Purification and characterization of HP1 Cox and definition of its role in controlling the direction of site-specific recombination.

The protein that activates site-specific excision of the HP1 genome from the Hemophilus influenzae chromosome, HP1 Cox, was purified. Native Cox consists of four 8.9-kDa protomers. Tetrameric Cox self-associates to octamers; the apparent dissociation constant was 8 μM protomer, suggesting that under reaction conditions Cox is largely tetrameric. Cox binding sites consist of two direct repeats of the consensus motif 5′-GGTMAWWWWA; one Cox tetramer binds to each motif. Cox binding interfered with the interaction of HP1 integrase with one of its binding sites, IBS5. This competition is central to directional control, as shown by studies on mutated sites. Both Cox binding sites were necessary for Cox to fully inhibit integration and activate excision, although Cox continued to affect recombination when the single binding site proximal to IBS5 remained intact. Eliminating the IBS5 site completely prevented integration but greatly enhanced excision. Excisive recombination continued to require Cox even when IBS5 was inactivated. Cox must therefore play a positive role in assembling the nucleoprotein complexes producing excisive recombination, by inducing the formation of a critical conformation in those complexes.

Site-specific recombination systems belonging to the integrase family mediate the joining and separation of a variety of replicons (reviewed in Refs. 1 and 2). These fall into several classes, depending on their function, on the complexity of the DNA substrates, and on the number of proteins involved. The simpler systems, like those of phage P1 (3,4), the yeast 2 plasmid (5,6), and the xer system of Escherichia coli (7), mediate recombination at specific short sequences through the action of a recombinase belonging to the integrase family (8). These systems regulate copy number, unlink daughter molecules, and resolve dimeric species to ensure accurate partitioning of the relevant element during cell division. More complex site-specific recombination systems are responsible for inserting and excising phage genomes from their cognate host chromosomes. At least one DNA substrate for these latter systems includes multiple protein binding sites (9), and proteins other than the recombinase or integrase are required. At least some of these accessory proteins are provided by the host (10,11). This complexity most probably reflects the directionality of these systems; the protein requirements for integrative recombination do not promote the excision reaction and in fact inhibit it, while conditions allowing excisive recombination disfavor the integration reaction (12)(13)(14).
Work in this laboratory is aimed at understanding the mechanism and regulation of a complex site-specific recombination system derived from the temperate phage HP1 of Hemophilus influenzae Rd, a relative of the P2-186 family of phages (15). The integration and excision of HP1 conform in outline to the analogous reactions first analyzed in and its host Escherichia coli (9). The HP1 genome is inserted into a single site in the host chromosome by site-specific recombination, in a reaction executed by HP1 integrase and stimulated by the bacterial DNA bending protein integration host factor (IHF) 1 (16,17). The substrates for integration are the 500-bp phage attachment site, attP (18), and an 18-bp host site, attB (19); the HP1 attB site coincides with the anticodon stem-loop sequence of a tRNA gene, and HP1 is one of a diverse group of integrating elements targeted to tRNA genes (20,21). The integration reaction produces an integrated prophage flanked by the hybrid attachment sites attL and attR, which are the substrates for the excision reaction. Excisive recombination, like integration, requires HP1 integrase and IHF but also requires an additional HP1-encoded protein, HP1 Cox (22). HP1 Cox has two effects on site-specific recombination in vitro. It inhibits the integrative reaction of attP and attB, and it activates the excisive reaction of attR and attL (22). In these two effects are mediated by the phage-encoded DNAbinding protein, Xis (23)(24)(25), in cooperation with the host protein Fis (11,26). Xis controls directionality by altering the relative affinities of Int for two of its binding sites, increasing the affinity for one site (P2) and reducing its interaction with the P1 site; population of the P2 site by Int appears to be vital for excision to occur (27). These changes result from cooperative protein-protein interactions and from changes in bending of the attachment site DNA (24,28,29).
The sequence of HP1 Cox is unrelated to that of Xis. HP1 Cox is, however, clearly similar to the Cox protein of P2 and the Apl protein of phage 186, which belong to the helix-turn-helix class of DNA-binding proteins (30,31). P2 Cox and 186 Apl have been implicated in directional control of recombination and in regulating the expression of early lytic functions (32)(33)(34)(35). A similar regulatory function for HP1 Cox has been identified recently. 2 To understand how HP1 Cox controls the direction of recombination, the protein has been purified to apparent homogeneity, and its physical properties have been characterized. The location on att site DNA where Cox binds has been identified and the affinity and stoichiometry of binding were determined. Competitive footprinting studies combined with the results of attachment site mutations provided evidence for the mechanism by which directionality is controlled in HP1.

EXPERIMENTAL PROCEDURES
Materials-Plasmid DNA was purified with QiaPrep Kits (Qiagen, Inc) and Magic columns (Promega). Media were from Difco, and antibiotics were from Sigma. Protein was assayed by the Bio-Rad protein assay using BSA as a standard. Phosphocellulose (Whatman P-11) was washed and equilibrated according to the manufacturer's instructions. AmpliTaq and other PCR reagents were from Perkin-Elmer, Inc. DNase I was from Life Technologies, Inc. Restriction enzymes, T4 polynucleotide kinase, the Klenow fragment of DNA polymerase I, and T4 DNA ligase were from New England Biolabs. Highly purified HP1 integrase was provided by S. Waninger. E. coli IHF was purified from E. coli K5746 as described (52).
Bacterial Strains and Plasmids-E. coli DH5␣ was grown in solid or liquid LB medium supplemented with antibiotics as needed. The construction of pCOX1, an expression vector containing the cox gene under the control of a heat-inducible promoter, has been described (22). Plasmids containing attachment sites (22) were pHPC120, which contains the HP1 attP site; pHPC121, which contains the H. influenzae attB site; pHPC122, which contains the recombinant attL site; and pHPC123, which contains the recombinant attR site. Plasmid pHPC130 was constructed by amplifying a 220-bp fragment (residues 481-701 of HP1) by PCR using pHPC120 as a template and primers 1.5R and 2.3L. The amplified product was digested with BamHI and SalI and ligated with pUC18 that had been treated with the same enzymes. Similarly, a 247-bp fragment (residues 251-498 of HP1) was amplified using pHPC120 as template and primers 0.7R and 1.5L and was inserted into the pUC18 BamHI and SalI sites to create pHPC140.
Mutations were introduced into individual protein binding sites using a modification of the restriction PCR method (36) with pHPC120 as the template. For each desired mutation, two separate PCR amplifications were performed: one with the pUC M13R primer and the AE19 primer, which removed an EcoRI site from the plasmid, and one with the pUC M13F primer and one of four mutagenic primers (CS1, CS2, CS12, IBS5) that altered the requisite binding sites. Amplified products were separated on 1% low melting agarose, and recovered with Qia-Quick gel extraction columns. The two products were then used as templates in another round of PCR with the M13F and M13R primers. The amplified product of this reaction was again purified on low melting agarose, digested with EcoRI and HindIII, and inserted into pUC18 prepared with the same enzymes. Plasmids susceptible to EcoRI cleavage should contain the desired mutation. Clones were screened for the presence of a restriction site introduced in the mutagenic primer, and positive clones were chosen and verified by sequencing to confirm the presence of the mutation.
SDS-Polyacrylamide Electrophoresis-Samples were precipitated with 7% trichloroacetic acid and resuspended in loading buffer containing 2% SDS. Samples were boiled for 5 min and separated by electrophoresis for 90 min at 160 V using conditions developed to separate low molecular weight proteins (37). Gels were stained with Coomassie Blue R-250, visualized on an Ambis CCD camera, and stored as TIFF files using Adobe Photoshop.
Purification of HP1 Cox-An overnight culture (5 ml) of E. coli DH5␣(pCOX1) was diluted into 1 liter of LB containing carbenicillin (100 g/ml) in a 2.5-liter Fernbach flask. The culture was incubated with vigorous shaking at 37°C. When A 600 ϭ 0.6, the culture was transferred to a 42°C shaker bath, and incubation was continued for 135 min. Cells were collected by centrifugation (5,000 ϫ g for 10 min) and resuspended in a final volume of 50 ml with CG buffer (50 mM Tris-phosphate, pH 7.5, 10 mM EDTA, 3 mM DTT, 10% glycerol, and 0.2 M KCl). To inhibit proteases, phenylmethylsulfonyl fluoride (50 g/ml), pepstatin (1 g/ml), and leupeptin (5 g/ml) were added. The suspension was sonicated on ice in three 3-min bursts at 40% duty cycle and 40% power with a Branson W-350 cell disrupter with a 0.5-inch tip. The sample was then centrifuged for 20 min at 30,000 ϫ g, and the precipitate was discarded.
To the crude extract (45 ml) at 4°C, saturated ammonium sulfate in CG buffer (84 ml) was slowly added, and the mixture was gently stirred for 30 min and centrifuged for 20 min at 30,000 ϫ g; the precipitate was discarded. Saturated ammonium sulfate (172 ml) was added to the supernatant solution with gentle stirring to a final concentration of 85%. The mixture was centrifuged for 20 min at 30,000 ϫ g, the supernatant was discarded, and the precipitate was resuspended in 10 ml of 85% ammonium sulfate in CG buffer and stored overnight at 4°C. The precipitate was collected by centrifuging at 30,000 ϫ g for 20 min, and the supernatant solution was discarded. The precipitate was resuspended in 10 ml of DG buffer (50 mM Tris-phosphate, pH 8.3, 10 mM EDTA, 1 mM DTT, 10% glycerol) containing 0.2 M KCl, and ammonium sulfate was removed by gel filtration using PD-10 desalting columns (Pharmacia Biotech Inc.) according to the manufacturer's instructions.
A column of phosphocellulose (2.5-cm diameter, 20-ml packed bed volume) was equilibrated with 0.2 M KCl in DG buffer, and the dissolved ammonium sulfate fraction (14 ml) was applied to the column at a flow rate of 1.2 ml/min. The column was washed with 80 ml of 0.2 M KCl in DG buffer and then with 60 ml of 0.6 M KCl in DG buffer, and Cox was eluted by washing with 1 M KCl in DG. The peak fractions (usually 10 -14 ml) were pooled and concentrated with an Amicon Centriprep-3, adjusted to a protein concentration of 1 mg/ml with 0.6 M KCl in CG buffer, and stored at Ϫ80°C.
Preparation of Radioactive Cox-pCOX1 was introduced into E. coli BL21(DE3), and the cells were grown at 37°C in MD media (6 g/liter Na 2 HPO 4 , 3 g/liter KH 2 PO 4 , 1 g/liter ammonium chloride, 0.5 g/liter sodium chloride, 120 mg/liter magnesium sulfate, 0.2% glucose, 0.5 mg/liter thiamin, 20 mg/liter L-methionine, 40 mg/liter all other Lamino acids except leucine) to an A 600 of 0.5. L-[ 3 H]leucine (2.5 mCi, 151 Ci/mmol, Amersham Life Sciences) was then added, and Cox expression was induced by temperature shift as above. In this expression system, Cox was completely insoluble. Trial experiments showed that the precipitate dissolved completely in 6 M guanidinium chloride (GdCl) and renatured to give a fully active product. Accordingly, induced cells were resuspended in 5 ml of solubilization buffer (50 mM Tris-phosphate, pH 7.5, 10 mM EDTA, 10% glycerol, 10 mM DTT) containing 6 M GdCl. After 5 min at 0°C, the sample was centrifuged at 30,000 ϫ g for 15 min, the pellet was discarded, and GdCl was removed from the supernatant solution on a PD-10 gel filtration column in solubilization buffer. Solid ammonium sulfate was added to the effluent from the PD-10 column to 65% of saturation, the mixture was stirred at 4°C for 15 min, and the precipitate was centrifuged down (30,000 ϫ g for 20 min) and discarded. The supernatant solution was desalted over a PD-10 gel filtration column. The desalted effluent was fractionated on phosphocellulose, concentrated, and stored as described for the unlabeled protein. The final Cox sample was Ͼ90% pure, and the specific activity was 75,000 cpm/g, or 673 cpm/pmol of Cox monomer.
Extinction Coefficients-The molar extinction coefficients for native and denatured Cox in the ultraviolet were determined by the method of Gill and von Hippel (38). Samples of Cox were extensively dialyzed against either 6 M GdCl or against buffer (50 mM Tris-phosphate, pH 7.5, 15 mM EDTA, 0.2 M KCl, and 10% glycerol). The absorption spectra of the native and denatured proteins were determined. As expected, Cox showed maximum absorbance at 276 nM, corresponding to the absorption maximum of the single tyrosyl residue; Cox contains no tryptophan (22). The extinction coefficient for the denatured protein, calculated from the amino acid composition, was 1450 M Ϫ1 cm Ϫ1 at 276 nM; the absorbance for Cox in buffer at the same concentration corresponded to an extinction coefficient for the native protein of 1547 M Ϫ1 cm Ϫ1 .
To compare the concentration of Cox determined by UV absorption with that obtained in the Bradford assay, a single sample was assayed by both methods. The concentration measured by absorbance at 276 nM was 188 Ϯ 6 g/ml, and that obtained by dye binding with BSA as a standard was 199 Ϯ 8 g/ml, indicating that the latter method accurately measured the Cox concentration.
Gel Filtration-A column (0.8 cm 2 ϫ 29 cm) of fine Sephadex G-100 was packed and equilibrated in a solution containing 50 mM Tris-phosphate, pH 7.5, 10 mM EDTA, 10% glycerol, 1 mM DTT, and 0.5 M KCl. The column was developed with equilibrating buffer at a flow rate of 0.2 ml/min and was calibrated with blue dextran 2000 and standard proteins (lysozyme, ovalbumin, and BSA). Fractions (0.4 ml) were collected. Standard proteins were located by the Bradford assay, and [ 3 H]Cox was assayed by scintillation counting.
Sedimentation Velocity-Linear sucrose gradients (5-20%, w/v) were made as described (39). Stock sucrose solutions were prepared in 50 mM Tris-phosphate, pH 7.5, 15 mM EDTA, 1 mM DTT, and 0.5 M KCl. Samples (200 l) were applied to the top of the gradients. Sedimentation was performed at 4°C in a Beckman SW50.1 swinging bucket rotor for 20.5 h at 47,000 rpm in a Beckman L800 ultracentrifuge, producing an 2 t value of 1.79 ϫ 10 12 radians 2 /s. After centrifugation, the contents of the tubes were collected from the bottom by positive displacement; 2-drop fractions were collected. The volume of each fraction (90 -110 l) varied with the sucrose concentration and was determined from gradients run in parallel for this purpose.
Protein in each fraction was detected by A 280 measurements (for standards), Bradford assays (for unlabeled Cox), or scintillation counting (for [ 3 H]Cox). The method of McEwen was used to calculate sedimentation coefficients (40). Molecular weights were estimated either from the ratio of distances traveled to the ratio of molecular weights or by the formula s 1  At higher concentrations, the data were not well fitted by lines, and the plots showed upward curvature. In these cases, the data were fitted to the following equation (43,44).
based on a monomer-dimer equilibrium (n ϭ 2) in which the "monomer" consisted of four Cox protomers (M ϭ 36,000). The data were analyzed by nonlinear least squares regression performed with Ultrafit software (Elsevier-BIOSOFT, Cambridge, UK). The partial specific volume (v) of Cox, calculated from amino acid composition (45), was 0.740 ml/g at 4°C. Solvent densities () were calculated as described (42). Retardation on Agarose Gels-Plasmids were digested with HindIII and EcoRI, and the fragment mixture was purified using Magic columns (Promega, Inc). DNA fragments (2 g) were incubated with proteins in 0.1 M Tris-phosphate, pH 7.5, 5 mM EDTA, 0.07 M KCl, 15% glycerol, and marker dyes (bromphenol blue/xylene cyanol). After 15 min at 37°C, the samples were separated by electrophoresis on 2.5% agarose gels in 0.5 ϫ TBE buffer (25 mM Tris, 50 mM boric acid, 2.5 mM EDTA) at 4 V/cm for 3 h. The gel was stained with ethidium bromide, destained, and photographed on an Ambis CCD system.
Retardation on Polyacrylamide Gels-Plasmids (pHPC120, -170, or -180) were digested with HindIII, labeled at the 3Ј ends with the Klenow fragment of DNA polymerase I using [␣-32 P]dATP, and purified using Promega Magic columns. The DNA was digested with DraI, and the fragments were separated by electrophoresis on 1% low melting point agarose gels. The 250-bp fragment containing the binding sites was recovered and purified using QiaQuick gel extraction columns. Binding was assayed in mixtures (20 l) containing the DNA fragment (10,000 cpm; 10 ng) in 50 mM Tris-phosphate, pH 7.5, 10 mM EDTA, and 10% glycerol, varying amounts of protein, and KCl to a final concentration of 0.07 M. After 15 min at 37°C, samples were separated by electrophoresis on 10% polyacrylamide gels in 0.5 ϫ TBE buffer (25 mM Tris, 50 mM glycine, 2.5 mM EDTA) at 13 milliamperes constant current for 2 h. The distribution of radioactivity was captured using Fuji BAS1000 imaging plates (0.5-1 h of exposure), and results were quantitated with Fuji MacBAS 2.0 software.
Stoichiometry Measurements-Fragment retardation on polyacrylamide gels was done as above. The specific activity of the DNA was adjusted by dilution with unlabeled fragment to provide approximately 1000 cpm/lane. This minimized spillage of the 32 P signal into the 3 H channel. After separation, the DNA bands were located using imaging plates; gel slices containing bands of interest were excised, transferred to screw-cap tubes, and digested with a solution of 21% H 2 O 2 in 17% (w/v) HClO 4 (0.5 ml/tube) for 14 -16 h at 65°C. Control tubes containing gel slices with known amounts of either 3 H-labeled protein, or [ 32 P]DNA were used to estimate quenching by acrylamide. After incubation, the contents of the tubes were mixed with 15 ml of Ecolite scintillation fluid, incubated for at least 4 h at room temperature in the dark, and counted on a Beckman LS7000 scintillation counter with a dual label counting program. The specific activity of Cox (478 cpm/pmol) and of DNA (3000 -7000 cpm/pmol) were determined from counts obtained from the control tubes. Cox concentrations were determined by the Bradford assay, and the concentrations of DNA were determined spectrophotometrically at 260 nm.
DNase I Footprinting-A 250-bp fragment from pHPC130 was used for these experiments. To prepare fragments with a radioactive top strand, pHPC130 was digested with EcoRI, the ends were filled in with the Klenow fragment of DNA polymerase I and [␣-32 P]dATP (Ͼ3000 Ci/mmol, Amersham Corp.), and the sample was digested with HindIII. The 250-bp singly labeled fragment was purified on a 1% low melting point agarose gel and recovered with the QiaQuick gel extraction kit (Qiagen, Inc). To prepare fragments with a radioactive bottom strand, pHPC130 was first treated with HindIIII, and the ends were filled in by repair synthesis as before and then digested with EcoRI. The DNA fragment was purified as described above and had a specific activity of 1250 cpm/ng.
Singly end-labeled DNA (50,000 cpm; 40 ng) in 100 l of buffer (50 mM Tris-phosphate, pH 7.5, 5 mM EDTA, 20% (v/v) glycerol, and 0.07 M KCl) and the indicated amounts of protein were incubated for 10 min at 37°C; 2 l of buffer containing 0.5 M CaCl 2 and 0.5 M MgCl 2 was added followed by an amount of DNase I determined by trial. The reaction was incubated at 25°C for 2 min and stopped by adding 125 l of a solution of 5 M ammonium acetate, 0.1 M EDTA containing tRNA (20 g/ml). Samples were precipitated with 2.5 volumes of ethanol at Ϫ80°C, centrifuged at 14,000 ϫ g for 20 min, resuspended in 100 l of 0.3 M sodium acetate, and reprecipitated. The precipitate was rinsed with 70% ethanol and dried at 65°C for 10 min. DNA was resuspended in 7 l of Sequenase loading buffer (U.S. Biochemical Corp.) and stored at Ϫ20°C.
Taurine-containing 6% polyacrylamide sequencing gels were prepared as described (46). Gels were prerun in 0.8 ϫ TTE buffer (80 mM Tris, 25 mM taurine, 0.5 mM EDTA) for 45 min at 65 watts constant power on an IBI STS-45 sequencing apparatus. Samples were boiled for 2 min and rapidly cooled in an ice water bath for at least 30 s before loading. When the run was completed, the gel was removed and transferred to Whatman 3 mm paper, dried, exposed to a Fuji BAS1000 imaging plate for 2 h, and scanned with MACBAS 2.0 software.
Recombination Assays-Linear substrates were prepared by digestion of plasmids with EcoRI. The ends were labeled by repair synthesis with the Klenow fragment of E. coli DNA polymerase I and [␣-32 P]dATP. DNA was purified with Magic columns (Promega, Inc).
Integration assay reactions (20 l) contained 50 mM Tris-phosphate, pH 7.5, 15 mM EDTA, 6 mM spermidine, 20% glycerol, 90 mM KCl, 100 fmol of supercoiled pHPC120, 100 fmol of linear pHPC121 (500 cpm/ fmol), and 75 ng of purified E. coli IHF. Purified HP1 integrase (100 ng) was added to start the reaction. The reactions were incubated at 37°C for 30 min in the standard assay; stopped by adding 1% SDS, 100 g/ml proteinase K, and sample loading buffer; and electrophoresed for 3 h on a 0.9% agarose gel at 90 V. The gel was stained with ethidium bromide and dried, and radioactivity was detected and quantitated on a Fuji BAS1000 imaging system.
Excisive recombination assays were performed under similar conditions as described (22); the routine assay used supercoiled pHPC122 and linearized, labeled pHPC123 as substrates; other substrates used were as indicated. The standard amount of HP1 integrase in these reactions was 600 ng. Excision reactions were carried out at 37°C for varying lengths of time depending on the substrate being studied. Excision was determined by taking 20-l samples from a larger reaction mixture at the indicated times.

RESULTS
Purification of HP1 Cox-The HP1 cox gene had been inserted into a high copy number plasmid with its expression controlled by the P L promoter and the heat-inducible cI 857 repressor. Heat induction of E. coli containing this plasmid (pCOX1) produced extracts containing a 9-kDa polypeptide corresponding to the expectation for HP1 Cox; this polypeptide constituted 5-10% of the total protein (22). HP1 Cox was purified from such extracts by the methods described under "Experimental Procedures." The results of a representative purification are summarized in Table I, and the polypeptide composition of the fractions, determined by polyacrylamide gel electrophoresis in the presence of SDS, are presented in Fig. 1. HP1 Cox appeared to constitute over 95% of the final fraction from phosphocellulose chromatography. Between 4 and 6 mg of purified Cox were obtained from each liter of induced culture.
Low concentrations of Cox proved difficult to measure. Its absorption spectrum, predictable from its amino acid composition, has a maximum at 276 nm and is entirely accounted for by its single tyrosyl residue. It has no enzymatic activity, and its effects on site-specific recombination require rather high concentrations. To follow Cox at low concentrations it was labeled with [ 3 H]leucine as described under "Experimental Procedures." E. coli DH5␣ containing the expression plasmid pCOX1 grew very poorly on appropriately supplemented minimal medium. When E. coli BL21(DE3), which grew well on minimal medium was used, Cox was extensively expressed and efficiently labeled but was completely insoluble. This insoluble aggregate dissolved readily in 6 M GdCl, and Cox was renatured by passage over a gel filtration column as described above or by dialysis. The renatured protein eluted from phosphocellulose under the same conditions as did the protein prepared without denaturation and was indistinguishable from standard Cox preparations in recombination assays. Excisive recombination proceeded at 2.8 fmol/min with protein prepared by denaturation and renaturation and 3.0 fmol/min with Cox purified without denaturation.
Molecular Properties of Native HP1 Cox--The amino acid sequence of Cox, deduced from the gene sequence, predicts a molecular weight of 9,100 for the monomeric polypeptide. Limited determination of the amino acid sequence of SDS gelpurified Cox (performed by the Johns Hopkins Protein-Peptide Facility) showed that the N-terminal Met residue had been removed. The protomer molecular weight is therefore 8,950, a value supported by the SDS-PAGE results. To determine the size of the native protein, several hydrodynamic methods were used.
Gel Filtration-Gel filtration of HP1 Cox on Sephadex G-100 at several initial concentrations suggested that Cox was a multimeric protein and that Cox multimers were capable of self-association. The results of one series of measurements are shown in Fig. 2. At low concentrations (0.2 M), Cox eluted as a single peak at a position corresponding to a Stokes radius of 18 Å and a molecular weight of approximately 36,000. These values suggested that native HP1 Cox is a tetramer of 8.9-kDa protomers. At 2 M Cox, a second more rapidly eluting peak was apparent; this second peak corresponded to a Stokes radius of 35 Å and a molecular weight of 67,000, suggesting that Cox had formed octamers or formed dimers of the tetramer. At high concentrations (22 M), Cox fractionated almost entirely as the rapidly eluting species, indicating that it was largely associated into octamers under these conditions. Sucrose Gradient Velocity Centrifugation-Several initial concentrations of Cox were examined by band sedimentation through sucrose gradients (39) together with ovalbumin, lysozyme, and BSA standards; the results are shown in Fig. 3. At low initial concentrations (0.5 and 5 M), nearly all of the protein sedimented slightly more slowly than ovalbumin. However, 25 M samples of Cox sedimented as two well defined species, one comigrating with the peak seen at lower Cox concentrations and the other sedimenting more rapidly than BSA. In the 50 M sample, only the peak sedimenting ahead of BSA was obtained. The sedimentation coefficients (40) for these two species were 5.5 S (for the rapidly sedimenting peak) and 3.0 S  (for the slower peak). The relative molecular weights of these two species were calculated as described under "Experimental Procedures." The value for the 5.5 S peak was 72,700 Ϯ 6,600, and for the 3.0 S peak it was 35,800 Ϯ 3,200, corresponding to 8.1 Ϯ 0.7 protomers and 4.0 Ϯ 0.4 protomers, respectively. These data confirm the results of gel filtration and suggest that Cox is present in an equilibrium of tetramers and octamers, tetramers constituting the principal species at concentrations of less than 5 M Cox protomers.
Equilibrium Sedimentation-To confirm these results, radioactive Cox was subjected to equilibrium sedimentation. At a loading concentration of 0.3 M Cox, plots of ln(c) against the square of the radial distance from the center of rotation could be fit by straight lines, indicating that the protein was monodisperse (Fig. 4A). The ratio of the slope of this line to the slope of the plot for BSA run in parallel (Fig. 4B) gave a molecular weight for Cox of 37,000. This value agrees well with that of the smaller of the two species observed by gel filtration and in velocity sedimentation and corresponds to a tetramer of Cox protomers. Higher initial concentrations of Cox (30 M) produced decidedly nonlinear plots (Fig. 4C), suggesting the presence of multiple species.
To examine this in more detail, data from three initial concentrations (10, 20, and 50 M Cox protomer) were replotted using the nonlinear Lamm equation, which describes the distribution of interacting components in terms of a dissociation constant and a stoichiometry of interaction (44) as described under "Experimental Procedures." The data from two experiments at 50 M Cox monomer are shown in Fig. 5, with the best fit line from the Lamm equation and the final values of the stoichiometry (n) and dimerization constant (K dim ). The data were well fitted by a model in which there was an equilibrium between tetramers (n ϭ 1) and octamers (n ϭ 2) of Cox with an apparent dissociation constant K dim , of 2.0 M (tetramer, equivalent to 8 M protomer). Identical results were obtained when the 10 M and 30 M samples were fitted (data not shown). Values for K dim varied from 1.8 to 2.2 M, and the stoichiometry values were all within an error of 2.0. Attempts to fit the data obtained from runs using a starting concentration of 1 M or 3 M failed to produce realistic solutions using this equation, which would be expected if the protein was largely monomeric at these concentrations.
A sample of Cox was denatured in 6 M GdCl and analyzed by equilibrium sedimentation (not shown). Denatured Cox produced a linear plot with a slope corresponding to a molecular mass of 7.8 Ϯ 1.2 kDa, which agreed with the molecular mass of the Cox protomer.
All these hydrodynamic data strongly indicate that under the conditions used to assay recombination native Cox is a 36-kDa tetramer of four identical 8.9-kDa protomers. At high concentrations, the 36-kDa species self-associates to form octamers or, more properly, dimers of tetramers.
Location of HP1 Cox Binding Sites-Cox most probably affects recombination by binding to attachment site DNA. To approximate the location of the Cox binding site, the ability of the protein to specifically retard the mobilities of a set of DNA fragments derived from the attP site was tested, as described under "Experimental Procedures." The portions of the attachment site used in these assays are shown in Fig. 6A, and the Cox-dependent retardation of a mixture of nonspecific (plasmid backbone) and specific (attachment site) DNA fragments was examined using agarose gels. The results are shown in Fig. 6B. Cox specifically retarded the insert from pHPC120, containing the whole of attP, and that from pHPC130, containing the right arm (bp 481-701) of the site. No specific shifting was detected with pHPC140, which contains the left arm (bp 251-498) of attP. Nonspecific binding, but no specific shifting, of pUC19derived segments occurred at high concentrations of Cox. These results indicate that Cox binds specifically to the portion of the phage attachment site to the right of the strand exchange points; this segment of the attP site is transferred to the attL site upon recombination. Cox binding produced two discrete shifted bands, suggesting that two Cox binding sites are present in this segment.
To locate the HP1 Cox binding sites more precisely, DNase I protection experiments were performed on the right arm of the attachment site, derived from the insert in pHPC130. The ability of Cox to affect the DNase I cleavage pattern on the top strand of attachment site DNA is shown in Fig. 7. The presence of HP1 Cox affected susceptibility to DNase I over ϳ45 bp. As the Cox concentration was increased, cleavage was clearly suppressed in the region between positions 588 and 634. In addition, cleavage was enhanced at positions 590 and 633, near the boundaries of the regions of suppression. These data, combined with the results of footprinting on the bottom strand (shown in Fig. 12) are aligned with the sequence in Fig. 8. This alignment suggests that the direct repeats of 5Ј-GGTMAWWWWA starting at residue 602 are most probably the binding sites for Cox. These two 10-bp repeats, designated CS1 and CS2, are separated by a single base pair. These motifs are indicated in Fig. 8.
Mutational Inactivation of the HP1 Cox Binding Sites-To determine if these motifs in fact constitute the binding site(s) for Cox, mutations were introduced into them as described under "Experimental Procedures," to produce an attP site and an attL site with presumptively defective Cox binding sites. The changes introduced in these constructions are shown in Table II. The effects of these changes were examined by mobility retardation assays on agarose gels, with the results shown in Fig. 6. Fragments containing mutations in CS1 were retarded, but only at Cox concentrations at least 10-fold higher than those needed to completely shift wild-type segments. Only a single shifted species was detectable. The retardation of fragments containing a modified CS2 site could not be distinguished from nonspecific effects, since control fragments were shifted at the same Cox concentrations. The changes introduced appear to have reduced the specific interactions of Cox to numbers inside the rectangles correspond to integrase binding sites. pHPC120 contains the complete attP segment, pHPC130 contains the right half alone, and pHPC140 contains the left half alone. pHPC170 and pHPC180 contain the complete attP site with mutations in the Cox binding sites described under "Experimental Procedures." Panel B examines the ability of Cox to retard DNA fragments from these five plasmids. In all cases, the DNA (2 g) was cleaved with HindIII and EcoRI, incubated with Cox, and separated on agarose gels as described under "Experimental Procedures.  Affinity and Stoichiometry of Binding-The interaction between Cox and its DNA sites was further studied using retardation on acrylamide gels. DNA fragments containing the region surrounding the Cox binding sites at attP were incubated with Cox and separated on polyacrylamide gels as described under "Experimental Procedures." As Cox concentrations increased, an initial retarded band (shift 1) appeared, which was soon accompanied by a second diffuse band (shift 2) at lower mobility. At higher Cox concentrations, complexes accumulated in the wells of the gel, suggesting aggregation due to nonspecific binding. In the concentration range giving specific retardation, binding curves were hyperbolic, as shown in Fig. 9; the Cox concentration at which half the input DNA was retarded was 92 nM (protomer). With DNA fragments without Cox binding sites, high concentrations of protein shifted the DNA to aggregates that remained in the wells (data not shown).
The binding of Cox to variant segments with either the CS1 or CS2 motif inactivated, described above, was examined similarly. The results, shown in Fig. 10, permit several conclusions. First, eliminating either CS1 or CS2 markedly weakened the specific interaction of Cox with the fragments. Modification of CS2 had a more profound effect, since specific retardation of the fragment from pHPC180 required Cox concentrations higher than those producing significant nonspecific retardation. Second, the two motifs are clearly not equivalent, as the results of retardation on agarose also indicated. With the fragment from pHPC170, containing a modified CS1 site and an intact CS2 site, Cox still produced two shifted complexes; higher concentrations of Cox were required to shift these fragments, and the shift 2 complex was more clearly resolved into two separate bands. Aggregation occurred at concentrations lower than those needed to retard 50% of the input DNA and was detectable before the shift 2 band appeared. Retardation of the fragment from pHPC180, containing an altered CS2 motif, also required high Cox concentrations. Only a single shifted species migrating identically with the shift 1 band found with wild-type DNA was produced with this fragment. Because of the nonspecific aggregation, apparent dissociation constants were not measured for this substrate.
The amounts of Cox and of DNA in the shifted species were determined by using [ 3 H]Cox and [ 32 P]DNA. The shifted bands from retardation experiments on acrylamide gels were excised, and radioactivity from each isotope was measured using dual label scintillation counting as described under "Experimental Procedures." Shifted bands from lanes containing different concentrations of Cox from three separate experiments including that in Fig. 9 were used to determine the stoichiometry values; these are summarized in Table III. The results clearly show that the samples from shift 1 contained 4 mol of Cox protomer/ mol of DNA, corresponding to the binding of a single tetramer. Because the shift 2 bands were diffuse, several small slices were taken from the top and bottom of the shift 2 region. In every case, the ratio of Cox to DNA in the shift 2 material was 8:1.
Stoichiometries were also determined for the shifted species shown in Fig. 10, where fragments containing inactivated CS1 and CS2 motifs were analyzed. The results are also summarized in Table III. The two shifted bands produced with the fragment from pHPC170 (CS1 inactivated) had Cox:DNA ratios of 4:1 and 8:1, while the single shifted species observed with pHPC180 (CS2 inactivated) had a molar ratio of 4:1. The aggregated material in the wells contained 36 Cox protomers/mol of 250-bp DNA fragment.
Effects on Recombination-The ability of Cox to inhibit the integration reaction when the attP site contained mutations in either or both Cox binding sites was measured, with the results shown in Fig. 11. When the attP site contained a mutation in the CS2 site, either alone or in combination with a mutated CS1 site, integrative recombination was not inhibited at any

Sequences of HP1 binding sites
The two Cox binding sites, CS1 and CS2, and their consensus and the changes made to inactivate these sites are indicated. Numbers are from Fig. 4. Lowercase residues indicate changes. The sequence of IBS5 is also shown; the two 8-bp direct repeats are underlined, and the sequence of the inactivated IBS5 site is given, with the changed residues in lowercase.  9. Interaction of Cox with the attP DNA fragment containing the CS1 and CS2 binding sites. Retardation of the HindIII-DraI fragment from pHPC120 by the indicated concentrations of Cox (protomer) was measured as described under "Experimental Procedures." The image was captured with a Fuji MacBAS1000 imaging system, and quantitated using MacBAS software. The fraction of DNA retarded by the protein was estimated by quantitating the radioactivity remaining at the unshifted position at each concentration. The gel corresponding to this plot is shown in the left panel of Fig. 10.
Cox concentration tested. Mutations in the CS1 site were less effective in eliminating inhibition by Cox; reactions with this substrate were inhibited approximately 30% at Cox concentrations that gave over 80% inhibition with the wild-type attP substrate.
Mutations in the Cox binding sites affected excisive recombination similarly. The time course of excision was measured at two levels of Cox, with attL substrates containing mutated Cox binding sites; the results are shown in Table IV. A defective CS2 site on the attL substrate eliminated any detectable excision. When the attL substrate contained an intact CS2 site and a mutated CS1 site, the rate of excision was 45% of that found with the wild-type substrate.
HP1 Cox-mediated Inhibition of Integrase Binding-The Cox binding sites in the attachment region are very close to the integrase binding site IBS5. The footprint produced by HP1 integrase at IBS5 extended to residue 616 on the top strand and to residue 626 on the bottom strand (47). This boundary lies well within the Cox footprint and extends to the CS2 site motif. This overlapping of footprints suggested that the binding of Cox might affect the binding of integrase to IBS5. To examine this question, DNase I footprinting was carried out in the presence of HP1 integrase with or without the addition of Cox; the results of one such experiment are shown in Fig. 12. The strong footprint observed at the IBS5 site in the presence of integrase alone was diminished by increasing concentrations of Cox; at the same concentrations, protection of the CS1 and CS2 sites by Cox was apparent. At the highest levels of Cox, the IBS5 footprint was almost entirely eliminated. At these concentrations, integrase continued to produce protection of the IBS4 and IBS6 sites, and there was no evidence of any novel interactions of integrase with attachment site DNA. These data indicate that the effect of Cox on integrase binding was confined to the IBS5 site and consisted entirely of preventing the interaction of the recombinase with this site.
Cox Activity in the Absence of IBS5-The simplest hypothesis to account for the above results is that Cox both inhibits integration and activates excision by displacing integrase from IBS5. To test this hypothesis directly, mutations eliminating the IBS5 site from the attP site and from the attL site were constructed; the changes made are shown in Table II. Introducing a mutant IBS5 site into the attP substrate had a profound effect on recombination; less than 0.2% of substrate was converted to product after 16 h of incubation, a level that is at the limit of detection for this assay. As expected, the addition of Cox to integration reaction mixtures containing the mutant attP substrate had no detectable effect. These results support the proposal that Cox inhibits integrative recombination by displacement of integrase from IBS5, since preventing integrase from binding to this site by mutation completely eliminated attP activity.
To determine the effects of eliminating IBS5 from attL on the excision reaction, this substrate was constructed and tested, with a rather surprising result. The attL substrate containing a mutant IBS5 was not only competent for excisive recombina- FIG. 10. Interaction of Cox with wild-type and mutant attP DNA. Conditions were as in Fig. 9, with Cox concentrations as indicated. The left panel shows the results with the fragment derived from pHPC120 (wild type), and the right panel shows the results using fragments containing mutated Cox binding sites. The six lanes to the left contained a fragment from pHPC170 with mutations in the CS1 site, while the six lanes to the right of this panel contained the fragment from pHPC180, with mutations in the CS2 site. tion but was a considerably better substrate than the wild-type attL, as shown by the results in Table IV. Excision with the mutant IBS5 attL was at least 15-fold greater than with the wild-type attL substrate, and this difference was manifested at two concentrations of Cox. Fig. 13 compares the amounts of integrase and of Cox required for optimal recombination using the wild-type and mutant attL substrates. The increased reaction occurred without any change in the optimal amount of integrase needed for excision (Fig. 13B). However, with the attL substrate lacking IBS5, much lower levels of Cox were required for optimal activity (Fig. 13A). It is important to note that the data of Fig. 13 clearly show that Cox was required for excisive recombination even when IBS5 was eliminated from the substrate, since no excision was detected in the absence of the protein. When the IBS5 mutation was combined with the Cox binding site mutations, the same pattern was observed as with the integration assay, as shown in Table IV; mutations in CS2 completely eliminated excision, while mutations in CS1 reduced activity slightly more than 2-fold. These results argue that Cox does not activate excision simply by affecting the interaction of integrase with IBS5. The interaction of Cox with attachment site DNA must produce an additional change required for the excision reaction to occur.
The HP1 excision reaction requires at least one supercoiled substrate (22). All of the excision measurements described employed a supercoiled attL substrate and a linear attR substrate;  this combination is between 2-and 3-fold more effective a substrate pair than the alternative. It is important to emphasize that the effects of mutating the various binding sites were proportionately the same when the topological states of the two substrates were exchanged, and the supercoiled substrate contained the attR site. DISCUSSION A fundamental property of complex site-specific recombination systems is the capacity to control the direction of reaction. Directional control prevents the products of one round of recombination from undergoing the reverse reaction under the same conditions. In HP1 recombination this control is mediated by HP1 Cox. As described here, Cox was readily purified to apparent homogeneity from a strain of E. coli overexpressing it. Physical measurements showed that native HP1 Cox has a molecular weight of 36,000 and consists of four protomers. At high concentrations, tetrameric Cox self-associated to form octamers; under the conditions optimal for site-specific recombination, the tetrameric form of HP1 Cox should predominate.
HP1 Cox protected a specific region of attachment site DNA from attack by DNase I; this region contains two close direct repeated copies of 5Ј-GGTMAWWWWA; this consensus occurs nowhere else in the HP1 attachment site. The early promoter cluster contains three directly repeated copies of this consensus, and Cox appears to regulate early gene expression by interacting with this cluster. 2 These are the only instances where a closely spaced pair of consensus sequences occur within the HP1 genome. Mutations in this consensus sequence in attachment site DNA prevented specific Cox binding and eliminated the effects of Cox on recombination.
Gel retardation experiments showed that each consensus motif was bound by a single Cox tetramer. Binding to the paired sites was not cooperative, and it occurred in two steps. The appearance of the second shifted species as a diffuse doublet with a constant ratio of Cox to DNA suggests that this species may exist in different conformations. One likely candidate for this conformational change in the doubly bound species is the association of the two Cox tetramers into an octameric complex. The CS2 site appears to be bound more tightly than the CS1 site, implying that the residues where the two sites differ must contribute to the strength of the interaction. The presence of Cox at CS2 appeared to enhance the specific interaction of Cox with a mutated CS1 motif, since a second specifically shifted complex was formed on this fragment, albeit at considerably higher protein concentrations. This suggests that the unchanged residues in the mutated CS1 site may be sufficient to align a second Cox molecule properly despite the reduced affinity. The converse, enhancement of binding to a mutated CS2 by binding to the CS1 motif, was not detectable under the conditions tested. Studies on modified binding sites in which the spacing between motifs is varied, as well as measurements of Cox-induced bending of DNA, will be needed to clarify the various modes of Cox binding.
The binding sites for HP1 Cox, P2 Cox, and 186 Apl all appear to have a center to center spacing of 10 or 11 base pairs. The CS1 and CS2 sites, written as GGTM-N 6 -7 -GGTM, clearly present common features of each motif on the same face of the DNA helix; the putative sites for P2 Cox and 186 Apl are also arranged with the repeated motifs separated by approximately one helical turn (33,34). This arrangement has been reported for other proteins with helix-turn-helix motifs, including the CII (48) and P22 C1 proteins. The P22 C1 protein is a tetramer of 10-kDa subunits with a sedimentation coefficient of 3.3, similar to that of HP1 Cox (49). The stoichiometry of binding of P22 C1 to DNA has not been determined, so it is not clear whether this protein shares the intriguing combination of char-acteristics exhibited by HP1 Cox, where the tetramer binds singly to an asymmetric DNA site.
The DNA segment protected by Cox overlaps the footprint produced by HP1 integrase at one of its binding sites, the type II site 5 or IBS5. This overlap in footprints raised the possibility that Cox affected the interaction of HP1 integrase with this site. Competitive footprinting experiments showed that this was indeed the case. In the presence of increasing concentrations of Cox, the footprint produced by integrase at IBS5 was eliminated, while the footprints at the other integrase binding sites on this arm of the attachment site (IBS4 and IBS6) were unchanged.
Since the only effect of Cox in these experiments was to reduce or eliminate integrase binding to IBS5, mutations were introduced at this site and at the Cox binding sites. These modified attachment sites were assayed to establish the mechanism of action of Cox on recombination. Inactivation of the IBS5 site completely abolished integrative recombination, indicating that this site is absolutely required for reaction in this direction. Elimination of this site from the attL substrate did not interfere with excision but rather produced a 20-fold increase in the rate of excision. Occupation of IBS5 by integrase is clearly not required for excision, and it most likely inhibits reaction in this direction. The binding of HP1 integrase to IBS5 must therefore act as the fundamental switch in determining whether integration or excision is possible. At the same time, it is clear that displacing integrase from IBS5 is not the sole function of Cox in controlling directionality, since Cox is absolutely required for excision even when IBS5 is removed from attL. Cox therefore must play a structural role in organizing the excision-competent nucleoprotein complex on attL. Furthermore, Cox must affect interactions between integrase and DNA when bound to its site, and not by directly interacting with free integrase.
These conclusions clarify the mode of directional control in HP1 recombination and permit its comparison with the analogous processes operating in other phages. In the case of and its close relatives, directionality is mediated by the phageencoded Xis protein. Xis, in combination with the host Fis protein, binds to specific sites within the attachment site (11,24,25). This binding increases the binding of Int to a site immediately adjacent to the Xis site while reducing its interaction at a more distal site (27). In essence, Xis binding reverses the relative affinities of Int for these two sites. These changes make the attP intasome incompetent for integrative recombination while activating the attR nucleoprotein for excision. Directional control in HP1 differs in important respects from this pattern. The chief difference is that HP1 Cox has only a negative effect on integrase binding, displacing the recombinase from a site where it is essential for integration and inhibitory for excision. No additional or novel binding sites for integrase are recruited by binding of Cox. At the same time, Cox does have a positive role in activating the excision reaction; these positive effects of Cox must be the direct consequence of its binding to attL. These results indicate that the excision reactions of HP1 and resemble each other less than the corresponding integration reactions do. This is not surprising, since the suite of proteins involved in HP1 integration are clearly related to their counterparts, while HP1 Cox is unrelated to Xis.
Because the sequence and genetic organization of HP1 make it a member of the P2-186 family of phages, it is of interest to compare our results with what is known about directional control in this group. HP1 Cox is clearly related to P2 Cox and 186 Apl; these proteins possess helix-turn-helix motifs, and are multifunctional, since they regulate early transcription as well as prophage excision (33,35). Initial studies in vitro showed that P2 Cox is similar to HP1 Cox in activating excision and inhibiting integration (35). Since the P2 Cox footprint overlaps one binding site for P2 Int, it seems likely that P2 Cox interferes with Int binding at this neighboring site. HP1 Cox and P2 Cox may inhibit integration by the same mechanism, displacing integrase from a key binding site. The second effect of Cox on HP1 recombination, the activation of the excisive reaction, most probably involves a specific conformation of attL induced by Cox binding. This positive effect on recombination may involve mechanisms specific to the individual system. This is because the number and arrangement of binding motifs for the excision protein differ in the three systems, as do the structures of HP1 Cox and Apl. Six potential Cox binding sites arranged as two sets of three closely spaced 9-bp direct repeats are present in the P2 attachment region (50), while Apl appears to bind to five directly repeated motifs in the 186 attachment region (51). These more numerous sites produce more extensive footprints covering 70 (P2; Ref. 35) to 80 bp (186; Ref. 33), in contrast to the 45-bp footprint produced by HP1 Cox on its two motifs. HP1 Cox and Apl also differ in their native structures. As shown here, HP1 Cox is a tetramer of identical subunits, while Apl is monomeric at all concentrations examined (51); the size of P2 Cox has not been reported. The more numerous motifs in the 186 site imply that several of these must be occupied to activate 186 excision. Presumably, each motif is occupied by a single Apl protomer, which contrasts with the presence of a tetramer of HP1 Cox on each of its motifs. It is possible that Apl might self-associate once bound to DNA to produce an arrangement resembling that produced by HP1 Cox. However, the available data indicate that distinctive DNA-protein interactions characterize each of these systems, suggesting that different nucleoprotein conformations produce excisive recombination in each case. It must be borne in mind that these excision-activating proteins also function as regulators of early transcription. Their structures and properties are presumably compromises reflecting the detailed requirements for discharging both functions, which must be integrated with the cellular processes of different bacterial hosts.
Understanding the precise mechanism by which HP1 Cox activates excisive recombination will require a complete description of the effects of its binding on the conformation of attL DNA. While simple steric interference is sufficient to account for its effects in inhibiting integrative recombination, this aspect of HP1 Cox function may also be mediated by changes in the conformation of one arm of the attP intasome. Studies to address these questions are in progress.