Three-dimensional structure of the anthrax toxin pore inserted into lipid nanodiscs and lipid vesicles

Abstract

A major goal in understanding the pathogenesis of the anthrax bacillus is to determine how the protective antigen (PA) pore mediates translocation of the enzymatic components of anthrax toxin across membranes. To obtain structural insights into this mechanism, we constructed PA-pore membrane complexes and visualized them by using negative-stain electron microscopy. Two populations of PA pores were visualized in membranes, vesicle-inserted and nanodisc-inserted, allowing us to reconstruct two virtually identical PA-pore structures at 22-Å resolution. Reconstruction of a domain 4-truncated PA pore inserted into nanodiscs showed that this domain does not significantly influence pore structure. Normal mode flexible fitting of the x-ray crystallographic coordinates of the PA prepore indicated that a prominent flange observed within the pore lumen is formed by the convergence of mobile loops carrying Phe427, a residue known to catalyze protein translocation. Our results have identified the location of a crucial functional element of the PA pore and documented the value of combining nanodisc technology with electron microscopy to examine the structures of membrane-interactive proteins.

Determining the structures of the membrane-inserting components of protein toxins that act within cells is critical for understanding how these toxins translocate their enzymatic cargoes across membranes. The three proteins that comprise anthrax toxin combine to form two binary toxins that contribute to the lethality accompanying infections by Bacillus anthracis. Protective antigen (PA), after binding to a cell-surface receptor and being proteolytically activated, oligomerizes to form a heptameric prepore. The prepore binds the enzymatic lethal factor and/or edema factor proteins, and, after the complexes are trafficked to the endosome, the acidic environment induces the prepore to form a pore in the membrane and to translocate the enzymatic moieties to the cytosol (1). A primary goal for understanding this process is to determine the mechanism by which the PA pore functions as a pH gradient-driven translocation machine to transport the enzymatic toxin components across membranes. Even though the x-ray crystallographic structure of the heptameric PA prepore was solved in 1997 (2), a structure of the PA pore has eluded researchers because of aggregation difficulties.

In a prior study we used the chaperonin GroEL as a molecular scaffold to avoid this aggregation problem. We showed that the PA pore formed stable complexes with GroEL and solved a structure of a functional PA pore at ∼23–25 Å resolution by using negative-stain EM (3). To gain a better structural understanding of this protein translocation machinery, it is imperative that one obtain structures of the pore inserted into membrane bilayers, and we chose a model bilayer system, the apolipoprotein A1-derived lipid nanodisc, for further studies (4). Unlike other systems such as liposomes or detergent micelles, nanodiscs are relatively monodisperse (∼100 Å in diameter) and stable. Furthermore, protein–nanodisc complexes have proven to be easy to construct, are readily purified and concentrated, and are applicable to single particle structural analysis (5). We have now succeeded in assembling PA-pore lipid-nanodisc and PA-pore:vesicle complexes and have reconstructed the nanodisc/vesicle-inserted PA pore by using EM of negatively stained specimens.

Mutational analysis in recent years revealed a mobile luminal loop of the PA prepore containing a Phe residue that is crucial for protein translocation (Phe427) (6,7). There is evidence that the seven Phe427 residues converge as the pore is formed, generating a luminal structure, called the Phe clamp, that catalyzes translocation of the unfolded lethal factor or edema factor from the acidified endosome compartment to the cytosol (1). Our previous structural study with the GroEL-stabilized pore revealed a flange within the lumen (3). This flange appears in all reconstructions performed thus far and is clearly visible in both the PA-pore reconstructions within lipid vesicles and lipid nanodiscs. We applied an unbiased molecular dynamics approach, normal mode flexible fitting (NMFF) (8–11), to a truncated version of the prepore structure and the EM protein density map of the domain 4-deleted PA pore and found that this analysis placed Phe427-containing loops of the seven monomers into the flange density within the pore lumen. These results therefore identified the location of a crucial element of the pore’s protein translocation machinery.

Results

The use of nanodisc technologies to solubilize the anthrax toxin pore was pursued with the hope that this approach might provide a general platform for assembling and viewing pore–toxin and/or pore–receptor complexes. Demonstrating that we could obtain a preliminary structure of the PA–nanodisc complex would aid our quest for higher resolution cryoEM structures and could provide insights into how the PA pore mediates pH-dependent protein translocation.

There are numerous examples where detergent-solubilized or newly synthesized membrane proteins have been introduced directly into preformed lipid nanodiscs (12, 13), and initially we focused on forming the PA pore in the presence of preformed nanodiscs. We exploited the fact that the nanodisc protein construct employed (MSP1D1) contained an engineered His tag at the N terminus. We formed the nanodisc, immobilized it by using Ni-nitrilotriacetate (NTA) resin, created and inserted the PA pore directly into the immobilized nanodisc during solution acidification, and finally eluted the nanodisc–PA-pore complexes. This approach proved problematic, however, as only larger nanodiscs (∼200–250 Å in diameter) were found to contain inserted PA pores. Nonetheless, during the course of these studies, we discovered that the PA prepore oligomer bound to the Ni-NTA resin under low stringency (0–5 mM imidazole) loading conditions. From this observation, we reasoned that it might be possible to bind the prepore to the resin, initiate the transition from prepore to pore directly on the Ni beads, and assemble the nanodisc around any exposed hydrophobic tips. This approach gave a good yield of PA pore inserted into lipid nanodiscs, plus pore inserted into lipid vesicles, aggregates of the pore, empty vesicles, and empty nanodiscs. We used this approach to prepare samples for examination by negative-stain EM.

About equal amounts of PA pore inserted into nanodiscs and into vesicles. The diameter of nanodiscs varied from roughly 100 to 400 Å, with most ranging between 100 and 250 Å. It has been reported (12) that the size varies in a stepwise manner, reflecting the varied number of lipids packing into individual nanodiscs. The larger variety in sizes observed within the negative-stained fields may also reflect a small degree of flattening of the nanodisc absorbed perpendicular to the carbon film. Under normal lipid loading, one should obtain 2 mol MSP1D1 per disc. We avoided large-scale pore aggregation during elution of the PA nanodiscs and PA vesicles from the Ni-NTA resin by eluting at pH 7.5. Some prepores were observed, indicating that conversion to pore in the bead-bound state was incomplete. Aggregated pores observed were no longer in the form of an amorphous mass, as visualized in our previous work (3), but rather clusters of recognizable pore structures.

Large numbers of nanodisc-inserted PA pores (1,760) and vesicle-inserted PA pores (2,320) were picked from 33 micrographs (Fig. 1B and D). There were numerous instances where multiple PA pores were inserted into a nanodisc, from the same side, the opposite side, or both. Multiple PA pores in a nanodisc complicated image analysis, and therefore we restricted our analysis to nanodisc particles with single inserted pores. Two-dimensional averages of the particles were generated with a reference-free alignment procedure from nanodisc-inserted PA pores (Fig. 1C) and from vesicle-inserted PA pores (Fig. 1E). The two images differ little from each other, except that the vesicle membrane appears thinner than the nanodisc membrane. The hollow stem and lumen extending to, and penetrating through, the lipid bilayer (nanodisc or vesicle) can be observed in both nanodisc-inserted and vesicle-inserted averages. The density of the pore protein at the border between the stem and lipid bilayer is greatly attenuated in both averages.

Fig. 1.

Raw EM images and two-dimensional averages. (A) A representative field of PA WT negatively stained EM micrograph. (Scale bar: 100 Å.) Red arrows point at PA inserted into nanodiscs. Yellow arrows point to vesicle-inserted PA. Aggregated particles can be observed in the upper right corner. (B) Representative raw particles of nanodisc-inserted PA WT pore. (C) Two-dimensional average of nanodisc-inserted PA pore. (D) Representative raw particles of vesicle-inserted PA WT pore. (E) Two-dimensional average of vesicle-inserted PA WT pore.

Three-Dimensional Reconstruction of WT PA Pores in Nanodiscs and Vesicles.

Two independent three-dimensional reconstructions, from nanodisc-inserted PA-pore particles and vesicle-inserted PA-pore particles, were performed, each yielding a structure at 22-Å resolution (Fig. 2A–D and E–H). The resolution was estimated by Fourier shell correlation with a cutoff when the value dropped to 0.5 (Figs. S1 and S2). Both structures were close to that previously determined for the PA pore with GroEL as the aggregation suppression agent. The structures are mushroom-shaped with a relatively thin tubular stem and a thicker cap with seven globular domains extending outward, and they stand on either the flat surface of the nanodisc or the rounded surface of the vesicle. The nanodisc is visible with the surface representation at 100% protein volume (Fig. 2A), but the vesicle is not (Fig. 2E), possibly because of averaging of vesicles with different curvatures. The stem of the PA pore seems to disappear adjacent to the nanodisc or vesicle when displayed as a surface representation. The cap of the PA pore is similar in appearance in both nanodisc-inserted and vesicle-inserted reconstructions. Density corresponding to domain 4 of PA is missing from the surface representation of both reconstructions, as was the case with our earlier PA-pore structure with GroEL. The constriction within the lumen of the cap region observed in our previous reconstructions (3) appears closed in the new reconstructions (Fig. 2B and F, white arrows). The reconstruction has a height of ∼180 Å (top of the cap to the bottom of the nanodisc), cap width of ∼130 Å, and stem width of 38 Å (Fig. 2A). The vesicle and connections between the PA pore and the lipid can be faintly seen in a projection image (Fig. 2C and G). It is noted that the vesicle’s lipid bilayer exhibits significantly more curvature than the nanodisc. A central slice of the PA-nanodisc or vesicle reconstructions clearly shows that the lumen of the pore traverses the stem and the lipid bilayer in both reconstructions (Fig. 2D and H).

Fig. 2.

Three-dimensional reconstruction of PA WT pore inserted in nanodiscs (A–D) and in vesicles (E–H). The reconstructions were low-pass filtered at their nominal resolutions of 22 Å. Surface representations of PA WT incorporated in nanodiscs (A) and vesicles (E). The yellow arrows denote the positions of domain 3. Surfaces displayed to include 100% of the molecular volume. (B and F) Central cuts of the reconstructions are made to display the interior structure. The white arrows mark the location of the prominent flange seen in all cross-sections of the pore. (C) and (G) represent projections of the structures. (D) and (H) represent central slices.

Structure of Domain 4 Deletion Mutant.

Domain 4 of PA is expected to have little contact with the rest of the molecule and was not resolved in any of our WT PA-pore structures. We speculated that deleting domain 4 from the molecule could be helpful for reconstruction of the rest of the molecule by eliminating smeared density due to movement of the domain 4. The domain 4 deletion mutant could also be beneficial to use for NMFF of the crystallographic structure of the PA prepore into the EM pore structure, because the fitting routine relies on the raw electron density map, including that contributed from the smeared density of domain 4. Because the PA prepore formed from the domain 4 deletion mutant is less stable than WT and easily converts to the pore, the Ni-NTA-dependent pore formation was performed at higher pH (pH 8.5).

Domain 4 deletion pores (1,455) inserted into nanodiscs and vesicles were selected from 22 negatively stained micrographs, and a 23-Å resolution structure was calculated (Figs. S1 and S2). Vesicle and nanodisc particles were combined here because no significant difference in the protein structures was seen when they were compared with the WT images. The use of the mutant did not significantly affect the resolution (Figs. S1–S3). The surface representation shows a smaller domain 3 (yellow arrows in Fig. S4A) compared to previous WT reconstructions (yellow arrows in Fig. 2A and E), which may reflect the loss of contributions from the smeared domain 4 density in the WT PA reconstructions. The negative stain was lighter for the interior of the domain 4 deletion mutant specimens, which may explain why there appears to be a smaller open luminal space within the interior of this pore construct (Fig. S4). The projection image and the slice show the lumen within the cap and stem mostly filled, but the flange region connection is still observed (Fig. S4). The connection to the membrane appears more solid, which may also be because of the lighter stain accumulation on the surface of the membrane.

NMFF Analysis of EM Structure and X-Ray Crystallographic Coordinates.

The initial PA-pore EM structure fit well with the prepore crystal structure in domains 1 and 3 (see correlation coefficient below), but domain 2 extended outward in the PA prepore crystal structure compared with this region into the EM structure envelope. This extension of domain 2 indicates that there must be significant inward movement of this domain as the beta strands comprising residues 275–352 unfold and refold into the 100-Å extended β-barrel structure (3, 14, 15).

To investigate these rearrangements, the EM structure and the 2.8-Å x-ray crystallographic coordinates (2) were used to perform dynamic fitting on the basis of normal mode analysis with the NMFF program package (11). Normal modes are preferred for the flexible fitting because they resemble the large low frequency conformational changes commonly observed in proteins (9, 10). The NMFF analysis is preferred over the independent fitting of disconnected domains, because these fits have to incorporate all structural constraints of the connected hinge regions, allowing one to obtain an energetically reasonable structure (16).

The initial correlation coefficient between computed and measured electron densities fitting the prepore PA into the PA-pore cap region was 0.83 (Figs. 3 and 4). During the refinement, only the lowest frequency normal modes that display constructive 7-fold symmetry modes were selected as search directions. NMFF without this symmetry constraint resulted in unequal subunit movements. The correlation coefficients after interactive fits into the electron density maps is 0.89 with the domain 4 truncated PA pore. The rms difference between the initial truncated prepore structure and the fitted pore cap region structure is 4.6 Å (Figs. 3 and 4). The rmsd at the monomer level is 3.5 Å, which mainly consists in a rearrangement of domain 2 position relative to other domains. In the final fit (Fig. 3B and D), Asp425 is shown in red as a marker of the flexible Phe-clamp loop, and it is evident that there is substantial movement of the loop into the luminal flange and toward the symmetry axis of the pore. The residues surrounding Phe 427, namely, Asp 425 and Ser 428, move by as much as 9.5–11 Å.

Fig. 3.

Comparison of initial and final NMFF analyses. Backbone models of the truncated PA heptamer before (A) and after (B) NMFF illustrate the conformational adjustments that occur as the lower half of domain 2 moves into the EM electron density of the luminal flange in the lower cap region. Four subunits of the heptamer are shown to visualize the changes within the pore lumen. Asp425, two residues removed from the unresolved Phe427, is shown in red space-filled representation. There also appears to be outward movement of the top of the PA cap. Top views of the initial (C) and final (D) NMFF analysis shows the movement of the Phe-clamp loop structures toward the 7-fold axis of the pore (red arrow heads).

Fig. 4.

NMFF of the x-ray coordinates of the truncated PA prepore into the EM protein density shell of the PA cap region. The Protein Data Bank coordinates of domains 1, 2, and 3 of the prepore (1TZ0) minus β-strands 1–4 of domain 2 (which form the β barrel stem of the pore) were used. The correlation coefficient for both fits was high (0.89) for the PA-pore cap region fitting into the EM density map. The solid surface (transparent) and the C^α backbone (white) ribbon representations were generated by using Chimera (UCSF).

Discussion

Biophysical measurements with planar lipid bilayers (17) have demonstrated the formation of ion-permeable and translocation-competent PA pores (18–20), but until now there has been no visualization of a PA pore inserted into lipid membranes. We succeeded in inserting the PA pore into both vesicles and ∼100 Å lipid nanodiscs with good efficiency. Our ability to induce the prepore-to-pore transition from an immobilized state as the nanodisc assembled, thus avoiding off-pathway PA-prepore aggregation that occurs in solution, was a key to success. Over the past decade, attempts to obtain high concentrations of PA pores in solution for structural analysis have been unsuccessful largely because of aggregation of the pore. Earlier we demonstrated that the chaperonin GroEL could stabilize the PA pore during the prepore-to-pore transition and thereby increase the concentration of observable pore structures. In the current study we were able to avoid aggregation in solution by binding the prepore to an inert matrix (Ni support resin) and initiating the transition to the PA pore from an immobilized state. The first indication that this alternative approach could be useful was when we observed that aggregates of released PA contained significant amounts of resolvable PA pores (3). When this transition was performed in the presence of assembling vesicles and nanodiscs, a significant population of PA pores in nanodiscs and vesicles was observed, allowing reconstructions of the PA–pore nanodisc/vesicle complexes from negative-stain EM single particle analysis.

The dimensions of the pore agree with our previous determination, with minor modifications. The structure obtained from the GroEL stabilization method showed a PA-pore barrel that was abnormally wide (50 Å) compared to the biochemically and biophysically predicted 14-strand structure (33 Å) (15). When correspondence analysis was applied to the GroEL-stabilized pore, many classes emerged on the basis of heterogeneity in the orientation of the barrel structures (3; see Supplemental Data). A significant portion of both the GroEL bound and free PA pores separated into populations defined by bent barrel structures. It was noted that the structures of the barrel may have been distorted because of negative staining artifacts such as grid adherence. In contrast, the nanodisc or vesicle PA-pore structures did not show any sign of bent barrels. The averaged and reconstructed membrane-inserted barrel structures appeared essentially normal to the plane of the nanodisc or vesicle. There were instances of angled insertions into the nanodisc, but even in these cases the beta barrel showed no bending. For these newer images, the width of the barrel approaches the width predicted for an extended 14-stranded beta barrel (∼33 Å predicted vs. 38 Å observed). Also, the distinct hydrophobic tip inserts into the lipid bilayer of the vesicle or nanodisc with the expected penetration distance of 40 Å. The distance from the bottom of the nanodisc to the cap-proximal beginning of the beta barrel was measured as 100 Å, in agreement with the previous predicted structure of the pore (14). We estimated the penetration distance by subtracting the distance (60 Å) from the beginning of the beta barrel to the surface of the lipid bilayer, defined as the center of highly accumulated stain at the nanodisc/vesicle protein interface. The structures observed herein show an obvious anomaly, an apparent lack of protein density at the nanodisc/vesicle protein interface. This missing density could have any of several possible explanations, such as negative staining artifacts (stain accumulation) or heterogeneity in the barrel orientation insertion into the lipid bilayer (angled vs. normal).

The most significant and functionally relevant structural feature of the PA-pore structures is within the pore lumen. In the lipid-inserted and GroEL-stabilized PA-pore structures, a luminal flange within the cap region is prominent in every pore reconstruction that we have determined to date. Such a flange is not seen in the low-pass filtered prepore crystal structure. The flange appears to be more prominent (i.e., connected) in the methylamine tungsten stain (pH 7) than in the GroEL-stabilized uranyl acetate stained images (pH 4.5), but, regardless of the pH, NMFF placed the flexible Phe-clamp loop within these internal densities. Two residues (Asp426 and Phe 427) within the Phe-clamp loop were not resolved in the crystal structure, but the position of the adjacent residue, Asp425, could place Phe427 close to the pore 7-fold axis, consistent with a constriction annulus within the lumen. Notably, fitting with the NMFF program involves no bias that might target this loop into the flange. The positioning of this phenylalanine flexible loop into these flange structures is consistent with previous biophysical measurements that implicate the loop containing Phe427 as a crucial protein structural motif that controls protein translocation. Mutating this particular Phe strongly impairs toxicity and translocation across model lipid bilayers (1). Furthermore, electron paramagnetic measurements of a spin probe attached to Cys in PAF427C gave strong evidence of relocation of Phe427 towards the 7-fold axis during prepore-to-pore conversion, such that these residues of the seven subunits are within 10 Å in the pore conformation (21).

We have consistently observed that domain 4 is not resolved in any of our EM reconstructions, most likely because of the inherent flexibility of this domain. Domain 4 is mainly involved in binding to host cell receptors and appears to have little or no contact with the remainder of the pore structure. Indeed, the removal of domain 4 in our truncated PA-pore structures of elimination of irrelevant smeared density improved the fit of crystallographically derived electron density into the EM density map.

There are many advantages of constructing PA–pore nanodisc complexes for structural analysis. In our studies the predominant 1∶1 stoichiometry simplified the reconstruction analysis and may enable us to visualize possible structural changes of the PA pore (particularly within the interior PA lumen) induced by physiological pH shifts. In addition, we may be able to determine if the binding of the enzymatic moieties of anthrax toxin, such as the lethal factor, results in larger scale distant changes within the stabilized PA-pore nanodisc complex. A cryoEM reconstruction of the prepore–lethal factor complex suggested that lethal factor binding resulted in a large asymmetric change in the prepore structure leading to an expansion of the interior luminal space (22).

Methods

Expression and Purification Of Proteins.

Both WT PA and domain 4-truncated constructs were overexpressed recombinantly in the periplasm of Escherichia coli BL21 (DE3). WT PA was purified by anion-exchange chromatography (23), and WT prepore was purified by anion-exchange chromatography after activation of PA with trypsin (24). The domain 4-truncated mutant was constructed by inserting the stop codon TAA between the codons for Arg595 and Phe596 by using site-directed mutagenesis. The domain 4-truncated PA monomer was found predominantly in the flowthrough material (> 90% pure) after anion-exchange chromatography. The domain 4-truncated PA prepore converts to pore at pH 8.0, and the nicked PA was therefore exchanged into 20 mM Tris, pH 9.0 buffer before anion-exchange chromatography. With 20 mM Tris, 0.5 M NaCl, pH 9.0 buffer as the elution buffer, the domain 4-truncated PA prepore eluted at 400 mM NaCl.

Preparation of PA Pore in Nanodiscs and Lipid Vesicles.

The PA prepore was found to interact with Ni-NTA resin. The histidine tag was cleaved from membrane scaffold protein 1D1 (MSP1D1; provided by the Sligar laboratory; all genetic constructs are available from AddGene (http://www.addgene.org/pgvec1)) (12) with AcTEV protease (Invitrogen). The removed His tags and any MSP1D1 molecules that still contained uncleaved His tags were removed with PerfectPro Ni-NTA Agarose (5 PRIME). PA prepore (2 nmol) was added to 100 μL (wet volume) PerfectPro Ni-NTA Agarose in 50 mM NaCl, 50 mM Tris-HCl, pH 7.5 (buffer A) for WT PA or in 50 mM NaCl, 50 mM Tris-HCl, pH 8.5, for domain 4 deletion PA mutant and incubated on ice for 5 min. The resin was washed twice with 500 μL 50 mM NaCl, 50 mM sodium acetate, pH 5.5 (buffer C). The resin was incubated in the second wash of buffer C for 5 min before being washed with buffer A twice to adjust the pH back to 7.5. Under these conditions, the PA pore showed little to no reversion to the prepore state. A MSP1D1 and palmitoyloleoyl phosphatidylcholine (POPC) mixture diluted into the Ni-NTA resin slurry containing bound PA pore to yield a final mixture (500 uL) that contained 0.4 μM Ni-NTA resin bound PA heptamer, 4 μM MSP1D1, and 400 μM POPC in buffer A containing 25 mM sodium cholate. It is important to note that the MSP1D1, POPC, and sodium cholate for the final mixture were initially combined and incubated for 10 min before it was added to the Ni-NTA resin slurry containing bound PA pore. The mixture, including the resin, was dialyzed with Spectra/Por Membranes molecular weight cutoff: 12–14,000 (Spectrum) in excess buffer A for 8–12 h for each of three buffer changes to remove the cholate detergent. The Ni-NTA resin was collected after dialysis, washed three times with 500 μL containing 500 mM NaCl, 50 mM Tris-HCl, pH 7.5, and then washed with buffer C twice. The resin was treated with 100 μL 50 mM imidazole, 50 mM NaCl, 50 mM sodium acetate, pH 5.5, and the eluted material was collected and prepared for electron microscopic examination.

Negatively Stained EM and Image Analysis.

For EM grid preparations, a diluted mixture of the PA-nanodisc/PA-vesicle suspension, estimated to be approximately 1 μM in PA, was applied to carbon-coated Cu 300 mesh grids and stained with 1% methylamine tungstate. Images were recorded on film (Kodak SO163) by using a minimal-dose protocol at a magnification of 60,000× with a JEOL 1200EX electron microscope at a defocus of 0.6–0.7 μm. The micrographs were digitized with a Microtek ScanMaker i900 scanner at a pixel size of 5.4 Å on the specimen. The power spectra of the micrographs were visually inspected to check for astigmatism and other optical faults. The defocus of each micrograph was estimated by using the Alternating Conditional Expectation MATLAB toolbox (25). Particles were selected and corrected for phase inversion with EMAN (26). Image alignment and 3D structural analysis was performed with SPIDER (27). Two-dimensional averages were generated by using a reference-free alignment procedure. A simple cylinder (radius of 30 Å and height of 170 Å) was used as an initial model for the three-dimensional reconstruction. The three-dimensional structure was calculated by projection matching (28) to the simple cylindrical model and subsequently to the resulting reconstructions in an iterative fashion until less than 5% of the particles move between classes. Model projections were calculated by using 5” angular increments. Because the PA pore is a heptamer, 7-fold symmetry was applied during the reconstruction. Nanodiscs and vesicles were masked on each picked particle, and the masked images were used for the projection-matching procedure. The final three-dimensional structures were generated with the raw particles without the mask by using the Euler angle assignment from the final round of projection matching. Three-dimensional structures were displayed by using University of California–San Francisco (UCSF) Chimera (29). Surface representations were shown at 100% volume of the proteins (excluding domain 4 in WT) calculated by using a general protein density of 1.35 g/cm³. The nanodisc and vesicle regions were masked during threshold estimation for 100% protein surface volume representation.

NMFF.

The NMFF procedure was used for the fitting (11). NMFF uses a linear combination of low frequency normal modes in an iterative manner to optimally deform the structure so it can conform to the lower-resolution structure. Normal modes were computed by using a simplified potential (30) and the rotation–translation block method (8). It has been previously demonstrated that NMFF could be performed by using simple C^α-based models (11), allowing one to obtain faster calculations while giving results similar to those carried out by using an all-atom approach. Thus, only the C backbones were taken into account in the refinement.

The NMFF algorithm was adapted to take into account the 7-fold symmetry of the PA-pore structure. Only the lowest 10 frequency normal modes that display constructive 7-fold symmetry modes were selected as search directions. Once the correlation coefficient converged and no longer showed significant changes, an all-atom structure for the final C^α-based model was reconstructed by using Pulchra (31). The structure was subsequently energy minimized. The molecular graphics images schematically shown as ribbon representations and surface views illustrated in Figs. 3 and 4 were produced by using the UCSF Chimera package from the Computer Graphics Laboratory, University of California, San Francisco (supported by NIH P41 RR-01081) http://www.cgl.ucsf.edu/chimera/ (29).