The C-terminal domain of ParB is critical for dynamic DNA binding and bridging interactions which condense the bacterial centromere

The ParB protein forms DNA bridging interactions around parS to form networks which condense DNA and earmark the bacterial chromosome for segregation. The mechanism underlying the formation of ParB nucleoprotein complexes is unclear. We show here that the central DNA binding domain is essential for anchoring at parS, and that this interaction is not required for DNA condensation. Structural analysis of the C-terminal domain reveals a dimer with a lysine-rich surface that binds DNA non-specifically and is essential for DNA condensation in vitro. Mutation of either the dimerisation or the DNA binding interface eliminates ParB foci formation in vivo. Moreover, the free C-terminal domain can rapidly decondense ParB networks independently of its ability to bind DNA. Our work reveals a dual role for the C-terminal domain of ParB as both a DNA binding and bridging interface, and highlights the dynamic nature of ParB networks.


INTRODUCTION
Bacterial chromosomes are actively segregated and condensed by the ParABS system and condensin (1). In Bacillus subtilis, this machinery is physically targeted to the origin proximal region of the chromosome by eight palindromic DNA sequences called parS (consensus sequence 5′-TGTTNCACGTGAAACA-3′) to which the ParB (Spo0J) protein binds (2). These nucleoprotein complexes act as a positional marker of the origin and earmark this region for segregation in a manner somewhat analogous to eukaryotic centromeres and their binding partners.
ParB is an unusual DNA binding protein. In addition to sequence-specific interactions with the parS sequence, the protein also spreads extensively around the site for about 18 kbp (2)(3)(4). The mechanistic basis for this behaviour is not well understood and a matter of active debate. Earlier models envisioned a lateral 1D spreading around parS to form a filament (3,5), principally because spreading can be inhibited in a polar manner by "roadblocks" placed to the side of parS sequences. However, ParB foci appear to contain fewer proteins than are necessary to form a filament, and single molecule analyses using direct imaging (6) and magnetic tweezers (7) have shown that binding of DNA by ParB is accompanied by condensation. These "networks" were inferred to be dynamic and poorly-ordered, consisting of several DNA loops between distally bound ParB molecules. In cells, they are presumably anchored at parS sites by sequence-specific interactions but must also contain many interactions with non-specific DNA (nsDNA), as well as self-association interactions that bridge ParB protomers to form DNA loops. Modelling suggests that a combination of 1D spreading and 3D bridging interactions can explain the condensation activity and recapitulate the polar effect of roadblocks on ParB spreading (8). Recently, single-molecule imaging of the F-plasmid SopB led to a broadly similar model, defining ParB networks as fluid structures that localise around parS using a "nucleation and caging" mechanism (9).
Despite these recent experiments converging on DNA bridging models to explain the ParB spreading phenomenon, the mechanism underpinning this behaviour remains unresolved.
In particular, the relationship between these dynamic nucleoprotein complexes and the molecular architecture of the ParB protein is unclear and is the subject of the work presented here.
The genomically-encoded ParB proteins comprise three distinct domains (Figures 1A and S1A-C). Our current understanding of their structure is limited to the N-terminal domain (NTD) which binds ParA (10-13) and the central DNA binding domain (CDBD) which binds parS and possibly also nsDNA (14, 15). A crystal structure of Thermus thermophilus ParB lacking the C-terminal domain (CTD) revealed a compact dimer in which the helix-turn-helix (HtH) motifs were symmetrically arranged in a fashion that appeared suitable for binding to the palindromic parS sequence ( Figure S1D) (14). Complementary analysis of the CTD by analytical ultracentrifugation suggested that it also formed a dimer, and it was argued that this protein interface might promote spreading interactions. More recently, a structure of Helicobacter pylori ParB, in which the protein was also truncated by removal of the CTD, showed a strikingly different conformation, where the NTD had moved away from the CDBD domain to form a tetrameric self-association interface ( Figure S1E) (16). In this structure, the CDBD was independently bound to a parS half site, and it was argued that the tetramerisation of the NTD could be responsible for bridging interactions between specific and nsDNA bound to the CDBD. This has yet to be tested experimentally.
In previous work, we hypothesised that ParB contains a second DNA binding locus for nsDNA that functions independently of the helix-turn-helix motif ( Figure 1B) (7). This idea was attractive to us for several reasons. Firstly, in a single DNA binding locus model, it is not straightforward to reconcile the strict localisation of ParB networks to just a few parS sites (and their surroundings) with the limited discrimination between specific and nsDNA binding that is observed in vitro (a <10-fold apparent difference in affinity) (7,8). Secondly, although binding to parS protects the CDBD region from proteolysis, high concentrations of nsDNA afford no such protection, implying that it binds elsewhere on the protein (7). Thirdly, the distantly-related ParB protein from plasmid P1 provides a precedent for a second DNA binding locus in a Type I centromere binding protein, and highlights the CTD as the putative candidate region (17). However, the lack of any structural information for the CTD of a genomically-encoded ParB prevents a rigorous comparison of the systems because the primary structure similarity in this region is negligible.
In this work, we have probed the role of the CDBD and CTD of B. subtilis ParB using a combination of structural, biochemical, single molecule and in vivo approaches. We find that the CDBD is responsible for specific recognition of parS, and that the CTD provides both a second nsDNA binding site and a self-association interface that promotes bridging interactions and DNA condensation.

RESULTS
Conserved helix-turn-helix motifs in the CDBD are essential for parS recognition, but dispensable for non-specific DNA binding and condensation Genetic and structural analyses have suggested that residue R149 may be critically important for specific binding to parS at the HtH locus (6,16,18,19). To probe the role of the HtH motif using biochemical techniques, we compared binding of parS by wild type ParB and ParB R149G using electrophoretic mobility shift assays containing Mg 2+ cations (TBM-EMSA). As reported previously, inclusion of divalent cations in both the gel composition and running buffer enables the clear differentiation of specific and nsDNA-binding activities of ParB (7). As expected, binding of wild type ParB to parS-containing DNA produced a distinct band shift corresponding to the ParB2-parS complex, as well as poorly migrating species at high [ParB] (Figure 2A). These latter complexes are also formed on DNA that does not contain parS, and are therefore indicative of ParB bound to nsDNA flanking the central parS sequence. ParB, and mutants thereof, were purified to homogeneity ( Figure S2A). EMSA experiments with ParB R149G fail to produce the specific ParB2-parS complex whereas the formation of nsDNA complexes is largely unaffected (Figures 2A). The retention of nsDNA binding activity in ParB R149G is further supported by data using gels lacking Mg 2+ ions (TBE-EMSA) ( Figure S2C), as well as a solution-based protein-induced fluorescence enhancement (PIFE) assay ( Figure S2B), in which an increase in Cy3 intensity reports ParB binding. For wild type ParB, the data were fitted to the Hill equation yielding an apparent Kd of 361 ± 14 nM and Hill coefficient of 3.2 ± 0.3 in reasonable agreement with published data (7,20). ParB R149G produced a similar binding isotherm yielding a moderately weaker Kd of 493 ± 18 nM. This apparent Kd was not significantly altered when the Hill coefficient was not shared between datasets indicating the cooperativity of binding was not impaired in this ParB variant.
We next investigated the ability of ParB R149G to condense DNA tethers using magnetic tweezers ( Figure 2B). We previously showed that wild type ParB mediates progressive condensation of DNA substrates which is reversible by both force and protein unbinding (7).
The condensed state is not highly ordered and its formation is not dependent upon parS sequences, indicating that nsDNA binding is sufficient for condensation. At a concentration sufficient for efficient condensation by wild type ParB (1 µM), ParB R149G did not fully condense DNA, although fluctuations of the DNA tether were consistent with minor condensation events that do not greatly affect the mean extension value measured (data not shown). However, at moderately elevated concentrations (3-fold), reversible condensation did occur and was qualitatively equivalent to wild type behaviour ( Figures 2C   and S2D).
Together, these data show that mutation of the HtH motif effectively eliminates the ability of ParB to interact specifically with its cognate parS site, while nsDNA binding and condensation is only moderately affected (2-3 fold reduced). This is consistent with the idea that nsDNA binding may occur at a second DNA binding locus.
The structure of BsParB CTD reveals a dimer with a putative DNA binding interface We next used solution NMR to determine the structure of the ParB CTD alone (see Table   S1 for structure validation and statistics and Figure S3A for an assigned 1 H-15 N HSQC spectrum). The structure forms a well-defined dimer containing two α-helices and two βstrands per monomer in a α1-β1-β2-α2 arrangement ( Figures 3A-B). The dimer interface is formed via an intermolecular β-sheet and two domain-swapped C-terminal helices. This is somewhat similar to that seen in the P1 and SopB ParB proteins (Figures S3B-D) (15, 17), but there are also significant differences especially in the N-terminal region: the α1 helix in our structure is replaced by an additional β-strand in the CTDs of P1 ParB and SopB.
Analytical ultracentrifugation, native mass spectrometry and circular dichroism thermal melt scans further confirmed that the CTD was primarily dimeric in solution and measured a Tm of 68ºC ( Figures S3E-F). NMR H-D exchange data revealed that the dimer is fully exchangeable (which will become important in later experiments) with the most stable Hbonds being those in the α2 helix and the intermolecular H-bonds between the two β2 strands (data not shown). These secondary structure elements are at the centre of the hydrophobic core which is made up of several Ile, Val and Phe residues in the β-sheet and several Ile/Leu residues in the α2 helix ( Figure 3C). The α1 helix forms a leucine zipper with the α2 helix, where alternating Leu residues interdigitate ( Figure 3D). A striking feature of the structure is a highly electropositive face of the dimer arising from several conserved Lys residues (Figures 3E-F) analogous to the plasmid-encoded SopB and P1 ParB proteins ( Figure S3G-I).
The CTD binds DNA non-specifically via a lysine-rich surface To test the idea that the lysine-rich surface we had observed might bind to DNA, we performed TBE-EMSAs with the isolated CTD. These showed that the CTD was indeed able to bind dsDNA ( Figure 4Ai) resulting in the formation of a "ladder" of bands of decreasing mobility. This is highly reminiscent of patterns formed by full length ParB under the same conditions ( Figure 4Aii and (7)) except for the presence of smaller gaps between the "rungs" as would be expected for a protein of much smaller size. The CTD was also shown to bind to hairpin oligonucleotides as short as 10 bp and to ssDNA ( Figure S4A and data not shown).
We do not see substantial differences in the affinity of ParB for DNA substrates with different sequences and so this binding activity appears to be non-specific (data not shown). Native mass spectrometry of complexes formed between the CTD and a 15 bp duplex DNA revealed a stoichiometry of 1 DNA per dimer ( Figure 4D). This is in contrast to the P1 ParB system where the CTD can bind two 16-mers (17).
To further probe the putative DNA binding surface, we performed a titration of the 10 bp hairpin DNA against the isotopically-labelled CTD dimer ( Figure 4E, assigned 1 H-15 N HSQC spectra are shown in Figure S4B). Residues with large chemical shift perturbations (CSPs, Δδ >0.08) are either directly involved in DNA-binding or undergo a conformational change as an indirect result of DNA-binding, and these were mapped onto the structure ( Figure 4F).
Two regions of interest were identified: D231-V233, and K252-K259, which are found on the intermolecular β-sheet face and proximal loop regions to form a large, concave and positively-charged interaction surface ( Figure 4G).
To confirm that this surface was responsible for DNA binding we substituted several Lys residues with Ala and monitored the effect on DNA binding using EMSA and PIFE assays.
In the first instance, a dual K255A/K257A substitution was studied in the context of both the CTD-only construct (CTD KK ) and the full length ParB protein (ParB KK ). CTD KK displayed a greatly reduced affinity (~50-fold) for DNA, but the binding was not completely abolished ( Figure 4Bi). CD thermal melt analysis confirmed that this defect was not attributable to global misfolding ( Figure S4C). The full length ParB KK variant showed a sigmoidal DNA binding isotherm in a PIFE assay, indicating strong positive cooperativity as observed for wild type ParB but the apparent Kd was 6-fold weaker ( Figure S4D). In EMSA assays, this variant showed no defect in specific binding to parS as would be expected ( Figure S4E).
Somewhat more surprisingly however, these lysine substitutions appeared to have a negligible effect on nsDNA binding when assessed using the TBE-EMSA assays ( Figure   4Bii). This may well reflect the complexity that arises when a partially defective nsDNA binding locus is physically attached to the wild type CDBD domain (which is still fully competent to bind DNA).
We next designed a triple K252A/K255A/K259A variant with the aim of fully dissipating the positive charge density across the surface of the CTD, rather than only targeting the loopproximal regions. EMSA analysis showed that DNA binding was completely abolished in CTD KKK up to concentrations of 50 µM (Figures 4Ci and S4F). CD thermal melt analysis showed that CTD KKK was equivalently folded to wild type CTD at ambient temperatures, but with a reduced Tm (53ºC) indicating a moderate destabilising effect of the mutations ( Figure   S4G). Interestingly, analysis of full length ParB KKK showed a clear and consistent defect in all nsDNA binding assays used. TBM-EMSA gels showed that ParB-parS complexes were still formed (albeit in apparently lower yield, Figure S4I), whereas TBE-EMSA gels showed a complete eradication of the discrete lower mobility bands which arise from nsDNA binding (Figures 4Cii). Moreover, nsDNA binding was undetectable using the PIFE analysis ( Figure   S4H). Interestingly, EMSA analysis showed that DNA-bound ParB KKK networks do still form as very low mobility species that form co-operatively at high ParB concentrations. This property might reflect the retention of a functional HtH motif in the ParB networks.

DNA binding by the CTD is essential for DNA condensation and bridging in vitro
We next exploited our double-and triple-lysine mutant ParB proteins to test the role of the DNA-binding activity associated with the CTD in forming condensed ParB networks. These networks have been extensively characterised previously for wild type ParB using magnetic tweezers with single tethered DNA substrates (7) and also in TIRF-based microscopy (6).
Unlike full length ParB, the CTD was not capable of condensing DNA tethers under any condition tested, even up to 5 µM CTD2 concentrations and under applied forces as low as 0.02 pN ( Figure S5Ai). This is consistent with the expected requirement for multiple protein-protein and/or protein-DNA interfaces to promote DNA looping and condensation.
Incubation of full length ParB KK with single DNA tethers resulted in defective DNA condensation compared to wild type ParB ( Figure 5A). When it was observed, condensation was sudden (rather than progressive, as for wild type ParB) and full condensation required the applied force to be dropped to an exceptionally low value (0.09 pN) ( Figure S5Aii). The DNA molecules also showed unusually large steps when decondensed by force, suggesting that ParB KK was infrequently stabilising in cis DNA-bridging interactions between isolated DNA regions (data not shown). Co-incubation of ParB KKK  The CTD can both inhibit the formation of, and decondense, ParB-DNA networks in vitro The CTD potentially acts as both an oligomerisation interface and also a site of nsDNAbinding. Therefore, we hypothesised that the CTD might have a dominant negative effect on full length ParB by competing for the DNA and protein interfaces that mediate the formation of ParB networks in the magnetic tweezers.
Purified CTD completely inhibited the formation of the condensed state if pre-incubated with wild type ParB and DNA under the high stretching force regime ( Figures 6A and S6A). We also tested whether the introduction of free CTD to pre-condensed tethers was able to disrupt ParB-DNA networks. Condensed ParB networks were completely stable in a flow of free ParB on the timescale of these experiments, and the DNA tethers were also able to recondense following force-induced decondensation ( Figure 6Bi). However, the inclusion of excess free CTD rapidly disrupted ParB networks, with some degree of decondensation observed in 94% of all the molecules tested ( Figure 6Bii). Moreover, those molecules which did not decondense spontaneously could be stretched by force, but were then unable to recondense when permissive forces were restored. This ability of the CTD to decondense ParB networks demonstrates that the protein-protein and/or protein:DNA interfaces that maintain the condensed state under a low force regime are dynamic (i.e. they are exchanging while the overall structure of the network is maintained).
We have shown above that the CTD binds tightly to nsDNA. Therefore, its ability to prevent condensation and induce decondensation might simply reflect competition for the nsDNA that becomes available during exchange of ParB:DNA interfaces. Indeed, we have shown previously that free DNA is a potent inducer of network decondensation in the MT apparatus (7). To test the idea that the CTD dimerisation interface is also important for maintaining the condensed state, we repeated our experiments with the CTD KK and CTD KKK constructs, which are defective and apparently unable (respectively) to bind nsDNA. Both mutant proteins were as effective as wild type in preventing condensation ( Figures 6C and S6B), and both were able to induce decondensation in approximately 95% of all molecules tested (Figures 6Bii and 6D). This strongly suggests that CTD-dependent ParB network dissipation is primarily mediated by competition for the CTD dimerisation interface and further confirms that the CTD KKK construct is folded. This competition presumably results from the formation of heteroligomers between full length ParB and the CTD, which disrupts interactions that are essential for condensation ( Figure S6C).

The CTD is critical for the formation of ParB foci in vivo
To test the importance of the CTD dimerisation and DNA binding interfaces in vivo, we compared the ability of wild type and mutant ParB-GFP proteins to form foci in B. subtilis cells when expressed from the endogenous locus. Wild type ParB-GFP formed discrete foci around oriC as expected (3,6,18,(21)(22)(23)(24)(25). In contrast, ParB KKK -GFP failed to form discrete foci (Figure 7Ai-ii) despite wild type expression ( Figure S7A). Interestingly, the triple-mutant protein appeared to localise non-specifically to the nucleoid, perhaps as a result of residual DNA binding by the HtH motifs, suggesting that ParB KKK -GFP retained the ability to dimerise.
A ParB L270D+L274D construct, designed to prevent dimerisation of the CTD, was completely unable to form ParB foci (Figure 7Aiii-v) despite being expressed at approximately wild type levels ( Figure S7B). The complete deletion of the CTD by truncation to E222 or E227 resulted in the same phenotype (data not shown).
Our attempts to purify recombinant ParB L270D+L274D failed because the protein was insoluble upon overexpression in E. coli. This raises the possible caveat that the loss of function associated with this dimerisation mutant in vivo might reflect mis-folding. Therefore, we also investigated whether the free CTD was able to interfere with dimerisation in vivo, thereby causing a dominant negative effect on ParB function. A B. subtilis strain was engineered with a C-terminal gfp fusion replacing the endogenous spo0J and the unlabelled CTD-only gene inserted at an ectopic locus downstream of a Phyperspank promoter, designed for high protein expression that is tightly-controlled with IPTG ( Figure 7B). Overexpression of the CTD caused ParB-GFP foci to become diffuse ( Figure 7C), although expression levels of endogenous ParB-GFP were unaffected ( Figure S7C).
ChIP-qPCR analysis allowed us to more directly characterise the effect of CTD expression upon ParB spreading (i.e. the enrichment of ParB at and widely around parS sites).
Spreading was measured around a single parS site (359.20º) and used a locus towards the terminus to monitor background 'enrichment' (146.52º) ( Figure 7D). As expected, in the absence of CTD expression, ParB was highly enriched not only at parS sites (~40-fold), but also for several kilobase pairs around parS ( Figure 7E). Overexpression of CTD significantly decreased the signal around parS (~4-fold enrichment), indicating that it interferes with spreading. Western blotting of cells grown under equivalent conditions to the ChIP-qPCR assay and using the same batch of polyclonal anti-ParB antibody suggests that CTD is not preferentially recognised over the endogenous ParB copy ( Figure S7D). Note that the reduced signal observed for the parS fragment does not necessarily indicate defective specific binding because the PCR product at parS is much larger than the 16 bp parS site or the 24 bp footprint of a ParB dimer (3).
Finally, we determined the consequence of decreased ParB spreading in vivo induced by CTD overexpression by measuring the rate of DNA replication initiation. ParB normally inhibits the activity of Soj, a regulator of the master bacterial initiation protein DnaA (26).
Marker frequency analysis showed that CTD overexpression stimulated the frequency of DNA replication initiation, indicating that regulation of DnaA by Soj was adversely affected ( Figure S7E). Together, these results are consistent with our in vitro observations, and support a model in which dynamic ParB-DNA networks are dependent upon ParB oligomerisation and DNA-binding interfaces in the CTD.

DISCUSSION
ParB proteins form long-distance bridging interactions on DNA substrates to form foci that facilitate chromosomal partitioning reactions (4,5,27). These ParB foci are anchored at parS sites and interact non-specifically around a single site for ± 18 kbp (2, 28). This ParB "spreading" activity appears to be a conserved property across chromosomal and plasmid segrosomes, yet the protein interaction interfaces involved have remained elusive, particularly for the genomically-encoded systems (6). This is, in part, due to the highly variable structures of ParB proteins and their cognate centromere sequences, even within the type I subclass of which B. subtilis ParB is a member (29, 30). Increasing evidence indicates that ParB spreading is the result of a DNA-bridging activity mediated by ParB-ParB oligomerisation interfaces (6-9, 31). However, a complete understanding of the relationship between ParB structure and function has been hindered by the lack of any full length structure for a chromosomally-encoded ParB. Indeed, the organisation of the N-terminal (NTD), central DNA-binding (CDBD) and C-terminal (CTD) domains appears to be quite complex ( Figures S1A-C). For the type I ParB protein class, there is evidence to suggest that dimerisation and/or tetramerisation can occur at the NTD and CDBD, and that dimerisation can occur at the CTD (14-16). A combination of some or all of these activities must support ParB oligomerisation.
To directly address the putative role of the CTD in spreading (14), we resolved the first structure of a genomic ParB CTD. The structure revealed a conserved lysine-rich surface and we showed that this binds to DNA in an apparently non-specific manner. This novel DNA binding locus is structurally distinct from the sequence-specific DNA binding site for Our CTD structure facilitated the design of separation of function mutations to test the importance of the dimerisation and DNA binding activities using a variety of in vitro and in vivo readouts of ParB function. We showed that the CTD is not required for parS binding, and that this is instead dependent on the HtH motif found within the CDBD domain as predicted in several previous studies (6,16,18,19,29,(34)(35)(36). In contrast, the CTD is essential for the formation of nsDNA complexes that are observed as ladders of decreasing mobility in EMSA assays. Mutant proteins that were unable to bind DNA at the CTD locus were severely defective in both DNA condensation assays in vitro and ParB foci formation assays in vivo. Moreover, ParB proteins that were designed to be unable to form oligomeric structures by mutation of the CTD-CTD dimerisation interface were completely unable to form ParB foci and the free CTD domain was able to disrupt ParB networks independently of its ability to bind DNA. Together, our observations strongly support the idea that the CTD is essential for both DNA binding and for ParB-ParB bridging interactions that support DNA condensation.
We propose that the presence of two DNA binding loci in ParB can help to explain how ParB networks are anchored at parS in vivo. Importantly, this architecture resolves the paradoxical observation that the apparent specificity for parS in vitro (<10-fold greater affinity for parS versus nsDNA) is insufficient to explain the strict localisation of ParB around just 8 sites in a ~4 Mbp genome (7,8). In a two DNA binding site model, specific and non-specific binding can be semi-independent activities that are architecturally-coupled only when ParB oligomerises into networks. This model can also explain why DNA condensation does not require parS in vitro, whereas the absence of parS sites prevents the formation of ParB-DNA foci in vivo ( (6,7,9,37) and this work). In a test tube, whenever ParB is present at concentrations that licence oligomerisation, it is always in large stoichiometric excess over binding sites and all available DNA will be bound. In cells, the situation is very different because there is a limited pool of ParB (6). Specific interaction with parS preferentially anchors the ParB network at parS, leaving a vast number of unoccupied sites. If parS sites are absent in cells, ParB might still form networks, but these would not be anchored at specific sites and would therefore fail to form foci, as has been observed experimentally (37, 38). A rigorous proof of these ideas will require a modelling approach that will be the subject of future work.
Previously, high-resolution SIM and ChIP-seq data have suggested that ParB-DNA partition complexes involve stochastic and dynamic binding of ParB to both DNA and other ParB proteins, resulting in the formation of fluid intra-nucleoid "ParB cages" on DNA (9). This view is consistent with the disorder observed in magnetic tweezers assays (7), and with the dominant negative effect of the free CTD domain on ParB networks shown here. However, a recent structural study of H. pylori ParB concluded that a novel tetramerisation interface within the NTD was also likely to be important in bridging (16). Moreover, spreading could be facilitated by parS-dependent conformational changes that act as nucleation points for networks (8, 14). A more complete understanding of ParB network formation and its regulation will be required to underpin future studies on how ParB acts together with ParA and condensin to orchestrate efficient chromosome segregation.

ParB overexpression and purification
ParB, and the variants R149G, K255A+K257A and K252A+K255A+K259A, were overexpressed and purified as described (7). CTD, and mutants thereof, were His-tagged and purified to homogeneity as follows. Cell pellets, produced as described (7) For structure determination by NMR, the CTD was dual isotopically (13C and 15N) labelled during overexpression in M9 media, as described previously (46), and subsequently purified as above. structures from this iteration went on to be water refined. Spin diffusion correction was used during all iterations (52). Two cooling phases, each with 30,000 steps were used. Torsion angle restraints were calculated using TALOS+. Standard ARIA symmetry restraints for 2 monomers with C2 symmetry were included (53). Structural rules were enabled, using the secondary structure predictions made by TALOS+. The HD exchange experiment showed 29 NH groups to be protected after 8 mins. Initial structure calculations were conducted without hydrogen bond restraints. Hydrogen bond donors were then identified and corresponding hydrogen bond restraints included in later calculations. Calculations were conducted using a flat-bottom harmonic wall energy potential for the distance restraints until no consistent violations above 0.1 Å were observed. The final calculation was then performed using a log-harmonic potential (54) with a softened force-field (55). Structures were validated using the Protein Structure Validation Software (PSVS) suite 1.5 (56) and CING (57). The chemical shifts, restraints and structural co-ordinates have been deposited with the BMRB and PDB for release upon publication.

EMSA experiments
The specific and nsDNA-binding activity of ParB was analysed by TBM-and TBE-PAGE as described (7) Where Y is the measured increase in fluorescence, Bmax is the maximal increase in fluorescence, h is the Hill coefficient and Kdapp is the apparent dissociation constant. When comparing wild type and mutant binding isotherms, the data were well-fitted using a shared value for the Hill coefficient (i.e. there was no evidence for changes in binding cooperativity as a result of the mutations studied). Standard errors for the fitted parameters were calculated in GraphPad Prism.

Instrument and samples
We used a home-made magnetic tweezers setup similar in design to that described in (58,59) and detailed in (60). In brief, images of 1 μm 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 is used to determine the spatial coordinates of beads with nm accuracy in x, y and z. To have a measurement of the degree of induced decondensation, we determined a condensation ratio, Cr ( Figure 6D), which was calculated simply as: where z0 is the expected extension at 0.34 pN measured before ParB injection and z is the equilibrium extension after induced-decondensation. z was determined from average extensions of 120 data points at 390 seconds after the cell volume was completely exchanged. These data were acquired at 60 Hz and filtered down to 3 Hz.   To measure the amount of DNA bound to ParB, GoTaq (Promega) qPCR mix was used for the PCR reactions and qPCR was performed in a Rotor-Gene Q Instrument (Qiagen) using serial dilution of the immunoprecipitate and the total DNA control as the template.

Force-extension curves and work calculation
Oligonucleotide primers were then designed that amplify at an interval of ~500-1000 bp away from parS 359˚ and were typically 20-25 bases in length and amplified a ~200-300 bp PCR product (Table S2).