordered protein complexes has led to mechanistic insights into one of the most complex cellular processes to be addressed in this level of detail. The multiexperiment analysis applied in this study bypasses obstacles to conventional structural biology approaches, which are unsuited to describing larger dynamic complexes containing substantial intrinsic disorder. This approach thus helps point a way forward for mechanistic and structural dissection for the class of complex, dynamic, and flexible systems that carry out most cell processes.
All four subunits of the Cys-free yeast ESCRT-I (Vps23C133A,C344A, Vps28C101A, Vps37C123A, and Mvb12C48A/C54A/C61A) and double Cys mutants were co-expressed from the pST39 polycistronic vector (Boura et al., 2011). Vps23 was expressed as a fusion protein with an N- terminal His6 tag followed by a TEV cleavage site. The ESCRT-I protein complex was expressed in Escherichia coli BL21 cells by induction with 0.3 mM isopropyl thiogalactoside (IPTG) at an optical density of 0.8 and at 28°C overnight. The protein was affinity purified using Ni-NTA resin (Qiagen, Venlo, The Netherlands) in accordance with the manufacturer’s instructions. Immediately upon elution, the buffer was supplemented with 1 mM ethylenediaminetetraacetic acid (EDTA) and cleaved by TEV protease overnight. The complex was further purified by size-exclusion chromatography on a Superdex 200 column (GE Healthcare, Waukesha, WI, USA).
To prepare the pseudo-Cys-free yeast ESCRT-II complex, all Cys that are not Zn ligands in the NZF1 or NZF2 domains, were replaced by Ala, leading to the following constructs: Vps22C68A, C96A, C103A, C146A, C176A, Vps25C38A,C103A,C176A,C181A, and Vps36C161A,C464A,C486A,C539A. Artificial codon-optimized genes bearing these mutations were purchased from GenScript (Piscataway, NJ). The Vps25 and Vps36 subunits were cloned into the pST39 vector (Tan, 2001), and Vps22 subunits were cloned into the pRSFD vector. At the 5′ end of the VPS36 gene, a sequence encoding an N- terminal His6 tag was inserted, followed by a TEV cleavage site. Double Cys mutants were prepared using the QuikChange kit (Stratgene, La Jolla, CA, USA) or by overlapping PCR. Yeast ESCRT-II complex wild-type and mutants were expressed and purified as described (Wollert and Hurley, 2010). The ESCRT-II complex was expressed in E. coli BL21 cells in medium supplemented with 10 μM ZnCl2 upon induction with 0.2 mM IPTG at optical density 0.8 at 18°C overnight. Affinity purification on Ni-NTA resin was followed by TEV cleavage overnight. The complex was further purified by size-exclusion chromatography on a Superdex 200 column (GE Healthcare). Proteins were concentrated to ~5 mg/ml, flash frozen in liquid nitrogen, and stored at −80°C until use.
For spin labeling, the double Cys mutants were incubated overnight at 4°C with 5 mM MTSL. For fluorescent dye labeling, the ESCRT-I complex double Cys mutant was incubated overnight at 4°C with 5-fold molar excess of Alexa488 and Alexa594 maleimide. To label yeast ESCRT-II, the double Cys mutants were incubated overnight at 4°C with a 2-fold molar excess of Alexa488 and Alexa594 maleimide. Unreacted dyes were removed by size-exclusion chromatography on a Superdex 25 column (GE Healthcare). The labeling efficiency for MTSL-labeled proteins was ~100% because we could not detect any unlabeled material using mass spectroscopy. An average of ~1 Alexa dye per Cys was estimated spectroscopically with a donor to acceptor (Alexa488 to Alexa594) ratio of ~1:2. Labeled proteins were concentrated to ~30 μM for MTSL-labeled proteins and to ~1 μM for fluorescently labeled proteins, flash frozen in liquid nitrogen, and stored at −80°C until use.
To avoid possible cysteine-mediated aggregation, the no-cysteine-residue mutant of ESCRT-I complex (Mvb12C48A/C54A/C61A, Vps37C123A, Vps28C101A, and Vps23C133A,C344A) and the no-nonessential-cysteine residue ESCRT-II complex (Vps22C68A, C96A, C103A, C146A, C176A, Vps25C38A,C103A,C176A,C181A, and Vps36C161A,C464A,C486A,C539A) were used for SAXS data collection. Prior to measurement, samples were dialyzed overnight at 4°C against 50 mM Tris (pH 7.4), 150 mM NaCl, 3 mM β-mercaptoethanol, 1% glycerol, and 0.18% ascorbic acid. Data were collected for a concentration series from 5 to 0.5 mg/ml at SSRL beamline BL4-2. Data reduction and analysis were performed using the beamline software SAStool.
Pulse-EPR measurements were performed on 25–30 μl of sample loaded into quartz capillaries with 2.00 mm i.d. by 2.40 mm o.d. (Fiber Optic Center, Inc., New Bedford, MA, USA). The protein concentration of ESCRT-I mutants with two nitroxide spin labels attached was 20–50 μM in a 20 mM Tris, 100 mM NaCl, (pH 7.4), and 10% w/v glycerol buffer, and the ESCRT-I: ESCRT-II protein molar ratio was 1:1. Prior to insertion into the instrument, the sample-containing capillaries were flash frozen in a dry ice/ isopropanol bath. The DEER data were recorded at 50 K or 80 K on a Bruker Elexsys-E580 spectrometer fitted with an ER4118X-MS3 split-ring resonator (Bruker Biospin, Billerica, MA, USA) at X-band frequency. Data were acquired using a four-pulse DEER sequence (Pannier et al., 2000) with a 16 ns π/2 and two 32 ns π observe pulses separated by a 28 ns π pump pulse. The dipolar evolution times were typically 2.0–2.5 μs. The pump frequency was set to the center maximum of the nitroxide spectrum, and the observer frequency was set to the low field maximum, typically 65–70 MHz higher. The phase-corrected dipolar evolution data were processed assuming a three-dimensional background using the DeerAnalysis2009 package (Jeschke et al., 2006).
SmFRET experiments were performed using a confocal microscope system (MicroTime200, PicoQuant, Berlin, Germany). The donor fluorophore (Alexa Fluor 488) was excited by a linearly polarized dual mode (CW/pulsed) 485 nm diode laser (LDH-D-C-485, PicoQuant) in the CW mode at 45 μW through an oil-immersion objective (Plan Apo, NA 1.4, 100×, Olympus, Shinjuku, Tokyo, Japan). Donor and acceptor (Alexa Fluor 594) fluorescence, emitted from molecules freely diffusing through the illuminated volume, were collected by the same objective, divided into two channels, and focused through optical filters onto single-photon avalanche diodes (SPAD; PerkinElmer Optoelectronics SPCM-AQR-15, Waltham, MA, USa). The labeled proteins were diluted to 40 pM for ESCRT-I complex and to 160 pM for ESCRT-II complex into a 20 mM Tris buffer (pH 7.4) with 100 mM NaCl and 3 mM β-mercaptoethanol with bovine serum albumin (1 mg/ml). The measurement for ESCRT-II was performed at a higher concentration (160 pM) to compensate for molecules lost because of adsorption on the surface of the glass coverslip. However, from the comparison of the frequency of fluorescence bursts between ESCRT-I and ESCRT-II, the effective concentration of ESCRT-II was estimated to be below 40 pM. An analysis described in Gopich (2008) shows that it is very improbable to have two molecules simultaneously in the illuminated volume in both ESCRT-I and ESCRT-II experiments. In case of experiments done on the ESCRT-I/II supercomplex, the labeled protein was mixed with the unlabeled binding partner proteins at 1 μM concentration. Other experimental details can be found in Chung et al. (2009).
The apparent FRET efficiencies, Eapp = nA / (nA + nD), calculated from the number of donor (nD) and acceptor (nA) photons in 2 ms bins containing more than 90 photons were, after correcting for background and donor leakage into the acceptor channel, converted to true FRET efficiencies, E = nA / (nA + γ nD), where the correction factor γ accounts for differences in the detection efficiencies of the donor and acceptor channels and differences in the quantum yields of the donor and acceptor dyes. γ = 2 was determined from measurements of donor lifetimes in the presence (τDA) and absence t (τD) of acceptor, which yield the true FRET efficiencies from E = 1 – τDA/τD (Merchant et al., 2007). In the FRET efficiency histograms, the number of counts at each value of the FRET efficiency represents the number of 2 ms bins containing more than 90 photons.
In the ensemble refinement, the number, N, of structures in the ensemble and the weights, wk, assigned to individual structures were varied. To avoid introducing regularization-dependent artifacts into the refinement procedure, the simulation structures were refined directly against the measured data. The ensemble-averaged SAXS intensity was calculated from model coordinates exactly as in the original EROS method (Różycki et al., 2011). The background-corrected, ensemble-averaged DEER dipolar evolution functions were calculated exactly as in our recent study on ESCRT-I (Boura et al., 2011) using the rotamer library of spin-labeled Cys (Polyhach et al., 2011). Also the single-molecule FRET efficiency histograms were calculated as previously (Boura et al., 2011). Briefly, possible locations of the fluorescence dyes were sampled according to a distance distribution that was previously obtained from molecular simulations (Merchant et al., 2007). Dye positions that overlapped with the proteins or with the other dye were excluded from the analysis, with a distance cutoff of 6 Å. All orientations of the donor and acceptor transition dipoles were assumed to be equally probable. The mean FRET efficiency for a dye pair attached to residues i and j in structure k was calculated as
where R0 = 54 Å is the Förster radius, and the average is performed over all positions α and β of the dyes attached to sites i and j, respectively. Using these mean efficiencies, Ek,(i,j), and the weights, wk, assigned to structures in the course of refinement, the FRET efficiency histograms were constructed for all dye pairs (i,j) by the recoloring method (Gopich and Szabo, 2007) according to the distributions in the number of photons in the individual bursts measured in experiment. The deviations between the measured and calculated FRET data for any residue pair (i,j) were quantified by
where Nb is the number of bins in the FRET efficiency histograms, Em is the FRET efficiency that corresponds to bin m, Nobs(Em) is the number of efficiency counts assigned to bin m in experiment, N(Em) is the theoretical number of counts in bin m, and σ2(Em) = Nobs(Em) is the statistical error of counts assuming the Poisson distribution of photon bursts. In this approach, we correctly account for shot noise and compare the computed and measured FRET efficiency histograms directly. In these calculations, dynamics among the clusters of conformations was not included in the analysis; however, given the large number of clusters, the shapes of the histograms would not be affected greatly. Moreover, in the case of ESCRT I, the bin-size dependence of the FRET efficiency histograms indicated the absence of interconversion of clusters on the time scale of the bin size (i.e., 500 μs).
The minimal ensemble refinement procedure (Boura et al., 2011) was modified for simultaneous fitting of SAXS, DEER, and FRET data. Deviations from SAXS and DEER data were quantified by χ2SAXS and χ2DEER as in (Boura et al., 2011). The free energy function,
measures the deviation from the different experiments, giving equal weight to each measurement. The last term in G is a penalty for the size, N, of the ensemble that is controlled by a parameter μ. To obtain a minimal ensemble of structures that capture the different experiments, G was minimized numerically with respect to the number, N, of structures and their relative weights, wk. Starting from a small value, μ was increased until all experiments could be fit simultaneously within the respective statistical uncertainties, χ2 ≈ 1.
We thank E. Tyler for generating Figure 8 and Movie S1. SAXS data were collected at the Stanford Synchrotron Radiation Laboratory, a national user facility operated by Stanford University on behalf of the U.S. Department of Energy, Office of Basic Energy Sciences. The SSRL Structural Molecular Biology Program is supported by the Department of Energy, Office of Biological and Environmental Research, and by the National Institutes of Health (NIH), National Center for Research Resources, Biomedical Technology Program. B.R. was supported by a Marie Curie International Outgoing Fellowship within the 7th European Community Framework Programme. E.B. was supported by an Intramural AIDS Research Fellowship. This work was supported by the Intramural Program of the NIH, National Institute of Diabetes and Digestive and Kidney Diseases (G.H., J.H.H., and W.A.E.), the Intramural AIDS-Targeted Antiviral Program of the Office of the Director, NIH (J.H.H.), and an NIH grant (GM072694 to D.S.C.).
Supplemental Information includes four figures and one movie and can be found with this article online at doi:10.1016/j.str.2012.03.008.