Introduction

Accurate selection of the division site is essential in bacteria for maintaining cell size, ensuring proper chromosome segregation, and preventing the formation of anucleate minicells. Cell division is governed by a complex protein machine known as the divisome (3-5). The spatiotemporal organization of this protein complex is directed by the dynamic arrangement of a bacterial tubulin homolog, FtsZ (6, 7). FtsZ forms treadmilling protofilaments that condense into a ring-like structure at the division site (8-10), recruiting several cell wall synthesis proteins, such as division-specific transglycosylases and transpeptidases (11, 12). Together, these proteins synthesize the bacterial peptidoglycan cell wall. To ensure that FtsZ polymerizes only at the correct location, many bacterial species employ the Min system, a conserved pathway that prevents aberrant polar septation events.

The Min system was identified through the formation of small, round, anucleate cells that divide close to the cell poles (27). The gene locus responsible for the “miniature” cell phenotype in E. coli was termed minB (28, 29), later shown to encode the three structural genes minC, minD and minE (30). MinC is the actual inhibitor of FtsZ-polymer formation (31, 32). The MinC protein contains an N- and C-terminal domain (33), both of which contribute to FtsZ inhibition. The C-terminal domain binds to the C-terminal tail of FtsZ, whereas the N-terminal domain, which also mediates MinC dimerization, binds at the interface of two FtsZ subunits (34-36). MinC-mediated inhibition of FtsZ has also been demonstrated in B. subtilis (37-39), suggesting a conserved mechanism across species.

The spatial localization of MinC is regulated by MinD, a Walker-type ATPase that belongs to the ParA/MinD family of proteins (37, 40, 41). These partitioning ATPases are involved in positioning cargo within bacterial cells (42, 43). While the molecular details of partitioning differ among ParA-like proteins, a common feature is their use of binding to a large intracellular surface as a spatial template (42, 43). ParA proteins typically bind non-specifically to DNA, covering the nucleoid and exploiting it as a track for cargo positioning (44). In contrast, MinD proteins bind to the membrane via an amphipathic helix (45, 46). In E. coli, membrane binding of MinD strictly depends on ATP binding and subsequent dimerization (40). Dimerized MinD recruits MinC and thereby activates MinC to inhibit FtsZ oligomerization. In E. coli, MinE further regulates MinD localization by binding to the membrane and interacting with MinD, stimulating ATP hydrolysis and triggering MinD release from the membrane. Slow nucleotide exchange in cytosolic MinD leads to an oscillatory cycle of membrane attachment and detachment, resulting in pole-to-pole oscillation of MinCD (48-50). This oscillatory behavior generates a temporal minimum of MinC at midcell, such that Z-ring assembly is restricted exclusively to this position. The oscillation can be reconstituted in vitro on planar membranes, demonstrating the robustness of this self-organizing system and a prime example of pattern formation. (51).

The Gram-positive model bacterium B. subtilis likewise encodes the minC and minD genes (52, 53), and loss of either results in the characteristic minicell phenotype. Thus, a role for the Min system analogous to that in E. coli has been proposed (37, 54). However, B. subtilis lacks a MinE homolog, and the Min proteins do not oscillate from pole to pole (37, 54). Instead, static imaging shows that MinD accumulates at the cell poles and at active division sites. Initial genetic and cell biological studies demonstrated that the polar scaffold protein DivIVA is essential for polar MinD localization (37). The interaction between DivIVA and MinD is mediated by the integral membrane protein MinJ (55, 56). MinJ is also part of the divisome and interacts with multiple division proteins (56). Loss of MinJ causes defects in divisome disassembly, as proteins such as FtsL and PBP2b remain associated with the nascent cell pole after division in ΔminJ cells (57). This function is supported by recent findings that MinD promotes FtsZ polymer disassembly at the cell poles, and that deletion of minD inhibits polar peptidoglycan remodeling (58).

We have recently shown that all components of the B. subtilis Min system, including DivIVA and MinJ, dynamically localize within the bacterial cell (57, 59-61). Interestingly, even the scaffold protein DivIVA, although highly conserved in many Gram-positive phyla, is not statically anchored at the poles but instead redistributes along the membrane from poles to septa (59-61). In contrast, DivIVA homologs in actinobacteria show greatly reduced mobility, consistent with their role in apical growth (60).

Despite these advances, the mechanism by which MinD activity and localization are regulated in the absence of MinE has remained unclear, raising the possibility of additional, unidentified regulatory factors or alternative modes of ATPase control. In this study, we address these long-standing questions by combining in vitro biochemical analyses with single-molecule imaging in vivo. We define the ATPase and membrane-binding properties of B. subtilis MinD and examine how its dynamics are shaped by interactions with MinJ and MinC. Our findings reveal an unexpected mechanism of MinD regulation that operates independently of MinE-like factors, providing a revised understanding of how the non-oscillatory B. subtilis Min system robustly controls cytokinesis and prevents aberrant septation events.

Results

Purification and gel filtration of His-MinD

Since localization and function of MinD appear to be strongly tied to its ATPase activity, the first logical step in describing and understanding details of the ATPase cycle was identification of the relevant biochemical characteristics, to eventually be able to correlate these properties to in vivo observations. To this end, B. subtilis minD was heterologously expressed in E. coli, yielding an N-terminally His-tagged fusion protein (Fig. S 1), as C-terminal MinD fusions have been shown to interfere with membrane binding (62). Since MinD is known for being membrane-associated, a well-established E. coli MinD purification protocol was adapted (63), where buffers were supplied with excess Mg²⁺-ADP to potentially reduce MinD self- and membrane interaction. Based on the respective western blot and gel filtration chromatogram (Fig. S 1 and Fig. S 2), His-MinD is not able to dimerise when purified in presence of excess Mg²⁺-ADP, as evidenced by a main peak at the expected sizes of the monomeric form (∼31 kDa, 16.11 ml) and the absence of a peak at the expected size of the dimeric form (∼62 kDa, 14.65 ml, Fig. S 2). However, Western blotting of Ni-NTA and size-exclusion chromatography (SEC) fractions indicates the presence of multiple His-tagged degradation products alongside full-length His-MinD (Fig. S 1). Because these fragments retain the His-tag, they co-purify during Ni-NTA and can co-elute or form non-covalent assemblies with full-length MinD, increasing the heterogeneity of the sample and likely producing the minor shoulder observed at an apparent molecular weight of ∼44 kDa (15.4 ml, Fig. S 2). Unfortunately, the Mg²⁺-ADP excess in the purification buffers did not seem to stop His-MinD membrane interaction, as a significant fraction of the recombinant protein were later detected to be lost in the cell pellet. However, sufficiently high His-MinD concentrations could be obtained to proceed with biochemical analysis.

B. subtilis MinD displays full ATPase activity when incubated with ATP and liposomes

Being able to purify His-MinD allowed us to investigate the biochemical conditions affecting the MinD ATPase cycle next. For this purpose, we choose to use the commercially available enzyme coupled EnzChek™ phosphate assay kit (Thermo Fisher Scientific) to measure ATPase activity. The activity was determined by measuring the P_i production rate (in µM min⁻¹) as a function of His-MinD concentration. We first tested His-MinD ATPase activity at protein concentrations between 0.5 µM and 16 µM. Under these conditions, only baseline ATPase activity below the detection limit was observed (consistent with Fig. 1 D, 0 mg ml^-1 liposomes condition). Since MinD interacts with the membrane and it was expected that the ATPase cycle is involved in controlling this interaction, we next measured ATPase activity in the presence of liposomes extruded from E. coli total lipids (Avanti). Remarkably, this led to a robust increase in ATPase activity of His-MinD, appearing to be linearly dependent on protein concentration (Fig. 1 A & B). To be able to compare the activity level to the published E. coli MinD in vitro ATPase data (64-66), the specific activity is shown as a specific activity in nmol mg^-1 min^-1 (Fig. 1 B). Interestingly, activity levels of recombinant B. subtilis MinD were found to be on par with fully active E. coli MinD in conjunction with MinE and phospholipids (64). Moreover, the ATPase activity did not show cooperative behavior compared to E. coli MinD at lower concentrations (64). We further measured that ATP hydrolysis can be titrated with both ATP and liposomes and is described by Michaelis-Menten kinetics (Fig. 1 C & D). Again, the hereby-obtained V_max values (19.50 nmol mg^-1 min^-1 and 22.05 nmol mg^-1 min^-1, respectively) indicate similar ATPase activity levels between MinD from E. coli and B. subtilis.

Fig. 1.

MinC and the PDZ domain of MinJ have different effects on MinD ATPase activity

Even though ATPase activity levels were similar, proteins that directly interact with MinD could potentially modulate this activity in vivo as seen for MinD:MinE in E. coli (64-66). To get the complete picture of MinD activity, investigating the potential effects of both known B. subtilis MinD interactors MinC and MinJ was therefore necessary. For this purpose, both proteins were subjected to purification attempts. While purification of the cytosolic His-MinC could be successfully established using affinity chromatography followed by SEC (Fig. S 3), the intermembrane protein His-MinJ proved more challenging. Since several different purification attempts of the full-length protein yielded no stable product, a truncated form containing only the soluble C-terminal PDZ domain of MinJ fused to an N-terminal His-tag (His-PDZ) via a flexible linker was expressed instead (Fig. S 4). PDZ domains are often found in membrane-associated proteins and known to selectively interact with the C-terminus of partner peptides or proteins to sequester them to highly specific sites of the cell, both in pro- and eukaryotes (67). Although His-PDZ could be isolated via Ni-NTA affinity chromatography, it was unstable during SEC and consistently lost from the column, precluding further purification by this method. Hence, further experiments were performed using His-PDZ directly after affinity chromatography.

To test the effect of His-MinC and His-PDZ on His-MinD ATPase activity, the previously established enzymatic assays were repeated in the presence of either His-MinC or His-PDZ (Fig. S 5). Addition of His-MinC resulted in a very small but statistically significant decrease in MinD ATPase activity (∼5%). Given the small sample size this effect should be interpreted cautiously. However, it could be caused by the experimental set-up, since the high amounts of MinC (6 µM) that have been used might sequester a small fraction of MinD away from membrane surfaces that are required for efficient ATP hydrolysis, due to direct interaction of MinC and MinD.

In contrast, addition of His-PDZ led to a slight but reproducible increase in MinD ATPase activity, which prompted us to repeat the experiment at different His-PDZ concentrations (Fig. S 5). All His-PDZ concentrations significantly increased His-MinD ATPase activity compared to the control (1 µM His-PDZ = 25% ± 6.8%, 3 µM His-PDZ = 37 ± 12.5 %; 5 µM His- PDZ = 22% ± 2.8 %). We speculate that the PDZ domain may promote MinD:MinD encounters, facilitating dimerization rather than directly enhancing catalytic turnover. Because only the soluble PDZ domain could be purified, it remains unclear how full-length, membrane-anchored MinJ would influence MinD activity in vivo. The magnitude of the effect might differ when the PDZ domain is present in its native membrane-associated context.

MinD ATPase activity is triggered by membrane binding

To determine whether interaction with the membrane is sufficient to stimulate MinD ATPase activity, a MinD mutant that is defective in membrane binding was required. MinD interacts with the membrane through an amphipathic α-helix located at the N-terminus, thus being the membrane targeting sequence (MTS) (Fig. 2 A). Exchanging the hydrophobic isoleucine residue in the apolar core of the MTS with a charged glutamate (I260E) has been shown to abolish membrane binding of MinD by perturbing the amphipathic properties, demonstrated through fluorescence microscopy (68). Accordingly, a His-MinD-I260E variant was constructed and purified via the established purification protocol. The yield of the purification was extensively higher compared to the wild type protein (Fig. S 6), hinting towards a successful abolishment of membrane interaction.

Fig. 2.

We next repeated ATPase activity assays with the purified, recombinant I260E mutant, and remarkably, activity remained below the detection limit of the assay, while a positive control of His-MinD wild type protein displayed regular activity (Fig. S 7). In summary, these data collectively suggest that membrane binding is necessary and the key trigger for ATP hydrolysis of B. subtilis MinD, without the apparent need for another stimulus.

Next, we wanted to compare in vivo and in vitro data. Therefore, a B. subtilis strain for microscopy was constructed, where the native copy of MinD was replaced by a HaloTag-MinD fusion protein (Fig. 2 B, HaloTag will henceforth be referred to as Halo). We then introduced the I260E mutation, resulting in a strain producing Halo-MinD-I260E as the only copy. Microscopic analysis of a Halo-MinD-I260E with TMR dye confirmed drastically reduced membrane binding through its almost exclusive cytosolic localization (Fig. 2 B). Furthermore, cell length and the number of minicells increased drastically, comparable to a minD knockout (Tab. S 1).

Construction of catalytic MinD mutants

To determine the role of ATP-mediated dimerization in membrane binding, we explored further mutations that affect the ATP binding region and thus the dimerization process. Due to the conserved nature of Walker-type MinD/ParA-like ATPases, residues involved in catalytic activity have been well characterized (69-71). We therefore constructed expression plasmids encoding three different MinD variants (illustrated in Fig. 2 A): This first mutation, MinD-G12V, produces a steric clash within the active site, so that ATP can still be bound, but dimers cannot form (69, 70). The second mutation, K16A, completely disrupts ATP binding and hence results in monomeric protein (69). The third variant encodes the mutation D40A, which should interfere with ATP hydrolysis (70, 71). Therefore, it was expected that MinD-D40A forms a trapped nucleotide sandwich dimer similar to Soj or E. coli MinD (70, 71). All MinD variants were successfully expressed and purified and resulted in similar protein yields when compared to the wild type protein (Fig. S 1, Fig. S 8, Fig. S 9 and Fig. S 10). When evaluating the gel filtration chromatograms, His-MinD-G12V elute in a single peak, which can be interpreted as a monomer when compared to the wild type His-MinD elution peak (Fig. S 2). His-MinD-D40A on the other hand also eluted in a single peak, albeit being wider and eluting earlier, indicating a mainly dimeric conformation. It eluted between the expected dimer and monomer positions during SEC as a wide peak (Fig. S 2). As His-MinD-D40A cannot hydrolyze ATP, it should be dimeric and likely also form small oligomers or transient multimers. A mixture of monomer/dimer/oligomer likely produced the broad, left-shifted peak whose center is between monomer and dimer. Only His-MinD-K16A reproducibly formed a rather unexpected shoulder in its main elution peak, which could hint towards further interaction with the gel filtration matrix or other potential self-interactions. All of the recombinant His-MinD mutants displayed negligible ATPase activity when tested using the EnzChek phosphate assay kit (Fig. S 7 C). This was expected, as all of these MinD variants (G12V, K16A and D40A) were mutated in residues which significantly alter the ATP binding site. In all cases, activity levels remained consistently near the detection limit, precluding reliable quantitative interpretation.

To investigate in vivo effects of these variants, the same mutations were introduced into the strain expressing Halo-MinD in B. subtilis, respectively, and subsequently imaged via wide-field microscopy (Fig. 2 B). All strains producing MinD mutants displayed an increased cell length and minicell production comparable to ΔminD (Tab. S 1), as it was expected in cells containing dysfunctional Min proteins. Surprisingly, both monomeric MinD variants (G12V and K16A) were still observed to be membrane associated, however resulting in the apparent loss of the polar-septal MinD localization (Fig. 2 B). Finally, the locked dimer D40A mutant appeared to be mainly membrane associated, thereby forming large foci indicating accumulation of MinD, but much less dispersed when compared to G12V or K16A (Fig. 2 B). In summary, the time-resolved MinD gradient appears to be lost in all MinD mutants, including D40A. This suggests that the ATPase cycle and thus quick cycling of MinD is indeed necessary for formation of the observed dynamic gradient, as it has been previously suggested (61).

Monomeric MinD binds the membrane with high efficiency

Membrane binding appeared to be a key factor of the ATPase cycle and since the microscopy-data indicated binding of MinD monomers to the membrane, we next aimed to determine biochemically, if the respective MinD variants are capable of binding to membrane and if so, to characterize their respective binding kinetics. To this end, we employed Bio-layer interferometry (BLI), a technique capable of quantifying binding strength and protein interaction, as well as identifying reaction kinetics. Liposomes were immobilized on a biosensor to create a membrane-like surface for subsequent MinD binding assays. After successful saturation of the sensor with liposomes, MinD-liposome association and dissociation was measured, a scheme of the experimental set-up is depicted in Fig. 3 A. As a negative control for unspecific binding, MinD-sensor interaction was tested on empty SA sensors. Furthermore, unspecific binding to liposome-saturated sensors was tested using bovine serum albumin (BSA). Both resulted in negligible, diffuse signal (data not shown).

Fig. 3.

Kinetics of His-MinD binding was tested first, using different protein concentrations between 1 µM and 16 µM in presence of ATP. Binding to the sensor-bound liposomes scaled with increasing protein concentration, resulting in an equilibrium dissociation constant K_D = 7.17 µM and an association rate constant k_on = 3.03 × 10³ M^-1 s^-1 as a reference point for the wild type protein (Fig. 3 B, Tab. 1). Unsurprisingly, His-MinD-I260E showed a more than 10-fold reduced affinity to the membrane with a K_D = 74.36 µM (Tab. 1), which is also apparent in the binding graph (Fig. 3 B). The strong increase in K_D is based on the over 16-fold lowered association k_on = 0.18 × 10³ M^-1 s^-1, while the dissociation k_off = 13.45 × 10^-3 s^-1 was also slightly lower when compared to wild type MinD. Next, both monomeric mutants G12V and K16A were probed, and to our surprise, capable of membrane interaction (Fig. 3 C & D). This is in stark contrast to E. coli MinD, where dimerization needs to precede membrane binding to achieve sufficient affinity (45, 46, 65, 71, 72). Similar to I260E, the G12V mutant displayed an around 10-fold reduced membrane affinity equilibrium with K_D = 75.38 µM (Tab. 1). However, G12V only exhibited a ∼5 fold reduced binding affinity with k_on = 0.63 × 10³ M^-1 s^-1 when compared to the wild type MinD, while the k_off = 47.8 × 10^-3 s^-1 represents the fastest dissociation rate constant between all MinD variants and more than 2-fold increase compared to wild type MinD (Tab. 1). This indicates a higher turnover and thus a quicker release of the protein from the membrane. Markedly different from this, K16A displayed a K_D of 1.36 µM, but not much lower binding rates (k_on = 2.03 × 10³ M^-1 s^-1, Tab. 1 & Fig. 3 C). Instead, the low K_D can be mostly attributed to the reduced dissociation (k_off = 2.77 × 10^-3 s^-1), implying a much longer retention of His-MinD-K16A at the membrane (Fig. 3 D). Since G12V can bind ATP while K16A cannot, these findings reveal that presence of ATP in the binding pocket affects association and dissociation of MinD. Last, the His-MinD-D40A mutant protein was probed, resulting in an almost 10-fold reduced dissociation equilibrium K_D = 0.74 µM (Tab. 1, Fig. 3 E). Here, the strongly reduced membrane dissociation (k_off = 1.76 × 10^-3 s^-1) causes the drastic change in K_D, which we expected for the trapped dimer, as it could position the two amphipathic helices into a more optimal orientation of the apolar residues towards the membrane, resulting in stronger binding and no capability of hydrolyzing ATP and thus relieving the dimer (68, 70, 71).

Tab. 1.

In summary, these BLI data demonstrate that B. subtilis MinD is capable of binding membrane in both monomeric and dimeric form, while the presence of ATP in the binding pocket appears to affect both membrane binding and dissociation.

MinC and the PDZ-domain of MinJ have no significant impact on MinD membrane kinetics

In vivo, the integral membrane protein MinJ recruits MinD to poles and septum, which in turn recruits MinC to form a functional complex that inhibits FtsZ-ring assembly. However, the mechanistic details of how MinJ influences membrane association of MinD remain unclear. Does MinJ actively recruit MinD from the cytoplasm, or does it primarily sequester MinD already bound to the membrane? Furthermore, does MinJ promote MinD dimerization or oligomerization, potentially enhancing ATPase activity and triggering membrane dissociation? Similarly, MinC binding to MinD may influence the stability or turnover of MinD at the membrane, yet its impact on association/dissociation kinetics is unknown. To address these questions directly, we employed BLI to test the effect of His-MinC and His-PDZ on His-MinD membrane kinetics.

After validation of the experimental setup (see Material and Methods), we next tested how His-MinC and His-PDZ affect MinD-liposome binding kinetics by subjecting a liposome coated sensors to solutions containing 8 µM His-MinD and varying concentrations [1, 2, 4, 8 and 16 µM] of His-MinC or His-PDZ, respectively (Fig. S 11, Tab. 2). MinD controls appear twice in the table, because the experiments were performed using two independent purification batches. The corresponding His-MinC and His-PDZ samples were each paired with the His-MinD batch they contained. Due to batch-specific differences in His-MinD stability and degradation, we did not directly compare measurements across batches. Interestingly, neither His-MinC nor the His-PDZ domain significantly altered the membrane binding kinetics of His-MinD in BLI experiments. The extracted K_D, k_on, and k_off values differed only marginally between conditions (Tab. 2), and these apparent shifts fell within the fitting uncertainty and typical inter-run variability of BLI. This contrasts with the slight increase in His-MinD ATPase activity observed in the presence of His-PDZ (Fig. S 5), which had led to the hypothesis that PDZ might promote MinD oligomerization at the membrane, enhancing ATP hydrolysis. However, the discrepancy between the two systems and the high lipid concentration (21mg/ml) in the ATPase assay versus a liposome-coated sensor in BLI may account for the different outcomes. In the BLI experiment, limited surface area and distinct spatial organization of liposomes may restrict the ability of the soluble PDZ domain to effectively cluster MinD, unlike full-length MinJ, which is membrane-anchored via its transmembrane helices. Thus, while the PDZ domain alone appears insufficient to modulate membrane binding kinetics of MinD, it remains plausible that the full-length MinJ could exert a more pronounced effect through spatial coordination at the membrane.

Tab. 2.

Single-molecule localization microscopy confirms in vitro data

To validate the biochemical in vitro results, we decided to utilize single-molecule localization microscopy (SMLM) with focus on single-molecule tracking (SMT), a technique allowing to observe and evaluate MinD dynamics in vivo inside B. subtilis cells. We began SMLM analysis by plotting a heat-map of the intracellular localization of all individually recorded Halo-MinD molecules as well as the mutated variants, respectively, on normalized cells (Fig. 4 A). This representation did not only provide a robust internal control for the imaging and post- processing process for the well characterized localization of MinD, but also revealed systemic differences in localization caused by the respective mutations. The first obvious difference is reduced MinD enrichment at the cell center and poles in the G12V or K16A variants. (Fig. 4 A). Here, the membrane-bound MinD fraction seems to be much more evenly distributed along the membrane, similar to what we observed during wide-field microscopy (Fig. 2 B). This result does not only confirm the ability of MinD monomers to bind the membrane (Fig. 3 C & D), but also fuels the speculation that MinJ might be unable to recruit MinD in its monomeric form, so that polar and septal recruitment by MinJ requires dimeric MinD. In combination with the information obtained from BLI results (Fig. 3), this could mean that the membrane itself serves as proxy for MinD dimerisation and subsequent recruitment, quickly followed by ATP hydrolysis and membrane detachment of MinD. In contrast, but in agreement with wide-field imaging (Fig. 2 B), the Halo-MinD-D40A mutant does appear to frequently form foci and larger local accumulations, with little protein in the cytosolic fraction (Fig. 4 A). This, again, could be caused by local MinJ interaction and recruitment in combination with the missing capability of membrane detachment or with increased lateral MinD interactions (73). Finally, the I260E mutant appears evenly distributed throughout the cell (Fig. 4 A), as it was expected.

Fig. 4.

Independently, all protein trajectories were subjected to a stationary localization analysis (SLA) (Fig. 4 B). This was done by first grouping all trajectories into an either mobile or static population by testing if the molecule left a circle with a diameter of 97 nm (equaling the pixel size of the imaging setup) within at least five consecutive frames. Trajectories that were classified as mobile were further sub-grouped into mixed and free trajectories, where only free trajectories never presented a confinement event (74). In agreement with previous data, the D40A mutant displayed the highest fraction of static tracks (Fig. 4 B), and by far the lowest proportion of truly free trajectories (48.5 %). In contrast, more than 90% of Halo-MinD-I260E molecules were classified as free with only few trajectories classifying as static (Fig. 4 B), accentuating the absence of interaction with the membrane or other structures. When comparing Halo-MinD to the monomeric mutants G12V and K16A, the number of static tracks was visibly reduced in the mutants. Since both monomers are unable to dimerise, it is less likely that they can form larger clusters, which we observed in a previous study (61). Between both monomeric mutants, K16A displayed a slightly larger number of static trajectories (Fig. 4 B), which is in alignment with BLI data, where the K16A mutant dissociates much less effective from the membrane (Fig. 2 C & D, Tab. 1).

Next, we examined the average intracellular mobility of Halo-MinD compared with the mutant variants by analyzing the mean-squared displacement (MSD, Fig. 4 C). In MSD analysis, the traveled distances of recorded trajectories are plotted against a given time lag (Δt) as a function of Δt, here one imaging frame (24 ms), where the slope can indicate the speed (75-77). As expected, the I260E mutant displayed the fastest average mobility, since it does not bind the membrane effectively anymore (Fig. 4 C). In contrast, the D40A mutant was the slowest recorded protein, indicated by the only mild incline (Fig. 4 C). Since we expect most of the proteins’ population to be membrane bound, and furthermore observed larger foci that suggest cluster formation, this was expected. Halo-MinD wild type protein appeared to be more mobile in comparison (Fig. 4 C), as it should exist in both monomeric and dimeric conformation and either freely diffusive in the cytosol or interact with the membrane. Last, the two monomeric mutants G12V and K16A displayed extremely similar average speeds that were both faster than the wild type protein (Fig. 4 C). Since these MinD variants cannot form dimers, general diffusion should be faster, as there should be no MinD dimers or oligomers, and possibly no recruitment by MinJ.

If the underlying movement in MSD analysis can be described by simple Brownian motion, the gradient of the curve is proportional to the diffusion coefficient D (77). We know, however, that MinD interacts at least with MinC, MinJ, the membrane and itself, in combination with the fact that bacterial cells exhibit natural confinement. With these information one cannot assume simple Brownian motion, therefore we have chosen jump distance (JD) analysis as a more accurate assessment to describe Halo-MinD dynamics (Fig. 4 D & E). In JD analysis, the distances molecules travel between subsequent frames are plotted in a probability distribution (Fig. 4 E), and diffusion coefficients are obtained by curve fitting (77), in this case with a three-component model separating fast, slow and confined molecules (Fig. 4 D & E). To accurately compare population sizes, the simultaneous option was chosen in the SMTracker software, fixing the diffusion coefficients for every MinD variant (74). To also obtain individual, variant specific diffusion coefficients, the same data was additionally fitted individually (non-simultaneous SQD analysis in SMTracker2, Tab. 3).

Tab. 3.

Generally, the results of JD analysis correlated well with our in vitro data. When dissecting the different populations in the wild type, more than a third of all proteins appear to be confined (36.6%, Fig. 4 D), which likely comprises membrane bound protein that is further stabilized by protein-protein interactions via lateral interaction of MinD with itself, and possibly MinC and MinJ. The slow mobile population was the largest (45.7%, Fig. 4 D), and likely summarizes different conformations of MinD that interact with the membrane, but are still diffusive and not captured or stabilized through further interactions. The third, fast diffusive population (17.7%, Fig. 4 D) likely reflects cytosolic MinD. These results stress the importance of the MinD ATPase cycle for its localization and dynamics, as these are severely altered in the MinD mutants.

In detail, the probability distributions of Halo-MinD-D40A and I260E (Fig. 4 E) illustrate the strong shift between MinD molecules in a confined (D40A) and a diffusive state (I260E) best. When examining the different populations, most Halo-MinD-D40A molecules were classified as confined (54.2%) and slow mobile (40.2%), whereas I260E only displayed ∼9% of confined molecules (Fig. 4 D), matching prior results and expectations of the membrane binding mutant. When comparing D40A to wild type Halo-MinD, the largest shift is observed between the fast mobile and confined populations, shifting towards a distinctively more confined regiment in the D40A mutant (36.6% WT, 52.2% D40A; Fig. 4 D) and visible disappearance of the fast population in the probability distribution (Fig. 4 E). These results are in line with the previous observations, as a trapped dimer of MinD will remain at the membrane at much higher efficiency (Fig. 3 E & Tab. 1), and appeared to form foci and clusters (Fig. 2 B, Fig. 4 A). G12V and K16A mutants on the other hand showed an increase in fast mobile populations (WT 17.7%, G12V 25.8%, K16A 26.6%; Fig. 4 D E), which could be attributed to a reduced interaction with MinD interaction partners (MinC, MinD, MinJ and plasma membrane). This notion is supported by the decrease in confined molecules for both G12V and K16A mutants compared to the wild type (WT 36.6%, G12V 23.7%, K16A 29%; Fig. 4 D & E), which are expected to be membrane bound and likely stabilized through other protein-protein interaction, which may be reduced or absent in the monomeric conformation. This notion is also supported by the absence of a MinD gradient in epifluorescence images of G12V and K16A mutants (Fig. 2 B), a pattern usually formed through sequestering via MinJ. Finally, when comparing both monomeric variants, the K16A mutant displays a noticeable shift between the slow diffusive (G12V 50.5%, K16A 44.3%; Fig. 4 D) and confined populations (23.7% G12V, 29% K16A; Fig. 4 D). The larger confined population of Halo-MinD-K16A is consistent with the BLI results (Fig. 3 D, Tab. 1), where K16A appears to have a lower dissociation from the membrane, increasing its likelihood of confinement, possibly caused by the inability to bind ATP.

In summary, these results emphasize the importance of the MinD ATPase cycle for functionality of the Min system in B. subtilis. Even though B. subtilis MinD is not capable of oscillation, its intracellular dynamics are highly dependent on the functionality of the MinD ATPase domain and its catalytic site, which is reflected by the variety of localization and interaction defects displayed by the MinD mutants on a molecular level, that all result in severe division anomalies.

MinC and MinJ have different effects on MinD protein dynamics in SMLM

MinC and MinJ interact with MinD in vivo, suggesting protein dynamics should be affected by this. However, in our ATPase and BLI experiments, MinC had only minor effects on ATPase activity and membrane binding kinetics of MinD, while the PDZ domain slightly enhanced ATPase activity without altering membrane binding kinetics of MinD. To validate these observations in vivo and to identify differences between MinD interacting with a cytosolic PDZ-domain or a full-length MinJ, we performed the same SMLM workflow and analysis pipeline described in the previous section on cells expressing Halo-MinD in ΔminC and ΔminJ backgrounds and compared them to the wild type background (Fig. S 12, Tab. 4).

Tab. 4.

When comparing the average Halo-MinD distribution between the respective strains, the ΔminC mutant exhibited a similar intracellular localization pattern as the wild type, with strong membrane enrichment and highest concentrations at midcell and poles (Fig. S 12 A). However, ΔminC cells interestingly exhibited a slightly reduced cytosolic signal and a corresponding increased membrane association. As expected, ΔminJ cells lacked the characteristic septal accumulation. Instead, Halo-MinD was distributed relatively uniformly along the membrane, consistent with MinJ being required for efficient sequestering of MinD to the future division site and cell poles.

In the stationary localization analysis (Fig. S 12 B), Halo-MinD in the ΔminC strain displayed only a modest increase in the static fraction compared to wild type (23.4 % from 20.1 %). This shift can be attributed to a reduction in the mixed, likely membrane-interacting population (from 19.0 % to 14.7 %), whereas the freely mobile fraction remained essentially unchanged (61 % to 62 %). These results indicate that in the absence of MinC, less MinD moves along the membrane and a slightly larger proportion becomes immobilized. Although the effect size is small, it suggests that MinD may form larger or slightly more stable membrane-associated assemblies when MinC is absent in vivo. In the ΔminJ background, the static Halo-MinD population surprisingly increased (20.1 % to 28.2 %), together with respective reductions in both the freely diffusing cytosolic pool (61.0 % to 56.6 %) and the mixed, likely membrane-interacting population (19.0 % to 15.1 %). This redistribution suggests that MinJ contributes to efficient MinD turnover on the membrane, in line with MinD ATPase data with His-PDZ (Fig. S 12). In the absence of MinJ, MinD appears to remain membrane-bound for longer, consistent with impaired detachment or reduced cycling between membrane-associated states. One possible explanation is that MinD may form more stable or aggregation-prone assemblies when MinJ is absent, whereas the presence of MinJ could promote more dynamic membrane association and hence sharper local accumulations and cycling of MinD, necessary for enriched polar and septal localization. Jump-distance analysis with fixed diffusion coefficients corroborated these findings (Fig. S 12 C & D), showing that ΔminJ cells contained a substantially enlarged confined population of MinD (27.9 % vs. 44.9 %), together with a large reduction of the slow diffusive population (50.9 % vs. 37.8 %). Fitting the data without fixed diffusion coefficients produced the same trend (Tab. 4), indicating that the shift reflects MinD spending more time in the respective state rather than actual changes in the underlying diffusive coefficients.

In the ΔminC strain, Halo-MinD shows a slight redistribution toward the confined state, with an increased fraction of very slow or immobile molecules (27.9 % vs. 36.8 %) and a corresponding decrease in both the fast and intermediate populations (50.9 % vs. 45.8 % and 21.2 % vs. 17.4 %, respectively, Fig. S 12 C & D). SQD analysis without fixed diffusion coefficient (Tab. 4) shows that the diffusion coefficients of the fast and intermediate populations are reduced compared with wild-type, indicating that the mobile molecules themselves move more slowly in average. Combined, this could suggest that in the absence of MinC, membrane-bound MinD is more prone to forming clusters or larger aggregates, which restrict mobility and reduce the efficiency of its membrane cycling.

Together, these data suggest that MinJ promotes productive membrane cycling of MinD, and in its absence, MinD accumulates more often in a confined, likely non-productive membrane-associated state, whereas MinC plays a comparatively minor regulatory role, where it passively contributes to MinD mobility by preventing excessive clustering and facilitating proper membrane-associated dynamics.

Discussion

In bacteria, the oscillation of MinCDE in E. coli is one of the best studied examples of a reaction-diffusion system, and the two core components MinD and MinE cycle between membrane and cytosol in an ATP-dependent manner, generating a standing wave with a time-resolved minimum at midcell (15, 83). Here, binding and unbinding of the components to the membrane surface influences the diffusion coefficients and controls the dynamic activity of MinC (84). While the E. coli MinCDE system is one of the best-studied examples of intracellular pattern formation in bacteria (15, 84-86), the Min system in B. subtilis remains less well characterized. In a previous study (61), we suggested that gradient formation depends on ATPase-driven MinD dynamics, backed by PALM and FRAP analysis as a basis for a mathematical model. However, due to a lack of biochemical data, we had to assume that the B. subtilis MinD cycles between membrane-bound and cytosolic localizations based on ATP-binding and hydrolysis, similar to E. coli MinD (61). A further gap in our knowledge of the B. subtilis Min system concerned the activation mechanism of MinD in absence of a MinE homolog.

To address this gap, we first analyzed B. subtilis MinD biochemically in vitro. Purified MinD showed basal ATPase activity that drastically increased upon liposome addition. Although E. coli MinD similarly requires lipids, it additionally depends on MinE for full activation (64). The reported maximum ATPase rate, around 20 nmol mg⁻¹ min⁻¹ (64, 65), was nearly identical to that of B. subtilis MinD, indicating that membrane binding alone is sufficient to trigger robust ATP hydrolysis in the Bacillus system. When comparing to other MinD/ParA ATPases that have been biochemically characterized, the activity level of B subtilis MinD is on the lower end of the spectrum with a k_cat of 36.27 h⁻¹. Soj, the ParA homologue in B. subtilis, reaches more than 2-fold higher catalytic activity (83.6 h^-1) when fully activated by supplementing Spo0J (ParB) and DNA (87), similarly to Soj in Thermus thermophiles combined with Spo0J and parS containing DNA (70). Studies investigating the activity levels of ParA in Caulobacter crescentus, purified in presence of ATP, determined a k_cat of 0.79 min^-1 (47.4 h^-1) in absence of ParB, increasing almost 5-fold to 3.7 min^-1 (222 h^-1) in presence of ParB (88). A more recent study found the C. crescentus ParA k_cat increase from 5.8 hr⁻¹ to 120 hr^-1 upon addition of ParB (89).

These findings suggest that membrane association is the primary trigger for MinD ATPase activity in B. subtilis. Specifically, the comparison of maximum velocities, activity stimulation factors, and K_M values between the E. coli and B. subtilis MinD proteins suggests that a functional homolog of the MinE protein is not necessary in B. subtilis. At the same time, our biochemical data indicate that MinJ/PDZ increases MinD ATP hydrolysis rates at least to a modest extent, (between 22% and 37%, Fig. S 5 B), pointing to a broader regulatory influence of MinJ on MinD that becomes evident from additional mechanistic analyses discussed below. In contrast, the presence of MinC led only to a slight decrease in MinD ATPase activity in our APTase assays (∼ 5%, Fig. S 5 A). This modest inhibitory effect suggests that MinC does neither directly stimulate MinD turnover nor inhibit it, and may instead stabilize MinD in a conformation or assembly state with mildly reduced catalytic activity, also discussed below.

With the goal of testing binding of isolated MinD to the membrane, we designed a Bio-layer interferometry experiment in which we immobilized liposomes and measured binding of purified MinD proteins. Importantly, without addition of ATP (and ADP present from purification) we observed only very weak MinD binding to the liposomes (Fig. S 13), indicating that MinD membrane binding and dissociation is coupled to the ATP hydrolysis cycle or the nucleotide binding state. In presence of ATP, MinD readily binds to the liposomes with high affinity. The ParA/MinD family is well studied, and mutations producing strictly monomeric (G12V, K16A), dimeric (D40A), or membrane-binding-deficient (I260E) MinD variants are well established (68-70). The most intriguing observation of the binding experiments is that MinD monomers can apparently bind to the liposome membrane in vitro. Both monomeric mutants (G12V and K16A) are readily binding to the membrane, but the release of the proteins from the membrane surface differs drastically. While the K16A mutant, unable to bind ATP, has a slow membrane release rate (k_off rate of only 2.77^*10^-3 s^-1), the G12V monomeric mutant, which can still bind ATP, is released fast from the membrane (k_off rate 47.8^*10^-3 s^-1). The D40A mutant of MinD, which is thought to be a stable ATP-bound dimer, releases with very slow kinetics (k_off rate 1.76^*10^-3 s^-1). This indicates that ATP hydrolysis accelerates MinD release from the membrane and that ATP binding itself influences MinD association and dissociation kinetics, although we cannot entirely exclude that the mutations introduce subtle structural changes that affect membrane affinity. These findings crucially support a reaction diffusion model in which the MinD ATPase cycle is essential for correct pattern formation of B. subtilis MinD. When testing the impact of MinC on MinD membrane-binding kinetics, no significant differences were observed, suggesting that MinC does not substantially influence the binding dynamics of MinD, in-line with the collective view of MinC as a passenger rather than a driver of MinD dynamics (84). It would have been desirable to study the interaction of MinD with reconstituted full-length MinJ proteins to specifically investigate the influence that MinJ has on membrane interaction of MinD. Unfortunately, we did not succeed in the purification of a full length MinJ protein so far. MinJ contains seven predicted transmembrane helices and displays a pronounced instability, rendering purification difficult. While the binding kinetics were unaffected during BLI, our in vitro ATPase experiments combining the soluble PDZ domain of MinJ with MinD already demonstrate that MinJ possibly modulates MinD:membrane interactions to some extent through an effect on the hydrolysis efficiency.

To clarify this point and to better understand the isolated MinD membrane interaction under in vivo conditions, we turned to state-of-the-art microscopy (SMLM) (92, 93).

SMLM data comparing the behavior of MinD in wild type vs. ΔminJ background (Fig. S 12) collectively suggest that MinJ stabilizes correct pattern formation of MinD. By promoting efficient membrane detachment and maintaining MinD in a dynamically cycling state, MinJ appears to ensure that MinD remains properly redistributed throughout the cell cycle rather than becoming sequestered in non-productive clusters. This organizational role likely underlies the ability of MinJ to position MinD, and thus MinC, at the appropriate cellular sites (55, 56), thereby supporting robust spatial regulation of division. However, a short-lived MinJ mediated stabilization of MinD-dimers at the membrane would be comparable to the stabilizing effect that MinE has on MinD oligomers in E. coli (90, 91), and was previously observed in our in vivo FRAP studies (61). In contrast to MinJ, MinC did not seem to have a major impact on MinD dynamics in vivo. When observing MinD in absence of minC, the main effect was a subtle reduction in MinD cycling activity, shifting a small fraction of MinD into more static or confined membrane-associated states. In the wild type situation, MinC may passively restrict uncontrolled aggregation by capping or destabilizing certain interfaces, or conversely transiently stabilize short-lived assemblies required for regulation.

The locked MinD dimers (D40A mutant) frequently nucleated into larger membrane clusters. This cluster formation is underlined by the drastic increase of the confined/immobile population (54.2%) of D40A in SMLM. This increase in the static population comes to the expense of freely diffusive protein (∼5.6 %), indicating that the vast majority of MinD D40A is membrane associated. We observed a clear enrichment of these MinD clusters at the cell poles and septa, lending support to the notion that MinD dimers are specifically stabilized/recruited by MinJ. This interpretation is supported by previous data showing that in absence of MinJ, MinD proteins are randomly localized along the membrane in clusters, but not enriched at poles or (in a minJ mutant background rare) septa (57). In line with this idea, the monomeric MinD variants (G12V, K16A) are evenly bound to the membrane surface along the entire cell length.

The fact that most monomeric MinD appears membrane associated in vivo also supports our in vitro measurements in which monomeric MinD variants efficiently bind the membrane, a notion strongly supported by a recent report that examined localization of the same monomeric variants in epifluorescence microscopy (94). The time-resolved MinD gradient is lost in all mutants - including D40A - which should still be recruited by MinJ, suggesting the ATPase cycle and thus MinD cycling is indeed necessary for proper formation of a gradient, as suggested before (61). Our data are in perfect agreement with recent in vitro data collected with the E. coli MinD using mass-sensitive particle tracking (MSPT) (73). MSPT analysis revealed that E. coli MinD forms lateral interactions, leading to higher order oligomers (trimer, tetramer and even higher oligomers). It was speculated that dimerization of E. coli MinD at the membrane leads to conformational changes opening an additional interaction surface of MinD for lateral interaction (73). These lateral MinD interactions were thought to recruit further monomeric MinD from solution to the membrane, thereby forming a nucleation for larger clusters. The MinD cluster formation that we observed in B. subtilis using SMLM (61) are likely in vivo evidence for this cooperative membrane binding. It should be noted that the MSPT analysis was not sensitive enough to detect monomer binding to the membrane, but this was explicitly not ruled out (73). Addition of a second amphipathic helix to the E. coli MinD however, was shown to promote monomer binding of E. coli MinD to the membrane (66). Thus, the difference between the E. coli and B. subtilis MinD is likely the higher membrane binding capacity of the Bacillus protein. The observed cooperativity in membrane binding and thus nucleation, as shown for the E. coli MinD in vitro (66, 73) and here for the B. subtilis MinD in vitro and in vivo, is required for pattern formation by conferring nonlinearity (61, 95, 96).

Consistent with this model, our SMT data revealed three major dynamic states of MinD: fast cytosolic diffusion, a slower membrane-associated pool, and a confined fraction corresponding to clustered or multimeric assemblies. Most of the membrane-binding-deficient mutant I260A shifted drastically into the fast, cytosolic population, validating this assignment. Conversely, the locked-dimer (D40A) mutant showed a pronounced increase in confined, slow-moving molecules, in line with its strong tendency to nucleate clusters. Monomeric variants (G12V, K16A) displayed reduced confinement, supporting the view that dimerization is required for higher-order assembly or recruitment. Together, these observations indicate that MinD dynamically cycles between cytosol and membrane in an ATP-dependent manner. Thus, we conclude that membrane bound dimers of MinD have a much stronger tendency to interact with and recruit binding partners (including self-nucleation). This finding is further supported by a recent report that analyzed MinD dynamics in vivo using diffraction limited microscopy (94). In this study, a MinD D40A mutant hyper-recruited MinC, leading to cluster formation of MinCD at the membrane. Deletion of MinC restored the MinD gradient formation, indicating that MinC might indeed be part of a MinC:MinD co-polymer, as suggested for the E. coli system (97). This assumption is in good agreement with our in vivo findings, where self-nucleation seems to increase drastically for the D40A mutant or in absence of MinJ. In contrast, monomeric G12V and K16A mutants do not display this behavior, in line with monomeric MinD variants not being able to interact with MinC, which was similarly observed through widefield microscopy in the same study (94).

Collectively, our data reveal that the MinD ATPase cycle is essential for regulated membrane binding and release, which is crucial for the formation of a sharp MinD gradient at the septum (61). However, in contrast to E. coli MinD, monomeric B. subtilis MinD can bind the membrane with higher affinity, suggesting that MinD also dimerizes at the membrane surface. Dimer formation is essential for ATPase activity, as indicated by our in vitro measurements, and is necessary for efficient membrane detachment. The main characteristics of the E. coli MinD with respect to membrane binding, self-interaction leading to nucleation, and membrane detachment are therefore shared in the B. subtilis protein. The difference, that ATP hydrolysis of the B. subtilis MinD does not require a trigger other than membrane binding, makes sense in a system that does not need to oscillate, but rather accumulate and sustain a sharp gradient at the current site of division. However, the balance of a dynamic membrane cycle of MinD seems to be fine-tuned by the interaction with MinC and especially MinJ, which appears to keep MinD in a productive state and protects it from aberrant aggregation, while recruiting it towards poles and septum, the intended site of action of MinC.

Materials and Methods

Bacterial strains, plasmids and oligonucleotides

Plasmid and strain construction of all relevant constructs is described in the supplementary information. Oligonucleotides, plasmids and strains used in this study are listed in Tab. S 2, Tab. S 3 and Tab. S 4, respectively.

Media and growth conditions

E. coli was grown on lysogeny broth (LB) agar plates containing 10 g L^-1 tryptone, 5 g L^-1 yeast extract, 10 g L^-1 NaCl and 1.5% (w/v) agar at 37°C overnight using appropriate selection (kanamycin 50 µg ml^-1, chloramphenicol 35 µg ml^-1).

B. subtilis was grown on nutrient agar plates using commercial nutrient broth and 1.5% (w/v) agar at 37°C overnight. To reduce inhibitory effects, antibiotics were only used for transformations and when indicated, since allelic replacement is stable after integration (kanamycin 5 µg ml^-1, spectinomycin 100 µg ml^-1).

For microscopy, B. subtilis was inoculated to an OD₆₀₀ 0.05 from a fresh overnight culture and grown in MD medium - a modified version of Spizizen Minimal Medium (98) – at 37°C with aeration in baffled shaking flasks (200 rpm) to OD₆₀₀ 0.5-0.8. MD medium contains 10.7 mg ml⁻¹ K₂HPO₄, 6 mg ml⁻¹ KH₂PO₄, 1 mg ml⁻¹ Na₃ citrate, 20 mg ml⁻¹ glucose, 50 µg ml⁻¹ L-tryptophan, 11 µg ml⁻¹ ferric ammonium citrate, 2.5 mg ml⁻¹ L-aspartate and 0.36 mg ml⁻¹ MgSO₄ and was always supplemented with 1 mg ml⁻¹ casamino acids. Subsequently, cultures were diluted to OD₆₀₀ 0.05 in fresh, pre-warmed MD medium and grown to OD₆₀₀ 0.5 (exponential phase).

Purification of His-MinC, His-PDZ, His-MinD and variants

For heterologous expression of His-MinC, His-PDZ and His-MinD and its variants, freshly transformed BL21(DE3) cells carrying the respective pET-28a expression vector (Tab. S 3) as well as pLysS (His-MinC, His MinD and variants only) were transferred to LB medium with appropriate selection (kanamycin 50 µg ml^-1; chloramphenicol 35 µg ml^-1 in presence of pLysS) and grown overnight (37°C, 200 rpm). The next day, fresh LB medium was inoculated 1:150 from the overnight culture and grown to an OD₆₀₀ of 0.5 at 37 °C and 200 rpm. Since His-PDZ tended to form inclusion bodies under these conditions, its expression culture was instead grown at 18 °C and 120 rpm without IPTG induction. For all other constructs, protein expression was induced with 1 mM IPTG. Cells expressing His-MinC and all His-MinD variants were harvested after 3 h of induction, whereas His-PDZ cultures were collected after overnight incubation. Cells were harvested by centrifugation (4 °C, 5000 × g, 15 min) and stored at -80°C. For purification, a protocol of E. coli MinD purification was adapted from (63): the cell pellet was resuspended on ice in lysis buffer (50 mM NaH₂PO₄, 500 mM NaCl and 10 mM imidazole; pH 8.0), supplemented with protease inhibitor cocktail (Roche), 0.2 mM Mg²⁺-ADP and 0.4 mM TCEP (His-MinD and variants only) and ∼250 U mL^-1 DNase I (Roche), and subsequently lysed in a pre-cooled (4°C) French pressure cell (Amico; 3 × 1,200 psi). The debris was then removed via centrifugation (10,000 g, 30 min, 4°C) and the supernatant was passed through a 1 ml Ni-NTA column (Protino, Macherey-Nagel) utilizing liquid chromatography (ÄKTApurifier). Here, the resin was washed with 10 ml binding buffer A (50 mM NaH₂PO₄, 500 mM NaCl, 20 mM imidazol; pH 8) before elution buffer B (50 mM NaH₂PO₄, 500 mM NaCl, 250 mM imidazol; pH 8, (and 0.2 mM Mg²⁺-ADP and 0.4 mM TCEP for His-MinD and variants)) was added sequentially at concentrations of 5% (10 ml) 10% (20 ml), and finally 100% (10 ml). Recombinant protein eluate was collected in 0.5 ml fractions. Fractions containing high concentrations of protein (determined by chromatogram analysis, sodium dodecyl-sulfate polyacrylamide gel electrophoresis (SDS-PAGE) and Western blotting using Penta-His antibody (QIAGEN)) were pooled and concentrated using Amicon filter devices with 10 kDa molecular weight cut-off (Millipore; Merck). Additionally, protein aggregates were removed by pelleting them through centrifugation (10,000 g, 10 min, 4°C).

Samples were further purified by size-exclusion chromatography through a Superdex200 Increase 10/300 GL column (Cytiva) at a flow rate of 0.75 ml min^-1 using gel filtration buffer C (50 mM HEPES-KOH, 150 mM KCl, 10% (v/v) glycerol, 0.1 mM EDTA, pH 7.2). For His-MinD and its variants, buffer C was supplemented with 0.4 mM TCEP and 0.2 mM Mg²⁺-ADP immediately before use. His-PDZ was not subjected to size-exclusion chromatography because it could not be recovered after SEC. Instead, it was used directly following affinity purification. To calibrate the column and allow identification of relevant fractions, a gel filtration standard (Bio-Rad) was run according to the manufacturers protocol and analyzed using the same conditions, respectively. Fractions containing recombinant protein were pooled and either directly used for the respective assays, stored at 4°C for a maximum of one week or flash-frozen and stored at -80 °C. Final protein concentration was determined using Bradford assay with bovine serum albumin (BSA) as standards. Protein purity and stability was confirmed via SDS-PAGE using 12% polyacrylamide gels and Western blotting using Penta-His antibody (QIAGEN) (Fig. S 1, Fig. S 3, Fig. S 4, Fig. S 6, Fig. S 8, Fig. S 9 and Fig. S 10).

ATP hydrolysis assays

ATP hydrolysis was analyzed in continuous reactions using the EnzChek Phosphate Assay Kit (Thermo Fisher Scientific), according to the manufacturer’s manual. Reaction volumes were reduced to 100 µl and assayed in flatbottom 96-well plates (Greiner-UV-Star 96-well plates). Reactions were measured at 37 °C continuously every minute over a time-course of 3 h in a Tecan Infinite 200 Pro with the Software Tecan i-control v.2.0. If not indicated differently, the reactions contained either 0-16 µM His-MinD or the respective variant/protein combination as well as 2 mM Mg²⁺-ATP and 0.2 mg ml^-1 liposomes. Liposomes were made from E. coli total extract (Avanti) by first drying and then resuspending and rehydrating lipids in gel filtration buffer C (50 mM HEPES-KOH, 150 mM KCl, 10% (v/v) glycerol, 0.1 mM EDTA). Next, the lipids were extruded 40 times through 100 nm polycarbonate membranes (Nuclepore™ Track-Etched Membrane 0.1 µm, Whatman). Before starting the reactions with 2 mM Mg²⁺-ATP, the plate was preincubated for 10 min in order to eliminate phosphate contamination.

Data were first analyzed with Excel (normalized, subtraction of ATP auto hydrolysis and subtraction of no substrate control), a linear regression was used to determine the hydrolysis rate per minute. Data visualization and further analysis was performed using GraphPad Prism version 9.4.0 for Windows, (GraphPad Software). Statistical comparisons between His-MinD alone (n = 2) and His-MinC/His-PDZ conditions (n = 3 each) were performed using Welch’s t-test (two-tailed), which is appropriate for unequal variances and unequal sample sizes.

AlphaFold prediction and structure-based analysis

A structural model of B. subtilis MinD was calculated using AlphaFold2 (99) installed on a local computer (Fig. 2 A), where the structural homology search was performed against the full AlphaFold database as of 11.02.2024 (99).

A structural model of B. subtilis MinJ was calculated using AlphaFold3 (100), where the structural homology search was performed against the full AlphaFold database as of 01.02.2025 (100).

Bio-layer Interferometry

Measurements were performed on the BLItz platform (Sartorius) using Streptavidin (SA) Biosensors (Sartorius, 18-5019). The BLItz device was operated with the Software BLItz Pro (version 1.3.1.3) using the advanced kinetics protocol modified as shown in Table S5 and essentially as described before (101-103).

Beforehand, Liposomes were made from a 33:1 mixture of E. coli total extract (Avanti) and DSPE-PEG₍₂₀₀₀₎-biotin (Avanti) by first drying and then resuspending and rehydrating lipids in gel filtration buffer C (50 mM HEPES-KOH, 150 mM KCl, 10% (v/v) glycerol, 0.1 mM EDTA). Next, the lipids were extruded 40 times through 100 nm polycarbonate membranes (Nuclepore™ Track-Etched Membrane 0.1 µm, Whatman). Biosensors were hydrated for at least 10 min in a 96-well plate in 200 µl gel filtration buffer C.

As a starting point, unspecific binding was tested. For this, non-biotinylated His-MinD was used to be loaded onto the Streptavidin biosensor. For the binding assays, all measurements were performed in black reaction tubes with 200 µl gel filtration buffer C or the indicated solution (Tab. S 5). After measuring the initial baseline, biotinylated liposomes were allowed to bind the biosensor. Another baseline was measured in the same tube as before. For association, 200 µl solutions of His-MinD and variants in different concentrations and/or in combination with His-MinC or His-PDZ, respectively, were exposed to the sensor, first without and later with addition of 2 mM Mg-²⁺ATP (final concentration) just before use. Furthermore, unspecific binding to liposome-saturated sensors was tested using bovine serum albumin (BSA, 16 µM). For dissociation, the sensor was exposed to buffer again. Each measurement was performed with a fresh biosensor.

For BLI experiments involving His-MinD and one of its direct interactors, we first assessed whether His-MinC or His-PDZ can directly interact with His-MinD in vitro. To this end, a liposome-coated BLI sensor was first dipped into a 8 µM His-MinD solution. Subsequently, the sensor was exposed to solutions containing either 8 µM His-MinD together with 8 µM His-MinC or 8 µM His-PDZ, respectively. As a negative control, a sample containing 8 µM His-MinD and 8 µM BSA was used. Both His-MinC and His-PDZ induced a significant increase in the equilibrium binding signal upon interaction with immobilized His-MinD, consistent with specific complex formation. In contrast, the BSA control showed only a minimal signal increase, confirming the specificity of the observed binding. To further validate the sensor specificity, individual proteins were tested. Only His-MinD exhibited robust binding to the liposome-coated sensor, while His-MinC and His-PDZ showed negligible signal, indicating non-specific interactions (Fig. S11). These initial tests support the expected specificity of the MinD:MinC and MinD:PDZ interactions.

Kinetic analysis and fitting were performed by the BLItz Pro Software (version 1.3.1.3). Data was exported and visualized using GraphPad Prism version 9.4.0 for Windows, (GraphPad Software).

The BLItz Pro Software analysis follows a 1:1 binding model, where:

with k_on = rate of association or “on-rate” and k_off = rate of dissociation or “off-rate”.

Formula for fitting the association (a = slope):

Formula for fitting the dissociation:

Resulting in the equilibrium dissociation constant K_D:

Fluorescence microscopy

Microscopy slide preparation

For epifluorescence imaging, B. subtilis cells were mounted on 1.5% MD agarose pads using 1.5 × 1.6 cm “Gene Frames” (Thermo Scientific), and incubated 10 min at 37°C before microscopic analysis.

For SMLM, slides and coverslips were first cleaned by overnight storage in 1 M KOH, carefully rinsed with ddH2O, and subsequently dried with pressurized air. Next, 1.5% (w/v) low melting agarose (Sigma-Aldrich) was dissolved in MD medium. Medium was sterile filtered (0.2 µm pore size) shortly before being used to remove particles. To produce flat, uniform, and reproducible agarose pads, 1.5 × 1.6 cm “Gene Frames” (Thermo Fisher) were utilized, and pads were allowed to solidify for 1 h at room temperature to be used within the next 3 h.

Staining of HaloTag

For staining, 1 mL of bacterial culture was moved to a 2 mL microcentrifuge tube and incubated with 5 nM (SMLM) or 50 nM (epifluorescence) HaloTag TMR ligand at 37 °C for 10 min. Cells were harvested (4000 g, 3 min, 37 °C) and washed four times in pre-warmed MD medium, before being mounted on the previously described microscopy slides.

Epifluorescence imaging

For strain characterization (Fig. 2 B), microscopy images were taken on a Axio Observer 7 microscope (Zeiss) equipped with a Colibri 7 LED light source (Zeiss) and an OrcaR² camera (Hamamatsu) using a Plan-Apochromat 100×/1.4 oil Ph3 objective (Zeiss). TMR fluorescence was visualized with a 90 HE LED filter set (QBP 425/30 + 514/30 + 592/30 + 709/100) (Zeiss) and excited at 555 nm (100% intensity of Colibri 7 LED) for 400 ms. The microscope was equipped with an environmental chamber set to 37 °C. Digital images were acquired with Zen Blue (Zeiss) and analyzed and edited with Fiji (104) ImageJ2 (105). For measurement of cell length and minicells, cells were stained with NileRed (1 µg/ml final concentration) at OD₆₀₀ 0.5 and incubated for further 20 minutes (200 rpm, 37°C) before mounting them on an agarose pad. To minimize bias during analysis, cell lengths and number of minicells were identified using the Fiji (104) plugin MicrobeJ 5.11c (106) on phase contrast images with manual correction by screening the results against the simultaneously acquired membrane stain images and sorting out mislabeled cells. For cell length measurements, MicrobeJ modules options, exclude on edges, shape descriptor, segmentation and chain were used with default settings, except for following parameters in options (area 1-max; length 1-max; width 0.6-1; sinuosity 1-max; angularity 0-0.5) and chain (tolerance 0.1; area 1.5-max). To identify minicells, MicrobeJ modules options, exclude on edges, shape descriptor and segmentation were used with default settings, except for following parameters in options (area 0.15-1; length 0.5-1.2; width 0.3-1; circularity 0.95-max).

SMLM imaging

SMLM imaging was performed with an Elyra 7 (Zeiss) inverted microscope equipped with two pco.edge sCMOS 4.2 CL HS cameras (PCO AG), connected through a DuoLink (Zeiss), only one of which was used in this study. Cells were observed through an alpha Plan-Apochromat 63x/1.46 Oil Korr M27 Var2 objective, yielding a final pixel size of 97 nm. During image acquisition, the focus was maintained with the help of a Definite Focus.2 system (Zeiss). Fluorescence was excited with a 561 nm (100 mW) laser, and signals were observed through a multiple beam splitter (405/488/561/641 nm) and laser block filters (405/488/561/641 nm) followed by a Duolink SR QUAD (Zeiss) filter module (secondary beam splitter: LP 560, emission filters: EF BP420-480 + BP495-550).

Cells were illuminated with the 561 nm laser (50% intensity) in TIRF mode (62° angle). For each time lapse series 10,000 frames were taken with 20 ms exposure time (∼24 ms with transfer time included) and 50% 561 nm intensity laser.

SMLM analysis

For average localization analysis (Fig. 4 A), a single-molecule localization table was constructed from a 2D Gaussian fitting in the ZEN 3.0 SR (black) software (Zeiss) with a peak mask size of 9 and a peak intensity-to-noise ratio of 6, and further allowing molecules to overlap. Filtering was used to minimize noise, background, and out-of-focus emitters, discarding events that did not emit within a Point spread function size (PSF) of 100-200 nm and did not display a photon count of 50-600. Further analysis was performed using individual Fiji (104) scripts for the cell outlines and R (4.2.2) scripts in R Studio (2023.12.1+402) for the corresponding localization events (107, 108). Here, the remaining events were transformed to align the orientation of all cells in which the signals were detected, to be able to summarize all recorded localizations in one normalized cell per strain. The localization frequency is displayed as a heatmap using the gradual viridis scaling from the ggplot2 R package (109), where the highest frequency (normalized frequency 1) is displayed in yellow and the lowest frequency (normalized frequency 0) in blue (Fig. 4 A).

For single-molecule tracking, spots were identified with the LoG Detector of TrackMate v6.0.1 (110) implemented in Fiji 1.53 g (104), an estimated diameter of 0.5 μm, and median filter and sub-pixel localization activated. The signal-to-noise threshold for the identification of the spots was set at 7. To limit the detection of particles to single-molecules, frames belonging to the bleaching phase of TMR were removed individually from the time lapses prior to the identification of spots. Spots were merged into tracks via the “Simple LAP Tracker” of TrackMate, with a maximum linking distance of 500 nm and no frame gaps allowed. Only tracks with a minimum length of 5 frames were used for further analysis, yielding a minimum number of total tracks per sample of 2596.

To identify differences in protein mobility and/or behavior, the resulting tracks were subjected to stationary localization analysis (SLA), mean-squared-displacement (MSD) and square displacement (SQD) analysis in SMTracker 2.0 (74). SLA analysis was performed with a confinement circle radius of 97 nm and a minimum number of 5 steps to be considered static. The average MSD was calculated for four separate time points per strain (exposure of 20 ms - τ = 24, 48, 72 and 96 ms), followed by fitting of the data to a linear equation. The last time point of each track was excluded to avoid track-ending-related artifacts. The cumulative probability distribution of the square displacements (SQD) was used to estimate the diffusion constants and relative fractions of up to three diffusive states. Diffusion constants were determined simultaneously for the compared conditions, therefore allowing for a more direct population fractions comparison. Additionally, the analysis was performed for each strain non-comparatively, yielding individual diffusion coefficients, displayed in Tab. 3.

Data availability

All strain and plasmids are available on request.

Acknowledgements

This research was supported by the Deutsche Forschungsgemeinschaft (DFG) within the Transregio Collaborative Research Center (TRR 174) “Spatiotemporal Dynamics of Bacterial Cells”. We thank Leendert Hamoen (Amsterdam) and Henrik Strahl (Newcastle) for communicating unpublished results prior to publication. We are also grateful to Anna Becker and Vivian Hennig (Kiel) for help in the initial stages of this project as part of their BSc theses.

Additional information

Author Contributions

M.B. and H.F. conceived the study, H.F. constructed the strains, performed the in vivo and in vitro experiments incl. microscopy. C.D. performed in vitro experiments. H.F., C.D. and M.B. analyzed the data, H.F. and M.B. wrote the manuscript.

Funding

Deutsche Forschungsgemeinschaft (DFG) (TRR174,P5)

• Marc Bramkamp

Additional files

Supplemental Material