Traditional textbook representations of the prokaryotic cytoplasm show an amorphous, unstructured amalgamation of proteins and small molecules in which a randomly arranged chromosome resides. The development and application of a swathe of microscopic techniques over the last 10 years in particular, has shown this image of the microbial cell to be incorrect: the cytoplasm is highly structured with many proteins carrying out their assigned functions at specific subcellular locations; bacteria contain cytoskeletal elements including microtubule, actin and intermediate filament homologues; the chromosome is not randomly folded and is organized in such a way as to facilitate efficient segregation upon cell division. This review will concentrate on recent advances in our understanding of subcellular architecture and the techniques that have led to these discoveries.
If you can't find a tool you're looking for, please click the link at the top of the page to "Go to old article view". Alternatively, view our Knowledge Base articles for additional help. Your feedback is important to us, so please let us know if you have comments or ideas for improvement.
The study of microbial cell architecture is not new, but the recent discovery that microbial cells contain complex cytoskeletal elements just like more ‘advanced’ eukaryotes has provided important insights into how microbial cells are organized. This review will cover how microscopic techniques have increased our understanding of the subcellular organization of microbial cells, and will also present some personal predictions on what the future holds.
One of the first things to address is size: where does architecture begin and end? In this review, I shall briefly cover structures that range in dimension from encompassing most of the cell that can be visualized by light or fluorescence microscopy to much smaller molecular complexes detected by electron microscopy (EM). All of these macromolecular complexes are relevant to the structure of the cell and to the organization of the cytoplasm, and thus contribute to the subcellular architecture of the cell. The bulk of these studies have been performed with model microbial organisms Bacillus subtilis, Caulobacter crescentus and Escherichia coli but many of the findings are likely to be highly conserved through the prokaryotes.
Nucleoid structure and organization
Although microbial genomes vary widely in size, structure and complexity, many bacteria contain a single circular chromosome that must be compacted approximately 1000-fold to fit in a cell. In addition, different parts of the cell cycle overlap so that multiple rounds of chromosome replication may be ongoing as the cell divides. It seems obvious then that microbial genomes need to be highly organized to ensure that accurate chromosome segregation coordinated with cell division during such complex cell cycles.
Typical negative stain electron micrographs of microbial cells revealed the nucleoid to be a lighter (less densely) stained area located within the middle of the cell. Very little structure was apparent, although a reconstruction of the nucleoid made from cryo-substituted cells revealed a central core with numerous domains projecting out from the centre towards the cytoplasmic membrane (Hobot et al., 1985). The multiple domains projecting from the central core were predicted to contain transcriptionally active regions of the genome, permitting simultaneous transcription, translation and transertion of membrane proteins into the cytoplasmic membrane, which in turn could drive chromosome segregation (Woldringh and Nanninga, 1985). However, the bacterial nucleoid is a very fragile structure that collapses and condenses on fixation, particularly with the strong fixatives that are often used in the preparation of samples for transmission EM. The use of intrinsically fluorescent protein fusions has allowed us to visualize whole nucleoids, subnucleoid regions and specific genetic loci in live cells, and has provided important information on the dynamic organization of the nucleoid. Nevertheless, we still know very little about anything more than the gross structure of the nucleoid in live cells.
In a typical rod-shaped microbial cell (such as E. coli or B. subtilis), the nucleoid appears to be a roughly oval-shaped structure in slow growing or resting cells and bi-lobed in exponentially growing cells (e.g. Sharpe et al., 1998). This structure appears the same using DNA-specific dyes such as 4,6-diamidino-2-phenylindole (DAPI) or fluorescent-tagged DNA-binding proteins such as histone-like proteins (Shellman and Pettijohn, 1991; Kohler and Marahiel, 1997) and RNA polymerase (Lewis et al., 2000; Cabrera and Jin, 2003). The analysis of the finer structure of the nucleoid in unfixed cells is limited by the resolution of the light microscope (1–200 nm), although the nucleoid probably fills most of the living cell, with the DNA concentration being greatest at the middle of the nucleoid, and decreasing in concentration towards the cytoplasmic membrane.
Analysis of chromosome organization and segregation in live bacteria has not involved great leaps in imaging technology, but has relied on the development of genetic tools to enable the dissection of chromosome organization. Information on the organization of chromosomes in live cells came through an adaptation of an ingenious approach first used in yeast, where tandemly repeated lac operators (lacO) were inserted into a specific genetic locus where they could be subsequently decorated by ectopically expressed fluorescently labelled LacI (Robinett et al., 1996). The use of this system, combined with time-lapse imaging of bacteria on agarose pads supplemented with media, allowed the precise marking of loci at any region within the chromosome and the subsequent mapping of their dynamic movement throughout the cell cycle. Thus, regions adjacent to oriC, the terminus terC situated diametrically opposite oriC, and the regions in between were mapped in live B. subtilis ( Telleman et al., 1998). Origins were the most polar-located loci in exponentially growing cells, whereas termini were located at mid-cell, and other loci were located at a position between oriC and terC commensurate with their position within the chromosome. This conservation of the linear organization of genes in 2D DNA space and 3D cell space has also been observed using fluorescent in situ hybridization and fluorescent protein-labelling in E. coli and C. crescentus. (Jensen and Shapiro, 1999; Niki et al., 2000; Lau et al., 2003).
One of the problems with using the lacO cassettes, which contain 256 tandemly arranged lacO sites, was that the large number of repeated sequences predisposed the region to recombination events. A recent paper by Lau et al. (2003) has addressed this issue by redesigning the cassettes so that the sequence between the operators is randomized, leading to a substantial increase in the stability of chromosomal inserts. The authors also refined this approach to enable double labelling of the loci within the cell by combining a tet repressor system with the lac system (Lau et al., 2003). Thus, lacO sites can be decorated with LacI-ECFP, and tetO with TetR-EYFP fusions (Fig. 1A). The simultaneous visualization of two separate loci allows more precise correlation of the position of a locus within the cell at particular points of the cell cycle.
Viollier et al. (2004) have taken advantage of this improved system to obtain highly detailed spatial information on the organization of the C. crescentus chromosome by mapping the location of 112 different loci. The loci were distributed approximately equally on either side of oriC, and using an automated image processing programme developed in-house, the position of the loci relative to cell length and fluorescently tagged oriC for several hundred cells from each strain were mapped to obtain average locations. These data confirmed the linear arrangement of these loci along the long axis of the cell with a resolution of 125–250 kb. The rate of segregation of 10 loci was also examined in detail, and this revealed that the loci are segregated chronologically in the order they are replicated, and that this segregation occurs rapidly at a rate similar to that reported for oriC regions. This compelling data suggests that there is an active segregation mechanism that operates on most or all loci to ensure efficient partitioning and maintenance of organization of the chromosome.
In B. subtilis, chromosomal origin regions (oriC regions) appear to be organized into a large (approximately 400 kb) domain whose formation is dependent on the ParA/ParB-like Soj/Spo0J protein pair (Glaser et al., 1997; Marston and Errington, 1999; Quisel et al., 1999). Spo0J is a DNA-binding protein that binds to eight parS sites distributed on either side of oriC (Lin and Grossman, 1998), and the role of Soj is to organize Spo0J–parS complexes into a large oriC supercomplex (Marston and Errington, 1999; Quisel et al., 1999), although it is also a transcription regulator (Cervin et al., 1998). The precise mode of activity of Soj in the organization of the oriC supercomplex is still unclear as it binds a region of DNA corresponding to approximately one genome equivalent, and it is not present on every chromosome in every cell (unless the cell carries a spo0J mutation). The precise role of Spo0J is also unclear but it was thought to be important in chromosome segregation. Certainly, Spo0J has a role in segregation of oriC regions, but there is only a mild vegetative cell segregation phenotype in its absence (Ireton et al., 1994), and it does not appear to be involved in bulk chromosome segregation (below). Despite the lack of a chromosomally encoded ParA/ParB system in E. coli, origin regions still appear to be arranged into a large domain as determined by fluorescence in situ hybridization experiments (Niki et al., 2000). The requirement for origin regions to condense into bacterial centromeric structures is probably a generally conserved property, and the role these structures play in chromosome segregation has been covered in other recent reviews (see Sherratt, 2003; Gerdes et al., 2004).
The study by Lau et al. (2003) also indicated that there was a cellular-positioning mechanism involved in coordinating chromosome segregation and cell division (Fig. 1B and C). Thus, the origin regions first, then DNA replication factories, then cell division proteins and finally terminus regions assemble at the mid-cell position (in a slow growing cell undergoing a single round of DNA replication). Interestingly, the authors found that the terminus regions were frequently positioned asymmetrically within the cell, and that the assembled cell division apparatus (as visualized with an FtsZ-YFP fusion) acted as a physical barrier to the movement of the terminus to a mid-cell position (Lau et al., 2003). In some cases, the termini segregated into daughter cells before cytokinesis, but in many instances, one terminus was observed to segregate and pass through the closing septum at a late stage of cytokinesis. This phenomenon of postseptational DNA segregation was first observed during the highly asymmetric segregation event that involves the transfer of DNA into the small prespore compartment at the onset of sporulation in B. subtilis and is caused by the activity of a protein called SpoIIIE (Wu and Errington, 1994; Wu et al., 1995). SpoIIIE is expressed throughout the normal vegetative growth cycle (Foulger and Errington, 1989) and is a highly conserved protein, similar to E. coli FtsK (Begg et al., 1995). It was subsequently shown that SpoIIIE/FtsK localizes to the septum in dividing cells and is important in the segregation of terminus regions that become trapped in the closing septum (Sharpe and Errington, 1995; Steiner and Kuempel, 1998; Lemon et al., 2001). However, it was not clear how FtsK was able to segregate sister chromosomes into the correct compartment during this event. The asymmetry reported by Lau and coworkers, combined with the proposed physical block to terminus movement imposed by the cell division ring, would provide directionality to FtsK, ensuring that the septal region was clear of DNA on completion of cell division (Fig. 1B and C; Lau et al., 2003). This is likely to be a rare event in wild-type cells, but this indicates that specific mechanisms have evolved to ensure the preservation of chromosome integrity even on trapping of the nucleoid in a closing septum. Strong evolutionary pressure in order to develop such a fail-safe mechanism was probably imposed by the overlapping nature of the replication and division cycles in bacteria.
The oriC region appears to be organized into a large domain that is segregated very early in the replication cycle. The terminus region is designed to ensure faithful segregation even if DNA becomes trapped in a closing septum. Is the nucleoid subdivided into additional functional domains? In organisms capable of rapid growth such as B. subtilis and E. coli, RNA polymerase (RNAP) becomes concentrated into specific subnucleoid regions termed transcription foci (TF), which represent sites of rRNA transcription, at mid- to high-growth rates (Lewis et al., 2000; Cabrera and Jin, 2003). Both of these organisms contain multiple genes encoding rRNAs (rrns) (B. subtilis has 10 rrns per chromosome and E. coli has 7). In addition, the rrns tend to be clustered close to oriC, although they are distributed over about half of the chromosome in both organisms. It was considered possible that rrns are clustered into TF to form the prokaryotic equivalent of a nucleolus, but dual-labelling studies in B. subtilis involving the colocalization of Spo0J (equivalent to oriC domains) with LacI-labelled rrns, indicated this was not the case. Instead, TF resulted from the high level-loading of RNAP onto the rrns (7 out of a total of 10) that were situated within, and close to the oriC domain (Davies and Lewis, 2003). Thus, in bacteria, there is little evidence for dynamic redistribution of chromosomal loci to form functional subnucleoid sites involved in the synthesis of specific subclasses of RNA, and so preservation of the linear organization of loci within the genome is likely to indicate the importance of this arrangement to ensure efficient chromosome segregation.
Now we know that the DNA replication machinery is organized into large replication factories. During a single round of replication, these factories are located at the mid-cell in B. subtilis and E. coli (Lemon and Grossman, 1998; Lau et al., 2003), but migrate from polar to mid-cell positions during the replication cycle in C. crescentus (Jensen et al., 2001). The localization of DNA replication to a single subcellular domain has led to the hypothesis that the physical act of DNA synthesis provides directionality for DNA segregation that is essential for the bulk movement of DNA during the replication cycle (Lemon and Grossman, 2001). In general, it had been assumed that DNA synthetic enzymes would move bidirectionally around the chromosome during replication as the replication machinery is small in size compared to the chromosome. However, chromosomes are relatively flexible and malleable polymers, whereas the DNA replication machinery represents a complex comprising a PolC–PolC or PolC–DnaE leading–lagging strand complex with associated lagging strand clamp loaders, primase, helicase, single-stranded binding proteins and many other enzymes. It was also assumed that DNA replication simply required a single leading and lagging strand polymerase per replication fork, but the ease with which factories could be visualized within the cell using GFP-tagged PolC indicated that the number of polymerase molecules in the factory region was far higher than had been suspected (Lemon and Grossman, 1998). It is possible that additional polymerases are required to aid in recombination repair DNA synthesis. This is supported by the fact that RarA, which is similar to γ-τ and the Holliday junction branch migration protein RuvB, is a replication factory marker in E. coli (Lau et al., 2003).
Further work using a yeast 2-hybrid screen also showed that a large number of additional proteins interacted with DNA polymerase and other replication proteins, indicating that replication factories could represent huge subcellular biosynthetic sites (Noirot-Gros et al., 2002). Subsequent screening of the replication factory interacting proteins using GFP fusions showed that many of these proteins did not colocalize with replication factories. However, this approach will not identify protein subpopulations that could be involved in these interactions and not all the interacting protein localization patterns have been reported yet (Meile et al., 2003). There is still much to be learnt about replication factories, for example, how closely juxtaposed are the two replication forks, what proteins are located in replication factories, what the DNA polymerase stoichiometry is in replication factories, and what the structural organization of replication factories is. Clearly, they appear to be sites of high recombinatorial activity as well as DNA synthesis. Recent data also suggest that although the two replication forks are located close together, they probably do not form a replication fork superfactory, as the two forks can be periodically resolved during the replication cycle (Migocki et al., 2004).
Therefore, despite the approximately 1000-fold condensation of DNA within the cell, the 2D genetic organization of the chromosome is preserved in 3D cytoplasmic space and, although we do not have high resolution information on the conformation of the prokaryotic chromosome, it is abundantly clear that these huge molecules are highly organized spatially within the cell.
Cytoskeletons: tubulin, actin and intermediate filaments
One of the most unexpected findings to emerge from the microscopic analysis of bacteria over the last few years was the discovery that bacteria contain cytoskeletons with homologues to all the eukaryotic cytoskeletal elements. Thus, bacteria contain a microtubule (tubulin) protein called FtsZ, multiple actin-like proteins including MreB and Mbl, and intermediate filament proteins such as CreS in C. cresentus, indicating the very ancient origin of these types of intracellular filaments (Fig. 2; Rothfield et al., 1999; Jones et al., 2001; Ausmees et al., 2003; Kruse et al., 2003). In eukaryotes, microtubules are involved in various transport processes and in forming the mitotic spindle during chromosome segregation, actin forms a dynamic cytoskeleton involved in cell-shape regulation and mobility as well as the cytokinetic ring during cell division, and intermediate filaments provide mechanical support to the cell and nucleus.
Image deconvolution has been used to great effect to improve image quality and allow us to build 3D images of bacterial cytoskeletal structures. Traditionally, the technique involved calibrating each objective in identical media to that used for data acquisition to determine the point spread function (Agard et al., 1989). In simple terms, light focused on a particular plane, when viewed from above that point, will scatter both above and below the point of focus, spreading the signal and thus reducing the resolution of the image (this phenomenon is called point spread). The point spread function allows data to be processed to remove all light above and below the plane of focus, leaving a higher resolution image of only in-focus signal. Manual determination of point spread functions is an extremely tedious process, and most integrated acquisition/processing programmes used these days employ algorithms that calculate a theoretical point spread function from a series of parameters provided by the user (such as emission wavelength, objective magnification, refractive index and objective numerical aperture). In addition to processing an image to remove out-of-focus light, 3D reconstruction and deconvolution require the acquisition of a stack of images in the Z plane moving through the cell. Many confocal systems and specialized widefield systems include integral Z (or focus) drives so that a series of images can be obtained in defined steps. These systems have been designed with eukaryotic imaging in mind and are often unable to move in steps of less than 100 nm. This is too large for many bacteria, which are about 800 nm thick, as only 8–10 in-focus images would be obtained in a Z series. Nose-mounted Z drives and integral drives included in some of the new generation fluorescent microscopes permit Z steps of as little as 10 nm, although the margin of error is likely to be quite large under such demanding conditions. Generally, good results can be obtained using 20–50 nm steps.
Other factors become significant issues when obtaining large Z series for deconvolution and reconstruction. When using immunostained cells, samples are often mounted in a medium that contains antifading agents that mop-up free radicals generated by the excitation light that quench fluorescence signal. Using antifading agents has not been reported yet with live samples containing fluorescent protein fusions. Nevertheless, the image intensity of a sample is progressively reduced as a result of photobleaching during the acquisition of an image stack, which can significantly affect the quality of processed data. This can become a major problem when acquiring image stacks at 20 nm steps that may contain 60 or more images. The shortest exposure that gives a good signal above the background should be used to minimize quenching, and some programmes contain a function to account for photobleaching and also lamp flicker to help improve data quality. Finally, deconvolution can be an extremely time-consuming process, and it is very important to ensure that the data is of the highest possible quality before embarking on further image processing. Selecting a single cell, or a small field also speeds up the deconvolution process and also simplifies subsequent analysis of reconstructed images. Analysis of cytoskeletal proteins has lent itself well to techniques like image deconvolution.
Because of the small size of prokaryotes, there appears to be little requirement for microtubule transport systems similar to those found in eukaryotic cells. Instead, the tubulin homologue FtsZ is essential for cell division, forming a cytokinetic ring at mid-cell early during the division process (Fig. 2A; Errington et al., 2003). In association with a number of other cell division proteins, FtsZ then constricts, concomitantly with septal peptidoglycan synthesis to bisect cells and to form two new daughters. Using image deconvolution and 3D reconstruction, it has been possible to look at FtsZ and other division protein structures in vivo (Fig. 2A). When looked at from the side, FtsZ appears to form two spots opposite each other at mid-cell, or a band bisecting the cell (Fig. 2A, top). However, when the reconstruction is rotated towards the viewer, it can be seen that ring structures, called Z-rings have formed (Fig. 2A, bottom). Very occasionally cells are oriented on a slide so that they are end-on rather than side-on. In these situations, Z-rings can be observed without the need for image processing and serve to confirm the accuracy of the processing of the stacks of side-on cells.
We know much about the phenomenology of cell division, such as when Z-rings assemble and what proteins interact with them to form a division complex, but much remains to be elucidated: although we know the molecular structure of FtsZ and how it relates to that of tubulin (Löwe and Amos, 1998), it appears that FtsZ does not form tubules like tubulin (see Carballido-López and Errington, 2003a). We do not know how FtsZ polymers are structurally organized at the division site or the structural nature of the interactions FtsZ forms with other cell division proteins. As we do not know the structure of the FtsZ cytokinetic ring, we can only speculate as to the precise mechanism of cell division (see below), and this remains an important area that is the focus of considerable research effort in a number of laboratories around the world.
The role of actin-like proteins is much more diverse than tubulin-like proteins within prokaryotes. The actin superfamily is very large, and includes many proteins not known to polymerize, such as hexokinase (Bork et al., 1992). Traditionally, the shape of bacteria was thought to be determined by the peptidoglycan cell wall, as removal of the cell wall in isotonic solutions led to the formation of sphere-shaped protoplasts from rod-shaped cells (Henning et al., 1972). Many of the proteins involved in cell-shape determination were enzymes involved in peptidoglycan, teichoic acid or teichuronic acid synthesis. However, a few of the cell-shape determinant mre genes were not biosynthetic enzymes but were members of the actin superfamily (Bork et al., 1992). The localization of two such proteins, MreB and Mbl was investigated in B. subtilis using immunofluorescence and fluorescent protein fusions, and led to the startling discovery that both of these proteins formed helical filaments (Fig. 2B; Jones et al., 2001). MreB and Mbl do not interact to form a hybrid filament but each protein forms a filament of different pitch and length, so that MreB forms a short (0.73 ± 0.12 µm) pitch filament that assembles around the mid-cell position, whereas Mbl filaments have a longer pitch (1.7 ± 0.28 µm) and traverse the entire cell length. MreB was shown to have a role in the maintenance of cell width, whereas Mbl was involved in maintenance of the long axis of the cell. In C. crescentus, MreB appears to play a direct role in controlling cell shape by acting as an organizer for a PBP2–peptidoglycan synthesis complex involved in cell elongation (Figge et al., 2004), although it does not appear to play such a role in B. subtilis (Scheffers et al., 2004). It now seems that MreB also plays a role in chromosome segregation, which is severely perturbed on deletion or removal of MreB from E. coli, B. subtilis and C. crescentus cells (Kruse et al., 2003; Soufo and Graumann, 2003; Gitai et al., 2004). A plasmid RI-encoded homologue of MreB called ParM has also been shown to form cytoskeletal filaments and to segregate the plasmid through the dynamic formation of ParM filaments, rather like an actin polymerization process (Møller-Jensen et al., 2002; 2003). These actin-like proteins comprise a family called the type II ParA-like ATPases (Gerdes et al., 2000; see below). Moreover, MreB from C. crescentus has been shown to be important in the determination of cell polarity, ensuring that a range of proteins involved in stalk biogenesis, cell division, DNA replication and flagellar assembly localize to the correct cell pole during development (Gitai et al., 2004). The pleiotropic effects of these actin-like proteins on cell shape, division, polarity, plasmid and chromosome segregation lend further weight to the proposal that they represent true actin homologues that form a structural scaffold within the cell on which a variety of other proteins assemble to carry out specific activities at particular locations.
The proteins MreB and Mbl are highly dynamic, with a variety of localization patterns being observed in all the systems listed above, including filaments spanning the cell length, mid-cell breaks presumed to represent segregation of filaments before cell division and, particularly in C. crescentus, constriction of filaments to a mid-cell ring (Carballido-López and Errington, 2003b; Shih et al., 2003; Figge et al., 2004). Deconvolution microscopy has been used to great effect to visualize these filaments, but it is extremely difficult to obtain very high resolution images of the filaments because of the small size of microbial cells. In addition to the reported filament dynamics, MreB filaments can often be observed as a single cable-like filament spiralling along the cell. However, this sometimes appears to be a continuous filament that runs both up and down the cell length, rather like Mbl (Fig. 2B). The structure of the filament has important implications for the mechanism of action of this protein and warrants further investigation.
Less is known about intermediate filaments, although they appear to be a diverse group of unrelated proteins that carry out (possibly) related structural roles in the cell. The recent report by Ausmees et al. (2003) that cell shape in C. crescentus is dependent on a prokaryotic intermediate filament CreS should result in the rapid accumulation of further information on the dynamics of these filaments. Intermediate filaments in C. crescentus appear to be closely juxtaposed with the cell membrane, and are responsible for conferring the vibrioid shape (C. crescentus lacking CreS filaments are rod-shaped). CreS filaments represent true architectural structures within the cell, as their conformation dictates cell shape, and they do not serve as scaffolds for proteins orchestrating a host of other functions like MreB (Gitai et al., 2004). In exponentially growing cells, C. crescentus does not form chains, but grows as differentiating pairs and CreS forms a curved filament along the inside edge of the vibrioid cells (Fig. 2C, top). However, filamentous cells in which CreS filaments could be observed to run the length of the cell, appear in old stationary phase cultures (Fig. 2C, bottom). Using image deconvolution and 3D reconstruction, CreS filaments were shown to have a left-handed pitch and to follow the shortest route along a helical filament, a property that might cause the cell to curl into a vibrio (Fig. 2C; Ausmees et al., 2003). Assuming this is the case, one would imagine that there would be some form of close contact between the cytoplasmic CreS filaments and the cell wall.
In addition to containing cytoskeletal equivalents to microtubules (FtsZ), actin cytoskeletons (MreB family and type II ParA ATPases) and intermediate filaments (CreS), a fourth type of dynamic cytoskeletal structure has also now been described in bacteria that is involved in both cell division and chromosome partitioning. Type I ParA ATPases (Gerdes et al., 2000) form a large superfamily with a series of subclasses that are involved in plasmid and chromosome partitioning (ParA-like proteins) and cell division site selection (MinD-like proteins). This family of proteins, although capable of performing a plasmid partitioning function similar to type II ParA ATPase actin-like proteins, bears no resemblance to other known cytoskeletal proteins. Crystallographic studies showed that MinD from Pyrococcus furiosus was most similar in structure to the nitrogenase iron protein (NIP) which is an Fe:S-binding ATPase component of the nitrogen-fixing enzyme nitrogenase (Hayashi et al., 2001). Earlier time-lapse studies on ParA protein Soj from B. subtilis and MinD from E. coli showed that these proteins form patches over nucleoids (Soj) or at cell poles (MinD) which oscillated across the cell over a period of seconds (MinD) to minutes (Soj) (Hu and Lutkenhaus, 1999; Marston and Errington, 1999; Quisel et al., 1999; Raskin and de Boer, 1999; Autret et al., 2001). The oscillatory characteristics of Soj were difficult to interpret as it was not present in all cells, although it was shown to be important in the condensation of widely distributed Spo0J–parS complexes (see above). The oscillatory role of MinD appears to be to prevent FtsZ ring formation at sites other than the mid-cell region, as over a time-averaged period, it was present at mid-cell for the least amount of time, thereby permitting Z-rings to form at the site of minimal MinD division inhibitor concentration (see Meinhardt and de Boer, 2001). Microscopic examination of these dynamic proteins gave no indication that higher-order structures were being formed. However, recent studies have shown that both E. coli MinD and ParA from the E. coli plasmid pB171 form highly dynamic cytoskeletal-like filaments within the cytoplasm that are distinct from actin-like filaments (Fig. 2D; Shih et al., 2003; Ebersbach and Gerdes, 2004). These important discoveries would not have been made without the aid of image deconvolution as it is impossible to define any evidence of filamentous structures from unprocessed images of pB171 ParA (see Fig. 2D, top; Ebersbach and Gerdes, 2004). Because of the highly dynamic nature of these proteins, it was not possible to obtain data for deconvolution from live cells, and fixatives had to be used. However, the possibility that the helical filaments arose as a result of an artefact of fixation could be dismissed in the case of MinD as careful examination of micrographs revealed evidence of helical structures in both fixed and non-fixed cells (Shih et al., 2003). Upon image processing, spiral-like filaments could be clearly observed and, in the case of ParA from pB171, the filaments appear to spiral around the outer edge of the nucleoid, forming a more concentrated patch at the end (Fig. 2D, bottom; Ebersbach and Gerdes, 2004). Time-course analysis of non-fixed cells showed that localization was dynamic and oscillatory with ParA moving from one end of the nucleoid to the other resulting in the positioning of pB171 over the mid-point of the nucleoid. Therefore, it seems that the oscillatory activity of these proteins arises through formation and turnover of highly dynamic filaments that span the cell.
MinD and ParA are located in different clades of the dendogram produced by Gerdes et al. (2000) and thus, are distantly related in both sequence similarity and function. It is remarkable therefore that MinD from B. subtilis, which is more similar to MinD from E. coli than is pB171 ParA, does not oscillate within the cell but instead forms stable caps at the ends of cells (Marston et al., 1998). No examination by deconvolution of these static MinD structures has been published to date, but clearly it would be of interest to determine whether the static polar MinD caps in B. subtilis also represent filamentous structures or not. Nevertheless, the localization of DivIVA that recruits MinD to cell poles in B. subtilis was shown to form rings and discs but not helices (Harry and Lewis, 2003). Perhaps, in addition to recruiting it to the cell poles, DivIVA also inhibits an intrinsic oscillatory function of MinD.
Image deconvolution of both MreB and Mbl enabled direct visualization of cytoskeletal filament structure within the cell and, in particular, of filament pitch and length (Jones et al., 2001). These filaments appear to behave in a very similar way to F actin, and Mbl filaments have been shown to cycle with similar kinetics (Carballido-López and Errington, 2003b). This analysis was carried out using a technique called fluorescence recovery after photobleaching (FRAP), a technique commonly used to analyse protein dynamics in eukaryotic cells. FRAP has also been used with great success to determine the dynamics of FtsZ ring assembly (Stricker et al., 2002). In bacteria, FRAP requires the use of live cells containing a functional fluorescent protein fusion, as it is not possible to microinject fluorescently labelled proteins into such small cells, although fluorescently tagged HU has been introduced into E. coli using chemical shock (Shellman and Pettijohn, 1991). Firstly, using a highly focused laser in conjunction with either a widefield (Stricker et al., 2002) or confocal microscopy system (Carballido-López and Errington, 2003b), a region within a cell is bleached using high intensity laser pulses. The sample is then monitored at regular intervals for recovery of fluorescence within the bleached region. Image analysis can then be used to quantify the recovery and the rates of protein turnover determined. Using this approach, Mbl filaments were shown to turn over with a half-life of 510 ± 90 s (Carballido-López and Errington, 2003b), whereas FtsZ filaments turn over with a much shorter half-life of only 31.7 ± 3.4 s (Stricker et al., 2002). These analyses help us to understand the composition of the structures seen under the microscope; it is not possible to determine with light microscopy whether the observed structures represent single or multiple filaments, or even tubes, or whether the filaments are continuous or discontinuous. FRAP analysis has helped formulate testable hypotheses to help address these issues.
Mbl (and MreB) filaments have to equipartition upon cell division, and clear breaks generating two equipartioned filaments have been observed in dividing B. subtilis and E. coli cells (Carballido-López and Errington, 2003b; Shih et al., 2003), indicating that the composition of these structures is highly dynamic. Detailed analysis of Mbl filament dynamics indicated that they were not continuous structures, but most likely comprise a series of overlapping non-polar protofilaments that form a single, apparently continuous filament. For example, FRAP experiments with filamentous cells indicate that fluorescence recovers at the same speed from both sides of a bleached region (Carballido-López and Errington, 2003b). Such a model is supported by the fact that there are approximately 13 000 molecules of Mbl per cell, which would be sufficient to form a single filament that could span the cell a number of times (Jones et al., 2001). Indeed, nearly all of the fluorescence attributable to Mbl appears to be in the filaments, indicating that most of the protein is in a polymerized form at any one time (Carballido-López and Errington, 2003b). The small cytoplasmic pool is proposed to be present as protofilaments that can dynamically exchange with the large filamentous structures in a similar manner to actin cable dynamics in yeast (Carballido-López and Errington, 2003b).
The dynamics of Z-rings have also been examined by FRAP (Stricker et al., 2002). Only about 30% of the total available FtsZ is incorporated into a Z-ring (Stricker et al., 2002), and a relatively high background level of cytoplasmic FtsZ can be seen in cells containing GFP fusions or on immunofluorescence of wild-type cells (e.g. Fig. 2A; Stricker et al., 2002). FRAP analysis showed that FtsZ turnover in the Z-ring via exchange with the cytoplasmic pool was extremely rapid, with the half-life for recovery being 31.7 ± 3.4 s in predivisional cells, and 35.6 ± 14.2 s in dividing cells with a visible division furrow (Stricker et al., 2002). Thus, FtsZ filaments remain highly dynamic throughout the entire cell division cycle. Furthermore, the analysis of a rare end-on cell, from which the viewer was able to focus directly down the long axis of the cell onto the Z-ring, indicated that there was no directional cycling of the filament, but that exchange occurred at multiple random locations around the ring (see Fig. 5 in Stricker et al., 2002). These analyses prompted the authors to reaffirm their belief in a model of isodesmic assembly of Z-rings from protofilaments. The concentration of FtsZ within the cell is approximately 11 µM and at these concentrations the 15 000 molecules of FtsZ within the cell will spontaneously assemble into around 180 GTP-dependent protofilaments, each comprising an average of 80 subunits. A Z-ring represents about 30% of the total FtsZ, and thus would comprise a structure of approximately 56 overlapping protofilaments that would be able to dynamically exchange with the remaining 124 cytoplasmic protofilaments (Stricker et al., 2002). However, additional data suggest the importance of cooperative FtsZ filament assembly and it is unlikely that cytoplasmic FtsZ is mainly in the form of such large protofilaments (Caplan and Erickson, 2003). Microscopic examination of Z-ring dynamics indicates that rings assemble extremely rapidly and that the cytoplasmic pool of FtsZ appears free to diffuse throughout the cell (Margolin, 1998). Also, the free diffusion within the cytoplasm of FtsZ protofilaments comprising an average of 80 subunits with a molecular weight of approximately 3.2 MDa seems unlikely given the data of Elowitz et al. (1999) showing that proteins much larger than about 500 kDa are not free to diffuse within the cytoplasm. Therefore, any cytoplasmic FtsZ protofilaments must also be highly dynamic structures that oscillate rapidly between free monomers or small oligomers that are free to diffuse throughout the cytoplasm, and larger protofilaments that are incorporated into the Z-ring. A detailed discussion of FtsZ polymerization dynamics has been recently reviewed (Romberg and Levin, 2003). In B. subtilis, the increase in production of FtsZ at the beginning of sporulation may be enough to shift the equilibrium to formation of larger oligomeric assemblies, resulting in FtsZ filaments that spiral towards the cell poles at this developmental stage (Ben-Yehuda and Losick, 2002).
Electron tomography and other 3D analysis techniques
Light microscopy has proven extremely useful in the analysis of many new aspects of microbial cell biology, and has shown itself to be a rapid, robust, sensitive approach in addition to being ideally suited for time-lapse analysis of dynamic events within the cell. However, the approach is limited by the resolution that can be obtained, and many structures within the microbial cell are not particularly amenable to light microscopic analysis.
Over the past few years, 3D analysis techniques have been refined using electron microscopic techniques (electron tomography). Briefly, the sample is mounted on a stage including a goniometer. The stage can be precisely tilted within the electron beam so that images of the sample can be obtained over a range of tilt angles. These stages may tilt through either one or two planes; single- or double-tilt tomography (Penczek et al., 1995). The 2D projection images can then be processed to obtain high resolution 3D data on intact cells. This approach is still in its infancy in microorganisms, although some impressive results are now beginning to be published and the technique clearly holds great future potential.
Despite the recent interest in microbial cytoskeletons, it has been known for many years that various microorganisms contain a wide range of cytoplasmic filaments (e.g. Bermudes et al., 1994) and many proteins can be induced to form filaments under suitable in vitro conditions or when over-expressed in vivo (e.g. Beck et al., 1978; Okada et al., 1994). The spirochete bacteria Treponema sp. contain a filamentous ribbon structure comprising 2–9 filaments adjacent to the cytoplasmic membrane composed of the CfpA protein (Fig. 3A; Hovind-Hougen, 1976; Izard et al., 1999). These filaments do not confer the spiral shape on the cells, but have been implicated in cell division and chromosome segregation, although CfpA is not similar to the tubulin or actin superfamilies (Izard et al., 2001). Nevertheless, reconstruction of electron tomographs revealed features strikingly similar to other cytoskeletal proteins (Fig. 3A). The filaments were 5–6 nm in diameter and spaced about 10 nm apart (Izard et al., 2004). A number of proteinaceous structures were observed to form interfilament bridges, and interaction with the cytoplasmic membrane was also seen (Fig. 3A, bottom; Izard et al., 2004). The association of numerous proteins with cytoskeletons is a common feature in eukaryotes and, although the identity of the bridging proteins is not known, this is an encouraging result as other studies have so far failed to identify significant protein interaction with structures such as MreB filaments (J. Errington, pers. comm.), although there is growing circumstantial evidence to suggest that such proteins must exist (e.g. Møller-Jensen et al., 2003; Gitai et al., 2004).
Tomography has also been used to analyse stationary-phase chromosome structures that are adopted by bacteria to protect fragile DNA polymers during periods of dormancy. Paracrystaline toroidal structures formed between the DNA and Dps protein, and intermediates involved in their formation were studied in E. coli (Frenkiel-Krispin et al., 2004). Image deconvolution techniques have been used previously to analyse the structure of chromosomes in spores that must survive many years in often harsh conditions, and these also form toroids associated with stable acid soluble proteins (SASPs) in B. subtilis (Sharp and Pogliano, 2002). Therefore, in addition to the association with small protein ‘protectants’, the formation of highly compact toroids may be a universal approach to preserve the genetic integrity of a cell during dormant periods in both spore and non-spore forming bacterial cells.
At what point does cell architecture end? Tomography presents a bridge between the analysis of individual macromolecules observed by very high resolution electron microscopy, such as single particle analysis, and gross structural analysis by light microscopy. Ribosomes are huge particles (2.5 MDa) with many associated cofactors that form polysomes with other translating ribosomes on mRNA, and thus represent substantial subcellular structures. As a result of cotranscriptional and translational coupling, polysomes are also connected via mRNA to transcription elongation complexes creating a physical link between the architecture of the nucleoid and cytoplasm. Likewise, enzymes such as pyruvate dehydrogenase are present as huge multimeric subcellular complexes that, according to the data of Elowitz et al. (1999), would be unable to diffuse through the cytoplasm. These are well known examples of large, multicomponent complexes of bacteria but there might be many more, such as the cytoplasmic σB stress regulator complex from B. subtilis (Chen et al., 2003a). Recent developments with large scale proteomic screens to identify the plethora protein–protein interactions within the cell suggest that large but possibly unstable macromolecular complexes might be very common indeed (Sali et al., 2003; Aloy et al., 2004). Therefore, it also seems appropriate to consider subcellular architecture with reference to the macromolecular composition of the cytoplasm using techniques that are briefly discussed below.
Single particle reconstruction is a relatively new transmission electron microscope technique that is used to make 3D reconstructions from 2D images (Frank, 2002). Images of negatively stained, or more preferably, cryo-fixed samples spread on a grid are classified according to their orientation. To ensure maximum preservation of the structure of these very delicate complexes, very low radiation doses are used in image acquisition (10e Å−2). Because of the very low signal to noise ratio, images within a particular class are stacked and averaged to reduce the background and increase the average signal. Average images from multiple classes in different orientations can then be used to obtain a density projection that gives 3D structural information on the samples. This approach has been used with great success to analyse ribosomal complexes and ribosome dynamics during the translation cycle and in the analysis of eukaryotic transcription complexes (see Frank, 2002). The replication factory represents an ideal subject for single particle reconstruction as the atomic-resolution structure of many of its components has been determined, and thus, these structures could be modelled into the cryo-EM density maps to build up a detailed structure of this molecular machine.
Scanning force microscopy has also been used with some success to analyse biological complexes, particularly since the development of techniques to image samples in liquid under quasi-physiological conditions (see Bustamante et al., 1997). Although a great deal of information can be obtained on nucleoprotein complexes (Dame et al., 2003), this approach is relatively low resolution, and it is unlikely that individual proteins in complex mixtures could be identified with confidence. However, a recent development for the higher resolution analysis of protein complexes is that of tapping mode atomic force microscopy (AFM). In standard AFM, an ultrafine carbon cantilever is used rather like a stylus from an old-fashioned record player and is dragged over the surface of an object. Structural information can be extrapolated from the amplitude changes and are transmitted by the cantilevers. In tapping mode AFM, the cantilever is made to vibrate so that, rather than being dragged across an object, it taps across it; a process that increases resolution and allows delicate protein complexes to be analysed (Stark et al., 2001). This approach has recently been used with impressive results to probe the structure of the τ-helicase complex of DNA polymerase (Haroniti et al., 2004). The authors showed that the stoichiometry of the complex was τ5-helicase6, and that the τ pentamer complex was arranged rather like a set of horns on the donut-shaped helicase hexamer. Thus, tapping mode AFM presents an alternative to more traditional approaches to understanding the structure of multiprotein complexes.
It seems that we stand on the doorstep of some great developments in microscopy that will enable us to visualize what is happening within cells from whole cell resolution right down to the level of molecular complexes by non-invasive techniques (see below). Nevertheless, significant hurdles remain, and it is important to recognize the limitations of certain approaches. The use of model systems such as E. coli, B. subtilis and C. crescentus has shown how important the understanding of the cell cycle is to various analyses (e.g. Sharpe et al., 1998; McAdams and Shapiro, 2003; Sherratt, 2003). If we do attempt to build up an understanding of the architecture of cells at the molecular level, it is important to bear in mind the stage of the cell cycle that is being analysed and to interpret data accordingly, as we are now beginning to appreciate just how dynamic individual proteins and macromolecular structures are within the cell in processes, such as DNA replication, cell division and cell shape regulation.
Light microscopic techniques such as time-lapse acquisition, image deconvolution and fluorescence recovery are being used to great effect to shed light on aspects of chromosome segregation, cell division and cytoskeletal organization (see above). The increasing adoption of these techniques indicates that many more data will be forthcoming over the next few years. However, fluorescence resonance techniques that have been used extensively in eukaryotes have yet to be widely adopted in microorganisms. Fluorescence resonance energy transfer (FRET) and a related approach, fluorescence lifetime imaging (FLIM) are techniques that can be used to analyse protein–protein interactions both in vivo and in vitro (Chen et al., 2003b). Briefly, if two proteins tagged with fluorophores, where the emission spectrum of fluorophore A overlaps the excitation spectrum of B, are situated very close together (<10 nm), some of the emission energy of A will be transferred non-radiatively to B, causing it to fluoresce. Thus, at excitation wavelengths for A, there should be a decrease in fluorescence emission of A and a resulting increase in fluorescence of B where FRET occurs. This drop in fluorescence of A and increase in fluorescence of B can be directly measured. The development of spectral variants of GFP opened the way to the use of fluorescent proteins for analysing protein–protein interactions and protein dynamics in situ (Miyawaki et al., 1997). FRET has been used to show the interaction of two regulatory proteins ScpA and ScpB with the chromosome condensing protein SMC in B. subtilis (Mascarenhas et al., 2002). However, the simple theory of the technique is difficult to apply in live cells for the following reasons: (i) fluorescent proteins have very large excitation and emission spectra, so images have to be extensively corrected for crossover fluorescence resulting from fluorescence of YFP at CFP excitation wavelengths and vice versa (Youvan et al., 1997; Gordon et al., 1998; Xia and Liu, 2001); (ii) fluorescent proteins are not particularly bright fluorophores and FRET is inefficient, so extremely small changes in fluorescence intensity need to be accurately detected; and (iii) in non-fixed cells, movement of fluorophores or the plane of focus between exposures will affect the data. Nevertheless, if appropriate controls are performed, it is possible to obtain good data. An alternative to FRET, which might circumvent these problems, involves protein fragment complementation similar to that of α-complementation of β-galactosidase used in blue-white clonal selection. Fluorescent proteins can be dissected into non-fluorescent fragments that reform a functional composite fluorescent protein on reassociation (Ghosh et al., 2000; Hu et al., 2002). Thus, if proteins containing these fusions interact, the fluorescence detected in situ delineates sites of interaction (Hu et al., 2002). There is no reason to believe that such an approach could not be used with success in bacteria as well, and this technique might also be a useful method to analyse protein complexes in situ.
Can higher resolution electron microscopic analysis be used on fully hydrated vitrified cells? Electron tomography clearly offers a way to initiate the analysis of cytoplasmic organization at the molecular level (see above). A further refinement of this technique is to cryo-freeze samples rather than chemically fix them (electron cryo-tomography; Grimm et al., 1998; Baumeister, 2002; Medalia et al., 2002; Plitzko et al., 2002). This approach involves the rapid freezing of samples (vitrification) in liquid nitrogen or ethane, thereby preventing the formation of ice crystals and fixing cellular components in space and time without the need for chemical fixatives that cause artefacts through their chemical activities. This technique has yielded very impressive results in eukaryotic systems in such applications as the structural analysis of regions of the cytoplasm (Medalia et al., 2002). In addition, approaches are being refined to enable the analysis of cytoplasmic organization down to the macromolecular level (Fig. 3B; Böhm et al., 2000). Crystallographic structures can be rotated through all possible orientations to fit electron cryo-tomographic density maps, making it possible to identify three structurally similar, but different macromolecular complexes from in vitro mixtures (Fig. 3B; Böhm et al., 2000; Plitzko et al., 2002). The ability to analyse cytoplasmic structure at this level represents one of the holy grails for cell biologists. Although great strides are being made in this area, a number of issues prevent the widespread application of this technique for the time being. Firstly, the protein concentration of the bacterial cytoplasm is extremely high (Grimm et al., 1998); so high that it is rarely possible to obtain sufficient contrast using electron microscopy to identify subcellular structures. Secondly, because of the delicate nature of fully hydrated cryo-fixed samples, great care must be taken not to expose samples to too high an electron dose (<2000–5000 e nm−2; Plitzko et al., 2002) compounding the problems of obtaining sufficient image contrast to identify structures/molecules and a sufficient number of images to create suitably high resolution 3D images (Penczek et al., 1995; Plitzko et al., 2002). The approach is also limited to samples < 1 µm thick without access to even more specialized cryo-sectioning equipment (Plitzko et al., 2002). Although the limits of sample thickness may not pose a great problem for many microbial systems, it may be necessary to treat cells to remove outer membranes, flagella, etc. so that structures within the cytoplasm can be observed at higher contrast (see Izard et al., 2004).
And what of situations where there will be insufficient image contrast to obtain informative tomographs? An area where tomography would be of considerable use is in the investigation of the molecular organization of the cytokinetic ring, and in particular the structure of FtsZ filaments as we still do not understand the mechanism of Z-ring constriction. Because of the very low contrast of Z-rings in the cytoplasm (no division rings have been reported in bacterial tomographs so far), it is likely that further technological developments will be required before we can analyse these structures with confidence in situ. One possible solution could be the combination of electron cryo-tomography with the phenomenon of cathodoluminescence in which photons emitted through excitation by the electron beam are detected (Thiberge et al., 2004). Thus, fluorescence from fluorescent protein fusions could be detected at an order of magnitude higher resolution than is possible with a light microscope, although signal brightness and autofluorescence from cytoplasmic proteins may represent an initial hurdle to be overcome (Thiberge et al., 2004).
It seems that we are increasingly moving towards an integrated approach to understanding cytoplasmic organization and subcellular architecture, with no single method being able to provide all the necessary details. Combinatorial approaches have now been developed whereby the structural organization of large multiprotein complexes isolated by TAP-tag procedures can then be analysed by electron microscopy and molecular modelling (Aloy et al., 2004). This information can subsequently be used to position these complexes within the cytoplasm using tomography (Sali et al., 2003). While such approaches are extremely challenging computationally, methodology is being refined while computer processing power continues to increase, and such analyses will become more widespread in the future as more groups embrace multidisciplinary approaches to structural analyses.
In conclusion, the future is bright for the investigation of subcellular architecture, and in the coming years, we can expect to learn a great deal more about microbial cytoskeletal dynamics and structure, chromosome organization and structure, and the macromolecular organization of the cytoplasm. A feast for the eyes indeed!
I would like to acknowledge the support of the ARC through grants A00105184 and DP0449482 to my laboratory. Because of the wide scope of this review, I have had to be selective in dealing with topics and apologize to all those whose work has not been included in this review that has contributed to the various fields discussed. Thanks to Nora Ausmees, Wolfgang Baumeister, Jeff Errington, Kenn Gerdes, Anton Hillebrand, Jaques Izard, Christine Jacobs-Wagner, Dave Sherratt and Joerg Sievers for permission to use and the supply of figures for this review. I would also like to thank Gerry Wake, Shelagh Martin, Geoff Doherty, Tony Pugsley and my reviewers for helpful comments on the preparation of this manuscript.