|Home | About | Journals | Submit | Contact Us | Français|
Functioning as key players in cellular regulation of membrane curvature, BAR-domain proteins bend bilayers and recruit interaction partners through poorly understood mechanisms. Using electron cryomicroscopy, we present reconstructions of full-length endophilin and its N-terminal N-BAR domain in their membrane-bound state. Endophilin lattices expose large areas of membrane surface, and are held together by promiscuous interactions between endophilin's amphipathic N-terminal helices. Coarse-grained molecular dynamics simulations reveal that endophilin lattices are highly dynamic, and that the N-terminal helices are required for formation of a stable and regular scaffold. Furthermore, endophilin accommodates different curvatures through a quantized addition or removal of endophilin dimers, which in some cases causes dimerization of endophilin's SH3 domains, suggesting that the spatial presentation of SH3-domains rather than affinity governs the recruitment of downstream interaction partners.
The cell membrane is a dynamic barrier whose shape is constantly remodeled with high spatial and temporal accuracy during essential cellular processes ranging from cell motility and signaling, to maintenance and generation of organelles (Gallop and McMahon, 2005; Hurley et al., 2010). While critical for normal cell function, the mechanisms by which cells control membrane remodeling are only beginning to come into focus (Dawson et al., 2006). At the molecular level, proteins of the Bin/Amphiphysin/Rvs (BAR) domain superfamily have emerged as major players in membrane remodeling (Farsad et al., 2001; Frost et al., 2009). Divided into four distinctive subfamilies, N-terminal BAR (NBAR), extended-FCH-BAR (F-BAR), inverse BAR (I-BAR) (Ren et al., 2006), and Pinkbar proteins (Pykalainen et al., 2011), the majority of BAR-domain proteins are scaffolding proteins that combine a membrane-remodeling BAR-domain with additional protein:protein interaction or catalytic domains, which are responsible for the biological coupling of specific processes to distinct membrane curvature states. At the structural level, BAR-domains are dimers of an antiparallel helix bundle (Peter et al., 2004; Shimada et al., 2007; Weissenhorn, 2005) that display various degrees of intrinsic curvature. The simple design of these modules allows cells to generate a large range of different curvature states, including membrane invaginations, like tubules, or extrusions like filopodia (Guerrier et al., 2009; Lee et al., 2002). At the mechanistic level, high-resolution structures and spectroscopic studies have inspired two major models for how BAR-domains accomplish changes in membrane curvature. One model holds that membranes are remodeled through a pure scaffolding mechanism in which a “banana-shaped” BAR-domain binds the bilayer through electrostatic interactions and imposes its intrinsic curvature on the substrate bilayer. In the second model, curvature sensing and generation are thought to critically depend on the membrane insertion of amphipathic wedges such as the N-terminal helix 0 (H0) that are found in N-BAR proteins such as endophilin and amphiphysin (Chernomordik and Kozlov, 2003; Farsad et al., 2001; Johannes and Mayor, 2010). There is good experimental evidence in support of both models, yet mechanistic details are largely unknown, in part because direct visualization of BAR-domain proteins in membrane bound states has proven to be challenging.
The structure of membrane-bound CIP4 F-BAR domain demonstrated that bending of the membrane can be realized solely through scaffolding (Frost et al., 2008). However, the same study also revealed a second physiologically relevant binding mode in which the F-BAR domains engage planar bilayers through an alternate, flat binding interface. This suggested that the repertoire and mechanistic complexity of BAR:membrane interactions likely is larger than was previously anticipated. In contrast, how insertion of amphipathic sequences into the bilayer promotes curvature sensing and generation remains less clear. A recent study showed that N-BAR domains of endophilin are able to form ordered surface lattices on membranes. However, the study focused on tubular structures with very narrow diameter. Individual N-BAR domain dimers were not resolved, and details about how the N-BAR domains interact with the membrane were not provided (Mizuno et al., 2010). Here we present reconstructions of full-length endophilin and its N-terminal N-BAR domain bound on tubules whose diameters are in a range that is relevant for action by downstream interaction partners, such as the GTPase dynamin (Chappie et al., 2011; Faelber et al., 2011; Roux et al., 2010; Sundborger et al., 2011). Our key findings are that at their structural level N-BAR lattices are fundamentally different from F-BAR scaffolds, that highly promiscuous and generic interactions between N-terminal H0 helices are essential for the formation of ordered scaffold structures, and that the geometry of the N-BAR endophilin lattice in some cases results in dimerization of the SH3-domain. Combined with our observation that the concentration of SH3-domains close to the scaffold surface is independent of curvature, our study suggests that recruitment of downstream partners is driven by steric selection rather than thermodynamic affinity.
To directly visualize the design principles of membrane-bound N-BAR lattices, we used electron cryomicroscopic imaging, maximum likelihood classification and an iterative helical real space reconstruction approach (Egelman, 2007)(Sorzano et al., 2004) to determine classaverages and the 3D-structures of tubules decorated with either the endophilin N-BAR domain or full-length endophilin (Figure 1, Figure S1, Supplemental Table 1). Starting from data sets of ~75,000 overlapping segments for endophilin, and ~35,000 segments for tubes coated with the endophilin N-BAR domain, we were able to determine structures of tubules of different diameters for each sample (see supplemental methods). In addition, we also obtained a classaverage for another N-BAR protein, amphiphysin (~2000 overlapping segments, Figure S1A) to test whether the smaller number of amphipathic wedges in amphiphysin (two versus four in endophilin) resulted in a change in overall scaffold design.
Classaverages obtained from our data sets showed a distinctive pattern where dark and light regions alternated along the vertical axis of the membrane tubule with a period of ~50Å (Figure 1, Figures S1A, B). Moreover, the pattern was highly conserved regardless of differences in the number of amphipathic wedges (endophilin vs amphiphysin), the presence/absence of domains other than the NBAR domain (full length endophilin/amphiphysin vs endophilin N-BAR domain), or different curvatures of the underlying membrane (Figure S1C). This suggested that N-BAR lattices were highly pliable, which was further supported by the observation that tube diameters readily and frequently changed even along a single tube.
To better understand the lattice design, we generated 3D-reconstructions of full-length endophilin (Figure 1) and endophilin's N-BAR domain (Figure S1E). In the reconstructions, individual N-BAR-domains were easily recognizable (Figure 1). Inspection of the structures revealed that tip-to-tip interactions between consecutive N-BAR dimers did not seem to play a major role in stabilizing the lattice and also seemed to switch between two modes. In the 25nm and 32nm tubes, the N-BAR domains were oriented perpendicular to the tube's long axis and appeared to have little to no direct tip-to-tip contact, while in the 28nm tubes, direct tip-to-tip interactions seemed to link consecutive dimers that were inclined by ~10° with respect to the tube's long axis. Cross-sections of these tubes revealed that the distinct orientations coincided with lattices that differed by an integral number of N-BAR dimers around the circumference of the tube (not shown). Moreover, N-BAR dimers in adjacent rows of the lattice did not show direct lateral interactions between the coiled-coil core domains of N-BAR dimers, which had been identified as a defining structural feature in scaffolds formed by the F-BAR protein CIP4 (Frost et al., 2008). This marked difference in lattice design exposed large areas of continuous membrane surface between adjacent lattice ridges (e.g. ~350 nm2 from one notch to the next in 28 nm tubules) that, independent of curvature, were spaced ~50Å apart from each other.
The absence of lateral interactions between the N-BAR cores, and the apparently weak or absent tip-to-tip interactions between consecutive N-BAR dimers suggested that interactions between the amphipathic helices of endophilin played an important role in lattice formation and stability. To better understand the role of these structural elements, we sought to identify their positioning in our reconstructions. For this purpose, reconstructions were thresholded to tightly fit the backbone of endophilin's high resolution N-BAR crystal structure to emphasize the molecular envelope of the protein components (pdb: 1ZWW). At this contouring level, additional densities bridged adjacent windings of the surface lattice and likely represented contributions from the amphipathic helices (Figure 1B). Based on the envelope, the reconstructions suggested that H0-helices from adjacent rows of the N-BAR lattice interacted in an antiparallel fashion, placing the helix pair roughly parallel to both the membrane surface and the longaxis of the tube (Figure 1B). In contrast to the H0-helices, positioning of the insert helices was less certain because the volume elements accounting for this helix were less well defined than the envelope for enclosing H0 (Figure 1B). Whether this poorer definition was due to flexible linkers that connected the insert helix to the BAR-domain core, or was caused by a spread of orientations of the insert helix could not be resolved at the resolution of our reconstructions. However, the molecular envelopes suggested that the insert helix was tilted towards the bilayer core, and did not make direct contacts with insert helices from the adjacent N-BAR domain dimer (Figure 1B).
The conclusion that interactions between the H0-helices from neighboring N-BAR dimers were essential to lattice formation and stability was independently established by coarse-grained molecular dynamics simulations (Figure 2A, see supplemental methods and Figure S2 for details). The simulations revealed that the oligomer structure was retained during the course of simulation for the NBAR system (Figure 2A, “H0+BAR”), while the oligomer rapidly became scrambled for the system where H0 had been deleted (Figure 2A, “BAR”). Notably, even in the presence of H0, the lattices had order parameters of only ~0.8 (Figure S2B). This further emphasized that endophilin lattices were very dynamic, and explained why segment sorting and classification were necessary to obtain reliable classaverages for reconstruction of the volumes.
To further test the importance of H0:H0 interactions for the formation of the lattices, mutant endophilins with serial deletions lacking 3, 6, 9, 12, 16 or 20 residues respectively along H0 were tested for their ability to tubulate liposomes. Of these mutants the Δ2-17 and Δ2-21 mutants did not tubulate liposomes (not shown). However, the shorter partial deletion mutants were still functional in this assay, but displayed a lower efficiency than wild-type protein if the amounts of endophilin were lowered to the smallest amount necessary to yield robust tubulation (not shown). Moreover, cryoEM data collected from tubes formed by the Δ2-10 and Δ2-13 mutants lacked the characteristic striation pattern that was observed for tubes formed by wild-type endophilin (Figure 2C). This suggested a higher degree of lattice disorder in the tubes formed by the mutant endophilins, and further supported the idea that antiparallel interactions between H0-helices were necessary for the formation of a well-defined lattice structure.
To further characterize the interactions between H0-helices, we generated twelve single Cys-mutants along H0 in a cysless endophilin background and determined their ability to spontaneously form crosslinked dimers in their membrane bound state (Figure 3A, Figure S3). Surprisingly, all mutants were able to crosslink and the efficiency increased towards the C-terminal end of H0, with the exception of the Q9C-mutant, which for unknown reasons crosslinked less well than other mutants in this region of H0. At the same time, all mutants retained their ability to tubulate liposomes (Figure S3D for examples). This indicated that the mutant endophilins were functional, and that the observed crosslinks occurred in the context of a scaffold. The latter implied that the helices were able to interact in many different registers, including pairings on either side of the helix. To acknowledge this observation, H0-helices are represented by generic cylinders in Figure 1B to indicate that the observed densities represent an ensemble average over all possible lateral alignments of H0-pairs. Consistent with the crosslinking experiments, all atom molecular dynamics calculations revealed a very smooth energy landscape for H0:H0 interactions, with exception of two configurations that were slightly more stable, and a more general trend that small overlaps towards the N-terminal ends of the helices were energetically slightly less favorable than more extensive overlaps (Figure 2B).
The hypothesis that lattice formation depended on lateral interactions between H0-helices implicitly made the prediction that both tubulation and crosslinking of the various mutants should be sensitive to the presence of chemically synthesized H0-peptide. As shown in Figure 3B, this condition was met as two different H0-peptides (wild-type and a T14D-mutant peptide that mimicked a naturally phosphorylated H0 variant (Kaneko et al., 2005)) significantly reduced crosslinking for all but the most N-terminal mutants of H0. The latter observation not only served as an internal negative control, but was also consistent with the slightly less favorable interactions that were observed in the molecular dynamics simulations (Figure 2B). In further agreement with the crosslinking results, tubulation was reduced in the presence of sub-milimolar concentrations of the peptide, and almost absent at nominal peptide concentrations of 2mM that were used to carry out the competitive crosslinking experiments (Figure 3B). Moreover, wild-type endophilin H0 peptide also blocked tubulation of another NBAR protein, amphiphysin 2, with the same concentration dependence, but had a much less inhibitory effect on tubulation by the F-BAR domain of FBP17 that does not employ amphipathic wedges to induce membrane curvature (Figure 3B, Figure S3E for full titration curves of N-BAR proteins). This suggested that the presence of the free peptide in the membrane did not interfere with membrane bending per se, but acted by disrupting the cooperative assembly of a coherent lattice that required a permissive spatial positioning of H0-helices. A rough estimate revealed that in fully assembled scaffolds, H0 is present at a concentration of ~14mM within the leaflet that accommodates the wedging element. With this in mind, the inhibition we observed at a nominal peptide concentration of 2mM was deemed significant. At this concentration, the peptide alone did not cause tubulation or disruption of the bilayer (not shown). However, our reconstruction revealed that insertion of the much higher concentrations of amphiphatic wedges in the actual scaffolds caused significant stress on the bilayer. This stress manifested itself in the absence of density for the bilayer leaflet closest to the N-BAR domain if the reconstructions were thresholded to emphasize the protein component of the ensemble (Figure 1B). It should be pointed out, that the missing leaflet was visible at a less significant thresholding level. However, at this level the definition of the BAR-domains was lowered to the point where individual dimers were no longer visible. Thus, the reconstructions shown in Figure 1 were the best overall compromise how to represent the structures.
As is the case in many BAR-domain proteins, endophilin employs src3-homology domains for the recruitment of its two downstream interaction partners, dynamin and synaptojanin (Chang-Ileto et al.; Llobet et al.; Ringstad et al., 1997; Takei et al., 1999). How each of these partners is selectively recruited is unknown and prompted us to compare reconstructions of membrane-bound full-length endophilin with reconstructions of the membrane-bound N-BAR domain alone to determine whether the SH3-domains adopted well-defined spatial patterns above the scaffold surface. Overlay of the reconstructions of the 25nm tubes for both proteins failed to show any significant differences. This indicated that the SH3-domains were highly mobile above the scaffold surface, which caused their density to be averaged out during the reconstruction. In contrast, overlay of the 28nm tubes revealed additional densities in the case of the full-length endophilin lattice (Figure 4, Figure S1E for data of N-BAR only tubes). The additional densities were large enough to accommodate the backbone of the high-resolution structure of an endophilin SH3-domain dimer (pdb: 3IQL, (Trempe et al., 2009)). Interestingly, the putative SH3-domain dimers were located above the region where the tips of two consecutive endophilin dimers come close to each other and thus provided additional lattice contacts in this case (Figure 4). Based on this model, placement of Cys-residues into the putative dimerization interface should allow the spontaneous formation of disulfide bridges. To test this prediction, we generated a Thr320Cys mutant in an otherwise cysless background. Consistent with our model, this mutant efficiently formed crosslinks when bound to bilayers, which set it apart from wild-type endophilin A1 in which the nearby Cys294 in the SH3-domain failed to form significant amounts of disulfide linked dimers. The fact that the crosslinking was less than 100% was likely due to the fact that only a fraction of the tubes in any given population had the correct diameter that would allow dimerization to occur (Figure S1D). Nevertheless, the pronounced difference in the behavior of the T320C mutant and wild-type endophilin A1 was consistent with the idea that SH3-dimers can form above the surface of some, but not all endophilin lattices.
Based on their structure, BAR-domains are very similar to each other overall (Frost et al., 2009). Once assembled into scaffolds, however, lattices of F-BAR (Frost et al., 2008) and N-BAR domains are remarkably different. While extensive lateral contacts between the coiled-coil regions of the F-BAR protein CIP4 are a defining hallmark of the coats it forms on membranes, such interactions were entirely absent from any of the N-BAR lattices that we reconstructed. Similarly, tip-to-tip interactions between consecutive F-BAR dimers made a significant contribution to controlling the curvature state of the bilayer (Frost et al., 2008), yet did not seem to make notable contributions to the formation of N-BAR lattices. These fundamental differences in lattices design both led to the same conclusion that antiparallel interactions between H0-helices from N-BAR domains in adjacent rows of the lattice are essential for scaffold assembly and the stabilization of membrane curvature because the limited or absent tip-to-tip interactions between consecutive N-BAR domain dimers would be too weak on their own to maintain the highly curved membrane tubules (Lyman et al., 2010). As a direct consequence, the lateral association of H0-helices along the tube axis and between adjacent ridges of the lattice exposed large areas of membrane surface, which may be critical to allow access for proteins like the GTPase dynamin or the inositol 5-phosphatase synaptojanin that in endocytosis act downstream of N-BAR proteins (Ringstad et al., 1997; Sundborger et al., 2011; Takei et al., 1999) (Figure 5). Also very different from the F-BAR domain of CIP4, cross-sections of endophilin-covered tubes revealed that they differed by an integral number of dimers around the circumference of the tube (not shown). This suggested that the N-BAR domains accommodated different curvatures primarily through a “quantized” addition/removal of dimers, which contrasts with CIP4 scaffolds, in which different curvatures are accommodated by a combination of lattice re-orientation and non-integer changes in F-BAR dimers around the circumference of the tubule (Frost et al., 2008).
One of the most surprising and counterintuitive findings of our work was that interactions between the H0-helices appeared to be extremely pliable and non-specific. Fortifying the crosslinking data, the observation that the endophilin H0-peptide was able to poison lattice formation by another N-BAR protein, amphiphysin 2 was, perhaps, the most striking illustration of the nonspecific nature of H0 interactions because the sequences of the H0 helices from both proteins do not share high homology. The notion that these types of amphipathic wedges provide a generic, low-affinity amphiphilic hook seemed at odds with their indisputable importance for establishing ordered arrays. Specifically, our findings raised the question how such generic interactions allow for proper biological function. In part, an answer may arise from the observation that N-BAR decorated tubes are very flexible in vitro and readily accommodate changes in diameter even within a given lattice. If flexibility were a functional requirement, then pliable interactions between the main lattice contact points would be necessary, and could not be accomplished with a scaffold design as seen in FBAR lattices of CIP4, where extensive lateral interactions between F-BAR coiled-coil domains create a very rigid cast (Frost et al., 2008). Similarly, a low-affinity interaction facilitates lattice disassembly, which may be necessary after a given process has run to completion. Contemplating possible reasons other than lattice dynamics, a highly specific recognition between amphipathic wedges may also not be necessary since the concentration of unrelated segments is likely to be small in the vicinity of the site where membrane remodeling occurs. While without doubt a final explanation for the perplexing lack of specificity will still need to be found, it seems that part of an answer may turn out to lie in a cell's need to mount a versatile, yet robust response in cases that require assembly of N-BAR lattices.
While the resolution of our reconstructions did not allow the assignment of a clear position for endophilin's insert helix, a reasonable fit for this helix within the molecular envelope would require it to be tilted towards the membrane core (Figure 1B). Interestingly, the direction of tilt would be the opposite of what was observed in the only crystal structure in which the insert helix was resolved (pdb: 2Z0V), and also would contrast with recent spectroscopic evidence showing that a membrane bound insert helix bound at an angle that made it slope away from the hydrophobic core (Jao et al., 2010). Notably, these latter studies had to be performed on liposomes that were too small to support tubulation because it would hinder interpretation of the spectra. Trying to reconcile these disparate observations suggests that in order to efficiently induce curvature at a macroscopic scale, the bilayer environment must be permissive for lateral H0:H0 interactions to occur and to allow the insert helix to undergo a conformational change that would cause it to tip towards the hydrophobic core of the bilayer. How these two components individually contribute to curvature sensing, stabilization and generation remains an unanswered question.
By nature, most BAR-domain proteins are scaffolding components that drive curvature-dependent biological processes through the recruitment of other proteins. Selection of interaction partners in many cases depends on SH3-domains, which often are able to interact with multiple different proteins (Ferguson et al., 2009; Solomaha et al., 2005). This promiscuity raises the important question how BAR-dependent scaffolds accomplish specificity in recruiting only those components that are required to drive any given biological process. The question becomes more pressing in light of the observation that, based on our reconstructions, endophilin coats generate effective SH3-domain concentrations of 3-5mM within a 100Å annulus above the scaffold surface. Notably, this concentration is independent of curvature as the same number is obtained for tubes of different width. Even more unsettling, calculations for the scaffolds formed by the F-BAR domain of CIP4 (Frost et al., 2008) yield the same result.
Considering that proline-rich target peptides typically bind SH3-domains with nanomolar to low micromolar affinities (Demers and Mittermaier, 2009), our reconstructions reveal that selection of downstream interaction partners cannot be based on binding affinities because the high local concentration of SH3-domains would cause non-selective recruitment, even of low affinity binding partners, to any type of BAR-domain lattice, and in a curvature independent manner. Our finding that SH3-domains dimerize above the surface of some but not all lattices offers a potential solution to this vexing problem since it suggests that unique spatial presentation signatures rather than binding affinities identify the type of BAR-scaffold and communicate the curvature state of an underlying membrane to potential interaction partners. For instance, the 28nm tubes present SH3-domains as dimers. This spatial arrangement may be particularly useful for recruitment of the GTPase dynamin that has two spatially close proline rich segments within its proline rich domain. Incidentally, a diameter of 28nm of the endophilin-covered tube is also in the middle of the range of diameters that have been implicated with various mechanistic models for dynamin-dependent membrane fission (Faelber et al.). Consequently, the 28nm tubes not only may recruit dynamin more readily than tubes of different diameters, but also may aid membrane scission to occur by matching the curvature state of the target membrane with what is required for dynamin action.
cDNA fragments encoding rat endophilin A1, rat endophilin A1 N-BAR domain (1-247), rat amphiphysin 1 and amphiphysin 2 were subcloned into pGEX6P-1 (GE Healthcare, Piscataway, NJ) via polymerase chain reaction. Cysteine Mutants were generated by Site Directed Mutagenesis (QuikChange, StrataGene, Santa Clara). cDNA for Rat endophilin A1, rat amphiphysin 1 and rat amphiphysin 2 were kindly provided from P. DeCamilli, Yale University. Fusion proteins were bacterially expressed and purified first on a GST-glutathione affinity column (GE Healthcare, Piscataway, NJ). The GST tag was cleaved by PreScission protease, followed by gel filtration chromatography (See supplement information for composition). Endophilin H0-truncation mutants were cloned and generated in pET24 (Novagen/EMD Chemicals, Gibbstown, NJ) to improve yield. H0-Truncation mutants were generated by Site Directed Mutagenesis (QuikChange, StrataGene, Santa Clara). The fusion proteins were purified on a Talon-metal-affinity column (Clonetech, Mountain View, CA). Aliquots of 4 mg/mL (endophilin NBAR) protein, 10 mg/mL (endophilin), 25 mg/ml (endophilin H0-truncation mutants) and 2 mg/mL (amphiphysin 1 and amphiphysin 2) were stored at -80º C.
For all experiments, synthetic lipids were used (Avanti, Alabaster, AL). For the imaged sample we prepared lipids with the following composition (w/w): 50% DOPS, 45% DOPE, 5% Cholesterol (endophilin) and 75% DOPS, 25% DOPE, 5% Cholesterol (amphiphysin, endophilin N-BAR). These mixtures were dried under a stream of dry argon with gentle vortexing in glass vials, dissolved in absolute hexane, dried with argon again, and desiccated under high-vacuum for 1 h. Lipids were then hydrated with buffer A (50 mM K-Aspartate,10 mM Tris/HCl, 1 mM EGTA, pH 7.5), sonicated and used immediately or stored in aliquots at -80° C. The in vitro tubulation was performed with liposomes (0.1–0.25 mg/mL) equilibrated at room temperature before adding the protein at a lipid/protein ratio (w/w) of 1.4:1 (endophilin N-BAR) or 1:1 (amphiphysin 1, N-BAR).
Endophilin cysteine mutants were reduced in buffer A containing 1mM tris(2-carboxyethyl)phosphine) (TCEP) before adding liposomes. After an additional incubation period, TCEP was removed by repeated centrifugation and resuspension of the pellet with TCEP-free and EGTA-free buffer A. The resuspended pellet was incubated in buffer A containing 1mM Cu-OPhenanthroline for 1 h and the reaction was stopped with 1mM EDTA. The pellet and supernatant were analyzed by non-reducing SDS-PAGE and stained with Coomassie Blue. An aliquot of the pellet of all tested mutants was assessed by Electron microscopy. For competition experiments with H0-peptides, nominal peptide concentrations of 2mM were used.
The tubulation reaction was screened using 1% uranyl acetate-stained samples and a Tecnai 12 microscope (Philips, FEI, Eindhoven, The Netherlands) operating at 120 kV. Images of unstained samples were acquired at a sample temperature of -170° C on a Tecnai F20 Twin transmission electron microscope operating at 120 kV, and recorded with a Tietz F415 4k × 4k pixel CCD camera using the Leginon data collection software (Suloway et al., 2005) at nominal magnifications of 29kx, and defocus values of –1.5μm to –2μm. Electron cryo micrographs used in the figures were contrast enhanced to increase visibility of fine molecular features. Detailed methods can be found supplemental experimental procedures.
Fourier–Bessel reconstruction proved to be impossible due to the lack of high-resolution features in the power spectrum from these tubules. Moreover, variations in the diameter over the length of even a single tubule precluded reciprocal space averaging. We therefore decided to sort overlapping segments of the tubules into defined classes using maximum likelihood methods (XMIPP; (Sorzano et al., 2004) before initiating reconstruction by an Iterative Helical Real Space Reconstruction (IHRSR) single-particle algorithm as implemented in SPIDER (Egelman, 2007; Frank et al., 1996). The resolutions of our reconstructions (Table S1) were calculated with the program RMEASURE (Sousa and Grigorieff, 2007). See supplement procedures for a detailed description of the reconstruction strategy.
We thank Pietro De Camilli for critical discussions and feedback on the initial manuscript. The majority of data for this work were collected at the National Resource for Automated Molecular Microscopy, which is supported by the National Institutes of Health though the National Center for Research Resources’ P41 program (RR017573). We are also indebted to David Morgan for letting us use the cryoEM facility at Indiana University. This work was funded by Deutsche Forschungsgemeinschaft (C.M.) and the Cancer Research Institute (C.M.) as well as PHS grants DA24101 (V.M.U), GM094479 (V.M.U.), GM063796 (GAV) from the National Institutes of Health.
Publisher's Disclaimer: This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final citable form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.
Volume files for the reconstructions have been deposited at EMDB under the following accession numbers: N-BAR-only (25nm) EMD-10394, N-BAR-only (28nm) EMD-24978, Endophilin (25nm) EMD-24979, Endophilin (28nm) EMD-24980, Endophilin (32nm) EMD-24981