Molecular Anatomy of ParA-ParA and ParA-ParB Interactions during Plasmid Partitioning*

Background: Small homodimeric δ2 (ParA) and ω2 (ParB) and parS mediate pSM19035 partitioning. Results: The δ2 ATPase is a modular protein. Conclusion: δ2·δ2 and δ2·ω2 interacting domains are juxtaposed. Significance: ATP, nonspecific DNA, and ω2 bound to parS induce multiple transitions on δ2. Firmicutes multidrug resistance inc18 plasmids encode parS sites and two small homodimeric ParA-like (δ2) and ParB-like (ω2) proteins to ensure faithful segregation. Protein ω2 binds to parS DNA, forming a short left-handed helix wrapped around the full parS, and interacts with δ2. Protein δ2 interacts with ω2 and, in the ATP-bound form, binds to nonspecific DNA (nsDNA), forming small clusters. Here, we have mapped the ω2·δ2 and δ2·δ2 interacting domains in the δ2 that are adjacent to but distinct from each other. The δ2 nsDNA binding domain is essential for stimulation of ω2·parS-mediated ATP hydrolysis. From the data presented here, we propose that δ2 interacts with ATP, nsDNA, and with ω2 bound to parS at near equimolar concentrations, facilitating a δ2 structural transition. This δ2 “activated” state overcomes its impediment in ATP hydrolysis, with the subsequent release of both of the proteins from nsDNA (plasmid unpairing).

Faithful segregation of newly replicated genomes to daughter cells requires specialized partitioning machinery. In eukaryotic cells, the well known microtubule-based mitotic spindle apparatus, which guides the separating chromosomes to the different ends of the dividing cell, drives the chromosome segregation process (1,2). In contrast, the mechanism of chromosome separation and plasmid partition in bacterial cells is not well understood. Three main types of systems have been described that ensure faithful partition of low copy number plasmids (reviewed in Refs. [3][4][5][6]. Of these, the ParAB system is the most widespread in plasmids and is the only type present also in bacterial chromosomes (3)(4)(5)(6). Unraveling the molecular basis of this plasmid partitioning system is therefore important to help us to understand chromosomal segregation. The ParAB partition system requires three or more components: two trans-acting dimeric proteins (ParA and ParB), one or more cis-acting parS site(s), and, in most cases, the host's chromosomal DNA (3)(4)(5)(6). The Walker box ParA ATPases can be subdivided into two distinct families based on their size or structure. The large (Lg) 3 Walker box ATPases (e.g. Escherichia coli P1-ParA (Lg) or F-SopA (ParA-Lg) plasmids) bind and hydrolyze ATP, and then ParA-ADP specifically binds to its cognate site to regulate expression of the ParAB locus. When hydrolysis is delayed, ParA⅐ATP* binds nonspecific DNA (nsDNA), forming a carpet on the DNA in vitro (7)(8)(9)(10)(11). This gradient-like distribution of the ParA protein on the nucleoid is an essential step for faithful plasmid segregation in vivo (12). The small (Sm) Walker-box ParA ATPases, which lack the specific DNA binding domain, can be divided into two subgroups. The first group includes Streptococcus pyogenes pSM19035-␦ 2 and Bacillus subtilis chromosome-encoded Soj (ParA-Sm), which in the ATP-bound form bind to nsDNA (13,14). DNA binding by the small ParA ATPases (pSM19035-␦ 2 and Soj) is a critical step for faithful segregation of plasmids and chromosomes (13,15). In contrast, the small ParA ATPases of the second group (e.g. Salmonella newport TP228-ParF plasmid) form bundles in the absence of DNA (16,17).
The interaction of the ParA and ParB components, which leads to proper separation of plasmid copies, has been extensively studied in plasmids of the ␥-Proteobacteria class. These studies provide the foundation for filament-and non-filamentbased modes of plasmid segregation. In the filament-based modes, ParA (Sm), when bound to ATP, assembles into fila-ments or bundles, and the partition complexes are mobilized by linear contractile filaments reminiscent of the spindle mechanism in eukaryotes (thread pushing or pulling model) ( Fig. 2A) (16,17). Alternatively, ParA (Sm), in the ATP-bound form, assembles by forming nucleoprotein filaments, and the partition complexes are mobilized by helical or linear filaments as a cargo (filament-pulling model) (Fig. 2B) (26,27). In the nonfilament-based models (diffusion-ratchet and DNA relay models), ParA dimers or small oligomers bind to the nucleoid (Fig.  2C) (9 -12, 28, 29). In the diffusion-ratchet model, a propagating ParA (Lg) ATPase gradient is the driving force for movement of the partition complexes (9 -11), whereas in the DNA relay model, the forces that drive segregation are generated by a ParA (Sm) gradient and the elastic forces within the DNA molecule (29). The segregation machinery of plasmids and bacteria of the Firmicutes phylum, which are evolutionarily separated by more than 1,500 million years from Proteobacteria (a genetic distance larger than the one between humans and plants), is less understood. It is likely that the characterization of the mechanism of active partitioning of low copy number plasmids of the Firmicutes phylum might contribute to the general understanding of ParAB-mediated plasmid and chromosome segregation in bacteria.
The structure of ␦ ϩ14 (having 14 extra N-terminal residues) bound to ATP␥S and Mg 2ϩ includes all 284 residues of the WT ␦ protein (15). ATP-bound ␦ 2 4 possesses a U-shaped structure, with one arm and a part of the joining region representing a monomer (Fig. 1B). The Walker domains face the cleft, and the bottom of the U-shaped dimer is negatively charged, whereas the C-terminal upper tips are positively charged (Fig. 1B) (15). Previous mutagenesis data showed that the nsDNA binding domain lies at the tip of the arm (Fig. 1B) (14). Protein ␦ 2 shows a low but significant degree of sequence identity with the P1-ParA 2 (Lg) (ϳ24%) and the Soj (ParA-Sm) (ϳ22%) ATPases, but the sequence identity drops to Ͻ20% when compared with the TP228-ParF (ParA-Sm) ATPase. However, the quaternary structures of these four ParA ATPases, three from Gram-negative and one from Gram-positive bacteria, are conserved (Fig.  1C).
At present, the ␦ 2 regions involved in oligomerization and in the interaction with 2 are poorly defined, and they have been mapped in this work. We show here that the ␦ 2 ⅐ 2 and ␦ 2 ⅐␦ 2 interacting domains are adjacent to but distinct from each other and suggest some functional transitions that are crucial for the activation of the ␦ 2 ATPase activity.

Experimental Procedures
Strains and Plasmids-The B. subtilis strain and the plasmids used for segregation studies are indicated in Table 1. The plasmids used for overexpression of proteins were propagated in E. coli BL21(DE3) (pLysS) and are listed in Table 1.
The ␦ gene encodes two co-linear polypeptides, a 298-residues (␦ ϩ14 ) and a WT 284-residue product. The plasmid pCB746-borne WT ␦ gene was used for site-directed mutagenesis and discrete deletions from the N-and C-terminal regions.
The exposed and positively charged Lys residues at the C-terminal end of ␦ 2 were substituted by Ala (␦ 2 K242A, ␦ 2 K259A/ K260A) or Ser (␦ 2 K248S), and the negatively charged residue 211 was replaced by Ala (␦ 2 D211A), as described previously (34). Many of the initially designed C-terminal deletion mutants were insoluble. To overcome this inconvenience, ␦ 2 variants containing or lacking residues within random coil regions around the initially programmed C-terminal deletion were constructed. Finally, the codons at positions 255, 227, 197, and 164 of the WT ␦ gene were fused to His 6 codons, leading to a series of constructs truncated from the 3Ј-end.
Plasmid Copy Number and Plasmid Stability Test-The number of plasmid copies per cell was estimated by quantitative PCR. Normalization was done with two distinct chromosomal genes as described earlier (35). The number of plasmid-containing cells was determined by replica-plating onto chloramphenicol-supplemented LB plates. The frequency of plasmid loss was calculated as described previously (35).
Chemicals, Enzymes, Proteins, and DNA-All chemicals were pro-analysis grade and purchased from Roche Diagnostics (Mannheim, Germany). DNA restriction and modification enzymes and nucleotides were from New England Biolabs (Frankfurt, Germany) and Sigma-Aldrich (Madrid, Spain). Ultrapure acrylamide was purchased from Serva (Heidelberg, Germany). The 2 or ␦ 2 variants were purified as described for WT 2 and ␦ 2 proteins (15,31,33). pBC30-borne parS2 DNA (source of parS DNA) was purified as described (31). parS DNA is expressed as moles of DNA molecules, and this was estimated using a molar extinction coefficient of 6,500 M Ϫ1 cm Ϫ1 at 260 nm. The protein concentrations were determined by absorption at 280 nm using molar extinction coefficients of 2,980 M Ϫ1 cm Ϫ1 for 2 and 2 ⌬N19 and 38,850 M Ϫ1 cm Ϫ1 for ␦ 2 and its variants. Concentrations of all the proteins are expressed as moles of protein homodimers. All structural images were generated using the PyMOL Molecular Graphics System, version 1.5.0.4 (Schrödinger, LLC).
Protein Cross-linking-The cross-linking agent DSS was used to study protein-protein interactions as described previously (36). Cross-linking was performed by incubating parS DNA with ␦ 2 or its mutant variants in buffer B (50 mM Tris-HCl (pH 7.5), 50 mM NaCl, and 10 mM MgCl 2 ) supplemented or not with 1 mM ATP for 15 min at 37°C. The assays were performed in the presence (ϩ) or absence (Ϫ) of 2 prior to the addition of the indicated concentrations of DSS at various 2 /␦ 2 molar ratios. Alternatively, parS DNA was preincubated with one of the proteins (5 min, 37°C), and the preformed complex was incubated with the second protein (5 min). Then DSS was added, and reactions were left for 10 min at 37°C and then stopped by the addition of 10 l of stop buffer C (50 mM Tris-HCl, pH 7.5, 400 mM glycine, 3% ␤-mercaptoethanol, 2% SDS, 10% glycerol) and subjected to SDS-PAGE. Cross-linked protein bands were excised from the gel and identified by tryptic digestion coupled to MALDI-TOF-TOF analysis as described (14).
Protein⅐DNA Complexes-For EMSA, gel-purified 423-bp ␣-32 P-labeled HindIII-KpnI parS DNA was incubated with various amounts of WT 2 , WT ␦ 2 (or its variants), or both proteins together in buffer B containing or lacking 1 mM ATP or 1 mM ADP for 15 min at 37°C in a 20-l final volume. The reaction was stopped by the addition of loading buffer (1 mM EDTA, 0.1% bromphenol blue, and 0.1% xylene cyanol), and the samples were subjected to 4% or 6% PAGE. Gel electrophoresis was conducted using 1ϫ TAE as running buffer, at 200 V, 4°C, and the gels were dried prior to autoradiography.
To obtain K D(app) values, the concentration of free DNA and protein⅐DNA complexes was densitometrically determined under non-saturating conditions from differently exposed autoradiographs of EMSA gels. Protein concentrations that transfer 50% of the labeled DNA into complexes are approximately equal to the K D(app) under conditions where the DNA concentration is much lower than the K D(app) .
ATPase Activity Assays-The ATPase activity was assayed by thin layer chromatographic separation of the reaction products as described (15). The indicated proteins in the presence or absence of parS DNA were incubated for 180 min at 37°C in buffer B containing 5 mM ATP in a 20-l reaction mixture.

Results
The C-terminal Region of ␦ 2 Contributes to Plasmid Segregation-To investigate the contribution of the C-terminal region of ␦ 2 to faithful partitioning, a plasmid stability test was used. A set of plasmid-borne WT and mutant and ␦ genes, transcribed from their native promoters (P), termed P ␦ (parS1) and P (parS2) (Fig. 1A), were tested. Under the growth condi-tions used in this study, each cell contained, on average, ϳ8 Ϯ 1 plasmid copies, and the different mutations (see Table 1) did not affect plasmid copy number (data not shown).
As already observed, the and ␦ gene products, the presence of at least one parS site, 2 ⅐␦ 2 interaction, ATP binding and hydrolysis, and the nsDNA binding domain of ␦ 2 were necessary and sufficient for stabilization of an otherwise unstable plasmid in B. subtilis cells (Table 2) (15,37).
In the filament pulling model, the polymerization onto nsDNA as shown for pB171-ParA 2 (Sm) is essential (26). The structure and mutagenesis analysis of the symmetric ␦ 2 ATPase revealed that nsDNA binding domain occurs at the tip of the U (Fig. 1B) (14). If ␦ 2 polymerizes on nsDNA, it is likely that one of the external arms (e.g. the negatively charged C-terminal end) of ␦ 2 might be in close proximity with the opposite arm (e.g. the positively charged N-terminal end) of another ␦ 2 (Figs. 1B and 2B). To test this hypothesis, ␦ 2 variants lacking the N-(␦⌬N20) or C-terminal (␦⌬C255) region were constructed. Protein ␦ 2 ⌬C255 also lacks the putative IF2 domain predicted to be involved in bundle formation ( Fig. 2A) (17).
When the ␦ gene was replaced by the ␦⌬N20 variant, whose product lacks the first 20 residues, or by ␦⌬C255, whose product lacks the last 29 residues (15), plasmid segregation was only marginally affected by ϳ1.5and 3-fold, respectively (Table 2). However, when the ␦ gene was replaced by the ␦⌬C197 or ␦⌬C164 genes, which not only lacked the negatively charged C-terminal domain but also lacked the nsDNA binding domain, faithful plasmid segregation was abolished ( Table 2). These results suggest that the DNA binding domain is essential, but the N-and C-terminal ends are dispensable for accurate plasmid partitioning. These data were not consistent with the filament-based models (Fig. 2, A and B).
The ␦ Dimer of Dimer Interacting Domain Does Not Overlap with the nsDNA Binding Domain-To further evaluate the contribution of the C-terminal region to the different activities of ␦ (i.e. ATPase, nsDNA binding, and ␦ 2 ⅐ 2 and ␦ 2 ⅐␦ 2 interaction), the exposed and positively charged Lys residues at the C-terminal end of ␦ 2 (␦K242A, ␦K248S, and ␦K259A/K260A) were substituted by Ala or Ser, the genes were cloned, and their products were overexpressed and purified (Fig. 3). To ascertain the importance of these basic residues for DNA binding, the exposed and negatively charged Asp-211 residue was replaced by Ala. The mutated gene was cloned, and its product (␦D211A) was purified.
In the presence of ATP, dimers and higher order oligomers were observed with ␦D211A, ␦K242A, or ␦K248S upon the addition of the DSS cross-linking agent, but ␦K259A/K260A only formed dimers ( Fig. 3A) (14).
Wild-type ␦ 2 binds nsDNA with a K D(app) of 140 Ϯ 31 nM, and ␦ 2 D60A, which binds but does not hydrolyze ATP, binds nsDNA with a K D(app) of 75 Ϯ 19 nM (Table 3) (14, 15). ATP␥S FIGURE 1. Relevant genome organization, critical residues for ␦ 2 interaction with nsDNA, and structural comparison of ParA ATPases. A, genome organization of the duplicated part of plasmid pSM19035. The rep (involving copS, RNA III, repS, and the leading (oriS) and lagging (ssiA) replication origins) and segB loci (segB1 (, ⑀, and ) and segB2 (␦, , and parS sites, P copS , P ␦ , and P )) are highlighted. The promoters (P), the mRNAs, and the genes of the relevant regions are shown and denoted as boxes, wavy lines, and rectangles, respectively. The upstream region of the promoters of the copS (P copS ), ␦ (P ␦ ), and (P ) genes, which constitute the cis-acting centromere-like parS sites, are enlarged. The variable number of contiguous 7-bp iterons (heptads) and their relative orientation are symbolized by arrows (3 or 4). The promoters repressed by 2 (double red spheres) are indicated. B, mapping of the nsDNA binding domain in ␦ 2 . Electrostatic potential surface representation of ␦ 2 in the ATP␥S⅐Mg 2ϩ -bound form (PDB code 2OZE) displayed using PyMOL. The surface charge of ␦ 2 is negative near the bottom of the U and positive at the tips of the arms of the U. The relevant ␣-helices involved in nsDNA binding (in green) map to the tips of the arms of its U-shaped structure. Relevant residues for nsDNA binding, residue Asp-60 (involved in ATP hydrolysis), and ATP␥S⅐Mg 2ϩ for one of the monomers are indicated, but they are repeated twice in the dimer structure. A lateral view highlighting the U-shaped form of ␦ 2 is shown. C, superimposition of full-length monomer structures of ␦ (in green) and T. thermophilus Soj (ParA-Sm) (in blue) (PDB code 2BEK), with P1-ParA (in yellow) (PDB code 3EZ6) and with TP228-ParF (ParA-Sm) (in orange) (PDB code 3K9H). The superimpositions were done using PyMOL.
or ADP failed to promote ␦ 2 binding to nsDNA ( Table 3). Removal of a negatively charged residue (␦ 2 D211A) from the nsDNA binding domain increased binding to nsDNA with K D(app) of 25 Ϯ 1.5 nM ( Table 3). Removal of positively charged residues (␦ 2 K242A, ␦ 2 K248S, or ␦ 2 K259A/K260A) reduced binding to nsDNA as measured by EMSA (K D(app) of Ͼ1.6 M) (Fig. 3, C and D). However, electron microscopic analysis showed that the ␦ 2 K242A mutant bound to the nsDNA with an efficiency comparable with WT ␦ 2 , which suggests that the mutation does not affect the on rate of protein-nsDNA complex formation (34). 5 It is likely that the ␦ 2 K242A⅐nsDNA complex has a higher off rate than the WT ␦ 2 ⅐nsDNA complex (see below).
Deletion of the nsDNA Binding Domain of ␦ Does Not Affect Dimer-Dimer Interaction-The ␦ genes coding for the previously mentioned deletion variants from the C-terminal end (Table 1) were cloned, and their products were overexpressed and purified. Proteins ␦ 2 ⌬C255 (which lacks the negatively charged C-terminal end and part of the predicted IF2 (17)), ␦ 2 ⌬C227 (which also lacks the ATP-interacting residues), ␦ 2 ⌬C197 (which also lacks the nsDNA binding domain), and ␦ 2 ⌬C164 (which also lacks part of the hypothetical IF1 region) were analyzed under in vitro conditions (Fig. 4A).
The ATPase activity of the ␦ 2 variants (␦ 2 ⌬C227, ␦ 2 ⌬C197, and ␦ 2 ⌬C164) was not stimulated by the addition of parS DNA or parS bound to 2 (Fig. 4C). Upon addition of 2 bound to parS DNA, the ATPase activity of ␦ 2 ⌬C255 was slightly stimulated (Fig. 4C). This is consistent with crystallographic data that show that three ␦ residues, Lys-238, Ser-240, and Tyr-265, which map to the C terminus, are also involved in ATP binding. Ser-240 forms hydrogen bonds with the adenosine, and Lys-238 and Tyr-265 form hydrogen bonds with the amino group of the two ATP␥S molecules in the ␦ 2 -ATP␥S structure (see below) (15). Thus, all three residues have activities that are crucial for full ␦ 2 -mediated ATP hydrolysis, and ␦ 2 ⌬C255, which only lacks Tyr-265, shows an intermediate phenotype (Fig. 4C).
The ␦ 2 ⅐ 2 Interacting Domain Does Not Map to the C-terminal End of ␦ 2 -Protein 2 binds to parS, forming transient 2 ⅐parS complexes (K D(app) of 5 Ϯ 1 nM, half-life Ͻ1 min) (24,32). The interaction of 2 bound to parS with ␦ 2 or apo-␦ 2 led to longer-lived 2 ⅐parS complexes (K D(app) 0.7 Ϯ 0.1 nM, half-life ϳ34 min) (14). It is likely that a poorly defined region of ␦ 2 , upon interacting with the unstructured N-terminal domain of 2 , facilitates structural transitions that enhance the stability of 2 ⅐parS complexes (14). To test whether 2 interacts with ␦ 2 through its C-terminal domain and whether such protein-protein interaction facilitates the binding of ␦ 2 or 2 to parS DNA, two different approaches were undertaken. First, in the presence of ATP, preformed 2 ⅐parS DNA complexes were incubated with limiting ␦ 2 , ␦ 2 K242A, ␦ 2 K248S, or ␦ 2 K259A/K260A concentrations. The preformed 2 ⅐parS complexes enhanced ternary complex formation with WT ␦ 2 ( 2 ⅐parS⅐␦ 2 ) to K D(app) 5 A. Volante and J. C. Alonso, unpublished results.

TABLE 1 Strains and plasmids
a Mid-copy number (7-9 copies/cell) pHP14 derivatives bearing the ␦ and genes (or their variants) under the control of their own promoters (P ␦ (parS1) and P (parS2); see Fig. 1).

TABLE 2
Effect of the different variants of the pSM19035 partitioning system in the faithful segregation of an unstable vector B. subtilis cells containing the Parvector or its derivatives (8 Ϯ 1 copies/cell) bearing the par locus of pSM19035 or its variants were grown in antibiotic-free LB medium at 30°C, and the frequency of plasmid loss during exponential growth was measured after 100 generations. The genes were transcribed from their native promoters.

Par genes
a This information was reported previously (15), but the experiments were reproduced here for comparison. ␦ 2 ⅐nsDNA alone (Table 3). When WT ␦ 2 was replaced by ␦ 2 K242A, ␦ 2 K248S, or ␦ 2 K259A/K260A, the preformed 2 ⅐parS complexes enhanced the formation of ternary complex from K D(app) of Ͼ1600 nM to K D(app) of 40 -85 nM (Table 3). Second, in the presence of limiting 2 concentrations (0.75 nM), C-terminal deletion mutants, which lacked the DNA binding domain, cannot form ternary complexes but may help 2 to bind to parS DNA. In the presence of limiting 2 , ␦ 2 ⌬C255, ␦ 2 ⌬C227, ␦ 2 ⌬C197, or ␦ 2 ⌬C164 enhanced 2 ⅐parS complex formation from K D(app) of 5 Ϯ 1 nM to K D(app) of Յ0.75 nM (Fig.  4, D and E). The interaction of 2 with ␦ 2 ⌬C164 facilitated the formation of 2 ⅐parS complexes, but these protein⅐DNA complexes were found to be less stable in comparison with the WT complex (Fig. 4, E and F). These results suggest that the deletion mutants still interact with 2 and facilitate 2 ⅐parS complex formation. To compare these results, the data presented in Fig.  4, D and E, were quantified. As revealed in Fig. 4F, the concentration of ␦ 2 (or its mutant) sufficient to facilitate 2 ⅐parS DNA complex formation was significantly enhanced, with a K D(app) of 30 -50 nM, which is 3-5-fold below the K D(app) for ␦ 2 ⅐nsDNA (K D(app) of 140 Ϯ 31 nM). It is likely, therefore, that the ␦ 2 variants interact with 2 with an efficiency similar to that of WT ␦ 2 (Fig. 4F). These data altogether support the conclusion that (i) interaction of the preformed 2 ⅐parS DNA complex with ␦ 2 K242A, ␦ 2 K248S, or ␦ 2 K259A/K260A facilitates ternary complex formation; (ii) the C-terminal region of ␦ 2 (residues 164 -284) is dispensable for interaction with the 2 ⅐parS DNA complexes; (iii) removal of the 120 C-terminal residues from ␦ 2 still enhances 2 ⅐parS complex formation, but the complexes formed are less stable (Fig. 4E); and (iv) the ␦ 2 concentration required for interaction with 2 bound to parS DNA (Fig. 4, D and E) or to facilitate 2 ⅐parS complex formation (Fig. 4F) is 3.5-4-fold lower than for interaction with nsDNA.
The Central Domain of ␦ 2 Interacts with 2 -To identify the region(s) of ␦ 2 involved in the interaction with 2 , both proteins were cross-linked in vitro by DSS in the presence of parS DNA, and the different protein bands were gel-purified (Figs. 5A (ag) and 6A (a-f)), subjected to limited proteolysis, and analyzed by MALDI-TOF-TOF as described previously (14). With this technique, under high sequence coverage, the regions not involved in protein-protein interaction should be proteolyzed by trypsin, and the generated polypeptides should be detected with the expected molecular mass. On the other hand, if trypsin does not reach the covalently bound proteinprotein regions, the expected polypeptide(s) should be absent. Alternatively, the interacting polypeptides, due to cross-linking among them, will have a higher molecular mass and therefore will also be absent. It is worth noting that our analysis was qualitative; hence, we considered a region involved in the protein-protein interaction if there was no detection of the polypeptide corresponding to that region, although it was present in the control (i.e. in the absence of the interacting partner).
First, 2 -bound to parS DNA was analyzed. Protein bands a, b, and c in Fig. 5A corresponded to monomers, dimers, and tetramers of 2 , respectively. In band a, the polypeptide corresponding to the ␤-strand (residues 22-31) of 2 (7.9 kDa, peptide coverage ϳ89%), in the presence of parS DNA, was not observed (Fig. 5B). This finding is consistent with data showing that 2 interacts with parS DNA through this ␤-strand (25).
The cross-linked ␦ proteins in the presence of nsDNA, leading to bands d, e, f, and g, were gel-purified, subjected to limited tryptic digestion, and followed by MALDI-TOF-TOF analysis (Fig. 5, A and B). Here, ␦ polypeptides encompassing intervals 139 -215 and 242-265 were not detected (Fig. 5B). Within the 139 -215 interval, which lacks trypsin sites, the ␦⅐␦ interaction was previously mapped (15). Indeed, in the crystal structure, a ␦ dimer is stabilized by a hydrophobic surface patch of otherwise solvent-accessible surface area, augmented by two reciprocal intersubunit salt bridges formed between Arg-119 and Asp-189 of each monomer (15).
Within the 242-265 interval, which contains three trypsin sites, the nsDNA binding domain was previously mapped (14). The cross-linked ϳ64-kDa protein in the absence of nsDNA (band h) was gel-purified, subjected to limited tryptic digestion, and followed by MALDI-TOF-TOF analysis (Fig. 6B). In the absence of nsDNA, a ␦ polypeptide encompassing the 242-247 interval was detected (Fig. 6C). However, these trypsin sites were not seen in ␦ 2 bound to nsDNA (Fig. 5B, band f). It is likely that the ␦ 2 -␦ 2 conformation adopted in the presence of nsDNA blocks trypsin's approach to these cognate sites following removal of the nsDNA.
In the presence of preformed 2 ⅐parS DNA complexes, at about stoichiometric 2 /␦ 2 ratios, ␦ and formed a new discrete band e (Fig. 5A, lanes 3 and 4). When the order of addition was altered or the relative molar ratio of 2 /␦ 2 was modified (3:1 or 0.3:1), bands e (as in Fig. 5A) and eЈ were observed (Fig. 6A,  lanes 3 and 6). Limited trypsin proteolysis and MALDI-TOF-TOF analysis of this gel-purified band e (ϳ40 kDa; Figs. 5A and 6A) showed that this novel polypeptide was composed of (7.9 kDa) and ␦ (32.6 kDa) cross-linked in a 1:1 stoichiometry with sequence coverage of 65.8% for ␦ and 81.3% for . The gelpurified band eЈ (ϳ48 kDa) (Fig. 6A, lanes 3 and 6) was composed of two monomers and one ␦.
When band e was analyzed, six discrete regions of ␦ and two discrete regions of were not observed (Fig. 5B). Polypeptides 1-10 and 47-52 (coverage Ͼ77%) of were missing. We propose that the unstructured N-terminal region of 2 bound to parS, upon interaction with ␦ 2 (Fig. 5B), undergoes a structural transition that might involve ␣-helix formation, as suggested by protein structure predictions (34). These ⅐␦ complexes might be stabilized by residues 47-52 of , because this region was also not detected when both proteins were incubated (Fig. 5B). When ␦ peptides missing in the gel-purified 40 kDa band e were compared with those of bands d (␦) and f (␦ 2 ) (coverage Ͼ60%), the two discrete segments, encompassing residues 88 -119 and 216 -223, were also not detected, suggesting that these regions may correspond to the -interacting region (Fig. 5, B and C). Within the 88 -119 missing interval, there is a potential trypsin cleavage site at position 116. Because we have not identified the cross-linked product, we cannot rule out the possibility that the 88 -119 or the 88 -116 region could be targeted by . Consistent with this, the region consisting of residues 88 -116 or 88 -119, comprising the ␣4 -␣6 interval, is surface-exposed in the ␦ 2 structure (Fig. 5C) (15).
Protein ␦ 2 ⌬C227 was incubated with 2 ⅐parS DNA, and then DSS was added. A novel band of ϳ34 kDa (denoted as band e*), which was composed of one and one ␦⌬C227 monomer, was observed at substoichiometric /␦⌬C227 ratios (Fig. 5A, lanes  7 and 8). In band e*, four discrete regions of ␦ and two of were not observed by MALDI-TOF-TOF analysis (Fig. 5B). When the peptides absent in this gel-purified 34 kDa band were compared with those absent in bands d and f, again the region detected previously with the WT protein (88 -119 or 88 -116 interval) was missing. The other potentially interacting region, the 216 -223 interval, was present when ␦ 2 was replaced by ␦ 2 ⌬C227. Therefore residues 216 -223, which were previously attributed to DNA binding, were not further evaluated. It is likely that the main ⅐␦ interacting domain maps to the central region of ␦ 2 or ␦ 2 ⌬C227 (residues 88 -119; the ␣4, ␣5, ␤3, ␤4,  and ␣6 interval) and is oriented opposite the face of the DNA binding domain (Figs. 5C and 7C). In contrast, using yeast or bacterial two-hybrid systems, it was suggested that the ParA⅐ParB interacting domain of the (F-SopA (ParA-Lg)) or the (pSM19035-␦ 2 ) ParA ATPase maps to the C-terminal end (i.e. residues 198 -284) (38,39). Because the previous results suggested that ␦ 2 ⌬C197 still interacts with 2 bound to parS (Fig.  4E), it will be of significant interest to determine the F-SopA⅐SopB (ParA⅐ParB-Lg) interacting domain by other methodologies.
The Central Region of ␦ 2 Is Involved in Dimer-Dimer Interface-To define the region(s) of ␦ 2 involved in dimer of dimer formation, ␦ 2 was preincubated with preformed 2 ⅐parS DNA complexes, the ternary complexes were treated with DSS, and the newly covalently bound protein was analyzed as described above (Fig. 5A, termed band g). A comparison of bands g (ϳ98 kDa, ␦ 2 ⅐␦) with bands d (ϳ32 kDa, ␦) and f (ϳ64 kDa, ␦ 2 ) revealed that one discrete region, encompassing residues 68 -119, was not detected in band g (Fig. 5B). This region constitutes the ␣3-␣6 interval in the crystal structure and is an exposed surface distant from the DNA binding domain (Figs. 5 (B and C) and 7 (A and C)). However, the 68 -119 interval partially overlaps the ⅐␦ interacting domain (residues 88 -119; interval ␣4 -␣6) (Figs. 5B and 7C). The ␦ 2 ⅐␦ interacting region is close to the ATP binding and hydrolysis domains (Walker AЈ (residues 51-61) and Walker B (residues 142-147)) (Fig. 7, A and C).

Discussion
pSM19035 partitioning, which depends on the dynamic interaction between the ␦ 2 ATPase, 2 , parS, and the host chromosome, is a multistep process with discrete functional and/or structural transitions necessary for faithful segregation. The small Walker box ATPase ␦ forms dimers that are stabilized by FIGURE 5. A central domain of ␦ is required for 2 ⅐␦ 2 interaction. A, DSS cross-linking assay showing the interaction between 2 (1 g) and ␦ 2 (or ␦ 2 ⌬C227), at 0.5:1 and 1:1 2 /␦ 2 ratios, in the presence of parS DNA. Cross-linked products were separated by 10 -15% SDS-PAGE. Bands a-g were gel-purified and subjected to limited trypsin proteolysis, followed by MALDI-TOF-TOF identification of the polypeptides. B, the identified peptide sequences are shown in light () or dark gray (␦) boxes, and the missing regions are shown in dotted lines. C, the relevant region of ␦ involved in the interaction with or with itself is highlighted in the crystal structure in purple (␣4-␣5-␤3-␤4-␣6). In this structure, ATP␥S⅐Mg 2ϩ is shown in orange and yellow, and residue Asp-60 (Walker BЈ) is shown in green (PDB 2OZE, displayed using PyMOL). The experiments were performed more than three independent times. a hydrophobic patch that occupies about 2197 Å 2 /subunit of otherwise solvent-accessible surface and augmented by two reciprocal intersubunit salt bridges formed between Arg-119 and Asp-189 of each monomer (see Ref. 15). Protein ␦ 2 , upon binding to nsDNA, forms discrete clusters on the nsDNA (4 -6 ␦ 2 /blob) (24), perhaps by interacting through the 68 -119 region with other ␦ 2 molecules (see Fig. 7, A and C). In contrast, using a bacterial two-hybrid assay, the ␦ monomer-monomer interface was mapped to residues 104 -105 (39), and using yeast two-hybrid and immunoprecipitation assays, the chromosome-encoded Pseudomonas aeruginosa ParA (Sm) monomermonomer interface was mapped to an interval equivalent to residues 89 -105 of ␦ 2 (40). To reconcile the apparent discrepancy between the crystal structure of ␦ 2 ⅐ATP␥S⅐Mg 2ϩ (15), our data (Fig. 5B), and those published by other groups (39,40), we propose that the bacterial or yeast two-hybrid assays cannot discriminate between the monomer and the dimer interface and that the authors were mapping the dimer-dimer interface (see Fig. 7A, center).
The interaction of the N-terminal region of 2 (residues 1-10) with the central region of ␦ 2 (residues 88 -119) leads to a discrete ternary ( 2 ⅐parS⅐␦ 2 ) complex (Figs. 5C and 7, A (left) and C). However, if parS DNA was omitted or 2 was replaced by 2 ⌬N19, such protein-protein interaction was not observed (14,33). We propose that the unstructured N-terminal region of 2 bound to parS, upon interaction with ␦ 2 , undergoes a structural transition with residues 1-9 folding into an ␣-helix, as predicted by different protein folding prediction programs (34). Such interaction, at stoichiometric 2 and ␦ 2 concentrations, may facilitate ␦ 2 -mediated ATP hydrolysis and ␦ 2 disassembly from the paired (parS⅐ 2 ⅐␦ 2 ⅐nsDNA) complexes. In the presence of an excess of 2 , the ATPase activity of ␦ 2 is inhibited (15), and the plasmid can be engaged in a new pairing complex with a ␦ 2 in the nucleoid.
While the functional characterization of the 2 ⅐␦ 2 interacting domains was being carried out during this study, the interacting domain of chromosome-encoded P. aeruginosa ParA (Sm) with ParB (Lg) was published (40). These authors mapped the ParB⅐ParA interacting region to within an interval equivalent to residues 89 -105 of ␦ 2 , by yeast two-hybrid and immunoprecipitation assays (40). This is in good agreement with the ⅐␦ interacting region reported here (Fig. 5B). However, it was unexpected because equivalent regions in both small ParA ATPases share very little sequence identity (Ͻ15%). Furthermore, plasmid-and chromosome-encoded large ParB proteins, which are of the helix-turn-helix family, bind specifically and nonspecifically to DNA and spread over many kb, leading to bridging, looping, and condensing of the nsDNA (18 -21). Con-  6) were incubated with ␦ 2 or 2 as indicated (at a 3:1 2 /␦ 2 ratio in lane 3 or at a 0.3:1 2 /␦ 2 ratio in lane 6), and the specific interactions were analyzed using DSS. Cross-linked products were resolved by 10 -15% SDS-PAGE. The stained protein bands e (ϳ40 kDa) and eЈ (ϳ48 kDa) were gel-purified and subjected to limited trypsin proteolysis and MALDI-TOF-TOF analysis of the polypeptides. Tetrameric to hexameric complexes are marked (lanes 4 -6). B, ␦⅐␦ interaction in the absence of nsDNA. The DSS-cross-linked product (lane 2) was separated by 10 -15% SDS-PAGE. Band h was gel-purified and subjected to limited trypsin proteolysis, followed by MALDI-TOF-TOF identification of the polypeptides. C, the identified peptide sequences are shown in boxes, and the missing regions are shown in dotted lines. The N-and C-terminal regions are denoted.
versely, the plasmid-encoded small ParB proteins (e.g. 2 ), which are ribbon-helix-helix proteins, specifically interact with parS without significant spreading (24). It is likely that the unrelated P. aeruginosa ParB (Lg) and 2 might expose an equivalent domain recognized by an equivalent region in the ParA (Sm) ATPase.
The results presented here are inconsistent with the ParAB filament-based (thread-pushing or -pulling ( Fig. 2A) or filament-pulling (Fig. 2B)) models. In the thread-pushing or -pulling model, the central (IF1; see Fig. 4A) and the C-terminal regions (IF2) of the TP228-ParF (ParA-Sm) (Fig. 7B) should be involved in bundle formation, as depicted in Fig. 2A (16, 17). However, ␦ 2 ⌬C164, which lacks IF2 and part of IF1, forms dimers and small oligomers (Fig. 4B). We have shown by atomic force microscopy that upon ATP binding, ␦ 2 free in solution forms discrete blob-shaped structures containing 2-3 ␦ 2 molecules (24) rather than long bundles in the absence of nsDNA. It is likely that DNA-independent bundle formation ( Fig. 2A) and ␦ 2 blobs on nsDNA (Fig. 2C) are two mutually exclusive states, and ␦ 2 bundles, if formed in vivo, might be for storage purposes, as proposed for RecA (41), rather than for plasmid segregation. Indeed, many of these large or small ParA proteins might form bundles (e.g. F-SopA), but such assembly is inhibited in the presence of nsDNA (42).
In the filament-pulling model, pB171-ParA 2 (Sm) binds and polymerizes on nsDNA, with the filaments pulling plasmids apart by depolymerization (Fig. 2B) (26,27). If pB171-ParA 2 (Sm) interacts with and polymerizes on nsDNA, as depicted in Fig. 2B, the N-and C-terminal domains should play a crucial role in plasmid segregation. However, deletion of the N-termi-nal (␦ 2 ⌬N20) or C-terminal (␦ 2 ⌬C255) domain did not abrogate accurate plasmid segregation (Table 2). Alternatively, once the tip of U-shaped ParA 2 (Sm) binds nsDNA, ParA 2 (Sm) may polymerize through unknown regions. To accommodate our results, we have to assume that ParA 2 polymerization onto nsDNA leads to double filaments, because the dimer of dimer interface maps at the bottom of the U-shaped protein (Fig. 7A, center) and is free for interacting with a preformed filament. Protein ␦ 2 fails to form nucleoprotein filaments (15,24). However, pseudofilaments on nsDNA were observed in the presence of ␦ 2 and a very large excess of 2 (Ͼ300-fold over its K D(app) ) (15). Dissection of these "( 2 ⅐␦ 2 ) n ⅐nsDNA filaments" revealed that they were composed of discrete ␦ 2 blobs stabilized by the nonspecific interaction of 2 with nsDNA (34).
Our data support non-filament-based models and are compatible with the diffusion-ratchet or DNA relay models. Because these models diverged at later stages (11,29) and we are analyzing the early ones, they are considered as a single model (Fig. 2C). From previous data and data presented here, we conclude that (i) the unstructured N-terminal region of 2 (denoted in red in Fig. 2C), which is dispensable for binding to parS DNA, is involved in the interaction with ␦ 2 (25, 33); (ii) ␦ 2 , upon ATP binding, binds to nsDNA and forms discrete blob-shaped structures containing 5-6 ␦ 2 /blob rather than filaments on the nsDNA in vitro (14,24), and it is homogeneously distributed on the nucleoid in vivo (34); (iii) the interaction of ␦ 2 bound to nsDNA with 2 bound to parS DNA promotes a structural transition in both proteins (see Figs. 5C and 7A); (iv) this protein-protein complex captures and moves the plasmids from any cytosolic position toward FIGURE 7. Summary of the nsDNA, ⅐␦, and ␦ 2 ⅐␦ 2 binding motifs of ␦ 2 identified in this work. A, ribbon diagram of ␦ 2 with key motifs highlighted in green, cyan, purple, and brown. From left to right, ␦ 2 was rotated 90°to highlight the different motifs. Boxes of broken lines indicate the interacting regions. B and C, sequence alignment of relevant regions of the small ParA ATPases (␦ 2 , Thermus aquaticus Soj and plasmid TP228 ParF). B, nsDNA binding motif (␣10 to part of ␣12; green). C, sequence alignment of the ␦ 2 region involved in ␦ 2 ⅐␦ 2 (␣3 to ␣4; cyan) and ⅐␦ (␣4 to ␣6; purple) interaction derived from our cross-linking experiments (see Figs. 5A and 6A). Sequence alignments were generated using ClustalW and displayed using Jalview version 2.7.