To understand further the role of the nucleoid and the min system in selection of the cell division site, we examined FtsZ localization in Escherichia coli cells lacking MinCDE and in parC mutants defective in chromosome segregation. More than one FtsZ ring was sometimes found in the gaps between nucleoids in min mutant filaments. These multiple FtsZ rings were more apparent in longer cells; double or triple rings were often found in the nucleoid-free gaps in ftsI min and ftsA min double mutant filaments. Introducing a parC mutation into the ftsA min double mutant allowed the nucleoid-free gaps to become significantly longer. These gaps often contained dramatic clusters of FtsZ rings. In contrast, filaments of the ftsA parC double mutant, which contained active MinCDE, assembled only one or two rings in most of the large nucleoid-free gaps. These results suggest that all positions along the cell length are competent for FtsZ ring assembly, not just sites at mid-cell or at the poles. Consistent with previous results, unsegregated nucleoids also correlated with a lack of FtsZ localization. A model is proposed in which both the inhibitory effect of the nucleoid and the regulation by MinCDE ensure that cells divide precisely at the midpoint.
Escherichia coli normally divides at the midpoint of the long axis of the cell to produce two daughter cells of equal size. The targeting of cell division is highly precise, with a deviation of less than 1% off centre (Trueba, 1982). However, the mechanism for the precision in the selection of this site is not well understood. One hypothesis, advanced as the nucleoid occlusion model, suggested that the targeting and activation of the division site is determined by the position of nucleoids (Mulder and Woldringh, 1989; Woldringh et al., 1990). According to this model, all positions along the cell length are competent for cell division, but cell division is prevented at positions occupied by nucleoids. The model then proposes that the nucleoid-mediated inhibition is subsequently released upon nucleoid segregation, unveiling the central division site.
To test this model, we have previously examined the localization of the essential cell division protein FtsZ (Lutkenhaus and Addinall, 1997; Margolin, 1998) in anucleate cells that were formed in mukB and parC mutants defective in nucleoid segregation. We found that FtsZ rings were frequently present at or close to the midpoint of the anucleate cells, indicating that nucleoids were not essential for the positioning of FtsZ rings (Sun et al., 1998). However, FtsZ rings were usually positioned off-centre in parC filaments containing a large, central, unsegregated nucleoid. This result suggested that FtsZ ring formation might be prevented by unsegregated nucleoids, leaving open the possibility that the wild-type nucleoid, while not essential for FtsZ ring positioning, might have an inhibitory role in this process. Many FtsZ rings in the anucleate cells were near but not exactly at the centre (Sun et al., 1998), indicating that the precision of FtsZ ring localization was decreased in anucleate cells in contrast to the high precision in wild-type cells. Taken together, these results raised the possibility that some type of nucleoid-mediated occlusion might be important for the precise central positioning of the FtsZ ring in wild-type cells.
Another model for division site selection has proposed that there are a limited number of predetermined division sites and that these sites differentiate before the onset of septation (Rothfield et al., 1990). In this model, potential cell division sites are present at the cell poles as well as at the cell centre; the polar sites are proposed to be remnants of old central division sites (Teather et al., 1974). The selection of the proper mid-cell site instead of the polar sites requires the co-ordinated functions of the minCDE gene products, MinC, MinD and MinE (de Boer et al., 1989; Rothfield and Zhao, 1996). Under normal conditions, MinC and MinD co-ordinately inhibit cell divisions at the poles and prevent the reuse of old division sites, while MinE suppresses MinCD inhibition at the central site. In agreement with this model, cells divide at the poles and produce minicells when the minCDE operon is deleted (de Boer et al., 1989). Polar divisions occur at the expense of central divisions, resulting in the formation of short filaments in addition to minicells (Teather et al., 1974). Consistent with its role as a cell division inhibitor, overproduction of MinCD blocks cell division at all sites, generating long filaments, while overexpression of MinE suppresses this inhibition by MinCD (de Boer et al., 1989). The absence of MinE is sufficient to block cell division at all sites as long as MinCD is present. Inhibition by MinCD appears to occur by directly preventing FtsZ ring formation and can be suppressed by increasing FtsZ levels (Bi and Lutkenhaus, 1990; 1993; de Boer et al., 1990). Moderate overexpression of FtsZ and FtsA induces minicell formation in otherwise wild-type cells (Ward and Lutkenhaus, 1985; Begg et al., 1998), probably by titrating out MinCD. Isolation of MinCD-resistant FtsZ mutants further supports the idea that FtsZ is the target of MinCD inhibition (Bi and Lutkenhaus, 1990). However, no interaction between FtsZ and MinCDE was detected in a yeast two-hybrid system (Huang et al., 1996), suggesting that the inhibition might be complex and possibly indirect.
The recent finding that MinE localizes near the central site and that the localization is independent of FtsZ strongly supports the topological role of MinE in division site selection (de Boer et al., 1989; Raskin and de Boer, 1997). However, the viability of a minCDE deletion strain argues that minCDE is limited to a regulatory role in cell division and is not required for formation of the cell division machinery or for the division process. Comparison of the positioning of MinE rings in wild-type cells and FtsZ rings in anucleate cells of mukB and parC mutants suggested that the pattern of ring positioning in these two cases is quite similar (Raskin and de Boer, 1997; Sun et al., 1998). In both cases, rings are located in a zone near the cell centre, but not always precisely at the centre. This prompted us to hypothesize that, in the absence of the nucleoid, MinCDE may be the only guide for targeting FtsZ ring assembly in the cell, and the precision of FtsZ localization is decreased because MinE localization itself is also not precise.
To understand further the biogenesis and selection of the division site, here we have characterized the localization of FtsZ in ΔminCDE mutants. Our results show that, in the absence of minCDE, FtsZ rings assemble promiscuously in nucleoid-free regions of the cell. We conclude that all positions along the cell length are competent for FtsZ ring formation and that minCDE regulates FtsZ ring localization.
Localization and distribution of FtsZ rings inΔminCDE mutants
Polar FtsZ rings have been found to be associated with polar division in min mutants (Bi and Lutkenhaus, 1993), but the overall properties and patterns of FtsZ localization in min mutants have not been explored. To compare FtsZ localization in ΔminCDE mutant and wild-type cells, we transduced the ΔminCDE::kan allele from strain PB114 into E. coli strain MG1655. One typical transductant was purified and designated strain WM947. When grown in rich medium at 37°C, strain WM947 produces a mixture of cells of normal size, minicells and short filaments, similar to the behaviour of strain PB114 (de Boer et al., 1989). As reported previously (Mulder et al., 1990; Åkerlund et al., 1992), some filaments exhibited an irregular distribution of nucleoids and larger than normal chromosome-free regions at the cell pole, as detected by DAPI staining (data not shown).
We then examined the localization of FtsZ in WM947 by immunofluorescence staining with affinity-purified anti-FtsZ antibody. Both polar and non-polar FtsZ rings were present in the ΔminCDE mutant cells (Fig. 1A, C, D and F, arrows), in agreement with a previous study using immunoelectron microscopy to visualize FtsZ (Bi and Lutkenhaus, 1993). Importantly, most of the short filaments contained FtsZ rings at all clear gaps between nucleoids (Fig. 1B, single arrowheads; Fig. 1E, all arrows). This indicates that the cell division delay in most of these filaments is not caused by a lack of FtsZ rings. Further examination of FtsZ staining also suggested that some rings stained more brightly than others, even within the same filament (Fig. 1A–C, forked arrows; most apparent in Fig. 1C). This variation in brightness was not generally observed in ftsA filaments, for example (data not shown). If we assume that the brightness of the FtsZ immunofluorescence signal reflects the quantity of assembled protein, then this result suggests that some rings in ΔminCDE filaments may contain more FtsZ molecules than others. If these rings mature at different times, then divisions might occur at random time intervals, as observed previously (Åkerlund et al., 1992). The formation of multiple rings in min mutants was not the result of increased FtsZ protein concentrations in the cell, because quantitative immunoblotting experiments demonstrated that FtsZ levels normalized to cell protein were unchanged in these mutants as well as the other mutant combinations described below (data not shown).
Examination of FtsZ ring positions in ΔminCDE cells revealed that some rings were located neither at mid-cell nor at ‘potential division sites’, defined as sites within a filament that could give rise to newborn cells of normal size, such as one-half or one-quarter of the cell length (Fig. 1A–F, horizontal arrows; Fig. 1G–I). We examined the spatial distribution of FtsZ rings in approximately 150 cells of the ΔminCDE mutant (Fig. 2A). In cells shorter than 3 μm, most of the non-polar FtsZ rings were localized at or near the cell centre. However, in cells longer than 3 μm, some rings were not localized at mid-cell or quarter sites. This apparently random distribution can be explained most easily by the alteration of the distance between a given pole and a fixed potential division site by the formation of one or more minicells from that pole. Another possible explanation is that nucleoid positioning is altered in many min mutant cells and that this altered positioning influences FtsZ ring placement. To rule out the possibility that the apparent irregularity of FtsZ ring placement in ΔminCDE cells was caused by a technical problem, we measured FtsZ ring positions in MG1655, the parent strain of WM947. As shown in 2Fig. 2B, most FtsZ rings were exactly at or very close to mid-cell. For the 130 cells examined, the standard deviation from 0.5 (the normalized mid-cell position, as measured from the cell poles) was 0.013; this scatter may have been introduced solely by measurement errors.
Correlation of FtsZ ring positions with chromosome staining in these mutants showed that most FtsZ rings were localized between nucleoids. This suggested that aberrant FtsZ localization in some filaments may have been caused by irregular nucleoid positioning, as postulated above. We also noticed that unsegregated nucleoids were often associated with missing FtsZ rings (data not shown), as described previously (Sun et al., 1998), supporting the notion that such nucleoids may inhibit FtsZ localization. However, rings were occasionally found in some filaments in which nucleoids were not visibly segregated (Fig. 1A–C, double arrowhead), suggesting either that nucleoid-mediated inhibition of FtsZ ring assembly may not always occur or that segregation occurred in these cases but was not detectable with our staining technique.
Nucleoids may play an important role in the precision of FtsZ ring localization
In our previous study, we found FtsZ rings at or near the centre of anucleate cells, suggesting that assembly of the FtsZ ring at mid-cell does not require the presence of nucleoids. The precision of FtsZ ring localization was, however, reduced in anucleate cells compared with wild-type cells (Sun et al., 1998). Further examination of FtsZ ring position in approximately 130 anucleate cells produced by the parC mutant revealed that the average position of FtsZ rings as measured from a pole was 0.5 ± 0.062, where 0.5 is the normalized cell midpoint. The standard deviation of 0.062 is five times higher than that calculated for wild-type cells (0.013). This result suggests that the nucleoid may affect the precision of FtsZ ring placement.
FtsZ rings were present at most nucleoid gaps in cephalexin-inducedΔminCDE filaments
To investigate further the relationship between the positioning of FtsZ rings and nucleoids in ΔminCDE cells, we used cephalexin to block the action of FtsI in these cells. This served to prevent minicell division and to induce the production of long filaments. After 1 h of growth in cephalexin, the cells of both WM947 and the parent MG1655 became filamentous. These filaments lacked constrictions, as described previously for FtsI inactivation (Pogliano et al., 1997). In MG1655, FtsZ rings localized at some of the nucleoid gaps (often at the quarter sites) and were missing at others (Fig. 3A–C), similar to the missing rings reported previously in filaments lacking FtsI activity (Addinall et al., 1996; Pogliano et al., 1997). However, in ΔminCDE filaments (WM947) induced by cephalexin, FtsZ usually localized at most potential division sites as long as chromosome segregation was normal (Fig. 3D–F). The average cell length per ring in WM947 increased from 1.76 to 2.04 μm after cephalexin treatment, whereas the distance more than doubled (from 2.65 to 5.98 μm) in MG1655. These results suggest that the deficit of FtsZ rings in the absence of FtsI is caused by MinCD-mediated inhibition of FtsZ assembly and that the inhibition is relieved in the absence of min. Moreover, the average cell length per ring of 1.76 in WM947 versus 2.65 in MG1655 supports the idea that more FtsZ rings can assemble per unit length in ΔminCDE cells.
Aberrant nucleoid segregation in ΔminCDE mutants also became more obvious when FtsI was inactivated (Fig. 3H, both arrows), probably because the longer cells facilitated the detection of abnormalities. Again, the aberrant nucleoids usually correlated with lack of FtsZ rings, but one or two rings were sometimes found when the aberrant nucleoids were very long, apparently unsegregated and covered many potential sites of division.
Double or triple FtsZ rings inΔminCDE and fts filaments
Occasionally, double FtsZ rings were observed in ΔminCDE mutant cells (Fig. 1D–F, double arrowheads; most apparent in Fig. 1F). This phenomenon became more evident when filaments were induced by cephalexin treatment. Arrows in Fig. 3D highlight some double (single arrowhead) and triple (double arrowhead) FtsZ rings in cephalexin-induced ΔminCDE mutant filaments. To rule out the possibility that double ring formation is caused by a non-specific effect of cephalexin treatment, we introduced ΔminCDE into a ftsI(ts) mutant (AX655) by P1 transduction. The resulting strain was designated WM955. Strain WM955 grew very slowly at 28°C, and the cells were more filamentous than either the ΔminCDE mutant or the ftsI(ts) mutant alone at this permissive temperature (data not shown). After shifting to the non-permissive temperature of 42°C for 2 h, the cells became extremely long filaments. The localization of FtsZ in WM955 filaments was similar to its localization in WM947 when FtsI was inactivated by cephalexin; double FtsZ rings were also found in some filaments (Fig. 3G–I, leftward arrow).
To determine whether the formation of double FtsZ rings in the cephalexin-treated ΔminCDE mutant or the ftsI ΔminCDE double mutant is a general property of the ΔminCDE mutant or specific to inactivation of FtsI, we examined FtsZ localization in a ΔminCDE ftsA(ts) double mutant. This mutant, WM952, was constructed by transducing the ΔminCDE::kan allele into strain AX621 (ftsA1882). Incubation of WM952 at the non-permissive temperature induced the production of long filaments. As was observed in the FtsI-inactivated ΔminCDE filaments, there were regions with apparently poor nucleoid segregation (Fig. 3J–L, rightward arrows), and FtsZ rings were present at every potential site at which the chromosomes were obviously well segregated (Fig. 3J–L). In addition, about 35% of these filaments had at least one double or triple FtsZ ring (Fig. 3J–L, leftward arrows). In contrast, FtsZ ring localization in ftsA minCDE+ filaments was similar to that reported previously (Addinall et al., 1996): FtsZ rings were missing at some sites and no double or triple rings were found (data not shown).
Localization of FtsZ inΔminCDE parC double mutants
The double or triple rings found in ΔminCDE ftsI and ΔminCDE ftsA double mutants were usually located in gaps between nucleoids that were clearly larger than normal gaps (Fig. 3E, H and K, leftward arrows). We reasoned that FtsZ might be able to form higher numbers of clustered rings if the gaps between nucleoids were even longer. This prompted us to examine FtsZ ring formation in cells that generated large nucleoid-free regions. To produce such cells, we transduced the ΔminCDE::kan mutation into JV47, an MG1655 derivative containing the parC(ts) mutation, which is conditionally defective for topoisomerase IV and therefore chromosome segregation (Zechiedrich and Cozzarelli, 1995). The resulting parC(ts) ΔminCDE double mutant was designated WM940.
When grown at 28°C, the permissive temperature for parC, the cell size and nucleoid distribution of the double mutant strain WM940 was indistinguishable from that of the single min mutant, WM947. However, after shifting the temperature to 42°C for 1 h, most filaments contained unsegregated or poorly segregated nucleoids, and many of these were centrally located. FtsZ rings were positioned to either side of the unsegregated nucleoid mass in both the parC mutant JV47 (Fig. 4A–C, leftward arrows) and the ΔminCDE parC double mutant WM940 (Fig. 4D–F, leftward arrows). Interestingly, however, the nucleoid-free regions in WM940 were shorter than those in filaments of JV47 (compare Fig. 4B and E). This is presumably because of the occurrence of sequential polar divisions caused by the absence of MinCDE. Importantly, double rings were found in some filaments with longer nucleoid-free regions (Fig. 4D–F, downward arrows). When WM940 was grown at 42°C for more than 2 h, long chromosome-free regions were observed (Fig. 4H, rightward arrow), but 80–90% of these lacked FtsZ rings (data not shown). This phenomenon might result from delayed induction of the SOS response, as FtsZ rings were also missing in many parC filaments when JV47 was grown at 42°C for more than 2 h (data not shown). WM940 also produced fewer long anucleate cells than did JV47, probably because minicell divisions predominated. Despite these complications, dramatic FtsZ spirals and multiple FtsZ rings were found in nucleoid-free regions of some filaments and anucleate cells of strain WM940 at this time point, as shown in 4Fig. 4G–I (horizontal arrows).
Cell division occurs at essentially all positions in polar chromosome-free regions in the absence of minCDE
To determine whether there is a specific site for polar cell divisions in the parC ΔminCDE double mutant (WM940), we measured the sizes of anucleate minicells and rods resulting from cell divisions at improper sites. The sizes of small minicells with diameters below 0.4 μm were impossible to measure accurately by light microscopy. We therefore examined the sizes of minicells, which ranged in length from 0.4 μm to 1 μm, and rod-shaped anucleate cells, which we arbitrarily defined as those cells without DNA that were longer than 1 μm. The size distribution is shown in Fig. 5. Approximately 70% of the total anucleate cells were minicells, with diameters from 0.4 to 0.9 μm. Cells longer than 2.1 μm, which represented less than 3% of the total, were not included in Fig. 5; some of these cells were up to 4 μm long. Nevertheless, the length distribution data shows that the sizes of minicells and anucleate rods covered a wide range. Minicells of different sizes are also highlighted in 4Fig. 4G–I (upward arrows). This suggests that division events can occur at essentially any position relative to the cell pole in the absence of the chromosome and minCDE. The predominance of minicells as opposed to longer anucleate rods can be explained if FtsZ rings assemble promiscuously near the cell poles when the anucleate regions are still small. Such FtsZ rings would lead to frequent minicell formation, which in turn would further reduce the length of the anucleate regions.
Interestingly, we found bands that appeared to be FtsZ rings in some minicells and short anucleate rods (data not shown) that appeared as convincing as FtsZ rings in longer anucleate rods (Fig. 4A–C, rightward arrows). It is impossible, however, to evaluate FtsZ ring placement in these cells, because of their small size and the small number of anucleate rods that contained these apparent FtsZ rings. However, if these bands are FtsZ rings, they support the hypothesis that FtsZ rings assemble promiscuously in the absence of MinCDE and the nucleoid.
Clusters of FtsZ rings form in the long chromosome-free regions ofΔminCDE parC ftsA triple mutant filaments
The presence of multiple FtsZ rings in large nucleoid-free regions of some ΔminCDE parC double mutant filaments prompted us to construct a strain that would yield large nucleoid-free regions without the complications of the parC mutant when incubated extensively at the non-permissive temperature. Because minicell division was the probable reason behind the shortened anucleate regions in WM940, we sought to eliminate these divisions by introducing a cell division mutation into this strain that would not affect FtsZ rings. The ftsA mutation seemed to be a good candidate. Therefore, we constructed a ΔminCDE parC ftsA triple mutant, strain WM990, by P1 transduction of parC::Tn10 into the ΔminCDE ftsA(ts) strain WM952. As a MinCDE+ isogenic control, we also transduced parC::Tn10 into AX621, the parent strain of WM952, to generate strain WM954.
After growth at 42°C for only 1 h, both WM990 (ΔminCDE parC ftsA) and WM954 (parC ftsA) produced very long filaments containing alternating unsegregated nucleoid aggregates and long nucleoid-free regions (Fig. 6B, E, H and K). Dramatic clusters of FtsZ rings were present in most of the nucleoid-free regions of WM990 filaments (Fig. 6A–F, arrows). As many as six rings were observed in a single anucleate segment. The most tightly packed ring clusters contained one ring every 0.5 μm on average, which is significantly higher than the normal 2–3 μm spacing of FtsZ rings in min+ filaments. In striking contrast, often only one or two FtsZ rings were found in each of the nucleoid-free regions of WM954 filaments, which were at least as large as those in WM990 cells (Fig. 6G–L, regions not highlighted by arrows). These results suggest that all positions along the cell length are competent for FtsZ ring formation in the absence of MinCD and the nucleoid and support the idea that MinCD and the nucleoid normally prevent FtsZ ring formation at most positions.
We characterized the frequency and positioning of FtsZ rings in filaments of the parC ftsA mutant, WM954, in more detail. In about 85% of large (longer than 1 μm) nucleoid-free regions, only one FtsZ ring was present and was located immediately adjacent to the nucleoid. Remarkably, in only 2% of the large nucleoid-free regions were FtsZ rings not immediately adjacent to the nucleoid. Moreover, in about 1% of the chromosome-free regions counted, two FtsZ rings were present and located at each side of the gap. One example of this rare phenomenon is highlighted by arrows in 6Fig. 6G–I. These anucleate regions with two FtsZ rings were usually longer than 3.6 μm. Double FtsZ rings or rings that appeared to be significantly wider than single rings were found in less than 10% of the anucleate regions examined; one of these relatively rare events is highlighted in 6Fig. 6J–L (arrows). To confirm that the clustering of FtsZ rings in the ΔminCDE ftsA parC triple mutant (WM990) is not specific to the ftsA mutation, ΔminCDE parC double mutant cells (WM940) were treated with cephalexin to inactivate FtsI, and FtsZ rings were characterized in the resulting filaments. As expected, multiple rings were also found in the nucleoid-free gaps (data not shown), although the number of rings per gap was often lower than for the triple mutant.
Finally, we measured the distance of FtsZ rings from the cell poles in the polar anucleate regions of WM954 (parC ftsA) filaments. Most rings were located at a distance of 1.9 ± 0.44 μm from the poles, consistent with their localization at potential division sites. In support of the idea that Min proteins may block FtsZ ring assembly at the poles in the absence of the nucleoid, we found that polar nucleoid-free regions shorter than 1 μm usually lacked FtsZ rings.
In this study, we have discovered that FtsZ assembles into a series of closely spaced rings in nucleoid-free segments of cells lacking the MinCDE proteins. This phenomenon became especially obvious when septation was inhibited, because the FtsZ ring cluster pattern was iterated dramatically within the long filaments. These results have prompted us to formulate a new model to explain FtsZ ring positioning, in which all positions along the cell length are competent for FtsZ ring formation in the absence of inhibition by MinCD and the nucleoid (see below). Consistent with this model, the ΔminCDE parC double mutant produces minicells and anucleate rods of various sizes ranging from smaller than 0.4 μm to as much as 4 μm. This indicates that divisions can occur at any position with respect to the pole and that there is no single polar division site. This idea supports and significantly extends a previous study of cell division site placement in min mutants (Åkerlund et al., 1992).
Although polar divisions occur at the expense of central divisions in min mutants (Teather et al., 1974), our data suggest that this is not because of a lack of FtsZ rings. Instead, this phenomenon can now be explained by the promiscuous assembly of polar and non-polar rings in the absence of MinCD inhibition. If the lack of MinCD releases the inhibition of FtsZ assembly into rings, for example, then when multiple rings form in the same cell, it is reasonable to propose that each ring might contain less FtsZ protein. This is conceivable because (i) there was clear variability in fluorescence intensity among multiple FtsZ bands in the same cell in the min mutants; and (ii) FtsZ levels, normalized to total protein, were unchanged in all the mutants described. If a minimum amount of FtsZ per ring is required for functioning of the ring, then division might be delayed until the first ring reached threshold thickness; subsequent divisions would be delayed as well and might be randomly timed. Furthermore, other cell division proteins, such as FtsA, may also become limiting under conditions in which too many FtsZ rings are formed, particularly if they also require a threshold level for function. FtsA is a good candidate for another rate-limiting protein, because both FtsZ and FtsA levels must be increased in order to stimulate extra division events (Begg et al., 1998). The ultimate result of these delays in division in min mutants, however they are caused, is the formation of both minicells and filaments.
After this paper was submitted, it was reported that minCD mutants of Bacillus subtilis often contain multiple FtsZ rings per cell when grown in rich medium (Levin et al., 1998). This phenomenon is similar to our findings with the E. coli minCDE mutant, although no clusters of FtsZ rings in B. subtilis were reported. Interestingly, B. subtilis min mutants cultured in minimal medium contained only one FtsZ ring per cell, suggesting that frequency of ring assembly might be influenced by nutritional status. It was not established whether this might be caused by increased levels of the protein or increased ability to assemble. To determine whether a similar type of regulation occurs in E. coli, we examined FtsZ ring patterns in the ΔminCDE strain WM947 grown in minimal glucose medium. We found that these cells contained multiple FtsZ rings and double FtsZ ring clusters typical of growth in rich medium (X.-C. Yu and W. Margolin, unpublished data). Moreover, when the triple mutant WM990 (ΔminCDE parC ftsA) was grown in minimal medium supplemented with appropriate amino acids and shifted to 42°C, typical FtsZ ring clusters were found (X.-C. Yu and W. Margolin, unpublished data). These results indicate that clustering and promiscuous assembly of FtsZ rings in E. coli are not influenced significantly by nutritional status.
The mechanism that triggers FtsZ ring assembly in vivo is unknown. The estimated concentration of FtsZ (10 μM) in the average E. coli cell (Lu et al., 1998) is much higher than the critical concentration for polymerization (1–2 μM) into protofilaments in vitro (Yu and Margolin, 1997; Mukherjee and Lutkenhaus, 1998a). This would suggest that FtsZ is primed to polymerize in vivo at any time and any location in the cell if an additional activator is not required for polymerization and/or FtsZ activity is not normally inhibited. FtsZ rings in vivo are, however, generally localized at very specific sites (Bi and Lutkenhaus, 1991; Addinall et al., 1996; Ma et al., 1997; Pogliano et al., 1997; Sun and Margolin, 1998), indicating that FtsZ ring assembly is highly regulated in vivo. Although it remains possible that an activator for FtsZ polymerization exists in vivo, no special factors except GTP are required for FtsZ polymerization in vitro (Lu and Erickson, 1998; Mukherjee and Lutkenhaus, 1998a,b). Alternatively, an inhibitor such as MinCD may play an important role in preventing unwanted FtsZ polymerization. It is well established that MinCD prevents polar FtsZ ring localization under normal conditions (de Boer et al., 1989; Rothfield and Zhao, 1996) and prevents assembly of the central FtsZ ring if overexpressed or if MinE is absent (de Boer et al., 1990; Bi and Lutkenhaus, 1993). Our results significantly extend these earlier models by suggesting that MinCD not only functions at mid-cell and at the poles, but also normally prevents polymerization of the FtsZ ring in all locations in the cell except the central site.
Based on the previous models and the present findings, we have developed a new model for the selection of cell division sites in E. coli (Fig. 7). This model combines elements of the nucleoid occlusion model as well as the predetermined sites model. In our model, all positions along the cell surface are competent for cell division, and unsegregated nucleoids inhibit FtsZ polymerization specifically in the regions of the cell that they occupy. This aspect is similar to the nucleoid occlusion model. The mechanism for such toporegulation of FtsZ ring assembly by an unsegregated nucleoid is unknown and may be indirect or direct. Nevertheless, the correlation between unsegregated nucleoids and the lack of FtsZ rings has been clearly established in our previous (Sun et al., 1998) and present studies and may require MukB.
Our model is also similar to the predetermined sites model, in that it proposes that the MinCDE proteins play a major role in the selection of division sites. However, we postulate that MinCD does not merely inhibit FtsZ polymerization at the pole but, in fact, inhibits FtsZ ring assembly at all positions. This argues against the notion of a specific polar division site as a remnant from the previous mid-cell division. In our model, the topological specificity factor for MinCD, MinE, localizes to a zone surrounding the cell centre in order to suppress MinCD inhibition locally (Fig. 7). This zonal localization of MinE is supported by the MinE–GFP fusion data (Raskin and de Boer, 1997), although the actual localization of wild-type MinE present at normal cellular levels might be more precise.
If MinCDE proteins are required for selection of the mid-cell division site so that FtsZ can assemble into a ring there, how might MinE localize and influence FtsZ localization? An unidentified receptor could be present for MinE, as proposed previously (Raskin and de Boer, 1997). Alternatively, periseptal annuli may provide localization cues (Rothfield and Zhao, 1996). However, a site-specific receptor for FtsZ, as proposed previously (Raskin and de Boer, 1997), may not be required because FtsZ ring assembly can occur at all positions in the absence of MinCD and nucleoid inhibition. In the absence of nucleoid inhibition, we speculate that FtsZ is targeted to the zone defined by MinE. If it is assumed that MinE localizes to a zone and not precisely to mid-cell, then the precision of FtsZ ring localization should be reduced as well. This relatively imprecise, zonal localization of FtsZ is precisely what we observe in anucleate cells containing MinCDE.
Then why is FtsZ ring positioning so precise in wild-type cells? We propose that, in wild-type cells, the nucleoid segregation apparatus is centred, and the onset of segregation provides relief of the postulated nucleoid-mediated inhibition of FtsZ assembly by creating a small nucleoid-free gap. Because this gap would derepress FtsZ inhibition precisely at the centre of the cell (and the centre of the MinE zone), and FtsZ is usually above its critical concentration for assembly, rapid nucleation would occur at the precise centre of the cell.
Although we favour the model in Fig. 7, there are several alternative explanations for our data that should be left open for consideration. First, although many experimental controls were included, we cannot rule out the possibility that the specific combination of mutations used in this study might give rise to misleading results about FtsZ ring positioning. For example, perhaps specific potential division sites normally exist, but the effects of one or more of the mutations might relax the normal spatial limitation of these sites. We feel that this is unlikely because the triple mutant simply amplifies the ring clustering that was already hinted at in the ΔminCDE mutant. However, the model clearly needs to be tested further using other approaches, including the use of other ways of generating chromosome-free cell segments. A second possible explanation for the data is that there is in fact one specific high-affinity potential division site, but the absence of MinCDE relaxes FtsZ binding specificity, allowing it to bind to weaker low-affinity sites. If this were true, then it might be possible to find an intermediate level of MinCDE inhibition that would reveal a transition between high-affinity and low-affinity binding. Finally, although we favour the idea that gaps between segregated nucleoids allow FtsZ ring assembly, another way to explain the presence of multiple FtsZ rings in larger internucleoid gaps is that ring assembly might stimulate nucleoid segregation. It is notable that this idea is consistent with the apparent requirement of active FtsZ to synthesize new cell wall material at potential division sites (de Pedro et al., 1997).
Interestingly, FtsZ rings appear to localize preferentially to the edges of nucleoids in the parC ftsA double mutant, assuming that the nucleoid edges visualized by DAPI staining in fixed cells actually reflect the edges in intact cells. Although the reasons for this localization are unclear, one possibility is that nucleoids might also affect the extent of the MinE zone, increasing the likelihood that FtsZ will target to nucleoid edges under these conditions. Because MinD is required for MinE localization (Raskin and de Boer, 1997) and also shares sequence homology with DNA partitioning proteins, it is not unreasonable to speculate that the nucleoid may influence MinE localization. Further study of MinE ring formation in partition mutants, as well as simultaneous localization of nucleoids, FtsZ and MinE under different growth conditions, should provide further insights into the regulation of FtsZ ring assembly.
Bacterial strains, culture conditions and genetic techniques
The strains used in this study are derivatives of E. coli MG1655 (wild-type K-12 strain), AX621 (ftsA1882) and AX655 (ftsI2158) and are summarized in Table 1. JV47 is MG1655 carrying the parC281 mutation, which is linked to a Tn10 marker (zge-2393::Tn10) and was used in our previous study (Sun et al., 1998). WM940, WM947, WM952 and WM955 were constructed by transducing the ΔminCDE::kan allele from PB114 (de Boer et al., 1989), which was kindly supplied by W. Cook and L. Rothfield, into JV47, MG1665, AX621 and AX655 respectively. The Tn10-linked parC281 allele from JV47 was transduced into AX621 and WM952 to generate WM954 and WM990 respectively.
Table 1. . E. coli strains used in this study.
All strains were grown in LB (Luria–Bertani) medium containing 10 mg ml−1 tryptone, 5 mg ml−1 yeast extract, and 5 mg ml−1 NaCl. When required, cephalexin was used at a final concentration of 20 μg ml−1. Tetracycline and kanamycin were used at 12 μg ml−1 and 50 μg ml−1, respectively, for selection of the appropriate resistance markers. Transduction with bacteriophage P1 was performed as described previously (Sun et al., 1998). If the recipient strain for transduction was temperature sensitive, the plates were incubated at 28°C; otherwise transductants were grown and selected at 37°C.
Immunofluorescence staining in combination with phase contrast imaging and nucleoid staining with 4′,6-diamidino-2-phenylindole (DAPI) was performed as described previously (Sun et al., 1998) with minor modifications. The concentration of lysozyme used to permeabilize the cells varied from 0.8 to 4 mg ml−1 for different strains to obtain the best results. For ΔminCDE strains, the staining time with antibody was also varied to optimize signal brightness and to preserve resolution. Visualization of FtsZ rings in the triple mutant strain WM990 was technically more difficult than in wild-type cells, because the rings were usually faint under normal staining conditions. We suspect that this was probably because each ring may have contained less FtsZ, as many rings formed simultaneously (as discussed in the text). Overstaining, on the other hand, was sometimes a problem and resulted in diffuse bright staining. The reason for this phenomenon may be that the FtsZ rings within the clusters typical of WM990 cells were too close to each other to resolve easily when overstained. We found that staining for 80 min with both primary and secondary antibodies resulted in the best combination of FtsZ band resolution and brightness under our conditions. For FtsZ staining, our methanol fixation method (Sun et al., 1998) results in better resolution, lower background and greater reproducibility than the paraformaldehyde–glutaraldehyde fixation method (Addinall et al., 1996). Hiraga et al. (1998) also reported a similar method using methanol to fix cells, and they also found methanol fixation to be superior to the paraformaldehyde–glutaraldehyde fixation method. Fixation of WM990 cells using the paraformaldehyde–glutaraldehyde fixation method also revealed FtsZ ring clusters, although the overall resolution and signal-to-noise ratio was not as good. However, this ruled out the possibility that FtsZ ring clusters were somehow artefacts of the methanol fixation procedure.
We thank W. Cook and L. Rothfield for the PB114 strain. This work was supported by grant MCB-9513521 from the National Science Foundation.