The P1 Plasmid Partition Complex at pars THE INFLUENCE OF ESCHERICHIA COLI INTEGRATION HOST FACTOR AND OF SUBSTRATE TOPOLOGY*

The P1 ParB protein is required for active partition and thus stable inheritance of the plasmid prophage. ParB and the Escherichia coli protein integration host factor (IHF) participate in the assembly of a partition complex at the centromere-like sitepars. In this report the role of IHF in the formation of the partition complex has been explored. First, ParB protein was purified for these studies, which revealed that ParB forms a dimer in solution. Next, the IHF binding site was mapped to a 29-base pair region withinpars, including the sequence TAACTGACTGTTT (which differs from the IHF consensus in two positions). IHF induced a strong bend in the DNA at its binding site. Versions of pars which have lost or damaged the IHF binding site bound ParB with greatly reduced affinity in vitro and in vivo. Measurements of binding constants showed that IHF increased ParB affinity for the wild-type pars site by about 10,000-fold. Finally, DNA supercoiling improved ParB binding in the presence of IHF but not in its absence. These observations led to the proposal that IHF and superhelicity assist ParB by promoting its precise positioning at pars, a spatial arrangement that results in a high affinity of ParB for pars.

The prophage of bacteriophage P1 exists as an autonomously replicating plasmid in Escherichia coli (for review see Ref. 1). The plasmid copy number is very low (one to two/ host chromosome), and P1 requires an active partition system, pur, to promote proper plasmid segregation at cell division. P1 pur encodes two proteins, ParA and ParB, and contains a centromere-like site, pars, so called because it is required in cis for plasmid stability (Fig. 1) (2). It has been proposed that one or both plasmid-encoded proteins bind to pars, which leads to pairing of plasmids and association of the paired complex with the host partition apparatus at the nascent septum of the dividing cell (3, 4). Subsequent cell division then splits the complex (and the pairs) between the two daughter cells (3, 4). Studies in viuo (5) and in vitro (6, 7 ) showed that ParB recognizes and binds to pars. ParA is not required for this binding activity (5-7), and its action in partition is unknown. To date the only other known protein component of P1 partition is the E. factor, or IHF' (6). It was identified as a factor that stimulated ParB binding activity to p a r s (6). IHF is a site-specific DNAbinding protein, originally discovered as a requisite component of the phage X site-specific recombination system (8). It consists of two similar but nonidentical subunits, a and p (9), encoded by the bacterial himA and hip(himD) genes, respectively (10, 11). IHF participates in a variety of processes in E. coli, including recombination, replication, and transcription, as well as P1 plasmid partition (for review see Ref. 12).
Measured by a nitrocellulose filter retention assay in vitro, IHF stimulates the binding of ParB to pars although ParB shows site-specific binding activity without IHF (6). The natural p a r s site, ''pars-large," lies between a TaqI and Sty1 site in pur ( Fig. 1). DNase I protection studies on DNA fragments by Davis and Austin ( 7 ) showed that ParB interacts with sequences both upstream (parB-proximal) and downstream (parB-distal) of the DraI site in pars. They proposed that ParB recognizes the heptamer sequence 5' ATTTCAC because this is a common motif in all protected regions.
Deletion of the left end of p a r s (from TuqI to DraI) yields a version of pars, "pars-small," which still binds ParB but is no longer stimulated by IHF (6). In the absence of IHF, ParB binds equally to pars-large and pars-small. These observations led to the prediction that the TaqI-DruI deletion removed or destroyed the IHF binding site, and a reasonable match to the IHF consensus binding site was found within this region (6).
Previous studies (6) showed that IHF is a component of the wild-type partition complex in uiuo. A low copy number P1 derivative used to test partition in vivo is a X-P1 chimera that cannot integrate into the bacterial chromosome. Such X-miniP1 lysogens exist as P1 plasmids; they require the P1 replication and partition systems for proper maintenance (13). X-MiniP1 plasmids were less stable in cells without IHF (E. coli hip mutants) than in wild-type cells, but the defect was less severe than that caused by complete loss of ParB function (6). From this it follows that IHF assists ParB but is dispensable whereas ParB is absolutely required. However, competition experiments showed that IHF is always a component of the partition complex in wild-type cells. These experiments measured the ability of a plasmid carrying p a r s t o destabilize a different low copy number plasmid partitioned by pars. The phenomenon is attributed to the formation of heterologous' plasmid pairs during partition, resulting in a randomization The abbreviations used are: IHF, integration host factor; BSA, bovine serum albumin, DTT, dithiothreitol; DSP, dithiobis-(succinimidyl propionate); bp, base pair(s); kb, kilobase pair(s); SDS, sodium dodecyl sulfate; Hepes, 4-(2-hydroxyethyl)-l-piperazineethanesulfonic acid; FPLC, fast protein liquid chromatography.
For the purposes of this discussion, heterologous plasmids refer to plasmids with different (compatible) replicons; thus competition phenotypes will be caused by the partition system. of plasmid copies at cell division (3). Plasmids containing pars-small were unable to destabilize wild-type X-miniP1 plasmids (with pars-large, the natural context) in wild-type cells (14). However, in cells without IHF, pars-small was an effective competitor of pars-large (6). These experiments led to several important conclusions. First, they showed that IHF allows the partition system to discriminate between the two versions of pars. Second, the sequences upstream of the DraI site are not essential for ParB binding. Finally, wild-type X-miniPl partitions via the intact large site. Thus the wild-type p a r s site is pars-large and must contain all information required for both proteins to bind.
This study addresses key questions regarding the mechanism of IHF action within the partition complex and presents the first quantitative analysis of the IHF effect on ParB binding. Is the role of IHF to increase ParB affinity for pars? Where and how does IHF exert its effect? The influence of substrate topology on the formation of the partition complex as well as several physical properties of ParB protein have also been examined. The experiments presented here lead to the hypothesis3 that the precise geometry of the partition complex, promoted by IHF and superhelicity, is crucial for high affinity binding of ParB to DNA.

EXPERIMENTAL PROCEDURES
Bacterial Strains, Phage, and Plasmids-E. coli K12 strains (and relevant genotypes) were DH5 (recAI; Bethesda Research Laboratories) and BR4301 (DH5hip; 6). XKan-miniP1 is a kanamycin-resistant derivative of h-P1:5R-3 (5, 13). The X-miniPlparS-[Dra] insertion mutant derivatives were constructed by cloning the insertions onto a pBR322 (16) plasmid containing about 2 kb of P1 DNA and then crossing the mutations onto X-miniP1. The cloning procedures and genetic crosses have been described in detail previously (5). Briefly, an 8-bp SalI restriction enzyme linker was inserted into the DraI site in p a r s (Fig. l ) , producing pBEF140. Next, the 1.2-kb kanamycin resistance gene from Tn903 (from pUC4-K; Pharmacia LKB Biotechnology Inc.) was cloned into the new Sal1 site, producing pBEF15O. pBEF150 was crossed with X-P1:5R, selecting for kanamycin-resistant progeny, producing X-miniP1parS::kan. pBEF140 '' A preliminary version of these observations and model has been presented previously (15). was crossed with Xkan-P1:5R, selecting for recombinant phage that had picked up the entire plasmid (ampicillin resistant). Lysogens of these progeny were induced, ampicillin-sensitive phage were isolated and plaque purified, and phage DNA was restriction mapped. About 50% of the progeny phage had picked up the SalI insert, yielding X-miniP1parS::Sal.
A variety of pars-containing plasmids were used in this study. The pBR322 derivatives pALA207 (pars-large, or pars') and pBEF127 (pars-small) have been described previously (2, 6). pBEF146 (parS::Sal) contains an 802-bp Sau3Al fragment (the p a r s fragment analogous to that in pALA207) in the pBR322 BamHI site. pBEF143 (DNase footprinting substrate) contains the P1 TaqI-EcoRV pars fragment (Fig. l ) , converted to an EcoRI fragment with synthetic linkers. The vector for pBEF143 was pBEF141, a deletion derivative of pBR322 which lacks all sequences between PuuI and EcoRI but retains the EcoRI site.
Low copy number p a r s derivatives were in pMF3, a miniF derivative (17). pBEF161 contains the Sau3Al pars-large fragment from pALA207, and pBEF162 the BamHI pars-small fragment from pBEF127. For DNA bending studies, the vector plasmid was pBend5 (a gift from Sankar Adhya, NCI), a derivative of pBend2 (18 Proteins-Purified E. coli IHF was generously provided by Howard Nash (NIMH). ParB Fraction 111 is Bio-Rex 70 purified ParB described previously (6). Protein molecular mass standards were from Sigma and Bio-Rad.
Spectroscopic Analysis of ParB-This procedure was modified from that in Ref. 20. All absorption spectra were measured on a Cary 118 spectrophotometer, and the data were analyzed on a VAX computer. A model amino acid mixture of N-acetyl-L-tryptophanamide, Nacetyl-L-tyrosinamide, and N-acetyl-L-phenylalaninamide was prepared a t a ratio of 1:4:14. This corresponds to the ratio of tryptophan to tyrosine to phenylalanine in ParB, predicted from the DNA sequence (2). FPLC-purified ParB (Fraction IV) was concentrated by centrifugation in Centricon-30 filters (Amicon) and run over a 2-ml Sephadex G-25 column in freshly prepared 25 mM Tris-HC1, pH 7.5; 200 mM Na2S04; 0.1 mM EDTA; 0.1 mM DTT; and 10% glycerol. The UV absorption spectrum of native ParB, from 245 to 340 nm, was determined. After a 4-fold dilution with 8 M guanidine HCl, the absorption spectrum of this denatured ParB was measured and compared with that of the model amino acid mixture in the same buffer and guanidine HCl. The concentration of the denatured ParB and thus of the original native ParB was deduced from the known concentration of the model compounds. The absorption spectrum of native ParB showed a maximum a t 280 nm and an extinction coefficient of 0.34 (mg/ml)" cm" or 1.31 X M-' cm" (monomers). '"P-Labeled DNA Fragments-DNA fragments with 5' overhanging ends were 3' end labeled with [a-"'P]dATP or -dCTP, and blunt end fragments were 5' end labeled with [r-:"P]ATP as described (19). Fragments for DNase I protection studies were isolated from pBEF143. The 180-bp "lower s t r a n d fragment resulted from H i n d digestion of a 3' end-labeled EcoRI digest (the EcoRI end was formerly the P1 TuqI site; Fig. 1). The 452-bp "upper strand" fragment resulted from a PstI digest of a 3' end-labeled HindIII digest of pBEF143. Fragments were purified from agarose gels by electroelution in an IBI electroelutor. Maxam-Gilbert sequencing reactions were performed using a kit supplied by Du Pont-New England Nuclear.
3H Topoisomers of DNA-Supercoiled plasmid DNA was prepared and 'H labeled with HhaI methylase as described (6). DNA was labeled before conversion to other topological forms. Relaxed plasmid was prepared by treatment with calf thymus topoisomerase I, following the directions of the supplier (Bethesda Research Laboratories). Linear DNA was made by digestion with PstI and PuuI restriction ' S. Adhya, personal communication. endonucleases (which cut within the bla gene of pBR322). The double digestion ensured complete cutting and linearization but did not change the specific activity of the linear DNA. Nicked plasmid was prepared as described previously (21). All DNAs were extracted twice with phenol:CHCI:, (1:l) and twice with CHCI:,, precipitated with ethanol, washed with 70% ethanol, and resuspended in 10 mM Tris-HCI, pH 8; 1 mM EDTA. The topology of each preparation was confirmed by agarose gel electrophoresis.
Binding Assays-Filter binding reaction mixtures contained ["HI DNA, 0.1 mg of sonicated salmon sperm DNA/ml, and 0.1 mg of RSA/ml in binding buffer. After an incubation for 20 min at 30 "C, they were filtered through nitrocellulose filters (6) and quantitated by liquid scintillation counting. IHF binding reaction mixtures (10 p l ) contained 5 fmol of ["'PIDNA (each fragment) and 0.1 mg of RSA/ml in binding buffer with 10% glycerol. They were incubated for 20 min a t 30 "C, bromphenol blue was added to 0.001%, and the samples were loaded onto a 5 or 6% polyacrylamide gel (see legends t o Figs. 6 and 7) in TBE buffer. The gel was run for 4 h at 4 "C, dried, and exposed to film.

RESULTS
Properties of ParB-ParB protein has been purified to greater than 99% homogeneity using only two column chromatography steps. The final step, gel filtration ( Fig. 2), removed essentially all the small contaminants present after Bio-Rex 70 ion exchange chromatography (6). The purification was monitored by pars binding activity and yielded high levels of pure protein (about 20 mg from 2 liters of E. coli culture). Using the purified ParB (Fraction IV), the UV absorption spectrum and extinction coefficient were determined (see "Experimental Procedures") to measure protein concentration accurately in subsequent assays. At the absorption maximum of 280 nm the extinction coefficient of ParB was 0.34 (mg/ml)" cm", or 1.31 X M" cm" (monomers). The eluate was assayed for ParB binding activity (6) and protein concentration (22). One unit of ParB is defined as the amount of protein necessary to bind 1 fmol of pALA207 to a nitrocellulose filter (6). In the column profile shown here, 0.66 mg of protein (2.4 X 10' ParB units) injected yielded 0.53 mg of Fraction IV ParB (2.1 X lo6 units). This represents an 80% recovery of protein and 89% of activity. All ParB activity was contained within the major peak a t 8.0 ml. The purified ParB contained no detectable endo-or exonuclease activity when assayed in the presence of magnesium. The positions of molecular mass standards, run before and after ParB purification, are shown. The Stokes radii of-various standards (IgG, 52 A; alcohol dehydrogenase, 46 A; BSA, 35 A; ovalbumin, 28 A; myoglobin, 19 A) were used to estimate (23) that of ParB (inset). Right, gel electrophoresis of ParB. Three pg of the ParB load (Fraction 111) and peak (Fraction IV) were analyzed on a 12% polyacrylamide gel and stained with Coomassie Blue (24). The positions of molecular mass standards, run in the same gel, are indicated in kDa.
The molecular weight of a monomer of ParB, predicted from the DNA sequence (2), is 38,519 (although it runs with 44-kDa proteins on denaturing gels (Fig. 2)). The elution profile from the sizing column (Fig. 2) suggested a native molecular mass of about 140 kDa. However, sedimentation analyses implied a smaller size: ParB sedimented with, or slightly more slowly than BSA (67 kDa) (Fig. 3). Because gel filtration and sedimentation are sensitive to shape as well as to size, the simple interpretation is that ParB forms an asymmetric dimer. Tbe Stokes radius, estimated from the gel filtration data, is 47 A (Fig. 2), and the sedimentation coefficient from glycerol gradients is approximately 4 S. From these values, the native molecular mass of ParB was calculated to be approximately 80 kDa (23), consistent with its existence as a dimer.
T o control for differences caused by buffers and running conditions sedimentation was repeated several times. The results were similar under all conditions tested (i) in 200 mM Na2S04 (the FPLC buffer) and in 150 mM KC1 (the binding assay buffer); (ii) at 4 and 20 "C; (iii) in sucrose and in glycerol gradients; and (iv) with partially purified Fraction I11 (from the Bio-Rex column).
Chemical cross-linking experiments confirmed that ParB forms a dimer (Fig. 4). DSP is a thiol-cleavable cross-linking reagent that reacts with primary amines (mainly lysine residues) (25). ParB (10 pg/ml) was treated with DSP, and the products were separated on SDS-polyacrylamide gels in the absence of 8-mercaptoethanol. ParB was converted rapidly to a product that ran as a dimer-sized smear ( Fig. 4-4). This species is heterogeneous because of intra-as well as intermolecular cross-links. When the samples were electrophoresed in the presence of 2-mercaptoethanol, all cross-links were cleaved (Fig. 4B), converting ParB to a monomer species (now slightly larger because DSP adds molecular weight). Glycerol gradients (1.9 ml of 10-40%) were prepared in binding buffer with 1 mM DTT. ParB Fraction IV (9 pg) was diluted with 150 pg of BSA in binding buffer with 5% glycerol (volume = 80 pl), layered onto the gradient, and centrifuged for 18 h a t 4 "C at 55,000 rpm in a Beckman TLS-55 rotor. Three parallel gradients contained ovalbumin (150 pg) and ParB (9 pg), a mixture of Bio-Rad standards (thyroglobulin, IgG, ovalbumin, and myoglobin; 800 pg total), and ParB (27 pg). Approximately 35 fractions were collected from the bottom of the tubes and assayed for protein concentration and ParB activity. The data represent the ParB activity from the gradient with BSA (approximately 80% of activity was recovered) and the ASS values from the Bradford protein concentration determinations (22) of BSA, ovalbumin (the contribution of ParB to total protein was negligible), and IgG. Thyroglobulin (not shown) was recovered in the very bottom of the tube. The myoglobin peak was determined by polyacrylamide gel electrophoresis to be fractions 23-25. ParB alone or with ovalbumin sedimented to the same position as above. Separate experiments showed that ParB did not affect the sedimentation of either ovalbumin or BSA. The positions of IgG (7 S), BSA (4.3 S), ovalbumin (3.6 S), and myoglobin (2 S) were used to estimate the sedimentation coefficient of ParB at 4 S. ParB behaved identically at the two salt concentrations used routinely in the analyses of ParB; that is, the FPLC buffer and binding assay buffer (Fig. 4). The experiment was repeated at 10-fold higher (100 pg/ml) and 50-fold lower (0.2 pg/ml) concentrations with similar results (in the latter case, protein was detected by immunoblots rather than Coomassie staining; data not shown). Therefore, cross-linking and thus dimer formation were not functions of protein concentration (between 0.2 and 100 pg/ml). It is possible that higher oligomers form but are not cross-linkable in this experiment. To be consistent with the gel filtration and sedimentation data, however, the simple interpretation that ParB exists as a dimer in solution is preferable.
IHF Acts by Binding to and Bending pars-IHF is an accessory protein in P1 partition (6). It is a site-specific binding protein that recognizes the sequence 5' T/C AAnnnnTTGAT A/T (12). The closest match within p a r s is 5' TAACTGACTGTTT, immediately to the left of the DraI site (Fig. 5). DNase I footprinting of pars showed that IHF protected 29 bp including the predicted recognition sequence (Fig. 5). The binding site resembles other footprinted IHF binding sites; it includes the consensus adjacent to an A + T rich region. Goodrich et al. (26) have recently developed a computer program to assess the similarity among published IHF binding sites and generated an extended consensus recognition sequence. Using this program, the IHF site within p a r s shows a similarity score of 54.9, which is within the range of scores of other footprinted binding sites. Deletion of the sequences between Tug1 and DruI yields a version of pars, pars-small (Fig. l), to which ParB still binds, but this activity is no longer stimulated by IHF (6). thus pars-small is insensitive to IHF because the binding site has been removed.
IHF has been shown to bend DNA (27)(28)(29). In the XattP site, for example, IHF creates bends that help to juxtapose X Int protein binding sites properly (30, 31). Goodman and 0.1 mg BSA/ml; and the indicated amount of IHF. They were incubated for 15 min a t 30 "C, MgCI, (to 10 mM) and DNase I (to 50 ng/ ml) were added for 90 s, and the reaction was stopped by the addition of 60 p1 of 0.6 M ammonium acetate, 0.1 M EDTA, and 20 pg of sonicated salmon sperm DNA/ml. After phenol extraction and ethanol precipitation the samples were resuspended in 80% formamide, 10 mM NaOH, 1 mM EDTA, 0.1% xylene cyanol, and 0.1% bromphenol blue and electrophoresed on sequencing gels in TBE buffer. On each gel, the Maxam-Gilbert C+T and A+G sequencing reactions of the same DNAs are the left two lanes. The position of the 13-bp IHF consensus sequence is shown as a filled box, and the heptamers proposed as ParB recognition sites (7) are unfilled boxes. A control fragment (the pBR322 EcoRI-EcoRV fragment) showed no protection by IHF (not shown). R, sequence of p a r s (2; GenBank Accession number X02954) indicating the region protected from DNase I cleavage by IHF as thick black lines. The protection on the lower strand can be defined unambiguously because the flanking bases were clearly unaffected by the presence of IHF (see A ) . The corresponding protection on the upper strand is correct +1 bp on each side. The IHF consensus is shown within the shaded box and the proposed ParB heptamers within the dashed boxes.
Nash have shown recently (32) that an IHF binding site can be replaced by an exogenous bend, arguing that IHF acts through the bend and not by IHF-Int protein-protein contact. The organization of pars, ParB binding sites flanking the IHF binding site (Fig. 5B), strongly suggested that IHF served to bend pars and bring distant ParB sites closer together. Because this bend is a key element in any three-dimensional model of the structure of the partition complex, the action of IHF was explored further. p a r s was cloned into the bending vector pBend5 (see "Experimental Procedures"). Digestion of the resulting plasmids ( p a r s was cloned in both orientations) with a variety of restriction endonucleases produced a series of fragments of identical length, in which p a r s was present in circularly permuted positions along the DNA. DNA-IHF complexes were analyzed by polyacrylamide gel electrophoresis; because the fragments were the same length, differences in mobility should reflect differences in shape. As predicted, IHF bends p a r s (Fig. 6). Fragments cut with three different enzymes (with bound IHF) showed differential mobility, and these differences were also dependent on the orientation of p a r s with respect to pBend5 sequences (Fig. 6A). Theoretically, fragments with the bend in the center will migrate more slowly than those with the bend at the end (33). The average center of the bend is thus extrapolated to be the center of the most slowly migrating fragment. The experiment was repeated with a larger number of restriction enzymes, and the relative mobilities of IHF-bound uersus -free fragments were plotted as a function of the center of each fragment. When plotted in this fashion, both orientations of p a r s gave the same result: the center of the bend mapped within the IHF binding site (Fig. 6B).
p a r s fragments electrophoresed anomalously even in the absence of bound protein. First, they consistently migrated more slowly than would be predicted from their size. In this experiment, the 255-bp p a r s fragments migrated with 310-bp DNA marker fragments (not shown). Second, the permuted set of unbound p a r s fragments did not comigrate with each other (Fig. 6A). Both observations indicated the presence of an intrinsic bend or distortion. The p a r s site, although very A + T rich, does not contain any obvious blocks of adenine residues in phase with the helix, the "A-tract b e n d sequences (34). The shift was insufficient to map this distortion precisely, and it is unknown whether this intrinsic structural feature of pars is important for ParB or IHF binding.
Construction of IHF Binding Site Mutations-The parssmall site described above is missing all sequences upstream of the Dm1 restriction site (Fig. 1) and thus lacks more than just the IHF binding site. IHF binding site mutations were constructed by inserting sequences into the DraI site ( Figs. 1  and 5B). An 8-bp Sal1 restriction enzyme linker insertion created parS::Sal, and a 1.2-kb kanamycin resistance gene insertion produced parS:kan. These "pars-[Dra]" mutations should leave all ParB sites intact but change their spacing by 8 bp and 1.2 kb, respectively. The parS:Sal site was characterized biochemically (Fig. 7). Two different techniques have been used to measure protein binding. IHF binding without ParB was measured by gel mobility shift assays and was about 10-20-fold weaker to parS::Sal than to pars (Fig. 7A). Some specificity was apparently retained, as the parS:Sal fragment was a slightly better substrate than a control fragment (included in the same reaction mixture), measured as disappearance of the unbound fragments with increasing IHF concentration. The bound DNA appeared as a shifted smear, presumably caused by some dissociation of the complexes within the gel. Thus the insertion mutation quantitatively and qualitatively damaged IHF binding. In filter binding assays, ParB could bind to parS:Sal (Fig. 7B), but this binding was insensitive to IHF (Fig. 7C). Since DNA. IHF complexes do not stick to nitrocellulose filters (6, 9) this technique measures ParB binding directly (Fig. 7B) and IHF binding by its ability to stimulate ParB binding activity (Fig. 7C). In the absence of IHF, ParB bound equally to pars-large, parS:Sal, and pars-small (Fig. 7B). Therefore, parS::Sal and pars-small were equivalent ParB substrates in uitro.
IHF Increases ParB Affinity in Vitro-The filter retention assay provides a quick and quantitative measure of ParB activity. Using this assay, IHF increases the affinity of ParB for pars. However, because some protein-DNA complexes do stick poorly to nitrocellulose, it was important to show that IHF increases ParB affinity for p a r s rather than simply ParB affinity for the filters. Therefore, unlabeled plasmids containing pars-large, pars-small, and parS::Sal were asked to compete with "H-labeled pars-large plasmids in the presence of limiting ParB (and excess IHF). The rationale for the competition was as follows. In the absence of IHF, ParB binding to all three p a r s versions was identical (Fig. 7B). Only parslarge was stimulated by IHF ( Fig. 7C; 6). If this stimulation represents increased ParB affinity for pars, then only unlabeled pars-large should compete with "H-labeled pars-large. Conversely, competition by all unlabeled p a r s versions would signify that IHF influences the ability of ParB bound to different versions ofparS to stick to nitrocellulose. The results demonstrated that pars-large was a better ParB substrate than pars-small or parS:Sal because only pars-large was an effective competitor of 3H-labeledparS-large (Table I). Therefore, IHF increases ParB affinity for the wild-type site, parslarge.

The
Previous experiments showed that IHF substantially increased the affinity of ParB for pars (6). These binding experiments were performed at a DNA concentration (3 nM) too high to estimate directly binding constants of the "highaffinity" complex. ParB binding in the presence of IHF was repeated at very low DNA concentration to measure ParB affinity. The dissociation constant, K d , equals the free protein concentration at 50% maximal binding; this is approximately equal to the total protein concentration if the DNA concentration (and thus bound protein) is a negligible proportion of  I pars Competition in vitro Nitrocellulose filter retention assay measuring ParB binding to 3H-labeled pALA207 in the presence of a 2-fold excess of unlabeled competing plasmid. Binding reaction mixtures (20 pl) contained 110 fmol of 'H-labeled pALA207,220 fmol of unlabeled plasmid containing the indicated pars site, 75 nM IHF, and 6 nM ParB (120 fmol of dimer). The mixtures were assembled on ice, and the proteins were added last. ParB was the limiting component; DNA and IHF were in excess. be estimated from the protein concentration at 50% maximum binding. DNA concentration was reduced by including less DNA in a greater reaction volume (2 ml). At 0.5 PM plasmid, 50% was bound at 10 PM ParB (dimer) (Fig. 8A). The apparent K d (apparent because one does not know whether all the protein was active, or the stoichiometry of ParB binding) was lo-" M ParB.
In the absence of IHF binding activity was so low that it was technically impractical to use the same reaction conditions; maximal binding was achieved at 0.4 PM ParB (30 pg/ ml). Under the original reaction conditions (3 nM DNA, 20 pl), 50% of the substrate was bound at approximately 0.1 C~M ParB (Fig. 8B). Nevertheless, the apparent Kd is still much greater than the substrate concentration. Thus by this rationale, IHF stimulates ParB binding by about 10,000-fold.
Nonspecific DNA carrier was omitted to assess the contributions of both specific (to pALA207) and nonspecific (to pBR322) DNA binding to total binding activity. In the presence of IHF and at low ParB concentration, binding was very specific (Fig. 8A). Without IHF, ParB disciminates very poorly between specific and nonspecific DNA although specific binding activity can be analyzed by including unlabeled nonspecific DNA in the reaction (Fig. 8B). Simplistically then, IHF increases specific recognition of p a r s by ParB. I n uiuo, of course, recognition of a specific sequence always occurs in the presence of a vast excess of nonspecific competitor.
DNA.IHF complexes are detected on filters by their ability to stimulate ParB binding to pars ( Fig. 7C;6). At 0.5 PM DNA, the DNA concentration was still too high to measure the Kd for IHF binding to pars in the presence of ParB; 50% of the pars plasmid was bound at 2 PM IHF (data not shown). It was technically difficult to lower the DNA concentration further. On linear fragments, the Kd for IHF was about 10 nM (data not shown). Therefore IHF affinity for pars was greater in the intact partition complex. It is likely that both DNA supercoiling and ParB contribute to this increase in IHF affinity (>5,000-fold; see "Discussion").
IHF Affects ParB Affinity in Viuo-Competition, or incompatibility, assays were used to examine the effect of IHF on partition in uiuo. Experimentally, one tests the ability of a plasmid containing a given version of p a r s t o destabilize a second resident plasmid partitioned by P1 par. The resident plasmid used in all experiments was X-miniP1, partitioned by different versions of pars. Thus the resident pars sites were tested within a par system that was as close to the natural context as possible, with the Par proteins supplied in cis from their natural promoters. The pars-[Dra] insertion mutations were crossed into X-miniP1, producing X-miniP1parS:Sal and X-miniP1parS::kan. These particular IHF-insensitive mutations were used because the original pars-small site cannot be tested directly in X-miniP1; the TaqI-DraI deletion removes the end of the parB gene as well as part of p a r s (Fig. 1). The competing "test" pars sites were pars-large and pars-small. They were cloned into pMF3, a low copy number miniF derivative (17), producing pBEF161 and pBEF162, respectively.
Both X-miniP1parS:Sal and parS::kan behaved similarly in all tests. Both were strongly competed by pBEF16l(parSlarge) and pBEF162(parS-small), even in the presence of IHF (Table 11). In wild-type cells pBEF161 was a better competitor than pBEF162. X-miniPlparS-large, on the other hand, was only competed by pBEFlGl(parS-large). In IHF mutants, pBEFlGl(parS-large) and pBEF162(parS-small) competed with all versions of pars. These results confirm that IHF allows the partition system to discriminate between parslarge and -small. They also show that plasmids partitioned by these IHF-insensitive versions of p a r s were susceptible to competition by pars-large.
The results were similar when the test sites were on high copy number plasmids except that the destabilization was more severe, expected with a higher concentration of competing sites (data not shown; 6,35). As competing sites, parS::Sal (on pBEF146) behaved identically to pars-small (on pBEF127).
These observations are consistent with a direct relationship between the in vitro affinity for ParB and the capacity to

TABLE I1 pars Competition in uiuo
The resident X-miniP1 plasmids were partitioned by the indicated p a r s sites (Fig. 1). Note that pars-large is pars', the wild-type partition site. The stability of each X-miniP1 (tested separately) was measured as percent plasmid retention in the population after 16 ? 2 generations in LB medium with ampicillin (50 pg/ml) (selection maintained for the competing plasmid only) (6). The copy numbers of pMF3, pBEF161, and pBEF162 were identical and unaffected by the presence of X-miniP1 (data not shown). compete in uiuo, if one assumes that ParB occupancy of pars is the rate-limiting step for plasmid pairing. (The models for par site-mediated competition assume that competition depends on, and is proportional to, plasmid pairing (3, 4).) Therefore, pair formation and stability are proportional to ParB affinity. Pairs containing two high affinity sites will be favored over high affinity:low affinity pairs, which will be favored over pairs of two low affinity sites. Consequently, the low affinity site should compete with another low affinity site but not with the high affinity site. Conversely, the high affinity site should always compete with both high and low affinity sites. This is consistent with the behavior of the various pars sites (Table 11). In wild-type cells, pars-large was the high affinity site, and pars-small, parS::Sal, and parS::kan were low affinity sites. In the absence of IHF, parssmall and pars-large became similar competing sites, indicating similar affinities for ParB. These data are consistent with the biochemical studies (Figs. 7 and 8 and Table I) that show that IHF significantly alters ParB affinity for the intact p a r s site.
Although similar, X-miniPlparS-large and X-miniP1-parS:Sal were not identical in the absence of IHF. The latter was always slightly less stable than the former, regardless of the competing plasmid. Although the differences were small, they imply that the spacing changes created by the pars-[Dra] insertions have affected something in addition to IHF binding. This may reflect subtle effects on ParB binding, on subsequent partition steps, or on binding of other (as yet undefined) proteins.
A different type of experiment confirmed that IHF increases ParB affinity for p a r s in. uiuo. In previous studies it was observed that very high levels of ParB specifically destabilize both high and low copy number plasmids containing pars ( 5 ) . The effect was shown to be a segregation defect, explained by the aggregation of ParB-bound plasmids. The destabilization of X-miniP1 was so severe that lysogens could not form colonies. If IHF influences ParB affinity for pars, then removal of IHF should reduce the lysogenization defect. The plasmid pBEF104 contains the parB gene in the trp promoter vector, pRPG48 ( 5 ) . The ratio of colony-forming lysogens in DH5(pBEF104) to DH5(pRPG48) was The experiment was repeated in cells lacking IHF; the lysogenization ratio in DH5hip(pBEF104) to DH5hip(pRPG48) was 0.21. The DH5hip(pBEF104) X-miniP1 lysogens were unstable (data not shown), but they could form small colonies. p-Galactosidase measurements (36) ofparB-1acZprotein fusions The PI Plasmid Partition Complex (transcribed from the trp promoter) showed that ParB expression was not lower in DH5hip cells (data not shown). Therefore, the lack of IHF diminished but did not eliminate the inhibitory effects of excess ParB, evidence that ParB binding to pars was reduced in the absence of IHF.
Topology Strongly Influences ParB Binding-During the development of the ParB binding assay, it was observed that ParB preferred supercoiled plasmid DNA; linear DNA fragments carrying p a r s were poor substrates for ParB. Subsequently, supercoiled, relaxed and nicked circular, and linear forms of the pars-containing plasmid, pALA207, were examined (Fig. 9A). Supercoiled DNA was clearly the best substrate. In addition, however, ParB could distinguish between relaxed circular and linear DNA. Plasmids linearized with different restriction enzymes, and short DNA fragments, gave identical results (data not shown). Competition experiments confirmed that supercoiled DNA was a higher affinity substrate. 3H-Labeled supercoiled pALA207 (pars-large) plasmids were mixed with an excess of unlabeled supercoiled, relaxed circular, or linear forms (Table 111). Both relaxed circular and linear forms were poor competitors compared with supercoiled plasmid although relaxed and linear pALA207 competed slightly better than nonspecific supercoiled pBR322 DNA at high competitor concentration.
It appeared that both superhelicity and plasmid shape (i.e. circular uersus linear shape) influenced the binding activity of 3H-labeled substrates (Fig. 9A). Subsequent experiments indicated that the supercoiling effect required ParB and IHF and that ParB was responsible for the shape discrimination. In the absence of IHF, all circular molecules behaved simiarly, and all were better substrates than linear DNA (Fig. 9B). Thus without IHF, ParB binding was insensitive to supercoiling but sensitive to shape. IHF stimulated ParB binding to all forms (compare the ParB concentrations in Fig. 9, A and B) although the stimulation on supercoiled circles was greater than on relaxed ones. At higher IHF concentrations, supercoiled circles were still better substrates than relaxed ones (Fig. 9C), implying that superhelicity does more than just improve IHF affinity. At the concentration of IHF used in Fig. 9A (25 nM), all IHF binding sites should be occupied fully (Figs. 7A and 9C). Therefore the free energy of supercoiling was used only in the complete partition complex.
The effect of plasmid shape on ParB binding in Fig. 9 seems puzzling since ParB is a site-specific binding protein, and the ends of the linear forms are at least 1.5 kb from pars. The competition experiment in Table I11 provides a preliminary explanation. Although poor competitors of 3H-supercoiled plasmid, linear forms behaved similarly to relaxed circular forms. This observation suggests that ParB affinity for both forms may be similar, but ParB-relaxed circular DNA complexes are trapped more efficiently than ParB-linear DNA complexes on nitrocellulose. This interpretation needs to be confirmed, but it seems the simplest way to rationalize why

0.0
3H-linear molecules appear to be such poor substrates. Nevertheless, the important conclusion here is that superhelicity is required for high affinity protein binding to pars.

DISCUSSION
One of the earliest steps in P1 partition is the assembly of a protein complex at pars. This study describes several biochemical and biological characteristics of this partition complex and its components: ParB, IHF, and the p a r s DNA substrate. In particular, the role of IHF and DNA supercoiling has been examined. (i) IHF greatly increases ParB affinity for p a r s (Fig. 8 and Table I). (ii) IHF binds to and bends p a r s (Figs. 5 and 6). (iii) The IHF binding site is between ParB binding sites ( Fig. 5; 7). Only the sequences downstream of the IHF site (between the DraI and Sty1 sites; Fig. 5) are necessary for low affinity ParB binding in the absence of IHF ( Fig. 7B; 5 , 7) whereas the intact pars-large site is required for high affinity (IHF-stimulated) ParB binding. (iv) The complex containing IHF and ParB strongly prefers a superhelical substrate (Fig. 9). The IHF-induced bend in p a r s will change the proximity and alignment of ParB binding sites with respect to each other. These data suggest that tight ParB binding requires that ParB contact recognition sequences on each side of the IHF site simultaneously (Fig. 5 ) and that this contact requires the bend provided by IHF and the free energy of supercoiling. In this model, the wild-type partition complex contains pars-large DNA specifically wrapped around ParB and IHF. The precise geometry of the partition complex, promoted by IHF and superhelicity, is required for high affinity ParB binding. ParB binding increases IHF affinity as well; thus ParB and IHF bind cooperatively with respect to each other.
The requirements for IHF and supercoiling and the organization of multiple binding sites are seen in the XattP.Int. IHF complex, the intasome. Many elegant studies have established that Int, IHF, and DNA supercoiling cooperate to allow Int protein to contact distant, multiple binding sites in a specific three-dimensional way (27)(28)(29)(30)(31)(32)37). The important function of IHF is probably to provide bends in the DNA substrate (32). Richet et al. (30) have shown that the major role of superhelicity is proper positioning of recombination proteins on their sites to favor formation of the intasome. The observations described in this study and summarized above lead to the proposal that IHF and supercoiling play similar architectural roles at pars and that this is a general feature of IHF-assisted complex formation.
A recent recurring theme in the study of protein-nucleic acid complexes is "higher order nucleoprotein structures," formed by multiple protein-DNA and protein-protein interactions (38). Such structures allow protein-mediated communication of distant DNA sites, or "action at a distance" (38, 39). This study suggests that PI has borrowed IHF from its host to assemble such a structure at pars. This parallels  Competition with different topological forms Nitrocellulose filter retention assay measuring ParB binding to "H-supercoiled pALA207 (pars-large) in the presence of a 2-fold or 10-fold molar excess of unlabeled competing plasmid. Reaction mixtures (20 el) contained 0.55 nM 3H-labeled pALA207, unlabeled plasmid as indicated, 25 nM IHF, and 0.58 nM ParB dimer; they were treated as described in Table I Partition complex formation is strongly stimulated by DNA supercoiling. Although these experiments do not establish how superhelicity affects protein binding the observation that both ParB and IHF are required for this stimulation supports the idea that pars DNA is wrapped around both proteins. Such wrapping should constrain supercoils and thus be energetically favored (40). DNA supercoiling can have other effects, for example, DNA melting and the formation of unusual DNA structures such as cruciforms (41). However, there is no a priori reason to invoke helix-opening events in partition. Therefore, the simplest and most likely role of supercoiling at pars is the proper positioning ParB and IHF at pars, favoring the formation of a higher order structure.
The model of the partition complex should eventually incorporate some additional information. First, important protein-protein contacts and ParB stoichiometry are still speculative. The bend-swap experiments in attP (32) imply that IHF supplies the bend, not specific IHF-ParB contacts. Simultaneous ParB binding to sequences on both sides of the IHF site requires ParB-ParB contact, but the number of ParB molecules involved is unknown. Binding experiments at high DNA concentration (when the equilibrium is pushed toward complex formation) indicated that two dimers are sufficient to bind a pars plasmid to nitrocellulose5 (Table I). However, stoichiometry must be determined on isolated partition complexes.
In addition, the contribution of intrinsically bent or kinked sequences (the DNA secondary structure) has not been assessed. Aberrant electrophoretic mobility has been reported for a variety of DNA sites involved in higher order nucleoprotein structures, such as the attP site (43), the X origin of replication (44), and the Fis binding site (45). It has been suggested that such secondary structure may be important for complex assembly.
Partition without IHF-The fact that partition works without IHF indicates that ParB must bind to pars in vivo in its ' B. E. Funnell, unpublished results. absence. The intracellular concentration of ParB may be high enough to promote some ParB binding. In fact, preliminary experiments with antibodies directed against ParB indicated about 2,000-4,000 dimer~/cell.~ In a cylinder the size of an E. coli cell (3 pm long by 1 pm in diameter), this corresponds to micromolar ParB concentrations. Alternatively, it is possible that some other host factor or factors substitute for IHF in vivo.
Affinity Versus Specificity-The in vivo competition suggested that ParB has similar affinities for parS::Sal, parS:kan, and pars-small because parksmall destabilized both X-miniPlparS-[Dra] insertion mutants (Table  11). I n vitro, pars-small and parS::Sal behaved identically (Fig. 7B and Table I). One experiment by Martin et al. (14) does suggest that pars-small is not exactly identical to pars-[Dra] insertion mutants. They reported that a miniF derivative partitioned by pars-small was not destabilized by coresident X-miniPlparS+ and proposed two alternative competition or incompatibility states for pars, incB' (pars-large) and incBd (pars-small), determined by IHF (14,35). They also showed recently that the incBd specificity difference disappeared when the competing plasmid was of high copy number, using a X-miniP1 with a pars-[DraI] insertion mutation (35). The experiments presented here show that X-miniP1 partitioned by the pars-[DraI] insertion mutants was always competed by plasmids carryingpars-large, regardless of the presence of IHF and of the copy number of the competing plasmid (Table  11). Thus it is very unlikely that IHF is the determinant of the observed specificity difference with the miniF experiment (14). The lack of upstream ParB sequences in pars-small, differences in context (i.e. the vector sequences), or differences in the amount of partition proteins when supplied in trans, may be responsible for the insensitivity of this miniF derivative to X-miniP1. Nevertheless, all experiments with the pars-[DraI] mutants support the idea that the in vivo action of IHF is the same as its role in vitro; it increases ParB affinity for wild-type pars.
Partition with IHF-The observation that wild-type X-miniPl is insensitive to competition by pars-small (Table 11; 6) shows that wild-type partition complexes contain IHF. Therefore, the nucleoprotein structure that IHF promotes is the substrate for subsequent steps in partition. However, ParB binding, rather than this structure, is sufficient (although not optimal) for partition since both IHF and the upstream ParB site are dispensable. It is unclear why P1 has chosen to use IHF to assist ParB and has conserved the complexity of the pars site. It has been suggested that IHF is a link to the physiological state of the cell (12). For example, IHF makes ParB binding sensitive to supercoiling (Fig. 9). There is evidence that intracellular supercoiling varies during transcription (46) and in response to certain environmental stimuli such as osmolarity (47) and anaerobiosis (42). Alternatively, higher order nucleoprotein structures may increase the fidelity of a reaction (38). Because IHF improves ParB affinity for specific sites more than its nonspecific DNA binding activity (Fig. 8) P1 plasmids may be less likely to pair with heterologous plasmids when the ParB IHF .pars complex is used. Finally, one can argue that a strong system is always better than a weak one. In the laboratory, time scales are brief relative to those that exert selection pressure during evolution. It has been observed consistently in this laboratory that the par system is slightly less efficient without IHF (Table 11; 6). In nature, after thousands of generations of growth, this difference in efficiency may yield cell populations practically devoid of the plasmid. Thus a weaker par system may confer little more advantage than no par system at all, explaining the evolutionary pressure to conserve the strong ParB binding site.
Finally, it remains to be determined how the partition complex a t pars interacts with ParA and functions in subsequent events in partition. The identification of other host components and the characterization of the interactions involved are the next steps in defining the process of plasmid partition.