While the absence of any cytoskeleton was once recognized as a distinguishing feature of prokaryotes, it is now clear that a number of different bacterial proteins do form filaments in vivo. Despite the critical roles these proteins play in cell shape, genome segregation and cell division, molecular mechanisms have remained obscure in part for lack of electron microscopy-resolution images where these filaments can be seen acting within their cellular context. Here, electron cryotomography was used to image the widely studied model prokaryote Caulobacter crescentus in an intact, near-native state, producing three-dimensional reconstructions of these cells with unprecedented clarity and fidelity. We observed many instances of large filament bundles in various locations throughout the cell and at different stages of the cell cycle. The bundles appear to fall into four major classes based on shape and location, referred to here as ‘inner curvature’, ‘cytoplasmic’, ‘polar’ and ‘ring-like’. In an attempt to identify at least some of the filaments, we imaged cells where crescentin and MreB filaments would not be present. The inner curvature and cytoplasmic bundles persisted, which together with their localization patterns, suggest that they are composed of as-yet unidentified cytoskeletal proteins. Thus bacterial filaments are frequently found as bundles, and their variety and abundance is greater than previously suspected.
The molecular mechanisms that enable bacterial cells to maintain characteristic shapes, establish polarity, segregate their genomes and divide are just beginning to be understood (Gitai, 2005). While it has been known for decades that eukaryotic cells use a complex network of cytoskeletal filaments to accomplish all these tasks, it was not until relatively recently that prokaryotes were thought to even have a cytoskeleton (Margolin, 1998). This view persisted in part because filaments had been seen only in certain rare or special cases by traditional electron microscopy techniques (i.e. imaging plastic-embedded sections) (Bermudes et al., 1994). More recently, fluorescence light microscopy (fLM) has shown that many bacterial proteins localize in filament-like patterns. FtsZ, MreB and crescentin are most likely to be among those that form true filaments, because all three are structurally related to known filament-forming eukaryotic proteins (Amos et al., 2004) and all three form filaments in vitro. FtsZ is a tubulin homologue that organizes the cell division machinery into a ring at the division plane (Oliva et al., 2004). MreB is a structural homologue of actin involved in forming rod-like cell shapes and in segregating genomes (van den Ent et al., 2001; Gitai, 2005). Crescentin shows strong similarities to eukaryotic intermediate filaments, and is responsible for the characteristic crescent shape of Caulobacter crescentus (Ausmees et al., 2003).
While fLM has revealed that each of these proteins localize in elongated, filament-like patterns in special positions within cells, its resolution is too limited to provide molecular details. Electron cryotomography (ECT) is an emerging technology that can reveal the three-dimensional structure of small, intact cells or thin regions of larger cells in a near-native state (Kürner et al., 2004). Indeed, tomographic techniques have already revealed cytoskeletal filaments in a few bacterial species (Izard et al., 2004; Kürner et al., 2005; Komeili et al., 2006; Scheffel et al., 2006). In order to characterize the bacterial cytoskeleton in greater depth we imaged the widely studied model prokaryote C. crescentus using state-of-the-art ECT. C. crescentus is a favourite model system for studies of cell shape, establishment of polarity, genome segregation and cell division. The cells can be easily synchronized, facilitating investigations of their asymmetric cell cycle (Viollier and Shapiro, 2004). Importantly, C. crescentus is the only bacterial species known at this time to contain homologues of all three major classes of cytoskeletal filaments (actin-related, tubulin-related, and intermediate filament-related), as represented by MreB, FtsZ and crescentin (respectively).
Several different filament bundles exist in exponentially growing cells
Log-phase C. crescentus cultures were plunge-frozen across electron microscope grids. Individual cells were imaged repeatedly in a 300 kV, field-emission-gun, energy-filtered transmission electron cryomicroscope while being incrementally tilted about one or two axes (Iancu et al., 2005) (see Experimental procedures and Movie S1). Three-dimensional reconstructions (tomograms) were then calculated from the series of tilted images (see Movie S2). Forty log-phase, wild-type CB15N cells were imaged, but only the thirty tomograms in which the vast majority of the intact cell was present were used for calculating the frequency of occurrence of each filament type.
The reconstruction in Fig. 1A is typical of our results: the surface layer, the inner and outer membranes, and even the peptidoglycan layer between the membranes are all clearly visible. As in all the reconstructions, the cytoplasm was full of dense, ∼25 nm particles, the majority of which are likely to be ribosomes. Large, typically spherical granules likely composed of poly-β-hydroxybutyrate were also present. The length of each cell and the occasional presence of flagella, pili, stalks and mid-cell constrictions were used as landmarks to determine the approximate cell cycle stage of each cell.
The most surprising feature of the cell shown in Fig. 1 was, however, the bundle of filaments lying just inside the inner membrane on the concave side of the cell at approximately its midpoint. In three dimensions, the filaments resembled a stack of three ribbons, each ∼390 nm long, ∼7 nm thick and ∼22 nm wide. The ribbons were stacked ∼10 nm apart (all spacings reported are measured from the centre of one filament to the centre of the next) with a much shorter, fourth ribbon on the ‘top’ face nearer the mid-line of the cell. The ribbons rested flat against the inner membrane, following its broad curvature along the cell's long axis (Fig. 1A and B). When projected 35 nm along an axis parallel to their widths, a periodicity of ∼8 nm became apparent (Fig. 1C). Here and below, gross measurements such as length and width are approximate because each ribbon in the stack varied slightly. Measurements of small features such as filament thicknesses are approximate because noise and the size of each voxel in the final reconstructions (2.4 nm) obscured finer detail. Similar bundles were seen in six other cells (23% of the total) where the ribbons had remarkably similar structures and proportions, but were found at differing distances from the poles (Fig. 2). This pattern of filaments was seen in swarmer, stalked and dividing cells, but no correlation was seen between filament positions or shape and cell cycle progression. We will refer to these filaments as ‘inner curvature ribbons’.
In other cells, two additional types of filament bundles were seen that resembled the inner curvature ribbons in that they also appeared in stacks of ribbon-like densities with similar proportions. We interpret them here as distinct types because their positions within the cell were different in ways that would be critical for function, and because of subtle differences in dimension and straightness.
One of these additional bundle types (the second overall) typically occurred in groups of three to five ribbons, each 5 to 6 nm thick and spaced ∼11 nm apart (Fig. 3). This bundle was the most prevalent of the four described here, and was seen in 13 of the cells (43%). Because filaments of this type were found in the cytoplasm away from any membrane, we refer to them as ‘cytoplasmic ribbons’. They were seen at various distances from the poles for each of the different cell cycle stages in which they occurred (swarmer, stalked, predivisional and divisional), arguing against any cell cycle-dependent pattern of occurrence. Their lengths were quite variable, ranging from ∼230 to ∼560 nm, and they aligned roughly with the long axis of the cell.
A third filament bundle type was seen in just two cells (7%) and consisted of a group of three or four ribbons, each 4 to 5 nm thick and ∼20 nm wide, spaced ∼9 nm apart (Fig. 4). In both cases the ribbons were found near a cell pole, so we refer to them as the ‘polar ribbons’. These ribbons followed a slightly curved path for a distance of ∼230 nm along the inner membrane towards the opposite end of the cell. In contrast to the previous two types, these bundles appeared to be directly attached to the membrane at or near the pole; there were some discontinuities within individual ribbons; and instead of being smooth and flat across their widths, the ribbons seemed corrugated, perhaps reflecting the subunit spacing of individual component filaments within each ribbon. These differences cannot be explained by differences in orientation compared with the cytoplasmic ribbons because some of the cytoplasmic filaments were oriented similarly to the polar ribbons. For example, the cytoplasmic ribbons in Fig. 3 were actually at a similar angle to the xy plane as the polar ribbons shown in Fig. 4.
The fourth (and last) type of filament bundle was observed in just one cell (nominally 3%), and consisted of 10 filaments arranged with a pyramid-shaped cross section forming an arch just inside the inner membrane on the convex side of the cell (Fig. 5). The individual filaments were ∼5 nm in diameter, hexagonally packed ∼12 nm apart (Fig. 5B), and lay within a plane perpendicular to the long axis of the cell. A couple of the filaments were over 200 nm long, but their full length was unclear because the anisotropic point spread function blurred their ends (Fig. 5C). The bundle was found two-thirds the length away from the stalked pole of a predivisional cell, adjacent to what appeared to be the earliest beginning of a constriction. These filaments did not form a complete ring around the cell, but rather were visible on only one side. Nevertheless we shall refer to this formation as the ‘ring-like bundle’, as the filaments were arranged in a strikingly regular partial ring around the cell.
In some cases, two bundle types were seen in the same cell: both inner curvature and polar ribbons were seen individually with cytoplasmic ribbons. Polar ribbons were not seen in cells with inner curvature ribbons; however, the cell containing the ring-like bundle did not appear to have any other filament types.
Filaments persist in crescentin knockouts and A22-treated cells
Suspecting that some of the observed bundles might be composed of one of the previously identified cytoskeletal proteins in C. crescentus (crescentin, FtsZ, or MreB), where possible we imaged cells where these filaments were expected to be absent. (No adequate labelling methods exist yet for ECT.) While FtsZ and MreB are essential for cell function and thus have no viable knockouts, crescentin knockout strains are viable and have been described in the literature as having a straight rather than curved phenotype (Ausmees et al., 2003). The small molecule A22 causes depolymerization of MreB filaments (Gitai et al., 2005). Thus we imaged crescentin knockout cells with and without A22 treatment, as well as wild-type cells treated with A22. The inner curvature ribbon cannot be either crescentin or MreB, because it appeared clearly in crescentin knockouts (Fig. 6A and B) and A22-treated cells (Fig. 6C and D). Likewise, the cytoplasmic ribbon cannot be either crescentin or MreB, because it appeared in crescentin knockouts treated with A22 (Fig. 6E and F). Curiously, we have not seen inner curvature ribbons in the 10 A22-treated crescentin knockout cells that we have imaged so far, suggesting perhaps that there is a complex interaction between MreB, crescentin and the inner curvature ribbons. Although we did not see polar ribbons and the ring-like bundle in these experiments, because their frequencies were so low in wild-type cells we cannot judge whether or not they were eliminated in the knockout cells.
Through direct imaging with state-of-the-art ECT, we have discovered what appear to be four distinct types of cytoskeletal filament bundles in log-phase C. crescentus: inner-curvature, cytoplasmic, polar and ring-like (see Fig. 7 and Movie S3 for 3-D views of the different filaments and their locations within the cell). When compared with existing fLM results for crescentin, FtsZ and MreB, there were both intriguing consistencies and puzzling inconsistencies. The position and path of the inner curvature ribbons seemed to be the same as the fLM localization for crescentin, for instance, but their limited frequency (they were seen in only about a quarter of the cells) and length (only a third to a quarter of the cell length) were different. Ultimately their presence in crescentin knock-out cells demonstrated conclusively that this bundle was not composed of crescentin. Likewise, the shape and position of the ring-like bundle was consistent with fLM localizations for either FtsZ or the particular configuration of MreB found in predivisional cells (when MreB collapses into a ring at the cell's future constriction site) (Figge et al., 2004); but instead of being a complete ring as expected, the bundle was seen only on one side of the cell and the observed interfilament spacing and overall arrangement of the filaments did not match anything seen in vitro for FtsZ or MreB (Lowe et al., 2004). The polar and cytoplasmic ribbons were unanticipated by existing fLM results.
Thus two obvious questions arise: first, what are these filaments; and second, where are FtsZ, MreB and crescentin? Concerning the first question, the inner curvature, cytoplasmic and polar ribbons must be as-yet unidentified cytoskeletal elements. By imaging crescentin knockout strains and A22-treated cells, we showed that at least the inner curvature and cytoplasmic ribbons are neither crescentin nor MreB. The polar ribbons are also unlikely to be crescentin or MreB because their extent and localization do not seem to match fLM results for either of these proteins. Likewise, none of these three ribbons match the localization pattern of FtsZ known by fLM. While the overall similarity of all three ribbon bundles suggest that they could be composed of the same protein, subtle (but reproducible) differences and their distinct localization patterns indicate that they could be different proteins.
We speculate, however, that the ring-like bundle is FtsZ, and is actually our first view of that structure in vivo. We favour the interpretation that this is FtsZ rather than MreB because the cell seems to be just beginning to constrict near the filament, a function known to be associated with FtsZ. Nevertheless much more information will be needed to confirm this interpretation and to track the development of the structure throughout cell division. Curiously, no similarly prominent filament bundles were seen here in more deeply constricted cells (Fig. 2A, for instance), in agreement with the results from a similar study targeting dividing C. crescentus cells (Judd et al., 2005).
Concerning the second question (the location of the known cytoskeletal proteins), we have speculated above that the ring-like bundle is FtsZ. While it is a possibility that resolution limitations are preventing us from seeing crescentin and MreB, this is unlikely because the filaments we did see had diameters similar to the known diameter of MreB. Perhaps MreB and crescentin filaments are small, highly curved or do not function as bundles and thus were not immediately recognizable by the simple visual inspection done here. It is also possible that fLM results exaggerate the extent to which real filaments actually exist in the cell. Filaments like crescentin or MreB could be highly dynamic like FtsZ (Anderson et al., 2004), for instance. If so, perhaps only short segments actually exist within the cell at any given moment. If depolymerized subunits remained even somewhat localized in the vicinity of the transient filament, they would still produce the apparently constant and elongated signal seen by the lower-resolution fLM. Consistent with this idea, numerous thin lines of density were seen here that could have been short single filaments, and we are now developing computational searches to explore their locations and statistical significance. One report claims to have visualized MreB (Kürner et al., 2005), but in a different context. The authors suggested that together with the fibril protein, MreB formed the spiral bundle they saw in Spiroplasma melliferum involved with motility. While the structure of MreB in C. crescentus may not necessarily be the same, nothing resembling such clear and continuous helical filaments immediately inside the cell membrane were seen here.
To gain more information, three lines of investigation are now being pursued. In addition to the computational searches mentioned above, various overexpression, deletion, and functional mutants are being imaged. Correlated fluorescent light microscopic and electron cryotomographic investigations of individual cells are also in progress. Nevertheless these first results have already revealed that large filament bundles exist within C. crescentus, and the bacterial cytoskeleton is more complex than previously thought.
CB15N cells were grown at 30°C in M2G minimal media with glucose as the carbon source (Ely, 1991) and synchronized according to published protocols (Tsai and Alley, 2001). One millilitre of aliquots was removed from the synchronized culture at 0, 15, 30, 50, 70 or 90 min post synchrony, gently pelleted and rinsed in M2 salt solution, then resuspended in 30–50 μl of supernatant. The cells were plunge-frozen as soon as possible after pelleting, but in some cases were kept on ice for several minutes until the freezing procedure could be finished. A 3X-concentrated solution of the 10 nm colloidal gold was applied to glow-discharged, 200 mesh, copper/rhodium grids with R2/2 carbon-on-film Quantifoil™ (Quantifoil Micro Tools, Jena, Germany) and the grids allowed to dry. A 5X-concentrated solution of 10 nm colloidal gold (Ted Pella, Redlands, CA) was added to the cells immediately before plunge freezing. A 4 μl drop of the gold-and-cell solution was manually applied to one side of the prepared grids and the grid automatically blotted and plunged into liquid ethane using a Vitrobot (FEI Company, Hillsboro, OR). Frozen grids were stored under liquid nitrogen until use, and kept below −165°C during loading and data collection.
Crescentin knockout strains
Crescentin knockout strains (CJW 763 and CJW 1208) were grown in peptone yeast extract (PYE) medium (Ely, 1991) at 30°C until they reached an optical density at 600 nm between 0.2 and 0.4. Two millilitres of the cell suspension were gently pelleted, the pellet rinsed in M2 Salt solution and resuspended in 30–50 μl of supernatant. After adding the colloidal gold solution the cells were immediately plunge-frozen as described above. This strain has been sequenced previously and shown to definitely not contain crescentin (Ausmees et al., 2003), and in our experiments appeared straight as expected.
10 μg ml−1 of A22 (ChemBridge) dissolved in 100% methanol was added to 2 ml of the log-phase cells grown in PYE medium and stored on ice for 5 min, conditions shown previously to disrupt MreB filaments (Gitai et al., 2005). The cells were spun down gently, resuspended in 30–50 μl of supernatant, combined with the colloidal gold and immediately plunge-frozen as described above. Cells in parallel cultures were seen by light microscopy to bulge and distort into lemon shapes 6–8 h after A22 treatment as expected, confirming the activity of our A22.
Images were collected using an FEI Polara™ (FEI Company, Hillsboro, OR, USA), 300 kV FEG transmission electron cryomicroscope operating in energy-filtered mode (slit width 20 or 30 eV), on a Gatan Ultrascan CCD camera (Gatan, Pleasanton, CA) such that each of the 2048 × 2048 pixels represented 1.2 nm on the specimen plane. Image tilt-series were acquired semiautomatically at 12 μm underfocus with the predictive UCSF-Tomo software package (Zheng et al., 2004) using a cosine-based dose-partitioning algorithm and 1° or 0.5°−0.7° tilt increments for dual-tilt (Iancu et al., 2005) or single-tilt data sets, respectively, over tilt ranges of at least ±60°. All of the figures use single tilt data. After the sensitivity of the CCD camera was carefully calibrated, the cells were seen to endure exposures in excess of 200 e–/Å2 without any visible structural damage, so this total dose was used to image each cell, delivered either in one single tilt-series or split between two orthogonal series.
Following removal of outlier pixels created by X-ray events, tilt-series were binned twofold, normalized, aligned and used to calculate three-dimensional reconstructions using the IMOD software package (Mastronarde, 1997). Reconstructions were low-pass-filtered to the resolution of the first zero of the CTF (∼1/4.9 nm−1), but no other ‘de-noising’ was used. Filaments were found and measured visually using IMOD. Figures were produced with IMOD and Amira (Mercury Computer Systems).
The authors thank Christine Jacobs-Wagner for the crescentin knockout strains and discussions and L. Shapiro and Z. Gitai for discussions. This work was supported in part by NIH Grants P01 G66521 and R01 AI067548 to G.J.J., DOE Grant DE-FG02–04ER63785 to G.J.J., a Searle Scholar Award to G.J.J., a grant from the Danish Natural Sciences Research Council to R.B.J., and gifts to Caltech from the Ralph M. Parsons Foundation, the Agouron Institute, and the Gordon and Betty Moore Foundation.