Sequence Specificity and Biochemical Characterization of the RusA Holliday Junction Resolvase of Escherichia coli *

The RusA protein of Escherichia coli is an endonucle- ase that resolves Holliday intermediates in recombination and DNA repair. Analysis of its subunit structure revealed that the native protein is a dimer. Its resolution activity was investigated using synthetic X-junc- tions with homologous cores. Resolution occurs by dual strand incision predominantly 5 (cid:42) of CC dinucleotides located symmetrically. A junction lacking homology is not resolved. The efficiency of resolution is related in-versely to the number of base pairs in the homologous core, which suggests that branch migration is rate-lim-iting. Inhibition of resolution at high ratios of protein to DNA suggests that binding of RusA may immobilize the junction point at non-cleavable sites. Resolution is stim- ulated by alkaline pH and by Mn 2 (cid:49) . The protein is un-stable in the absence of substrate DNA and loses (cid:59) 80% of its activity within 1 min under standard reaction condi- tions. DNA binding stabilizes the activity. Junction resolution is inhibited in the presence of RuvA. This obser- vation probably explains why RusA is unable to promote efficient recombination and DNA repair in ruvA (cid:49) strains unless it is expressed at a high level. The RusA protein coli an that Purification of RusA— Frozen cells (70 ml) from 5 liters of induced culture were thawed at room temperature and then mixed on ice with 25 ml of 5 M NaCl, 0.5 ml 200 m M DTT, 1 ml of 10% (v/v) Triton X-100. Preliminary trials revealed that a high salt concentration was required during cell lysis to reduce loss of RusA. Cells were lysed by freezing in liquid nitrogen and rethawing, and cell debris was removed by centrifugation at 40,000 rpm for 60 min at 4 °C, using a Kontron TST 41.14 rotor. Despite the presence of high salt, (cid:59) 70% of the RusA was removed with the cell debris. All subsequent steps were carried out at 4 °C. Purification of RusA was monitored throughout by SDS-PAGE analysis of column fractions and by assays for junction cleavage activity, using J12 DNA as a substrate. The cleavage assays also provided a measure of nonspecific nuclease contamination. The supernatant (85 ml, 978 mg of protein) fraction containing the soluble RusA was dialyzed overnight against 4 liters of buffer A containing 1 m M phenylmethylsulfonyl fluoride and 0.5 M KCl, centrifuged at 15,000 rpm for 10 min to remove traces of insoluble material, and applied to a phosphocellulose column (2.6 (cid:51) 8.5 cm, 45-ml bed volume) equilibrated in the same buffer. The column was washed with 100 ml of the same buffer before eluting bound proteins with a 450 ml of gradient of 0.5–1.0 M KCl in buffer A. RusA eluted between 0.7 and 0.9 M KCl. Peak fractions were pooled and dialyzed against 2 liters of buffer A containing 0.1 M KCl. SDS-PAGE analysis revealed that RusA was (cid:46) 90% pure at this stage. Minor protein bands and trace exonuclease contaminants were removed by fractionation on three further columns. The dialyzed phosphocellulose pool (144 ml, 47 mg of protein) was loaded on a DEAE Bio-Gel A column

The RusA protein of Escherichia coli is an endonuclease that resolves Holliday intermediates in recombination and DNA repair. Analysis of its subunit structure revealed that the native protein is a dimer. Its resolution activity was investigated using synthetic X-junctions with homologous cores. Resolution occurs by dual strand incision predominantly 5 of CC dinucleotides located symmetrically. A junction lacking homology is not resolved. The efficiency of resolution is related inversely to the number of base pairs in the homologous core, which suggests that branch migration is rate-limiting. Inhibition of resolution at high ratios of protein to DNA suggests that binding of RusA may immobilize the junction point at non-cleavable sites. Resolution is stimulated by alkaline pH and by Mn 2؉ . The protein is unstable in the absence of substrate DNA and loses ϳ80% of its activity within 1 min under standard reaction conditions. DNA binding stabilizes the activity. Junction resolution is inhibited in the presence of RuvA. This observation probably explains why RusA is unable to promote efficient recombination and DNA repair in ruvA ؉ strains unless it is expressed at a high level.
The RusA protein of Escherichia coli is an endonuclease that resolves Holliday intermediates made during homologous genetic recombination and DNA repair (1). Proteins of this type were first described in bacteriophage-infected bacterial cells (2)(3)(4)(5), but have since been found in bacteria (6,7), yeast (8 -10), and mammalian cells (11,12). The T4 enzyme, endonuclease VII, cleaves a variety of DNA secondary structures including Holliday junctions (2), cruciforms (13), branched structures (14, 15), heteroduplex loops (16), and mismatches (17). This broad substrate spectrum is consistent with a role both in recombination and in the removal of branch structures from the phage DNA prior to packaging (18). RuvC protein of E. coli is highly selective for Holliday junctions (19 -21), which suggests a more specialized role in recombination. It binds the four-way duplex structure of a Holliday junction and folds the DNA in an open configuration that allows symmetrical strands to be located within the catalytic core of each subunit and cleaved to yield nicked duplex products (22)(23)(24). Strand cleavage has been shown to involve four acidic residues in the protein (25,26) and occurs 5Ј of a T that has to be located within a certain sequence context for efficient resolution (27,28). This sequence-dependence reflects both topological factors that dictate folding of the junction (28) and the need to make specific protein-DNA contacts (29).
RusA was discovered through its ability to promote recombination and DNA repair in strains lacking RuvC. Genetic analysis of this alternative resolvase provided the first indication that RuvC cannot cleave junctions efficiently in vivo without the associated activities of RuvA and RuvB (1,30). These two proteins assemble at junctions to form a highly specialized RuvAB complex that drives the branch point along the DNA (31)(32)(33)(34). They presumably help RuvC to target cleavable sequences by holding the junction in an open configuration (RuvA) and moving it along the DNA (RuvB), possibly through formation of a RuvABC "resolvasome" complex (34,35). RusA has no requirement for RuvAB. However, efficient recombination and DNA repair does need RecG which like RuvAB drives branch migration of Holliday junctions (30, 36 -38). Whether or not this reflects a need to locate junctions at cleavable sequences remains to be determined.
The 14-kDa RusA polypeptide is encoded by a gene (rusA) located within the defective prophage, DLP12, and is probably of bacteriophage origin (36). The rusA gene is normally expressed poorly, if at all, and can be deleted with no apparent effect. However, it can be activated to suppress the ruv mutant phenotype by insertion of either IS2 or IS10 (formerly called rus-1 and rus-2 mutations, respectively) upstream of the gene to promote transcription, or by cloning in a multicopy plasmid (1,30,36). Expression from multicopy plasmids is needed to achieve full suppression in the presence of RuvA, which suggests that the binding of Holliday junctions by RuvA can prevent their resolution by RusA (1,30). In this work, we detail a new purification of RusA and investigate the native form of the protein. We also report its general biochemical properties, its sequence specificity, and the inhibition of its resolution activity by RuvA.
Enzymes and Reagents-RuvA (40) and RuvC (41) were purified as described elsewhere. T4 polynucleotide kinase was from Pharmacia Biotech Inc., and [␥-32 P]ATP from Amersham. All other reagents were from Sigma or BDH and were of analytical grade.
Chromatography-Heparin-agarose, Reactive Blue 4-agarose, and double-stranded DNA cellulose were purchased from Sigma. Phosphocellulose P11 was from Whatman, and DEAE Bio-Gel A from Bio-Rad. A pre-packed gel filtration column (Superose 12 HR 10/30) was from Pharmacia. All chromatography was carried out at 4°C in buffer A (20 mM Tris-HCl, pH 8.0, 1 mM EDTA, 0.5 mM DTT, 1 10% glycerol). Phenylmethylsufonyl fluoride and KCl or NaCl were added to buffer A as indicated.
Overexpression of RusA-500-ml batches of strain N3757 transformed with pAM151 were grown with aeration at 37°C in Luria-Bertani broth containing 10 g/liter NaCl, 100 g/ml ampicillin, and 20 g/ml chloramphenicol. At a cell density corresponding to an A 650 of 0.5, RusA was induced by adding isopropyl-1-thio-␤-D-galactopyranoside to a final concentration of 2 mM and incubating for a further 3 h. The cells were chilled on ice, harvested by centrifugation, resuspended in 10 ml of lysis buffer (100 mM Tris-HCl, pH 8.0, 2 mM EDTA, 5% glycerol), and stored at Ϫ80°C until required. Under these conditions, RusA was induced to 15-30% of the total cell protein.
Purification of RusA-Frozen cells (70 ml) from 5 liters of induced culture were thawed at room temperature and then mixed on ice with 25 ml of 5 M NaCl, 0.5 ml 200 mM DTT, 1 ml of 10% (v/v) Triton X-100. Preliminary trials revealed that a high salt concentration was required during cell lysis to reduce loss of RusA. Cells were lysed by freezing in liquid nitrogen and rethawing, and cell debris was removed by centrifugation at 40,000 rpm for 60 min at 4°C, using a Kontron TST 41.14 rotor. Despite the presence of high salt, ϳ70% of the RusA was removed with the cell debris. All subsequent steps were carried out at 4°C. Purification of RusA was monitored throughout by SDS-PAGE analysis of column fractions and by assays for junction cleavage activity, using J12 DNA as a substrate. The cleavage assays also provided a measure of nonspecific nuclease contamination. The supernatant (85 ml, 978 mg of protein) fraction containing the soluble RusA was dialyzed overnight against 4 liters of buffer A containing 1 mM phenylmethylsulfonyl fluoride and 0.5 M KCl, centrifuged at 15,000 rpm for 10 min to remove traces of insoluble material, and applied to a phosphocellulose column (2.6 ϫ 8.5 cm, 45-ml bed volume) equilibrated in the same buffer. The column was washed with 100 ml of the same buffer before eluting bound proteins with a 450 ml of gradient of 0.5-1.0 M KCl in buffer A. RusA eluted between 0.7 and 0.9 M KCl. Peak fractions were pooled and dialyzed against 2 liters of buffer A containing 0.1 M KCl. SDS-PAGE analysis revealed that RusA was Ͼ90% pure at this stage. Minor protein bands and trace exonuclease contaminants were removed by fractionation on three further columns. The dialyzed phosphocellulose pool (144 ml, 47 mg of protein) was loaded on a DEAE Bio-Gel A column (2.6 ϫ 4.5 cm, 24-ml bed volume) and washed with 50 ml of the same buffer. Bio-Gel A does not bind RusA under these conditions, but retains contaminants that do not fractionate from RusA in later chromatographic steps. The flow-through (ϳ200 ml, 32 mg of protein) was applied in aliquots of 50 ml to a double stranded DNA-cellulose column (1.0 ϫ 5.1 cm, 4-ml bed volume), washed with 8 ml of buffer A containing 0.25 M KCl, and bound proteins eluted with a 40-ml gradient of 0.25-1.0 M KCl in buffer A. RusA eluted between 0.5 and 0.7 M KCl. Peak fractions from the four aliquots were pooled (80 ml, 28 mg of protein), dialyzed against 2 liters of buffer A, and applied to a heparinagarose column (1.0 ϫ 7.6 cm, 6-ml bed volume). The column was washed with 12 ml of buffer A containing 0.25 M KCl and bound proteins eluted with a 60-ml gradient of 0.25-1.0 M KCl in buffer A. RusA eluted between 0.5 and 0.7 M KCl. No contaminating bands could be detected in these fractions with Coomassie Blue or silver-staining of SDS gels. The peak fractions were pooled and dialyzed into storage buffer (20 mM Tris-HCl, pH 8.0, 1 mM EDTA, 0.5 mM DTT, 50% (v/v) glycerol), and stored in aliquots at Ϫ80°C.
Protein Concentrations-All protein concentrations were determined by a modified Bradford method, using a protein assay kit from (Bio-Rad) and bovine serum albumin (Pharmacia) as standard. Amounts of RuvA, RuvC, and RusA are expressed as moles of the monomeric protein.
Oligonucleotides-Oligonucleotides were synthesized on an Applied Biosystems 380B DNA synthesizer using cyanoethyl chemistry. Each oligonucleotide was deprotected, precipitated in ethanol, and purified on a 12% (w/v) polyacrylamide gel containing 7 M urea. The bands containing full-length oligonucleotides were cut out and extracted from the gel by soaking in water overnight.
Construction of Junction DNA Substrates-X-junctions were made by annealing four partially complementary oligonucleotides, each of approximately 50 nucleotides in length. The point of strand crossover is either fixed centrally within the structure (static junction, J0) or free to branch migrate within a central core of homology (mobile junctions J2, J3, J4, J11, J12, and J26). The number after the letter J indicates the size of the homologous core in base pairs. The sequences of the oligonucleotides used for J0 (35), J2 (26), J3 (35), J4 (35), J11 (27,35), J12 (37,44), and J26 (45) have been described. J4 was made by annealing oligonucleotides 5, 7, 9, and 10 (35). J0 has been referred to elsewhere as X-static (35) and J11 as junction A (27). Annealing followed the procedures described (46). One of the strands was labeled at the 5Ј end prior to annealing using [␥-32 P]ATP and T4 polynucleotide kinase. Junctions were purified by nondenaturing 10% PAGE and electroelution. Unlabeled junctions were made as described above, except that oligonucleotides were annealed in equimolar ratios and the resulting substrates were not purified by PAGE. Unlabeled linear duplex DNA was made in the same way using oligonucleotides 1 and 5 described previously (37).
Junction Cleavage Assays-Cleavage of 32 P-labeled junction DNA by RusA and RuvC was assayed at 37°C in buffer CB (25 mM Tris-HCl, pH 8.0, 1 mM DTT, 100 g/ml bovine serum albumin, 6% (v/v) glycerol) or SCB (25 mM Tris-HCl, pH 8.0, 1 mM DTT, 100 g/ml bovine serum albumin, 10% (v/v) glycerol), with either MgCl 2 or MnCl 2 added as indicated (10 mM in standard resolution reactions). Reactions (20 l final volume) were terminated by adding 5 l of stop mixture (2.5% SDS, 200 mM EDTA, 10 mg/ml proteinase K) and incubating for a further 10 min at 37°C to deproteinize the mixture. DNA products were analyzed by native PAGE, using 10% gels in TBE (90 mM Tris borate, pH 8.0, 2 mM EDTA). Gels were dried, and labeled products were detected using a Molecular Dynamics PhosphorImager (Model 425) and by autoradiography. Reactions were quantified using ImageQuant software (Molecular Dynamics) to analyze PhosphorImages. For time courses, 20-l samples were removed at intervals from bulk reactions and processed as described. To assess activity over a wide range of pH, assays were conducted in a series of different buffers, each covering a narrow range of pH, with a pH overlap between each buffer type.
Mapping Cleavage Sites-Separate preparations of junction DNA, each 5Ј-32 P-labeled in a different strand, were incubated with RusA in SCB buffer for 10 min at 37°C. Deproteinized samples were extracted with phenol, and the DNA was precipitated with ethanol, resuspended in sequencing gel loading dye (0.3% (w/v) bromphenol blue, 0.3% (w/v) xylene cyanol, 10 mM EDTA, pH 7.5, 97.5% formamide), and denatured by boiling for 5 min before analyzing by denaturing PAGE, using 12% gels (42). Sequencing ladders of each labeled oligonucleotide generated using a Maxam-Gilbert sequencing kit (Sigma) were loaded on the same gel to provide markers. Gels were dried, and analyzed by autoradiography and PhosphorImaging. Cleavage sites were mapped by reference to the sequencing ladder. A 1.5-base allowance was made to compensate for the nucleoside eliminated in the sequencing reaction.

RESULTS
Purification and Physical Analysis of RusA-In previous work, we described a recombinant plasmid (pAM151) for the overexpression of RusA and purified some of the protein from strain N3757 transformed with this construct (1). To obtain larger quantities of the protein for physical analyses, we devised a new purification ("Materials and Methods"). This yielded 24 mg of RusA (in 21 ml) from 5 liters of induced cells. The recovery was much lower than expected given that RusA accounted for 15-20% of the total cell protein. A large fraction of the induced protein was in an insoluble form that was removed with the cell debris during centrifugation of the lysed cells. Subsequent studies revealed that solubility can be improved by growing cells and inducing RusA at 25°C (data not shown), allowing much better recovery of the induced protein by the same method. The protein purified was free of any visible contaminants as judged by Coomassie Blue and silver staining of SDS gels (Fig. 1A, lane c, and data not shown).
To investigate the native form of RusA, 11 g of the protein was analyzed by SDS-PAGE with and without prior boiling in SDS sample loading buffer. The boiled sample showed a single band of RusA (Fig. 1A, lane c). It migrates as a 15-kDa species, which is close to its predicted molecular mass of 13.8 kDa (1). Without boiling, a second band is clearly visible. Its migration suggests a molecular mass of about 30 kDa, which is consistent with a dimer of RusA. A further sample of the protein was treated with glutaraldehyde before boiling in sample loading buffer containing SDS. Fig. 1B shows a 30-kDa protein species in the treated sample (lane b). The only other band visible corresponds to the monomer species seen in the untreated control (lane c). The absence of any other band suggests that glutaraldehyde specifically cross-links a dimer of RusA.
The ability to detect a dimer on denaturing SDS gels without prior cross-linking (Fig. 1A, lane b) suggests that a substantial fraction of the purified protein is in this form. To determine the molecular mass, a sample of the native protein was dialyzed against buffer A containing 1 M KCl and applied to a Superose 12 HR column. The elution profile (Fig. 1C) showed a single sharp peak. SDS-PAGE analysis of the fractions confirmed that this peak coincided with the peak of RusA (data not shown). By comparison with protein standards analyzed using the same buffer, the molecular mass of the protein at the peak was determined to be 24 kDa. This is nearly twice the value of 13.8 kDa predicted from the amino acid sequence of the protein (36). Given the high salt conditions, this result strongly supports the idea that RusA is a dimer in solution. We also analyzed RusA in the presence of 0.15 M KCl. A substantial fraction of the protein eluted faster than the 670-kDa marker, suggesting a tendency to form aggregates.
Autoinhibition of Junction Cleavage-Previous studies (1) showed that RusA resolves X-junctions to nicked duplex products. Resolution can be monitored by electrophoresis of the reaction products on a nondenaturing polyacrylamide gel ( Fig.  2A). Using this simple assay, we investigated the resolution activity of RusA over a range of protein concentrations in reaction buffer containing 10 mM MgCl 2 or MnCl 2 . Fig. 2B shows the result of a typical experiment. Resolution is clearly sensitive to the concentration of RusA, peaking sharply at 50 nM in the Mn 2ϩ reactions and at 200 nM in the Mg 2ϩ reactions. Above these levels, the activity of RusA is very clearly inhibited. At subinhibitory concentrations, RusA is substantially more active in Mn 2ϩ than it is in Mg 2ϩ (Fig. 2B, inset). This difference disappears as RusA becomes inhibitory. Furthermore, even with 5 M RusA in the reaction, resolution is not totally abolished. Similar results were obtained using junction J3 (data not shown), although the fraction of junction resolved was generally higher (see below). Autoinhibition in this case reached a lower limit of ϳ30% junction resolution (data not shown).
Buffer Requirements and Time Course of Resolution-To determine the optimal conditions for resolution by RusA, we investigated the activity under a variety of reaction conditions, using J12 with RusA at 100 nM. RusA showed a requirement for divalent metal ions with a broad optimum of 10 -30 mM for Mg 2ϩ (Fig. 3A). Replacing Mg 2ϩ with Mn 2ϩ stimulated resolution in the 2-10 mM range, but at 20 mM and above Mn 2ϩ was substantially less effective than Mg 2ϩ (data not shown). No activity was detected with Zn 2ϩ , Ca 2ϩ , or Cu 2ϩ . RusA was insensitive to NaCl up to 200 mM, but was inhibited at higher concentrations and inactive at 500 mM (Fig. 3B). Activity in Mg 2ϩ buffer was stimulated at alkaline pH, and increased with increasing pH throughout the range tested (Fig. 4A). In contrast, the activity in Mn 2ϩ buffer was optimal at pH 7.5. Up to pH 8.5, it was also significantly higher than in Mg 2ϩ buffer. Fig. 4B shows the time course of the reaction in the presence of 10 mM Mg 2ϩ under two different buffer conditions, one at pH 7.5 in Tris buffer, and one at pH 9.5 in glycine buffer. As expected, significantly more of the junction was resolved at the higher pH. In both cases the reaction was over within the first 10 -15 min even though substantial fractions of the substrate remained unresolved. The failure to cleave all the junction molecules is surprising given that the protein (in terms of dimers) was present in at least a 50-fold molar excess. It means either RusA is inactivated rapidly during the reaction or that a fraction of the junction DNA, determined by the pH, is resistant to cleavage. It could also reflect a combination of these factors.
Sequence Specificity of Junction Cleavage-RusA resolves J12 by cleaving strands 1 and 3 (Fig. 5) 5Ј of the CC dinucleotide located symmetrically within the homologous core (1). The same site is targeted by RuvC (47). This coincidence could reflect some special feature of J12 because subsequent studies revealed that RuvC has a preference for cleaving strands 3Ј of a T within the context (A/T)TT2(G/C), where 2 identifies the cleavage point (27). To see if resolution by RusA is sequencespecific, we therefore investigated its ability to resolve junction J11. This junction (Fig. 5) is resolved very efficiently by RuvC, with Ͼ90% of strand cleavage at ATT2G, which matches the consensus target for RuvC (27). Fig. 6 shows that J11 is resolved by RusA to the same extent as J12. In contrast, RuvC cleaves J11 much more efficiently than J12.
The inability of RusA to respond to the RuvC target sequence in J11 suggested that it had a different requirement for efficient junction cleavage. We decided therefore to map the strand cleavages in J11. Four preparations of J11, each 32 P-labeled in a different strand, were incubated with RusA under standard cleavage conditions in the presence of either Mg 2ϩ or Mn 2ϩ . The products were resolved by electrophoresis on denaturing polyacrylamide gels and the cleavage sites in each strand mapped as described under "Materials and Methods." Fig. 5 summarizes the results of the mapping, while the amount of cleavage detected at each site is presented in Table I. The data revealed that RusA resolves J11 by symmetrical strand cleavage within the homologous core. In Mg 2ϩ , Ͼ90% of the cleavage was 5Ј of the CC in strands 2 and 4 (site 1). Very minor cleavage was also detected at a second site in strands 1 and 3 (site 2). Reactions done in Mn 2ϩ showed slight stimulation of cleavage at site 2, but site 1 was still strongly preferred. No cleavage was detected at the RuvC target site in strands 1 and 3.
The base 5Ј of the major RusA cleavage site differs between J11 (G) and J12 (T). Both junctions are resolved equally by this protein, suggesting the difference is of no consequence. However, the homologous cores of J11 and J12 differ substantially in sequence (Fig. 5), which could mask a base-specific effect at this position. We therefore analyzed cleavage of J26. The central region of this junction is identical in sequence to J12, but the flanking regions are different and extend the homologous core by 7 bp on either side (Fig. 5). Fortuitously, each of these regions contains a CC dinucleotide, one with a 5Ј T and the other with a 5Ј G. RusA resolves J26 with about the same efficiency as J12 (Fig. 4, Table I, and data not shown). In Mg 2ϩ buffer, substantial cleavage was detected at two sites ( and Table I, sites 1 and 2). Both are 5Ј of a CC dinucleotide. Site 1, which accounts for about two-thirds of the resolution activity, has a 5Ј T, and is identical to the major cleavage site in J12. Site 2, which accounts for most of the remainder, has a 5Ј G. Minor cleavages occurred at the remaining CC dinucleotide (site 3) and 5Ј of a C at the right end of the homologous core as drawn (site 4). The low cleavage at site 3 is surprising given that it has a T 5Ј of the CC and, like site 1, is located in strands 1 and 3. A similar observation has been seen with RuvC, which cleaves the ATTG target in strands 2 and 4 with high efficiency, but hardly touches the same sequence in strands 1 and 3 or the TTTG in strands 2 and 4, which also matches the consensus target sequence (48). The failure to cleave at these sites indicates that the general sequence context is also important for resolution. RusA activity was stimulated by Mn 2ϩ , with increased strand cleavage at sites 1, 3, and 4 at the expense of site 2. Additional cleavage was also seen at a new position (site 5, Fig. 5 and Table I).
The failure of RusA to target site 3 efficiently raises the possibility that resolution is not sequence-specific in terms of protein-DNA interactions, but is targeted to particular locations where the point of strand crossover might accumulate due to the local sequence context. In theory, the junction in the substrates used is free to branch migrate within the confines of the homologous core. However, previous studies with J26 and other similar junctions revealed that a major fraction of the molecules have the junction located at rather few positions, sometimes only one (24,47,48). This non-random distribution might have a significant bearing on the apparent sequencespecificity of resolution. We therefore tested RusA on J2, which has the junction constrained to a homologous core of two A:T base pairs, and also on J0, a static junction with G:C base pairs at or near the crossover point (Fig. 5). No nicked duplex products were detected (data not shown), which means that RusA cannot resolve either structure. However, a low level of cleavage was detected 5Ј of the CC dinucleotide in strand 3 of J0 (Table I). These data support the conclusion that junction resolution by RusA involves sequence-specific interactions between the protein and the DNA and that it occurs by a dual incision mechanism targeted predominantly 5Ј of CC dinucleotides.
Effect of Homologous Core Size-The results described in the previous sections revealed that factors other than the presence of a target sequence can have an overriding effect on the efficiency of resolution by RusA. The CC dinucleotide target identified using J11, J12, and J26 is located within an homologous core ranging from 11 to 26 bp (Fig. 5). To see if branch migration could be a limiting factor, we constructed two new junctions with homologous cores of either 4 or 3 bp, each containing the sequence TCC as a target for strand cleavage (Fig. 5, J4 and J3). Both are resolved very efficiently, with ϳ68% of J4 and ϳ73% of J3 converted to nicked duplex products within 10 min in standard Mg 2ϩ buffer at pH 7.5 (Fig. 7A, lanes b, c, e, and f, and data not shown). As expected, cleavage occurs specifically 5Ј of the CC dinucleotide, although there may be some minor cleavage between the two cytosines (Fig. 7B, lanes c, d, k, and l; Table I). The levels of resolution seen with these two junctions are substantially higher than with J11, J12, and J26, which reached a maximum of 40 -50% (Fig. 4B, Table I, and data not shown). Resolution of J3 and J4 could be stimulated by Mn 2ϩ (Table I), but the effects were modest compared with those seen with J12 or J26 (Fig. 4A, Table I). These data suggests that the length of the homologous core has a substantial effect on the fraction of substrate molecules that can be cleaved by RusA.
Stability of RusA-To gain further insight into the factors governing RusA activity, we monitored the time course of resolution of J3 at three different concentrations of protein, and compared the results with the time courses for resolution of J2 by RuvC under similar conditions. At 1 nM, RusA resolved ϳ4% of the J3 DNA within 2 min of starting the reaction (Fig. 8A). No further activity was detected over the remaining period of incubation. Increasing RusA 10-fold resulted in resolution of about 54% of the DNA, an increase of ϳ13-fold. Again, most of the resolution occurred within the first 2 min, but significant activity was detected for up to 5 min. At 100 nM RusA, 76% of the junction was resolved, and activity was detected over an even longer period (up to 15 min). However, the initial rate was slow compared with that seen with 10 nM RusA. We assume this reduced rate of cleavage is a further reflection of the autoinhibition seen with RusA at high concentrations (Fig. 2). The fact that resolution reached a clear plateau in all three cases, and that the activity was detected over a longer time scale the higher the initial concentration of protein, suggested that RusA is rapidly inactivated during the incubation. The results contrast sharply with those observed with RuvC. First, the initial rate of cleavage was roughly proportional to the concentration of protein over the 10 -100 nM range tested. Second, junction resolution continued over the 30-min period of incubation. Indeed, the time courses at 10 and 20 nM protein suggest that RuvC loses very little of its activity during this period.
To investigate the stability of RusA more directly, the protein (at 10 nM) was incubated in reaction buffer lacking either DNA or Mg 2ϩ , or both DNA and Mg 2ϩ , before adding the missing ingredients to assay for resolution activity. No loss of activity was detected over 60 min at 0°C under any of the conditions tested (Fig. 9, and data not shown). In the absence of DNA, ϳ80% of the resolution activity was lost within 1 min at 37°C (Fig. 9). The presence of Mg 2ϩ made little difference (data not shown). However, when RusA was incubated for 30 min at 37°C in the presence of junction DNA (without Mg 2ϩ ), it retained Ͼ90% of its resolution activity. This protection by substrate DNA was seen even when the concentration of RusA was increased to 100 nM, without increasing the amount of DNA. 91.8 Ϯ 3.7% of the initial resolution activity was present after 30 min preincubation. At 1 nM protein, 100% of the initial activity was retained over this time scale. Junction molecules were present at a concentration of only 1 nM in these studies. It is surprising that this amount could stabilize a 50-fold molar excess of RusA dimers. Although RusA binds preferentially to junctions, it has quite a high affinity for linear duplex DNA, which enables it to bind the arms of the junction to give a series of protein-DNA complexes that appear as a ladder of retarded species in bandshift assays (1) (see Fig. 11B). To see if this could help to stabilize RusA, we incubated the protein at 37°C with unlabeled linear duplex DNA before adding labeled J3 and Mg 2ϩ to assay resolution activity. Resolution activity decayed  rapidly under these conditions, but less so than in the absence of DNA (Fig. 9), which confirms that binding to linear duplex DNA affords some protection. The instability of RusA activity means that inactive protein accumulates rapidly during the course of a resolution reaction. This inactive protein could interfere with further activity. To investigate this possibility, a sample of RusA was inactivated by incubation for 10 min at 37°C in the absence of DNA. Its inactivation was confirmed by assaying resolution activity. A sample of the inactive protein was analyzed by SDS-PAGE. Only one band of protein was detectable on the gels, even with silver staining (data not shown). This band had the same mobility as the native species, which suggests that inactivation is not caused by proteolysis. A further sample was used to test for the ability to bind 32 P-labeled J12 DNA in a standard bandshift assay (see "Materials and Methods"). No bandshifts were detected with 12-100 nM protein in the reaction (data not shown), which established that the inactive protein can no longer bind DNA. Over the range of protein tested, native RusA forms a series of well defined protein-DNA complexes (1) (see Fig. 11B). The idea that inactive RusA binds unproductively to junction DNA can therefore be eliminated. However, we have not eliminated the possibility that inactive protein interferes with res-olution by interacting directly with the native protein.
Junction Turnover by RusA-The rapid inactivation of RusA when free of DNA would be expected to limit its ability to cycle from one junction to another during the course of a reaction. To gain some indication of junction turnover, resolution of 32 Plabeled J3 DNA was assayed in reactions containing a 40-fold excess of unlabeled J12 DNA. RusA was used at protein to junction (total) ratios in the range of 0.025-0.5. Fig. 10 shows that the amount of J3 resolved is reduced compared with control reactions done in the absence of unlabeled J12 DNA, when the protein to junction ratios ranged from 1 to 20. If RusA functions as a dimer, a single turnover would be expected to resolve from 1.25 to 25% of the J3 DNA, which is very close to the range observed (1.6 -35%, Fig. 10). However, the expected values are based on the assumption that all molecules of J3 are equally cleavable and that the preparation of RusA used is 100% active. Both of these assumptions are likely to be incorrect given the sequence specificity of strand cleavage and the instability of the protein. If the four possible locations for the strand crossover in J3 (Fig. 5) are equally represented, and only junctions at one of these positions can be cleaved, RusA would have to cycle through 4 molecules on average before finding one it could cleave. We conclude therefore that RusA probably acts catalytically. However, the number of junctions resolved on average by each enzyme molecule is likely to be rather low (Յ6 from the data in Fig. 10) unless of course the RusA preparation has a substantial fraction of inactive protein and/or the distribution of the strand crossover point is skewed unfavorably.
Inhibition of RusA Resolution Activity by RuvA-Previous studies revealed that the ability of RusA to fully suppress the recombination, and DNA repair-deficient phenotypes of ruvB and ruvC mutants requires the absence of RuvA protein (30). One plausible explanation for this effect is that RuvA sequesters Holliday intermediates and blocks their resolution. To investigate whether RuvA can block RusA directly, we monitored resolution activity in the presence of increasing amounts of RuvA. We used junction J11 for this purpose so that RuvC could be used as a control. Both proteins were used at a final concentration of 100 nM. Fig. 11A shows the results of a typical experiment of this type. In the absence of RuvA, 70% of the junction was resolved by RuvC (lane b) and 40% by RusA (lane j). Titrating RuvA into the reaction had a substantial inhibitory effect on both RuvC (lanes c-i) and RusA (lanes k-q), with Ͼ90% inhibition achieved at 100 nM and 400 nM RuvA, respectively (lanes e and o). A bandshift assay showed that RusA and RuvC bind J11 with approximately equal affinities. RusA (Fig. 11B,  lanes c-g) gives a ladder of retarded complexes, which is quite characteristic of the protein (1), whereas RuvC (Fig. 11B, lanes  j-n) gives a single well defined complex (19). These complexes were not detected in the presence of inhibitory concentrations of RuvA (data not shown). They were replaced by a well defined RuvA-junction complex (complex II) that sandwiches the junction point between two tetramers of RuvA (35,49). We assume this complex prevents RusA or RuvC from gaining access to the junction. From the results shown in Fig. 11A, its formation in vivo would certainly explain the inhibition of RusA's ability to function during recombination and DNA repair (30). DISCUSSION In this paper we have described an improved scheme for the purification of RusA and investigated the subunit structure of the protein by gel filtration, glutaraldehyde cross-linking, and SDS-PAGE. These studies revealed that the native protein is a dimer. It is stable in 1 M NaCl but is dissociated to monomers by boiling. However, without boiling, a substantial fraction of the dimers remain intact during SDS-PAGE, which suggests that the constituent monomers are tightly bound. A dimeric structure for RusA is consistent with its ability to resolve Holliday intermediates, an activity that requires dual strand incision at the junction point (24,29). Several of the Holliday junction resolvases studied to date have been shown to bind junctions as dimers (29,50,51). In the case of RuvC, the target strands are folded into the catalytic core of each subunit and cleaved in a coordinated and sequence-dependent manner (22,23,24,27).
Using small synthetic X-junctions, we found that RusA also cleaves junctions at specific sequences. Our previous studies with J12 had revealed that RusA introduced symmetricallyplaced nicks in strands of the same polarity (1). Remarkably, the vast majority of the incisions were at the same site cleaved by RuvC (GT2CC). However, this proved to be a coincidence reflecting the fact that RuvC cleaves 3Ј of a T and RusA 5Ј of a CC dinucleotide. When we analyzed resolution of junctions (J11 and J26) containing the RuvC target sequence, ATT2G, we found that RusA did not cleave this site at all. The junctions were resolved efficiently, but in both cases Ͼ90% of the activity was targeted 5Ј of a CC dinucleotide located symmetrically within the homologous core, either at G2CC (J11) or both G2CC and T2CC (J26). There was very little evidence of cleavage at non-symmetrical sites, which supports the notion that RusA acts as a dimer. From the junctions analyzed, a CC dinucleotide is essential for efficient resolution by RusA; a single C is not enough. It may be significant therefore that RuvC prefers to cleave 3Ј to a TT. Previous studies of T4 endonuclease VII and of T7 endonuclease I revealed that each of these enzymes resolves J26 predominantly by symmetrical strand cleavage at one of three sites within the homologous core (45). None of these sites is common to both enzymes. Cleavage at all six sites occurs next to a pyrimidine, and in five out of the six occurs between two pyrimidines. The CCE1 resolvase of Saccharomyces cerevisiae also targets cleavage between two pyrimidines (51). This preference for cleaving the DNA between or next to a pyrimidine dinucleotide by five different resolvases suggests that some general feature of these dinucleotides, or their purine complements, affects folding of the junction and its interaction with the protein. RuvC has been shown to bind junctions in an open configuration and even then will resolve the structure only when the target sequences are in the non-crossover strands (22,23). The fact that four of the enzymes tested on J26 target different pyrimidine dinucleotide sites indicates that each enzyme also has to make specific contacts with bases flanking the cleavage site. It reinforces the fact that both topological factors and sequence-dependent effects determine the efficiency of resolution (28,29).
By constraining a TCC target to homologous cores of 4 and 3 base pairs, we found that the efficiency of junction resolution by RusA depends not only on DNA sequence but also on the number of base pairs in the homologous core. For instance, RusA resolves J11 and J12 with approximately equal efficiency, whereas under the same reaction conditions it resolves nearly twice as much of J3. Furthermore, J26 is cleaved as efficiently as J11 or J12 despite having more than twice the length of homologous core. However, it also has two sequences that are targeted efficiently by RusA, as opposed to one in both J11 and J12. These observations raise the possibility that branch migration may be a limiting factor for RusA, although we cannot rule out other junction-specific effects.
Under the conditions used, RusA rarely resolved more than 85% of the substrate DNA even when the junction was constrained. This limitation can be attributed in part to the instability of the protein. Most (ϳ80%) of its resolution activity is lost within 1 min at 37°C when incubated in the absence of junction DNA. This means the protein is likely to have difficulty cycling from one molecule to another without losing activity. The X-junctions used in this work that are cleaved by RusA have homologous cores of 3-26 base pairs within which the junction itself can branch migrate. If the junction point is distributed and RusA, like RuvC (24), resolves molecules only when the crossover is at a target sequence, then branch migration would be required to achieve 100% resolution. Previous studies have shown that junctions fold into an antiparallel stacked-X in the presence of divalent metal ions such as Mg 2ϩ , and have to unstack before branch migration can occur (15,(52)(53)(54)(55). RuvA protein, which acts with RuvB to catalyze branch migration, solves this problem in vivo by holding the DNA in an open configuration that enables branch migration to be driven over many base pairs in a single step (31,34). With naked junctions, the stacking and unstacking of the DNA between each base pair of branch migration is rate-limiting (56).
Branch migration of free DNA is unlikely to be a significant factor in determining the efficiency of resolution by RusA over the time scale of our experiments. The spontaneous rate of junction branch migration in Mg 2ϩ (56) is considerably faster than the rate of decay of RusA. Therefore, even if each RusA dimer was limited to cleaving one junction, it should be possible to resolve all the junction molecules by having a large excess of protein over DNA. What we actually found was that beyond a certain level, increasing the protein to DNA ratio reduced resolution activity, which is contrary to expectation. In the two cases examined, there was a lower limit to the amount of junction resolved at high protein concentrations. The limit was 10 -15% for J12 and 30% for J3, which suggests that it is related to the length of the homologous core. We suspect therefore that the autoinhibition observed is due to a mass effect of protein binding, which would limit the junction folding and unfolding needed to allow branch migration. This "freezing" of the junction would limit resolution to those molecules located at a cleavable sequence. Given the instability of the free protein, binding of at least some of the junctions at non-cleavable positions readily explains the ceiling on junction resolution at more modest ratios of protein to DNA.
We found that resolution could be increased substantially by replacing Mg 2ϩ in the buffer used with Mn 2ϩ , or by increasing the pH. This could be due to some specific effect on the enzyme. However, very similar findings were reported previously for RuvC (41,48), raising the possibility that it might be related to a more general effect on protein-DNA interaction. Furthermore, the stimulation of RusA by Mn 2ϩ was much greater with J12 than with J4 or J3, again suggesting some possible relationship with branch migration. Mn 2ϩ and alkaline pH may destabilize the interaction of RusA with junction DNA, thereby allowing more freedom for branch migration before the protein decays. The fact that very little of the additional resolution in Mn 2ϩ involved cleavage at new sites argues against any general reduction in cleavage specificity.
The similarity between RusA and RuvC revealed by these studies is striking. However, there are significant differences, apart from the sequence specificity. RusA seems much more active in resolution than RuvC. This was particularly noticea- ble when RusA was used at a concentration that did not cause autoinhibition (Fig. 8). It is difficult to quantify because the sequence-specificity requires the use of different junctions for the two proteins. However, it is likely to be substantial given that RusA is also prone to decay, which limits its turnover of junctions. RuvC appears to be particularly stable under standard reaction conditions and can withstand prolonged incubation at room temperature in the absence of substrate DNA (41). It is clear from genetic studies that RuvC cannot function efficiently in vivo without RuvAB (1,30,36). It most probably functions as part of a RuvABC resolvasome complex assembled on junction DNA (35). In vitro resolution reactions containing RuvC alone may therefore not reflect its true activity. RusA on the other hand appears to function alone, although its ability to promote recombination and DNA repair does require the presence of RecG (30). It can certainly do without RuvAB; that is how it was discovered (30). Indeed, its activity is inhibited by RuvA both in vivo (30) and in vitro (this work). The reason why RecG activity is needed is less clear. There is no evidence for a direct interaction between the two proteins. One possibility is that RecG helps RusA by moving junctions located at noncleavable sites.
The ability of RusA to act alone is consistent with its bacteriophage origin (36). This may also explain both its high activity and instability. Presumably, a short burst of potent resolvase activity would be needed late in the infection cycle to resolve junctions prior to packaging DNA. There would be no need to evolve a protein that retained its activity over a long time scale. In this respect, RusA may resemble other bacteriophage enzymes such as bacteriophage T4 endonuclease VII. However, its sequence specificity and failure to resolve a static junction is more characteristic of a specialized Holliday junction resolvase than of an enzyme required to remove branch structures from DNA.