Structure and function of a spectrin-like regulator of bacterial cytokinesis

Bacterial cell division is facilitated by a molecular machine—the divisome—that assembles at mid-cell in dividing cells. The formation of the cytokinetic Z-ring by the tubulin homologue FtsZ is regulated by several factors, including the divisome component EzrA. Here we describe the structure of the 60-kDa cytoplasmic domain of EzrA, which comprises five linear repeats of an unusual triple helical bundle. The EzrA structure is bent into a semicircle, providing the protein with the potential to interact at both N- and C-termini with adjacent membrane-bound divisome components. We also identify at least two binding sites for FtsZ on EzrA and map regions of EzrA that are responsible for regulating FtsZ assembly. The individual repeats, and their linear organization, are homologous to the spectrin proteins that connect actin filaments to the membrane in eukaryotes, and we thus propose that EzrA is the founding member of the bacterial spectrin family.

B acterial cell division is directed by cytoskeletal proteins. The bacterial homologue of eukaryotic tubulin, FtsZ 1 , polymerises in the presence of GTP 2 and is attached to the membrane at the mid-cell position to form the Z-ring 3,4 . The hydrolysis of GTP to GDP results in the dissociation of FtsZ polymers into monomers that disengage from the Z-ring 5 . Hence, the Z-ring is highly dynamic and constantly exchanges FtsZ molecules with the pool of its freely diffusible monomers 6 . The Z-ring is stabilized and anchored to the cytoplasmic membrane by interacting with FtsA, a bacterial homologue of actin, which is itself capable of polymerizing at the membrane 7 . FtsZ and FtsA recruit other cell division proteins, including cell wall synthesis enzymes and many proteins of uncharacterized function, to the mid-cell to form the mature divisome 8 . This large macromolecular assembly synthesizes the cell wall that will form the new poles of the cell after division 9 .
The polymerization of FtsZ into the Z-ring is a necessary and tightly regulated step in divisome assembly. The short C-terminal tail of FtsZ is essential for self-assembly and for interactions with its regulators 3 . Z-ring assembly is regulated by multiple, widely conserved cell division proteins, including FtsA, ZapA, the protease ClpX, SepF and EzrA. FtsA both recruits FtsZ filaments to the cell membrane and counteracts the clustering of FtsZ filaments on it 10 . ZapA and SepF promote the bundling of FtsZ filaments into higher-order structures in vitro [11][12][13] . SepF additionally stabilizes the Z-ring at mid-cell in vivo 14 by binding to both FtsZ and the membrane 13 .
By contrast, the integral membrane protein EzrA was identified as an inhibitor of FtsZ polymerization and its absence results in the inappropriate formation of multiple FtsZ rings, not only at mid-cell but also at the cell poles 15,16 . However, EzrA is also one of the first proteins to be recruited to the mid-cell by FtsZ 8 , suggesting it has an additional regulatory or scaffolding role in the formation of the divisome. As an inhibitor of FtsZ polymerization, EzrA is assumed to be required to maintain the dynamics of the Z-ring as it constricts during cell division 16 . Other evidence supports a positive role for EzrA in the assembly of the divisome. In Bacillus subtilis, EzrA recruits the major bi-functional, glycosyltransferase-transpeptidase peptidoglycan synthase, penicillin-binding protein 1 (PBP1, aka PonA) 17 to the septum. Bacteria require the synthesis of new peptidoglycan to grow, and peptidoglycan must be deposited as cells divide to ensure a continuum of cell wall around both halves of the dividing cell. Consequently, PBPs that synthesize new cell wall material must localize to areas of cell growth or division. Deletion of ezrA changes the cellular localization patterns of both FtsZ and PBP1, causing marked growth defects and alterations in cell morphology, division and elongation 17,18 . Generally, ezrA knockouts display a delay in cell division, resulting in cells about twice as long as wild type in B. subtilis 13,15,19 and larger, spherical cells in Staphylococcus aureus 18 . The ezrA knockout has been described as lethal to S. aureus 20 , Streptococcus pneumoniae 21,22 and Listeria monocytogenes 23 , and thus constitutes a novel target for antibacterial agents. In B. subtilis, the concurrent knockout of ezrA with other divisome components including zapA 11 , sepF 14,17 or gpsB 17 causes severe growth defects. Therefore, EzrA plays key roles both as a regulator of cell division and in the maintenance of cell size and shape.
EzrA comprises a single, N-terminal trans-membrane (TM) spanning helix followed by an B60-kDa cytoplasmic domain 15 , which sequence analysis suggests forms an a-helical coiled-coil. To understand further the role of EzrA during cell division, we report the crystal structure of the intact, cytoplasmic domain of B. subtilis (Bs) EzrA at 3.95 Å resolution. Its three-dimensional structure comprises five, B100 amino-acid long repeats, each of which forms an anti-parallel triple helical bundle; these are arranged linearly to form a complete semi-circle with a diameter of 120 Å. The individual repeats and their linear head-to-tail arrangement are reminiscent of the spectrin family of eukaryotic cytoskeletal proteins that associate with actin filaments and membrane-bound proteins. Two copies of the spectrin repeats are also observed in a 2.6-Å resolution structure of a 20-kDa cytoplasmic fragment of EzrA from S. aureus (Sa), which has enabled direct molecular comparisons to be made between eukaryotic spectrin repeats and those found in EzrA. FtsZbinding sites on the BsEzrA arch are mapped and a binding site that inhibits FtsZ oligomerization is identified. We propose a role for EzrA that is analogous to that of spectrins in eukaryotes, in linking the dynamics of the bacterial cytoskeleton to the membrane-anchored components of the divisome, the proper localization of which is essential for the bacterial cell cycle.

Results
EzrA has a semi-circular structure. The crystal structure of B. subtilis EzrA lacking the N-terminal 21 residue TM helix (that is, BsEzrA 22-562 ) was solved by selenomethionine single wavelength anomalous dispersion (SAD) to a resolution of 3.95 Å. Despite extensive crystallization screening, the reliable diffraction limit could not be extended. However, through the use of highly redundant data, the experimentally phased electron density maps of BsEzrA 22-562 ( Supplementary Fig. 1a) enabled the building of 96% of the 540 amino acids in the structure. The only regions that could not be built were six residues at the N-terminus and a loop between Ser177 and Asn190, which the sequence predicted to be disordered. Although only a few side chain features are visible at this resolution, the final refined model, which includes sidechains, has geometric and crystallographic refinement parameters that compare very favourably with other structures of similar resolution ( Table 1). The final model correctly aligns methionine residues with the experimentally determined positions of the selenium atoms from the SAD analysis ( Supplementary Fig. 1b). Overall, the BsEzrA 22-562 monomer is a curved rod, forming a complete semicircle with a diameter of 120 Å. Groups of three a-helices pack together along its length (Fig. 1a), except at the C-terminus where there is a four-helix bundle. The formation of a semicircle confers the potential of the C-terminal four-helix bundle to interact with the membrane at the same time that the N-terminus of EzrA is embedded in it through its N-terminal TM helix; this can be modelled trivially by manually extending the N-terminal a-helix of repeat 1 ( Supplementary Fig. 1c). An antiparallel dimer is observed in the BsEzrA 22-562 crystal lattice as a result of crystallographic symmetry ( Supplementary Fig. 2a). The dimerization of BsEzrA 22-562 in the crystal buries some 3,300 Å 2 of surface area (B9% of the total) in an extensive, molecule-long interface. The anti-parallel dimer could not be detected in solution by sedimentation velocity, which indicates that the monomer and an oligomer of unknown arrangement are in equilibrium ( Supplementary Fig. 2b). However, bacterial twohybrid analysis in vivo shows an interaction between the N-and C-terminal regions of BsEzrA ( Supplementary Fig. 2c), which is consistent with the anti-parallel dimer seen in the crystals and which is the only oligomeric arrangement in the crystal that would enable the N-terminal TM helices of both subunits to insert into the same face of a membrane in vivo.
To obtain higher resolution structural information on EzrA, several fragments from different EzrA homologues were subjected to crystallization. A 190 residue N-terminal fragment of S. aureus EzrA, SaEzrA 24-214 (that also lacks its N-terminal TM helix), yielded crystals that diffracted to 2.6 Å resolution. Its structure (Fig. 1b, Supplementary Fig. 1d) was solved by molecular replacement using model phases from the equivalent portion of BsEzrA 22-562 , and was refined to a final crystallographic R free of 0.25 ( Table 1). The structures of SaEzrA 24-214 and BsEzrA 22-562 , which share only 20% sequence identity, superpose with root mean square deviations (RMSDs) of 2.4 Å on 154 paired Ca atoms ( Supplementary Fig. 3a). The BsEzrA 22-562 and the shorter SaEzrA 24-214 rods comprise multiple copies of the same B100 amino acid, triple helical bundle; BsEzrA 22-562 contains five such repeats and SaEzrA 24-214 contains just two. The individual repeating units can all be superimposed with RMSDs of between 1.5 and 2.3 Å on paired Ca atoms ( Supplementary Fig. 3b).

Structural homologues of EzrA occur only in eukaryotes.
A search for structural homologues of BsEzrA 22-562 using PDBefold 24 , DALI 25 and BLAST-3D 26 using default matching criteria did not identify any structures that superimposed convincingly along the entire length of the structure; only short segments could be aligned onto structural homologues. Representative matches from a BLAST-3D search included the eukaryotic cytoskeletal proteins vinculin 27 and a-actinin 28 , the nuclear importers karyopherin 29 and importin 30 and four-helical coiled coils such as the bacterial chemotaxis receptor, CheA 31 ( Supplementary Fig. 4a). The same searches were repeated with the structure of SaEzrA 24-214 as the query. PDBefold identified five matches, all of which resembled the hits from the BLAST-3D search with BsEzrA 22-562 ; one match was to vinculin and the other four were to members of the spectrin superfamily to which a-actinin belongs.
Although EzrA structures could be aligned partially onto other proteins that also have an up-down-up arrangement of helices, the match with the spectrin superfamily of structures stood out as most significant. Despite having an up-down-up helix configuration, vinculin, karyopherin, importin and CheA all lack discrete triple-helical bundles; in all of these protein families, each helix packs intimately against three, rather than two, other helices. In addition, although DALI searches matched both EzrA structures with other rod-shaped triple helical bundles, such as BAR domains 32 (Supplementary Fig. 4a), in these proteins there is only a single, triple helical bundle that is longer (4150 amino acids) than the repeat unit in EzrA. The linear head-to-tail arrangement of interconnected B100 amino-acid triple helical bundles appears to be unique to spectrins and to EzrA. Finally, no matter which search tool was used, the C-terminal four-helix bundle of EzrA superimposed only weakly on matching structures in the PDB, but it is notable that it has a superficial resemblance to the helical bundle fold of the actin-binding calponin homology domains 33 , which can be found at the termini of spectrin-type proteins.
EzrA is the founding member of the bacterial spectrin family. To our knowledge, the structural similarity of EzrA or any other bacterial protein to spectrins had not been predicted. From the logic below, we conclude that EzrA is the founding member of a  new family of bacterial spectrin-like proteins: (i) the eukaryotic structural homologues of EzrA share the same extended rod-like character, comprising repeating units of B100 amino acids arranged as triple helical anti-parallel bundles in a head-to-tail linear array (Fig. 2a, Supplementary Fig. 4b). Eukaryotic spectrins contain 4-20 of these units, called spectrin repeats, whereas EzrA contains 5 of them. (ii) Four of the five BsEzrA 22-562 spectrin repeats and the two in SaEzrA 24-214 can be superimposed on representative eukaryotic spectrin repeats with RMSDs of between 1.9 and 2.6 Å on matched Ca atoms (Fig. 2b); this is a high degree of similarity that is comparable to RMSDs among the EzrA spectrin repeats. Moreover, each superimposition accounts for greater than 70% of the EzrA spectrin repeat with paired amino acids coming from all three of the helices. (iii) The structural similarity of EzrA to the spectrin superfamily extends to the junctions and interfaces among adjacent repeats. As with all other crystallized spectrins, the repeat junctions in EzrA are formed from long continuous a-helices that form the third helix of one repeat and the first helix of the next. As also observed in several spectrins [34][35][36] , the SaEzrA 24-214 structure reveals intimate interactions at the interface between repeats involving the linker helix and loop residues. For example, the linker residues Tyr135 and Ile131 interact with loop residues Thr62, Leu56 and Tyr193 (Supplementary Fig. 5a,b). (iv) Although there are multiple examples of spectrin structures containing dual repeats, arranged like EzrA in a head-to-tail linear arrangement (Figs 1a and 2a, Supplementary Fig. 4b), the semi-circular curvature (Figs 1a and 2a) of BsEzrA 22-562 is unique. The principal axes of adjacent repeats in BsEzrA 22-562 intersect with angles of 37, 38, 33 and 53°, resulting in the pronounced overall bend. The angle between the two repeats in SaEzrA 24-214 is 34°, suggesting that full-length EzrA from S. aureus will also have a pronounced bend. The magnitude of these angles in EzrA is similar to that between erythrocytic b-spectrin repeats 14 and 15 (ref. 35), which is 37° (  Fig. 2c). Furthermore, a study of the six-repeat plakin domain of desmoplakin in solution by small-angle X-ray scattering was consistent with the plakin adopting an overall L-shape 37 , but BsEzrA 22-562 is the first structure with a sufficient bend between enough repeats to generate a full semi-circle. The absence of longer chains of spectrin repeats from the PDB may simply reflect a crystallization bias towards the shorter, relatively rigid, dual spectrin repeats that predominate in the database.
Nevertheless, there are facets of the EzrA structure that differentiate it from spectrins. For instance, there is no sequence homology between EzrA and any spectrin, although short segments of both proteins share a heptad repeat periodicity of hydrophobic amino acids ( Supplementary Fig. 6). Although the orientations of the three a-helices in the EzrA repeats are similar to those of spectrin repeats (Fig. 2a,b), the relative positions and therefore the connectivities of the helices are different (Fig. 2d). For example, the first 'up' helix in the EzrA repeat, helix A (blue, Fig. 2d), is superimposed on the third 'up' helix in the spectrin repeat from a-actinin, helix C (red, Fig. 2d). The second 'down' helix in the EzrA repeat is superimposed well on the a-actinin equivalent (helix B, coloured yellow in Fig. 2d), whereas the third 'up' helix in the EzrA repeat, helix C, is superimposed on the first 'up' helix in the a-actinin spectrin repeat, helix A. In other words, the left-handedness of the connectivity between helices in eukaryotic spectrin repeats (helices A-C are arranged in a counter-clockwise manner when viewed end-on) is reversed to be right-handed in EzrA (helices A-C are arranged clockwise).
Guided by limited proteolysis studies and the crystal structure, two truncated BsEzrA fragments were produced. BsEzrA  encompassed repeats one and two and BsEzrA 216-562 included the other three and the C-terminal domain (Fig. 1c). BsEzrA  bound to an SPR chip surface coated with immobilized BsFtsZ with fast kinetics, enabling equilibrium binding levels to be measured (Fig. 4a). Fitting the binding isotherm to a 1:1 binding model yielded a dissociation constant (K d ) of B13 mM for the BsFtsZ:BsEzrA 22-217 complex (Fig. 4b). The BsEzrA 216-562 fragment also bound to the BsFtsZ-coated chip, but with considerably slower kinetics ( Supplementary Fig. 7a), and fitting the sensorgrams to a 1:1 binding model yielded a K d of 2 mM. The specificity of the observed interactions is supported by the lack of any interaction of either BsEzrA 131-217 or bovine serum albumin with the FtsZ chip surface under equivalent buffer conditions (dashed cyan sensorgrams, Fig. 4a and Supplementary Fig. 7c).
The binding of both BsEzrA 22-217 and BsEzrA 216-562 to BsFtsZ implies that there are either two distinct FtsZ-binding sites on an EzrA monomer or a single site that extends across the interface between repeats 2 and 3. However, to be consistent with the crosslinking of BsFtsZ to repeat 1, a single binding site at the interface between repeats 2 and 3 would need to extend also to spectrin repeat 1, implying an intimate association of FtsZ with repeat 2. However, a fragment encompassing only repeat 2, BsEzrA 131-217, showed no detectable interaction with immobilized BsFtsZ (red dashed traces, Fig. 4a) in SPR, despite it having appreciable secondary and tertiary structure content ( Supplementary Fig. 8a,b). Therefore, it seems most likely that there are at least two distinct FtsZ interaction sites on an EzrA monomer, one of which is in spectrin repeat 1.
Effect of EzrA on FtsZ oligomerization. The BsEzrA-BsFtsZ interaction was analysed to map regions of EzrA responsible for the inhibition of FtsZ oligomerization. GTP-induced increase in light scattering provides a measure of the polymerization of FtsZ monomers into filaments and the bundling of filaments into higher order structures; we refer to these processes collectively as 'oligomerization'. Previous studies have shown that a twofold excess of BsEzrA over BsFtsZ reduced FtsZ oligomerization between two-and tenfold 16,38 . The pH was increased from 6.5, which was used in previous studies with this assay, to 7.3, to reduce the aggregation of BsEzrA 22-562 . Under these conditions, a twofold excess of BsEzrA 22-562 inhibited BsFtsZ oligomerization about twofold (Fig. 5a,b), consistent with previous studies 16,38 .
Remarkably, identical concentrations of BsEzrA 22-217 in this assay inhibited BsFtsZ oligomerization almost completely (Fig. 5a,b), whereas a twofold excess of BsEzrA 216-562 reduced BsFtsZ oligomerization by about one-third. An equimolar mixture of BsEzrA 22-217 and BsEzrA 216-562 also completely inhibited BsFtsZ assembly (Fig. 5a,b). The stronger inhibition of BsFtsZ assembly by the smaller  Fig. 8c). The relative levels of inhibition of BsFtsZ assembly by different BsEzrA fragments must be interpreted cautiously, but the marked inhibition by BsEzrA 22-217 points once more to a functionally important interaction of the N-terminal region of EzrA with FtsZ.
To ascertain whether the marked inhibition of BsFtsZ oligomerization by BsEzrA 22-217 reflects direct inhibition of FtsZ polymerization into filaments, or the blocking of filament bundling into higher order structures, its effect on the GTPase activity of BsFtsZ was examined. Head-to-tail polymerization of FtsZ into filaments is dependent upon GTP binding 39 , and dissociation of FtsZ polymers into monomers is dependent upon GTP hydrolysis 5 . Therefore, either the inhibition of GTP binding or the stimulation of the GTPase activity could result in the inhibition of FtsZ head-to-tail polymerization by EzrA. The presence of BsEzrA 22-217 did not markedly affect the binding of the fluorescent BODIPY-GTP-g-S to BsFtsZ ( Supplementary  Fig. 9a), a result that is consistent with the very slight reduction  (1.6-fold) in GTP affinity for BsFtsZ in the presence of BsEzrA measured previously 40 . The GTPase activity of BsFtsZ was measured in the presence and absence of BsEzrA 22-217 , to determine whether EzrA affected GTP hydrolysis and thus FtsZ oligomerization. Both BsEzrA 22-562 and BsEzrA 22-217 had marginal effects on GTP turnover in a coupled GTPase assay ( Supplementary Fig. 9b) and in a malachite green assay that monitored release of free phosphate (Supplementary Fig. 9c). It would thus seem that EzrA has little impact on the binding of GTP by FtsZ or on its hydrolysis.
Finally, we examined the impact of BsEzrA 22-217 on the oligomerization of BsFtsZ 1-366 , which lacks the C-terminal 17 amino acids required for bundling of FtsZ filaments 41,42 . Consistent with previous observations 41 , the change in light scattering on adding GTP to BsFtsZ 1-366 was reduced markedly, by B15-fold relative to wild-type BsFtsZ ( Supplementary  Fig. 10a). As has been observed previously with BsEzrA 22-562 (ref. 43), EzrA  had no impact on the residual GTP-induced oligomerization activity of BsFtsZ 1-366 (Supplementary Fig. 10). These results suggest that EzrA does not affect the head-to-tail polymerization of FtsZ into filaments but instead largely acts by preventing the bundling of FtsZ filaments into higher-order assemblies.

Discussion
Spectrins are ubiquitous proteins involved in the organization of the cytoskeleton and membrane processes in animals. A typical spectrin has an elongated molecular shape due to multiple, connected repeats of three anti-parallel a-helices, which are slightly curved and coil around each other. Individual spectrin repeats are approximately 50 Å long and 20 Å wide and are similar in length (99-122 amino acids). Their primary aminoacid sequences differ significantly, but the bundling of the helices is stabilized by regularly spaced hydrophobic side chains. EzrA shares all of these features with the spectrin superfamily of proteins, and the structural similarity of the EzrA repeats with those of other spectrin-like proteins is striking (Fig. 2). Although many proteins are known to comprise bundles of antiparallel a-helices, EzrA is, to the best of our knowledge, the only bacterial protein that is formed from serial spectrin-like repeats.
Structure-based sequence alignments do not show any meaningful sequence homology either among the five EzrA repeats, or between EzrA and spectrins. Such a lack of sequence conservation is not without precedence within the spectrin superfamily 44 . The spectrin repeat is a degenerate sequence established predominantly on the heptad repeat pattern of hydrophobic amino acids in the primary sequence, a characteristic of lefthanded helical supercoiled proteins 45 . The degeneracy is exemplified by the observation that there are only four invariant residues in an alignment of the four spectrin repeats in human a-actinin 2, one invariant residue in an alignment of the ten spectrin repeats in the human Duchenne muscular dystrophy protein and none in the seventeen spectrin repeats of human erythrocyte b-spectrin. Therefore, even the tryptophan in the first helix in the repeat, originally described as critical for thermodynamic stability of the spectrin repeat 46,47 , is not strictly maintained. The first and third repeats in EzrA each contain a highly conserved proline residue in helix A; Pro45 of SaEzrA in repeat 1 (Supplementary Fig. 11a) is conserved in eight out of nine phylogenetically diverse paralogues, whereas Pro228 (Pro227 in BsEzrA) in repeat 3 ( Supplementary Fig. 11b) is conserved in all of them. Although sequence analyses of spectrins have identified that a proline is found frequently in helix B 48 , neither helix A nor C has been reported to be interrupted by proline residues.
There is significant variation in the relative orientation of adjacent repeat units in spectrin structures 35,49 . However, EzrA is the first spectrin-like repeat-containing structure in which there is a pronounced enough bend between a sufficient number of repeats to form a complete semicircle. Molecular dynamics studies suggest the flexibility of the junction regions in spectrins is constrained appreciably, with the force constants for changing the angle between adjacent repeats determined 36 to be between 20 and 60 kcal mol À 1 rad À 2 . A considerable input of energy is thus likely to be necessary to convert the semicircular EzrA into a straight rod. The solvent contents of the EzrA crystals are unusually high (BsEzrA 22-562 80%; SaEzrA 24-214 74%); 480% of all entries in the PDB have a solvent content o60% and only 1.1% have a solvent content 475%. The high solvent content of these crystals does not concur with the release of sufficient energy during crystallization to drive thermodynamically unfavourable events. Therefore, the semicircular shape is highly unlikely to represent an artefact induced by crystal packing. The extended curved structure is a property EzrA shares with SepF, an FtsZ bundling protein required for normal septum formation. B. subtilis SepF assembles into polymeric arcs 13 and rings 12 with a diameter (55 nm) larger than that of the EzrA semicircle (12 nm). FtsZ filaments are proposed to bind on the outside of the SepF curve, whereas the inner face interacts with the surface of the invaginating septum membrane 13 . An analogous intimate association of the much smaller inner face of the EzrA semicircle with the membrane in vivo would require extreme curvature of the lipid bilayer, which has not been reported to date.
A question raised by the BsEzrA 22-562 structure is the orientation of full-length EzrA relative to the lipid bilayer. A tilted orientation of the EzrA arch with respect to the membrane could result from the curvature of the membrane, the bending of the TM helix or flexibility in the junction between the TM helix and the cytoplasmic domain. The conserved proline in helix 1 of repeat 1 (Pro45 in S. aureus EzrA) may confer some additional flexibility in the junction between the body of the cytoplasmic domain and the membrane. However, an intimate association of the EzrA cytoplasmic domain with the surface of the membrane, as seen in the curve-shaped helical BAR domain proteins 32 , is not supported by electrostatic potential calculations on either of the EzrA structures presented here. A tilted orientation of the arch would position the C-terminal four-helix bundle away from the surface of the membrane. This domain contains a conserved motif between residues 505 and 511, termed the 'QNR patch' (Supplementary Fig. 1c), that is important for mid-cell localization 38 . The QNR patch has been proposed recently to bind directly to FtsZ, based upon the inhibition of FtsZ oligomerization by different EzrA fragments in a light scattering assay in vitro 50 .
In the simplest model for the orientation of EzrA, the TM helix would form a straight extension of the first few turns of helix A of the first spectrin repeat; secondary structure predictions suggest a continuous uninterrupted helix extends into the first helix in the cytoplasmic domain. This orientation would result in the cytoplasmic domain of EzrA forming an arch with the C-terminal helical bundle positioned close to the membrane surface ( Supplementary Fig. 1c). Although the sequence register in this region of the structure must be regarded cautiously, because of the paucity of methionine residues in this region of the sequence, this model would position key solvent-exposed residues in the QNR patch (R510, R512 and F511) to interact with the membrane, and preclude their direct interaction with Z-ring FtsZ polymers ( Supplementary Fig. 1c). This suggests rather that the QNR patch controls EzrA localization through direct interaction with the membrane or specific membrane-associated protein(s).
An arch over the membrane could allow EzrA to regulate FtsZ oligomerization by trapping FtsZ filaments in the gap between the membrane and the inside of the EzrA curve. Molecular modelling with the Methanococcus janaschii FtsZ dimer 51 and the Thermotoga maritima FtsA:FtsZ complex 7 reveals there is sufficient space beneath the arch, in both monomeric and dimeric forms of EzrA, to accommodate an FtsZ filament or even to simultaneously accommodate FtsA-mediated membraneanchored FtsZ and FtsA protofilaments (Fig. 6). Trapping FtsZ protofilaments inside the arch is intuitively an effective means to disrupt lateral interactions between protofilaments and could additionally anchor contractile protofilaments to the membrane. An individual filament inside the arch cannot engage in lateral interactions with filaments outside of it (Fig. 6), providing an additional level of Z-ring regulation. The ability of the arch to accommodate FtsZ protofilaments could be modulated depending on the cellular location of EzrA by the curvature of the cell membrane and/or by competing interactions of EzrA with other divisome components such as GpsB. In addition, or alternatively, FtsA and FtsZ may interact with the outside face of the EzrA arch only, leaving the inner surface for mediating alternative protein-protein interactions.
A bacterial origin for the eukaryotic cytoskeleton has been suggested previously, in part, because of the similarity in structure and function of FtsZ to tubulin and FtsA to actin, although the primary amino-acid sequences are poorly related 1,52,53 and their structures differ in some respects. For example, two C-terminal a-helices (residues 381-437), which cover the surface of tubulin and are intimately associated with the globular body of the protein, are not visible in many FtsZ structures from different organisms [53][54][55][56][57] or are replaced by b-strands 51 . In addition, the relative positioning of the four subdomains differs in FtsA and actin 58 , and these two proteins are further distinguished by the unique presence in FtsA of subdomain 1C 58 . Our study extends the similarities between the eukaryotic and prokaryotic cytoskeletons to proteins that contain spectrin-like repeats. An established function of spectrins in eukaryotes is to connect actin filaments to other membrane-associated proteins and, at least in the case of plectin, to also regulate actin dynamics 59 . An analogous role can thus be envisaged for EzrA during cell division, co-ordinating FtsA/FtsZ dynamics inside the cell to peptidoglycan biosynthesis on the outside, an activity that is essential for cytokinesis and which is effected by the recruitment of PBP1 to the site of division by its interaction with EzrA 17 .

Methods
Full details of the Methods are supplied in the Supplementary Information.
Expression plasmid construction. The genes encoding fragments of Bs and SaEzrA and T7 epitope-tagged FtsZ were cloned into pET-based vectors. The plasmids pBS58 and pCXZ for overexpressing FtsZ were a gift from David Adams (Newcastle University).
Expression and purification of EzrA and EzrA fragments. Recombinant EzrA proteins were produced using E. coli strain BL21(DE3) carrying the appropriate plasmid. Expression was induced in Luria Bertani (LB) medium with isopropyl b-D-1-thiogalactopyranoside or by auto-induction. For the isopropyl b-D-1-thiogalactopyranoside -induced expression of selenomethionine-labelled proteins, the methionine-auxotrophic host strain B834 (DE3) was grown in a defined, minimal medium based on M9 supplemented with appropriate ions, vitamins and 19 amino acids; methionine was replaced by selenomethionine. For the BsEzrA proteins, the cells were harvested by centrifugation, before the cell pellet was resuspended in lysis buffer. Cells were lysed by sonication and clarified by centrifugation. The supernatant was filtered and then purified by Ni-nitrilotriacetic acid (Ni-NTA) affinity chromatography with elution effected by imidazole. The hexahistidine tag was removed by overnight incubation with thrombin, before further purification by size exclusion chromatography. The purified proteins were snap-frozen in liquid nitrogen for storage. SaEzrA 24-214 was produced without a tag, and was purified by ammonium sulphate fractionation and anion exchange chromatographic steps.
Expression and purification of BsFtsZ. Untagged BsFtsZ was co-expressed with E. coli FtsQ, FtsA and FtsW using plasmids pCXZ and pBS58, as described previously 60 . FtsZ was purified by ion exchange chromatography and serial ammonium sulphate precipitation at 20, 30 and 40% (w/v). The proteins precipitated with 40% ammonium sulphate were resuspended and purified further by size exclusion chromatography, before snap-freezing in liquid nitrogen for storage. T7-tagged BsFtsZ was purified by Ni-NTA affinity chromatography following the same procedures as used for His-tagged BsEzrA fragments, and the His-tag was removed by thrombin.
Crystallization and structure determination. BsEzrA 22-562 was crystallized using a mother liquor of 50 mM MES/NaOH, pH 6.7, 1 M tri-sodium citrate, 0.1 M ammonium sulphate, 0.2 M non-detergent sulphobetaine-201. The structure of EzrA was solved by single wavelength anomalous dispersion from a single selenomethionine-labelled crystal. The heavy atoms' substructure and initial phasing were determined in SHELX 61 . SHELXE 61 and PHASER 62 were used for subsequent density modification. The atomic model was built by manually docking a-helices into the electron density in COOT 63 . The amino-acid sequence was assigned by matching the positions of selenium atoms to methionine positions in the protein sequence. SLOOP 64 was used to find loops that best fit the electron density map. The geometry of the loop-helix junctions was corrected using RAPPER 65 and PHENIX. REFINE 66 was used for refinement of the model. SaEzrA 24-214 was crystallized in a buffer of 34% 2-methyl-2, 4-pentanediol, 0.1 M Tris-HCl, pH 7.5, and its structure was solved by molecular replacement using equivalent atoms from the BsEzrA 22-562 structure as a search model in  FtsA has been oriented to place the membrane-anchored C-terminus close to the lipid bilayer and its binding site for FtsZ near the FtsZ filament. The FtsZ filament has been positioned to place its C-terminal FtsA-binding region (disordered in FtsZ crystal structures) close to the FtsA filament. An FtsZ filament trapped beneath the EzrA arch cannot bundle together with filaments outside the arch. Although the crystallographic dimer is represented, the same effect on bundling could be made by EzrA monomers. PHASER 62 and refined to convergence using COOT 63 and REFMAC5 (ref. 67). Summaries of the data collection and model refinement statistics are provided in Table 1, and Molprobity 68 was used for structure validation reports.
EDC crosslinking. EzrA fragments were mixed with T7-tagged FtsZ and cross-linked with EDC before analysis by SDS-polyacrylamide gel electrophoresis (SDS-PAGE) and western blotting. To map the crosslink site on EzrA, crosslinked proteins were purified from non-crosslinked proteins by Ni-NTA before NTCB cleavage, SDS-PAGE and immunoblotting with an anti-T7 antibody.
Light scattering. Light scattering measurements used a Varian Cary Eclipse fluorimeter at 30°C, with all samples in a buffer of 50 mM Tris-HCl, pH 7.3, 20 mM KCl, 3 mM magnesium acetate. Samples were pre-incubated at 30°C, centrifuged in a microfuge and the supernatant transferred to a quartz microcuvette. GTP was added to a final concentration of 1 mM and light scattering recorded for at least 30 min. The GTP-induced change in light scattering in the presence of EzrA and FtsZ was normalized relative to the average GTP-induced scattering change for FtsZ alone.
Surface plasmon resonance. All SPR experiments used a Biacore X100 instrument with the chip surface at 25°C. Proteins were immobilized on Biacore CM5 chip surfaces by standard protocols. The binding data were all fit using the Biacore X-100 evaluation software; for affinity fits, the response on the surface 4 s before the end of the injection were fit as a function of the concentration of the injected EzrA fragment. Each SPR titration shown was performed independently at least twice.
GTPase assay. The GTPase activity of FtsZ was measured under the same buffer conditions as used for the light scattering assays. In an NADH oxidation-coupled continuous regenerative assay 69 , 150 ml samples were incubated at 30°C in a microplate reader after the addition of GTP to 1 mM. NADH consumption was measured, which is coupled to GDP production, from the decrease in absorption at 340 nm. For malachite green assays, the release of inorganic phosphate was measured 20 min after adding GTP to 2 mM. A standard curve was generated using phosphate ions in the concentration range 50-450 mM.