Architecture of the ring formed by the tubulin homologue FtsZ in bacterial cell division

1Membrane dynamics during cytokinesis are some of the most fundamental processes in biology, yet are poorly understood at the molecular and mechanistic level. During prokaryotic cell division the cell membrane and the cell envelope constrict, eventually leading to cell separation. In most bacteria and archaea, this is guided by a ring structure containing the bacterial tubulin homologue FtsZ protein, which polymerises in a GTP-dependent manner. During constriction, the FtsZ ring decreases in diameter through an unknown mechanism. The C-terminal tail of FtsZ links it to other components of the divisome, an ensemble of many proteins that facilitates essential functions during the cell division process, most importantly remodelling of the cell envelope. Components of the divisome engage in cell wall synthesis (PBPs), synchronisation with chromosome dimer resolution (FtsK), lipid II cell wall precursor flipping (FtsW or MurJ), and many components currently have no known function (reviews:; ).

2In Escherichia coli, binding of the FtsZ tail to ZipA and possibly more importantly to FtsA anchor the FtsZ ring to the membrane. FtsA is a bacterial actin-like protein that forms domain-swapped, canonical actin-like protofilaments that are membrane associated through FtsA's C-terminal amphipathic helix. Several cellular regulatory processes influence the onset and progression of cell division through mechanisms that directly act on FtsZ. For example, SulA is induced during the SOS stress response and sequesters monomers, stopping FtsZ polymerisation. In E. coli, both nucleoid occlusion and the oscillating, pole-protecting MinCDE system contain components that inhibit FtsZ function within the ring directly.

3Although progress has been exhilarating over that past 20 years or so, some of the most fundamental questions still remain: what happens during FtsZ ring constriction? How are the filaments arranged in the ring? What drives constriction?

4Many different models have been proposed for the mechanism of FtsZ-based constriction (reviewed in; ). Essentially, three different approaches have been taken to validate the models: in vivo imaging of FtsZ constrictions using fluorescently labelled proteins. Electron cryotomography of frozen hydrated cells without labelling and, thirdly, in vitro reconstitution experiments with pure, fluorescently labelled proteins. The most recent results emanating from those studies are that the rings appear to show strong fluorescence intensity variations that may suggest that the FtsZ ring is discontinuous. Equally, tomography data have been interpreted to show scattered individual FtsZ filaments, some precise distance away from the membrane. Reconstitution experiments with FtsZ and FtsA showed dynamic behaviour and liposome constrictions. However, obtaining detailed molecular and mechanistic information regarding the FtsZ ring, particularly the arrangement of individual filaments and subunits within a constricting Z-ring, has remained a formidable challenge.

5In this study, we obtained high-resolution images of the FtsZ ring in Caulobacter crescentus and Escherichia coli by means of electron cryotomography. Furthermore, we reconstituted a minimal constriction force-generating system from purified components in vitro, encapsulating Thermotoga maritima FtsA (TmFtsA) and FtsZ (TmFtsZ) in liposomes of sizes corresponding to those of a bacterial cell. We produced images and three-dimensional maps of filaments arranging themselves into ring structures around the liposome perimeters that coincided with constriction sites. The observed FtsZ ring architectures in C. crescentus and E. coli cells and in liposomes favour a mechanism of FtsZ-based cell membrane constriction that is accompanied by filament sliding, as was proposed previously.

6We started out by visualising division sites in an unmodified C. crescentus strain (NA1000/CB15N) because the thin Caulobacter cells are most suitable for electron cryotomography. When a log-phase culture was plunge-frozen and imaged, many dividing cells could be found. At the division sites, a series of dots arranged in a single line were found (Figure 1A, top). Careful analysis of cellular tomograms (Video 1 and Figure 1A, bottom) revealed that the dots were in fact 2D projections of filamentous structures encircling the cell and likely forming a continuous ring, disrupted in the images at the top and bottom by the missing wedge of the tomography method. The filaments were at a distance of 15 ± 2 nm from the inner membrane (Figure 1B), as previously reported. The long-standing problem of the missing wedge in electron tomography caused by our current inability to tilt the specimen much beyond 65° (Figure 1A, bottom, white triangle, see also Figure 1—figure supplement 1 for more details on the missing wedge problem related to this study) makes it impossible to follow features all the way around the cell's perimeter. It is important to note, however, that the protein filaments are visible and uninterrupted everywhere the missing wedge allows it, as can be gauged from the disappearance of the cell membrane and envelope (Figure 1A,B). We could detect the filamentous rings in 20 out of 28 dividing cells after tomography. Of the 8 without obvious filamentous structures, the division had progressed too far in 3 and 5 were of poor quality because of the cell's orientation with respect to the tilt axis (Figure 1—figure supplement 1). Hence, the finding of complete rings is supported by the fact that perfect coincidence of any hypothetical gaps in constricting rings with the missing wedges in all of the analysed tomograms, found after all in entirely random orientations, would be remarkably implausible.

7To investigate the generality of these findings, we imaged unmodified E. coli B/r H266 cells, which we chose because of their thinness. Although the E. coli cells were thicker than C. crescentus, we found the wild-type filaments still discernible, most obviously when imaged along the long axis of cells (Figure 1C and Figure 1—figure supplement 2, especially B and D; Video 2), and these were 16 nm away from the IM. Based on these observations we concluded that E. coli Z-rings were, like in C. crescentus, probably continuous and consisted of single-layered bands that are 5–10 filaments wide.

8In order to investigate if the filaments imaged in wild-type cells so far contained FtsZ protein, we over-expressed FtsZ(D212A), a mutant protein that hydrolyses GTP much more slowly in E. coli B/r H266 cells. When over-expressed to 2.5-fold total FtsZ (Figure 1D), the protein formed a wide single layer of filaments at the division site (Figure 1E, see also Video 3 and Figure 1—figure supplement 3), very similar to the bands seen in unmodified cells, but wider and containing more filaments as would be expected because there is now more FtsZ protein in the cell and filament dynamics have been reduced because of the GTPase-reducing mutation D212A. Figure 1F (Figure 1—figure supplement 3A–C) provides a view rotated by 90°, showing again that the filamentous ring was located approximately 16 nm away from the inner membrane (IM). The band of filaments most likely consisted of doublets of individual protofilaments, as is indicated in Figure 1G, which shows a cell with higher expression level (Figure 1—figure supplement 3A–C). The filaments were on average 6.8 nm apart (n = 17, distance between centres of adjacent filaments within a doublet, Figure 1G, lower). It is currently not known what lateral interactions between FtsZ filaments cause this arrangement or if it is facilitated by other proteins.

9Because no specific label for electron cryotomography currently exists that works in E. coli, we decided to further confirm the identity of the filaments as being composed of FtsZ by systematic perturbations of the system in four (a–d) separate experiments with subsequent imaging by electron cryotomography (Figure 1—figure supplement 5, Supplementary file 1A,B). (a) Introducing extra amino acids into the flexible linker within FtsZ that separates the globular N-terminal domain of FtsZ from the small C-terminal helix that binds FtsZ's membrane anchor, FtsA, increased the distance between the FtsZ ring and the IM from 16 nm to a somewhat variable 16–21 nm (Figure 1H and Figure 1—figure supplement 4A–C). (b) Removing the C-terminal FtsA-interacting helix and replacing it with a membrane-targeting sequence (mts) from MinD protein shortened the distance from the IM to 10 nm. No constrictions of the cells were observed (Figure 1I and Figure 1—figure supplement 4D). Therefore, despite it having been used in earlier studies, we agree with previous findings that this construct is non-functional in vivo and we did not include it in our subsequent in vitro investigations below. (c) Removing the C-terminal FtsA-interacting helix from FtsZ detached the filaments from the membrane, making them appear throughout the cytoplasm (Figure 1J and Figure 1—figure supplement 4E). (d) Finally, removing most amino acids between the C-terminal FtsA-interacting helix and the globular body of FtsZ caused a minicell phenotype. Because of their size, minicells produce tomograms of higher quality and it was possible to determine the longitudinal subunit repeat of the filaments to be around 4 nm, very close to the expected value of 4.2 nm for FtsZ (Figure 1K and L and Figure 1—figure supplement 4G). We conclude that the filament localisations reacted to our perturbations as expected for FtsZ and the subunit repeat was the same as for all known FtsZ protofilaments. The minicell tomogram (Figure 1K) is another indication that the filaments most likely encircle entire cells.

10Encouraged by reports that simultaneous over-expression of FtsZ and its membrane anchor FtsA led to additional division sites, we imaged E. coli cells in which extra FtsZ(D212A) and FtsA were produced from a bicistronic expression vector, by electron cryotomography (Figure 2A–E). Providing just these two proteins in excess (twofold to fourfold total vs WT) produced a severe phenotype with many extra constrictions visible (Figure 2A,B). Since in these cells the normal FtsZ to FtsA ratio had been altered to be close to 1:1, from normally 5 FtsZ:1 FtsA, FtsA filaments became visible at the constricting division sites (Figure 2C–E). Actin-like FtsA binds to the membrane directly via its C-terminal amphipathic helix and polymerises into canonical actin-like protofilaments. In this way, an artificially strong ‘FtsA ring’ became apparent, indicating that FtsA is located between the IM and FtsZ, being 8 nm away from both (Figure 2D,E).

11We then demonstrated that the FtsZ(D212A) and FtsA-over-expressing E. coli cells, we had imaged at high-resolution with electron cryotomography, were dividing and separating, despite looking quite distorted. For this, we employed structured illumination microscopy (SIM) on live cells (Figure 2F). The cells showed a strong minicell phenotype, while performing many cell divisions randomly distributed along the cell length. This indicated to us that the extra constrictions and division sites were functional in the sense that they led to complete cell separation (abscission).

12We concluded that the ability to produce extra constriction sites and divisions by just providing more FtsZ and FtsA indicates that these two proteins may be central components of the IM constriction force generator that is localised within the inner division apparatus and that it may be possible to reconstitute membrane constriction with just the two of them. Indeed, this has recently been reported and has been imaged at low resolution using fluorescence microscopy, although not providing any molecular insights.

13So, we then used purified FtsZ and FtsA proteins for in vitro reconstitution experiments, in order to observe constriction. We avoided fluorescently tagged proteins as they have been shown to introduce artefacts. We therefore used completely unmodified TmFtsZ and TmFtsA proteins from Thermotoga maritima, both of which are easy to obtain and handle and have crystal structures available. It should be noted that extra care had to be taken in order to obtain proteins that did not have their disordered but important C-terminal tails cleaved during purification. Just to confirm that TmFtsZ and TmFtsA formed structures similar to the E. coli counterparts in vivo, we over-expressed TmFtsZ and TmFtsA in E. coli and imaged the sample by electron cryotomography (Figure 2G). Minicells were formed and, at constriction sites, they contained filaments that closely resembled the filament arrangement observed here for E. coli FtsA and FtsZ over-expression in E. coli (Figure 2C). The distance of TmFtsZ to the IM was shorter at 12 nm; this was expected because the linker between the very C-terminal TmFtsA-interacting helix and the body of TmFtsZ is much shorter, measuring around nine amino acids. Minicell formation might indicate that TmFtsA and TmFtsZ interacted with the E. coli cell division machinery or even supported membrane constriction on their own, but we did not investigate this further.

14When added onto a flat lipid monolayer, TmFtsZ and TmFtsA formed striking spirals (Figure 3A). The filaments forming the spirals tended to form weak doublets and in the centre of the spirals, white material was visible that may have been lipid that had been pushed up by the spiral constricting, possibly via a sliding filament mechanism as has previously been observed for FtsZ alone by AFM. Intriguingly, much larger dynamic chiral spirals of polar FtsA and FtsZ filaments have recently been reported on supported lipid bilayers, but the exact relationship with our observation is currently unclear as treadmilling and no constriction were observed on the supported bilayers.

15Since the FtsZ ring in vivo does not act on flat membranes, we then switched to liposomes formed from E. coli lipid extract. First, TmFtsZ and TmFtsA were added to pre-formed liposomes so that the proteins remained on the outside (Figure 3B and Figure 3—figure supplement 1A). Reactions containing liposomes and proteins, as indicated, were vitrified and imaged by conventional 2D transmission electron cryomicroscopy. When no protein was added, the liposomes appeared as almost perfect circles (spheres in projection) and the bilayers were clearly visible as a double line, 5 nm apart (Figure 3B, top, Figure 3—figure supplement 2D). The addition of TmFtsA alone led to the formation of an additional layer, probably consisting of only partly polymerised protein, and no strong deformations were observed. TmFtsZ alone did not cause deformations or generation of an additional layer (Figure 3—figure supplement 2B). But when both TmFtsA and TmFtsZ were added, two extra layers, in addition to the liposome bilayer, became visible (Figure 3B). Particularly in the presence of nucleotide, strong negative curvature was induced, leading to deformations of the liposomes (Figure 3—figure supplement 1A). As was suggested previously, the co-polymerisation of FtsZ and FtsA will lead to bending and curvature because the subunit repeat lengths of FtsZ and FtsA are roughly 4 and 5 nm, respectively (Figure 3—figure supplement 1B and Figure 3—figure supplement 2A). We propose that the subunit repeat mismatch causes some deformations from the outside of the liposomes, especially when FtsZ polymerisation-inducing GMPCPP is present so that long filaments are formed that will exert more mechanical force.

16Since TmFtsZ and TmFtsA induce negative membrane curvature, we concluded that in order to reconstitute the actions of these proteins correctly, we needed to incorporate them on the inside of liposomes. For this, CHAPS detergent-solubilised E. coli lipid extract was mixed with the proteins at high concentrations and then diluted many-fold. Lowering of the detergent concentration by dilution led to spontaneous liposome formation with most of the proteins on the inside. We demonstrated with a time-lapse experiment that liposomes did not form around pre-existing FtsZ scaffolds since we observed that most liposomes were initially perfectly spherical and that they deformed over a 30-min period, after which most of them were heavily misshapen (Figure 3—figure supplement 2C,E).

17The liposomes were then analysed by transmission electron cryomicroscopy. Amazingly, when both TmFtsZ and TmFtsA were included on the inside of liposomes, clear constriction sites appeared and these occurred only when supported by filaments (Figure 3C–G). The liposomes were around 300 nm in diameter, similar to that of a small bacterial cell. The protein filaments were arranged into three distinct structures: arcs that were mostly single filaments, presumably formed by the co-polymerisation of TmFtsZ and TmFtsA, showing the characteristic curvature caused by the repeat mismatch; spirals with decreasing curvature that appeared in perfectly spherical areas of the liposomes; and rings of filaments, appearing as bands in projection, formed around liposome constriction sites, with varying diameters. It is very important to note that constrictions only appeared where there were rings of filaments. Filaments within the liposomes had very different curvatures, for example in Figure 3D, filaments seemed to go round the liposome at a very large diameter, compared to the ones in the constriction zone further to the left.

18When the constriction sites were imaged at higher magnification, it became possible to discern the TmFtsA and TmFtsZ filaments end-on (Figure 3H–J). The FtsA filaments were again sandwiched between the liposome membrane and the FtsZ filaments, just as in the images obtained with E. coli cells (Figure 2C,G). The architecture is easily explained with TmFtsA-binding to the inside of the liposome membrane via its amphipathic helix, presumably polymerising, and FtsZ polymerising on top of FtsA, binding to it via its C-terminal FtsA-binding peptide. Nucleotide presence had some influence on the appearance of these constricted liposomes as GTP addition produced the most bilobed liposomes. Constriction itself, however, was largely nucleotide hydrolysis independent since not adding any nucleotide produced constrictions as well, and they appeared even tighter (Figure 3E). Therefore, despite not being strictly required for constriction, we believe that the nucleotide only has an influence on the appearance thereof and this might be due to the fact that FtsZ forms much longer filaments with polymerisation-inducing GTP added and this would enable the filaments to span larger liposomes when the process starts. Given that TmFtsZ is a hyperthermophilic protein, significant hydrolysis of GTP to GDP is not expected.

19Next, we employed electron cryotomography to image the liposomes in three dimensions (Figure 4, Figure 4—figure supplement 1, Videos 4–10 ). Because the samples only contained lipid and two proteins, contrast was very high in the resulting tomograms (Video 4 and 5), making it possible to represent the volume data without segmentation, as single-threshold surfaces or as volume renderings (Video 6–10, Figure 4–source data 1 PyMOL session file). None of the figures or videos we present has been segmented, manually or automatically. Figure 4A, top shows a constricted liposome in stereo, highlighting the three distinct filament architectures in detail: arcs, spiral domes at the ‘poles’, and the filamentous ring, pulling and constricting the membrane. Note that the liposome was only deformed where the ring was located (and where it unfortunately touched the carbon grid at the top right). More examples in Figure 4A, bottom and Figure 4—figure supplement 1 show the same overall architecture, with the same mix of filament architectures (also Videos 6–10). The TmFtsAZ rings were between 30 and 90 nm in diameter, when we looked at several different liposomes, and the filaments were on average 7.8 nm apart (n = 16) laterally (Figure 4B,C) as compared to 6.8 nm seen in E. coli cells (within doublets). Because contrast was very high, the ring of filaments could be traced in most tomograms all around the inside of the liposome and this revealed that the filaments were not totally equidistant and often came into contact. This was also true for the in vivo situation in C. crescentus (Video 1) where the distance between the filaments (grey arrow) changes around the ring. The rings were always closed, continuous without gaps, with some possibly consisting of one double filament forming a helix (Figure 4—figure supplement 1 and Video 6) and others containing several shorter filaments in a helix-like arrangement (Figure 4A stereo view and Video 10). When the filaments were investigated end-on, they appeared to always be arranged close to 90° with respect to the liposome membrane tangent (Figure 4C), which was also observed for the filaments in cells (Figure 2C,G). All of these features are best demonstrated in Video 10, which provides an overview of the constriction and filament architectures and should be consulted to appreciate these findings properly.

20Map quality allowed us to fit the FtsZ crystal structure manually and roughly through a spline curve and arrive at a pseudo-atomic model for an FtsZ ring (Figure 5A), making it possible to judge distances and dimensions relative to the crystal structures. A more detailed view using sphere representation (Figure 5B) shows, again, that the filaments within a ring were not exactly equidistant (black arrows) but came into direct contact only at certain points. Fitting the FtsA crystal structure into the map as well revealed two closely associated filaments and showed that the outline fit of the tomographic density is extremely good, although exact orientations and locations of the subunits along the filament of the molecules can only be guessed in most places given the resolution limit. It should be noted, though, that peaks appear in many places indicating the centre positions of individual FtsZ molecules (Figure 5C, Figure 4–source data 1, a PyMOL session file).

21How do FtsZ and FtsA constrict liposomes? Is this likely to be related to the in vivo situation? Given that the filament architectures observed here in C. crescentus and E. coli and in constricting liposomes are so similar, we would suggest that the model we propose should be valid for both in vitro and in vivo, at least at some primordial level. FtsA forms (partial) filaments between the membrane and FtsZ filaments, and the filaments together encircle the constriction site while forming a single-layered small band of filaments. The entire structure is slightly helical, and shorter filaments overlap to form a continuous, closed ring.

22By imaging completely unmodified cells, utilising recent advances in cryo-EM and acquiring tomograms of cells parallel to the tilt axis, we concluded that the FtsZ ring in cells is most likely continuous, probably made of shorter overlapping filaments. Previous analysis of C. crescentus cells by cryo-ET also showed that the FtsZ ring consists of overlapping filaments inside the inner membrane, although not all cells showed continuous rings. Equally, results obtained with super resolution fluorescence microscopy techniques showed punctuated fluorescence, possibly indicating non-continuous rings. We think it is important to point out that fluorescence microscopy only images the labelled species and intensity fluctuations within the ring may have arisen from using non-functional GFP fusions and/or their over-expression. Or fluctuations coming from overlapping filaments may have been over-emphasised during image analysis because of very low signal-to-noise. Unfortunately, currently, cellular tomography data are too weak to be able to trace individual filaments and their ends with confidence, so we have no direct evidence from our in vivo data for the length of individual FtsZ filaments making up continuous rings.

23Given a continuous ring, at least in the in vitro situation, with no GTP turnover (or no added nucleotide) there is no filament shortening, meaning constriction requires the filaments to slide along each other as the ring decreases in diameter and the membrane deforms. That opposing forces from the filaments and the membrane surface are at work is most evident from the fact that the rings were always perfectly round, in contrast to the rest of the liposomes (Figure 4B, middle, Video 10).

24In this context, the filament spirals that form shallow ‘dome’ structures in some of the liposomes and on flat monolayers (Figures 3A and 4A, top and Figure 4—figure supplement 1) are most revealing. They appear to be the result of sliding and condensation and show decreasing filament curvature, but do not deform the lipid membrane. This can be explained because the constriction force, which acts in the plane of the spirals, will not exert any force on the membrane since it is tangential (Figure 5D, right). In contrast, if the filaments form a ring around the volume of the liposome (in the middle, not at the poles), the constriction force is perpendicular to the membrane and will lead to the membrane being pulled in (Figure 5D, left).

25Since in this model force generation is dependent on a closed ring, the system becomes self-regulating since constriction will only commence after a complete ring has formed. If the cell is too large or not enough FtsZ is available, constriction will not begin. It is important to note that only a closed, continuous ring is required, but it may consist of a number of shorter, overlapping filaments (as it did in the liposome reconstitutions, Figure 4).

26What drives constriction of the closed rings and filament sliding? We propose three possible mechanisms that may even act in concert: (a) maximising filament overlap via sliding, (b) increasing repeat mismatch, and (c) repeated filament shortening through nucleotide turnover.A. When the overlap between the filaments that are attracted to each other increases, more and more binding energy is produced. This has been proposed before to be theoretically sufficient for the constriction process (Figure 5E). The lateral spacing of 6.5–8 nm between filaments we report here is slightly larger than the thickness of FtsZ filaments and presumably also FtsA filaments. Previous in vitro work reported an interfilament distance of 5 nm using FtsZ-mts; however, this was after negative staining and dehydration and no constrictions were observed, possibly due to lateral interactions being too tight. In C. crescentus cells, the lateral spacing between filaments was found to be 9.3 nm previously and we report here distances of ∼7.8 (Figure 1A) in C. crescentus and ∼6.8 nm (Figure 1G) in E. coli. AFM using only FtsZ, but observing spiral condensation, provided an even larger distance of 12 nm. All of these measurements are averages with large variances. One may conclude that the filaments in the FtsZ ring interact transiently and direct contact is localised to only a few small regions within the ring at a time. This could facilitate the constriction process since the filaments have to be free to slide. It was suggested previously that instead of forming many intermolecular solid bonds, which would lead to avidity and a barrier to sliding, an attractive force over a longer distance would keep the filaments apart while interacting. This is more akin to the liquid state of matter, where many transient homotypic interactions, counteracted by thermal motion, lead to a fluid situation without absolute order but still keeping the molecules together.B. The second possible driver of constriction comes from the repeat length mismatch of FtsA and FtsZ. Although it is evident from our data that the curvature of individual filaments can not deform liposomes significantly from the inside, the decreasing diameter of the ring accompanied by increasing membrane curvature might enable more and more FtsA to be added, until an optimum curvature of the system has been achieved (Figure 4—figure supplement 2); this could provide additional energy and would also explain why FtsA exists at all and FtsZ is not directly attached to the membrane.C. Why does FtsZ hydrolyse GTP then? We speculate that when constriction starts at large diameters (1 µm in E. coli), longer GTP-induced FtsZ filaments are needed to reach reliably around the cell in order to produce overlap for the ring and 'force engagement' to start the process. However, increasing overlap, developing as the constriction progresses, might lead to a kinetic barrier of sliding through avidity and the filaments would then have to be shortened. Formally this provides a third possible driver of constriction, at least for large constriction distances.

27A. When the overlap between the filaments that are attracted to each other increases, more and more binding energy is produced. This has been proposed before to be theoretically sufficient for the constriction process (Figure 5E). The lateral spacing of 6.5–8 nm between filaments we report here is slightly larger than the thickness of FtsZ filaments and presumably also FtsA filaments. Previous in vitro work reported an interfilament distance of 5 nm using FtsZ-mts; however, this was after negative staining and dehydration and no constrictions were observed, possibly due to lateral interactions being too tight. In C. crescentus cells, the lateral spacing between filaments was found to be 9.3 nm previously and we report here distances of ∼7.8 (Figure 1A) in C. crescentus and ∼6.8 nm (Figure 1G) in E. coli. AFM using only FtsZ, but observing spiral condensation, provided an even larger distance of 12 nm. All of these measurements are averages with large variances. One may conclude that the filaments in the FtsZ ring interact transiently and direct contact is localised to only a few small regions within the ring at a time. This could facilitate the constriction process since the filaments have to be free to slide. It was suggested previously that instead of forming many intermolecular solid bonds, which would lead to avidity and a barrier to sliding, an attractive force over a longer distance would keep the filaments apart while interacting. This is more akin to the liquid state of matter, where many transient homotypic interactions, counteracted by thermal motion, lead to a fluid situation without absolute order but still keeping the molecules together.

28B. The second possible driver of constriction comes from the repeat length mismatch of FtsA and FtsZ. Although it is evident from our data that the curvature of individual filaments can not deform liposomes significantly from the inside, the decreasing diameter of the ring accompanied by increasing membrane curvature might enable more and more FtsA to be added, until an optimum curvature of the system has been achieved (Figure 4—figure supplement 2); this could provide additional energy and would also explain why FtsA exists at all and FtsZ is not directly attached to the membrane.

29C. Why does FtsZ hydrolyse GTP then? We speculate that when constriction starts at large diameters (1 µm in E. coli), longer GTP-induced FtsZ filaments are needed to reach reliably around the cell in order to produce overlap for the ring and 'force engagement' to start the process. However, increasing overlap, developing as the constriction progresses, might lead to a kinetic barrier of sliding through avidity and the filaments would then have to be shortened. Formally this provides a third possible driver of constriction, at least for large constriction distances.

30Continual depolymerisation and re-polymerisation through nucleotide turnover by FtsZ and/or FtsA, as shown in vivo by FRAP, might also ensure that the ring never reaches a highly condensed state and FtsZ monomer-sequestering inhibitors such as SulA remain able to stop the process at any time. However, it has been reported that the GTP hydrolysis-deficient FtsZ(D212G) mutant generates prominent constrictions of tubular liposomes in vitro and functions in cell division in E. coli. Furthermore, the use of non-hydrolysable GTP analogues did not impair the formation of condensed FtsZ structures on mica. Taken together with our result that constriction of liposomes may in principle be independent of GTP hydrolysis, these data question the alternative idea of force generation by filament bending upon GTP hydrolysis as has been suggested previously. We suggest that nucleotide binding/hydrolysis is required solely for filament growth/shrinkage as these are essential to maintain the dynamic state of the FtsZ ring in cells.

31Taken together, we envisage that in vitro liposome constrictions and in vivo cell division quite possibly utilise a different set of energetic drivers (a–c); for example, GTP hydrolysis was not required in vitro but clearly plays a role in vivo. And of course, it is likely that the cell wall synthesis in the periplasm, guided by the Z-ring through the divisome, provides additional force in cells. So far, wall-less bacterial L-forms have not shed light on this interdependence of cell wall synthesis and the FtsZ ring since artificial L-forms were found to divide by blebbing, most likely a mechanism that is not actively supported by cellular machinery. FtsZ-based cell division is not functioning in these L-forms and we suggest this may be because L-forms are too large for Z-rings to close, given the amounts of FtsAZ present.

32The described FtsZ filament arrangement might also provide a solution to the abscission problem: how do the membranes fuse at the end of division when the protein filaments are in between? Figure 5D, middle shows how an intermediate between the rings and the domes (as is present in the liposome constriction shown in Figure 4A, bottom right and Video 9) may explain abscission, since the protein ring would normally be in the way of membrane fusion/fission at the end. The change from a flat band of filaments towards the helical spirals enables inward force to be developed on each side of the constriction, with a helical spiral on each side. The spirals observed in the dome-like structures might even be remnants of such liposome abscission events, although we have no evidence for this (Figure 5D, right). It remains to be seen if FtsZ is involved in final abscission since it has recently been reported that FtsZ might leave the septum earlier.

33It is important to mention that membrane constriction with ESCRT-III and dynamin filaments has also been suggested to involve sliding helical filaments and similar arrangements to those depicted in Figure 5D, right were predicted for the ESCRT-III system.

34Finally, our reconstitution of cell division can easily be adapted to include other cell division proteins, such as the division site selection mechanism MinCDE, nucleoid occlusion, FtsZ cross-linkers such as ZapA, and many more middle and outer divisome components.

35Plasmids used in this work are listed in Supplementary File 1A. E. coli DH5α was used for cloning. Caulobacter crescentus NA1000/CB15N and E. coli B/r H266 were used for cellular electron cryotomography.

36Caulobacter crescentus was grown overnight in PYE medium at 30°C. The overnight culture was used to inoculate 50 ml of M2G medium. The culture was grown at 30°C until the OD reached 0.5. 11 µl of this culture was mixed with 1 µl of protein-A conjugated to 10 nm gold beads (CMC, Leiden) and applied to freshly glow-discharged 300 mesh Cu/Rh Quantifoil (3.5/1) grids. The grids were plunge-frozen into liquid ethane using a FEI Vitrobot (Mark IV) and stored in liquid nitrogen.

37E. coli cells (some containing relevant plasmids, for FtsZ mutant co-expression with endogenous wild-type FtsZ, Supplementary file 1, Tables A and B) were grown at 30°C in M9 minimal media supplemented with 0.4% glycerol until log-phase. Cells were then diluted into fresh M9 media with 0.02% arabinose (where needed, final concentration) and grown for 1–2 hr for FtsZ mutant protein expression.

38Thermotoga maritima FtsZ (TmFtsZ) and FtsA (TmFtsA) proteins were purified and 2D monolayers were prepared as described previously, taking extra care and verifying by ESMS that the C-terminal tails of both proteins were intact after purification as they are prone to proteolytic cleavage. This was not obvious from gels, sometimes.

3920 μl of E. coli total lipid extract (Avanti Polar Lipids, Alabaster, AL) chloroform solution at 10 mg/ml was dried in a glass vial (Wheaton, Millville, NJ) under a stream of nitrogen gas and left overnight under vacuum to remove traces of the solvent. The resulting thin lipid film was hydrated with 200 μl of TEN1007.5 buffer (50 mM Tris/HCl, 100 mM NaCl, 1 mM EDTA, 1 mM NaN3, pH 7.5), containing either TmFtsA at 20 µM or TmFtsZ at 60 µM or both proteins. After 10 min of incubation at room temperature, the solutions were sonicated for 1 min in a water bath sonicator and then 2.5 µl of sample was plunge-frozen onto Quantifoil R2/2 holey carbon grids (Quantifoil, Germany) using an FEI Vitrobot (FEI Hillsboro, OR). Samples were stored in liquid nitrogen.

4050 μl of E. coli total lipid extract chloroform solution at 10 mg/ml was dried in a glass vial under a stream of nitrogen gas and left overnight under vacuum to remove traces of the solvent. The resulting thin lipid film was hydrated with 50 μl of TEN1007.5 plus 20 mM CHAPS (Anatrace, Maumee, Ohio) and shaken vigorously at 800 rpm using a benchtop Eppendorf shaker for 2 hr. The lipid–detergent solution was sonicated for 1 min in a water bath sonicator. Subsequently, 50 μl of TmFtsZ (30 µM) and TmFtsA (10 µM) solutions supplemented with 0.5 mM MgGTP or MgGMPCPP (Jena Bioscience, Germany) or no nucleotide was added and left for 30 min at room temperature. Next, the mixture was gradually diluted within 10 to 20 min to 600 μl with TEN1007.5 or TEN1007.5 plus nucleotides (both without detergent) to trigger spontaneous liposome formation. 2.5 µl of the solution was mixed with 0.2 µl 5 nm IgG immunogold conjugate (TAAB, UK) and plunge-frozen onto Quantifoil R2/2 holey carbon grid using an FEI Vitrobot.

412D electron cryomicroscopy (cryo-TEM) images were taken on an FEI TECNAI Spirit TEM operating at 120 kV with a 2k × 2k CCD camera at a magnification of 42 k, corresponding to a pixel size of 0.25 nm. For electron cryotomography, samples were imaged using an FEI Polara or FEI Titan Krios TEM operating at 300 kV, equipped with a Gatan imaging filter set at zero-loss peak with a slit-width of 20 eV. A Gatan Ultrascan 4000 CCD camera binned to 2k × 2k or a 4k × 4k K2 Summit direct electron detector was used for data acquisition with SerialEM software. Cells or in vitro reconstituted systems were imaged at a magnification of 41 k, corresponding to a pixel size of 5.8 Å (with US4000), or at a magnification of 26 k, corresponding to a pixel size of 4.5 Å (for K2) at the specimen level. Specimens were tilted from approximately −60° to +60° (±65° for C. crescentus cells) with a 1° increment. The defocus was set between 8 and 10 µm, and the total dose for each tilt series was around 120 e/Å2 for in vitro reconstitution samples and 150–200 e/Å2 for cells.

42Tomographic reconstructions from tilt series were calculated using RAPTOR and the IMOD tomography reconstruction package, followed by SIRT reconstruction with the PRIISM software or the TOMO3D package. Measurements of distances between structures were carried out within IMOD. Videos showing liposomes were prepared with PyMOL.

43E. coli B/r H266 cells with plasmid pMZ124 were grown in LB medium at 30°C. At an OD600 of 0.2, FtsZ(D212A) and FtsA expression was induced by adding 0.02% arabinose. After 2 hr, cell membranes were stained with FM4-64 membrane dye and cells were mounted on an agarose pad and visualised using a Nikon N-SIM microscope in the 2D-SIM mode.

44FtsZ expression level in cells used for electron cryotomography experiments were examined with Western blots using rabbit anti-FtsZ primary antibodies (Agrisera, Sweden) and donkey anti-rabbit IgG conjugated with horseradish peroxidase (GE Healthcare) and detected with ECL blotting reagent.

45The Caulobacter crescentus tomogram shown in Figure 1A has been deposited in the EM databank with accession number EMD-2814. The edited tomogram of TmFtsAZ constricting a liposome, as shown in Figures 4A and 5A and Video 10, has been deposited with accession number EMD-2815.