Bacterial chemotaxis depends on signalling through large protein complexes. Each cell must inherit a complex on division, suggesting some co-ordination with cell division. In Escherichia coli the membrane-spanning chemosensory complexes are polar and new static complexes form at pre-cytokinetic sites, ensuring positioning at the new pole after division and suggesting a role for the bacterial cytoskeleton. Rhodobacter sphaeroides has both membrane-associated and cytoplasmic, chromosome-associated chemosensory complexes. We followed the relative positions of the two chemosensory complexes, FtsZ and MreB in aerobic and in photoheterotrophic R. sphaeroides cells using fluorescence microscopy. FtsZ forms polar spots after cytokinesis, which redistribute to the midcell forming nodes from which FtsZ extends circumferentially to form the Z-ring. Membrane-associated chemosensory proteins form a number of dynamic unit-clusters with mature clusters containing about 1000 CheW3 proteins. Individual clusters diffuse randomly within the membrane, accumulating at new poles after division but not colocalizing with either FtsZ or MreB. The cytoplasmic complex colocalizes with FtsZ at midcells in new-born cells. Before cytokinesis one complex moves to a daughter cell, followed by the second moving to the other cell. These data indicate that two homologous complexes use different mechanisms to ensure partitioning, and neither complex utilizes FtsZ or MreB for positioning.
Recent advances in bacterial cell biology have dramatically increased our appreciation of the exquisitely organized and dynamic subcellular architecture of bacterial cells (Gitai, 2005). Many macromolecular complexes occupy specific subcellular locations and exhibit dynamic behaviours (Rudner and Losick, 2010; Govindarajan et al., 2012), therefore, understanding spatial dynamics is essential to understand biological processes in bacteria. Chemotaxis and cell division are very different cellular activities but both are accomplished by the concerted actions of several molecular complexes that localize at specific sites in the cell (de Boer, 2010; Sourjik and Armitage, 2010). At cell division each daughter cell must inherit a complement of chemosensory proteins to allow efficient chemotaxis, meaning positioning of the chemosensory proteins must co-ordinate with cell division and it has been suggested that sophisticated mechanisms have evolved to orchestrate the spatial regulation of cell division and chemotaxis (Sourjik and Armitage, 2010; Ringgaard et al., 2011).
Chemotaxis is one of the best-studied model systems of signal transduction. The Escherichia coli chemotactic machinery has been extensively characterized (Sourjik and Armitage, 2010; Sourjik and Wingreen, 2012). The chemotaxis pathway starts with transmembrane chemoreceptors, which detect chemoeffectors and transmit signals to regulate the autophosphorylation of the cytoplasmic histidine kinase, CheA. The adaptor CheW facilitates the interactions between CheA and receptors. CheA transfers phosphoryl groups to the response regulator CheY. CheY-P released from the receptor complex diffuses to the flagellar motor and promotes a switch in the rotational direction of flagella (Sourjik and Armitage, 2010; Sourjik and Wingreen, 2012). Thousands of chemosensory proteins, including receptors, CheA and CheW, form large membrane clusters at the cell poles and smaller clusters along the cell body (Briegel et al., 2009; Sourjik and Armitage, 2010). Fluorescence microscopy and cryo-electron tomography suggest a preferential polar/subpolar localization of these clusters in bacterial species studied so far (Briegel et al., 2009; Sourjik and Armitage, 2010; Porter et al., 2011).
Three mechanisms have been proposed for the polar localization of membrane chemosensory clusters mainly based on the studies of E. coli and Vibrio cholerae. The stochastic self-assembly model suggests newly synthesized proteins either join existing clusters or nucleate new ones (Thiem and Sourjik, 2008; Wang et al., 2008; Greenfield et al., 2009). In this model, cluster nucleation is distance-dependent and away from existing clusters. This results in a periodic distribution of clusters along the cell body, with the cluster size decreasing in the order of old-pole, new-pole, midcell, quarters and so on. This mechanism ensures an even partitioning of membrane chemosensory clusters during cell division (Thiem and Sourjik, 2008; Wang et al., 2008; Greenfield et al., 2009). Cluster partitioning may involve anchoring the lateral clusters to pre-cytokinetic sites such that they are at new poles after cytokinesis (Thiem et al., 2007). Movement of polar clusters is also restricted by unknown mechanisms, possibly the specific membrane curvature or lipid composition of the cell pole (Rudner and Losick, 2010). An alternative helical insertion diffusion–capture model suggests that newly synthesized chemoreceptors are inserted in the membrane in a helical fashion (Shiomi et al., 2006), and then migrate to and form large clusters at the poles. A third ParC-mediated diffusion–capture model occurs in some polar-flagellated bacteria where a ParA homologue (see below), ParC, has a cell cycle-dependent, unipolar to bipolar localization pattern (Ringgaard et al., 2011; Yamaichi et al., 2012). ParC actively recruits chemosensory proteins to the cell poles before cytokinesis, ensuring every new-born cell has an old-pole cluster after cell division.
The chemosensory systems in the α-proteobacterium Rhodobacter sphaeroides represent a higher level of complexity, reflecting its physiological versatility (Porter et al., 2008; 2011; Sourjik and Armitage, 2010). It is an emerging model organism for studying signal transduction complexity found in many bacteria and more complex cells (Porter et al., 2011). R. sphaeroides has three major operons encoding chemosensory proteins, two of which are expressed under laboratory conditions. The chemosensory proteins of these two pathways form an E. coli-like conventional membrane-associated pathway with membrane-spanning chemoreceptors (Mcps), and a second novel cytoplasmic pathway with soluble chemoreceptors (transducer-like proteins, Tlps) (Porter et al., 2008; 2011; Sourjik and Armitage, 2010). R. sphaeroides requires both pathways for chemotaxis. Separation of chemosensory proteins may enable tuning of responses to independently sensed external and internal conditions. The metabolic versatility of R. sphaeroides is also accompanied by the differentiation of cell morphology and subcellular architecture (Slovak et al., 2005; Mackenzie et al., 2007). When growing photoheterotrophically, R. sphaeroides develops numerous invaginations of the cytoplasmic membrane (Niederman, 2006; Niederman, 2010; Tucker et al., 2010), and change their morphology from rod-shape to coccobacillus (Slovak et al., 2005). The dramatic rearrangements of the cytoplasmic membrane pose a great challenge for maintaining subcellular architecture. The eukaryote-like behaviour of intracytoplasmic photosynthetic membrane (Niederman, 2010) makes R. sphaeroides a promising model to study spatial regulation of membrane proteins during membrane differentiation.
The cytoplasmic chemosensory receptors localize as clusters at the R. sphaeroides midcell. After duplication, clusters appear equi-positioned (Thompson et al., 2006), a pattern reminiscent of that seen for certain plasmids (Gerdes et al., 2010; Lutkenhaus, 2012). Partitioning of cytoplasmic chemosensory clusters requires homologues of plasmid and chromosome DNA partitioning proteins PpfA (ParA homologue) and TlpT (ParB analogue), probably using the nucleoid as a platform to ensure daughter cells each inherit a cytoplasmic cluster (Thompson et al., 2006; Roberts et al., 2012). There is growing evidence that ParA homologues are involved in the spatial regulation of several other protein complexes, including polar membrane chemosensory clusters in certain bacterial species (Lutkenhaus, 2012). Strikingly, current studies demonstrate an important role of the nucleoid in ParA-mediated plasmid and protein complex partitioning (Gerdes et al., 2010; Jain et al., 2012; Lutkenhaus, 2012; Roberts et al., 2012), making it possible that the partitioning of cytoplasmic chemosensory clusters is co-ordinated with cell division.
The tubulin homologue FtsZ is the most critical component of the bacterial cytokinetic machinery (divisome) (Adams and Errington, 2009; de Boer, 2010; Erickson et al., 2010). FtsZ is found in nearly all bacteria, many archea and chloroplasts. FtsZ forms a ring-like structure (the Z-ring) at cytokinetic sites to establish a scaffold that sequentially recruits other divisome members, then constricts to initiate cytokinesis. In vitro, FtsZ subunits assemble into protofilaments that can form higher-order structures (Erickson et al., 2010). However, the in vivo ultrastructure of the Z-ring is still unclear. The formation and positioning of the Z-ring is co-ordinated with chromosome movement (Thanbichler and Shapiro, 2006; Adams and Errington, 2009; de Boer, 2010; Erickson et al., 2010). In addition to FtsZ, most rod-shape bacteria contain the actin homologue MreB (Shaevitz and Gitai, 2010). MreB in R. sphaeroides localizes to the midcell, moving to quarter positions before the next round of cytokinesis (Slovak et al., 2005).
Previous studies have suggested that chemosensory proteins in different species may use FtsZ, MreB or ParA-like systems to position the membrane cluster at cytokinetic sites or at specific poles. Our previous data have indicated that a ParA-like system positions the cytoplasmic cluster of R. sphaeroides to the chromosome and suggested connections between the cytokinetic site and the spatial regulation of cytoplasmic chemosensory clusters. Here, we studied the localization of FtsZ, MreB and the two chemosensory clusters in R. sphaeroides relative to each other during the cell cycle under different growth conditions. We found that FtsZ moves from the new pole to the midcell after cell division and, unlike E. coli, the membrane chemosensory clusters do not colocalize with FtsZ, but form large diffusing dynamic complexes in the membrane. Our data also suggest that these large chemosensory complexes are composed of unit-clusters. The cytoplasmic chemosensory clusters show more defined positioning, but again are independent of both FtsZ and MreB. Despite the considerable knowledge about FtsZ accumulated during the past two decades, it is still unclear how the Z-ring forms. This study also allowed us to develop a model for Z-ring formation in R. sphaeroides, which may be applicable to other bacteria.
Spatiotemporal dynamics of FtsZ during the cell cycle of R. sphaeroides
The spatiotemporal dynamics of FtsZ in R. sphaeroides have not been previously reported. We therefore characterized localization patterns of FtsZ before examining its positioning relative to chemosensory clusters. Since fluorescent protein fusions of FtsZ homologues are not fully functional in any bacterial species studied so far (Erickson et al., 2010), FtsZ was examined in a wild-type strain containing a low-copy inducible plasmid expressing ftsZ–yfp (yellow fluorescent protein). Leaky expression or induction with IPTG (isopropyl-β-d-thiogalactopyranoside) up to 10 μM did not affect cell growth and morphology. The typical ring-like structures (Fig. 1A, 0′) were seen (Table S1). Depending on the cell-cycle stage, the Z-ring was not always a continuous, uniform structure, but frequently showed gaps (Fig. S1A) and heterogeneous distributions (Fig. S1B), consistent with current super-resolution imaging studies (Fu et al., 2010; Strauss et al., 2012). The midcell Z-ring constricted and further invagination produced a pair of daughter cells sharing an FtsZ assembly. The completion of septation split the shared assembly into two independent spots localized at the new poles of the two cells (Fig. 1A, 30′ to 90′), similar to the cell cycle-specific pattern seen in Caulobacter crescentus (Thanbichler and Shapiro, 2006; Aaron et al., 2007).
A controversial aspect of current models of Z-ring constriction is whether or not FtsZ disassembles extensively and delocalizes from the Z-ring (Lan et al., 2009; Erickson et al., 2010; Strauss et al., 2012). Quantifying time-lapse images of dividing R. sphaeroides cells (n = 12) showed constant total intensity [0.98 ± 0.05-fold change (mean, with standard deviation used throughout)] of the Z-ring during constriction (from 0 to ∼ 40% decrease in the Z-ring radius), while the Z-ring density increased (12.8 ± 0.8% per 10% decrease in the Z-ring radius), suggesting the protein content remaining constant. The condensation of the Z-ring may provide contractile force for cytokinesis (Lan et al., 2009; Strauss et al., 2012).
We frequently observed noticeable differences in the intensities of FtsZ polar spots in sibling cells (Fig. 1A 90′ and Bi 30′). The relative intensities of polar chemosensory clusters can be used to distinguish between old and young poles (Ping et al., 2008). We followed the intensities of both FtsZ assemblies and polar chemosensory clusters through cell division (n = 20 cell pairs) and found that in 70% (P-value = 0.0184) the sibling cells with smaller polar chemosensory clusters (i.e. younger poles) received more FtsZ molecules after cytokinesis (1.89 ± 0.56-fold, P < 0.00003). This result suggests that although FtsZ content does not link to the ‘absolute age’ of cell poles, FtsZ may have asymmetric inheritance correlated to cellular asymmetry/polarity (Hallez et al., 2004; Macara and Mili, 2008).
The development of the Z-ring
It is unclear how the Z-ring forms (Adams and Errington, 2009; de Boer, 2010; Erickson et al., 2010). After cytokinesis, the polar FtsZ spots redistributed to the midcell and transverse FtsZ spatial gradients extended from the spots (Fig. 1Bi). The transverse FtsZ spatial gradients gradually encircled the midcell plane (Fig. 1Bii). In late-septation and new-born cells FtsZ spots and short fragments were often (Table S1) visible at the future cytokinetic sites (white arrowheads, Fig. S2A), suggesting they are Z-ring precursors. Patterns of spots/fragments at the midcell periphery and time-lapse imaging suggested the Z-ring developed from these precursors, and these precursors were derived from polar spots that had moved to the midcell (Fig. 1B and Fig. S2B). The distribution of FtsZ along the precursor fluctuated before reaching a more symmetric arrangement (Fig. 1C). Intriguingly, the development process reversed in 15.85% cells (Fig. 1Bi, 60′ to 90′, right cell; Fig. S2C). This reverse was not seen for extensively constricted Z-rings, suggesting the development process cannot reverse once the Z-ring matures or constriction has started. The translation inhibitor chloramphenicol has no apparent effect on Z-ring development (Table S1). By quantifying YFP intensity, we found that midcell nodes/Z-ring precursors and mature Z-rings contained a larger percentage of the total FtsZ–YFP molecules than polar spots (Table S1). However, there is a significant pool of FtsZ–YFP molecules that are not localized to visible assemblies at any given time in the cell.
Therefore FtsZ forms at least two kinds of subcellular structures through the R. sphaeroides cell cycle: spots and ring-like assemblies. Right after cytokinesis, FtsZ assembles into polar spots which then gradually redistribute to the midcell and may have dynamic localization before Z-ring formation. The Z-ring appears to initiate at midcell precursor ‘nodes’ (Fig. 1D), from which an FtsZ spatial gradient extends circumferentially. Initially, an asymmetric distribution of FtsZ exists but it eventually encircles the midcell plane with a more symmetric distribution. A correlation was found between the density of Z-ring/ring-precursors and total cellular intensity, but not for polar spots (Fig. S2D). These results suggest that the Z-ring becomes more dense with increasing FtsZ content, in agreement with a loose bundle model (Fu et al., 2010), while the polar spot has a more constant packing density. Hence, polar spots and Z-ring/ring-precursors may have different molecular architectures.
Prior to the formation of mature Z-rings, FtsZ spots are mobile and dynamic over our observation timescale of tens of minutes (Fig. 2). FtsZ spots were able to localize at both cell poles in the same cell (Fig. 2A, yellow arrowheads), raising the question of how FtsZ is positioned in R. sphaeroides. FtsZ spots appeared to integrate/separate and adjust positions (Fig. 2B), resulting in changes in spot intensity (Fig. 2). Due to the resolution limit we could not exclude the possibility that two spots came too close to be distinguished (Fig. 2B, 20′). However, the change in relative intensity (Fig. 2B, 10′ to 30′) suggests they integrated and reformed. This observation indicates that in R. sphaeroides FtsZ is repositioned to future cytokinetic sites via dynamic clusters rather than spiral structures proposed for other bacteria (Ben-Yehuda and Losick, 2002; Thanedar and Margolin, 2004; Thanbichler and Shapiro, 2006; Peters et al., 2007; Specht et al., 2013). However, we cannot exclude the possibility that other configurations of FtsZ assemblies may contribute to Z-ring formation in R. sphaeroides.
Strikingly, FtsZ has similar behaviours in photoheterotrophic cells (Fig. S3), suggesting that the dramatic rearrangement of the cytoplasmic membrane and the change in cell shape which occur during photoheterotrophic growth have no effect on the positioning of FtsZ and the cell division machinery.
Membrane chemosensory protein clusters do not localize at pre-cytokinetic sites
The stochastic self-assembly model for chemosensory clusters predicts that periodic localization and attachment of membrane chemosensory clusters to pre-cytokinetic sites results in new-pole clusters after cytokinesis (Thiem et al., 2007; Thiem and Sourjik, 2008; Wang et al., 2008; Greenfield et al., 2009). To test this in R. sphaeroides, we observed the positioning of membrane chemosensory clusters in cephalexin-treated filamentous cells. The functional YFP fusion of CheW3, expressed as a genomic replacement, was used as a marker for membrane clusters (Wadhams et al., 2003). Unlike E. coli, membrane chemosensory clusters in R. sphaeroides did not colocalize with pre-cytokinetic sites (as marked by FtsZ–CFP), indeed, no detectable membrane clusters were found in the vicinities of Z-rings (Fig. 3A and see below). Occasionally (21% of cells), Z-rings were not positioned at the mid- or quarter-cells, and concomitant exclusions of membrane chemosensory clusters were seen (Fig. 3A, lower panel). These results suggest that although stochastic self-assembly may participate in the positioning of membrane chemosensory clusters in R. sphaeroides, the final positions are not correlated with future cytokinetic sites. Indeed, the membrane clusters appear unable to localize to these sites.
The establishment of new-pole membrane chemosensory clusters
How does R. sphaeroides establish the bipolar localization of membrane chemosensory clusters if the clusters are not targeted to pre-cytokinetic sites? Unipolar localization of membrane chemosensory clusters (Fig. 3B and Fig. S4A) could be seen in some new-born cells, suggesting new-pole clusters form after cell division and cells eventually reach a bipolar pattern. Two possible scenarios are: (i) new-pole clusters are formed via de novo nucleation and assembly, and (ii) new-pole clusters are derived from elsewhere in the cell, with these clusters moving to the new pole. Tracking the membrane chemosensory clusters showed they are dynamic and move randomly in the membrane. Small clusters left old poles and moved along the length of the cell (Fig. 3B). The lateral movement of clusters suggests new-pole clusters can derive from lateral or even old-pole clusters, to become trapped by the curvature at the new pole (Thiem et al., 2007; Endres, 2009). Single-particle tracking showed that lateral chemosensory clusters have a diffusion coefficient D of 4.1 × 10−3 ± 3.0 × 10−3 μm2 min−1, with an average velocity of 17 ± 6 nm min−1, and a maximum of ∼ 80 nm min−1. Individual lateral clusters frequently showed pauses with varied time length (mean = 41.5 ± 28.1 min, n = 34 clusters; Fig. S5A and B). However, the mean square displacement (MSD) versus time interval plots of most trajectories (74%) can be fitted with a straight line (R2 = 0.97 ± 0.03), indicating normal Brownian diffusion. The pauses could be caused by transient encounters with obstacles like cell-wall synthesis machinery.
Inhibition of protein synthesis should impede the formation of new-pole clusters if they are formed de novo, while new-pole clusters derived from the diffusion of existing lateral clusters should be unaffected. Control and chloramphenicol-treated cells had similar ratios of unipolar cells, 32 ± 6% (n = 424) and 30 ± 3% (n = 653) respectively. By following the fate of unipolar cells using time-lapse imaging, we found that within 100 min, 55 ± 2% (n = 197) became bipolar in the presence of chloramphenicol, comparing to 62 ± 4% (n = 82) of control cells. This difference should be much greater if new-pole clusters formed de novo, because there would not be sufficient mature fluorescent chemosensory proteins to form new clusters at the onset of imaging. These results indicate new-pole clusters are derived from old-pole or lateral clusters. Directional movements would not be required for the proposed old-pole/lateral clusters-derived mechanism, since random movements coupled with polar trapping will eventually redistribute clusters to the new pole.
We tested this idea by using a different approach. We followed the behaviour of a chemoreceptor (McpG–GFP) in a ΔcheW2cheW3strain. In both E. coli and R. sphaeroides deletion of CheW results in diffuse chemoreceptors, somewhat more concentrated towards the curved polar regions but without cluster formation, with clusters reforming when cheW2cheW3are re-expressed (Wadhams et al., 2000 and this study). We followed the cluster formation when the expression of cheW2cheW3was induced by IPTG, and found that clusters first formed dispersed around the cell and the clusters then gradually moved to and accumulated at the cell poles (Fig. S6). Therefore, clustering of membrane chemosensory proteins can take place all around the cell membrane, and the resulted chemosensory clusters have a propensity to move towards and accumulate at the cell poles.
Taken together, the above results suggest that in a wild-type scenario, new-pole clusters are most likely derived from existing lateral or old-pole clusters.
FtsZ and membrane chemosensory protein clusters do not colocalize
After cell division the time taken for membrane chemosensory clusters to arrive at the new pole was very variable: from before FtsZ moved away from the new-pole (Fig. 3Bi and Fig. S4A) to after Z-ring formation (Fig. 3Biii and Fig. S4A), suggesting no direct coupling of the redistribution of FtsZ and establishment of new-pole clusters. However, FtsZ polar spots never colocalized with new-pole clusters, although diffraction-limited images occasionally show them to be close (Fig. 3Bi, 20′). Time-lapse imaging showed that membrane chemosensory clusters move into the very tip of the new pole once FtsZ polar spots are no longer present (Fig. 3Bi and iii). Similar relative localization patterns were seen for FtsZ and membrane clusters in photoheterotrophic cells (Fig. S4B). We sought to explore whether this is just a coincidence of the timing of two independent processes. One possibility is that certain concentration of chemosensory proteins is required before a new-pole cluster can form. However, there was no obvious correlation between the amount of chemosensory proteins in the cell and the appearance of new-pole clusters or movement of small clusters from the old pole (e.g. compare Fig. 3Bi to Biii; Fig. S4C). This observation also argues against a purely stochastic self-assembly process in the positioning of membrane chemosensory clusters in R. sphaeroides.
Blocking cytokinesis, which consequently delays the redistribution of FtsZ and formation of new-poles, should accelerate the relative timing of new-pole cluster formation if there is merely a coincidence of timing (Fig. 4, schematics). If the membrane curvature of the cell pole is the main determinant for polar chemosensory cluster localization (Endres, 2009), releasing cells from cephalexin treatment should also accelerate the formation of new-pole clusters (Fig. 4, left part of the schematics). If however the presence of Z-rings/FtsZ polar spots at the forming new pole affects the establishment of new-pole clusters, we should see a time delay (Fig. 4, right part of the schematics). Small membrane chemosensory clusters gradually moved into the new pole only after FtsZ started leaving (Fig. 4; seen in 88.6% of cells, n = 88). The simplest explanation is that chemosensory clusters move from the old poles or lateral sites, diffuse around the cell, but are excluded from the developing new poles by constricting Z-rings. The completion of cytokinesis allows the laterally accumulated clusters to diffuse into the new poles. Moreover, the data support the notion that new-pole clusters are derived from existing clusters moving into the new pole.
From the above data, it seems that the Z-ring inhibits/constrains membrane chemosensory clusters from accumulating at cytokinetic sites (Figs 3 and 4), while FtsZ polar spots might have a different effect (e.g. steric hindrance) at cell poles. Overexpression of FtsZ in R. sphaeroides resulted in randomly distributed spots (Fig. S7B). Close proximity of FtsZ and membrane chemosensory clusters was only observed for FtsZ spots, not for Z-rings (Fig. S7). Accordingly, membrane chemosensory clusters localized as proposed for a stochastic self-assembly pattern (i.e. the ‘default’ pattern) in FtsZ-overexpressing cells (Fig. S7B). This observation indicates that the Z-ring (or associated factors) modifies the localization pattern of membrane chemosensory clusters in R. sphaeroides.
As a huge macromolecular assembly that alters the cell envelope, the Z-ring/divisome may exclude other proteins from its vicinity. We sought to further explore this by following the dynamics of lateral chemosensory clusters relative to the Z-ring in cephalexin-treated cells. The average distance between Z-rings and their nearest lateral clusters is 324 ± 100 nm (n = 92). This result suggests a region covering ∼ 300 nm at both sides of the Z-ring is unfavourable for the localization of membrane chemosensory clusters in R. sphaeroides (Fig. S5A, arrowheads; Fig. S5C).
Membrane chemosensory proteins form dynamic unit-clusters
The polar ‘caps’ formed by membrane chemosensory proteins have been viewed as single large, static entities (Zhang et al., 2007; Ping et al., 2008; Schulmeister et al., 2008; Greenfield et al., 2009). However, our observations suggest that in R. sphaeroides, the polar caps are composed of smaller, dynamic clusters. The polar caps changed their shapes continuously (e.g. Fig. 3Biii) as these small clusters congregate and segregate (Fig. 5A). This suggests there may be a basic unit of membrane chemosensory proteins, and the polar caps or bigger clusters are formed from dynamic congregation–segregation of these unit-clusters (Fig. 5B, schematics). This idea is supported by quantifying the intensities of small clusters over time (Fig. 5B). We observed similar localization patterns for chemosensory receptors McpB and McpG (Fig. S8), suggesting this is not connected to the chemosensory protein fusion used. Therefore, membrane chemosensory proteins formed independently moving dynamic unit-clusters which may have a maximum size.
To estimate the size of unit-clusters, we ranked membrane clusters in snapshots according to their intensities. The distribution of cluster intensity shows a two-phase pattern (Fig. S9A) which can be explained by a unit-cluster model (Fig. 5C), with unit-clusters growing by assembling processes (first linear phase in Fig. S9A), with end of the linear phase representing the average size of unit-clusters. The second exponential phase represents the combination (congregation) of growing and mature unit-clusters (Fig. 5C). It should be noted that the first linear phase does not imply that the unit-clusters grow linearly, since exponential growth with a small increase rate could also result in a good linear fit. Instead, the two-phase pattern indicates the nature of dominant assembling components (e.g. oligomers or unit-clusters).
Since membrane chemosensory clusters are better separated in filamentous FtsZ-overexpressing cells, we performed intensity-ranking analyses on these cells for the average size of unit-clusters by correlating fluorescence intensity to quantitative fluorimetry data (Wilkinson et al., 2011). The data suggest an average of ∼ 1100 CheW3 proteins per unit-cluster. Mapping each cluster to the intensity-ranking curve by its localization reveals striking patterns (Fig. S9A). The first linear phase is mainly composed of lateral and new-pole clusters, while the second exponential phase is composed of new-pole and old-pole clusters (Fig. S9A). The distribution patterns suggest lateral clusters are basically unit-clusters, old-pole clusters are essentially congregations of multiple unit-clusters, and new-pole clusters can be made of either one or more unit-clusters. The average intensity of lateral clusters is ∼ 470 counts on our camera detector (corresponding to ∼ 920 CheW3 molecules), coinciding with the plateau-like region of intensity-ranking curve (Fig. S9B, dashed circle). The intensities of larger lateral clusters bunch together around a value (Fig. S9B, solid circle) about twice of that of the plateau-like region, suggesting these larger lateral clusters were composed of two unit-clusters. The intensity-ranking curve of new-pole clusters also shows a first linear phase (Fig. S9C). The intensity value at the transition point is ∼ 500 counts (corresponding to ∼ 970 CheW3 molecules). Taken together, the estimations of unit-cluster size from the intensity distributions of lateral and new-pole clusters in normal cells, as well as clusters in filamentous cells, suggest a value of ∼ 920–1100 CheW3 proteins per unit-cluster.
Cytoplasmic chemosensory clusters stay close to FtsZ assemblies during most of the cell cycle
The cytoplasmic chemosensory clusters are composed of homologues of the membrane clusters, and the daughter cells need to inherit both clusters on cell division to allow chemotaxis. We have shown that the cytoplasmic clusters move dynamically in R. sphaeroides cells with time-average positions at pre-cytokinetic sites and these clusters appear linked to chromosome movement via a ParA-like system (Thompson et al., 2006; Roberts et al., 2012). Consistent with this, snapshots show that cytoplasmic chemosensory clusters (marked by TlpT–YFP) were close to the Z-ring (Fig. 6A). For cells with multiple clusters, most of them (86.0%) have at least one cluster close to the Z-ring (Fig. 6B). By using the different FtsZ assemblies as indicators of cell-cycle stages, we found that the colocalization of cytoplasmic clusters and FtsZ was cell-cycle related (Fig. S10A and B). The frequent proximity of cytoplasmic chemosensory clusters and constricting Z-rings or polar FtsZ spots (Fig. S10A, last and first cells respectively; Fig. S10B) was unexpected, since it is believed that non-chromosome cytoplasmic cargos of ParA homologues move to the future cytokinetic sites before cytokinesis (Gerdes et al., 2010; Lutkenhaus, 2012). Time-lapse imaging shows that in two-cluster cells, one cluster moved to a pre-cytokinetic site first (Fig. 6C, white arrowhead), while another cluster stayed with the Z-ring and gradually moved to the other pre-cytokinetic site later (Fig. 6C, green arrowheads). The proximity to the Z-ring was often seen even after extensive constriction of the Z-ring (Fig. S10C, green arrowheads).
Our data suggest that cytoplasmic chemosensory clusters and FtsZ occupy similar subcellular locations, at least to within the optical resolution limit of our microscope. However, the movement of cytoplasmic clusters to the pre-cytokinetic sites before FtsZ (Fig. 6C and Fig. S10A and C) suggests cytoplasmic clusters do not use FtsZ assemblies as anchors for positioning. The asymmetric movement of the two cytoplasmic clusters to the future midcell (Fig. 6C and Fig. S10C) is consistent with assemblies moving with chromosome segregation (Lutkenhaus, 2012), suggesting a larger role for the chromosome in this process than previously postulated. Similar spatiotemporal dynamics of cytoplasmic chemosensory clusters was seen in photoheterotrophic cells (Fig. S10D), indicating that the reorganization of the cytoplasmic membrane does not interfere with cytoplasmic chemosensory cluster positioning.
Relative positioning between MreB, FtsZ and chemosensory clusters
The actin homologue MreB participates in the spatial regulation of different biomolecules in bacteria (Shaevitz and Gitai, 2010). The dominant models for the in vivo configurations of MreB cytoskeleton during the past decade are filamentous helices or rings (Shaevitz and Gitai, 2010). However, we observed circumferentially arranged GFP–MreB patches in most cells of a R. sphaeroides strain harbouring the genomic gfp–mreB fusion (Fig. 7A and B) (Slovak et al., 2005), which is consistent with current studies (Domínguez-Escobar et al., 2011; Garner et al., 2011; van Teeffelen et al., 2011; Swulius and Jensen, 2012). Since the GFP–MreB fusion is not fully functional (Slovak et al., 2005), we examined the localization of MreB in a merodiploid strain containing an IPTG-inducible copy of yfp–mreB in addition to the native genomic mreB copy. Leaky expression resulted in a similar patchy pattern of MreB localization (Fig. 7C and D).
We observed a cell cycle-specific colocalization between FtsZ and MreB. FtsZ nodes/rings and MreB colocalize at midcell until constriction, after which MreB relocalizes at future midcell before FtsZ (Fig. 7A and B) (Slovak et al., 2005). The localization patterns of the MreB ring-like arrangement and the Z-ring suggest two independent structures (Fig. 7B). MreB forms ring-like structures which colocalized with and are dependent on the Z-ring in C. crescentus and E. coli (Aaron et al., 2007; Vats and Rothfield, 2007). However, the establishment of medial ring-like arrangement of MreB seems to be independent of the Z-ring in R. sphaeroides (Fig. 7A and B) (Slovak et al., 2005), suggesting a novel mechanism to position MreB assemblies in this bacterium.
To explore the possible role of MreB plays in the positioning of membrane and cytoplasmic chemosensory clusters, we monitored the localizations of chemosensory clusters together with MreB. No correlation was found (Fig. 7C and D).
Our data show that FtsZ forms dynamic, cell cycle-dependent assemblies in R. sphaeroides, which can exhibit substantial reorganization during the formation of the Z-ring. The polar spots localized to the new poles immediately after cytokinesis redistribute to the midcell to initiate the formation of the Z-ring (Figs 1 and 3Bi). These polar spots could move as single or multiple independent entities. FtsZ spots change in their intensities, numbers and positions. Redistribution of FtsZ spots to the midcell allows the formation of Z-rings. FtsZ spatial gradients stretch out from nodes and eventually encircle the midcell plane, undergoing rearrangements and a more symmetric allocation is reached. Structures similar to FtsZ nodes have been seen in C. crescentus (Quardokus et al., 2001), E. coli (Sun and Margolin, 1998), sporulating Streptomyces coelicolor (Willemse et al., 2011), an ectosymbiotic γ-proteobacterium divides longitudinally (Leisch et al., 2012) and cyanelles (cyanobacteria-like endosymbionts) (Sato et al., 2009), suggesting a common mechanism for the formation of Z-rings.
One current model for Z-ring formation suggests that the Z-ring forms by the collapse of a helical structure (Ben-Yehuda and Losick, 2002; Thanedar and Margolin, 2004; Thanbichler and Shapiro, 2006; Peters et al., 2007). However, less than 1% of R. sphaeroides cells showed the extended spiral-like structures under medium induction (up to 7.5 μM IPTG). Instead, more than 50% of R. sphaeroides cells had clear Z-ring precursors (Table S1). One possibility is that similar Z-ring precursors are only transiently present in other bacteria, and the helical structures may represent a relaxed Z-ring or adjacent structures. The longer cell cycle of R. sphaeroides increases the chance of seeing the Z-ring precursors and serves as a good model to study Z-ring development.
Another nucleation site model (Addinall and Lutkenhaus, 1996; Pichoff and Lutkenhaus, 2005) suggests the formation of Z-ring initiate at single (Addinall and Lutkenhaus, 1996) or multiple (Pichoff and Lutkenhaus, 2005) midcell sites. Bidirectional polymerizations emerging from nucleation sites produce long FtsZ polymers that associate laterally to form Z-rings. For the nucleation model to be true, the Z-ring must be composed of a few long protofilaments that grow from the nucleation sites and encircle the midcell plane. However, in vitro and in vivo data suggest that FtsZ protofilaments from diverse species are only 100–200 nm long (Li et al., 2007; Erickson et al., 2010), much shorter than the circumference of a typical bacterium. Since the midcell node in R. sphaeroides contains the majority of total cellular FtsZ molecules (Table S1), further protein synthesis is not required for Z-ring formation (i.e. the relocation of polar spots and the reorganization of midcell nodes into mature Z-rings), and the Z-ring development process is reversible (Fig. S2C), we believe that the main function of midcell nodes is probably to allow reorganization/redistribution of FtsZ rather than de novo nucleation. Nevertheless, it is possible that the reorganization/redistribution of FtsZ may include nucleation–polymerization process taking place at the edges of the FtsZ nodes/gradients.
The positioning of membrane chemosensory clusters
Membrane chemosensory clusters accumulated at both poles, as seen in many bacterial species, with one obvious old pole and one new pole (Gestwicki et al., 2000). A group of polar-flagellated γ-proteobacteria show a modified, cell cycle-dependent bipolar pattern: new-born cells have a unipolar distribution of chemosensory clusters at the old pole, and becomes bipolar as the new pole matures (Ringgaard et al., 2011). Stochastic self-assembly/helical insertion (Shiomi et al., 2006; Thiem and Sourjik, 2008; Wang et al., 2008; Greenfield et al., 2009) and ParA homologue-mediated diffusion-and-capture (Ringgaard et al., 2011) mechanisms have been proposed for the two patterns respectively. C. crescentus represents an extreme with a sole unipolar localization programmed by its intrinsic asymmetry (Alley et al., 1992). In E. coli polar clusters seem relatively dynamic, but remain at the cell poles, while new clusters localized laterally at pre-cytokinetic sites, appearing tightly localized (Thiem et al., 2007). The situation in R. sphaeroides seems very different: the polar clusters appear to be formed from a number of smaller clusters, which appear to freely diffuse as large complexes in the lateral membrane with some polar retention. The lateral clusters are not static, and there is no accumulation at pre-cytokinetic sites. Indeed, the clusters appear to be excluded from the vicinities of Z-rings. A periodic pattern seen in cephalexin-treated filamentous cells is probably determined by a synergistic action of stochastic self-assembly and an exclusive effect associated (directly or indirectly) with the Z-ring. Taken together, stochastic self-assembly, diffusion-and-capture and a Z-ring-associated exclusive effect may co-ordinate to generate the spatiotemporal dynamics of membrane chemosensory clusters in R. sphaeroides.
Although FtsZ spots seem to only have a steric-hindrance effect on the localization of membrane chemosensory clusters, the Z-ring/divisome may create an environment with different physicochemical properties in its vicinity (Lopez-Montero et al., 2012), preventing membrane chemosensory clusters to localize in the Z-ring vicinity.
Dynamic unit-clusters of membrane chemosensory proteins
While many of the previous fluorescence and cryo-electron microscopy data have suggested the membrane chemosensory proteins form massive polar clusters, our data suggest that in R. sphaeroides there is an optimum cluster size and these smaller clusters are free to diffuse, but tend to accumulate at the poles. Similar congregation–segregation process (Thiem et al., 2007) and pattern of intensity-ranking plots (Ping et al., 2008) can be seen in E. coli. Quantification of fluorescence intensities from lateral clusters in normal and filamentous cells suggests an average cluster size of about 1000 CheW3 molecules. As quantitative fluorimetry data suggest an average of 8700 ± 260 CheW3 molecules per cell (with the majority present in membrane clusters) (Wilkinson et al., 2011), this suggests six to eight clusters per unit cell, which correlates well with the numbers seen in normal and filamentous cells when clusters were segregated (Figs 3A and 5A).
The clustering of the chemosensory proteins into large quaternary complexes may be important for the high level of sensitivity and gain in the bacterial chemosensory pathway (Bray et al., 1998; Gestwicki and Kiessling, 2002), but recent data suggest that signalling works through signalling teams that are smaller in size (Hansen et al., 2010). The size (upper limit) of clusters could be jointly determined by intracluster forces that hold constituting subunits and the nature of surrounding environments (Endres, 2009). It is likely that intercluster interactions create transient, unstable couplings between unit-clusters. With the same number of constituting proteins, several unit-clusters may generate better digital signals than a single huge cluster, hence improved signalling fidelity (Suzuki, 2012).
Recent data have suggested that when R. sphaeroides grows photoheterotrophically, rather than producing invaginations tightly connected to the inner membrane, the intracytoplasmic photosynthetic membranes bud from a limited number of sites on the inner membrane to form intracellular vesicles (Tucker et al., 2010). This is supported by our data as the behaviour of FtsZ and membrane chemosensory clusters in photoheterotrophic and aerobic cells are similar, suggests a relatively similar inner membrane organization.
The partitioning of cytoplasmic chemosensory clusters
Previous work has shown that the cytoplasmic chemosensory cluster segregates using a ParA-like protein, PpfA, which localizes to the chromosome surface, suggesting a plasmid-like segregation system (Thompson et al., 2006; Roberts et al., 2012). Dynamic oscillation of ParA homologues has been suggested to constantly adjust the inter-cargo distance and equi-partition cargos (Gerdes et al., 2010; Lutkenhaus, 2012). One feature of such mechanisms is a symmetric distribution of cargos. The cytoplasmic chemosensory clusters and FtsZ often colocalize to the same midcell region over a large portion of the cell cycle, but the two duplicated clusters move to daughter cells independently. We do not see equi-partitioning of the cytoplasmic chemosensory clusters, which may reflect the absence of oscillations seen for PpfA (Roberts et al., 2012) and suggests asymmetrically segregating chromosomes (i.e. only one copy of the duplicated chromosomes is translocated across the cell) could act as the anchor/carrier.
FtsZ and MreB
The cell cycle-specific colocalization between MreB and FtsZ is different in R. sphaeroides compared with C. crescentus and E. coli (Aaron et al., 2007; Vats and Rothfield, 2007): the medial ring-like arrangement of MreB forms before and are independent of the formation of the Z-ring (Fig. 7A and B) (Slovak et al., 2005). This observation suggests that there are diverse interacting modes in the homologous cytoskeletal networks in different bacteria. Moreover, it is possible that MreB assemblies at pre-cytokinetic sites may assist in the development process of the Z-ring. Although previous studies (Shiomi et al., 2006; Thiem et al., 2007) suggest MreB may not be involved in the positioning of membrane chemosensory clusters, the distinct localization pattern of MreB prompted us to investigate the role of MreB in the positioning of membrane and cytoplasmic chemosensory clusters in R. sphaeroides. However, no apparent correlation was found.
Taken together these data show that very different mechanisms are employed to ensure daughter cells inherit two homologous protein complexes generating chemosensory signals, both required for chemotaxis, one relying on the diffusing of large protein complexes and polar trapping, the other is partitioned using the segregating daughter chromosomes. Unlike previous observations, the membrane clusters form unit complexes which diffuse in the cell and neither chemosensory cluster has positioning linked directly to the positioning of FtsZ or MreB through the cell cycle. The data also show a detailed pattern of FtsZ positioning through the R. sphaeroides cell cycle, with polar spots moving to midcell to produce nodes from which the Z-ring develops. Therefore, our study provides new perspectives on the complete life cycle of the Z-ring and the dynamic nature of chemosensory clusters.
Bacterial strains and growth conditions
The strains and plasmids used in this study were constructed as described before (Porter et al., 2007) and are listed in Table S2. All R. sphaeroides strains were derived from WS8N and grown in succinate medium with antibiotics at 30°C. When appropriate, the antibiotics nalidixic acid and kanamycin were used at 25 μg ml−1. All experiments were performed with log-phase aerobic or photoheterotrophic (illuminated with light at 10 W m−2) cells. E. coli XL1-Blue was used for cloning. E. coli strain S17-1λpir was used for conjugal DNA transfer into R. sphaeroides (Porter et al., 2007). For inducing filamentation, cells were (i) treated with 2.5 μg ml−1 cephalexin for 3.5 h or (ii) induced with 250 μM IPTG overnight before imaging. To inhibit protein synthesis, chloramphenicol was added to the bacterial culture at a final concentration of 50 μg ml−1 (Romagnoli et al., 2002). For time-lapse imaging, cells pre-treated with chloramphenicol for 90 min were laid on agarose pads pre-soaked in medium containing chloramphenicol.
Microscopy and image analysis
Cells were immobilized on 0.8% agarose pads in succinate medium and observed using a DeltaVision platform (Applied Precision) equipped with a 100 × 1.4 NA Plan-Apochromat objective and a Coolsnap HQ camera. The live-cell filter set was used. Imaging settings were carefully selected to avoid saturation of the camera, and were kept constant for similar samples. To optimize the accuracy of protein colocalization, each z-stack was acquired with one channel immediately followed by another. Cells were optically sectioned into multiple slices spacing 80–120 nm and deconvolved using the DeltaVision softWoRx 4.0.0 restoration system. Cells were imaged at 25°C.
Images are, unless specified, maximal projections of z-stacks. However, for quantification of intensity, sum projections were used. Background intensity was subtracted. Quantification and analysis were performed in ImageJ (http://rsbweb.nih.gov/ij). Brightness and contrast adjustment, when performed, was applied uniformly to whole images. For single-particle tracking, distance measurement and colocalization categorizing, the centroid (for spherical chemosensory clusters), centre of mass (for irregular-shaped chemosensory clusters) and edges (for FtsZ assemblies) were used. Cell Counter (http://rsbweb.nih.gov/ij/plugins/cell-counter.html) was used as a plug-in for ImageJ to count and categorize assemblies and cells.
Strain JPA1418 cells were collected and resuspended in succinate medium to give a wide range of cell concentrations. Resuspended cells were transferred to black, clear-bottom 96-well plates (Corning) and fluorescence readings were taken using appropriate YFP filters on a FLUOstar Optima fluorescence plate reader (BMG LABTECH) with a gain setting of 2200. A cellular autofluorescence baseline was calculated using wild-type WS8N cells and subtracted from the measured values. For protein copy number quantification, fluorescence signal was compared with that obtained from known quantities of purified YFP, resuspended in the same number of WS8N cells (Wilkinson et al., 2011).
We thank Micron Oxford for help with microscopy. S.-W.C. was supported with an EPA Cephalosporin Scholarship by the Linacre College (Oxford, UK) and a Scholarship for Studying Abroad by Ministry of Education (Taiwan). M.C.L. was supported by a Royal Society University Research Fellowship and through BPSI. This work was supported by BBSRC.