Cell wall synthesis and remodelling dynamics determine division site architecture and cell shape in Escherichia coli

The bacterial division apparatus catalyses the synthesis and remodelling of septal peptidoglycan (sPG) to build the cell wall layer that fortifies the daughter cell poles. Understanding of this essential process has been limited by the lack of native three-dimensional views of developing septa. Here, we apply state-of-the-art cryogenic electron tomography (cryo-ET) and fluorescence microscopy to visualize the division site architecture and sPG biogenesis dynamics of the Gram-negative bacterium Escherichia coli. We identify a wedge-like sPG structure that fortifies the ingrowing septum. Experiments with strains defective in sPG biogenesis revealed that the septal architecture and mode of division can be modified to more closely resemble that of other Gram-negative (Caulobacter crescentus) or Gram-positive (Staphylococcus aureus) bacteria, suggesting that a conserved mechanism underlies the formation of different septal morphologies. Finally, analysis of mutants impaired in amidase activation (ΔenvC ΔnlpD) showed that cell wall remodelling affects the placement and stability of the cytokinetic ring. Taken together, our results support a model in which competition between the cell elongation and division machineries determines the shape of cell constrictions and the poles they form. They also highlight how the activity of the division system can be modulated to help generate the diverse array of shapes observed in the bacterial domain.

The bacterial division apparatus catalyses the synthesis and remodelling of septal peptidoglycan (sPG) to build the cell wall layer that fortifies the daughter cell poles. Understanding of this essential process has been limited by the lack of native three-dimensional views of developing septa. Here, we apply state-of-the-art cryogenic electron tomography (cryo-ET) and fluorescence microscopy to visualize the division site architecture and sPG biogenesis dynamics of the Gram-negative bacterium Escherichia coli. We identify a wedge-like sPG structure that fortifies the ingrowing septum. Experiments with strains defective in sPG biogenesis revealed that the septal architecture and mode of division can be modified to more closely resemble that of other Gram-negative (Caulobacter crescentus) or Gram-positive (Staphylococcus aureus) bacteria, suggesting that a conserved mechanism underlies the formation of different septal morphologies. Finally, analysis of mutants impaired in amidase activation (ΔenvC ΔnlpD) showed that cell wall remodelling affects the placement and stability of the cytokinetic ring. Taken together, our results support a model in which competition between the cell elongation and division machineries determines the shape of cell constrictions and the poles they form. They also highlight how the activity of the division system can be modulated to help generate the diverse array of shapes observed in the bacterial domain.
the denoised tomograms showed an elongated, triangular wedge of PG close to the invaginating IM (Fig. 1b). In cells at the final stages of cytokinesis, where IM fission was complete (Extended Data Fig. 2c), two layers of PG signal comprising the septum were readily visible (Fig. 1b). We performed subtomogram averaging to compare the envelope structure between the side wall and septum of dividing cells, which also showed two layers of PG signal within partial septa and a single layer of PG signal in the side wall (Extended Data Fig. 3a).
To investigate the mechanism of partial septum formation, we followed the constriction dynamics of each membrane. The IM was tracked using a superfolder green fluorescent protein (sfGFP) fusion to the IM-anchored Z-ring binding protein ZipA (ZipA-sfGFP), while constriction of the OM was followed using mCherry fused to the OM-lipoprotein Pal (Pal-mCherry) ( Fig. 1c and Supplementary Video 2). The Pal-mCherry and ZipA-sfGFP signal distribution at the division site confirmed the deeper constriction of IM with respect to the OM (Fig. 1d) observed by cryo-ET (Fig. 1a,b). The invagination rate of each membrane was calculated from kymographs (Fig. 1e,f and Extended Data Fig. 4). We found that the IM constriction rate increased faster than linear as the septum closed, with an average rate of 64.26 ± 33.98 nm min −1 , in line with previous measurements 20,21 . The increase in the OM constriction rate during division was less pronounced than that of the IM (Fig. 1f,g and Extended Data Fig. 4b-d). The different rates of change in constriction velocity between the two membranes account for the two membranes becoming increasingly separated as division proceeds, by 147 nm at late stages in cell division, which is in good agreement with our cryo-ET data. Thus, E. coli divides by a mixed constriction/septation mechanism, with the partial septum containing two layers of sPG signal ( Fig. 1b and Extended Data Fig. 3a). White arrowheads indicate where the PG layer appears to thicken from one to two layers, and black arrowhead indicates the side of the division site shown in the schematic overview below. Bottom row: corresponding labelled summary diagrams. The left two bottom panels correspond to arrowhead-marked top division side rotated 90° to the left. Segmented PG signal is not indicative of specific glycan strand network. c-e, Representative time-lapse series from 3 biological replicates of wild-type E. coli expressing Pal-mCherry and ZipA-sfGFP as OM and IM markers, respectively, imaged at 30 °C on M9 supplemented with 0.2% casamino acids and d-glucose. Fluorescence signals were deconvolved (see Methods). The yellow triangle marks division sites used for line scans of fluorescence intensity (FI) profiles (d) and kymograph analysis of cytokinesis (e). f, Average constriction velocities of the IM and OM were derived from the slopes of the fluorescence signals in kymographs (see Methods). Black line indicates mean. Two-sided unpaired Mann-Whitney test; **P < 0.01; N = 150 division kymographs. g, Instantaneous constriction velocities for ZipA (IM, green) and Pal (OM, magenta) are plotted against normalized cell width. Second order polynomial fits with 95% confidence intervals are shown. Scale bars: a and b, 100 nm; c, 2 µm; e, 200 nm (vertical) and 5 min (horizontal).
sPG synthesis and remodelling defines septal architecture. We next determined how the architecture of the division site and the dynamics of its constriction are altered by mutations affecting sPG synthesis and remodelling. The essential PG synthase of the divisome is formed by FtsW and FtsI (FtsWI) 22 . Following Z-ring assembly, a regulatory pathway is initiated that activates sPG synthesis by this synthase [23][24][25][26] (Fig. 2a). Activation is mediated in part via an interaction between FtsWI and the FtsQ-FtsL-FtsB (FtsQLB) complex 27 . Genetic evidence suggests that FtsQLB activation is stimulated by an essential peptide within FtsN 28 . Another domain of FtsN called SPOR concentrates the activation peptide at the division site through binding to sPG that was processed by PG amidases [28][29][30]  constriction and daughter separation 31 . The interplay between sPG synthesis activation by FtsN and amidase processing bringing more FtsN to the division site promotes a positive feedback loop, the sPG loop, that has been proposed to drive cell division 28 . We imaged several mutants defective in this process (Supplementary Tables 4  and 5): (1) Tables 1-3 and 6). All mutants displayed similar growth rates (Extended Data Fig. 6). Division sites from ftsN-∆SPOR cells resembled those observed previously for Caulobacter crescentus 33 (Fig. 2b). By cryo-ET, there appeared to be greater coordination between IM and OM constriction throughout division (Fig. 2b, Extended Data Figs. 3b-f and 5, and Supplementary Video 3). Tracking of membrane constriction dynamics confirmed that the rate of constriction for the two membranes was nearly identical in the mutant (Fig. 2c-f, Extended Data Fig. 4b-e and Supplementary Video 2). As a result of the close opposition of IM and OM, the wedge of sPG observed in filtered tomograms was not as elongated as in wild-type cells, and separate plates of material forming ahead of the wedge were not observed (Fig. 2b).
Cells defective for both amidase activators (∆envC ∆nlpD) formed a near-complete septum in which the constriction of the IM was accomplished without much observable invagination of the OM (Fig. 2b, and Supplementary Figs. 3b-f and 4b-e). NAD-filtering revealed signal corresponding to sPG that was even more clearly discernible as two distinct plates of material than in wild-type cells (Figs. 1b and 2b). Furthermore, the tomograms revealed a triangular wedge of PG material at the outer edges of the septa that was not previously observed in conventional EM analysis 34 and presumably serves as a roadblock to OM invagination ( Fig. 2b and Supplementary Video 4). It was not possible to measure membrane constriction dynamics in live cells of the ∆envC ∆nlpD double mutant because it grew poorly when expressing the fluorescent markers. However, measurements in a mutant lacking only EnvC, the dominant amidase activator 32 , revealed a substantial disparity between IM and OM constriction rates (Fig. 2c-f and Supplementary Video 2). Cryo-ET data of the ∆envC strain showed a similar division site architecture as the double ∆envC ∆nlpD mutant (Extended Data Figs. 3b-f and 5). Overall, the results indicate that impairing key components of the divisome converts the mixed constriction/septation division mechanism of E. coli to either a purely constriction or septation mode depending on the lesion.

Hyperactivated sPG biogenesis alters septal architecture.
To determine the effects of hyperactivated sPG biogenesis on division site architecture, we imaged cells of the ftsL* mutant. Strikingly, cryo-ET revealed an altered architecture in which the signal corresponding to the wedge of sPG observed in wild-type and other mutant cells was missing, and the envelope at the leading edge of the invagination was 50.8% thinner than in wild-type cells ( Fig. 2b and Extended Data Fig. 2d-f). Conversely, the envelope in the nascent polar regions adjacent to the leading edge of the invagination was 12% thicker in ftsL* cells than in wild-type cells, with the bulged areas containing more PG than normal (Extended Data Fig. 2e and Supplementary Video 5). As expected from previous measurements 26 , the average constriction velocity of the IM was much greater than in wild-type cells (Fig. 2c-e, Extended Data Fig. 4b-f and Supplementary Video 2). Surprisingly, the average rate of OM constriction was much slower than that of the IM (Fig. 2c-f and Extended Data Fig. 4), a difference that would normally be expected to give rise to cells with partial septa. However, such architectures were not observed in the tomograms (Fig. 2b, and Extended Data Figs. 3b-f and 5). The distances between the IM and OM remained relatively constant in all cells that were imaged (Fig. 2b, and Extended Data Figs. 2a-c, 3b-f and 5). This discrepancy is probably due to the PG binding activity of Pal causing the OM reporter to get stuck in the thicker PG that accumulates behind the closing septum and therefore to track poorly with the leading edge of the invaginating OM. Notably, upon closer inspection of kymographs of the ftsL* mutant, we noticed that one side of the cell constricted faster than the other (Fig. 2c). When we compared the constriction velocity for each side of the division site and assessed the degree of anisotropy (Extended Data Fig. 4a), ftsL* cells showed a higher but not statistically significant anisotropy score for both IM and OM constriction compared with other strains (Extended Data Fig. 7a-d).
Additionally, when cells were imaged in vertical orientation, the constriction of ftsL* cells was less isotropic than in wild-type cells, indicative of uneven closure of the division ring (Extended Data Fig. 7e). Thus, circumvention of the normal controls regulating sPG biogenesis in the ftsL* mutant results in aberrant division site geometry and abnormal thickening of the envelope at the poles. These cells also lack an observable sPG wedge, which may destabilize the division site and help explain why these mutants were originally found to lyse at elevated temperatures 35,36 . sPG degradation activates its synthesis. To better understand the mechanism(s) by which changes in division site architecture are caused by mutations altering divisome components, we measured the rates of sPG synthesis and degradation using two different cytological assays ( Fig. 3a and Extended Data Fig. 8a). The first assay used a pair of compatibly labelled fluorescent d-amino acids (FDAAs), YADA and HADA 37 , and the other used HADA and MurNAc-alkyne 20 . In both cases, cells were labelled extensively with an FDAA, pulsed with the second label for different lengths of time  Fig. 8l,m. c,d, Mean fluorescence intensity was measured at the division site for new (c) and old (d) PG. c, Data were fit to a linear regression to derive sPG synthesis rates. Data points represent median ± 95% confidence intervals. d, Reduction in old (YADA) fluorescence intensity was fit to a one-phase exponential decay curve. e, Mean septal PG hydrolysis rates were derived from decay curves in d. Points represent the average value of the three biological replicates and bars indicate mean + 1 s.d. N = 1,054 (wt), 716 (ftsN-∆SPOR), 819 (∆envC), 880 (ftsL*) cells. f, Side wall incorporation of new cell wall material (HADA fluorescence intensity) was measured after 8 min due to low signal intensities in earlier time points. Black line indicates median, one-way ANOVA with Dunnett's correction for multiple comparisons, significance of difference is tested relative to wild type; NS, P = 0.06; ***P < 0.001, ****P < 0.0001; N = 103 (wt), 107 (ftsN-∆SPOR), 101 (∆envC), 100 (ftsL*) cells. g, The ratio between sPG and side wall synthesis was calculated by dividing the mean HADA fluorescence intensity after the 8 min pulse. h-j, Labelling patterns observed for the pulse-chase experiment in cells with inhibited division by SulA expression (h). New cell wall material is labelled with Alexa488-labelled MurNAc-alkyne probes (yellow), while old material is stained with HADA (blue). Mean fluorescence intensity was measured along the side wall for both MurNAc-alkyne (i) and HADA (j) and fitted to a quadratic exponential Malthusian exponential growth function (i) or one-phase exponential decay (j). Data points represent median ± 95% confidence intervals. N = 578 (wt), 456 (ftsN-∆SPOR), 427 (∆envC), 501 (ftsL*) cells. k, Representative images from 3 biological replicates of indicated strains after 15, 20 and 30 min pulses with MurNAc-alkyne. Scale bar, 2 µm. and then fixed before visualization. The intensity of the second label appearing at midcell after the pulse was used as a proxy for sPG synthesis. Additionally, the signal intensity of the first label before and after the pulse was used as a proxy for sPG degradation. Both assays yielded qualitatively similar results ( Fig. 3a-  of FtsWI 20 . Notably, the dual FDAA assay detected an increased amount of old PG at the division sites before the HADA pulse, and this material appeared to be relatively stable during the time course. Additionally, bright foci of old material were also observed at the poles of many cells after extended YADA labelling ( Fig. 3b and Extended Data Fig. 8h,i). This accumulation of old material probably corresponds to the thickened areas of cell wall in nascent poles observed by cryo-ET of the ftsL* mutant (Extended Data Figs. 5 and 8j,k), reinforcing the conclusion that short-circuiting the normal controls governing sPG biogenesis not only leads to more rapid sPG synthesis and septal closure, but also aberrant accumulation of PG within the developing poles. Both the ftsN-∆SPOR and ∆envC mutants displayed a reduced rate of sPG synthesis relative to wild-type cells (Fig. 3b,c and Extended Data Fig. 8c,d). Although the sPG synthesis rates were similar, the two mutants differed in their rates of sPG degradation. The ftsN-∆SPOR mutant displayed relatively normal rates of sPG degradation, whereas ∆envC cells showed reduced turnover of sPG ΔenvC ftsL* as expected for a mutant lacking an amidase activator (Fig. 3d,e and Extended Data Fig. 8e). The combination of slower sPG synthesis with normal sPG degradation explains the well-coordinated constriction phenotype displayed by the ftsN-∆SPOR mutant in the cryo-ET analysis. The reduced rate of sPG synthesis in the ∆envC cells is notable because it indicates that proper sPG processing by the amidases is required for normal rates of sPG synthesis. This result along with the reduced rate of sPG synthesis observed for the ftsN-∆SPOR mutant provides strong support for the sPG loop model 28 .
sPG degradation is required for normal Z-ring formation. Cells lacking EnvC commonly displayed closely spaced sPG labelling consistent with the aberrant formation of adjacent division sites (Fig. 4a). Accordingly, closer examination of the localization of Pal-mCherry in these cells revealed that double bands of the OM constriction marker occurred at an elevated frequency over wild-type or ftsN-∆SPOR cells (Fig. 4b,e,f). These double constrictions were also observed in cryo-ET ( Fig. 4d and Extended Data Fig. 5) and were typically placed within the cell body. We also observed constrictions near cell poles, generating what appeared to be minicells (Fig. 4c,g). However, free minicells were not observed in the culture, suggesting that these aberrant poles were probably generated from a double constriction event within the cell body, one of which was aborted while the other completed division, generating a daughter with a polar constriction. Membrane blebs were also observed emanating from some of the developing septa of ∆envC cells, some of which appeared to lyse, suggesting there was a catastrophic failure in division.
The pattern of ZipA-sfGFP was also altered in the ∆envC mutant ( Fig. 4h and Supplementary Video 6). ZipA binds FtsZ and is a Z-ring marker 38 . Many ZipA-sfGFP structures in ∆envC cells were diffuse and/or malformed, indicating a difficulty condensing into the tight Z-ring structure typical of normal cells ( Fig. 4h and Extended Data Fig. 9). Consistent with this possibility, averaging the ZipA-sfGFP signals for a population of cells, either over a 10 min time window or over a 2 µm volume spanning midcell, showed that the fluorescence was more broadly distributed in the ∆envC mutant than in wild type or the other mutants ( Fig. 4i,j). Notably, this phenotype was not suppressed by combining the ftsL* mutation with ∆envC, indicating that it probably stems from the loss of sPG processing, not its collateral effect of reducing sPG synthesis ( Fig. 4h-j). A diffuse Z-ring phenotype has been observed for cells defective in FtsZ-binding proteins such as ZapA that are thought to bundle FtsZ polymers to condense the ring 39,40 . These results therefore suggest a previously unappreciated role for sPG hydrolysis by the amidases in Z-ring condensation and division site stability and/or placement.
Competition between elongation and sPG biogenesis. We took advantage of the PG labelling assays to quantify side wall PG synthesis ( Fig. 3f and Extended Data Fig. 8f) and found that it was inversely correlated with sPG synthesis. Side wall PG incorporation was highest in the ftsN-∆SPOR mutant, which had one of the lowest rates of sPG synthesis (Fig. 3c,f,g and Extended Data Fig. 8d,f,g). Conversely, side wall PG synthesis was lowest in the ftsL* mutant that made sPG most rapidly (Fig. 3c,f,g and Extended Data Fig. 8d,f,g). In support of a competition with cell division being responsible for the differing rates of side wall PG synthesis, the rates were found to be the same in all cells when cell division was blocked (Fig. 3h-k and Extended Data Fig. 8n-q).
Another measure of cell elongation activity is the circumferential motion of the Rod complex associated cytoskeletal element MreB around the cell cylinder [41][42][43] . We tracked the motion of an mNeon-Green fusion to MreB in wild-type and mutant cells using a combination of structured Illumination microscopy and total internal reflection fluorescence (SIM-TIRF) imaging. Consistent with the sPG synthesis measurements, the total number of directionally moving MreB filaments per area was significantly reduced in ftsL* cells (Fig. 5a,b, Extended Data Fig. 10a,b and Supplementary Video 7), which had an increased cell width (Extended Data Fig. 10c,d) indicative of reduced Rod complex activity 44 . All mutants displayed a similar density of directionally moving MreB filaments following the inhibition of cell division (Extended Data Fig. 10e,f and Supplementary Video 7), providing further support for a competition between the processes of elongation and division.
Notably, the interplay between cell elongation and division impacted the geometry of the division site and the shape of the daughter cell poles (Fig. 5c,d). The ftsN-∆SPOR mutant, which elongates more rapidly and constricts slower, displayed an elongated division site and a shallower OM invagination angle at midcell as compared with wild-type cells (Fig. 5e-g). This altered constriction geometry was also observable by cryo-ET and correspondingly gave rise to daughter cells with pointier poles than wild-type cells ( Fig. 5c-g). On the other hand, the rapidly constricting ftsL* mutant formed daughter cells with relatively blunt cell poles ( Fig. 5c-g).
We reasoned that the variation in division site and polar geometry among the different strains could be related to the activity of the Rod complex at or near the division site. The number of directionally moving MreB filaments in proximity (≤200 nm) to cell constrictions was therefore quantified (Fig. 5a,b and Supplementary Video 8). Such filaments were readily observed to pass through division sites in both early and late pre-divisional cells in all strains tested. Notably, however, the ftsN-∆SPOR mutant displayed more MreB tracks at the division site at late stages of division than all other strains, and the ftsL* mutant showed the least number of total MreB tracks at the division site (Fig. 5b). Thus, the density of MreB tracks at the division site for these cells correlates well with the steepness of the constriction site and the extent of cell pole elongation observed for the different strains. The outlier was the ∆envC mutant, which had an inverted trend of having fewer directionally moving MreB tracks at early division stages than at later points (Fig. 5b). We suspect that this change is due to the defect in sPG splitting, which causes a steep curvature of the inner membrane at early points in division that is probably unfavourable for MreB localization 45,46 . However, at later stages when sPG processing eventually allows for slow constriction of the OM, this curvature probably becomes more favourable for MreB localization, allowing elongation to occur near the division site to generate a shallow constriction such as that of the ftsN-∆SPOR mutant. Overall, these results not only provide strong support for a competition between the PG biosynthetic machineries involved in cell elongation and division, but also highlight the potential for this competition to define the morphology of the daughter cell poles.

Discussion
Architecture of the sPG layer. Here we combined cryo-FIB milling with cryo-ET to visualize the division site of E. coli in situ. In cells just starting to constrict, all three envelope layers appeared to be invaginating in concert, and little change in the sPG relative to the side wall PG was evident. However, the speed of IM invagination and sPG synthesis increases faster than PG splitting and OM constriction, leading to the formation of a partial septum (Fig. 6a) similar to that previously observed in fixed samples 9,10 . In NAD-filtered tomograms, a triangular wedge of what is likely to be sPG is observed at the lagging edge of the septum closest to the tip of the invaginating OM (Fig. 6a). The wedge thins as it approaches the leading edge of the closing IM. In this narrow portion of the septum, two dense tracks of material are often discernible, which correspond to the PG layers that will eventually fortify the daughter cell poles. In ftsN-∆SPOR cells with reduced sPG synthesis activity and slower IM constriction, a more uniform constriction of all envelope layers is observed, generating a division site architecture that resembles that of C. crescentus 33 (Fig. 6b). However, in cells defective in sPG splitting, OM constriction is almost completely blocked and a Gram-positive-like septum is formed, with two visible tracks of PG reminiscent of the two tracks observed in the developing septa of Staphylococcus aureus 7 (Fig. 6b). These results suggest that the activity of the same basic cell division machinery can generate different septal architectures observed in diverse bacteria. All that may be required is to change the relative activities of the sPG synthesis and remodelling systems. In cells defective in sPG processing by the amidases, the sPG wedge structure is more prominent than in wild-type cells and it appears to impede the invagination of the OM. We thus infer that amidases process this structure to allow constriction of the OM (Fig. 6a). Furthermore, because the sPG wedge is observed in deeply constricted wild-type cells as well as unconstricted amidase activation mutants, we suspect that the structure is dynamic, with its lagging edge being degraded as new wedge material is deposited at the leading edge. Such a spatial separation of synthesis and degradation would allow the sPG wedge to move in a treadmill-like fashion ahead of the OM as the septum closes.
The enzymes responsible for creating the sPG wedge remain to be identified, but our results with the ftsL* mutant suggest that it is not made by FtsWI. This mutant is thought to hyperactivate FtsWI 20,24-27 . Therefore, if the wedge were produced by the FtsWI synthase, the ftsL* mutant would be expected to produce a thicker or otherwise larger wedge. Instead, it lacks a wedge altogether, suggesting that enhanced FtsWI activity disrupts biogenesis of the sPG wedge by other synthases. An attractive candidate for this additional synthase is the class A penicillin-binding protein (aPBP) PBP1b. Inactivation of PBP1b is synthetically lethal with defects in FtsWI activation. The affected mutants were found to lyse due to septal lesions, suggesting that this aPBP promotes division site stability 24,47 . The location of the wedge at the lagging edge of the division site closest to the OM is also consistent with a role for PBP1b in its construction, given that this enzyme requires an OM lipoprotein for activity 48,49 . Thus, the outer fork of the division site where the wedge is located is the only place where aPBPs would be predicted to be functional. Although further work will be required to test this model, it provides an attractive explanation for the division of labour between the aPBP and FtsWI synthases during constriction, with the FtsWI synthase promoting ingrowth of the PG layer and the aPBPs providing backfill to stabilize the septum and prevent lysis.

The sPG activation loop.
Our results support the proposal that FtsN and the amidases cooperate in a positive feedback loop that promotes sPG synthesis 28  coli divides via a mixed constriction-septation mechanism in which a partial septum with two discernible plates of sPG is formed at later stages of the division process. A wedge structure is observable at the lagging edge of the septum where the dual layers of PG signal of the developing septum meet the single-layered signal of the side wall. Although not clearly resolved in the tomograms, we assume that the two layers of PG signal within the septum are probably connected by additional PG material (drawn as crosshatches). b, A constrictive mode of cell division is observed for the ftsN-∆SPOR mutant, where OM and IM invaginate at similar velocities due to lower sPG synthesis rates. The result is a V-shaped constriction that is similar to that formed by the distantly related Gram-negative bacterium C. crescentus. In contrast, inhibition of sPG hydrolysis causes a temporal separation of IM and OM constriction, leading to septation. These septa as well as the partial septa in wild-type cells are reminiscent of the Gram-positive bacterium S. aureus, which also displays two distinctive plates of sPG within its septa. c, The activities of the two major synthetic cell wall machineries, the Rod complex and the divisome, are anti-correlated probably due to competition for limited substrate (lipid II). The balance of their relative activities determines the shape of the cell division site and the resulting poles they form. Cells with higher Rod complex activity are thinner and form pointier poles, while cells with elevated divisome activity are shorter and wider, with blunt poles.
to be important for normal septal architecture. The ftsL* mutant hyperactivates the FtsWI synthase and eliminates the strict FtsN requirement for sPG biogenesis 26 . This short-circuiting of the normal division activation pathway not only causes the loss of the sPG wedge structure, but also promotes the aberrant accumulation of PG within the developing poles. Whether this accumulation results from inappropriate activation of PG synthesis by FtsWI or PBP1b, the improper turnover of the deposited material, or some combination of the two remains unknown. Nevertheless, what is clear is that bypassing the normal controls involved in sPG activation has adverse consequences on the architecture of the poles that are formed. We therefore infer that the normal divisome activation pathway serves an important function in coordinating different activities of the machinery to ensure that division is successfully completed once it is initiated and that the polar end products have a uniform surface.

PG hydrolysis and the Z-ring.
Our results have uncovered an unexpected connection between the activation of sPG processing by the amidases and the Z-ring structure, suggesting that there is feedback to the Z-ring from events downstream of sPG synthesis activation (Fig. 6a). Z-rings were found to be poorly condensed in mutant cells lacking the amidase activator EnvC (Fig. 4h-j). Additionally, closely spaced constrictions or areas of sPG biogenesis were also observed at an elevated frequency in these cells, suggesting that division sites are unstable and fail before they complete the division process ( Fig. 4a-g). Taken together, these results suggest the counterintuitive notion that sPG degradation by the amidases is required to stabilize the divisome, most probably via a positive influence on Z-ring condensation. Given that the amidases act on sPG in the periplasm, they are unlikely to directly modulate FtsZ activity. Rather, their effect is probably mediated through SPOR domain proteins such as FtsN and DedD that bind the amidase-processed glycans 29,30 . These proteins have transmembrane domains and N-terminal cytoplasmic tails, which in the case of FtsN is known to associate with the FtsZ-binding protein FtsA 50,51 . Thus, the status of sPG biogenesis in the periplasm could be communicated to the Z-ring in the cytoplasm using the binding of SPOR domain proteins to denuded glycans as a proxy. Whether the effect might be mediated simply by concentrating the cytoplasmic domains of SPOR proteins at the division site to modulate the activity of FtsZ-binding proteins or via more complex mechanisms requires further investigation, but the emerging picture is that the divisome activation pathway is not a one-way street from Z-ring formation to sPG synthesis and processing. The Z-ring probably also receives return stabilizing/activating signals from the PG biogenesis machinery.

Cell shape and the balance between division and elongation.
The idea that the cell elongation and division machineries may be in competition with one another has been discussed in the field for some time 52,53 . However, it has only been recently that evidence for such a completion has been presented 47,52,53 . Here we used several independent assays to demonstrate that septal and side wall PG synthesis rates are inversely correlated to each other, providing strong support for antagonism between the activities of the elongation and division systems, which most probably stems from a competition for the limited supply of the lipid II PG precursor. Importantly, our results indicate that this competition does not just affect cell width or how long or short cells are. It also influences the geometry of the septum and the shape of the daughter cell poles. Thus, modulation of the relative activities of the elongation and division systems is likely to play an important role in generating the diversity of shapes observed among different bacteria.

Methods
Media, bacterial strains and mutagenesis. Indicated strain derivatives of E. coli MG1655 used in this study are listed in Supplementary Tables 4 and 5. Bacteria were grown in LB (1% Tryptone, 0.5% yeast extract, 0.5% NaCl) or M9 media 54 each supplemented with 0.2% d-glucose and casamino acids. For selection, antibiotics were used at 10 µg ml −1 (tetracycline), 25 µg ml −1 (chloramphenicol) and 50 µg ml −1 (kanamycin, ampicillin). Mutant alleles were moved between strains using phage P1 transduction. If necessary, the antibiotic cassette was removed using FLP recombinase expressed from pCP20 55 . All mutagenesis procedures were confirmed by PCR. Fig. 1 summarizes the cryo-FIB/ cryo-ET pipeline utilized in this study. Bacterial strains were grown overnight in LB media, back diluted 1:1,000 and incubated with shaking at 37 °C and 250 r.p.m. to optical density (OD) 600 = 0.3. Cells were collected by centrifugation (2 min, 5,000 × g, r.t.) and resuspended in LB media to a final OD 600 = 0.6. This cell suspension (3 µl) was applied to Cflat-2/1 200 mesh copper or gold grids (Electron Microscopy Sciences) that were glow discharged for 30 s at 15 mA. Grids were plunge-frozen in liquid ethane 56 with an FEI Vitrobot Mark IV (Thermo Fisher Scientific) at r.t., 100% humidity with a waiting time of 10 s, one-side blotting time of 13 s and blotting force of 10. Customized parafilm sheets were used for one-side blotting. All subsequent grid handling and transfers were performed in liquid nitrogen. Grids were clipped onto cryo-FIB autogrids (Thermo Fisher Scientific).

Cryo-EM specimen preparation. Extended Data
Cryo-FIB milling. Grids were loaded in an Aquilos 2 Cryo-FIB (Thermo Fisher Scientific). The specimen was sputter coated inside the cryo-FIB chamber with inorganic platinum, and an integrated gas injection system was used to deposit an organometallic platinum layer to protect the specimen surface and avoid uneven thinning of cells. Cryo-FIB milling was performed on the specimen using two rectangular patterns to mill top and bottom parts of cells, and two extra rectangular patterns were used to create micro-expansion joints to improve lamellae instability 57 . Cryo-FIB milling was performed at a nominal tilt angle of 14°−18°, which translates into a milling angle of 7°−11°5 8 . Cryo-FIB milling was performed in several steps of decreasing ion beam currents ranging from 0.5 nA to 10 pA and decreasing thickness to obtain 100-200 nm lamellae.

Cryo-ET.
All imaging was done on an FEI Titan Krios (Thermo Fisher Scientific) transmission electron microscope operated at 300 KeV and equipped with a Gatan BioQuantum K3 energy filter (20 eV zero-loss filtering) and a Gatan K3 direct electron detector. Before data acquisition, a full K3 gain reference was acquired, and ZLP and BioQuantum energy filters were finely tuned. The nominal magnification for data collection was ×42,000 or ×33,000, giving a calibrated 4 K pixel size of 2.193 Å and 2.565/2.758 Å, respectively. Data collection was performed in the nanoprobe mode using the SerialEM 59 or Thermo Scientific Tomography 5.3 software. The tilt range varied depending on the lamella, but was generally from −70° to 70° in 2° steps following the dose-symmetric tilt scheme 60 Table 1).
Cryo-ET image processing. Acquired tilted super-resolution movies were motion corrected and Fourier cropped to 4 K × 5 K stacks, using 'framealign' from IMOD 61 . Tilt series were aligned using 'etomo' in IMOD 62 and 'Dynamo' . Contrast transfer function (CTF) estimation was performed in IMOD. CTF correction was performed using the 'ctfphaseflip' programme in IMOD 63 . CTF-corrected unbinned tomograms were reconstructed by weighted back projection with and without a SIRT-like filter and subsequently 2x, 4x and 8x binned in IMOD 62 . Bandpass filtering and summed projection of cryo-tomogram slices were performed in Dynamo 64-67 complemented with customized MATLAB scripts. Gaussian and NAD-filtering were performed in Amira (Thermo Fisher Scientific) for visualization purposes. NAD-filtering was applied using the command ' Anisotropic Diffusion' in 3D mode for 5 iterations. Gaussian filtering was done by applying the command 'Gaussian Filter' under 3D mode with a kernel size factor of 3. Whole 3D-volume FFT filtering was performed in IMOD.
Segmentation. Segmentation was performed on FFT filtered and NAD-filtered tomograms using Amira (Thermo Fisher Scientific) by non-biased semi-automatic approaches. Manual annotation was required every 10 slices, then Amira's interpolation function was applied to automatically trace slices in between. Annotation was done in two-dimensional (2D) slices where features of interest were visible by eye. The segmented PG signal is not indicative of specific glycan strand network, but rather serves as a visual guide to relevant cell wall features.
Curvature. Three-dimensional pole curvature rendering was performed in Amira by applying the command 'Curvature' on the basis of the triangulated 3D mesh and 'Shape Index' as implemented in Amira 68 . Shape index (SI) computes the surface scalar field, which is calculated as where C 1 and C 2 are the two principal curvatures. Shape index ranges from −1 to 1, negative values indicate negative curvature, positive values indicate positive curvature and values close to 0 indicate flatness of the surface. Values are normalized with respect to neighbouring triangles' SI values 68 (Fig. 5d).

Quantification of cryo-ET data. Division site dimensions.
Summed projection images of cryo-ET tomograms were used to quantitatively measure cell dimensions at the division site 69 . Measurements were performed in Fiji 70 using the 'point to point' measuring tool. Measurements were from IM to IM and from OM to OM.
Periplasmic space. Measurements of periplasmic space thickness were performed from the centre of the OM to the centre of the IM in the cell areas referred to here as 'side wall' , 'pole' and 'curve' as well as the invagination tip of the OM to the IM at the constriction division stage. Measurements from centre to centre of opposing IMs were performed in the cell area defined in this study as the septum (Supplementary Figs. 2 and 5). We used a customized macro in Fiji that measures 30 Euclidean distances from surface-to-surface areas 71 in nm, for example, from IM to IM at the septum and from IM to OM at the rest of the areas (side wall, pole, curve and initiation). For these 30 single measurements, the mean was calculated, yielding a final single value per defined subcellular localization, for example, septum, curve, pole and side wall.
From the full wild-type cryo-ET data set, particles were identified using 'dtmslice' interface in Dynamo 66,67,72 . In 4x-binned tomograms, subtomograms with a size of (777.6) 3 Å were extracted from 4x-binned tomograms. Initial angles were assigned following the normal of the IM. A starting reference generated from a random set of particles was used for both side wall and septum particles. A total of 16 iterations were used to align particles and obtain final averages. Final averages were generated from 8,076 subtomograms for the side wall and 212 particles for the septum. Notice that side wall regions were much more abundant in the cell than septum regions. EM densities were visualized in Chimera 73 .
Sample preparation for live cell imaging. Overnight cultures of indicated E. coli strains were grown in LB supplemented with appropriate antibiotics at 37 °C. The next day, cells were collected by centrifugation (2 min, 5,000 × g, r.t.) and washed 2× with M9 medium. Day cultures were back diluted (1:1,000) and grown in M9 (0.2% d-glucose, 0.2% casamino acids) supplemented with 50 µM Isopropyl β-D-1-thiogalactopyranoside (IPTG) and appropriate antibiotics at 30 °C until OD 600 = 0.2-0.4. For filamentation experiments, SulA was produced from pNP146 74 by the addition of 0.2% l-arabinose during the last 10 min of the incubation period. Cells were collected (2 min, 5,000 × g, r.t.) and resuspended in 1/10th of the original volume. Two microlitres of this cell suspension were added onto a 1% (w/v) agarose in M9 (0.2% d-glucose, casamino acids) pad supplemented with 50 µM IPTG and covered with a #1.5 coverslip. For filamentation experiments, the agar pad was also supplemented with 0.2% l-arabinose.
Live-cell imaging. All samples were imaged on a Nikon Ti-E inverted widefield microscope equipped with a fully motorized stage and perfect focus system. Images were acquired using a 1.45 NA Plan Apo ×100 Ph3 DM objective lens with Cargille Type 37 immersion oil. Fluorescence was excited using a Lumencore SpectraX LED light engine and filtered using ET-GFP (Chroma, 49002) and ET-mCherry (Chroma, 49008) filter sets. Images were recorded on an Andor Zyla 4.2 Plus sCMOS camera (65 nm pixel size) using Nikon Elements (v5.10) acquisition software. For subsequent deconvolution procedures, three 200 nm spaced Z-planes were acquired for both fluorescence channels using 100% LED output power and 50 ms exposure. Temperature was maintained at 30 °C using a custom-made environmental enclosure. After a 20 min acclimatization period, cells were imaged at a 2.5 min acquisition frame rate for a total observation time of 1-4 h.
Image processing for fluorescence microscopy. First, time-lapse series and Z-stacks were drift corrected using a customized StackReg plugin in Fiji 70,75 . Subsequently, fluorescence images were deconvolved using the classical maximum likelihood estimation algorithm in Huygens Essential v19.10 (SVI), employing an experimentally derived point spread function (PSF) from 100 nm TetraSpeck beads (Thermo Fisher Scientific). Image reconstruction was performed over 50 iterations with a quality threshold of 0.01 and a signal-to-noise ratio set to 20 for live-cell imaging and 40 for fluorescent cell wall probes in fixed samples. Background removal was set to 0 to preserve fluorescence intensity values best among different images. Chromatic aberrations between different fluorescent wavelengths were post-corrected using the chromatic aberration corrector in Huygens from the TetraSpeck bead template. The same image reconstruction parameters and chromatic aberration templates were applied to images that were compared to each other. Last, reconstructed fluorescence images were merged back to phase-contrast images and rendered for figure or movie display with Fiji.
Measuring cell envelope constriction dynamics. Fluorescent fusions to IM-anchored protein ZipA and OM-lipoprotein Pal allowed us to determine the respective positions of the different cell envelope layers during division. These cell envelope fiducial markers accumulate specifically during cytokinesis at the division site, which was critical for the generation of kymographs. Constriction dynamics of IM and OM were derived from kymographs generated using the Fiji plugin 'KymographClear' 76 and automatically split into forward and reverse trajectories using Fourier filtering. This filtering step allowed us to measure the constriction rate for each side independently. Constriction kinetics were derived by automatically extracting the fluorescent trajectories for ZipA and Pal using 'KymographDirect' 76 (Extended Data Fig. 4a). Anisotropy of the division process was determined by taking the ratio of the constriction velocities between the forward and reverse trajectories. Only cells where the division site displayed minimal signs of displacement except for constriction were analysed to eliminate confounding effects on the analysis by excessive cell movement (for example, pushing). This manual exclusion resulted in the rejection of approximately 15-20% of the cell division events. Applying these procedures, we found the constriction rate of the OM to be increasing over time, in contrast to a previously reported constant rate 77 . This might be explained by different image analysis procedures (for example, kymographs vs width measurements, Pal-mCh marker vs a combination of phase-contrast and FM4-64 dyes).

Measuring division site circularity of vertically imaged cells.
For vertical imaging of bacterial cells undergoing division, similar procedures as described previously 78,79 were applied. A silicon wafer containing 5.5 µm long and 1.5 µm wide photo-resist pillars was generated following high aspect ratio photolithography procedures with an adhesion layer. The dimension of these pillars reaches the practically feasible aspect ratio for photolithography designs and thus impedes increasing pillar length without concomitantly increasing width, precluding use of elongated or chaining division mutants for this imaging mode. A modified silanization surface treatment with plasma cleaning was applied to increase the surface hydrophobicity of the silicon wafer to minimize agarose accumulation. Agarose micro holes were generated by pouring degassed 6% agarose (w/v) in H 2 O on the silicon wafer. Agarose was allowed to solidify for 40 min at r.t., was peeled off, cut into 5 ×5 mm pieces, and incubated in M9 medium supplemented with 0.2% d-glucose, casamino acids, 25 µg ml −1 chloramphenicol and 50 µg ml −1 ampicillin overnight. Cells were grown as described for sample preparation for live-cell imaging and added on agarose pads. Cells that were not trapped in micro holes were washed off gently using 1 ml of growth medium. Five micrometre spanning Z-stacks (at a 200 nm step size) were acquired and subsequently deconvolved.
Circularity quantification was carried using the software package 'Morphometrics' 80 . Fluorescence signals were segmented using Laplacian algorithm in combination with the peripheral fluorescence setting. Circularity (C) is calculated in Morphometrics as: where P is the perimeter and A is the area enclosed by the circle and is a dimensionless measure. A perfect circle displays a circularity of 1, while increasing values correspond to less circular objects. Cells that were trapped tilted in agar holes were manually excluded from the analysis (15 out of 573 analysed cells).
Measuring Z-ring condensation from time-lapse data. Condensation of cytoskeletal elements was addressed using previously described procedures 40 . Briefly, five frames (corresponding to 10 min) from recorded time-lapse series were sum-projected in Fiji. Z-rings in these sum-projected images were then aligned along the length axis and average-intensity-projected into a single image. Fluorescence intensity was measured across the full width along the horizontal axis of the averaged projection image. Intensity values were normalized and their corresponding full width at half maximum (FWHM) values were calculated in MATLAB.
Measuring Z-ring condensation from 3D data. Similar procedures as outlined for measuring Z-ring condensation in time-lapse series were applied. Two micrometre spanning Z-stacks (at a 200 nm step size) were acquired to capture a full 3D view of a cell. Images were restored in Huygens as described above. Image volumes were sum-projected into a single plane, Z-rings extracted, aligned and averaged as described above. Fluorescence intensity profiles were measured identically as for time-lapse data. Snapshots for 3D maximum intensity projections were rendered in Huygens.

Measuring cell wall synthesis rates by biorthogonal MurNAc-alkyne probes.
Septal cell wall synthesis rates were measured as described previously 81,82 . MurNAc-alkyne was purchased as a custom synthesis product from Tocris following the procedures of ref. 81 . All experiments were carried out in ∆murQ background and in the presence of pCF436 83 for IPTG-inducible expression of AmgK and MurU. Filamentation was induced by expressing the FtsZ antagonist sulA from arabinose-inducible plasmid pNP146 74 . Overnight cultures were back diluted 1:1,000 into fresh LB containing 15 µg ml −1 gentamycin. Cells were grown at 37 °C until OD 600 = 0.4. Subsequently, 1.5 ml of cells were collected (2 min, 5,000 × g, r.t.) and resuspended in 300 µl LB containing 1 mM IPTG and 0.5 mM HADA to label all cell wall material with FDAAs. For filamentation experiments, SulA expression was induced by the addition of 0.2% l-arabinose. Samples were incubated by rotating at 37 °C for 30 min. Endogenous UDP-MurNAc production was inhibited by the addition of 200 µg ml −1 fosfomycin. After 10 min incubation, cells were washed twice in 1.5 ml LB, 1 mM IPTG and 200 µg ml −1 fosfomycin. Next, cells were incubated for 15 min in the presence of 0.2% (w/v) MurNAc-alkyne, 1 mM IPTG and 200 µg ml −1 fosfomycin at 37 °C. Cells were fixed using ice-cold 70% (w/v) ethanol for 20 min at 4 °C. Next, cell pellets were washed 3× with 1x PBS. Biorthogonal MurNAc-alkyne probes were labelled by click chemistry using 5 µM Alexa488 azide substrate according to the manufacturer's instruction. Samples were stored in 20 µl PBS at 4 °C and imaged within 48 h of the labelling experiment.
One micrometre spanning Z-stacks (separated by 200 nm) were acquired and subsequently deconvolved as described above. Z-stacks were sum-projected using Fiji. De novo septal PG synthesis was assessed by measuring the mean fluorescence intensity of NAM-Alexa488 along the division site using the line tool (width, 3 pixels). Levels of cell wall hydrolysis were assessed by measuring the overall reduction in HADA fluorescence as compared to baseline signal intensity derived from fixing cells before MurNAc-alkyne chase. Reduction in fluorescence intensity of FDAAs is indicative of cell wall remodelling mediated by amidases, endopeptidases or transglycosylases.

Measuring cell wall remodelling by FDAA incorporation.
For FDAA pulse-chase experiments, cells grown overnight were back diluted 1:1,000 in fresh LB and grown until OD 600 = 0.4 at 37 °C. For the filamentation experiment, sulA was expressed from pNP146 74 by the addition of 0.2% l-arabinose to cultures during the last 10 min of the incubation period. Subsequently, 1.5 ml of cells were collected (2 min, 5,000 × g, r.t.) and resuspended in 300 µl LB containing 0.5 mM YADA. Samples were incubated while rotating at 37 °C for 40 min. Cells were washed once in 1.5 ml LB and resuspended in 300 µl LB containing 0.5 mM HADA. Samples were incubated at 37 °C for either 2 min, 4 min or 8 min before immediate fixation with 70% ethanol. After fixation, cells were washed 3× in PBS, stored in the dark at 4 °C and imaged within 48 h. The same image acquisition and analyses procedures were carried out as described for MurNAc-alkyne probes. Fluorescence intensity values were fit to a linear regression for HADA and an exponential one-phase decay for YADA. Levels of cell wall hydrolysis were assessed by subtracting the average fluorescence intensity from cells fixed before chase (0 min) and the respective time point, and fit to a linear regression model. Reduction in fluorescence intensity of FDAAs is indicative of cell wall remodelling mediated by amidases, endopeptidases or transglycosylases. In addition to the division site, fluorescence intensity measurements were also performed along the side wall and polar region of the cells at the 8 min time point. For filamenting cells, HADA fluorescence intensity values were fit to a Malthusian exponential equation, assuming cells keep elongation at the same rate before SulA induction.
Bulk growth curve measurements. Overnight cultures of indicated E. coli strains were grown in LB supplemented with appropriate antibiotics at 37 °C. The next day, cells were collected by centrifugation (2 min, 5,000 × g, r.t.) and washed 2× with the respective growth medium (M9 or LB). Day cultures were back diluted (1:1,000) and grown in the respective media supplemented with corresponding IPTG concentration (50 µM for ZipA-sfGFP induction, 1 mM for AmgK/MurU expression) and appropriate antibiotics at 30 °C until OD 600 = 0.3. Cells were collected (2 min, 5,000 × g, r.t.) and resuspended to an initial OD 600 of 0.01 in a final volume of 100 µl. Growth curves were measured in a Tecan M-plex 96-well plate reader by OD 600 read out. Plates were incubated with shaking at 30 °C for a total of 18 h.
Cell shape quantification analyses. Bacterial cells were segmented and analysed from phase-contrast images using the software package 'Morphometrics' 80 . Results from Morphometrics were post-processed using customized MATLAB scripts to exclude erroneously segmented cell debris in live-image data on the basis of area. Cell width, length and pole curvature per segmented cell were directly extracted from Morphometrics. Since curvature (k = 1 r , where r is the radius of the cell cylinder) is dependent on the cell cylinder width, curvature values were normalized by multiplying half-cell width to each respective curvature value. Thus, spherical poles display curvature values of k = 1, while pointy (elongated) poles display elevated curvature values (k > 1) and flat (shortened) poles display reduced curvature values k < 1, respectively. We obtained division site curvature from both sides of the cell at the invagination site. The invagination site is defined as the narrowest segment of the cell, for example, lowest cell width value, that presents negative curvature on both sides of the cell body. Division site curvature was normalized to the half-cell width of the invagination site.
SIM-TIRF microscopy and MreB tracking. Samples were prepared as described for live-cell imaging. To block cell division, sulA was expressed from pNP146 74 by the addition of 0.2% l-arabinose during the last 10 min of the incubation period. Cells were added to high precision #1.5 coverslips (Marienfeld) and placed on a 1% (w/v) agarose pad in M9 (0.2% d-glucose, casamino acids, supplemented with 0.2% l-arabinose for filamentation experiments) and imaged at room temperature on a Nikon Ti2 N-SIM microscope equipped with N-SIM spatial light modulator illuminator, TIRF Lun-F laser combiner with 488 and 561 nm laser lines, an N-SIM 488/561 dual band dichroic mirror, SR HP Apo TIRF ×100 1.5 NA oil objective with automated correction collar and a Hamamatsu Orca Flash 4.0 camera attached to a Cairn Research Twimcam splitter with an ET525/50m or an ET605/70m emission filter (for MreB-sw-mNeonGreen or Pal-mCherry fusion, respectively). The refractive index of the immersion oil (1.512) (GE Healthcare) was optimized for MreB-sw-mNeonGreen signal and corrected using the automated correction collar for the Pal-mCherry fusion. Alignment of the 488 and 561 lasers for SIM-TIRF and 3D-SIM, and of the N-SIM optics and illumination was performed before each experiment at the image plane. First, a 3 min time-lapse series (at 3 s acquisition frame rate) in SIM-TIRF mode was collected using 20% laser power with 100 ms exposure time to follow MreB-sw-mNeonGreen dynamics. Then, a single slice of a 3D-SIM Pal-mCherry (40% laser power, 100 ms exposure) and a brightfield reference image was acquired. Raw fluorescence images were reconstructed using Nikon Elements 5.11 acquisition software with indicated settings: MreB illumination contrast 0.8, noise suppression 0.3 and blur suppression 0.05; Pal illumination contrast 3.75, noise suppression 0.1 and blur suppression 0.5. Only reconstructed images with a quality score ≥8 and passed SIMcheck quality test 84 were used for further analysis. Subsequently, MreB time-lapse series were overlayed over the reference channels in Fiji.
Particle tracking was performed in Fiji using the TrackMate v6.0.1 plugin 85 . MreB filaments were detected using the LoG-detector with an estimated radius of 0.3 µm. Spurious spots were filtered using a quality threshold of 50. Spots were linked using a Kalman filter with an initial search radius of 0.2 µm and search radius of 0.1 µm. No frame gaps were allowed. Only tracks consisting of ≥4 continuous spots (12 s) and that travelled less than 1 µm in total distance were kept for further analysis. To analyse the nature of the displacement of each track, the mean square displacement (MSD) was calculated using the MATLAB class msdanalyzer 86 . Slopes (α) of the individual MSD curves were extracted using the log-log fit of the MSD and the delay time τ. As the maximum delay time of 75% of the track length was used, tracks with an R 2 for log [MSD] versus log[t] below 0.95 indicative of a poor fit to the MSD curve were excluded from the analysis. MreB filaments engaged in active cell wall synthesis are displaced by the enzymatic activities of RodA and PBP2b 41-43 , hence their MSD curves display slopes of α ≈ 2 indicative of a transported particle motion above the rate of Brownian diffusion (Extended Data Fig. 10b). MreB filaments in constricting cells, as determined by the presence of Pal-mCherry foci at the division site, were analysed by fitting a 200 nm wide region of interest to the cell division site. Directional MreB tracks were deemed to contribute to the elongation of the division site. Early and late division stages were distinguished by the presence of two separated Pal foci or a continuous fluorescent signal across the cell, respectively. Statistical analysis. All data measurements were plotted and analysed using GraphPad Prism 9 (Version 9.3.1). In general, (log-) normal distribution was tested using Shapiro-Wilk test. For comparisons of two groups, significance was determined by two-tailed, unpaired Student's t-test with Welch correction and F-test for variance analysis. One-way analysis of variance (ANOVA) was used for comparison of more than two groups using the recommended post-test for selected pairwise comparisons. All experiments were carried out with at least 3 independent biological replicates. P values less than 0.05 were considered statistically significant.
Reporting summary. Further information on research design is available in the Nature Research Reporting Summary linked to this article.