Membrane binding properties of the cytoskeletal protein bactofilin

Bactofilins are a widespread family of cytoskeletal proteins with important roles in bacterial morphogenesis, chromosome organization and motility. They polymerize in a nucleotide-independent manner, forming non-polar filaments that are typically associated with the cytoplasmic membrane. Membrane binding was suggested to be mediated by a short N-terminal peptide, but the underlying mechanism and the conservation of this interaction determinant among bacteria remain unclear. Here, we use the bactofilin homolog BacA of the stalked bacterium Caulobacter crescentus as a model to analyze the membranebinding behavior of bactofilins. Based on site-directed mutagenesis of the N-terminal region, we identify the full membrane-targeting sequence of BacA (MFSKQAKS) and identify amino acid residues that are critical for its function in vivo and in vitro. Molecular dynamics simulations then provide detailed insight into the molecular mechanism underlying the membrane affinity of this peptide. Collectively these analyses reveal a delicate interplay between the water exclusion of hydrophobic N-terminal residues, the arrangement of the peptide within the membrane and the electrostatic attraction between positively charged groups in the peptide and negative charges in the phospholipid molecules. A comprehensive bioinformatic analysis shows that the composition and properties of the membrane-targeting sequence of BacA are conserved in numerous bactofilin homologs from diverse bacterial phyla. Notably, our findings reveal a mutual interdependence between the membrane binding and polymerization activities of BacA. Moreover, we demonstrate that both of these activities have a pivotal role in the recruitment of the BacA client protein PbpC, a membrane-bound cell wall synthase involved in stalk formation whose N-terminal region turns out to associate with the core polymerization domain of BacA. Together, these results unravel the mechanistic underpinnings of membrane binding by bactofilin homologs, thereby illuminating a previously obscure but important aspect in the biology of this cytoskeletal protein family.


Introduction
Cytoskeletal proteins mediate a variety of fundamental cellular processes in bacteria, including cell growth, cell division and DNA segregation (Cabeen and Jacobs-Wagner, 2010;Wagstaff and Löwe, 2018).Many important components of the bacterial cytoskeleton have eukaryotic homologs.The conserved cell division protein FtsZ, for instance, is a member of the tubulin superfamily (Löwe and Amos, 1998;Mukherjee et al., 1993), forming dynamic polymers at the cell division site that control the assembly and function of the cell division apparatus (McQuillen and Xiao, 2020).The morphogenetic protein MreB, by contrast, which governs cell growth and morphology in most rod-shaped bacteria (Jones et al., 2001;van den Ent et al., 2001), is homologous to actin (Rohs and Bernhardt, 2021).In addition, bacteria also contain intermediate-filament-like proteins, such as the coiled-coil-rich protein crescentin, which promotes cell curvature in the crescent-shaped alphaproteobacterium Caulobacter crescentus (Ausmees et al., 2003;Liu et al., 2024).Apart from these universally conserved cytoskeletal protein families, bacteria contain a variety of polymer-forming proteins that are typically absent from eukaryotes (Wagstaff and Löwe, 2018).A prominent member of this group of proteins is bactofilin (Kühn et al., 2010).
Bactofilin homologs are widespread among bacteria and involved in diverse cellular functions.In many species, they have important roles in cell shape determination, including the establishment or adjustment of cell curvature in Helicobacter pylori (Sycuro et al., 2010), Campylobacter jejuni (Frirdich et al., 2023), Leptospira biplexa (Jackson et al., 2018) and Rhodospirillum rubrum (Pöhl et al., 2024), the modulation of cell size in Chlamydia trachomatis (Brockett et al., 2021), the formation of buds in Hyphomonas neptunium (Pöhl et al., 2024), the stabilization of rod shape in Myxococcus xanthus (Koch et al., 2011) and Proteus mirabilis (Hay et al., 1999) and the synthesis of stalk-like cellular extensions in C. crescentus (Kühn et al., 2010), Asticcacaulis biprosthecum (Caccamo et al., 2020) and Rhodomicrobium vannielii (Richter et al., 2023).Other functions include DNA organization in M. xanthus (Anand et al., 2020;Lin et al., 2017) as well as cell motility in Bacillus subtilis (El Andari et al., 2015).Bactofilins are relatively small proteins characterized by a central Bactofilin A/B domain (InterPro ID: IPR007607) that is typically flanked by unstructured terminal regions of varying lengths.The central domain is ~110 amino acids long and folds into a compact right-handed β-helix with a triangular geometry, measuring roughly 3 nm along its longitudinal axis (Kassem et al., 2016;Shi et al., 2015;Vasa et al., 2015;Zuckerman et al., 2015).This basic unit polymerizes into extended non-polar filaments through alternating head-to-head and tail-to-tail interactions between individual subunits (Deng et al., 2019).The resulting protofilaments can further assemble into higherorder structures such as bundles, lattices and 2D crystalline sheets in vitro (Holtrup et al., 2019;Kühn et al., 2010;Pöhl et al., 2024;Sichel et al., 2022;Vasa et al., 2015;Zuckerman et al., 2015) and, potentially, also in vivo (Kühn et al., 2010).Previous studies have indicated that bactofilins are typically associated with the cytoplasmic membrane (Deng et al., 2019;Hay et al., 1999;Koch et al., 2011;Kühn et al., 2010;Lee et al., 2023).However, only a small subset of them, including certain gammaproteobacterial representatives (Hay et al., 1999), contain predicted transmembrane helices, suggesting that canonical bactofilins use an unconventional membrane-targeting mechanism that may have specifically evolved in this cytoskeletal protein family.Consistent with this notion, the membrane-binding activity of bactofilin homologs from Thermus thermophilus (Deng et al., 2019) and C. trachomatis (Lee et al., 2023) has recently been shown to depend on a short peptide in the N-terminal unstructured region of the proteins.However, the conservation of this peptide, the precise mechanism underlying its affinity for the lipid bilayer, and the interplay between membrane attachment and bactofilin polymerization still remain to be investigated.
In this study, we comprehensively analyze the membrane-binding behavior of bactofilins, using BacA from C. crescentus as a model protein.Previous work has shown that BacA forms membrane-associated polymeric sheets at the old pole of the C. crescentus cell, which recruit the cell wall synthase PbpC to stimulate stalk formation at this subcellular location (Billini et al., 2019;Hughes et al., 2013;Kühn et al., 2010).By systematically exchanging residues in the N-terminal region of BacA, we identify a short N-terminal peptide (MFSKQAKS) that acts as the membrane-targeting sequence (MTS) of BacA and pinpoint residues critical for its membrane-binding activity in vivo and in vitro.We then clarify the molecular mechanism underlying the membrane affinity of this sequence using molecular dynamics simulations.A comprehensive bioinformatic analysis suggests that the mode of action revealed for the MTS of BacA is conserved in a large number of bactofilin homologs from various different phyla.Importantly, our findings indicate a mutual stimulatory effect of BacA membrane association and polymerization as well as a key role of BacA assembly in the interaction with its client protein PbpC.Together, these findings reveal the mechanistic basis of membrane binding by bactofilin homologs and highlight the implications of this process for bactofilin function, thereby shedding light on a thus-far poorly understood aspect in the biology of this widespread cytoskeletal protein.

Results
The membrane-binding activity of BacA depends on an N-terminal amphipathic peptide Previous work has proposed a small N-terminal peptide as the membrane-targeting sequence (MTS) of canonical bactofilin homologs (Deng et al., 2019).In the case of C. crescentus BacA, the suggested region comprises the sequence MFSKQAKS at the N-terminal end of the protein (Figure 1A).To verify the importance of this peptide for BacA function, we compared the localization pattern of wild-type BacA with that of a mutant derivative lacking the predicted membrane-targeting sequence (MTS).For this purpose, the two proteins were fused to the monomeric yellow fluorescent protein mVenus and produced in a strain carrying an in-frame deletion in the endogenous bacA and bacB genes (Figure 1B).As expected, the wildtype protein was fully functional and formed distinct fluorescent foci at the stalked pole of the cells, which have been previously shown to reflect the formation of small patches of membrane-associated BacA polymers (Kühn et al., 2010).The mutant protein (Δ2-8), by contrast, was evenly distributed in the cytoplasm, supporting a role of the N-terminal region of BacA in membrane binding.The diffuse localization also suggests a defect in polymerization, which however may be a secondary effect resulting from the loss of membrane association, since previous work has shown that the unstructured terminal regions are not essential for bactofilin assembly (Sichel et al., 2022;Vasa et al., 2015).Notably, an amino acid exchange that disrupted the C-terminal polymerization interface of the Bactofilin A/B domain (F130R) (Deng et al., 2019;Vasa et al., 2015;Zuckerman et al., 2015) also led to an even distribution of BacA-mVenus within the cell (Figure 1B), supporting the idea that polymerization and membrane binding are mutually interdependent.
Apart from transmembrane segments, proteins commonly employ amphiphilic helices to associate with the inner face of the cytoplasmic membrane, as exemplified by the bacterial actin homologs FtsA (Pichoff and Lutkenhaus, 2005) and MreB (Salje et al., 2011).As a first step to clarify the mechanism underlying the membrane-binding activity of the MTS, we therefore determined whether it was able to adopt an amphiphilic helical structure.However, neither machine-learning-based pattern recognition, as employed by AmphipaSeek (Sapay et al., 2006) (Figure 1A), nor helical wheel analysis (Figure 1C) provided any support for this possibility, suggesting a different mode of action.
To obtain more insight into the function of the MTS, we went on to systematically exchange single or multiple amino acid residues in the N-terminal eight amino acids of BacA-mVenus (Figure S1) and analyze the localization patterns of the mutant proteins.Among the single exchanges, substitution of the hydrophobic phenylalanine residue at position 2 (F2Y and F2E) led to a disperse or, occasionally, patchy distribu-tion of the fusion protein (Figure 1B).Other single substitutions, by contrast, had relatively mild effects, including the formation of multiple and/or mispositioned bactofilin complexes (Figure S2).Notably, however, a double exchange replacing the two positively charged lysine residues in the MTS with polar serine residues (K4S-K7S) strongly increased the fraction of delocalized protein (Figure 1B), although distinct fluorescent foci or patches were still detectable, indicating residual membrane-binding activity.An even stronger effect, as reflected by a complete dispersion of the fluorescence signal, was observed when negatively charged glutamate residues were introduced at these positions (K4E-K7E).Collectively, these results point to a central role of F2 and K4/K7 in the association of BacA with the cytoplasmic membrane.
To further characterize the mutant BacA-mVenus variants, we analyzed their mobility within the cell using single-particle tracking (Figure 2A).As expected by its ability to interact with the membrane and polymerize, the wild-type protein exhibited a very low diffusional mobility.Mutant variants that were partially delocalized (F2Y and K4S-K7S) were significantly more mobile, and variants with an entirely diffuse localization (Δ2-8, F2E, K4E-K7E and F2E-K4E-K7E) displayed the highest diffusion rates.These results support the notion that the MTS is required for the formation of stable membrane-associated BacA assemblies.
As a direct means to analyze the membrane-binding activity of the fusion proteins, we next performed cell fractionation studies.To this end, cells producing different BacA-mVenus variants were lysed and membranes were separated from soluble components by ultracentrifugation.Western blot analysis revealed that a large part of the wild-type protein was detected in the membrane fraction, whereas the variant lacking the MTS (Δ2-8) was completely soluble (Figure 2B).A severe reduction in membranebinding activity was also observed for all fusion proteins lacking residues F2 and K4/K7 (Figure 2C), confirming the importance of these residues in the function of the MTS.
After having identified a role of the MTS in the association of BacA with membranes in vivo, we aimed to verify the membrane-targeting activity of this peptide in a defined system in vitro.For this purpose, we purified wild-type BacA and various mutant derivatives, using a cleavable N-terminal His6-SUMO affinity tag (Marblestone et al., 2006), which allowed us to obtain proteins without any non-native extensions that could potentially interfere with the analysis.This approach also made it possible to obtain two additional BacA variants, one lacking the N-terminal methionine of BacA (ΔM) and another one lacking the Nterminal methionine and containing a glutamate instead of the phenylalanine normally located at the second position of the MTS (xE).We then incubated the different proteins with small unilamellar vesicles (liposomes) made of phosphatidylglycerol, which has been identified as the most abundant membrane lipid in C. crescentus under standard growth conditions (Contreras et al., 1978;Stankeviciute et al., 2019).
Subsequently, the liposomes were collected by ultracentrifugation and analyzed for the amount of bound protein (Figure 3).Consistent with the in vivo data, a large proportion (~80%) of the wild-type protein was associated with the liposome pellet, whereas only a small fraction sedimented in control reactions lacking liposomes, likely reflecting large, poorly soluble polymer bundles.Variants lacking the entire MTS (Δ2-8) or carrying the F2E or K4E/K7E exchanges, by contrast, showed severe defects in membrane binding, with sedimentation efficiencies similar to those obtained in the liposome-free control reactions.Residual binding was still observed for the F2Y and K4S/K7S variants, consistent with the milder defects observed for these proteins in the localization and single-particle tracking studies (compare Figures 1B and 2A).Similarly, the absence of residue M1 (ΔM) only partially abolished the membrane association of BacA, indicating that the N-terminal methionine plays an important but not decisive role in the function of the MTS (Figure 3).Together, these findings verify the existence of an N-terminal MTS in BacA.Moreover, they identify the hydrophobic residues M1 and F2 and the two positively charged lysine residues K4 and K7 as key determinants of its membrane-binding affinity.

BacA polymerization and membrane binding are mutually interdependent
Our results showed that the disruption of the polymerization interface (F130R) resulted in a diffuse, cytoplasmic localization of BacA, suggesting that polymerization might be a prerequisite for efficient membrane binding (Figure 1C).To investigate this possibility, we clarified the membrane-binding activity of the polymerization-deficient F130R variant of BacA-mVenus.Cell fractionation experiments showed that the fusion protein was barely detectable in the membrane fraction (Figure 4A).In line with this finding, only marginal membrane-binding activity was observed for the untagged BacA-F130 variant in liposome-binding assays (Figure 4B).These results suggest that the MTS has a relatively low affinity for the membrane, so that a stable interaction can only be achieved by cooperative binding of multiple MTS-containing Nterminal tails, arrayed on the surface of bactofilin polymers.Further support for this idea came from domain swapping experiments that showed that the MTS of BacA was able to mediate stable membrane attachment when fused to a derivative of the polymer-forming protein crescentin from C. crescentus lacking its native membrane-targeting sequence (Figure S3).
Interestingly, the diffuse localization and high diffusional mobility of BacA variants containing a defective MTS (Figures 1C and 2A) indicate that there may also be a converse stimulatory effect of membrane binding on BacA polymerization.To further assess this possibility, we analyzed whether an MTS-defective BacA variant could regain the ability to form membrane-associated polymeric complexes if equipped with a heterologous membrane-targeting sequence.To this end, two copies of the N-terminal amphiphilic helix of MreB from E. coli (Salje et al., 2011) were fused to a BacA-mVenus variant lacking the native MTScontaining peptide (Δ2-8).Microscopic analysis revealed that the fusion protein condensed into distinct foci at the old pole or the cell center (Figure 4C), reminiscent of the aberrant localization patterns observed for mutant proteins with exchanges in the MTS that did not completely abolish membrane binding (compare Figure S2).The restoration of fluorescent foci was accompanied by a strong decrease in diffusional mobility, as determined by single-molecule tracking (Figure 4D).In addition, cell fractionation experiments showed that the fusion protein was highly enriched in the membrane pellet (Figure 4E).Importantly, when its polymerization interface was disrupted by introduction of the F130R exchange, focus formation was abolished, membrane binding was strongly reduced, and the diffusional mobility of the fusion protein increased to a value similar to that of an F130R variant containing the native MTS of BacA (Figure 4C-E).Collectively, these findings demonstrate that the polymerization and membrane-binding activities of BacA strongly stimulate each other.

Molecular dynamics simulations identify M1, K4 and K7 as key residues of the BacA MTS
To investigate the molecular underpinnings of the interaction between the MTS of BacA and the membrane, we turned to all-atom molecular dynamics (MD) simulations.For this purpose, peptides comprising the ten N-terminal bactofilin residues were simulated in the presence of a multi-component lipid bilayer that was composed of approximately 50% monoglucosyldiglyceride (GLY), 33% phosphatidylglycerol (PG) and 16% diacylglycerol (DAG) lipids (see Table S1 for details), approximating the native composition of C. crescentus cell membranes under standard growth conditions (Chow and Schmidt, 1974;Contreras et al., 1978;Stankeviciute et al., 2019).In addition to the wild-type peptide MFSKQAKSNN, our analysis also included the mutant F2Y and K4S-K7S variants (each in a separate simulation).
The starting structures of the peptides were modeled in an extended conformation without any secondary structure.At the beginning, a single peptide (wild-type, F2Y or K4S-K7S) was placed in the aqueous phase at a distance of ca. 3 nm from the lipid headgroups.During the 500-ns simulations, all three peptides rapidly bound to the membrane and established multiple contacts with the lipids at the membrane/water interface (Figure 5A).In doing so, the peptides did not adopt any persistent secondary structures but remained disordered, with a preference for somewhat extended conformations (Figure 5B,C).Notably, in control simulations in which the peptides were initially modeled as α-helices, the peptides unfolded rapidly, either in the aqueous phase or after binding to the membrane/water interface, eventually yielding the same results (within the statistical uncertainties) as the simulations starting from an extended conformation.
Excluding the initial phase of the simulations in which the peptides were not yet bound to the membrane, we then determined the density profiles of the three different peptides along the membrane normal (Figures 5D and S4).The results show that the N-terminal residues 1 and 2 insert deeply into the lipid bilayer, assuming positions below the phosphate headgroup region, while the positions of the remaining residues (3 to 10) are gradually shifted towards the aqueous phase.The peptides thus bind to the membrane in a tilted orientation, with their N-terminal region protruding more deeply into the hydrophobic parts of the lipid headgroup region (Figure 5B).Interestingly, in case of the F2Y variant, the presence of tyrosine at position 2 pulls residue M1 and Y2 further towards the headgroup/water interface (Figure S4B).
This result can be explained by the preferential hydration of tyrosine compared to phenylalanine, as is also reflected in the difference of the solvation free energies (ΔGsolv) of the corresponding side-chain analogs p-cresol (-25.6 kJ mol -1 ) and toluene (-3.2 kJ mol -1 ) (Wolfenden et al., 1981).
To characterize the interactions between the peptides and the membrane in more detail, the contacts between the individual peptide residues and the different lipid species in the bilayers were counted.The results confirm the above finding that residues 1 and 2 show the strongest association with lipids and that lipid interactions gradually decrease for residues further down the peptide chain (Figure 6A).Notably, among these interactions, contacts with anionic PG lipids dominate, even though the percentage of PG lipids (33%) in the simulated bilayer is smaller than that of GLY lipids (50%) (see above and Table S1), which can largely be attributed to the lysine residues at positions 4 and 7.In line with this result, the K4S-K7S variant shows a strong reduction in the number of PG contacts at these positions, which is partially compensated by a higher prevalence of GLY contacts, particularly for residues 1 to 5 (Figure 6A).The F2Y variant, by contrast, did not show a marked decrease in the number of total lipid contacts for position 2, with fewer PG contacts (of residues 1 and 2) again compensated by more GLY lipid contacts (Figure 6A).
Taken together, the lipid contact analysis shows that the N-terminal residues M1 and F2 establish the closest contacts with the lipid bilayer, and that the lysine residues K4 and K7 strongly interact with PG lipids.
The contact analysis described above only provides insight into the spatial proximity of the peptides and the lipids.To additionally determine the strength of the interactions, we analyzed the MD simulations in terms of the interaction energies between the different peptide residues and the lipid molecules (Figure 6B).For both the wild-type peptide and its F2Y variant, the strongest (most favorable) peptide-lipid interaction energy is seen for residues 1, 4 and 7, which can be attributed to strong electrostatic attraction between the PG lipids and the charged NH3 + group at the N-terminus (residue M1) as well as the side chains of residues K4 and K7, respectively.Notably, out of all residues, M1 interacts most strongly with the membrane.Nevertheless, its removal (ΔM) had only a moderate effect on the membrane-binding activity of BacA in vitro (Figure 3), corroborating the notion that the largest part of the interaction energy is provided by the N-terminal NH3 + group of M1 rather than its hydrophobic side chain.However, taken together, the two lysines K4 and K7 make an even larger contribution to the interaction, consistent with the significant reduction in membrane association observed for the K4S-K7S variants of BacA (Figures 1C,    2C and 3).It is interesting to note that in terms of the interaction energy, residue F2 does not seem to play a prominent role in the peptide-membrane association (Figure 6B), although the F2Y exchange strongly reduces the membrane-binding affinity of BacA (Figures 1C, 2C and 3).The deleterious effect of this exchange may be explained by the preferential hydration of tyrosine compared to phenylalanine, which is not reflected in the peptide-membrane interaction energy but in the difference in hydration freeenergy (see above).Furthermore, residues M1 and Y2 in the F2Y peptide are located in closer proximity to the aqueous phase (see Figure S4B), which leads to a substantial decrease in the favorable interaction energy of the N-terminal NH3 + -group with PG lipids that is only incompletely compensated by slightly more favorable interactions with GLY lipids (Figure 6B).Together, these results provide a rational basis for the experimental finding that the F2Y and K4S-K7S variants of BacA have a considerably lower membrane-binding affinity than the wild-type protein.

The N-terminal membrane-targeting sequence is conserved among bactofilin homologs
The results of the MD simulations confirmed that the short peptide (MFSKQAKS) at the N-terminus of BacA acts as an MTS.To assess whether this mode of interaction was more widespread among bactofilins, we set up a bioinformatic pipeline to analyze the conservation of the N-terminal region of bactofilin homologs from a broad range of bacterial lineages (Figure S5).To this end, all proteins containing an annotated bactofilin domain were retrieved from the UniProt database (UniProt Consortium, 2023).After the exclusion of entries from non-bacterial origin, we eliminated all bactofilin homologs containing predicted Nterminal transmembrane helices as a membrane anchor.These proteins were relatively rare (~3.2%) but widely distributed across the bacterial phylogeny, with the majority found in the gammaproteobacteria (Figure S5).To account for biases in the coverage of bacterial species, we next filtered out homologs with highly similar amino acid sequences.For each of the remaining proteins, we then determined the Nterminal region preceding the bactofilin domain.Based on the length of the MTS identified for BacA, we eliminated proteins whose predicted N-terminal tails were shorter than this minimal length, potentially due to misannotation of the translational start sites.The remaining proteins (14337) were then used to search for conserved N-terminal amino acid motifs.
A global analysis of the entire protein set yielded a conserved motif with the consensus sequence MFSKKKK, which resembled the MTS identified for C. crescentus BacA and contained both the two hydrophobic N-terminal residues that insert into the core of the lipid bilayer as well as the two lysin residues that make electrostatic interactions with the headgroups of PG lipids (Figure 7).This result suggests that the presence of an N-terminal MTS is a common feature of bacterial bactofilin homologs.However, this global approach might obscure lineage-specific adaptations of the MTS to evolutionary differences in membrane lipid composition.To address this issue, we conducted an additional analysis in which we sorted the proteins according to their phylogenetic origin before performing a motif search.In doing so, we focused on the phyla Pseudomonadota, Bacteroidota, Bacillota, Spirochaetota and Thermodesulfobacteriota, which contribute the majority of the bactofilin homologs in the UniProt database.This refined analysis confirmed the presence of conserved N-terminal motifs in each of the five phyla, although there is variability in the consensus sequences of these motifs both within and between phyla (Figure S6).A comparison of the most abundant motifs (Figure 7) revealed that bactofilin homologs from Pseudomonadota and Bacillota share highly similar N-terminal sequences, which largely correspond to the global motif described above.A similar N-terminal motif is also found in most bactofilin homologs from Bacteroidota, although these proteins often display a negatively rather than a positively charged residue at position 4.
The motifs identified for Spirochaetota and Thermodesulfobacteriota, by contrast, have fundamentally different consensus sequences, likely reflecting a different mode of membrane binding (Figures 7 and S6).
Together, these results strongly suggest that the majority of bactofilin homologs known to date feature an N-terminal MTS whose composition and mode of action are similar to that of C. crescentus BacA.

BacA polymerization promotes client-protein binding
The only client protein of BacA known to date is the cell wall synthase PbpC, a bitopic membrane protein with a long N-terminal cytoplasmic tail, which was shown to contain the determinants mediating its bactofilin-dependent recruitment to the stalked cell pole of C. crescentus (Hughes et al., 2013;Kühn et al., 2010) (Figure S7A).To clarify whether the interaction with this membrane-integral client protein contributes to the association of BacA with the membrane, we compared the localization pattern of BacA-Venus in wildtype and ΔpbpC cells.Importantly, the behavior of the fusion protein was unchanged in the mutant background, suggesting that PbpC does not have a significant role in BacA polymerization or membrane binding (Figure S7B).
To further clarify the interactions that promote the formation of bactofilin-client complexes at the cytoplasmic membrane, we aimed to pinpoint the parts of PbpC that mediate its recruitment to BacA.For this purpose, we retrieved the amino acid sequences of PbpC homologs and aligned their predicted cytoplasmic tails (Figure S8).This analysis identified four distinct segments, including a highly conserved peptide at the N-terminal end of PbpC (region C1), a proline-rich region, another highly conserved peptide (Region C2) and a region rich in positively charged amino acids immediately adjacent to the transmembrane helix.
To assess the importance of these segments, we generated mVenus-tagged variants of PbpC whose cytoplasmic tail either lacked region C1 (Δ2-13) or only comprised region C1, with the remaining parts being replaced by an unstructured region taken from the periplasmic protein DipM of C. crescentus (Izquierdo-Martinez et al., 2023) (Figure S9).Microscopic analysis showed that the removal of region C1 resulted in the loss of polar localization, whereas the chimeric variant showed the same localization pattern as the wild-type fusion protein (Figure 8A).Region C1 thus appears to be necessary and sufficient for the recruitment of PbpC to the cell pole-associated BacA assembly.To verify the ability of region C1 to directly associate with BacA, we generated a synthetic peptide comprising the N-terminal 13 residues of PbpC and analyzed its interaction with purified BacA polymers using biolayer interferometry.When biosensors carrying the immobilized peptide were titrated with increasing concentrations of BacA, we observed specific binding with an apparent equilibrium dissociation constant (KD) of 4.9 µM (Figures 8B    and S10).A similar assay with the polymerization-deficient F130R variant, by contrast, only yielded residual binding with very fast dissociation rates (Figure 8C).These results indicate that BacA assembly is a prerequisite for efficient and stable PbpC binding, likely because the high local accumulation of BacA within polymers leads to an increase in the avidity for its interaction partner.
Having identified a direct association of BacA with the N-terminal peptide of PbpC, we set out to map the PbpC-binding site of BacA using hydrogen-deuterium-exchange (HDX) mass spectrometry, a method that allows the detection of local shifts in the accessibility of backbone amide hydrogens induced by conformational changes and/or ligand binding (Konermann et al., 2011).When analyzed alone, the HDX profile of BacA showed very fast HDX in the N-and C-terminal regions flanking the bactofilin domain, which indicates a high degree of disorder and thus corroborates the predicted domain structure of BacA (Figure S11).
Upon incubation with the PbpC peptide, several peptides in the C-terminal half of the central bactofilin exhibited a small, yet significant decrease in the HDX rate (Figure 8C).The most prominent changes were observed in a region close to the C-terminal polymerization interface (Figure 8D), opening the possibility that inter-subunit interactions in the bactofilin polymer could potentially induce slight conformational changes in BacA that promote PbpC binding.Together, these findings show that polymerization is not only critical for the association of BacA with the cytoplasmic membrane but also for its interaction with the client protein PbpC.

Discussion
Bactofilins are widespread among bacteria and mediate the spatial organization of diverse cellular processes.Although the polymers they form are typically associated with the inner face of the cytoplasmic membrane, the vast majority of bactofilin homologs lack transmembrane helices, suggesting the existence of an alternative membrane-targeting mechanism.Consistent with this notion, the membrane-binding activity of the bactofilin ThBac from T. thermophilus has recently been localized to the unstructured Nterminal tail of the protein, and a sequence alignment based on several well-characterized bactofilin homologs suggested the existence of a conserved N-terminal motif in this protein region (Deng et al., 2019).
Our work considerably extends these initial findings by determining the precise composition of the Nterminal MTS, unraveling its mode of action and clarifying its conservation among bacteria.Moreover, we reveal a mutual interdependence of membrane-binding and bactofilin polymerization and provide first insights into mode of interaction between the bactofilin scaffold and membrane-associated client proteins.
Mutational analysis of C. crescentus BacA identified residues F2 and K4/K7 as key components of the BacA MTS (Figure 1B) .This finding is explained by the results of our MD analysis, which show that the insertion of F2 into the hydrophobic core of the lipid bilayer (Figure 5) is critical to ensure the proper positioning of the MTS in the membrane, thereby promoting electrostatic interactions of the N-terminal NH3 + group with the negatively charged lipid headgroups of phosphatidylglycerol.The positively charged ε-amino groups of K4 and K7, by contrast, directly interact with these lipid headgroups, making the largest contribution to the overall interaction energy (Figure 6).Notably, the N-terminal methionine residue of proteins is usually removed after translation when followed by an amino acid with a short side chain (Hirel et al., 1989).In the case of BacA, however, the bulky phenylalanine residue at position 2 presumably inhibits this process, so that residue M1 may be a genuine, conserved part of the MTS.Although M1 interacts most extensively with the lipid bilayer (Figure 6), its absence has only a moderate effect on the membrane-binding activity of BacA in vitro (Figure 3).This observation supports the idea that most of its interaction energy is contributed by the N-terminal NH3 + group, whereas the insertion of the M1 side chain into the hydrophobic core of the membrane only serves to tune the positioning of the N-terminus at the lipid-water interface.Consistent with the results of the MD simulations, the loss of either F2 or K4/K7 drastically impairs the membrane-binding activity of BacA (Figures 2C and 3).Moreover, it prevents the assembly of BacA into distinct higher-order complexes (Figure 1B), suggesting a link between BacA membrane association and polymerization.Indeed, a soluble BacA variant lacking the native MTS (Δ2-8) regains the ability to form polymeric complexes when fused to a heterologous MTS from E. coli MreB (Figure 4C), indicating that membrane binding is a prerequisite for efficient BacA polymerization.Conversely, polymerization appears to promote membrane binding, because the disruption of the BacA polymerization interface (F130R) strongly increases the fraction of soluble protein (Figure 4).These results point to a mutually cooperative binding process, whereby membrane association leads to a local increase in the concentration of BacA that stimulates its self-assembly into polymers and polymerization leads to a juxtaposition of MTS-containing N-terminal tails that increases the avidity of BacA for the membrane, thus shifting the equilibrium to the membrane-bound state.
Interestingly, substitutions in the MTS that did not completely abolish BacA assembly often led to the formation of multiple and/or mislocalized complexes (Figure S2).The mechanism determining the polar localization of BacA is still unclear.However, it is conceivable that bactofilin polymers are intrinsically curved and thus preferentially associate with membrane regions of positive or negative Gaussian curvature (Kühn et al., 2010;Pöhl et al., 2024).Alternatively, the MTS may preferentially bind certain lipid species that are enriched in the polar regions of the cell, with its modification leading to a change in the lipid specificity that interferes with this recruitment pathway.Exchanges at less critical positions of the MTS may affect these recruitment mechanisms by reducing the membrane-binding affinity and thus, indirectly, also the dimensions of BacA polymers or by altering their lipid specificity.
To assess the conservation of mode of membrane association established for BacA, we conducted a comprehensive comparison of the N-terminal regions of all bactofilin homologs known to date.This analysis revealed that amino acid sequences similar to the MTS of BacA are highly prevalent in a large number of bactofilins from the phyla Pseudomonadota, Bacteroidota and Bacillota, strongly suggesting that MTSmediated membrane binding is a common theme among this group of proteins (Figure 7).Notably, however, the N-terminal motifs identified for bactofilin homologs from Spirochaetota and Thermodesulfobacteriota are fundamentally different from those of the other phyla investigated (Figures 7 and S6), suggesting a different mode of membrane binding or an adaptation of the MTS to distinct membrane compositions.
Notably, a small subset of bacterial bactofilin homologs contain an N-terminal transmembrane domain, which mediates their stable insertion into the cytoplasmic membrane.While found in a diversity of bacterial phyla, these proteins are particularly abundant in the Pseudomonadota, especially in their gammaproteobacterial and alphaproteobacterial lineages (Figure S5).One of them, the bactofilin CcmA of P. mirabilis, has been previously investigated (Hay et al., 1999).It was shown to undergoes proteolytic processing and exist in two forms, a full-length form that is an integral membrane protein and a shorter form that lacks the N-terminal transmembrane segments.Intriguingly, the shorter form has an N-terminal sequence (MFSRKTE) that corresponds to the predicted consensus sequence of the MTS in Pseudomonadota and, indeed, behaves like a typical peripheral membrane protein.The specific functional properties conferred by these two distinct modes of membrane association are still unclear.However, it is tempting to speculate that the assembly of MTS-containing bactofilins could be dynamically regulated by cell cycledependent changes in membrane curvature or composition, whereas homologs with transmembrane domains may form static polymeric assemblies.Consistent with this idea, C. crescentus BacA only assembles at the old cell pole once its cell envelope bulges outward to initiate stalk formation, thereby establishing a region of positive Gaussian curvature, although its cytoplasmic levels remain constant throughout the cell cycle (Kühn et al., 2010).Similarly, the BacA homolog of the stalked budding bacterium H. neptunium consistently assembles in regions of positive Gaussian curvature, first localizing to the stalked pole and then, once the terminal segment of the stalk starts to expand into a bud, to the bud neck (Pöhl et al., 2024).
Instead of using an MTS or a transmembrane domain, some bactofilins may also interact with the membrane indirectly by binding to other membrane-associated proteins, as recently suggested for the bactofilin homolog CcmA from H. pylori (Sichel et al., 2022).It was therefore conceivable that the recruitment of its client protein PbpC could contribute to the membrane-binding activity of C. crescentus BacA.However, we did not observe any obvious localization defects in a ΔpbpC mutant (Figure S7A), suggesting that client binding is not critical for its proper assembly and positioning.
Notably, the determinants mediating the interaction of membrane-associated bactofilin scaffolds with their client proteins are still poorly understood.For M. xanthus BacP, the binding site for its soluble interactor PadC was shown to be located in the exceptionally long unstructured C-terminal tail, which contains a conserved KKKVVVKK motif potentially involved in the interaction (Anand et al., 2020;Lin et al., 2017).However, most bactofilin homologs have much shorter C-terminal tails, which are typically poorly conserved in sequence and, at least in certain cases, dispensable for function in vivo (Sichel et al., 2022), suggesting a direct interaction of client proteins with the bactofilin core domain.Our results indeed suggest that the N-terminal unstructured region of PbpC associates with a C-terminal region of the bactofilin domain, close to the polymerization interface.Importantly, BacA assembly appears to be critical for PbpC binding in vitro (Figure 8C), suggesting that inter-subunit contacts could potentially change the conformation of BacA such as to increase its affinity for its client protein.Alternatively, polymers may have a higher avidity for PbpC than BacA monomers.Polymerization thus acts as a trigger for both mem-brane association and the recruitment of client proteins, ensuring the tight regulation of bactofilin function.
Collectively, our findings provide important new insights into the function of bactofilins and the nucleotide-independent dynamics of their assembly.It will be interesting to analyze the effect of cell cycle-or stress-induced changes in membrane composition on the membrane-binding and polymerization behavior of MTS-containing bactofilin homologs.Moreover, it will be critical to investigate the mode of membrane binding for bactofilins with conserved N-terminal sequences that are clearly distinct from the global consensus motif.

Construction of plasmids and strains
The bacterial strains and plasmids used in this work are described in Tables S2 and S3.The oligonucleotides used for their construction are listed in Table S4.All plasmids were verified by DNA sequencing.C. crescentus was transformed by electroporation (Ely, 1991).Non-replicating plasmids were integrated into the chromosome by single homologous recombination at the xylX locus (Thanbichler et al., 2007).Proper chromosomal integration was verified by colony PCR.

Live-cell imaging
To prepare samples for microscopy, overnight cultures were diluted to an OD600 of 0.1 and cultivated for 1 h prior to the addition of 0.005% D-xylose.After a further 1 h (BacA variants) or 2 h (PbpC variants) of incubation, the cultures were diluted tenfold and samples (1.5 µl) of the suspensions were spotted on 1% agarose pads prepared with ddH2O.Images were taken with an Axio Observer.Z1 microscope (Zeiss, Germany) equipped with a Plan Apochromat 100x/1.45Oil DIC, a Plan Apochromat 100x/1.4Oil Ph3 M27 objective and a pco.edge 4.2 sCMOS camera (PCO, Germany).An X-Cite® 120PC metal halide light source (EXFO, Canada) and appropriate filter cubes (ET-CFP, ET-YFP or ET-TaxasRed; Chroma, USA) were used for fluorescence detection.Images were recorded with VisiView 3.3.0.6 (Visitron Systems, Germany) and processed with Fiji 2.14.0/1.54f(Schindelin et al., 2012) and Adobe Illustrator CS6 (Adobe Systems, USA).
The subcellular distribution of fluorescence signals was analyzed with BacStalk (Hartmann et al., 2020).

Single-particle tracking
For single-particle tracking, cells were cultivated in M2G minimal media at 28 ºC.In the early exponential growth phase, the expression of genes placed under the control of the xylX (Pxyl) promoter was induced by the addition of 0.005% D-xylose.After another 3 h of incubation, cells were spotted on coverslips (25 mm diameter; Menzel Gläser, Germany) and covered with 1% agarose pads prepared with ddH2O.All coverslips were cleaned before use by sonication in 2% (v/v) Hellmanex II solution (Hellma, Germany) for 30 min at 37 ºC, followed by rinsing in distilled water and a second round of sonication in double-distilled water.Images were taken at 30-ms intervals by slimfield microscopy (Plank et al., 2009), using an Olympus IX-71 microscope equipped with a UAPON 100x/ NA 1.49 TIRF objective, a back-illuminated electronmultiplying charge-coupled device (EMCCD) iXon Ultra camera (Andor Solis, USA) in stream acquisition mode, and a LuxX 457-100 (457 nm, 100 mW) light-emitting diode laser (Omicron-Laserage Laserprodukte GmbH, Germany) as an excitation light source.The laser beam was focused onto the back focal plane and operated during image acquisition with up to 2 mW (60 W cm -2 at the image plane).Andor Solis 4.21 software was used for camera control and stream acquisition.During the first ~500 frames, most fluorescent proteins in the cell were bleached to reach the single-molecule level.Subsequently, remaining and newly synthesized molecules were tracked over ~2,500 frames.Prior to analysis, the frames recorded before reaching the single-molecule level were removed from the streams, using the photobleaching curves as a reference, and the proper pixel size (100 nm) and time increment were adjusted in the imaging metadata using Fiji (Schindelin et al., 2012).Subsequently, cell meshes were determined using Oufti (Paintdakhi et al., 2016) and single particles were tracked with u-track 2.2.0 (Jaqaman et al., 2008).After the removal of all trajectories that were shorter than five steps, the diffusional behavior of the tracked particles was analyzed using SMTracker 2.0 (Oviedo-Bocanegra et al., 2021).

Cell fractionation
Overnight cultures were diluted to an OD600 of 0.1 and grown for 2 h prior to the addition of D-xylose to a final concentration of 0.03%.The cells were further cultivated for 1 h (bacA-mVenus alleles) or 2 h (creS-mNeonGreen) to allow the expression of the genes of interest.Subsequently, cells from 5 ml culture were harvested by centrifugation, washed with 0.2 M Tris-HCl (pH 8.0) and stored at -80 ºC.For further processing, the pelleted cells were resuspended in a buffer containing 60 mM Tris-HCl (pH 8.0), 0.2 M sucrose, 0.2 M EDTA, 100 µg ml -1 phenylmethylsulfonyl fluoride (PMSF) and 10 µg ml -1 DNase I and lysed by sonication at 30% amplitude for 7.5 min with alternating 10-s on and off phases (Model 120 Sonic Dismembrator; Fisher Scientific, USA).The lysates were centrifuged at for 10 min at 4,000 ×g (4 ºC) to remove incompletely lysed cells, followed by ultracentrifugation for 1 h at 133,000 ×g (4 ºC) to separate the soluble and membrane fractions.The supernatants were carefully removed, and the pellets were resuspended in an equal volume of 0.2 M Tris-HCl (pH 8.0).Both fractions were then subjected to immunoblot analysis with a polyclonal anti-GFP antibody (Sigma, Germany; Cat.#: G1544; RRID: AB_439690), a monoclonal anti-mNeonGreen antibody (Chromotek, Germany; Cat.#: 32f6; RRID: AB_2827566), anti-MipZ antiserum (Thanbichler and Shapiro, 2006), or anti-FlgH antiserum (Mohr et al., 1996) at dilutions of 1:10,000, 1:1,000, 1:10,000 and 1:10,000 respectively.Goat anti-rabbit immunoglobulin G conjugated with horseradish peroxidase (Perkin Elmer, USA) or goat anti-mouse immuneglobulin G conjugated with horseradish peroxidase (Sigma, Germany) were used as secondary antibodies.Immunocomplexes were detected with the Western Lightning Plus-ECL chemiluminescence reagent (Perkin Elmer, USA).The signals were recorded with a ChemiDoc MP imaging system (BioRad, Germany) and analyzed using Image Lab software (BioRad, Germany).

Protein purification
To overproduce His6-SUMO-BacA or mutant variants thereof, E. coli Rosetta(DE3)/pLysS cells transformed with the appropriate plasmids were grown overnight at 37 ºC, diluted 100-fold into 3 l of LB medium containing antibiotics and cultivated to an OD600 of ~0.6.Subsequently, the media were supplemented with IPTG to a final concentration of 1 mM, and the cells were incubated for another 4 h before they were harvested by centrifugation and stored at -80 ºC until further use.To purify the fusion proteins, the pelleted cells were resuspended in buffer B (50 mM Tris/HCl pH 8.0, 300 mM NaCl, 0.1 mM EDTA, 5% glycerol and 20 mM imidazole) supplemented with 10 µg/ml DNase I and 100 µg/ml PMSF and lysed by three passages through a French press at 16,000 psi to break cells.The crude cell extract was cleared by centrifugation for 30 min at 16,000 ×rpm (4 ºC) and loaded onto a 5 ml HisTrap™ HP affinity column (GE Healthcare, USA) equilibrated with buffer B. After washing of the column with 50 ml of buffer B, protein was eluted with a 50-ml linear gradient of imidazole (20-250 mM imidazole in buffer B).Fractions that were highly enriched in the protein of interest were supplemented with 200 µg Ulp1-His6 protease (Marblestone et al., 2006) and dialyzed overnight at 4 ºC against buffer C (50 mM Tris-HCl pH 8.0, 300 mM NaCl, 0.1 mM EDTA, 5% glycerol, 1 mM DTT) to cleave off the His6-SUMO tag and remove the imidazole.Subsequently, the protein solution was again applied to a 5 ml HisTrap™ HP affinity column to separate the untagged BacA protein from His6-SUMO and Ulp1-His6.The flow-through fractions were collected and analyzed by SDS-PAGE.Fractions containing the desired protein in high amount and purity were dialyzed against buffer D (50 mM Tris-HCl pH 8.0, 200 mM NaCl, 0.1 mM EDTA, 5% glycerol) (BacAΔ2-8) or buffer E (50 mM MOPS-NaOH pH 7.0, 200 mM NaCl, 0.1 mM EDTA, 5% glycerol) (all other BacA variants).The protein solutions were then concentrated using Amicon Ultra Centrifugal Filter (10 kDa MWCO) ultrafiltration devices (Millipore, Germany) and stored in small aliquots at -80 °C.

Preparation of liposomes
Liposomes were generated from 1-palmitoyl-2-oleoyl-sn-glycero-3-phospho-(1'-rac-glycerol) (16:0-18:1 PG) (10 mg/mL in chloroform; Avanti Polar Lipids, USA).Chloroform was evaporated in a rotatory evaporator, resulting in a lipid film that was left to dry overnight.Subsequently, the lipids were resuspended in liposome buffer (50 mM MOPS-NaOH pH 7.0, 200 mM NaCl), and the mixture was incubated at room temperature for 1 h with periodic vigorous agitation.To prepare small unilamellar vesicles (SUVs) with an average diameter of 100 nm, the final lipid solution (20 mg ml -1 ) was extruded at least ten times through a Mini Extruder (Avanti Polar Lipids, USA) equipped with polycarbonate membranes of 0.1 μm pore size, until the solution was clear.

Co-sedimentation assay
Protein (20 μM) was incubated for 20 min at room temperature with or without liposomes (0.4 mg ml -1 ) in binding buffer (50 mM MOPS-NaOH pH 7.0, 300 mM NaCl) in a total reaction volume of 100 µl.The mixtures were then centrifuged for 20 min at 100,000 ×g (20 °C) in a TLA-55 rotor (Beckman Coulter, Germany).After transfer of the supernatant to a reaction tube and resuspension of the pellet in 100 μl of liposome buffer (50 mM MOPS-NaOH pH 7.0, 200 mM NaCl), samples of the two fractions were mixed with SDS-PAGE sample buffer, heated at 95 °C for 10 min and loaded onto a 15% SDS-polyacrylamide gel.
After electrophoresis, protein was stained with Coomassie Brilliant Blue R-250.The gels were imaged in a ChemiDoc MP imaging system (BioRad, Germany) using Image Lab software (BioRad, Germany), and the intensity of protein bands was quantified using Fiji 2.14.0/1.54f(Schindelin et al., 2012).

Bio-layer interferometry
Bio-layer interferometry experiments were conducted using a BLItz system equipped with Octet® High Precision Streptavidin 2.0 (SAX2) Biosensors (Satorius, Germany).In the initial step, an N-terminally biotinylated PbpCaa1-13 peptide (Biotin-Ahx-MNDWTLPPYKFDD; GenScript, USA) was immobilized on the sensor.After the establishment of a stable baseline, association reactions were monitored with BacA at various concentrations or with different BacA variants of the same concentration.At the end of each binding step, the sensor was transferred into an analyte-free buffer to measure the dissociation kinetics.The extent of non-specific binding was assessed by monitoring the interaction of the analyte with unmodified sensors.All analyses were performed in BLItz binding buffer (50 mM MOPS/NaOH pH 7.0, 100 mM NaCl, 1 mM EDTA, 5% glycerol, 10 μM BSA, 0.01 % Tween) 3.0 (Waters™, USA), using peptides that fulfilled the following criteria: minimum intensity = 5,000 counts; maximum length = 40 amino acids; minimum number of products = 2; maximum mass error = 25 ppm; retention time tolerance = 0.5 min.After automated data processing with DynamX, all spectra were manually inspected and, if necessary, peptides were omitted (e.g. in case of a low signal-to-noise ratio or the presence of overlapping peptides).

Bioinformatic analysis
Nucleotide and protein sequences were obtained from the National Center for Biotechnology Information (NCBI) or UniProt (UniProt Consortium, 2023) databases, respectively.Sequences were compared and analyzed using the blastn, blastp or PSI-blast algorithm as implemented on the NCBI website.The presence of putative amphipathic helices was assessed using AMPHIPASEEK (Sapay et al., 2006).Multiple sequence alignments were generated with MUSCLE v3.8.31 (Edgar, 2004) and viewed and edited with Jalview (Waterhouse et al., 2009).
To analyze bactofilin homologues obtained from UniProt, a Python-based pre-processing procedure was used, involving the exclusion of non-bacterial entries, the removal of sequences with predicted transmembrane helices and the elimination of sequences lacking detailed phylogenetic information.Additionally, highly similar sequences (with over 90% similarity) from the same species were filtered out to correct biases due to over-sequencing.The N-terminal region of each protein was determined using hmmscan (HMMER 3.3.2) (Eddy, 2011) with the hidden Markov profile constructed from the seed alignment of the bactofilin domain downloaded from the Pfam database (Paysan-Lafosse et al., 2022).Subsequently, sequences with N-terminal regions shorter than 8 amino acids were eliminated.The resulting set of sequences was categorized by phylum, and MEME (Bailey and Elkan, 1994) was used to detect the 10 most probable motifs.The analysis employed the following settings: classical objective function, with zero or one occurrence per sequence, motif width ranging from 5 to 50 amino acids, and a 0 th -order background Markov model.

Molecular dynamics simulations
All molecular dynamics (MD) simulations were carried out with Gromacs (version 2021.1)(Abraham et al., 2015).The simulation systems were set up and prepared for MD simulation with the Charmm-GUI (Jo et al., 2008;Jo et al., 2009;Lee et al., 2016;Wu et al., 2014),using the Charmm36m force field (Huang et al., 2017) for the peptide and lipids together with the Charmm-specific TIP3P water model.The lipid composition of the simulated bilayer (Table S1) was chosen such that it mimics as closely as possible the C. crescentus membrane (Chow and Schmidt, 1974;De Siervo and Homola, 1980).Symmetric bilayers with 128 lipids in each monolayer were constructed, that is, the same number and type of lipids was present in the two leaflets.At the beginning of the simulations, a single bactofilin 10mer peptide (either the wild-type sequence MFSKQAKSNN, or the K4SK7S or F2Y mutants) were modeled as extended conformations and placed in the bulk water phase, located at least 3 nm away from the lipid headgroups.The net charge of the simulation box was neutralized with 150 mM KCl.After energy minimization (with steepest descent), MD simulations were carried out with periodic boundary conditions in the NpT ensemble at a constant temperature of 310 K, maintained via the Bussi velocity-rescaling thermostat (Bussi et al., 2007), and constant 1 bar pressure, maintained by semi-isotropic pressure coupling of the lateral (x,y) and normal (z) dimensions of the simulation box to a weak coupling barostat (Berendsen et al., 1984).Short-range Coulomb and Lennard-Jones 6,12 interactions were described with a buffered pair list (Páll and Hess, 2013) with potentials smoothly shifted to zero at a 1.2 nm distance, with forces switched to zero between 1.0 and 1.2 nm.The long-range electrostatic interactions were described with the particle mesh Ewald (PME) method with 0.12 nm grid spacing (Darden et al., 1993).The LINCS and SETTLE algorithms were used to constrain all protein bonds involving H-atoms and to keep the water molecules rigid, respectively, allowing to integrate the equations of motion with 2 fs time steps using the leap-frog integrator.The final production MD simulations were 500 ns long.

Statistics and reproducibility
All experiments were performed at least twice independently with similar results.No data were excluded from the analyses.To quantify imaging data, multiple images were analyzed per condition.The analyses included all cells in the images or, in the case of high cell densities, all cells in a square portion of the images.The selection of the images and fields of cells analyzed was performed randomly.targeting sequence is highlighted in red.The sequence at the botton shows that result of an amphipathic helix prediction for BacA using the AMPHIPASEEK software (Sapay et al., 2006).Residues predicted to be  predicted MTS (Δ2-8).Whole-cell lysates (W) as well as the soluble (S) and pellet (P) fractions of cells producing the indicated proteins were subjected to immunoblot analysis with an anti-GFP antibody, detecting the BacA-mVenus fusion protein.As controls, the same samples were probed with antibodies raised against the soluble cell division regulator MipZ (Thanbichler and Shapiro, 2006) or the membranebound flagellar L-ring subunit FlgH (Mohr et al., 1996)      The graph shows the total number of contacts between individual residues in the wild-type, K4S-K7S and F2Y peptides and the lipid bilayer as well as the number of contacts with PG lipids and GLY lipids.A contact between a peptide residue and a lipid was defined to exist if any two non-hydrogen atoms of the residue and a lipid molecule were within a distance of 0.5 nm to each other.Contacts were counted for each frame of the MD trajectories and averaged.Multiple contacts between a peptide and a lipid molecule were treated as a single contact, so that the number of contacts counted was either 1 or 0. The statistical errors plotted were obtained from the difference between the two different sets of 500-ns simulations, starting with peptides in an unfolded or α-helical conformation, respectively.(B) Energies of the interactions between individual residues in the wild-type, K4S-K7S and F2Y peptides and the lipid bilayer.The interaction energies plotted are the combined interaction energies of all Coulomb and van-der-Waals interactions in the force field averaged over the simulation trajectories.Scale bar: 1 μm.(B) Biolayer interferometric analysis of the interaction between PbpC1-13aa and BacA.A synthetic peptide comprising the first 13 amino acids of PbpC (PbpCaa1-13) was immobilized on a biosensor and probed with increasing concentrations of BacA.After the association step, the sensor was transferred to a protein-free buffer to monitor the dissociation reaction.The graph shows a representative titration series (n=3 independent replicates).(C) Comparison of the interaction of PbpCaa1-13 with BacA and its polymerization deficient F130R variant, performed as described in panel B. (D) Mapping of the PbpC binding site on BacA by hydrogen-deuterium exchange (HDX) mass spectrometry.The plots show the extent of deuterium uptake by three representative peptides obtained after tryptic digestion of BacA protein (2.5 µM) that had been incubated in the absence or presence of PbpCaa1-13 peptide (10 µM) for the indicated time periods (see Data S1 for the full set of peptides).(E) Mapping of the the differences in deuterium uptake observed at t=1000 s onto the solid-state NMR structure of BacA (Shi et al., 2015).The graph shows the total number of contacts between individual residues in the wild-type, K4S-K7S and F2Y peptides and the lipid bilayer as well as the number of contacts with PG lipids and GLY lipids.A contact between a peptide residue and a lipid was defined to exist if any two non-hydrogen atoms of the residue and a lipid molecule were within a distance of 0.5 nm to each other.Contacts were counted for each frame of the MD trajectories and averaged.Multiple contacts between a peptide and a lipid molecule were treated as a single contact, so that the number of contacts counted was either 1 or 0. The statistical errors plotted were obtained from the difference between the two different sets of 500-ns simulations, starting with peptides in an unfolded or α-helical conformation, respectively.(B) Energies of the interactions between individual residues in the wild-type, K4S-K7S and F2Y peptides and the lipid bilayer.The interaction energies plotted are the combined interaction energies of all Coulomb and vander-Waals interactions in the force field averaged over the simulation trajectories.spectrometry.The plots show the extent of deuterium uptake by three representative peptides obtained after tryptic digestion of BacA protein (2.5 µM) that had been incubated in the absence or presence of PbpCaa1-13 peptide (10 µM) for the indicated time periods (see Data S1 for the full set of peptides).(E) Mapping of the the differences in deuterium uptake observed at t=1000 s onto the solid-state NMR structure of BacA (Shi et al., 2015).

Figure 2 .
Figure 2. Verification of residues M1, F2 and K4/K7 as critical components of the BacA MTS.(A) Mobility of the indicated BacA-mVenus fusion proteins.Shown is the average mean-squared displacement (MSD) (± SD) as a function of time, based on single-particle tracking analysis.(B) Cell fractionation experiment investigating the membrane-binding activity of BacA-mVenus (WT) or a mutant variant lacking the from C. crescentus.(C) As in panel C, but for cells producing mutant BacA-mVenus variants with single or multiple amino-acid exchanges in the predicted MTS.Shown are representative images (n=3 independent replicates).The strains used are given in the legend to Figure 1B.

Figure 3 .
Figure 3. Co-sedimentation analysis of the association of various BacA variants with liposomes.The indicated proteins (20 µM) were incubated without (-) or with (+) liposomes (0.4 mg/mL) prior to ultracentrifugation.The supernatant (S) and pellet (P) fractions of each mixture were analyzed by SDS gel electrophoresis.Shown are scans of representative gels and a quantification of the average amount of protein (± SD) contained in the different fractions (n=3 independent replicates).

Figure 4 .
Figure 4. Interdependence of BacA polymerization and membrane binding.(A) Cell fractionation experiment investigating the membrane-binding activity of the polymerization-deficient F130R variant of BacA-mVenus in vivo (LY119).The analysis was performed as described for Figure 2B.(B) Co-sedimentation analysis of the association of BacA-F130R with liposomes in vitro, performed as described for Figure 3. (C) Role of polymerization in the membrane association of a BacA-mVenus variant carrying the membranetargeting sequence of E. coli MreB.Shown are phase contrast and fluorescence images of ΔbacAB mutants (LY103, LY123) producing either a BacA-mVenus variant in which the MTS is replaced by two tandem copies of the N-terminal amphiphilic helix of E. coli MreB ( MreB WT) or a polymerization-deficient variant thereof ( MreB F130A).Demographs summarizing the single-cell fluorescence profiles obtained from random subpopulations of cells are given next to the respective fluorescence images.The numbers of cells analyzed are: MreB WT (126), MreB F130A (169).The vertical red line indicates the junction between cell body and the stalk.Scale bar: 1 μm.(D) Mobility of the indicated BacA-mVenus fusion proteins.Shown is the average mean-squared displacement (MSD) (± SD) as a function of time, based on single-particle tracking analysis.(E) Cell fractionation experiment investigating the membrane-binding activity of MreB BacA-mVenus ( MreB WT) and its polymerization-deficient F130R variant ( MreB F130R) in vivo (LY103, LY123).The analysis was performed as described for Figure 2B.

Figure 5 .
Figure 5. Molecular dynamics simulation of the interaction between the BacA MTS and a model membrane.(A) Snapshot of the molecular dynamics (MD) simulation system showing the 10-mer peptide MFSKQAKSNN (BacA1-10; red) after binding to the lipid bilayer.The water is shown in surface representation.K + and Cl -counterions are not shown.(B) Close-up view of a representative snapshot from the MD simulation visualizing the binding mode of the peptide on the membrane surface.(C) Structural overlay of 40 snapshots from the MD simulation, taken after constant time intervals from the trajectory.(D) Density profiles of individual residues in the wild-type peptide along the membrane normal, i.e. the zcomponent of the distance vector from the center-of-mass (COM) of the bilayer, with the membrane midplane located at zero.The vertical dashed black line indicates the maximum of the density distribution of the lipid headgroup phosphates.

Figure 6 .
Figure 6.Contact numbers and interaction energies for different peptide-lipid bilayer interactions.(A)

Figure 7 .
Figure 7. Conservation of the N-terminal regions of bactofilin homologs in different bacterial phyla.The pie chart in the middle shows the relative distribution of the 14337 unique bactofilin homologs analyzed among the indicated bacterial phyla.The sequence logos give the the most widespread N-terminal motifs obtained either by a global analysis of all 14337 bactofilin sequences (global consensus) or by an analysis of subsets of these sequences from specific phyla.

Figure 8 .
Figure 8. Interaction of BacA with its client protein PbpC.(A) Localization patterns of different PbpC variants.ΔbacB ΔpbpC cells producing mVenus-PbpC (LY75) or mutant variants thereof lacking region C1 (LY76) or carrying an unstructured region from C. crescentus DipM in place of the unstructured region connecting region C1 and the transmembrane helix (LY77) were analyzed by phase contrast and fluorescence microscopy.The outlines of the cells are shown in the fluorescence images.Demographs summarizing the single-cell fluorescence profiles obtained from random subpopulations of cells are given next to the respective fluorescence images.The numbers of cells analyzed are: LY75 (158), LY76 (253), LY77 (119).

Figure 2 .
Figure 2. Verification of residues M1, F2 and K4/K7 as critical components of the BacA MTS.(A) Mobility of the indicated BacA-mVenus fusion proteins.Shown is the average mean-squared displacement (MSD) (± SD) as a function of time, based on singleparticle tracking analysis.(B) Cell fractionation experiment investigating the membrane-binding activity of BacA-mVenus (WT) ora mutant variant lacking the predicted MTS (Δ2-8).Whole-cell lysates (W) as well as the soluble (S) and pellet (P) fractions of cells producing the indicated proteins were subjected to immunoblot analysis with an anti-GFP antibody, detecting the BacA-mVenus fusion protein.As controls, the same samples were probed with antibodies raised against the soluble cell division regulator MipZ(Thanbichler and Shapiro, 2006) or the membrane-bound flagellar L-ring subunit FlgH(Mohr et al., 1996) from C. crescentus.(C) As in panel C, but for cells producing mutant BacA-mVenus variants with single or multiple amino-acid exchanges in the predicted MTS.Shown are representative images (n=3 independent replicates).The strains used are given in the legend to Figure 1B.

Figure 3 .
Figure 3. Co-sedimentation analysis of the association of various BacA variants with liposomes.The indicated proteins (20 µM) were incubated without (-) or with (+) liposomes (0.4 mg/mL) prior to ultracentrifugation.The supernatant (S) and pellet (P) fractions of each mixture were analyzed by SDS gel electrophoresis.Shown are scans of representative gels and a quantification of the average amount of protein (± SD) contained in the different fractions (n=3 independent replicates).

Figure 4 .
Figure 4. Interdependence of BacA polymerization and membrane binding.(A) Cell fractionation experiment investigating the membrane-binding activity of the polymerization-deficient F130R variant of BacA-mVenus in vivo (LY119).The analysis was performed as described for Figure 2B.(B) Co-sedimentation analysis of the association of BacA-F130R with liposomes in vitro, performed as described for Figure 3. (C) Role of polymerization in the membrane association of a BacA-mVenus variant carrying the membrane-targeting sequence of E. coli MreB.Shown are phase contrast and fluorescence images of ΔbacAB mutants (LY103, LY123) producing either a BacA-mVenus variant in which the MTS is replaced by two tandem copies of the N-terminal amphiphilic helix of E. coli MreB ( MreB WT) or a polymerization-deficient variant thereof ( MreB F130A).Demographs summarizing the single-cell fluorescence profiles obtained from random subpopulations of cells are given next to the respective fluorescence images.The numbers of cells analyzed are: MreB WT (126), MreB F130A (169).The vertical red line indicates the junction between cell body and the stalk.Scale bar: 1 μm.(D) Mobility of the indicated BacA-mVenus fusion proteins.Shown is the average mean-squared displacement (MSD) (± SD) as a function of time, based on single-particle tracking analysis.(E) Cell fractionation experiment investigating the membrane-binding activity of MreB BacA-mVenus ( MreB WT) and its polymerization-deficient F130R variant ( MreB F130R) in vivo (LY103, LY123).The analysis was performed as described for Figure 2B.

Figure 5 .
Figure 5. Molecular dynamics simulation of the interaction between the BacA MTS and a model membrane.(A) Snapshot of the molecular dynamics (MD) simulation system showing the 10-mer peptide MFSKQAKSNN (BacA1-10; red) after binding to the lipid bilayer.The water is shown in surface representation.K + and Cl -counterions are not shown.(B) Close-up view of a representative snapshot from the MD simulation visualizing the binding mode of the peptide on the membrane surface.(C) Structural overlay of 40 snapshots from the MD simulation, taken after constant time intervals from the trajectory.(D) Density profiles of individual residues in the wild-type peptide along the membrane normal, i.e. the z-component of the distance vector from the center-of-mass (COM) of the bilayer, with the membrane midplane located at zero.The vertical dashed black line indicates the maximum of the density distribution of the lipid headgroup phosphates.

Figure 6 .
Figure 6.Contact numbers and interaction energies for different peptide-lipid bilayer interactions.(A) The graph shows the

Figure 7 .
Figure 7. Conservation of the N-terminal regions of bactofilin homologs in different bacterial phyla.The pie chart in the middle shows the relative distribution of the 14337 unique bactofilin homologs analyzed among the indicated bacterial phyla.The sequence logos give the the most widespread N-terminal motifs obtained either by a global analysis of all 14337 bactofilin sequences (global consensus) or by an analysis of subsets of these sequences from specific phyla.

Figure 8 .
Figure 8. Interaction of BacA with its client protein PbpC.(A) Localization patterns of different PbpC variants.ΔbacB ΔpbpC cells producing mVenus-PbpC (LY75) or mutant variants thereof lacking region C1 (LY76) or carrying an unstructured region from C. crescentus DipM in place of the unstructured region connecting region C1 and the transmembrane helix (LY77) were analyzed by phase contrast and fluorescence microscopy.The outlines of the cells are shown in the fluorescence images.Demographs summarizing the single-cell fluorescence profiles obtained from random subpopulations of cells are given next to the respective fluorescence images.The numbers of cells analyzed are: LY75 (158), LY76 (253), LY77 (119).Scale bar: 1 μm.(B) Biolayer interferometric analysis of the interaction between PbpC1-13aa and BacA.A synthetic peptide comprising the first 13 amino acids of PbpC (PbpCaa1-13) was immobilized on a biosensor and probed with increasing concentrations of BacA.After the association step, the sensor was transferred to a protein-free buffer to monitor the dissociation reaction.The graph shows a representative titration series (n=3 independent replicates).(C) Comparison of the interaction of PbpCaa1-13 with BacA and its polymerization deficient F130R variant, performed as described in panel B. (D) Mapping of the PbpC binding site on BacA by hydrogen-deuterium exchange (HDX) mass