The bacterial actin-like cell division protein FtsA forms curved antiparallel double filaments upon binding of FtsN

Cell growth and division of walled bacteria depend on the synthesis and remodelling of peptidoglycan (PG). These activities are carried out by two multiprotein complexes, the elongasome and the divisome during cell elongation and division, respectively. Filaments of tubulin-like FtsZ form the cytoplasmic scaffold for divisome assembly, the Z-ring. In E. coli, the actin homologue FtsA anchors the Z-ring to the membrane and recruits downstream divisome components, including bitopic FtsN. FtsN is recruited late and activates the periplasmic PG synthase FtsWI. To start unravelling the activation mechanism involving FtsA and FtsN, we showed that E. coli FtsA forms antiparallel double filaments on lipid monolayers when also binding FtsN’s cytoplasmic tail, and that Vibrio maritimus FtsA crystallised as an equivalent double filament. We structurally located the FtsA-FtsN interaction site in FtsA’s IA-IC interdomain cleft and confirmed FtsA double filament formation in vivo using site-specific cysteine cross-linking. FtsA-FtsN double filaments reconstituted on and in liposomes preferred negative Gaussian curvature, as was previously shown for the elongasome’s actin, MreB. MreB filaments serve as curvature-sensing “rudders”, orienting insertion of PG around the cell’s circumference. We propose that curved antiparallel FtsA double filaments function similarly in the divisome: FtsA filaments, together with dynamic FtsZ filaments orient and concentrate cell-constricting septal PG synthesis in the division plane.


INTRODUCTION
In non-spherical, walled bacteria, cell shape is determined by the peptidoglycan sacculus, the load-bearing structure counteracting turgor pressure (Typas et al., 2011).
Insertion of new glycan strands into the sacculus is orchestrated by two related multiprotein complexes: the divisome, acting in cell division, and the elongasome (or rod complex), acting in cell elongation (Szwedziak and Löwe, 2013). In Escherichia coli, both complexes span the entire cell envelope and contain the cytoplasmic, membrane-binding and filament-forming actin homologues FtsA (divisome) and MreB (elongasome) (Pichoff and Lutkenhaus, 2005;Salje et al., 2011).
In E. coli and many other bacteria, FtsA is the main membrane anchor for the cell division ring, the Z-ring, which initiates and organises cell division (Geissler et al., 2003;Pichoff and Lutkenhaus, 2002). The Z-ring is mainly formed by filaments of the tubulin homologue FtsZ (Bisson-Filho et al., 2017;Nogales et al., 1998;Yang et al., 2017).
FtsZ treadmilling dynamics, driven by FtsZ's GTPase activity, were shown to be essential for the initial condensation of FtsZ filaments into the Z-ring, but seem to become dispensable after partial constriction of the septum (Monteiro et al., 2018;Squyres et al., 2021;Whitley et al., 2021).
Small amounts of FtsN are recruited early to the divisome in a FtsA-dependent manner (Busiek and Margolin, 2014). The FtsA-FtsN interaction is known to involve FtsA's for actin-like proteins uniquely positioned IC domain and FtsN's cytoplasmic tail, which comprises ~32 amino acids in E. coli (Baranova et al., 2020;Busiek et al., 2012;Pichoff et al., 2015). Two FtsN suppressor mutations in FtsA located in its IC domain (Bernard et al., 2007;Liu et al., 2015) further support the idea that FtsN binds the IC domain of FtsA. In FtsN, a conserved stretch of basic amino acids in its cytoplasmic tail is required for interaction with FtsA in vitro (Baranova et al., 2020;Busiek et al., 2012). In a zipA null background, in which the FtsA-FtsN interaction becomes essential, mutation of the basic stretch in FtsN was however permissible, whereas a D5N mutation was not (Pichoff et al., 2015). It has been proposed that interaction with FtsN depolymerises FtsA and, thereby, allows recruitment of downstream divisome components via binding sites that were (partially) occluded in the FtsA polymer (Pichoff et al., 2015).
However, little else has been reported regarding FtsA's polymerisation state in the divisome and its interaction sites with other divisome proteins. Despite the uniqueness of FtsA's IC domain amongst the actin-like proteins (van den Ent and Löwe, 2000), FtsA was shown to form single protofilaments that recapitulate some structural features of bona fide actin protofilaments (Szwedziak et al., 2012). Double filaments were shown to be the smallest functional unit of all other known actin homologues (Wagstaff and Löwe, 2018) and, indeed, Thermotoga maritima FtsA was shown to form double filaments (Szwedziak et al., 2012). Furthermore, several mutations in E. coli FtsA, originally described as ZipA suppressor mutations, were shown to facilitate double filament formation of FtsA on supported lipid monolayers (Schoenemann et al., 2018).
In contrast, the actin-like protein of the elongasome, MreB, adopts the canonical actinlike fold (van den Ent et al., 2001). MreB binds to membranes directly and forms curved antiparallel double filaments (Hussain et al., 2018;Salje et al., 2011;van den Ent et al., 2014). This enables a curvature-sensing mechanism that allows MreB filaments to align with the axis of highest principal curvature, the short axis of the cell in the case of rod-shaped cells, such as Bacillus subtilis (Hussain et al., 2018;Wong et al., 2019). Hence, the elongasome uses MreB double filaments as "rudders" to direct movement of its bipartite PG synthase RodA-PBP2 (equivalent to FtsWI in the divisome) and, thereby, directs insertion of new PG strands around the cell circumference (Dion et al., 2019;Garner et al., 2011). It is thought that the radially inserted PG hoops enforce rod shape by mechanically limiting cell width expansion.
Here, we showed that FtsA from the Gram-negative bacterium Vibrio maritimus crystallised as antiparallel double filaments. Using supported lipid monolayers, we demonstrated that E. coli FtsA forms very similar antiparallel double filaments upon binding the short, cytoplasmic tail of FtsN. Using crystallography, we showed that FtsN binds FtsA in the IA-IC interdomain cleft. Furthermore, we demonstrated the in vivo significance of FtsA double filaments by site-specific cysteine cross-linking in E. coli.
FtsA-FtsN filaments reconstituted on and in liposomes preferred negative Gaussian curvature, suggesting that FtsA double filaments in the divisome sense curvature using a similar mechanism as has been proposed for MreB filaments in the elongasome. We therefore propose that FtsA double filaments provide a curvatureguided mechanism to orient cell-constricting septal PG synthesis, which is further organised by dynamic FtsZ filaments into the required ring structure at the division site.

Vibrio maritimus FtsA crystallises as an antiparallel double filament
Previously obtained structures of FtsA are limited to the two organisms Staphylococcus aureus and Thermotoga maritima (Fujita et al., 2014;Szwedziak et al., 2012;van den Ent and Löwe, 2000). Encouraged by a recent study reporting double filaments formed by E. coli FtsA (Schoenemann et al., 2018), we reasoned that further FtsA crystal structures, especially from Gram-negative organisms such as E. coli, might provide new insights into the architecture of FtsA filaments and their function in the divisome.

E. coli FtsA forms antiparallel double filaments upon binding the cytoplasmic tail of FtsN, EcFtsN 1-32
As described previously , we found that E. coli FtsA forms "minirings" on supported lipid monolayers (Figure 2g, "WT + no peptide"). Given that filament architecture mutations and FtsN suppressor mutations in E. coli FtsA map to lateral interfaces in the VmFtsA double filament (Figure 1c), we hypothesised that FtsN could induce double filament formation of E. coli FtsA. The short, cytoplasmic tail of FtsN (comprising ~32 amino acids in E. coli) (Supplementary Figure S2f) interacted with FtsA as had been shown previously (Busiek et al., 2012;Busiek and Margolin, 2014;Pichoff et al., 2015). Using surface plasmon resonance (SPR), we showed that The short cytoplasmic tail of FtsN harbours two sequence motifs that are conserved among E. coli and related proteobacteria with similar FtsA and FtsN sequences: a conserved 3 R/KDY 6 (E. coli amino acid positions) motif near the N-terminus (Pichoff et al., 2015) and two to three stretches of basic amino acids, the most prominent of which in E. coli FtsN is 16 RRKK 19 (Busiek et al., 2012) (Supplementary Figure S2f, g).
In E. coli, a D5N mutation in the conserved N-terminal motif of FtsN was shown to impair FtsA-FtsN interactions in vivo (Pichoff et al., 2015), but did not affect binding or co-localisation with FtsA-FtsZ filaments on supported lipid bilayers in vitro (Baranova et al., 2020). In contrast, and somewhat confusingly, mutation of the basic 16 RRKK 19 stretch abrogated binding in vitro (Baranova et al., 2020;Busiek et al., 2012), yet was permissible in vivo under conditions for which the FtsA-FtsN interaction becomes essential (Pichoff et al., 2015). We therefore tested a set of EcFtsN 1-32 truncations and mutations in our SPR and monolayer assays to determine the effect of those modifications on FtsA-FtsN interactions and on double filament formation (Supplementary Figure S2h,i). Mutation or deletion of any one of the three stretches of basic amino acids in EcFtsN 1-32 reduced binding affinity for FtsA and, consequently, its ability to promote FtsA double filament formation. EcFtsN 1-32, D5N and EcFtsN 1-33, scrambled (Supplementary Table T 4), which has been described previously but was shortened through removal of its tags in this study (Baranova et al., 2020) We then tested whether any of the FtsN suppressor (Bernard et al., 2007;Liu et al., 2015) or filament architecture and ZipA suppressor mutations (Schoenemann et al., 2018) in FtsA would impact FtsN-induced double filament formation of FtsA ( Figure   1c). While the FtsN suppressor mutants FtsA E124A and FtsA I143L , similar to wildtype FtsA, formed "mini-rings" in the absence of EcFtsN 1-32 , they more readily formed double filaments than wildtype FtsA at three-(7 µM) and six-fold (13 µM) molar excess of EcFtsN 1-32 (Figure 2g). The filament architecture and ZipA suppressor mutants FtsA G50E and FtsA R286W (also known as FtsA*) formed fewer "mini-rings" than wildtype FtsA in the absence of FtsN (Figure 2g). For FtsA G50E , "mini-rings" and short double filaments were observed. FtsA R286W predominantly formed arcs, curved, short and single filaments. Similar to the FtsA E124A and FtsA I143L proteins, FtsA G50E and FtsA R286W formed many more double filaments than wildtype FtsA at three-(7 µM) and six-fold (13 µM) molar excess of EcFtsN 1-32 (Figure 2g) Table T Table T  The FtsA-FtsN interaction is reminiscent of that between PilM and PilN, two proteins involved in type IV pilus formation (Karuppiah and Derrick, 2011;McCallum et al., 2016). PilM, which also has a IC subdomain, is a structural homologue of FtsA, and PilN, similar to FtsN, is a bitopic membrane protein with a short, cytoplasmic tail comprising about 30 amino acids (Karuppiah and Derrick, 2011;McCallum et al., 2016). As mentioned, the low resolution of the electron density for the VmFtsN 1-29 peptide in the VmFtsA 1-396 -VmFtsN 1-29 co-crystal structure limited our ability to assign the peptide register and only peptide density associated with two FtsA monomers (PDB 7Q6I chains A and C) was interpretable. Based on the geometry in vivo, where FtsA binds to the cell membrane through its C-terminal amphipathic helix, we reasoned that the N-terminal half of VmFtsN 1-29 is expected to interact with FtsA, as the globular domain of FtsA was reported to be several nanometres away from the inner membrane in E. coli (Szwedziak et al., 2014). The C-terminal half of the VmFtsN 1-29 peptide would be confined to the proximity of the inner membrane, as it is linked to the transmembrane helix in the full-length FtsN protein (Supplementary Figure S2f We further studied the VmFtsA 1-396 -VmFtsN 1-29 interaction by nuclear magnetic resonance (NMR) spectroscopy. We assigned the peptide backbone of Gly-Gly-VmFtsN 2-29 , a VmFtsN 1-29 construct in which two glycine residues replace the N-terminal methionine (Supplementary Figure S3c). Next, we analysed peak broadening (reduction in peak intensity due to the slower tumbling rate of FtsA-bound Gly-Gly-VmFtsN 2-29 peptide) upon addition of equimolar amounts of VmFtsA 1-396 (Supplementary Figure S3d). Intensity reduction was most prominent in the stretch of amino acids Y6-G10 (Supplementary Figure S3d), suggesting a slightly shifted binding motif compared to our density-based assignment (M1-R8) (Supplementary Figure   S3e). We were not able to generate a good fit of these residues into the electron density map of the VmFtsA 1-396 -VmFtsN 1-29 co-crystal structure. For technical reasons, the NMR experiments were carried out at pH 6.0, whereas binding experiments were done at pH 7.7 and crystallisation was achieved at pH 8.5.

FtsA double filaments prefer negative Gaussian curvature
Based on the observation that VmFtsA 1-396 -VmFtsN 1-29 double filaments were curved in the co-crystal structure (Figure 3b), we hypothesised that the intrinsic curvaturepreference of MreB double filaments might also be a feature of FtsA-FtsN double filaments (Hussain et al., 2018;Salje et al., 2011). The curvature preference of MreB filaments enables a curvature-sensing mechanism that allows them to robustly align in cells with the axis of highest principal curvature, which corresponds to the short axis of rod-shaped bacteria such as E. coli and B. subtilis (Hussain et al., 2018;Wong et al., 2019). The elongasome uses the aligned MreB filaments to guide glycan strand synthesis around the cell's circumference, producing radially inserted glycan strands that are believed to mechanically reinforce rod shape (Dion et al., 2019;Garner et al., 2011;Hussain et al., 2018).

E. coli FtsA forms antiparallel double filaments in vivo
To complement and validate our in vitro findings, we investigated double filament formation of FtsA in vivo using site-specific cysteine cross-linking in E. coli. For greater feasibility of targeting the native ftsA locus, which is located in the dense division and cell wall (dcw) gene cluster, and to increase throughput with a number of mutants needed, we used a derivative of replicon excision for enhanced genome engineering through programmed recombination 2 (REXER 2) (Wang et al., 2016) (Supplementary Figure S4a). We inserted a neoR marker downstream of lpxC for positive selection.
For visualisation via Western blotting, we introduced a 3x HA-tag (comprising 40 amino acids including linkers) into the H7-S12 loop of FtsA van den Ent and Löwe, 2000). Cysteine mutations for in vivo site-specific cross-linking were designed based on the VmFtsA 1-396 double filament crystal structure (PDB 7Q6F) BMOE enters living E. coli cells and rapidly cross-links closely spaced thiols such as cysteine side chains in vivo. FtsA species were visualised by SDS-PAGE and Western blotting against the 3x HA-tag in FtsA ( Figure 5c). We detected efficient formation of crosslinked FtsA dimers for both the ftsA 3x HA, P98C, S118C and ftsA 3x HA, E199C, S252C double mutants, but only weak background signal spread across multiple species in single cysteine mutant controls. We also detected higher order polymers for the ftsA 3x HA, E199C, S252C mutant since the open symmetry of the longitudinal filament contact allows for more than two FtsA monomers to be cross-linked, through chaining.
To probe the lateral association in vivo further, we generated the two, additional FtsA single cysteine mutant strains ftsA 3x HA, D123C and ftsA 3x HA, Q155C , which probe the lateral FtsAi-FtsAi* and FtsAi-FtsAi*−1 interface of the FtsA double filament, respectively ( Figure 5d). These single cysteine mutants may cross-link to their symmetry mates because of the local C2 symmetry in each of the two lateral filament interfaces.
However, Cβ-Cβ distances in the VmFtsA double filament structure of 3.9 Å for ftsA 3x HA, D123C and 12.6 Å for ftsA 3x HA, Q155C indicated that cross-linking with BMOE with an expected cross-linking distance of ~8 Å might be inefficient, which prompted us to try thiol-directed cross-linkers of different lengths (Figure 5f). The ftsA 3x HA, D123C mutant showed more efficient cross-linking of FtsA using dibromobimane (dBBr) than using BMOE (Figure 5e, left). In case of the ftsA 3x HA, Q155C mutant, BMOE cross-linking did not lead to efficient formation of covalent FtsA dimers, whereas treatment with the longer maleimide cross-linkers 1,4-bismaleimidobutane (BMB) and bismaleimidohexane (BMH) did (Figure 5e, right). Taken together, our data strongly suggest that FtsA forms protofilaments in cells and, that these protofilaments are further arranged into antiparallel double filaments as suggested by the VmFtsA 1-396 crystal structure (PDB 7Q6F) .

DISCUSSION
Here, we found that FtsA forms antiparallel double filaments in E. coli and that in vitro the same filaments are induced by the binding of the cytoplasmic tail of FtsN. The only other polymerising actin-like protein known to form antiparallel and, therefore, apolar double filaments is the actin homologue of the elongasome, MreB, which serves as a "rudder" (Hussain et al., 2018;van den Ent et al., 2014;Wagstaff and Löwe, 2018).
Given the shared preference for negative Gaussian curvature displayed by FtsA-FtsN and MreB double filaments reconstituted in and on liposomes (Hussain et al., 2018;Salje et al., 2011;van den Ent et al., 2014) (Figure 4), we propose that MreB and FtsA share the same curvature sensing mechanism. Hence, we put forward a model for curvature-guided cell constriction by FtsA-FtsN double filaments, which serve as a "rudder" to align the direction of the divisome's PG synthesis activity, more precisely glycan strand synthesis, with the cell's circumference ( Figure 6a).
In our model, we discriminate three distinct phases in divisome assembly and maturation: unaligned FtsZ filaments at midcell, a fully assembled divisome aligned with the short axis of the cell, and a fully activated divisome synthesising septal PG and driving cell constriction ( Figure 6a). However, the temporal order of individual recruitment and activation events during divisome maturation remains largely unknown and would benefit from in vivo single molecule imaging of FtsA. FtsZ filaments are positioned at midcell to determine the division plane but are unaligned in the absence of an alignment mechanism (Figure 6a, left). After recruitment of divisome proteins and Z-ring condensation (Squyres et al., 2021;Whitley et al., 2021), FtsN-induced FtsA double filaments will align themselves and the other divisome components with the short axis of the cell, possibly aided by FtsZ treadmilling, which may provide a distribution mechanism that avoids FtsA filaments and divisome components to cluster in one spot (Bisson-Filho et al., 2017;McQuillen and Xiao, 2020;Yang et al., 2017) (Figure 6a, middle). We propose that curvature-sensing FtsA double filaments provide a solution to the FtsZ alignment problem noted previously (Du and Lutkenhaus, 2019). Consistently, the fraction of directionally treadmilling FtsZ filaments was reported to be decreased in a ΔftsA strain of B. subtilis (Squyres et al., 2021). Together FtsA double filaments and treadmilling FtsZ filaments we suggest align and evenly distribute divisome components in the narrow division plane. Most importantly, this will restrict movement of FtsWI, the bipartite PG synthase of the divisome, in such a way that cell-constricting septal PG synthesis follows the cell's circumference (Figure 6a, right). In this context, FtsN might function as an activation switch of the divisome by coordinating activities of FtsA and FtsWI. Recently, a more direct interaction between FtsA and FtsW has been proposed .
Based on bacterial and yeast two-hybrid assays, it has been suggested that FtsN promotes divisome maturation through depolymerisation of FtsA, which would free FtsA's IC domain and thereby allow recruitment of downstream divisome components (Pichoff et al., 2012;Pichoff et al., 2015Pichoff et al., , 2018. In contradiction to that we found that the short cytoplasmic tail of FtsN promotes FtsA polymerisation, or more precisely FtsA double filament formation (Figure 2c, d and g). Our work adds to previous evidence illustrating that FtsA can form different polymers Schoenemann et al., 2018;Szwedziak et al., 2012). Using FRET probes attached to FtsA, a complementary fluorescence microscopy study found FtsA* to be less polymeric than wildtype FtsA on supported lipid bilayers (Radler et al., manuscript in preparation). Addition of cytoplasmic FtsN peptide however induced FtsA* polymerisation to levels comparable to wildtype FtsA, again questioning the prevalent model that FtsN depolymerises FtsA. Further, our data clearly demonstrate that the FtsA double filament is compatible with FtsN binding, as could be the case for other downstream divisome components. Double filament formation will also position FtsA's IC domain close to the inner membrane (Supplementary Figure S1c), which could facilitate recruitment of downstream divisome components that possibly bind to FtsA's IC domain such as FtsQ (Baranova et al., 2020).
Taken together with recent reports establishing FtsWI (Taguchi et al., 2019) and RodA-PBP2 (Cho et al., 2016;Sjodt et al., 2020) as bipartite PG synthases, our data on the similarities between FtsA and MreB double filaments strengthen the previously proposed evolutionary relationship between the divisome and elongasome (Szwedziak and Löwe, 2013) (Figure 6b). In Chlamydia, one of the few bacteria that lack FtsZ, MreB was implied to organise cell division (Jacquier et al., 2015;Ranjit et al., 2020), yet another indication that divisome and elongasome share at least some basic principles of function.
In this context it is also interesting to speculate which of FtsZ's many functions in cell division "converted" the presumably ancestral elongasome into the divisome. Our model suggests that FtsZ is the long-range organiser of the division site which ensures that septal PG synthases are functioning in a single division plane, only, and are evenly distributed around the cell's circumference. FtsA on the other hand aligns these PG synthases with the orientation of the division ring, making sure septal PG glycan synthesis by the divisome goes around the ring's circumference. It will require further studies before we truly understand, or can reconstitute FtsZ-based cell division but our present study highlights the central role FtsA polymerisation might play in this process and, further, illustrates important similarities between elongasome and divisome mechanisms.

Expression plasmids
Expression plasmids (Supplementary Table T 2) were cloned by isothermal assembly using NEBuilder HiFi DNA Assembly Mix (NEB). Plasmids were transformed into E. coli MAX Efficiency DH5α (ThermoFisher) or C41(DE3) cells (Lucigen or Sigma) for plasmid propagation and protein expression, respectively.

Protein expression and purification
The amino acid sequences of all proteins used in this study are listed in Supplementary Table T 3. If applicable, removable tags that were cleaved during purification are indicated. All purification steps were carried out on ice or in a cold room at 4-6°C unless stated otherwise. Buffers were prepared in Millipore water (MPW), pH-adjusted at room temperature and filtered through a 0.2 μm PES filter.

SUMO protease (GST-SENP1)
GST-SENP1 (pTN_AN_902) was expressed in C41(DE3) cells. Cells were grown in 2xTY medium, supplemented with 100 µg/ml ampicillin, at 37°C. Expression was induced at OD600 = 0.6-0.8 by adding 0.5 mM IPTG. Cells were grown over night at 18°C and harvested by centrifugation. Cells were lysed in buffer SA (50 mM Tris/HCl, 150 mM NaCl, 2 mM TCEP, 1 mM EDTA, 5 % glycerol, pH 8.5), supplemented with DNase, RNase and cOmplete EDTA-free Protease Inhibitor Cocktail (Roche), using a cell disruptor at 25 kpsi (Constant Systems). The lysate was cleared by ultracentrifugation at 100,000x g for 30 min at 4°C. The supernatant was added to pre-washed Glutathione Sepharose 4B beads (Cytiva) and incubated for 2h at 4°C, with gentle stirring. The sample was passed through an Econo gravity flow column (Bio-Rad) and washed 5x with 50 ml buffer SA, followed by 2x 50 ml buffer SB (buffer SA with 500 mM NaCl) and 2x 50 ml buffer SA. The protein was eluted in 5x 5 ml buffer SA supplemented with 10 mM reduced glutathione. Peak fractions were pooled and concentrated with a Vivaspin 20 concentrator (30 kDa MWCO, Sartorius).

Full-length FtsAs
Cells were lysed in buffer LB2 (50 mM Tris/HCl, 500 mM NaCl, 5 mM TCEP, 10 mM Protein mass for each batch was verified by ESI-TOF mass spectrometry on a Micromass LCT mass spectrometer (Waters).
Search models used for molecular replacement are indicated in Supplementary Table   T 1. Models were rebuilt manually using MAIN (Turk, 2013) and COOT (Emsley et al., 2010) and refined using REFMAC (Murshudov et al., 1997) and PHENIX (Afonine et al., 2018). Models were validated using PROCHECK (Laskowski et al., 1993) and MOLPROBITY (Williams et al., 2018). Final statistics are summarised in Supplementary Table T 1, and the structure factors as well as atomic coordinates have been deposited in the Protein Data Bank (PDB) with accession codes 7Q6D, 7Q6G, 7Q6F and 7Q6I. Note that VmFtsN 1-29 (chains X and Y) density was modelled as VmFtsN 1-8 for refinement but deposited as polyAla.

Surface plasmon resonance (SPR)
SPR was performed using a Biacore T200 using CM5-sensor chips (Cytiva). Both reference control and analyte channels were equilibrated in binding buffer ( Figure S2i). After reference and buffer signal correction, sensogram data were fitted using Prism 8.0 (GraphPad Software Inc). The equilibrium response (Req) data were fitted to a single-site interaction model to determine Kd: where C is the analyte concentration and Rmax is the maximum response at saturation and B is the background resonance. Binding constants are given as mean.

Fluorescence polarisation (FP)
Peptides EcFtsN 1-32 -C and VmFtsN Dissociation constants were calculated using a two-step model: where FLo and FHi are the anisotropy changes at saturation of low and high affinity sites with binding constants of KDLo and KDHi, respectively. Binding constants were averaged from 5 replicates and are given as mean ± SEM.
Supernatant and pellet fractions were separated carefully and analysed by SDS-PAGE. Protein band intensities were quantified using ImageJ 2.1.0 (Abràmoff et al., 2004). Given are mean ± sd of two technical replicates for EcFtsA 1-405 and VmFtsA 1-396 , respectively.

Lipid monolayer assays and negative stain electron microscopy
Two-dimensional lipid monolayers were prepared from E. coli polar lipid extract (Avanti Polar Lipids) (Ford et al., 2001;Levy et al., 1999). Wells of a custom-made

Cryo-EM of EcFtsA-FtsN 1-32 double filaments on lipid monolayer
Lipid monolayers were prepared as described above, with the exception that Quantifoil Au R0.6/1 300 mesh grids (Quantifoil) were used. EcFtsA at 0.1 mg/ml (~2.2 µM) was mixed with 1 mM ATP and EcFtsN 1-32 at 22 µM in binding buffer (50 mM HEPES/KOH, 100 mM KAc, 5 mM MgAc2, pH 7.7), and incubated for 20 min at RT. Grids were blotted from the side after attachment of the monolayer using Whatman No 1 filter paper and inserted into a Vitrobot Mark III (ThermoFisher) set to 20°C and 100% humidity. 3 µl of mixed protein sample were applied to the carbon side of the grid, incubated for 30 s and blotted for 12-15 s (0.5 s drain time, -15 blot force) before plunge-freezing into liquid ethane maintained at −180°C using a cryostat (Russo et al., 2016). Grids were imaged using a Tecnai G2 Polara (ThermoFisher) operating at 300 kV. Movies were collected on a Falcon III direct electron detector at a pixel size of 1.38 Å, −3.3 to −4 µm defocus and a total dose of 100 e − /Å 2 using EPU (ThermoFisher) for automated acquisition. Data were processed using MotionCor2 (Zheng et al., 2017), CTFFIND4 (Rohou and Grigorieff, 2015) and RELION3 (Zivanov et al., 2018). A total of 104,660 particles were automatically picked and extracted, with the presented 2D class average corresponding to 14,602 particles (Figure 2e).
Presented micrographs were contrast adjusted and blurred for display purposes.

Hydrogen deuterium exchange mass spectrometry (HDX-MS)
VmFtsA The first round of analysis and identification was performed automatically by the DynamX software; however, all peptides (deuterated and non-deuterated) were manually verified at every time point for the correct charge state, presence of overlapping peptides, and correct retention time. Deuterium incorporation was not corrected for back-exchange and represents relative, rather than absolute changes in deuterium levels. Changes in H/D amide exchange in any peptide may be due to a single or multiple amides within that peptide. Time points were prepared in parallel and data for individual time points were acquired on the mass spectrometer on the same day.

Strain construction
The general cloning and recombination strategy is illustrated in Supplementary Figure   S4a. The helper plasmid pKW20 (NCBI ID: MN927219.1) (Wang et al., 2016) was used for genome engineering. pKW20 encodes for a constitutively expressed tracrRNA as well as Cas9 and -Red components under control of a L-arabinoseinducible promotor. Acceptor strain SFB123 was created by integrating a pheS T251A,A294G -hygR double selection cassette downstream of the lpxC gene using -Red recombineering (Yu et al., 2000). pheS T251A,A294G /pheS* is a negative selection marker encoding a mutant phenylalanyl-tRNA synthetase that confers toxicity through misincorporation of 4-chloro-phenylalanine (4-CP) during translation (Miyazaki, 2015).
We found a long (5 kB) homologous region upstream of FtsA to benefit recombination efficiency. SFB143 was used as donor strain during conjugation. SFB143 is an E. coli MDS42 thistrain transformed with the non-transferrable conjugative plasmid pJF146 (NCBI ID: MK809154.1) (Fredens et al., 2019). pJF146 bears an apR resistance marker and a truncated nick site of the origin of transfer. SFB143 cells were made chemically competent enabling parallelised transformation (Inoue et al., 1990).
Shuttle plasmid pFB483 was designed with a pMB1 origin of replication, a pheS T251A,A294G -hygR double selection cassette, a CRISPR array targeting ftsW and the region just upstream of secM, and a ccdB toxin gene (outsert) flanked by BsaI acceptor sites via Golden Gate assembly (Engler et al., 2008). CRISPR arrays were designed to mediate scarless excision. pFB483 was propagated in a ccdB survival strain.
Targeting constructs were split into 2-3 modules for insertion of single or double point mutations, respectively. Initially, the internal 3x HA-tag (120 bp including a XhoI restriction site) was inserted into ftsA using three modules, resulting in sTN001 (ftsA269-3x HA-tag-270(SW), lpxC::neoR), and propagated in modules in the following. Point mutations were introduced by PCR and modules assembled into pFB483 via Golden Gate assembly with BsaI. The final targeting construct also introduced a kanamycin-selectable neoR marker downstream of the lpxC gene.
Assembled shuttle vectors were transformed into SFB143 and selected on LB agar plates supplemented with 200 µg/ml hygromycin-B and 50 µg/ml apramycin at 37°C.
Acceptor cells (SFB123) were grown to stationary phase in 5 ml LB supplemented with 10 µg/ml tetracycline. 4 ml of culture were harvested by centrifugation and washed three times in LB, before being transferred into 50 ml LB supplemented with 10 µg/ml tetracycline and 0.5 % w/v L-arabinose. Cells were grown at 37°C for 1h, harvested by centrifugation and washed three times in LB. In the meantime, donor transformants were washed off the agar plates using 2 ml LB and left at room temperature. All cultures were resuspended in LB to an OD600 of 40. 12.5 µl of acceptor cells were mixed 87.5 µl of donor cells and spotted onto well-dried TYE plates in 10-20 µl drops.
Spots were air-dried, before plates were incubated at 30°C for 1h. Cells were washed off the plates with LB and transferred into 50 LB supplemented with 12.5 µg/ml kanamycin and 10 µg/ml tetracycline. Cells were grown at 37°C for 1h, harvested by centrifugation and plated on LB agar plates supplemented with 12.5 µg/ml kanamycin, 10 µg/ml tetracycline, 2 % glucose and 2.5 mM 4-CP. Strains were single colony purified, and verified by marker analysis and colony PCR followed by XhoI digestion and Sanger sequencing. Strains with desired point mutations were cured of pKW20 by repeated growth in LB in absence of antibiotics, diluted 1:10 6 and plated on TYE plates. pKW20 was typically lost in one of eight cells after 2-3 growth cycles. Strains were verified by marker analysis and Sanger sequencing of PCR products covering the targeting region. Strains used in Figure 5c were further whole genome sequenced on a MiSeq (Illumina). NGS data were analysed using breseq v0.35.1 (Deatherage and Barrick, 2014). Table T 5.

Growth on solid media
Strains were streaked on the same TYE plate and incubated at 37°C overnight. The next morning, strains were re-streaked on a fresh TYE plate and incubated at 37°C for 12h.

Growth in liquid media
Strains were inoculated into 5 ml LB and incubated at 37°C overnight. Cells were diluted 1/1000 into fresh LB into a 96-well flat bottom plate in octuplicate. The plate was incubated at 37°C in a Tecan microplate reader with regular shaking. Absorbance at 600 nm wavelength was measured every 5 min for 24h. OD600 values were background corrected, normalised to the maximum OD600 value of each well, averaged and plotted as mean ± sd.

DIC imaging of exponential phase cultures
Strains were inoculated into 5 ml LB and incubated at 37°C overnight. The next day, cells were diluted 1/1000 in 5 ml fresh medium and incubated at 37°C. 2-3 µl of exponential phase cells (OD600 = 0.2-0.3) were applied onto an agarose pad and imaged on a Nikon Eclipse E800 microscope equipped with a 100x oil objective and a Photometrics Iris 9 CMOS camera using a differential interference contrast (DIC) imaging setup. Presented images were contrast-adjusted for display purposes.
Recombinants were selected for neoR and tetR markers and against the pheS* marker.
Strains were cured of the helper plasmid pKW20 by growth in absence of selection. b,