ZapA stabilizes FtsZ filament bundles without slowing down treadmilling dynamics

For bacterial cell division, treadmilling filaments of FtsZ organize into a ring-like structure at the center of the cell. What governs the architecture and stability of this dynamic Z-ring is currently unknown, but FtsZ-associated proteins have been suggested to play an important role. Here, we used an in vitro reconstitution approach combined with fluorescence microscopy to study the influence of the well-conserved protein ZapA on the organization and dynamics of FtsZ filaments recruited to a supported membrane. We found that ZapA increases the spatial order and stabilizes the steady-state architecture of the FtsZ filament network in a highly cooperative manner. Despite its strong influence on their large-scale organization, ZapA binds only transiently to FtsZ filaments and has no effect on their treadmilling velocity. Together, our data explains how FtsZ-associated proteins can contribute to the precision and stability of the Z-ring without compromising treadmilling dynamics.

For cytokinesis, bacteria need to build two new cell poles at the division site. This process is 2 initiated by a dynamic ring of FtsZ filaments, which forms at the mid-cell early during the cell cycle, 3 where it persists during the inward growth of the cell septum until it disassembles just before 4 division is completed 1-4 . Recent studies have shown that this Z-ring not only defines the location 5 of divisome assembly, but that FtsZ treadmilling drives the movement of peptidoglycan synthases 6 around the cell diameter, which in turn controls remodeling of the cell wall at the division site 5,6 . 7 Due to the important role of FtsZ polymerization dynamics for cell wall synthesis, the spatial and 8 temporal organization of filaments within the Z-ring needs to be tightly controlled. Indeed, bacteria 9 contain various proteins that directly bind to FtsZ and contribute to the precision of cell division. 10 For example, mutants of Escherichia coli lacking one of the FtsZ-associated proteins ZapA, ZapB, 11 Using a novel automated image analysis, we studied the behavior of FtsZ and ZapA on three 1 different spatial scales: First, we analyzed the spatiotemporal pattern of the membrane-bound 2 filament network; next, we studied the underlying polymerization dynamics of filament bundles, 3 and finally, we quantified the behavior of single molecules. Our experiments and quantitative 4 analyses reveal that ZapA is able to align FtsZ filaments in a highly cooperative manner, giving 5 rise to two distinct states: a state of low spatial order with fast reorganization dynamics and a 6 state of high spatial order and slow reorganization dynamics. Importantly, we show that the 7 treadmilling velocities are identical in these two regimes, demonstrating that ZapA is able to 8 stabilize filament bundles without affecting FtsZ polymerization dynamics. Our results suggest 9 that the function of the ZapA-FtsZ interaction is to cooperatively align FtsZ filaments at the division 10 site, which defines the track for treadmilling and thereby increases the spatiotemporal precision 11 of cell division. 12 To study the effect of ZapA on the architecture and dynamics of the FtsZ filament network, we 3 took advantage of an in vitro system based on supported lipid bilayers and purified proteins (Fig.  4  2E). In the absence of ZapA, the correlation curves obtained for different time points overlapped 3 with each other (Fig. 2E, left; black to blue gradient), showing that spatial order stayed constant 4 during the experiment. The presence of ZapA increases the correlation length, an effect that 5 becomes more pronounced with time, indicating that ZapA increases the spatial order of the 6 system ( Fig. 2E, right; black to red gradient). We quantified this time-dependent increase in spatial 7 order by comparing the characteristic correlation length, #(%), which corresponds to the distance 8 at which (()) = 0.5 (Fig. 2E insets and Fig. 2F). We found that the presence of ZapA led to a 9 continuous increase in spatial order from # + = 0.97 ± 0.06 µm to # '' = 1.74 ± 0.21 µm at steady 10 state ( Fig. 2F, red, n = 8), while ρ stayed constant in its absence (Fig. 2F, blue, n = 6).

11
Next, we measured how increasing concentrations of ZapA affects the curvature and correlation 12 length at steady state, κSS and ρSS (Fig. 2G, Fig. S2B,C). Similar as for the bundle width (Fig. 1F), 13 we found a switch-like transition between two states with Hill coefficients of nH = 4.63 ± 0.84 and 14 nH = 4.68 ± 2.84 and with EC50 = 0.84 ± 0.05 and EC50 = 0.84 ± 0.17 for " '' and # '' , respectively. 15 Furthermore, neither ZapA R46A nor I83E had the same effect on the FtsZ filament network (Fig.  16 2H,I; Table S1). These results show that the system exists in two states: a state of high curvature 17 and low spatial order, at ZapA concentrations below 0.8 µM, and a state of high spatial order and 18 low curvature at higher concentrations. 19 ZapA increases persistence of the membrane-bound FtsZ filament network 20 Z-rings in cells lacking ZapA not only show loose arrangement, but are also highly dynamic, 21 transitioning back and forth between multiple locations in the cell 20 . In contrast, the Z-ring in 22 wildtype cells usually persists at the same position during the cell cycle. Consistent with these 23 observations, we realized that FtsZ filament bundles continuously reorganized in the absence of 24 ZapA, but appeared much more static when ZapA was present. To quantify the degree of 25 reorganization, we performed temporal autocorrelation analysis. This method measures the 26 similarity of fluorescence images as a function of a time lag Δt between different frames (Fig. 3A). 27 We found that the corresponding autocorrelation function, ϴ(Δt), rapidly decayed in the absence 28 of ZapA, consistent with a rapidly rearranging pattern (Fig. 3B, blue curve). In the presence of 29 6µM of ZapA, however, this decay was about 4-fold slower (Fig. 3B, red curve). We defined the 30 characteristic correlation time, /, for varying concentrations of ZapA by the halftime of a mono-reorganization and consequently short correlation times, but the system switched to slow 1 reorganization and long correlation times at saturating ZapA concentrations with a Hill coefficient 2 of 4.89 ± 5.13 and EC50 = 0.79 ± 0.21 µM (Fig. 3C). ZapA R46A, which does not bind to FtsZ, did 3 not affect reorganization dynamics ( Fig. 3D; p = 0.42), while ZapA I83E increased slightly, but 4 not-significantly the correlation time ( Fig. 3D; p = 0.08). This shows that tetrameric ZapA not only 5 affects the architecture of the membrane-bound filament network, but also its reorganization 6 dynamics in a highly cooperative manner. 7 Together, our in vitro experiments and quantitative image analysis show that ZapA promotes an 8 ultrasensitive switch in filament architecture and reorganization dynamics. In all instances, this 9 switch happens at ZapA concentrations around 0.75 µM, before the system saturates at ZapA 10 concentrations similar to FtsZ. Interestingly, this correlation is consistent with ZapA tetramers 11 having four FtsZ binding sites. 12 ZapA does not change FtsZ polymerization dynamics 13 So far, our analysis revealed that ZapA strongly increases the spatial order of the membrane-14 bound FtsZ filament network, and slows down reorganization dynamics. This apparent increase 15 in filament stability, however, could perturb the spatiotemporal dynamics of cell wall synthases in 16 vivo, as their motion is driven by treadmilling. Knocking out individual Zap proteins in vivo did not 17 show any significant change in the FtsZ treadmilling velocity 5 . However, due to their overlapping 18 functions, it is difficult to rule out a possible compensation by other FtsZ associated proteins in 19 these mutants. Accordingly, we wondered if FtsZ crosslinking by ZapA influences the 20 polymerization dynamics of membrane-bound filaments. To quantify the treadmilling velocity in 21 the absence and presence of ZapA, we first constructed differential time-lapse movies, where we 22 calculated the intensity differences between frames separated by a constant time delay (Fig. 4A  23 and Fig. S3; Movie S2). This step yielded new time-lapse movies of moving speckles that 24 represent either the growth or shrinkage of filament bundles in a given time (methods for details). 25 First, as ZapA was suggested to promote an antiparallel alignment of FtsZ filaments, we were 26 interested if the presence of ZapA leads to a change of filament orientation 12 . Comparing the 27 speckles between experiments with and without ZapA, we did not observe any significant 28 difference, neither in shape nor intensity (Fig. S3A). This shows that the orientation of filaments 29 in ZapA-induced bundles was similar as in bundles formed due to intrinsic lateral interactions 30 between FtsZ filaments. Furthermore, we observed that fluorescent speckles corresponding to 31 growth and shrinkage along the bundle roughly co-localized, and moved in a common direction 32 with similar velocity (Fig. 4B,C). This observation is consistent with a polar orientation of FtsZ filaments on the membrane and indicates that the mean filament length is shorter than the 1 resolution of our fluorescence microscope (Fig. S3B). 2 Importantly, as these speckles can be automatically detected by particle tracking methods, 3 differential imaging allows us to quantify the growth (0+) and shrinkage velocities (0-), of thousands 4 of treadmilling trajectories simultaneously (methods for details). Using this approach, we found 0+ 5 and 0-to be normally distributed with similar mean values consistent with treadmilling behavior 6 (0+ = 62.5 ± 4.5 nm/s, n = 4232 tracks, 10 samples; 0-= 51.84 ± 5.9 nm/s, n = 6302 tracks, 10 7 samples) (Fig. 4D,E, blue). These values were unaffected in the presence of excess ZapA (6.0 8 µM, 0+ = 58.8 ± 3.3 nm/s and 0-= 54.3 ± 3.5 nm/s) (Fig. 4D,E, red) and at all other ZapA 9 concentrations tested in our experiments (Fig. 4F,G). 10 The velocity autocorrelation along each treadmilling trajectory provides information about the local 11 directional persistence of treadmilling (  Table S1). 17 Together, these experiments and analyses show that despite its strong effect on the architecture 18 and reorganization dynamics of the FtsZ filament network, ZapA does not slow down or enhance 19 the underlying polymerization dynamics. 20 ZapA binds FtsZ only transiently 21 Next, we sought to better understand how ZapA could change the architecture of FtsZ filaments 22 without slowing down their treadmilling dynamics. To this end, we prepared fluorescently labeled 23 ZapA and imaged its behavior simultaneously with FtsZ. We found that ZapA co-localized with 24 FtsZ bundles on the membrane, however with a more discontinuous appearance ( We then analyzed the turnover rate of FtsZ and ZapA using fluorescence recovery after 27 photobleaching (FRAP) experiments ( Fig. 5B; Movie S4). For FtsZ, we found a mean recovery 28 half-time of 7.63 ± 1.30 s (n = 6) in the absence of ZapA and 6.66 ± 0.64 s (n = 7) at 6.0 µM ZapA, 29 consistent with our observation that treadmilling was unchanged ( Fig. 5C; p = 0.47). In contrast, 30 ZapA itself showed a faster turnover with a recovery half-time of only 3.01 ± 0.47 s (n = 6). This 31 difference in turnover is similar to the one found in living cells 8 . 32 To further corroborate these results, we performed single molecule experiments, where we added 1 small amounts of a Cy5-labelled protein, either ZapA or FtsZ, to a background of Alexa488-2 labelled FtsZ ( Fig. 5D; Movie S5). This not only allowed us to analyze the lifetime of FtsZ 3 monomers in the treadmilling filaments, but also the recruitment of ZapA from solution. In 4 agreement with our FRAP experiments, we found no difference in the lifetimes of FtsZ with and 5 without ZapA (FtsZ, 6.59 ± 0.28 s, n = 5; FtsZ + ZapA, n = 6.01 ± 1.24 s, n=6; p = 0.14) (Fig. 5E, 6 blue and red, respectively). Again, ZapA showed a faster turnover compared to FtsZ (0.94 ± 0.14 7 s, n = 5; Fig. 5D, purple; p = 1.44E-04; Table S1). Together with the treadmilling velocity, we can 8 now provide an estimate of the mean length of filaments, which is given as the product of lifetime 9 with the treadmilling velocity. We found an FtsZ filament length between 341.6 ± 50.0 and 412.0 10 ± 37.7 nm, when only FtsA was present, and between 302.4 ± 72.9 and 351. shown that FtsZ plays two roles: First, its polymerization into the Z-ring defines the location of 3 division in the cell. Second, FtsZ treadmilling was found to drive a circumferential motion of 4 peptidoglycan synthases, which is required for the homogeneous distribution of cell wall synthesis 5 at the division site. Given the importance of FtsZ polymerization dynamics for cytokinesis, the 6 question arises how the robustness and precision of cell division is achieved, while FtsZ filaments 7 are continuously turning over. FtsZ-associated proteins were known to contribute to the stability 8 of the Z-ring and consequently that of the cell division machinery, the underlying mechanism 9 however was not clear. 10 In this study, we show that ZapA aligns membrane-bound FtsZ filaments in a polar, parallel 11 orientation. This results in a straightening and stabilization of filament bundles, and consequently 12 an increase in the spatial order and spatiotemporal persistence of the filament network. 13 Importantly, ZapA did not change the treadmilling velocity of FtsZ filaments (see Table 1 for a 14 summary of results). Together, these observations lead us to conclude that ZapA is able to 15 increase the precision and robustness of the Z-ring, without compromising the function of the cell 16 division machinery (Fig. 6). 17 Furthermore, we found that ZapA promotes a switch-like transition between a state of low order 18 and fast reorganization dynamics to a state of high order and slow reorganization. What could be 19 the reason for the observed cooperativity? Due to its central four-helix bundle, the ZapA tetramer 20 can be assumed to have a rigid structure, which promotes the formation of straight filament 21 bundles as observed in previous studies 13 . Furthermore, we expect the ZapA tetramer to bind 22 with a higher affinity to filaments aligned in parallel, where all binding sites are in contact with 23 FtsZ. Accordingly, binding to and aligning filaments can increase the affinity of ZapA towards FtsZ 24 and therefore stimulate its own recruitment (Fig. 6). 25 Our experiments show that FtsZ, FtsA and ZapA self-organize into a highly ordered cytoskeletal 26 structure. How does this system compare to filament networks of other cytoskeletal proteins? 27 Similar to gels of actin filaments and molecular motors, the structures described here exist out of 28 equilibrium. In contrast to these active gels, however, the FtsZ filament network is driven out of 29 equilibrium due to the continuous consumption of energy during treadmilling and not by the activity 30 of motor proteins. Furthermore, the function of actin bundling proteins is to modulate the 31 architecture and mechanical properties of actin filament networks, which can also affect actin 32 polymerization dynamics 22 . In contrast, polymerization and depolymerization rates of FtsZ stay 1 constant even at saturating concentrations of ZapA. Accordingly, we believe that the cytoskeletal 2 networks of FtsZ, FtsA and ZapA represent a novel and distinct form of active biological material 3 despite the apparent absence of mechanical stresses that are usually present in gels composed 4 of filaments and molecular motors 23-25 . 5 6 Acknowledgements 7 We thank all Loose lab members for support and useful discussions, to the life sciences facility 8 (LSF) and the bioimaging facility (BIF) of IST Austria for assistance with general equipment and 9 TIRF microscopy, respectively, and to Georgia Squyres Phospholipids used in this paper, DOPC (1,2-dioleoyl-sn-glycero-3-phosphocholine) and DOPG 3 (1,2-dioleoyl-snglycero-3-phospho-(1'-racglycerol)), were purchased from Avanti Polar Lipids 4 (Alabaster, AL) and kept at -20°C as 25mg/ml stock solutions in chloroform; Sulfo-Cyanine-5-5 maleimide (Cy5) was acquired from Lumiprobe and Alexa Fluor® 488 C5-Maleimide (Alexa488) 6 was acquired from ThermoFisher Scientific; Nucleotides were acquired from ThermoFisher 7 Scientific or Jena Bioscience; Precision cover glasses for the homebuilt chambers were obtained 8 from VWR (thickness No. 1.5H, 24 x 50); E.coli strains were obtained from Lucigen; Strep-Tactin 9 resin was acquired from Iba lifesciences and Nickel resins were purchased from ThermoFisher 10 Scientific (HisPur TM Ni-NTA resin) or Macherey-Nagel (Protino Ni-TED resin); All the remaining 11 reactants and salts were obtained from Sigma, Merck, or Invitrogen and were of analytic or 12 spectroscopic grade. 13

Protein Biochemistry 14
Purification and labelling of FtsZ 15 FtsZ (UP P0A9A6) was cloned into a pTB146-derived vector which attached a N-terminal His6-16 SUMO tag plus seven additional amino acids (AEGCGEL) that provide a cysteine residue for 17 further fluorescent labelling (pML45, His6-SUMO-GCG-FtsZ). E.coli C41 (DE3) cells were 18 transformed with pML45 and grown in TB medium supplemented with ampicillin at 37°C, until 19 cells reached an OD600 of 0.8. After the addition of IPTG to a final concentration of 1mM, cells 20 grew for 5h at 37°C. Cells were harvested by centrifugation, pellets were frozen in liquid nitrogen 21 and kept at -80°C until further use. 22 For purification, pellets were thawed and resuspended in FtsZ buffer (50 mM Tris-HCl pH 8.0, 23 500mM KCl, 2mM β-mercaptoethanol, 10% glycerol) supplemented with 20mM imidazole and 24 cOmplete EDTA-free protease inhibitors cocktail (1tablet/50ml, Roche)) followed by incubation at 25 4°C for 15 min. Cells were lysed using a cell disruptor (Constant Systems) at a pressure 1.36kbar 26 and incubated with 1mg/ml DNase I (Sigma-Aldrich) and 2.5mM MgCl2 for 15min. The lysate was 27 then centrifuged (30 min, 60,000xg, 4°C) and the supernatant was incubated with nickel agarose 28 resin (HisPur Ni-NTA resin, Thermo Scientific) for 60 min at 4°C. The resin was extensively 29 washed with FtsZ buffer supplemented with 20mM imidazole and 30mM imidazole. The fusion 30 protein was eluted with FtsZ buffer supplemented with 250 mM imidazole. Fractions were during an overnight dialysis into FtsZ cleavage buffer (50mM Tris-HCl pH 8.0, 300mM KCl and 3 10% glycerol). The digested sample was passed several times through Ni-NTA resin, to remove 4 His6 containing molecules. The flow through was collected and active protein was enriched by 5 sedimentation of FtsZ filaments. For this, the protein was dialyzed into FtsZ polymerization buffer 6 (50 mM PIPES pH 6.7, 10 mM MgCl2) and incubated with CaCl2 and GTP for 15 min at 30°C. The 7 solution was then centrifuged (2 min, 20,000xg, RT) and a clear gel-like pellet containing 8 polymeric FtsZ was obtained. The pellet was resuspended into FtsZ storage buffer (50mM Tris-9 HCl pH 7.4, 50mM KCl, 1mM EDTA and 10% glycerol) and incubated with 100×molar excess of 10 7 Tris(2-carboxyethyl)phosphine hydrochloride (TCEP) for 20 min at RT for cysteine-labelling. 11 Five times molar excess of a thiol-reactive dye (Alexa488 or Cy5-maleimide) was added to the 12 solution and incubated overnight at 4°C during dialysis into FtsZ storage buffer. Finally, labelled 13 FtsZ was loaded on a PD10 desalting column to remove CaCl2, GTP and free dye. To form a supported lipid bilayer, a SUV suspension was diluted to 1mM in SLB buffer (50mM 25 Tris-HCl pH 7.5, 300mM KCl) and CaCl2 was added to a final concentration of 2mM. From this 26 mix, 50 µl were added to each self-made reaction chamber. Vesicles adsorb to the surface, 27 rupture and fuse with the clean hydrophilic glass to form a flat bilayer, which is further facilitated 28 by the presence of CaCl2. After 1h of incubation at room temperature, 50 µl of SLB buffer was 29 added to each chamber (for a final volume of 100µl) and sample was rinsed several times with For the self-organization assay, all experiments were performed by first adding an oxygen 1 scavenger system to the reaction chamber (0.2% d-Glucose, 0.016 mg/ml glucose oxidase, 2 0.002mg/ml catalase, 1mM dithiothreitol and 0.25mg/mL trolox) to prevent photobleaching. Then, 3 FtsZ (with 30% Cy5-or Alexa488-labeled protein) and FtsA were added to the reaction chamber 4 in a final concentration of 1.5 µM and 0.5 µM, respectively. To trigger polymerization and 5 membrane recruitment of FtsZ, ATP and GTP were added to the system to a final concentration 6 of 4 mM. ZapA was added in different concentration ranges before the addition of GTP. 7 Total internal reflection fluorescence microscopy (TIRFM) 8 All experiments were performed on an Inverted multipoint total internal reflection fluorescence 9 (TIRF) microscope (TILL Photonics) equipped with dual camera TIRF objectives (Andor 10 897straight (X-8449) 512x512 pixel and Andor 897 (X-8533) 512x512 pixel) and an image splitter 11 (Andor Tucam) equipped with a long pass of 580 and 640nm. Alexa488 and Cy5 dyes were 12 excited using 488nm and 642nm laser lines, respectively, and the emitted light was filtered by an 13 Andromeda quad-band bandpass filter. Images were typically obtained every 2s, with 50ms 14 exposure to minimize photobleaching and using a 100x Olympus TIRF (NA = 1.49 DIC) objective. 15

Image analysis and processing 16
For image processing, image stacks were imported using FIJI software 27 . For image analysis and 17 for visualization (supplemental movies) acquired time-lapse movies were usually first normalized 18 to create a constant overall intensity and compensate for the increasing intensity over time due 19 to protein binding to the membrane. 20

Quantification of Bundle Width 21
To estimate the width of membrane-bound FtsZ bundles, fluorescence images were binarized 22 using the adaptive threshold plugin for ImageJ. This plugin corrects for non-homogeneous 23 background intensities, overcoming the limitation of conventional threshold methods. The 24 background was removed using a local threshold corresponding to the mean intensity of an area 25 of 20x20 pixels without any further subtraction. The binarized time lapse movie was then 26 processed with the despeckle filter in ImageJ to remove small particles. Next, we calculated the 27 Euclidean Distance Map (EDM) of every frame using the Distance Mapping function of ImageJ. 28 This transformation results in a grey scale image, where the grey value of each pixel represents 29 the shortest distance to the nearest pixel in the background. Accordingly, bundle widths 30 correspond to the local peak intensities multiplied by 2.
The mean bundle width for every frame of the movie was calculated by identifying the peak 1 intensities for each line and column of the image using a MATLAB script. This value could then 2 be plotted as a function of time. The characteristic bundle width for different concentrations of 3 ZapA and respective mutants was obtained by taking the time-average at steady state. 4

Architecture analysis 5
For a quantitative description of the filament architecture, we first calculated an orientation field 6 of the pattern by calculating a gradient squared tensor at every position of a fluorescence 7 micrograph 28,29 . This analysis was performed either using the OrientationJ plugin for ImageJ or 8 using a custom python code based on the scikit-image package ( Fig. 2A). For this method, a unit 9 vector 2 3 = (5678, 7:;8) is assigned to all pixels in the image and their directional derivative is 10 Where the first eigenvector of the so-called structure tensor matrix, P X = 〈∇?(@, A), ∇?(@, A) B 〉, 26 defines the local dominant orientation. The final output is a two-dimensional orientation field, 1 is described as the rate of change in the local orientation in the direction perpendicular to that 2 orientation: 3 Where 5 corresponds to the axis perpendicular to the orientation. From this, we generated a color 5 map containing Z b@ c A d e, which represents the curvature in Y at position x, y (Fig. 2B). Different 6 colors (red and blue) indicate which way the curves bend (left or right). This depends on the way 7 we define the parameters, i.e. which way we walk along the curve. Accordingly, as the direction 8 of curvature is not a relevant parameter in our analysis, we plotted the distribution of absolute 9 values for every experiment (mirrored negative values). The mean curvature for each image 〈Z〉 10 was given by the mean half time of the mono-exponential decay fitted to these distributions (Fig.  11   2C). 12 To obtain a quantitative description of the spatial order of the filament network, we applied a To quantify the reorganization dynamics of the membrane-bound FtsZ filament network, we 2 performed a temporal correlation analysis using the ImageCorrelationJ plugin for ImageJ 30 . This 3 method calculates the Pearson cross-correlation coefficient (PCC) between two frames with an 4 increasing time lag between them (∆t) 31 . The mean intensities in local areas of 3x3 pixels were 5 used to calculate the PCC between two different images, which was then plotted as a function of 6 ∆t (Fig.3B). For highly dynamic image pattern, the PCC typically decays rapidly with ∆t, but slowly 7 for persistent structures. The corresponding correlation curve was fitted to a mono-exponential 8 decay, and the half-time (correlation time, /) was used to compare a set of individual experiments. 9 Treadmilling Velocity Analysis 10 To quantify treadmilling dynamics, we have developed an automated image analysis based on 11 differential image stack obtain from the ImageJ ImageCalculator command. First, two frames 12 separated by 10s were subtracted from one another to generate a new time-lapse movie showing 13 directionally moving fluorescent speckles. These speckles correspond to growth or shrinkage of 14 filaments at a given position, (Fig. 4A and S3A). Next, the TrackMate 32 , a particle tracking toolbox 15 available for ImageJ, was used to identify and track fluorescent speckles or filament treadmilling 16 and to obtain detailed information regarding their displacement and velocity. Moving speckles 17 were detected using the LoG (Laplacian Gaussian) detector with an estimated diameter of 1 µm. 18 We used TrackMate's quality criterion to select only the best 5% tracked particles and from these, 19 we discarded particles with a signal-to-noise ration lower than 0.7. To build the final trajectories 20 we used the "Simple LAP tracker" with a "Max Linking Distance" of 0.5µm, a "Maximal gap-closing 21 distance" of 1µm and "Max frame Gap" of 2 frames, and we only considered for analysis 22 trajectories longer than 6sec. 23 Reconstructed trajectories were further analyzed using a Matlab script based on @msdanalyzer 33 24 toolbox to obtain information regarding the velocity and directionality of the moving speckles. The 25 mean velocities obtained from TrackMate followed a normal distribution and were fitted to a 26 Gaussian function to extract the mean velocity at each condition (Fig.4). In addition, we calculated 27 the autocorrelation of the velocity vectors, i.e, angles of the normalized displacement vectors were 28 compared pairwise as a function of an increasing time interval (∆t) and the correlation coefficient 29 (Vcorr) was given by the cosine of the angle difference. This kind of analysis provided information 30 about directionality since random motion particles tend to show velocity vectors completely uncorrelated (Vcorr = 0 for all ∆t) while particles with a directed migration display highly correlated 1 velocity vectors (Vcorr > for all ∆t). 2

Single Molecule Tracking 3
Single molecule experiments were performed as described in ref. 34 . Briefly, individual FtsZ 4 monomers were resolved at single molecule level by mixing small amounts of Cy5-labelled FtsZ 5 (less than 2nM) to a background of 1.5 µM Alexa488-labelled FtsZ. To track individual ZapA 6 molecules, we followed a similar strategy by adding small amounts of ZapA-Cy5 to a background 7 of unlabeled ZapA, always in the presence of Alexa488-FtsZ 1.5µM. Experiments were typically 8 recorded using minimal laser power to visualize single molecules, with an exposure time of 50 ms 9 and a varying acquisition time (114ms to 1s) for further photobleaching correction 34 . Single 10 molecules were identified and tracked using the automated particle-tracking platform TrackMate 11 (https://imagej.net/TrackMate), with the following parameters: a particle size of 0.5 µm, particles 12 localized for at least two frames were considered, linking max distance and gap closing distance 13 of 0.5 µm, and 2 frames for the gap closing max. The mean lifetime of single molecules was 14 obtained by fitting a mono-exponential decay to the lifetime distribution. 15 Fluorescence recover after photo bleaching (FRAP) 16 For FRAP experiments, we allowed the system to reach the steady state and a small area of the 17 membrane was bleached using a high laser intensity. To obtain the half-time of the recovery, a