A polymerisation-associated conformational switch in FtsZ that enables treadmilling

Bacterial cell division in many organisms involves a constricting cytokinetic ring that is orchestrated by the tubulin-like protein FtsZ. FtsZ forms dynamic filaments close to the membrane at the site of division that have recently been shown to treadmill around the division ring, guiding septal wall synthesis. Here, using X-ray crystallography of Staphylococcus aureus SaFtsZ we reveal how an FtsZ can adopt two functionally distinct conformations, open and closed. The open form is found in SaFtsZ filaments formed in crystals and also in soluble filaments of E. coli FtsZ as deduced by cryoEM. The closed form is found within several crystal forms of two non-polymerising SaFtsZ mutants and corresponds to many previous FtsZ structures from other organisms. We argue that FtsZ undergoes a polymerisation-associated conformational switch. We show that such a switch provides explanations for both how treadmilling may occur within a single-stranded filament, and why filament assembly is cooperative.


Introduction
FtsZ is an ancient, filament forming, tubulin-like GTPase protein found in the vast majority of bacteria and archaea, where it acts as a central component of the cell division machinery (1)(2)(3). FtsZ is localised to the plasma membrane at future division sites resulting in the emergence of a ring structure around the centre of the cell, the Z-ring. FtsZ is anchored to the plasma membrane by other proteins, most often FtsA but also ZipA and/or SepF (4)(5)(6).
FtsA is a divergent actin homologue that forms copolymers with FtsZ and contains an amphipathic helix that facilitates membrane attachment (7).
After the localisation of FtsZ, a large number of other proteins are recruited to the division site. These proteins carry out remodelling and synthesis of cell wall during the division process. Together these proteins have been termed the divisome, although it is currently not clear which components, if any, form stable multi-subunit complexes during the process.
The precise molecular architecture of the Z-ring remains unclear, although it is probably composed of dynamic overlapping filaments along the circumference of the ring, at least during the later stages of the division process in rod-shaped model organisms such as Escherichia coli (8). It was already clear from early fluorescence microscopy studies that during the cell division process the Z-ring contracts with the invaginating septum (9). In vitro reconstitution experiments of FtsZ and FtsA with membranes showed that these two components alone deform membranes (10). Together with homology to force-generating eukaryotic tubulins this prompted the suggestion that FtsZ has a role in generating forces required for invagination. In contrast, observations of constrictions and divisions of cells with helical Z-rings, incomplete Z-rings, and divisomes with modified FtsZ properties, support the opposing idea that FtsZ does not provide an indispensable driving force for constriction (11)(12)(13)(14). The alternative candidate for force generation is cell wall remodelling. A third option is that cell wall remodelling and Z-ring dynamics are interlinked processes that work together to generate the forces needed for division to occur robustly and efficiently every time.
Recently, in vitro treadmilling of FtsZ filaments has been reported on supported bilayers (15,16) and also in vivo where FtsZ filaments were found to treadmill with components of the divisome around the division site (17,18). These findings have resurrected an old model of bacterial cell division: the template model, in which the closing septum constricts 4 by new cell wall material being deposited in concentric rings on the inside of old material by moving synthesis machinery, which in turn is guided or organised into a ring by dynamic FtsZ filaments. This idea fits into the third category of ideas about the role of FtsZ.
Treadmilling is a property of certain cytomotive filaments whose filament-forming subunit interfaces are made dynamic in time through nucleotide hydrolysis that is, in turn, triggered by the polymerisation reaction. Treadmilling requires a difference in the rate of net polymerisation and de-polymerisation at the so-called plus and minus ends of the filaments. FtsZ so far was not considered a good candidate for treadmilling behaviour largely because it is currently not known if any well ordered structures are formed in cells beyond single-stranded protofilaments. Differing rates of subunit addition at the two ends of simple single-stranded filaments are difficult to envisage (19).
Surprisingly, knowledge of FtsZ filament architecture is limited. Only one FtsZ crystal form, from Staphylococcus aureus (SaFtsZ, PDB IDs 3VOA, 3VO8), has revealed a straight protofilament of FtsZ, as might be expected from electron micrographs of many different FtsZ filaments (20). The arrangement of this crystalline filament is very similar to  (20,22). PC190273 reduces polymerisation cooperativity in vitro (23).

5
Isolated open form SaFtsZ monomers relax into the closed conformation during molecular dynamics simulations (24). Fluorescent analogues of PC190723 have recently been used to monitor opening and closing of the inter-domain cleft in solution as a function of FtsZ polymerisation state (25). Together, these results hint that the closed form of FtsZ seen in many crystals is the predominant conformation of monomeric FtsZ and vice versa that filamentous FtsZ in solution is in the open conformation seen in SaFtsZ filament crystals.
Currently what is lacking is robust structural evidence that this is the case.
FtsZ shares two properties with actin and tubulin that until now have been hard to explain.
Firstly, FtsZ exhibits cooperative assembly, with a critical concentration and a lag phase in assembly. This is not possible for a single-stranded, isodesmic filament with rigid subunits, and an assembly switch has long been hypothesised to explain this cooperativity (26).
Secondly, filament treadmilling is presumed to require multi-strandedness, which FtsZ may not have.
Here we demonstrate that FtsZs do indeed undergo a conformational switch, that this switch is associated with polymerisation and not nucleotide hydrolysis state and that switching provides a possible mechanism for both cooperative assembly and also for treadmilling, which has been proposed to be a key dynamic filament behaviour used to organise cell wall remodelling.

Results and Discussion
SaFtsZ-T66W and -F138A are polymerisation and GTPase compromised.
Two SaFtsZ mutations, F138A and T66W were designed to inhibit SaFtsZ filament formation, based on equivalent mutations inhibiting assembly of Methanocaldococcus jannaschii FtsZ (M164A (27) and T92W (28) respectively, polymerisation inhibition of T92W unpublished data). Both mutation sites are located on the 'top' surface of FtsZ, on the N-terminal, GTP-binding domain and are part of the longitudinal protofilament interface seen in crystals. Full-length, untagged, SaFtsZ wildtype, F138A, and T66W proteins were purified and characterised biochemically ( Figure 1). Filament formation in both SaFtsZ mutated proteins was compromised since no filament formation was detected by sedimentation ( Figure 1A) or negative stain electron microscopy ( Figure 1C) for either T66W or F138A in the presence of GTP or guanosine-5'-[(α,β)-methyleno]triphosphate (GMPCPP), a slowly-hydrolysable analogue of GTP. FtsZ GTPase activity is largely dependent on polymerisation as one subunit provides catalytic residues to the active site of the next subunit through residues in loop T7. Both mutants have weak GTPase activity ( Figure 1B), indicating that monomers may at least associate to form transient but functional active sites. In support of this, on addition of PC190723, the mutant proteins did form filaments detectable by sedimentation and electron microscopy in the presence of GTP and GMPCPP. We conclude that SaFtsZ T66W and F138A are polymerisation and GTPase compromised but retain some residual activities.

SaFtsZ adopts either a closed or an open conformation.
We solved five crystal structures of the globular domains of SaFtsZ T66W and F138A (Table S1, Figure 2). SaFtsZ constructs truncated to residues 12-316 were used to remove the N and C-terminal tails of FtsZ previously found to inhibit crystallisation.   In order to have two stable globular conformations FtsZ must have structural features that rearrange during switching. These features are best visualised using morph (32) interpolations between the structures shown in Figure 2B, as in Video S1. The large   Figure 5A. Like in tubulin, the nucleotide forms a large part of the interface between subunits, and it is thought that nucleotide hydrolysis is used to modulate interface affinity. These crystalline filaments have a 44 Å repeat, which corresponds well to repeat intervals seen in negatively stained FtsZ filaments from a number of species. As a result, it has been hypothesised that the 1FOf-like crystal filaments resemble soluble FtsZ filaments.
As discussed, previous work generated SaFtsZ structures in the closed form by extensive alteration of the T7 loop (PDB IDs 3WGL, 3WGK). These structures contain SaFtsZs that are clearly in the closed conformation ( Figure 2D), but are arranged in straight filaments in the crystal. However, these filaments are not the same as the open form filaments, with a much 11 smaller interface buried surface area of ~700 Å 2 as compared to ~1200 Å 2 for 1FOf and PDB ID 3VOA (calculated with PDBePISA server (35)), and a repeat of 45 Å. A dimer from a 3WGL pseudofilament is shown in Figure 5B. The longitudinal contact is made between residues from the bottom subunit at the N-terminus of H5 and the preceding loop (including residue F138), and the loop between H6 and H7. From the top subunit the T7 loop (replaced in these structures), one face of S9, and the loop at the N-terminus of H10 are involved. Interaction does not involve any of the residues on the other side of the interface, towards the phosphates of the nucleotide.
In contrast, we present four crystal forms where, for the first time, SaFtsZ is not arranged in straight, infinitely long filaments. In three of these, 2TCm, 3FCm and 5FCm we find in each case that one of the molecules in the ASU forms what looks superficially like a filament interface via its top face, and the other molecule in the ASU equivalently contributes a bottom face to another pseudo-interface. The crystals are therefore composed of pairs of poorly-interacting FtsZs (shown in Figure 5B) which pack via further crystal contacts that do not resemble interfaces. The pseudo-interfaces have subunitsubunit buried surface area (BSA) of 670-800 Å 2 , and look similar to the interfaces seen in the closed T7 mutant structures, only including residues from one side of the top face.
We know that our mutations inhibit filament formation (Figure 1), and we obtained crystals where these mutant SaFtsZs adopt the closed conformation, and fail to form bona fide interfaces. Hence we propose that these closed forms correspond to the conformation of monomeric SaFtsZ in solution. The pseudo-interface seen is perhaps best seen in this context as a consequence of crystallisation, which uses conditions that enhance proteinprotein interactions, and not a cause. If the protein will crystallise it is extremely likely that one of the major crystal contacts will imitate the longitudinal interface, as the interface regions are most likely sticky. However, we cannot rule out the possibility that this minimal interface (in silico repetition of which generates a highly curved filament) is a functionally relevant and stable way for FtsZs to interact in solution.
The fact that the phosphate end of the interface is not formed in any of the F138A closed form crystals supports the idea that it is the closed conformation that is not compatible with formation of bona fide interfaces, not the mutation per se -because the F138A mutation is within the pseudo-interface. The fact that we recovered the 1FOf crystal form 12 for the F138A mutant suggests that the conformational switch in this mutant is finely balanced.
The fourth closed form structure, 4FCs, as already mentioned, is arranged very differently within the crystal. The two pseudo-whole domain-swapped FtsZs formed by each pair of polypeptides both clearly adopt the closed conformation (Cα RMSD for comparable atoms in pseudo-monomer vs 2TCm 1.0 Å). The domain-swapped FtsZs do not make any crystal contacts that resemble filament interfaces. That a domain swap can happen, and that a domain-swapped FtsZ adopts a closed conformation, suggests that the two domains have a significant degree of independence and, more importantly here, that when SaFtsZ conformation is not modulated by polymerisation or pseudo-interfaces, the molecule adopts the closed conformation, as suggested previously by molecular dynamics (24).
We are able to significantly bolster the case that polymerisation into a straight filament is the driving force for the conformational switch we observe in crystal structures by presenting a medium (~8 Å) resolution electron cryomicroscopy (cryoEM) structure of wildtype, full-length E. coli FtsZ straight filaments, Fig 4A, contains open subunits, we propose that a polymerisation-associated conformational switch is a general property of FtsZs. One of the consequences of such a switch, namely permitting cooperative assembly of a single-stranded filament, have been discussed 13 previously (23,26,27,36), however, such a switch confers additional surprising properties.
The FtsZ conformational switch between monomer and filament provides filament-end asymmetry necessary for treadmilling.
In preparing this manuscript it became clear to us that theoretical considerations of treadmilling can be fraught with intellectual traps. Having run the gauntlet of these potential pitfalls we present simplified, yet robust, schema to explain how a polymerisation-associated conformational switch provides the end-asymmetry necessary for treadmilling within a single stranded filament. We focus on the specific case where nucleotide forms part of the filament interface (i.e. in a tubulin-like fashion). In these, solvent exposed NDPs (nucleoside diphosphate) are quickly exchanged with NTPs (nucleoside triphosphate), NTP hydrolysis is not immediate, and NTP interfaces are stronger than NDP interfaces; although many of the conclusions are the same for filaments where nucleotide is buried inside subunits and has an allosteric effect on interface strength (i.e. actin-like).
Treadmilling of cytoskeletal filaments is a useful dynamic property. Treadmilling filaments can be used to push or pull molecules in the cell without motor proteins as long as endtracking mechanisms or co-factors exist, and these behaviours can be made switchable with high flux through the filament (e.g. in eukaryotic anaphase microtubules). Recently it has been shown that FtsZ can treadmill in vitro with FtsA (16) and also alone (15), and that FtsA/Z treadmilling in cells guides septal cell wall remodelling (17,18). FtsZ is widely considered to function as a single filament. However, as has been noted previously (19), a single-stranded filament with the above properties and rigid subunits, without conformational changes, cannot do robust treadmilling.
Such a hypothetical filament with rigid subunits is shown in Figure 6A. Note that the location (top/bottom) of nucleotide binding to the monomer is not important. This filament treadmills if a nucleotide gradient along the filament exists, and the kinetic plus end (net growth) will be at the end with more NTP. On-rates cannot differ at the two ends because they are the same reaction, but off-rate at the minus (GDP) end will be greater than at the plus end, so a situation of net growth at one end and net shrinkage at the other can be produced by changing monomer concentration (addition reaction is 1 st order with respect to monomer, loss is 0 order). This does not represent robust treadmilling however, 14 as breakdown of terminal GDP interfaces is equivalent to breakdown of a GDP interface anywhere in the filament -and these processes will occur at the same rate as they are all 0 order. As noted previously (37) filament breakage and annealing could be an important facet of dynamics, but the filament in Figure 6A has a more fundamental limitation on it's biological usefulness: the direction of treadmilling is determined entirely by the history of the filament (the direction of the initial NTP-NDP gradient), so there is no coupling of kinetic and structural polarity -and the same filament could treadmill in either structuredefined direction.
Coupling of kinetic and structural polarity requires subunit addition and or loss to proceed via different stereochemical pathways at each structure-defined end of the filament. This is not the case in Figure 6A, the difference in off-rates is defined by the nucleotide gradient and not structural polarity, and we have already seen that there can be no difference in onrates. Filament systems can generate different stereochemistry for subunit addition at either end by being multi-stranded and having staggered subunits that undergo a conformational change, such as actin (38), or by using a longitudinal hooking mechanism, such as TubZ (39). Figure 6B shows our model for how a single filament very similar to the case in Figure 6A can couple its structural polarity to a defined kinetic polarity and thus usefully, and robustly, treadmill. The crucial difference between Figure 6A and B is the existence of a polymerisation-associated conformational switch, i.e. subunits are no longer rigid, but can exist in one of two conformations -one form associated with the polymer, the other adopted in the free monomer. The free energy cost of the conformational switch from closed to open is paid for by binding to a filament end, and in the other direction through nucleotide hydrolysis and exchange that makes the longitudinal NDP intersubunit interface unfavourable. Although formation of a NTP interface at either structurally-defined end has the same net energy change, the reaction pathways are stereochemically different, and will occur at different rates because two different pairs of molecular surfaces are involved in each case initially. This difference is illustrated in a non-rigorous fashion in the context of our structures in Figure 7.
Importantly, the scheme in Figure 6B also allows breakage at NDP interfaces within the filament to be different to loss of NDP subunits from each end: essentially NDP interfaces in the filament can be stronger because the energetic cost of breaking them, when losing subunits from ends only, is paid for by the favourable switch to the monomer conformation which the end subunits of two halves of the broken filament cannot adopt because they remain in filaments. Especially important to note here that the scheme in Figure 6B can also be drawn with nucleotide on the other side of the monomer, i.e. we are not making a prediction about which end of a single-stranded FtsZ filament is the kinetic plus end -an issue that will need further investigation (37,40,41).

Conclusion
Here

Negative stain electron microscopy
Full-length wild-type and mutant SaFtsZs were visualised by negative-stain electron microscopy. About 20 µL of sample, prepared as for sedimentation analysis of assembly (except protein was at 20 μM), were applied to formvar and carbon-coated copper grids, incubated for one minute, and then stained with 2% w/v uranyl acetate in water. Images were taken at several magnifications using a JEOL 1200 EX-II microscope operated at 100 kV equipped with a Gatan CCD camera.

Crystallisation
Crystallisation conditions were found using our in house high-throughput crystallisation platform, by mixing 100 nL truncated SaFtsZ T66W or F138A solution at 5 or 10 mg/mL, with GTP at 10 mM, with 100 nL of 1920 different crystallisation reagents in MRC vapour 19 diffusion sitting drop crystallisation plates. Conditions yielding crystals were optimised, and crystals from either the initial screens or subsequent optimization were selected for data collection. Conditions giving the crystals for which structures are presented are in Table S1.

Crystallographic data collection and structure determination
Diffraction images were collected from single frozen crystals at beamlines at either DLS (Diamond Light Source, Harwell, UK) or ESRF (European Synchrotron Radiation Facility, Grenoble, France) as indicated in Table S1, at 100 K. Diffraction images were processed with XDS, POINTLESS and SCALA software. Initial phases were determined by molecular replacement using PHASER with search models as indicated in Table S1. Models were rebuilt manually using MAIN and refined using REFMAC and PHENIX.REFINE.
Ramachandran plots and MOLPROBITY statistics were used to validate the structures.

Supplementary Material
Table S1 -Crystallographic and cryoEM data Video S1 -The SaFtsZ conformational switch This video shows a morph interpolation between SaFtsZ structures 1FOf and 3FCm, from two angles, with and without side chains.

Figure 3. All FtsZ structures (except apo SaFtsZs) can be grouped into two conformations, open and closed. A)
All previous FtsZ structures were obtained from the PDB as listed in (B). Chain A from each downloaded structure, and the five structures determined here, was extracted and aligned to the N-terminal domain (residues 12 -176) of 3VOA using the PyMOL align command (matches residues via sequence then minimises RMSD (root mean square distance) with 5 cycles of outlier rejection; except for PDB ID 1W5F and our structure 4FCs which are both domain-swapped, in these cases a pseudomonomer was generated for each, also S. aureus apo-structures (PDB IDs 3VO9, 3VPA), which have a very different conformation (20), were excluded. N-and C-terminal extensions were removed, and the aligned structures are shown in ribbon representation from the same view as in Figure 2B.     A) and B) FtsZ pairs were extracted from crystal lattices as described in the text. Structures are shown in cartoon representation, each chain is rainbow coloured blue to red, N terminus to C terminus. Nucleotide atoms are coloured by element. In each case the view is from the same orientation after the lower molecule is aligned to the NTD of the lower subunit from PDB ID 3VOA. 4FCs is shown with one chain coloured and the other in white to highlight the domain swap. EcFtsZ filament cryoEM density is shown at a threshold of 7.5 σ, the same as in D. Open structures (panel A) can be arranged to have 44 Å repeats using the favourable tubulin-like interfaces found in crystals 3VOA and 1FOf. Closed structures (B) have smaller, incomplete, inter-subunit interfaces within crystals and cannot be sensibly arranged to produce straight filaments with a 44 Å repeat. C) Typical micrograph of frozenhydrated EcFtsZ.GMPCPP filaments. Curved, straight, single and double/bundled filaments are seen. Inset shows magnified view. 44 Å repeat is visible by eye. D) Representative EcFtsz filament 2D classes produced by RELION E, F) FtsZ structures as indicated were fitted into the EcFtsZ cryoEM density using the CHIMERA volume viewer fitting tool. (E) A 1FOf 5-mer fits very well into the density, as does a 1FOf monomer -with both fits extremely similar. RMSD is for middle subunit in rigidly fitted 5-mer and monomer fitted into middle subunit density (F) Closed structures do not fit well into the electron density, and certainly not so that a repeating filament can be constructed. Some regions of poor fit are indicated with black arrowheads. Black arrows indicate rates roughly in proportion to their width, coloured arrows in (A) indicate rates that are equivalent. See main text for discussion of limitations and assumptions of these simplified models, particularly regarding implied orientation of molecules. A) An idealised rigid (lacking a conformational switch), tubulin-like, filament forming protein, for which addition/loss of a given NXP is isodesmic. This filament canot do robust treadmilling, as breakage is the same as minus end subunit loss, and it cannot couple structural and kinetic polarity B) A singlestranded version of (A) with a polymerisation-associated conformational switch, able to treadmill robustly and with coupled kinetic and structural polarities. The conformational switch allows filament breakage and subunit loss from ends to be different, and for the stereochemistry of subunit addition at either end to be different -meaning that addition will take place at different rates in a manner defined by structural polarity. will not represent the transition state of subunit addition at either end of a filament (nor even any position on the reaction pathway), but they illustrate the fact that the conformational switch will necessarily lead to stereochemically different reaction pathways at each end that allow the two ends to have different rates of subunit addition, linking structural and kinetic polarity.  a Values in parentheses refer to the highest recorded resolution shell. b 5% of reflections were randomly selected before refinement. c Percentage of residues in the Ramachandran plot (PROCHECK 'most favoured' and 'additionally allowed' added together).