The structure of a 15-stranded actin-like filament from Clostridium botulinum

Microfilaments (actin) and microtubules represent the extremes in eukaryotic cytoskeleton cross-sectional dimensions, raising the question of whether filament architectures are limited by protein fold. Here, we report the cryoelectron microscopy structure of a complex filament formed from 15 protofilaments of an actin-like protein. This actin-like ParM is encoded on the large pCBH Clostridium botulinum plasmid. In cross-section, the ~26 nm diameter filament comprises a central helical protofilament surrounded by intermediate and outer layers of six and eight twisted protofilaments, respectively. Alternating polarity of the layers allows for similar lateral contacts between each layer. This filament design is stiffer than the actin filament, and has likely been selected for during evolution to move large cargos. The comparable sizes of microtubule and pCBH ParM filaments indicate that larger filament architectures are not limited by the protomer fold. Instead, function appears to have been the evolutionary driving force to produce broad, complex filaments.

L arge DNA plasmids often encode segregation systems that ensure their faithful inheritance in bacteria. Functional ParMRC plasmid segregation systems comprise an actin-like polymerizing motor (ParM), a centromere-like specific DNA sequence (parC), and an adaptor protein (ParR), which attaches a plasmid, via the parC, to each end of the growing ParM filament or filament bundle 1,2 . Thus, the force generated from ParM polymerization is harnessed for plasmid segregation. Recently, structural studies have highlighted that ParMRC systems are highly plasmid specific. In particular, large variations in ParM filament geometries have been described that have differences in polarity, twist and the numbers of associating pairs of filament strands [3][4][5] . This variety of filament structures is likely required in cells that simultaneously maintain multiple plasmid types, where each unique ParM actin-like motor is dedicated to the survival of a single type, or small number of plasmid types. Clostridium botulinum often supports more than one plasmid, many of which are large and encode for neurotoxins 6,7 . Here we investigated the ParMRC cassette from pCBH, a 257 kb plasmid that carries the botulinum neurotoxin type B. This ParMRC cassette is also found on other C. botulinum plasmids, such as pCLK (267 kb) and pRSJ2_3 (245 kb), which encode neurotoxin types A and F, respectively.

Results
The pCBH ParMRC cassette. In order to determine that the putative pCBH ParMRC cassette encodes functional elements (Fig. 1a), we determined that pCBH ParM quickly assembled on addition of ATP monitored by light scattering (Fig. 1b). Phosphate release, following nucleotide hydrolysis, was measured to have delayed kinetics (Fig. 1b), and pCBH ParM disassembly was substantively slower, as observed by a gradual loss in light scattering (Fig. 1b). The critical concentration for assembly was estimated to be around 3 µM from the plot of maximum intensity values of light-scattering curves at different pCBH ParM concentrations (Fig. 1c). This compares with a similar value of 1.5-2 µM determined in vitro for the Escherichia coli R1 plasmid ParM 8 , for which the cellular concentration of ParM has been estimated to be 12-14 µM 9 . Thus, the filament assembly parameters are in line with this well-characterized segregation system. Titration of DNA fragments generated via PCR from pCBH parC with increasing levels of pCBH ParR resulted in a defined mobility shift to larger molecular size, consistent with a specific interaction between pCBH ParR and pCBH parC (Fig. 1d). Together these interactions are consistent with the identification of the pCBH ParMRC cassette as a plasmid segregation system, since the ParM polymerizes and the ParR is able to interact with parC.
CryoEM of the pCBH ParM filament. Electron microscopy (EM) of negatively stained specimens and subsequently cryoEM images indicated that the pCBH ParM filaments are substantially thicker and straighter than F-actin 10 (Fig. 2). Estimation of the persistence length of the pCBH ParM filaments from the cryoEM images is 35 μm, which compares to 11 μm for the actin filament by the same method, consistent with previous reports (10-11 μm) 11,12 . These estimations will be dependent on solution conditions, nucleotide state, and the thickness of the ice, however they indicate that the pCBH ParM filaments are substantially stiffer than actin. The pCBH ParM filaments could be imaged under a wide range of conditions including high physiological salt concentrations typically found in bacterial cells. The condition used   13,14 , indicating more than 90% of the particles are homogeneous. The 2D class averages indicated a complex filament architecture (Fig. 2d), as did the averaged Fourier transform calculated from 50 negatively stained filament images (Fig. 2b). Due to this complexity, the helical parameters were determined by cryoelectron tomography (Supplementary Fig. 1, axial rise 5.2 nm, twist −50.1°). These parameters refined to a distance 5.03 nm and twist −50.4°with the cryoEM data. Helical averaging of the cryoEM density, from each cross-section of the filament, based on these parameters led to a 4.7 Å map for the entire filament ( Supplementary Fig. 2a Fig. 3a and Supplementary Fig. 2).
pCBH ParM protomer structure. Using the E. coli R1 ParM cryoEM protofilament structure as a template, we designed a quadruple mutant of pCBH ParM that was predicted to prevent polymerization ( Supplementary Fig. 3a). Mutations to subdomain 2 (F42D, I46D) and subdomain 3 (S298D and R299D) were  expected to disrupt intra-protofilament subunit contacts, with subdomains 1 and 4. X-ray crystallography studies on the quadruple mutant ( Supplementary Fig. 3a), using the cryoEM density map (Fig. 3a) as a molecular replacement model, led to the 3.25 Å structure of the monomer (G-ParM, Fig. 3b), which was subsequently used as a guide to construct the filament protomer structure in the cryoEM density (F-ParM, Fig. 3a-c and Supplementary Fig. 3b). Two of the mutation sites (Phe42 and Arg299), from the quadruple mutant, were observed to participate in intra-protofilament subunit contacts. This accounts for the success of the quadruple mutant in the preventing filament formation and allowing for monomer crystallization. The major difference between the monomer and filament protomers was the folding of a disordered region into a helix in F-ParM (F helix, Fig. 3b, c). This region is equivalent to the DNase I binding loop in actin, which is generally disordered in G-actin and becomes ordered to form a defined extended loop in F-actin (Fig. 3d, e and Supplementary Fig. 4) [15][16][17] . The nucleotide-binding site in the pCBH G-ParM and F-ParM structures contain ADP and the magnesium ion is coordinated by Asp9, Asp195, and Gln168 equivalent to Asp11, Asp154, and Gln137 in actin, suggesting a common mechanism of ATP hydrolysis in which Gln168 (Gln137) directs the hydrolytic water towards the ATP γphosphate for nucleophilic attack ( Supplementary Fig. 4). Actin residue His161, which has also been proposed to be the catalytic base in ATP hydrolysis 18 , is not conserved in the pCBH ParM structure and has the opposite charge, Asp202 (Supplementary Fig. 4), indicating that the conserved glutamine may represent the evolutionary conserved residue important for hydrolysis 19 .
Despite the similarities to actin, pCBH ParM is a genuine ParM, since it has a closed β-barrel, a feature that is missing from actin and MreB structures (Fig. 4). Thus, at the protomer level, pCBH ParM adopts a fold that is typical for ParMs, which themselves are a subset of the actin superfamily fold.
pCBH ParM filament structure. By contrast, the pCBH ParM filament structure is far more complex in comparison to the known filament structures of actin, MreB, and other ParMs (   is completely different to the E. coli R1 plasmid ParM filament structure, in which two protofilaments tightly associate in a staggered, parallel arrangement 2 . Furthermore, two R1 ParM filaments can come together in an antiparallel, staggered architecture 2 . None of the R1 ParM inter-protofilament interactions resemble those observed in the pCBH ParM filament. Intra-strand contacts. Next, we examined the pCBH ParM filament structure to understand how a single protomer forms strands of different lengths and twists (Fig. 6e). The interactions between intra-strand protomers fall into two categories (Fig. 7a), and involve similar contact regions to actin intra-strand contacts 15 (Fig. 7b). Firstly, the central interaction region involves long-chain charged and polar residues between subdomains 2 and 4 from the lower protomer and subdomain 3 from the upper protomer ( Fig. 7a and Supplementary Fig. 3a). Secondly, hydrophobic residues on the F helix from the subdomain 2 of the lower protomer fit into the hydrophobic pocket between subdomains 1 and 3 of the upper protomer (Fig. 7a). The hydrophobic residues in this interaction have fewer rotatable bonds than in the central interaction site (Fig. 7a and Supplementary Fig. 3a). This architecture allows the hydrophobic interaction region to act like a ball-and-socket while the central interaction region is elastic due to the flexible long-chain interactions. The linker residues on either side of the F-helix are relatively flexible ( Supplementary  Fig. 5a), which adds further pliancy into the intra-strand contacts. Taken together, these features result in a malleable protofilament that can be twisted and extended to adopt the different strand conformations observed in the three-filament layers of strands (Fig. 6).
Inter-strand contacts. Finally, we inspected the pCBH ParM filament structure to determine the mechanism by which a single protomer forms many types of inter-strand interaction between strands of varying twist (Fig. 8a). The major type of inter-strand interaction occurs between complimentary basic and acidic patches on the protomers (Fig. 8b, c). Prominent in the acidic patch are two acidic loops (A1 loop and A2 loop, Figs. 3b, 5d and Supplementary Fig. 4), which adopt different positions across the basic surfaces of neighboring protomers in a lateral cross-section ( Supplementary Figs. 5 and 6), allowing for multiple types of inter-strand contacts. The A1 loop is missing from the actin structure (Figs. 3, 4 and Supplementary Fig. 4) and involved in minor contacts between the two parallel strands of the ParM R1 filament 2 . The A2 loop is extended in pCBH ParM relative to actin and ParM R1 (Fig. 4). This loop lies on the outside of the actin filament 15 and forms some secondary contacts between ParM R1 antiparallel filament doublets 2 . Thus, these loops are not used in assembling the actin filament and play minor roles in assembling ParM R1 filaments and filament doublets.
The helicity of the pCBH ParM central strand dictates the interactions between the intermediate and outer strands through twisting~50°per cross-section (compare the central strand with the blue and red protomers in Fig. 9 and Supplementary Fig. 7). In a single cross-section, the central strand forms significant antiparallel interactions with two opposing strands from the pseudo-hexagonal intermediate layer ( Supplementary Figs. 6 Fig. 9a-c). However, the strands in each layer experience the same range of contacts over the 360°twist of the central protofilament ( Supplementary Fig. 7 and Supplementary Movie 2). This design allows the 15 malleable strands to assemble into a stiff filament.

Discussion
In summary, we have demonstrated an interaction between pCBH ParR and pCBH parC and have determined the structure of the pCBH ParM polymerizing system. The characteristics of the pCBH ParMRC elements allow for a hypothesis for the mechanism of pCBH plasmid segregation in C. botulinum (Fig. 10). We propose that multiple copies of the adaptor protein ParR bind to the parC repeats on the pCBH plasmid in a circular arrangement, as determined for the pSK41 and R1 plasmids 20,21 . ParR from the R1 plasmid has been shown to bind to the barbed end of ParM monomers and associate with barbed ends of ParM filaments 22 . The pCBH ParM filament presents different numbers of barbed end protomers at the two ends of the filament, 6 and 9, which include two circular arrangements of 6 and 8 barbed end protomers. Thus, two copies of the pCBH plasmid can be expected to bind to a single elongating bipolar 15-stranded pCBH ParM filament, one at each end, via some of the copies of the circularly arranged plasmid-bound ParR. The remaining copies of plasmid-bound ParR bind to ParM monomers. Elongation of the pCBH ParM filament ends at the attachment site of the two plasmids will then proceed via the ParR-bound ParM monomers joining the filament with the release of the filament-bound ParRs. In this mechanism, the plasmids remain attached to the filament and ParM polymerization provides the force to drive the plasmids to the two extremes of the C. botulinum cell. The pointed ends of strands would elongate by incorporating free pCBH ParM to match the barbed-end strands' elongation rates, to maintain the integrity of the filament. The pCBH ParM filament appears to be highly adapted to segregating two large plasmids, which are 245-267 kb in size for plasmids containing this ParMRC system. In comparison, the E. coli R1 plasmid is 98 kb in size and is segregated by the antiparallel association of two 2-stranded ParM filaments 2,23 . The pCBH ParM "actin-like" filament has a similar number of protofilaments and diameter (26 nm) to chromosome-segregating microtubules 24 . The pCBH ParM filament is bipolar and stiff, and like microtubules, the 15 polymerizing strands are likely to exert greater combined force relative to typical two-stranded actin-like filaments.

Methods
Protein expression and purification. Constructs of ParM (AJD29063.1) and ParR (AJD29378.1) from plasmid pCBH (CP006901.1) were synthesized and cloned into pSY5, a modified pET-21d(+) (Novagen) plasmid encoding an 8-histidine tag, followed by a HRV 3C protease cleavage site, ahead of the N-terminus of the proteins 25 . Plasmids were transformed into BL21 (DE3) cells, grown to OD 600 of 0.8 and protein expression was induced with 1.0 mM IPTG overnight at 15°C. The cultures were then centrifuged at 4000×g and the cell pellets resuspended in 50 mM Tris-HCl, pH 8.0, 500 mM NaCl, 20 mM imidazole, 5% glycerol, 0.5 mg/ml lysozyme and protease inhibitor tablets (1 per 2 l culture, Roche) and lysed via sonication. The cell lysate was then clarified via centrifugation at 30,000×g for 30 min and loaded onto a HisTrap FF 5 ml (GE Healthcare) pre-equilibrated with 50 mM Tris-HCl, pH 8.0, 500 mM NaCl, and 20 mM imidazole. Following loading and washing of the column, HRV 3C protease was loaded in the same buffer for cleavage of tagged proteins over 12 h at 4°C. The cleaved protein was then eluted, pooled, and subjected to size-exclusion chromatography (Superdex 75 pg, GE Healthcare) in 40 mM HEPES, pH 7.5, 150 mM KCl, 2 mM MgCl 2 , and 1 mM DTT. Fractions were checked for purity via SDS-PAGE, and the pure fractions pooled and concentrated to between 5 and 10 mg/ml as determined by UV absorbance at 280 nm using an estimated A280 value calculated using PROTEIN CALCULATOR v3.4 (http://protcalc.sourceforge.net).
Polymerization and phosphate release assays. Assembly of pCBH ParM and the 1:0.5 ParM/ParR complex was initiated by addition of 2 mM ATP at 24°C in 20 mM HEPES, pH 7.5, 350 mM KCl, 2 mM MgCl 2 , and followed by light scattering (600 nm) at 90°using either a Perkin Elmer LS 55 spectrometer. The release of inorganic phosphate (Pi) upon nucleotide hydrolysis during pCBH ParM polymerization was measured at 24°C using the Phosphate Assay Kit (E-6646) from Molecular Probes. The absorbance at 360 nm was measured using Ultraspec 2100 pro (Amersham Biosciences).
Electrophoretic mobility shift assays. The standard reaction mixture (10 μl) for the pCBH parC/ParR experiments contained 2-320 µM of ParR, 40 mM HEPES, pH 7.5, 300 mM KCl, 2 mM MgCl 2 , 1 mM DTT, 1 mg/ml bovine serum albumin, 0.1 μg/μl sonicated salmon sperm DNA and 5% glycerol. The 5′ 6-FAM-labeled DNA fragment was generated via PCR from a pCBH parC (Supplementary Fig. 3c) encoding pUC57 plasmid synthesized by GenScript (Singapore). The primers were 5′ 6-FAM-M13-F and M13-R (Supplementary Table 3). 5′ 6-FAM is a single isomer derivative of fluorescein. The standard reaction mixture was mixed at room temperature for 10 min, followed by the addition of Sample preparation for EM. Frozen ParM protein was quickly thawed then diluted into polymerization buffer (40 mM HEPES, pH 7.4, 70 mM KCl, 7 mM MgCl 2 , 1 mM DTT) to 20 µM. ATP was then added to a final concentration of 3 mM and the reaction mixture allowed to incubate for 30 min at 25°C, to allow the filaments to form.
Transmission electron microscopy of negatively stained specimens. STEM100Cu Elastic carbon substrate grids (Ohkenshoji Co., Ltd.) were glow discharged (5 mA, 45 s) and used within one hour. 5 µl of the incubated reaction mixture was applied to the glow discharged grids and allowed to absorb for 1 min. The reaction mixture was then blotted away using filter paper and 5 µl of 1% uranyl acetate applied to the grid. Staining was allowed to proceed for 1 min before being blotted by filter paper. Prepared grids were then dried overnight in a dry box before imaging. Grids were observed with a Hitachi H-7600 electron microscope operated at 100 kV and at a nominal magnification of ×40,000. Films were scanned in 7 µm steps with a Zeiss Z/I Imaging PhotoScan 2000 scanner. Fifty filament images were extracted, unbent and averaged. A Fourier pattern was calculated. Many obvious layer lines were observed, indicating the filament structure is rigid and uniform.
Cryo-electron tomography and helical parameters determination. BSA-coated colloidal gold in the polymerization buffer (40 mM HEPES, pH 7.4, 70 mM KCl, 7 mM MgCl 2 , 1 mM DTT, and 20 µM ATP) was prepared by mixing 5 nm colloidal gold (Sigma) and BSA (Sigma) 26 . MultiA Cu 200 grids (Quantifoil) were glow discharged and used within an hour after discharging. 1.5 µl of the polymerized ParM solution and 1.5 µl of the colloidal gold solution were applied on the glow discharged grid, blotted for 5 s on the EM GP (Leica) and plunge vitrified using liquid ethane cooled by liquid nitrogen. Frozen grids were kept under liquid nitrogen for no more than 1 week before imaging. Data series for cryo-electron tomography was acquired with a Tecnai G2 Polara (FEI, Nagoya University) equipped with a FEG operated at 300 kV, CCD camera (Gatan US 4000) and an energy filter (Gatan GIF). Tilt series were recorded with 2°i ncrement from −60°to 60°, with the energy filter using a slit width of 20 eV, with −5 µm defocus and 1 e − dose per Å 2 per one image. Tomogram reconstruction was performed using IMOD 27 . Five tomograms were imaged and assembled. The tomogram with the most distinct layer line pattern ( Supplementary Fig. 2) was helically averaged in real space using different helical parameters. One helical parameter was selected (-50.1°twist/52 Å rise) which showed the best match with the averaged layer lines (Fig. 2b).
Cryo-transmission electron microscopy. R1.2/1.3 Mo400 grids (Quantifoil) were glow discharged and used within an hour. 2.5 µl of the incubated reaction mixture was applied on the glow discharged grids, blotted for 5 s on the EM GP (Leica) and plunge vitrified using liquid ethane cooled by liquid nitrogen. Frozen grids were kept under liquid nitrogen for not more than 1 week before imaging. For screening of good conditions for cryoEM imaging, grids were manually observed in a Tecnai G2 Polara (FEI, Nagoya University) equipped with a FEG operated at 300 kV and a minimal dose system. Images were captured at a nominal magnification of ×115,000 with an underfocus ranging from 1.5 to 3.5 µm and by subjecting the sample to a 2 s exposure time totaling~30 electrons per Å 2 of electron dose. Images were recorded on a GATAN US4000 CCD camera which was energy filtered by an in-column energy operated between 10 and 15 eV filter width with each pixel representing 1.8 Å at the specimen level at exposure settings. Grids for data collection were made via the Spotiton system 28,29 . For data collection, grids were semi-automatically imaged using Appion on a Titan Krios (New York Structural Biology Center) equipped with a FEG operated at 300 kV and a minimal dose system. Images were captured at a nominal magnification of ×91,000 with an underfocus ranging from −1 to −2.5 µm for 14 s of exposure time totaling~54 electrons per Å 2 of electron dose over 70 frames on a K2 Summit with each pixel representing a calibrated 1.331 Å at the specimen level.
Image analysis. 1786 raw 'movie' cryoEM data were collected and processed in RELION 2.1 13,14 . Drift correction was carried out using MotionCor2 30 , and the CTF for each micrograph calculated via CTFFIND4.1 31,32 . Filaments were manually picked with e2helixboxer from the EMAN2 software package 33 . 36,292 particles were extracted at a box size of 320 × 320 pixels. After 2D classification, 33,356 particles were selected. The helically averaged cryo-electron tomogram was used as the initial reference. Helical symmetry converged to -50.4°twist and 50.3 Å rise and the resolution reached 4.7 Å. Frames 3-42 of the raw movies were used for the final reconstruction, and the total dose was~30 electrons per Å 2 .
Strand averaging in real space. Local resolution was calculated by ResMap 34 (Supplementary Fig. 2b) and the intermediate layer showed the best resolution. To achieve higher resolution, the six strands in the intermediate layer were averaged in the real space. The region between 2.4 and 8.8 nm of the radius from the filament axis was extracted from half maps as the intermediate layer. The intermediate layer was segmented into six strands. The density for these six strands were aligned and averaged. The resolution of the averaged map (4.2 Å) was evaluated as a mean resolution by ResMap (Supplementary Fig. 2c). Persistence lengths of the pCBH ParM and actin filaments. Eighteen and 21 cryoEM filament images, which were longer than 350 nm, were randomly selected from the pCBH ParM data (Fig. 2c)  is plotted against distance x (Supplementary Fig. 8). The persistent length (L p ) was calculated by fitting Crystallization and monomer X-ray structure determination. The DNA encoding a quadruple mutant in the polymerization interfaces of pCBH ParM ( Supplementary Fig. 3a) was generated via two rounds of PCR using primers: F42D_147D_F and F42D_147D_R; and S298D_R300D_F and S298D_R300D_R (Supplementary Table 3). The resultant purified protein subjected to crystallization trials. 5 mM ATP was added to 5 mg/ml pCBH ParM mutant and incubated for 1 h on ice. Crystals were grown via hanging-drop vapor diffusion in drops containing 0.5 µl of protein/ATP and 1 µl of mother liquor (28% PEG 400, 0.1 M HEPES, pH 7.5, 150 mM MgCl 2 ) at 288 K. Native X-ray diffraction data were collected on a RAYONIX MX-300 HS CCD detector on beamline TPS 05A (NSRRC, Taiwan, ROC) controlled by BLU-ICE (version 5.1) at λ = 1.0 Å. Data were indexed, scaled, and merged in HKL2000 (version 715) 37 . Molecular replacement using the protomer cryoEM density map was carried out in the Phaser 38 followed by building in Coot 39 and refinement in PHENIX 38 . The model was built into the three-fold averaged electron density map. Group B factors and three-fold non-crystallographic symmetry constraints were maintained until the final round of refinement. The model quality was checked with the MolProbity 40 . Data collection and final refinement statistics are summarized in Supplementary Table 1.
Model building for a single strand. The missing loop (38)(39)(40)(41)(42)(43)(44)(45) in the crystal structure of pCBH ParM was modeled by Rosetta3. Two subunits of the resultant model were fitted into the averaged strand density map and then refinement into the density map was performed by Rosseta3 and then PHENIX 41 . The two subunits were subjected to symmetry restraints in the refinement.
Model building for the entire structure. Fifty-one subunits of the model for the averaged strand were fitted to three adjacent slices of the whole 4.7 Å map. Each subunit was fitted individually. The atomic models were flexibly fitted to the density map with MDFF on NAMD 2.12 software 42 . The calculation was performed under harmonic helical symmetry restraints on all atoms except hydrogens, and the solvent effect was modeled with generalized Born implicit solvent. In the MDFF, the main chain near the interface with the other subunits was allowed to move while other regions were fixed.
Reporting summary. Further information on research design is available in the Nature Research Reporting Summary linked to this article.

Data availability
The atomic coordinates and structure factors have been deposited in the Protein Data Bank under the accession code 6IXW. The coordinates and cryoEM maps for pCBH F-ParM (accession codes 6IZR and EMD-9757) and the averaged strand from the intermediate layer (accession codes 6IZV and EMD-9758) have been submitted to the Protein Data Bank and Electron Microscopy Data Bank, respectively. Other data are available from the corresponding authors upon reasonable request.