Structure of the Hemolysin E (HlyE, ClyA, and SheA) Channel in Its Membrane-bound Form*

Hemolysin E (HlyE, ClyA, SheA) is a pore-forming protein toxin isolated from Escherichia coli. The three-dimensional structure of its water-soluble form is known, but that of the membrane-bound HlyE complex is not. We have used electron microscopy and image processing to show that the pores are predominantly octameric. Three-dimensional reconstructions of HlyE pores assembled in lipid/detergent micelles suggest a degree of conformational variability in the octameric complexes. The reconstructed pores were significantly longer than the maximum dimension of the water-soluble molecule, indicating that conformational changes occur on pore formation.


* This work was supported by the Biotechnology and Biological Sciences
Research Council. The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked "advertisement" in accordance with 18 U.S.C. Section 1734 solely to indicate this fact. 1 4 The abbreviations used are: HlyE, hemolysin E; BTLE, brain total lipid extract; EM, electron microscopy; MRA, multi-reference alignment; MSA, multivariate statistical analysis; ␤-OG, n-octyl-␤-D-glucopyranoside; TCB, thrombin cleavage buffer. magnification was calibrated by imaging of negatively stained catalase crystals (Agar). Determination of Pore Symmetry-All alignments and multivariate statistical analyses (MSA) were carried out using the IMAGIC-5 software package (Image Science Software GmbH) (11,12). After direct alignment to the total average and cycles of multi-reference alignment (MRA) to selected averages, ϳ650 images of top views were selected from an initial set of over 1400 images of top and side views of HlyE pores in ␤-OG. "Top views" were images of pores viewed down the putative axis perpendicular to the membrane plane, whereas "side views" were images of pores viewed parallel to the putative membrane plane. For determination of the averages used in the MRA and for selection of the images, three different criteria were used: high overall quality as defined in IMAGIC, high total number of images in the corresponding MSA class, and representation of intact circular complexes.
The selected set of top views was subjected to additional cycles of MRA with application of symmetries ranging from 6to 12-fold to all references. This produced three class averages of different sizes, with no symmetry enforced: "big," consisting of 84 views (13%); "medium," consisting of 375 views (58%); and "small," consisting of 72 views (11%). Finally, images within each of these size groups were aligned by cycles of direct align-ment to the total average of the group (with no symmetry enforced) (Fig. 2, B-D, insets). After the final alignment, rotational power spectra (13) of the total average were determined in each subset (Fig. 2, B-D) using the Medical Research Council (MRC) package of programs (14,15).
Electron Microscopy of HlyE Pores in Mixed Brain Lipid/␤-OG Micelles-To determine the optimal conditions for imaging of individual pores in side views, protein, detergent, and lipid were mixed in varying final concentrations in the range of 125 to 325 g/ml for the HlyE, 0.3 to 3.8% for the ␤-OG (0.5-6ϫ critical micelle concentration), and 0.1 to 4 g/ml for the brain total lipid extract (BTLE) (Avanti Polar Lipids, Inc.). The mixtures were incubated at 37°C for 15 min and kept overnight at 4°C before being examined. (Fig. 3). For example, 20 l of 250 g/ml HlyE in 1.2% ␤-OG (2ϫ critical micelle concentration), containing 4 g/ml BTLE was prepared by mixing 10 l of 500 g/ml solution of HlyE in TCB with 10 l of 8 g/ml BTLE solubilized with 2.4% ␤-OG in TCB. A stock solution of 4 mg/ml lipid solubilized in 2% ␤-OG and TCB was obtained by drying 8 l of 20 mg/ml chloroform solution of BTLE in a stream of air at room temperature followed by the addition of 2.67 l of 30% water solution of ␤-OG, 29.2 l of deionized H 2 O, and 8 l of 5ϫ TCB. The mixture was vortexed briefly after the addition of each component and finally sonicated for 10 min at room temperature. Micrographs were recorded under similar conditions as those used for protein in detergent-only micelles but in low dose mode.
Electron Microscopy of Truncated HlyE Pores-Truncated HlyE was loaded on grids either after dilution of the initial stock solution in TCB or after reconstitution experiments identical to those described previously for the wild type protein (6). Micrographs were recorded as described above (Fig. 4A). The top view average (Fig. 4B) of the pores was obtained after cycles of direct alignment of 102 individual images to their total average followed by rotational averaging of the total sum of aligned images. The side view average was obtained after cycles of direct alignment of 161 individual side view images. Mirror symmetry was enforced along the vertical axis; this is equivalent to cylindrical averaging (Fig. 4C).
Correlation analysis of the nonsymmetrized total average of truncated pore side views was done by cross-correlating the average with side projections of rotationally averaged reconstructions of the full-length pore. The highest correlation was obtained for the nonassociation end of the full-length complex, with a maximal value found for the "wide" pore reprojection. Correlation to "squat" pore reprojection was 98% and to "narrow" pore reprojection was 95% of the maximum. The highest correlation to the association end half of the reprojection was again for the wide pore reprojection (70% of the maximum).
Three-dimensional Reconstruction of Full-length Pores-A small set of ϳ100 side view images of HlyE pores in ␤-OG was subjected to direct alignment to a total average, six cycles of MRA to selected averages (selected by the criteria specified above), and two cycles of three-dimensional reconstruction, reprojection, and MRA, to give a preliminary three-dimensional model of the HlyE pore. The latter was reprojected to produce a single reference for the direct alignment of a set of over 1000 individual images of side views of HlyE pores in BTLE/␤-OG micelles. After MSA of the aligned images, MRA and a second round of MSA, selected averages were used for calculation of the initial low-resolution three-dimensional reconstruction of the HlyE pore (Fig. 5). The initial reconstruction was refined through cycles of reprojection, MRA, MSA, and three-dimensional reconstruction. Projections were then sorted according to their shape and width into four groups, and four three-dimensional volumes were reconstructed from the projections in each group as described above. The reconstructions were refined by further cycles of reprojection, MRA, and MSA until convergence was reached. Eight-fold symmetry was applied to all three-dimensional reconstructions. From the initial set of four volumes, two merged during this further refinement leaving three unique reconstructions: squat pores, wide pores, and narrow pores. To correct for the widening of the reconstructions because of the flattening of the negatively stained pores, the size of the top view reprojection for each volume was compared with the size of the averages produced during the analysis of the experimentally obtained top views (Figs. 2, B-D, insets, and 6B) by magnification alignment. A zoom factor of 0.9 was found for all reprojections and applied to compensate for the effects of lateral flattening.
Determination of the threshold density for surface rendering was carried out by examination of two-dimensional contour maps of cross-sections of the volumes. The threshold was determined at the midpoint of the maximum density gradient.
Current data do not allow determination of the absolute handedness of the channel, but for representation purposes, an arbitrary hand was chosen (Fig. 6, C and D). Comparison of the cross-correlation coefficient between projections used for the reconstruction of the narrow pores and reprojections of the original and mirrored volumes of the other two reconstructions was used to determine the relative handedness of the squat pores and the wide pores (Fig. 6C).

RESULTS AND DISCUSSION
Pores in Lipid Vesicles-We have previously described EM experiments in which HlyE pores reconstituted in lipid vesicles formed large dome-like structures consisting of numerous closely packed pores (6). Similar clusters have been observed in electron micrographs of E. coli overexpressing HlyE in vivo (16). Precise analysis of these images is hampered by the variable size of the pores, the irregular nature of their close packing, and the curvature of the dome-like assemblies. This curvature appears to arise from the tapering V-shape of the individual pore assemblies, which are narrower at the end that binds in the membrane and wider on the exterior (Fig. 1B). Very rare observations of isolated pores in side view confirm this tapering (Fig. 1C). To obtain more information on the structure and stoichiometry of the pore, we attempted to collect many more images of isolated pores for single particle work.
Preparation of Pores for Single Particle Work-HlyE forms pores in a variety of detergents, but we found that in 1.2% of ␤-OG (2ϫ critical micelle concentration) the proportion of clearly identifiable individual pores was highest. In these preparations, all images of pores suitable for image analysis were of top views, i.e. views down the putative axis perpendicular to the membrane plane ( Fig. 2A), suggesting a preferred surface of interaction of the complex with the carbon film. Pores in side views (views parallel to the putative membrane plane) were generally seen in multilayered clusters ( Fig. 2A).
Symmetry of the HlyE Pore, Analysis of Top Views-Average images were selected as references for MRA on the basis of the number of images making up each average, low variance, and a fully intact circular structure. MRA was performed on selected top views using references with 6 -12-fold rotational symmetry applied. Four cycles of MRA resulted in the generation of three class averages of differing size and symmetry (Fig. 2, B-D, big (13%), medium (58%), and small (11% of all analyzed top views)). Rotational power spectra (13) showed conclusively that the big and the medium averages both had 8-fold symmetry, whereas the small average had 6-fold symmetry (Fig. 2, B-D). The existence of small but significant populations of pores with sizes considerably different from the overall average is in good agreement with the previous observations of pores reconstituted in lipid bilayers (6).
In some of the top views of pores (Figs. 1, B and C, and 3B) there is an apparent central density feature, although this is not present in all of the top views. This density is comparable with FIGURE 3. Imaging of HlyE at different brain lipid/n-octyl-␤-D-glucopyranoside ratios (w/w). A and B, diagram (A) and electron microscopy (B) of the observed types of pore aggregates at different relative lipid and detergent concentrations. 1 (ࡗ), pores in lipid vesicles; 2 (E), few pores possibly inserted in bilayer fragments or in discoidal micelles (24); 3 (F), spherical aggregates of pores; 4 (छ), paracrystalline "sheets" and increased fraction of individual pores in side views; 5 (‚), single pores aggregating into "chains"; 6 (OE), single pores with a preponderance of top views. Symbols represent observations at certain relative concentrations and vary according to the type of the pore aggregates found. The pores shown in Fig. 5A were formed under conditions 4 and 5 (Fig. 5A, inset) shown here. Scale bar ϭ 50 nm.
that of the background and most likely arises from a preferential accumulation of negative stain along the walls of the pores. Moreover, there is no indication of any central protein density in side views of pores (Fig. 5, C and D). Although previous EM, biochemical, and biophysical data have suggested approximately eight HlyE subunits (6,9,16), the present work constitutes the first direct observation of the oligomeric structure of HlyE.
Analysis of Side Views-Side views of isolated HlyE pores were extremely rare in the electron micrographs of samples prepared in ␤-OG alone. Instead, the majority of side views were of pores in small clusters ( Fig. 2A), unsuitable for image analysis. Extensive searches for alternative conditions, including addition of lipid to the protein/detergent mixture, produced a wide variety of different types of pore aggregates (Fig. 3) but very few individual pores. Mixtures of ␤-OG with defined phospholipids such as 1-palmitoyl-2-oleoyl-sn-glycero-3-phosphocholine (POPC) or with POPC and cholesterol were tested, but we found that mixtures of brain lipids and ␤-OG gave the most rapid and extensive assembly of pores. Even under these most favorable conditions most of the pores in side views were found in aggregates. As a control we examined electron micrographs of detergent/lipid mixtures without adding protein, but none of these showed pores or pore aggregates as found in the presence of protein.
Only one end of any individual pore assembly appeared to be involved in inter-pore aggregation, i.e. it is evident that the inter-pore interactions are either all head-to-head or all tail-to-tail. Thus the long "chain-like" pore aggregates and the pore pairs and triplets in side views are all built from protein pores, which appear to be specifically oriented with respect to one another in a symmetrical manner with an apparent inter-pore "association end" toward the center of the aggregate (Figs. 3 and 5A). This phenomenon allows the relative orientations of the pores in side views to be readily determined.
Low Resolution Three-dimensional Reconstructions-Attempts to obtain a three-dimensional map of the pore by using the random conical tilt approach (17) on the top view pores seen in ␤-OG were unsuccessful, probably because of the extensive flattening of the pores in the axial direction perpendicular to the carbon support. Similarly, attempts to produce a three-dimensional reconstruction by combining top views with side views were unsuccessful, probably because of the differential flattening of the pores in axial and lateral directions (18,19). Therefore, tomographic reconstruction from side view images (19,20), with application of the 8-fold symmetry expected from the predominant symmetry of top views, was used for generation of a low-resolution three-dimensional map of the pore assembly. A small initial set of ϳ100 images of pores in side views was selected from micrographs of HlyE in mixed lipid/␤-OG micelles and used to generate a starting three-dimensional model of the HlyE pore. This was used as an initial reference for direct alignment of a larger set of Ͼ1000 side view images of individual pores (Fig. 5B) formed under conditions corresponding to image 4 of Fig. 3B. Cycles of refinement by projection matching were applied until convergence was reached with a final set of three different reconstructions of the HlyE pore (Fig. 5, B-D), which were designated as squat (184 images), wide (326 images), and narrow (348 images). The resolutions of the reconstructions were estimated by Fourier shell correlation as ϳ30 Å (threshold value 0.5) (Fig. 5E).
Analysis of Three-dimensional Maps-The previous conservative model of the pore (6) suggests a tapered complex with an external diameter of ϳ110 Å at the top and a height of ϳ100 Å. Molecular envelopes of the three distinct species of pore identified also reveal slightly tapered assemblies. However, all three-dimensional reconstructions are significantly larger than those of the existing model (Fig. 6A).
Flattening-induced widening of the side view projections of cylindrical complexes is a known artifact in negatively stained specimens (18,19). When reprojections of the three-dimen-  Fig. 2). C, total average of 161 individual side view images of truncated pores (cf. Fig. 5, C and D). The threshold for the measurements is at the steepest gradient between the densities of the pore and the surrounding stain. D, mass spectrometry and sequencing data indicate that truncated pores are complexes built from two HlyE fragments containing the whole of ␣B, part of ␣C, part of ␣F, and the whole of ␣G (represented as ribbons). Sequences not confirmed in the truncated pores are represented as C␣-wire in this diagram of the water-soluble HlyE monomer (cf. Fig. 1A).
sional reconstructions are compared with the top view averages of HlyE pores (Fig. 6B), it is evident that the top view reprojections of both the squat and the wide pores have an appearance similar to the big top view average, a ring of relatively well contrasted subunits. The subunits are less contrasted in the reprojection of the narrow pores, similar to the medium top view average. In all cases, however, the reprojections of the reconstructions from side view images were wider than the top view averages and had to be rescaled to be matched (Fig. 6B). Magnification alignment of the reprojections and the top view averages confirm that the squat pores and the wide pores best correlate with the big top view averages, with a magnification correction factor of 0.9. The narrow pores correlate best with the medium top view average, also with a correction factor of 0.9.
These results suggest that, because of the similarity of their diameters (Fig. 6A), top view reprojections of negatively stained wide and squat pores are indistinguishable (Fig. 6B), and probably the group of big top view averages, found during process-ing of top view images, is a mixture of individual views of squat and wide pores. Analysis of the reconstructions also reveals that the minimum internal diameter is located at the association end of the pore for all three reconstructions (Fig. 6, A and C). The HlyE channel has been predicted to have a minimum diameter of ϳ30 Å (1,7,8). After correcting for the effect of flattening, we found the minimum inner diameter of the narrow pores to be comparable at ϳ35 Å.
Truncated HlyE-When incubated for several months at 4°C, HlyE pore preparations tend to partially hydrolyze, possibly because of traces of residual protease activity in the preparation. When observed by EM, the products of this partial hydrolysis are "short pores" (Fig. 4A), in which the top views (Fig. 4B) appear very similar to those of the complete pore, but the side views are significantly shorter (Fig. 4C) than the 120 -140-Å long full-length pores. Image analysis confirms that the side view average of these truncated pores (Fig. 4C) has a diameter comparable with the wider end of the squat or wide pores (Fig. 5D). These data, coupled with the absence of the narrow-

Structure of the Hemolysin E Channel
AUGUST 11, 2006 • VOLUME 281 • NUMBER 32 ing observed at the association end of the full pores (Fig. 4C), suggest that the truncated pores correspond to the nonassociation end of the full-length pores. This view is further supported by the absence of the pairwise associations characteristic of the full-length pores. Moreover, the images of the truncated pores show that they do not bind to any lipid bilayers tested (Fig. 4A), implying that it is the association end of the full-length pores that interacts with the membrane. Although only one band with mobility corresponding to a molecular mass of ϳ8 kDa was apparent by SDS-PAGE of truncated HlyE, mass spectrometry revealed that there were several hydrolysis products with molecular masses of ϳ8.07, 8.17, 8.28, 8.42, and 8.48 kDa. Two distinct N-terminal sequences were detected: SVLVGDIKTL and SNTVKQANKD. The first is situated at the N terminus of helix ␣B, from which a peptide with molecular mass of ϳ8 kDa would include the whole of ␣B and the first half of ␣C (Fig. 4D). The second sequence commences in the middle of helix ␣F, and an 8-kDa peptide will include the second half of ␣F and the whole of ␣G (Fig. 4D). The absence of helix A in the truncated pore complex suggests that this part of the toxin structure, which would have an expected mass of ϳ4 kDa, has also undergone proteolysis and is therefore not involved in or is peripheral to the hydrophilic end of the pore assembly. All of these findings indicate that the complex in the truncated pores corresponds to the end of the full pore assembly that protrudes outside the membrane and also give further support to the view that it is the head domain, not present in these preparations, that is responsible for membrane interaction.
Single particle analysis of top views of the HlyE pore revealed the presence of at least two types of octameric assemblies and a small group of hexameric pores. Formation of active complexes with varying numbers of subunits is not unusual for pore-forming toxins, e.g. pneumolysin forms pores containing between 30 and 50 subunits (20). The observation of pores with different diameters, but having 8-fold symmetry is interesting and may indicate different forms of the HlyE channel corresponding to different functional conformations (e.g. pore and prepore complexes). The three surface views are generated by restricting the resolution at 50 Å and rotationally averaging the corresponding reconstruction followed by surface rendering at the maximum density gradient. B, comparison between the symmetrized averages from Fig. 2, B and C, and top view reprojections (rescaled by 0.9) from the three reconstructions. In each image, the left half is from the symmetrized top view average, and the right half is from the reprojection. Views of the big top view averages are compared with reprojections of the squat pores (1) and the wide pores (2). Views of the medium top view averages are compared with reprojections of the narrow pores (3). Noise features on the symmetry axis are reinforced by the symmetry averaging. C, surface views of the reconstructions filtered at their resolution limit (Fig. 5E ). Volumes are corrected for flattening, and noise along the symmetry axis has been masked. The threshold for surface rendering is at the steepest three-dimensional density gradient. Two subunits have been removed for clarity. The association end is at the bottom. Scale bar ϭ 5 nm. D, docking of the x-ray structure to a three-dimensional map of the narrow HlyE pore. X-ray crystal structure of the water-soluble monomer is placed within the map with the head domain toward the association/putative membrane end of the reconstruction, as suggested from the analysis of the truncated pores. The three-dimensional map is generated by low-pass filtering of the narrow pore reconstruction at 50 Å and is represented by a mesh drawn at the same threshold level used to render the surface in A (image 3). The resolution is insufficient to reveal structural changes that might take place; the figure merely illustrates the need for a structural change to accommodate the increased length of the subunits within the pore assembly.
In the present study, we were able for the first time to observe individual side views of assembled HlyE channels (Fig. 5), allowing reconstruction of the HlyE pore. All three statistically different HlyE pore reconstructions have lengths significantly greater than the 100 Å of the water-soluble form of HlyE determined by x-ray crystallography. Even the shortest squat pores appear 20 Å longer than the x-ray structure, which is shown placed in the three-dimensional map of the narrow pores in Fig. 6D, the orientation of the HlyE molecule being based on observations for the truncated pores and in agreement with previous indications that the head domain is embedded into the membrane. The observed elongated dimensions are consistent with our more recent EM studies of lipid-inserted complexes where we have rarely obtained side views of individual pores inserted into vesicles (Fig. 1C). These views show the hydrophilic domains extending ϳ80 Å beyond the surface of the bilayer. If the total thickness of the membrane is 50 -60 Å (20 -22), the expected full-length of the HlyE channel would be as much as 130 -140 Å, consistent with our measurements of the complexes in micelles. This suggests that the conversion of the water-soluble toxin to the membrane-bound form may necessitate significant changes in conformation, as observed in other pore-forming toxins (23). For example, hinge movements within the head domain (Fig. 1A) could possibly elongate the molecule by ϳ25 Å, and indeed previous mutagenesis experiments suggest a possible movement between ␤1 and ␣D (9). Thus, the observations made here constitute important advances in our understanding of the mechanism of HlyE function, but it is clear that higher resolution data will be needed to fully characterize the major structural changes that are clearly involved in the transition between the water-soluble monomer and transmembrane pore complex.