Structural and biochemical analyses of Caulobacter crescentus ParB reveal the role of its N- terminal domain in chromosome segregation

The tripartite ParA-ParB-parS complex ensures faithful chromosome segregation in the majority of bacterial species. ParB nucleates on a centromere-like parS site and spreads to neighboring DNA to form a network of protein-DNA complexes. This nucleoprotein network interacts with ParA to partition the parS locus, hence the chromosome to each daughter cell. Here, we determine the cocrystal structure of a C-terminal domain truncated ParB-parS complex from Caulobacter crescentus, and show that its N-terminal domain adopts alternate conformations. The multiple conformations of the N-terminal domain might facilitate the spreading of ParB on the chromosome. Next, using ChIPseq we show that ParBs from different bacterial species exhibit variation in their intrinsic capability for spreading, and that the N-terminal domain is a determinant of this variability. Finally, we show that the C-terminal domain of Caulobacter ParB possesses no or weak non-specific DNA-binding activity. Engineered ParB variants with enhanced non-specific DNA-binding activity condense DNA in vitro but do not spread further than wild-type in vivo. Taken all together, our results emphasize the role of the N-terminal domain in ParB spreading and faithful chromosome segregation in Caulobacter crescentus.

INTRODUCTION exhibit variation in their intrinsic capability for spreading. We discover "maxi-spreaders" (e.g. ParB from Moorella thermoacetica) that spread over ~50 kb, while "mini-spreaders" (e.g. ParB from Caulobacter crescentus) spread only ~5 kb from a single parS site. We construct a series of chimeric proteins and find that the NTD is a determinant for the inter-species variation in spreading, at least in the case of Caulobacter and Moorella ParBs. In addition, we show that the CTD of Caulobacter ParB does not display non-specific DNA-binding and DNA condensation activities in vitro. Engineered Caulobacter ParB variants with an enhanced non-specific DNA-binding activity can condense DNA in vitro but do not spread further than wild-type protein in vivo. Overall, our results emphasize the key role of the NTD in ParB spreading in Caulobacter and highlights the inter-species variation that exists within the chromosomal ParB family.

RESULTS AND DISCUSSION
Co-crystal structure of the C-terminal domain truncated ParB-parS complex from Caulobacter crescentus revealed the multiple conformations of the NTD We sought to determine a co-crystal structure of a ParB-parS complex from Caulobacter crescentus. After screening several constructs with different length of ParB and parS, we obtained crystals of a 50 amino acid C-terminally truncated ParB (Ct-ParB) in complex with a 22-bp parS duplex DNA ( Fig.  1 and Fig. 2A). In solution, Caulobacter Ct-ParB also binds to parS, albeit weaker than a full-length protein (Fig. S1). Diffraction data for the Caulobacter Ct-ParB-parS complex were collected to a resolution of 2.9 Å, and the structure was solved by molecular replacement using the 3.1 Å structure of the Helicobacter Ct-ParB-parS complex and the 2.3 Å structure of apo-Thermus Ct-ParB as search templates. X-ray crystallographic data are summarized in Table 1.
The asymmetric unit of our co-crystal contains four copies of the Ct-ParB monomer and two copies of the full-size parS DNA (Fig. S2A). Each Ct-ParB monomer binds to a half parS site via the DNAbinding domain ( Fig. 2A-B). Since chain A and B are very similar to chain C and D, respectively (RMSD= 1.59 Å, Fig. S2B), we used chain C-D-parS complex for subsequent analysis ( Fig. 2A). Each Ct-ParB monomer consists of two domains: an N-terminal domain (NTD) (helices α1-α4 and sheets β1-β3) and a parS DNA-binding domain (DBD) (helices α5-α10) ( Fig. 1 and Fig. 2B). We previously reported a 2.4 Å co-crystal structure of a Caulobacter ParB (DBD only) in complex with parS (40), here we discuss the structure of the NTD in depth. We observed that helices α3 and α4 of the NTD are packed against the DBD and are connected to the rest of the NTD via a loop in between α3 and β4 ( Fig.1 and Fig. 2B). The rest of the NTD is comprised of a four-stranded β-sheet (β1-β4) and two surrounding helices (α1-α2) (Fig. 2B). The highly conserved arginine-rich patch (G 101 ERRWR), crucial for Caulobacter ParB spreading (10), resides on helix α2 (Fig. 1). We observed that while the DBD and the NTD α3-α4 are near identical between chain C and D (RMSD=0.19 Å, Fig. 2C) the rest of the NTD (α1-β4) adopts completely different arrangement ( Fig.  2C-D). The NTD (α1-β4) of chain C and D are oriented ~80 o apart from each other (Fig. 2D); this is due to a loop (hereafter, called the elbow) that connects α3 and β4 together ( Fig. 2C-D). The role of this elbow in orientating the NTD became clearer upon comparing the Caulobacter Ct-ParB-parS structure to two other available structures of chromosomal ParBs from Helicobacter pylori and Thermus thermophilus.

Structural comparisons of the Caulobacter Ct-ParB-parS complex to other ParB family members
In the co-crystal structure of the Helicobacter Ct-ParB-parS complex, ParB adopts an open conformation in which its NTD projects outwards to contact a nearby ParB monomer (Fig. 3A, Fig.  S3A) (29). In contrast, no such interaction was seen between the NTD of the two adjacent Caulobacter ParB monomers (Fig. 3A). By superimposing the structure of Helicobacter Ct-ParB onto the Caulobacter one, we observed that each NTD has a different orientation ( Fig. 3B-C). The Helicobacter ParB NTD extends outwards (an open conformation), while the Caulobacter ParB NTD points either inwards (chain D) or side-way (chain C) (a closed conformation) ( Fig. 3B-C). Superimposition of three chains showed that the elbow (residues 121-125) swivels around the α3 axis, allowing the NTD to adopt three distinct conformations (Fig. 3C). Sequence alignment of ~1800 ParB orthologs showed an enrichment for charged and polar uncharged residues in the elbow region (Fig. 3C). This amino acid preference is typically found in intrinsically disordered proteins (41,42) and might confer flexibility to the elbow region of ParB. A further structure superimposition showed that the NTD of an apo-Thermus Ct-ParB also adopts a closed conformation, most similar to chain D in the Caulobacter Ct-ParB-parS structure ( Fig. S3C-D). Altogether, our three-way structural comparison suggests that the NTD can adopt multiple open or closed conformations regardless of whether ParB is on or off DNA. Our finding contrasts with Chen et al (2015) study which proposed that a parS-binding event induces a transition (at the NTD) from a spreading-incompetent closed conformation to a spreading-competent open conformation (29). While all currently available structures of chromosomal ParB lack the CTD, it is reasonable to assume that the NTD is also flexible in a full-length protein on/off DNA. Indeed, a co-crystal structure of a full-length SopB (a Type-I ParB protein for F-plasmid segregation) with DNA was previously solved, but only the density for the central DBD was observed (43). Schumacher et al (2010) showed that the absence of density for the NTD and CTD of SopB was due to their extreme flexibility rather than proteolysis during crystallization (43). The multiple orientations of the NTD of ParB (on/off DNA) might allow a dynamic ParB-DNA network to form inside cells.
We observed the second level of flexibility at the N-terminal-most peptide (residues 1-64) of ParB. This amino acid region is extended in the Thermus Ct-ParB structure (pink dashed line, Fig. 3D) but folds back in the Caulobacter Ct-ParB to contribute the fourth strand to the core β-sheet at the NTD (green dashed line, Fig. 3D). The equivalent region was not observed in the Helicobacter Ct-ParB structure. Due to the alternate conformations of this N-terminal-most peptide and of the NTD as a whole, the ParA-interacting region (residues 1-30, Fig. 1 and Fig. 3D) can potentially explore a very large space surrounding ParB. This flexibility might be beneficial for the network of ParB-DNA complexes to "fly-fishing" for ParA molecules in vivo (15)(16)(17)(18).
The inter-species variation in spreading among ParB orthologs is dependent on the NTD ParB orthologs are divergent in sequence, especially at their C-terminal domain (CTD) (Fig. 1). Therefore, we wondered whether ParBs from different bacterial species have distinct capacities for spreading and if it is dependent on the variable CTD. Exploiting the conservation of the parS sequence in bacteria (5), we constructed an E. coli heterologous system that allowed us to compare the spreading ability of ten chromosomal ParBs by ChIP-seq (Fig. 4A). E. coli does not possess a native ParA-ParB-parS system. We inserted a single parS site at the ygcE locus on the E. coli chromosome (Fig. 4A). Genes encoding N-terminally FLAG-tagged ParBs were codon optimized and expressed individually in E. coli (Fig. 4B). FLAG-ParBs were produced to a similar level ( Fig.  S4) before bound DNA was immunoprecipitated using an α-FLAG antibody and deep sequenced to reveal the extent of spreading from a single parS to the flanking DNA. From the ten ChIP-seq profiles, we observed that the majority of ParB (seven out of ten, including Caulobacter ParB) spread ~5 kb surrounding a single parS (Fig. 4B). On the other hand, three ParBs are "maxi-spreaders" that spread between ~20 kb to ~50 kb (Fig. 4B). In particular, Moorella thermoacetica ParB, despite its lower expression in E. coli (Fig. S4), spread ten times more extensively on DNA than Caulobacter ParB (Fig. 4B). Also, by considering only the shape of ChIP-seq profiles, we noted that ChIP signals reduced to the background more gradually for Moorella ParB than Caulobacter one. We also noted that Bacillus ParB spread only ~5 kb surrounding a single parS; this is more restrictive than a previous reported ~10 kb spreading distance when ChIP-chip of Bacillus ParB was performed in the native bacterium (25). The reason behind this discrepancy is unknown. Due to the caveat that ParB spreading has been taken out of the context of the native organism, the biological significance of the inter-species variation in ParB spreading is unclear. Nevertheless, we utilized this inter-species variation to determine the domain responsible for spreading. To do so, we constructed a series of chimeric proteins in which different regions of a "mini-spreader" Caulobacter ParB were replaced with the corresponding regions of a "maxi-spreader" Moorella ParB (Fig. 5). These chimeric proteins were produced to the same level in the E. coli ygcE::parS host (Fig. S4), and α-FLAG ChIP-seq experiments were performed to determine the extent of spreading (Fig. 5). Replacing a ParAinteracting region of Caulobacter ParB with the corresponding region from Moorella ParB produced Chimera A that spread to the same extent as a wild-type Caulobacter ParB (Fig. 5). Similarly, Chimera B that had the CTD of Caulobacter ParB replaced by an equivalent region from Moorella ParB spread to the same extent as a wild-type Caulobacter ParB (Fig. 5). However, swapping the NTD (Chimera C) or both the NTD and the DBD (Chimera D) between Caulobacter ParB and Moorella ParB produced variants that are "maxi-spreaders" i.e. having a similar extensive spreading to the wild-type Moorella ParB (Fig. 5). Taken together, our ChIP-seq profiles suggested that the NTD, at least in the case of Caulobacter and Moorella ParBs, dictates their variation in spreading.
Engineering a lysine-rich surface into the Caulobacter ParB CTD resulted in variants with non-specific DNA-binding and condensation activities in vitro In addition to the NTD, the CTD of ParB from Bacillus subtilis also contributes to the formation of the ParB-DNA network (32,36,39). Bacillus ParB was reported to bind non-specific DNA to condense both parS and non-parS DNA in vitro; these activities were mediated by a positively charged lysinerich surface on the CTD (36,39). Whether the non-specific DNA-binding activity of the Bacillus ParB CTD is a shared feature among ParB orthologs is currently unknown, furthermore, the relationship between the in vitro DNA condensation and the in vivo spreading is not fully understood. To better understand this relationship, we sought to generate variants of Caulobacter ParB with an enhanced non-specific DNA-binding activity. Unlike Bacillus ParB, the CTD of Caulobacter ParB lacks a lysinerich patch (Fig. 6A), and the wild-type protein does not bind or binds very weakly to non-specific DNA in vitro (10) (Fig. 6B). To engineer a non-specific DNA-binding activity into Caulobacter ParB, we introduced additional lysine residues into its CTD. We systematically introduced a single (1K), double (2K), triple (3K), quadruple (4K), and quintuple (5K) lysine substitutions from the Bacillus ParB CTD into equivalent positions on the CTD of Caulobacter ParB (Fig. 6A). Ten variants were purified to homogeneity (Fig. S5A) and analyzed by a quantitative bio-layer interferometry assay that directly assessed their binding to a parS and a scrambled parS DNA (i.e. non-specific DNA) (Fig. 6B). All ten tested ParB variants retained their binding activities to parS (Fig 6B). We were unable to detect any noticeable non-specific DNA-binding activity for the 1K and 2K variants (Fig. 6B). However, a further introduction of lysine residues created 3K, 4K, and 5K variants that interacted with nonspecific DNA similarly to that of Bacillus ParB (Fig. 6B).
The non-specific DNA-binding property of Bacillus ParB CTD was previously shown to condense DNA in vitro by magnetic tweezer assay (32,36,39) (Fig. 7A). To test if the engineered non-specific DNA-binding activity of Caulobacter ParB (3K-5K) variants ( Fig. 6B) also leads to DNA condensation in vitro, we performed magnetic tweezer experiments on these variants and compared their activities to that of wild-type Caulobacter and Bacillus ParBs (Fig. 7B). We performed experiments at 1 µM concentration of proteins and different forces using an identical setup and conditions described in experiments with Bacillus ParB (32,36). The extension of a tethered DNA was tracked, and any observation of a decrease in extension that was substantially larger than by the applied force alone is an indication of DNA condensation (Fig. 7A). Bacillus ParB condensed both non-parS DNA (Fig.  7B) and parS DNA (32,36). On the contrary, Caulobacter ParB (WT) did not display any noticeable in vitro DNA condensation activity with either parS or non-parS DNA substrate at the tested concentration (Fig. 7B). However, upon incubating the 3K, 4K, or 5K ParB variants with tethered DNA resulted in a decrease in the DNA extension that was much greater than that attributable to the decrease in force alone (Fig. 7B). These results indicated that introduction of three to five lysine residues to the Caulobacter ParB CTD resulted in DNA condensation in vitro.

Caulobacter ParB variants with an in vitro DNA condensation activity did not spread more extensively in vivo
We then wondered whether the enhanced DNA-condensation activity of the ParB (3K-5K) variants leads to an increase in spreading in vivo. To test this, we performed α-FLAG ChIP-seq experiments on Caulobacter strains expressing individual FLAG-tagged ParB variant in a ParB (WT)-depletable background (44). Caulobacter cells that were completely depleted of a native ParB (WT) while producing the FLAG-tagged ParB (3K-5K) variants were viable (Fig. S5B), suggesting that the additional lysine residues at the CTD did not impair chromosome segregation in Caulobacter. As controls for ChIP-seq experiments, strains expressing FLAG-tagged versions of ParB (WT), a nonspreading FLAG-ParB (R104A) mutant (10), and a non-DNA-binding protein FLAG-YFP were included (Fig. 7C). Consistent with the previous report (10), the ChIP-seq profile of a FLAG-ParB (WT) showed a clear enrichment above the background in the ~10 kb region from 4030 to 4040 kb on the chromosome (Fig. 7C). The extensive ChIP-seq profile is consistent with ParB (WT) spreading on the chromosome in vivo. This contrasts with the ChIP-seq profile of a non-spreader FLAG-ParB (R104A) in which the enrichment was confined to just ~500 bp immediately surrounding parS sites (10) (Fig. 7C). The profiles of the FLAG-ParB (3K-5K) variants were less extended than the FLAG-ParB (WT). We also noted that the overall heights of the ChIP-seq profiles of ParB (3K-5K) are lower than that of ParB (WT). It is possible that ParB (3K-5K) might bind DNA non-specifically along the chromosome, thereby titrating ParB molecules away from the parS cluster, resulting in a lower concentration of DNA-bound ParB near parS. Another possibility is ParB (3K-5K) are defective at the parS nucleation step, however this scenario is less likely since ParB (3K-5K) retained their parSbinding activities in vitro (Fig. 6B) and expressed to a comparable level to wild-type protein in vivo ( Fig. S5).
At first, we were surprised to find that Caulobacter ParB has no or a very weak non-specific DNAbinding activity in vitro since both current models for ParB-DNA network formation ("spreading and bridging" and "nucleation and caging") require some degree of interaction between ParB and nonspecific DNA (31)(32)(33)(34)36). However, while we did not observe non-specific DNA-binding activity at 1 µM Caulobacter ParB in vitro, the local concentration of ParB near parS has been estimated to reach ~500 µM (five times higher than in a crystallization drop) inside Caulobacter cells (15). At this extreme concentration, it is entirely possible that the central DNA-binding domain can provide some non-specific DNA-binding activity (Fig 8). We also noted that the five strongest Caulobacter parS sites cluster more closely (within a ~5 kb DNA segment (10)), while the four strongest Bacillus parS sites are dispersed within a ~57 kb region on the chromosome (30). A closer clustering of tightly bound ParB-parS complexes might be more effective in increasing the local concentration of Caulobacter ParB, despite its much weaker non-specific DNA-binding activity. On the contrary, the concentration of ParB in Bacillus cells is lower than in Caulobacter (~140 dimers compared to ~360 dimers per origin of replication (15)) and parS sites are more dispersed in this bacterium, in this case an added non-specific DNA-binding activity might enhance the formation of a ParB-DNA network in Bacillus.

Final perspectives
In this study, we characterize Caulobacter ParB biochemically and structurally to compare to orthologous proteins from different bacterial species. The availability of the Caulobacter Ct-ParB-parS structure, together with the structures of apo-Thermus Ct-ParB and Helicobacter Ct-ParB-parS, allows us to propose that the NTD can adopt multiple alternate conformations with respect to the DBD regardless of whether ParB is on/off DNA. The multiple conformations of the NTD might be beneficial in promoting the formation of a loose but dynamic ParB-DNA network. This is consistent with both "spreading and bridging" and "nucleation and caging" models. We further show that the NTD, at least in Caulobacter and Moorella ParB, determines how far the protein spreads on the chromosome from a single parS site. Our results emphasize the key role of the NTD in the formation of the ParB-DNA network in Caulobacter cells (Fig. 8). Our co-crystal structure lacks the CTD, hence the role of this domain is less clear in Caulobacter. In Bacillus ParB, the CTD acts both as a dimerization and DNA-binding and bridging interface via its non-specific DNA binding and condensation activities, thereby contributing to the formation of the nucleoprotein network. Here, we discover that the Caulobacter ParB CTD lacks these activities, Caulobacter CTD might mainly function as a monomer-monomer dimerization interface (37) (Fig. 8B). Taken all together, we suggest that different bacteria might fine-tune the properties of their chromosomal ParBs and there is a noticeable inter-species variation in how each domain contributes to the optimal function of ParB inside cells.

ACCESSION NUMBER
The accession number for the sequencing data reported in this paper is GSE134665. Atomic coordinates for a protein crystal structure reported in this paper were deposited in the RCSB Protein Data Bank with the accession number 6T1F. Conflict of interest statement. None declared. Values in parentheses are for the outer resolution shell.

ACKNOWLEDGMENTS
is the ith observation of reflection hkl, 〈I(hkl)〉 is the weighted average intensity for all observations i of reflection hkl and N is the number of observations of reflection hkl.
c CC½ is the correlation coefficient between symmetry equivalent intensities from random halves of the dataset.
d The dataset was split into "working" and "free" sets consisting of 95 and 5% of the data respectively. The free set was not used for refinement.

Protein overexpression and purification
Plasmid pET21b::Caulobacter crescentus Ct-ParB-(His) 6 (Table S1) was introduced into E. coli Rosetta pRARE competent cells (Novagen) by heat-shock transformation. 10 mL overnight culture was used to inoculate 4 L of LB medium + carbenicillin + chloramphenicol. Cells were grown at 37˚C with shaking at 210 rpm to an OD600 of ~0.4. The culture was then left to cool to 28˚C before isopropylβ-D-thiogalactopyranoside (IPTG) was added at a final concentration of 1 mM. The culture was left shaking for an additional 3 hours at 30 o C before cells were harvested by centrifugation. Pelleted cells were resuspended in a buffer containing 100 mM Tris-HCl pH 8.0, 300 mM NaCl, 10 mM Imidazole, 5% (v/v) glycerol, 1 µL of Benzonase nuclease (Sigma Aldrich), 0.1 g of lysozyme (Sigma Aldrich), and an EDTA-free protease inhibitor tablet (Roche). The pelleted cells were then lyzed by sonification (10 cycles of 15 s with 10 s resting on ice in between each cycle). The cell debris was removed through centrifugation at 28,000 g for 30 min and the supernatant was filtered through a 0.45 µm sterile filter (Sartorius Stedim). The protein was then loaded into a 1-mL HiTrap column (GE Healthcare) that had been equilibrated with buffer A (100 mM Tris-HCl pH 8.0, 300 mM NaCl, 10 mM Imidazole, and 5% glycerol). Protein was eluted from the column using an increasing (10 mM to 500 mM) Imidazole gradient in the same buffer. Ct-ParB-containing fractions were pooled and diluted to a conductivity of 16 mS/cm before being loaded onto a Heparin HP column (GE Healthcare) that had been equilibrated with 100 mM Tris-HCl pH 8.0, 25 mM NaCl, and 5% glycerol. Protein was eluted from the Heparin column using an increasing (25 mM to 1 M NaCl) salt gradient in the same buffer. Ct-ParB fractions were pooled and analyzed for purity by SDS-PAGE. Glycerol was then added to ParB fractions to a final volume of 10%, followed by 10 mM EDTA and 1 mM DDT. The purified Ct-ParB was subsequently aliquoted, snap frozen in liquid nitrogen, and stored at -80˚C. Ct-ParB that was used for X-ray crystallography was further polished via a gel-filtration column. To do so, purified Ct-ParB was concentrated by centrifugation in an Amicon Ultra-15 3-kDa cut-off spin filters (Merck) before being loaded into a Superdex 200 gel filtration column (GE Healthcare). The gel filtration column was pre-equilibrated with 10 mM Tris-HCl pH 8.0, 250 mM NaCl. Ct-ParB fractions were then pooled and analyzed for purity by SDS-PAGE (Fig. S1A). Full-length Caulobacter ParB-(His) 6 and other ParB variants were also purified using the same procedure ( Fig. S1A and Fig.  S5A). After optimization of an initial hit, suitable crystals were cryoprotected with 20% (v/v) glycerol and mounted in Litholoops (Molecular Dimensions) before flash-cooling by plunging into liquid nitrogen. X-ray data were recorded on beamline I04-1 at the Diamond Light Source (Oxfordshire, UK) using a Pilatus 6M-F hybrid photon counting detector (Dectris), with crystals maintained at 100 K by a Cryojet cryocooler (Oxford Instruments). Diffraction data were integrated and scaled using XDS (46) via the XIA2 expert system (47) then merged using AIMLESS (48). Data collection statistics are summarized in Table 1. The majority of the downstream analysis was performed through the CCP4i2 graphical user interface (49).  Table 1). Analysis of the likely composition of the asymmetric unit (ASU) suggested that it would contain four copies of the Ct-ParB monomers and two copies of the 22-bp parS DNA duplex (Fig. S2B), giving an estimated solvent content of ~46.6%.

Reconstitution of parS DNA
Interrogation of the Protein Data Bank with the sequence of the Caulobacter Ct-ParB revealed two suitable template structures for molecular replacement: apo-ParB from Thermus thermophilus (38) (PDB accession code: 1VZ0; 46% identity over 82% of the sequence) and Helicobacter pylori ParB bound to parS DNA (29) (PDB accession code: 4UMK; 42% identity over 75% of the sequence).
First, single subunits taken from these two entries were trimmed using SCULPTOR (50) to retain the parts of the structure that aligned with the Caulobacter Ct-ParB sequence, and then all side chains were truncated to Cβ atoms using CHAINSAW (51). Comparison of these templates revealed a completely different relationship between the N-terminal domain and the DNA-binding domain. Thus, we prepared search templates based on the individual domains rather than the subunits. The pairs of templates for each domain were then aligned and used as ensemble search models in PHASER (52). For the DNA component, an ideal B-form DNA duplex was generated in COOT (53) from a 22bp palindromic sequence of parS. A variety of protocols were attempted in PHASER (52), the best result was obtained by searching for the two DNA duplexes first, followed by four copies of the DNAbinding domain, giving a TFZ score of 10.5 at 4.5 Å resolution. We found that the placement of the DNA-binding domains with respect to the DNA duplexes was analogous to that seen in the Helicobacter Ct-ParB-parS complex. After several iterations of rebuilding in COOT and refining the model in REFMAC5 (54), it was possible to manually dock one copy of the N-terminal domain template (from 1VZ0) into weak and fragmented electron density such that it could be joined to one of the DNA-binding domains. A superposition of this more complete subunit onto the other three copies revealed that in only one of these did the N-terminal domain agree with the electron density. Inspection of the remaining unfilled electron density showed evidence for the last two missing Nterminal domains, which were also added by manual docking of the domain template (from 1VZ0). For the final stages, TLS refinement was used with a single TLS domain defined for each protein chain and for each DNA strand. The statistics of the final refined model, including validation output from MolProbity (45), are summarized in Table 1.

Generation and analysis of ChIP-seq profiles
For analysis of ChIP-seq data, Hiseq 2500 Illumina short reads (50 bp) were mapped back to the Caulobacter NA1000 reference genome (NCBI Reference Sequence: NC_011916.1) using Bowtie 1 (29) and the following command: bowtie -m 1 -n 1 --best --strata -p 4 --chunkmbs 512 NA1000bowtie --sam *.fastq > output.sam. Subsequently, the sequencing coverage at each nucleotide position was computed using BEDTools (30) using the following command: bedtools genomecov -d -ibam output.sorted.bam -g NA1000.fna > coverage_output.txt. For analysis of E. coli ChIP-seq data, the same procedure as above was applied, except that short reads were map to the reference genome of the E. coli MG1655 (NCBI Reference Sequence: NC_000913.3). Finally, ChIP-seq profiles were plotted with the x-axis representing genomic positions and the y-axis is the number of reads per base pair per million mapped reads (RPBPM) using custom R scripts.

Measurement of protein-DNA binding affinity by bio-layer interferometry (BLI)
Bio-layer interferometry experiments were conducted using a BLItz system equipped with Dip-and-Read © Streptavidin (SA) Biosensors (ForteBio). BLItz measures the wavelength shifts (binding signal or response (R), unit: nm) resulting from changes in the optical thickness of the sensor surface during association or dissociation of the analyte. The streptavidin biosensor (ForteBio) was hydrated in a low salt binding buffer (100 mM Tris-HCl pH 7.4, 100 mM NaCl, 1 mM EDTA, and 0.005% Tween 20) for 10 min. Biotinylated double-stranded DNA was immobilized onto the surface of the SA biosensor through a cycle of baseline (30 s), association (120 s), and dissociation (120 s). Briefly, the tip of the biosensor was dipped into a low salt buffer for 30 s to establish the baseline, then to 1 μM biotinylated double-stranded DNA for 120 s, and finally to a low salt binding buffer for 120 s to allow for dissociation. Biotinylated double-stranded DNA harboring parS or a scrambled parS site (i.e. non-specific DNA) were prepared by annealing a 20-bp biotinylated oligo with its unmodified complementary strand in an annealing buffer (1 mM Tris-HCl pH 8.0, 5 mM NaCl). The sequences of oligos are listed in Supplementary Table S2. The oligos mixture was heated to 98 o C for 2 min and allowed to cool down to RT overnight. After the immobilization of DNA on the sensor, association reactions were monitored at 250 nM, 500 nM, and 1 μM dimer concentration of ParB (WT) or ParB variants for 120 s. At the end of each binding step, the sensor was transferred into a protein-free low salt buffer to follow the dissociation kinetics for 120 s. The sensor was recycled by dipping in a highsalt buffer (100 mM Tris-HCl pH 7.4, 1000 mM NaCl, 1 mM EDTA, and 0.005% Tween 20) for at least 1 min to remove bound proteins. All interaction kinetics profiles (sensorgrams) recorded during BLItz experiments were analyzed using the BLItz analysis software (ForteBio). Reactions were run in triplicate for each concentration of ParB used, and the equilibrium responses were recorded and averaged. The extent of non-specific binding was assessed by monitoring the interaction of proteins with unmodified sensors and was deemed to be negligible.

Magnetic tweezer assays
Magnetic tweezer experiments were performed using a home-made setup as described previously (32,36). Briefly, images of micro meter-sized superparamagnetic beads tethered to the surface of a glass slide by DNA constructs are acquired with a 100x oil immersion objective and a CCD camera. Real-time image analysis was used to determine the spatial coordinates of beads with nm accuracy in x, y and z. A step-by-step motor located above the sample moves a pair of magnets allowing the application of stretching forces to the bead-DNA system. We used horizontally-aligned magnets coupled to an iron holder. Applied forces can be quantified from the Brownian excursions of the bead and the extension of the DNA tether. Data were acquired at 150 Hz to minimize sampling artifacts in force determination. We used horizontally-aligned magnets coupled to an iron holder to achieve force up to 15 pN.
Fabrication of DNA substrates for magnetic tweezer experiments containing a single parS sequence with biotins and digoxigenins at the tails was described previously (32).The DNA molecules were incubated with 2.8 μm streptavidin-coated beads (MyOne, Invitrogen) for 10 min. Then, the DNAbead complex was injected in a liquid cell functionalized with anti-digoxigenin antibodies (Roche) and incubated for 10 min before applying force. Torsionally constrained molecules and beads with more than a single DNA molecule were identified from its distinct rotation-extension curves and discarded for further analysis. All the experiments were performed in a reaction buffer composed of 10 mM HEPES pH 7.5, 150 mM NaCl, 3 mM EDTA, 0.1% (v/v) Tween-20 and 100 μg/ml BSA.
Force-extension curves were obtained by decreasing the applied force in steps from 15 pN to ~0.02 pN for a total measuring time of 15 min. First, we measured the force-extension response for bare DNA molecules. Then, the force was reset to 15 pN and ParB variants were flown and incubated for 2 min before starting the measurement of a force-extension curve at the same magnet positions in absence of proteins. The force applied to each bead was determined based on the force-extension data of bare DNA molecules. Bare DNA curves were fitted to the worm-like chain model and fitted values of persistence length and contour length were used as a quality control. Molecules with a large discrepancy for contour length or persistence with respect to expected parameters (45 nm persistence length, 2.1 µm contour length) were discarded from the analysis.