On the Role of ATP Hydrolysis in RecA Protein-mediated DNA Strand Exchange

RecA protein promotes a substantial DNA strand exchange reaction in the presence of adenosine 5‘-03-(thi0)triphosphate (ATPyS) (Menetski, J. P., Bear, D. G., and Kowalczykowski, S. C. (1990) Proc. Nutl. Acad. Sci. U. S. A. 87, 21-25), calling into question the role of ATP hydrolysis in the strand exchange reaction. Here, we demonstrate that the ATPyS-mediated reaction can go to completion when the duplex DNA substrate is only 1.3 kilobase pairs in length. The ATPyS-mediated reaction, however, is completely blocked by a 52-base pair heterologous insertion in either DNA substrate. This same barrier is readily bypassed when ATP replaces ATPyS. This indicates that at least one function of recA-mediated ATP hydrolysis is to bypass structural barriers in one or both DNA substrates during strand exchange. This suggests that ATP hydrolysis is directly coupled to the branch migration phase of strand exchange, not to promote strand exchange between homologous DNA substrates during recombination, but instead to facilitate the bypass of structural barriers likely to be encountered during recombinational DNA repair.

The RecA protein is a central component in recombinational DNA repair pathways and homologous genetic recombination in Escherichia coli. In vitro, RecA protein promotes the pairing and exchange of complementary DNA strands in reactions that provide an important experimental paradigm for RecAfunction in vivo. Avariety of DNAsubstrates are used in these reactions, including circular single-stranded DNA (ssDNA)' and linear duplex DNA, and gapped circular duplex DNA and linear duplex DNA. In each case, one of the DNA substrates is at least partially single-stranded. The ssDNA is the site where RecA protein filament formation nucleates and where initiation of DNA strand exchange takes place. ssDNA is highly recombinogenic in vivo and may be the primary signal for SOS induction of the DNA repair genes (for review, see Cox, 1993;Kowalczykowski, 1991;Radding, 1991;Roca and Cox, 1990;West, 1992).
The RecA-mediated DNA strand exchange reaction can be divided into three distinct phases (see reviews cited above). During the first phase, called presynapsis, a stoichiometric * This work was supported by National Institutes of Health Grants GM-32335 (to M. M. C.) and GM-14711 (to R. B. I.). This paper is the third in a series on the function of RecA-mediated ATP hydrolysis. The first two papers in the series are Kim et al., 1992a and 1992b. The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked "aduertisement" in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.
$ To whom correspondence should be addressed. Tel.: 608-262-1181; The abbreviations used are: ssDNA, single-stranded DNA; ATPyS, adenosine 5'-0-(34hiotriphosphate); bp, base pair(s); kbp, kilobase paids); SSB, E. coli single-stranded DNA-binding protein; AMT, 4'aminomethyl-4,5',8-trimethylpsoralen; dsDNA, double-stranded DNA. F a : 608-265-2603. amount of RecA protein (one monomerhhree nucleotides) assembles in the 5' to 3' direction along the phosphate backbone of the ssDNA to produce a right-handed, helical, nucleoprotein filament. In the presence of ATP or ATPyS (an ATP analog that is not readily hydrolyzed by RecA), DNA within RecA filaments is extended and, if it is duplex, underwound. There are 6 RecA monomers and 18 nucleotides or bp of DNAhelical turn of the filament, and the axial risehp increases form 3.4 to 5.1 A. The nucleoprotein filament is the active species of RecA protein in the subsequent phases of the strand exchange reaction. In the second phase, called synapsis, homologous sequences in the ssDNA within the RecA filament and the duplex DNA are aligned. There is evidence for the formation of a novel triplestranded DNA pairing intermediate in this phase (Camerini-Otero and Hsieh, 1993;Rao et al., 1993;Stasiak, 1992). A strand switch or crossover occurs within the filament to form short regions of DNA in which complementary strands from the two different DNA substrates are paired.
This product DNA has often been referred to as a heteroduplex, whether or not it contains any bp mismatches or structural irregularities, to distinguish it from substrate DNA that is often identical in sequence. In this article, we use the term hybrid DNA to describe product duplexes made up of strands that are completely complementary, and we reserve the term heteroduplex to describe duplexes containing a mismatch or some other structural irregularity.
In the third and final phase of the strand exchange reaction, the region of hybrid or heteroduplex DNA is extended in a kind of facilitated branch migration. DNA pairing intermediates that have not completed DNA strand exchange are often referred to as joint molecules. RecA-mediated DNA strand exchange in the presence of ATP is unidirectional and slow, proceeding in the 5' to 3' direction with respect to the ssDNA at a rate of 3-10 bp s-'. The DNA strand exchange reaction promoted by RecA readily bypasses structural barriers in the DNA, including DNA lesions and even short (1-100 bp) heterologous inserts in one of the DNA substrates (Bianchi and Radding, 1983;Das Gupta and Radding, 1982;Jwang and Radding, 1992;Kim et al., 1992a;Livneh and Lehman, 1982;Rosselli and Stasiak, 1991). This latter property is probably critical to the function of RecA in recombinational DNA repair.
RecA protein is a DNA-dependent ATPase, and ATP is hydrolyzed during the strand exchange reaction with a monomer Kc,, of 20-24 min" (Schutte and Cox, 1987). ATP is hydrolyzed uniformly by RecA monomers throughout each nucleoprotein filament (Brenner et al., 1987;Kowalczykowski and Krupp, 1987;Schutte and Cox, 1987).
The molecular function of RecA-catalyzed ATP hydrolysis is not understood and is currently a subject of controversy. At least three observations indicate that ATP hydrolysis has only a minimal, if any, function in the DNA strand exchange reaction. First, the reaction is inherently inefficient from a thermo-20653 dynamic viewpoint. DNA strand exchange between completely homologous DNA substrates is an isoenergetic reaction since there is no net increase or decrease in DNA bp. However, in a typical strand exchange reaction using DNA substrates derived from bacteriophage M13 or 4x174, 100 or more ATPs are hydrolyzed per bp of hybrid DNA formed (Roca and . Second, substantial DNA strand exchange can occur in the presence ofATPyS (Kim et al., 1992a;Konforti and Davis, 1992;Menetski et al., 1990;Rosselli and Stasiak, 1990). Under narrowly defined conditions, about 2 kbp or more of hybrid DNA is formed within 2 min of reaction (Menetski et al., 1990). This result has recently been reinforced by the observation that a RecA mutant (RecA K72R) that binds but does not hydrolyze ATP can also promote a limited amount of DNA strand exchange in vitro (Rehrauer and Kowalczykowski, 1993). Third, eukaryotic proteins have been found which promote DNA pairing and strand exchange but do not hydrolyze ATP (Eggleston and Kowalczykowski, 1991).
However, RecA-mediated ATP hydrolysis is not readily dismissed as a molecular anomaly. The single ATP binding site in each monomer represents the most highly conserved part of the RecA amino acid sequence (Roca and . When RecA protein is bound to DNA, the Kcat for ATP hydrolysis by each monomer within a filament generally falls into a relatively narrow range of 20-30 min". In addition, the RecA K72R mutant exhibits a null phenotype for all RecA activities, providing evidence that ATP hydrolysis is critical to RecA function in vivo.' One generally agreed upon function of ATP hydrolysis is the recycling of RecA monomers in nucleoprotein filaments after a DNA strand exchange reaction. Filament assembly and disassembly are end-dependent processes, with disssociation of monomers occurring predominantly a t one end (the "-" end, located 5' with respect to the ssDNA that serves as a nucleation site for filament assembly) (Lindsley and Cox, 1990) and assembly occurring at the other (the "+" end; 3') (Register and Griffith, 1985). Assembly and disassembly exhibit different equilibria (Lindsley and Cox, 19901, implying a n involvement of ATP hydrolysis in one or the other process. ATPyS, which is not hydrolyzed by RecA, blocks disassembly (Lindsley and Cox, 1990), whereas ATP hydrolysis reduces the affinity of RecA protein for DNA (Menetski et al., 1988). Together, these results indicate that ATP hydrolysis by a RecA monomer at the (-) end of a nucleoprotein filament is likely to result in dissociation. Recycling of RecA monomers provides one function for ATP hydrolysis. However, ATP hydrolysis is not limited to or even enhanced in monomers at filament ends, and filament disassembly rarely accounts for more than a minute fraction of ATP hydrolytic events. Much evidence now indicates that filament disassembly does not play a mechanistic role in DNA strand exchange (Cox, 1994;Lindsley and Cox, 1990).
The DNA strand exchange reaction observed in the presence of ATPyS has provided an important new tool to explore the function of ATP hydrolysis by RecA. An examination of the limitations of the ATPyS-mediated reaction has begun to define a role for ATP hydrolysis in DNA strand exchange. These studies have shown that the DNAstrand exchange reaction does not bypass structural barriers in the DNA (Kim et al., 1992a;Rosselli and Stasiak, 1991) and does not accommodate four DNA strands (Kim et al., 199213) when ATPyS replaces ATP. The requirement for ATP hydrolysis in the bypass of DNA structural barriers provides an experimental link between the ATP hydrolytic activity and RecA function in DNA repair.
Another potential function of ATP hydrolysis is to render DNA strand exchange unidirectional. Two recent studies have R. Devoret, personal communication.
looked for a relationship between the direction taken by the strand exchange reaction and ATP hydrolysis and have arrived at very different conclusions. Rosselli and Stasiak (1990) found that DNA strand exchange in the presence of ATPyS was unidirectional, exhibiting the same polarity as the ATP reaction. In addition, they found no difference in the efficiency with which paired DNA intermediates were formed with ATP or ATPyS. Konforti and Davis (1992) arrived at the opposite conclusion, finding no evidence for a prevalent direction in the strand exchange reaction carried out with ATPyS. The present study was carried out to resolve this discrepancy and further explore the function of ATP hydrolysis by RecA protein. We provide evidence that ATP hydrolysis is required to render the DNA strand exchange reaction unidirectional. We further demonstrate that ATP hydrolysis has a substantial effect on the lengths of hybrid DNA formed during DNA strand exchange.

MATERIALS AND METHODS
Enzymes a n d Reagents-E. coli RecA protein was purified to homogeneity and stored as described previously . E. coli ssDNA-binding protein (SSB) was purified as described by Lohman et al. (19861, except that an additional step utilizing DEAE-Sepharose chromatography was included to ensure removal of single-strand exonucleases. The RecA protein and SSB concentrations were determined by absorbance at 280 nm, using extinction coeficients of Ẽ , , , = 0.59 A,,, mg" ml (Craig and Roberts, 1981) and = 1.5 A,,, mg" ml (Lohman and Overman, 19851, respectively. Restriction endonucleases were purchased from New England Biolabs. Tris buffer and ATPyS were purchased from Boehringer Mannheim. DEAE-Sepharose was purchased from Pharmacia Biotech. Inc. Creatine phosphokinase, phosphocreatine, proteinase K, low melting agarose, and ATP were purchased from Sigma. 4'-Aminomethyl-4,5',8-trimethylpsoralen (AMT) was purchased from Calbiochem. The purity of the ATP and ATPyS was assayed by thin layer chromatography, and each was shown to be at least 99 and 90% pure, respectively, with the major contaminant beingADP (Shibata et al., 1981).
DNA-Circular ssDNA and supercoiled circular duplex DNA from bacteriophage M13mp8 (7,229 bp) and M13mp8.1037 (8,266 bp) were prepared using methods described previously (Davis et al., 1980;Messing, 1983;Neuendorfand Cox, 1986). The bacteriophage M13mp8.1037 is the bacteriophage M13mp8 with a 1,037-bp sequence (EcoRV fragment) from the E. coli galT gene inserted into the SmaI site. The concentrations of ssDNA and dsDNA stock solutions were determined by absorbance at 260 nm, using 36 and 50 pg ml" A,,,", respectively, as conversion factors. DNA concentrations are reported in terms of total nucleotides. Full-length linear duplex (FIII) DNA was derived from M13mp8.1037 FI DNA by complete digestion with EcoRI or PstI endonuclease, using conditions suggested by the enzyme supplier. Linear duplex fragments of M13mp8.1037 were prepared by digestion with two restriction endonucleases, EcoRI and BglII or MscI and BamHI. The desired fragment was purified from low melting agarose gels by a published procedure (Sambrook et al., 1989).
Strand Exchange Reaction Conditions-Reactions involving ATP were performed a t 37 "C in a standard reaction buffer containing 25 mM Tris acetate (80% cation, pH 7.5), 10 mM magnesium acetate, 3 mM potassium glutamate, 1 mM dithiothreitol, 5% glycerol, and an ATP regeneration system (10 units ml-' creatine phosphokinase, 12 mM phosphocreatine). Reaction volumes were 130 pl, and the concentrations of DNA and proteins reported below are the final concentrations after the addition of all components. ssDNA (IO p~) was preincubated with RecA protein (6.6 J~M) for 10 min before the addition of duplex DNA (11.4 PM for the 8.3-kbp duplex substrate; 2.4 PM for the 1.7-kbp substrate; 3.0 p~ for the 2.2-kbp substrate). After incubation of this mixture for 10 min, ATP (3 mM) and SSB (2 PM) were added to start the reactions. The ATP regeneration system maintains the ADP concentration near zero for at least 80 min.
Reactions involving ATP$ were performed at 37 "C in a standard reaction buffer containing 25 mM Tris acetate (80% cation, pH 7.5),4 mM magnesium acetate, and 1 mM dithiothreitol. ssDNA (IO PM) was preincubated with SSB (0.9 p~) for 10 min before the addition of RecA protein (6.0 p~) and ATPyS ( 1 mM). After incubation of this mixture for 10 min, the reaction was initiated with duplex DNA (11.4, 2.4, and 3.0 p~ for the 8.3-, 1.7-, and 2.2-kbp substrates, respectively). Conditions for ATP and ATPyS reactions are different so that each proceeds optimally. The ssDNA molecules are in 2-fold molar excess with respect to the dsDNA molecules in all reactions.
Agarose Gel Assays-Aliquots (15 p1) of the strand exchange reactions described above were removed at the indicated times, and the reactions were stopped by the addition of 5 pl of gel loading buffer (25% glycerol, 15 m M EDTA, 0.025% bromphenol blue, 5% sodium dodecyl sulfate). These aliquots were stored on ice until the last time point was taken. Samples were electrophoresed overnight in an 0.8% agarose gel at 2-2.5 V cm".
Electron Microscopy-Visualization of reactions by electron microscopy was carried out by spreading the entire strand exchange reaction mixture after deproteinization and dialysis. Reactions were stopped after 60 min and incubated with proteinase K (1 mg ml-I final) and sodium dodecyl sulfate (0.5% final) at 37 "C for 30 min to remove the RecA protein and SSB. All samples were dialyzed into 20 m M NaCl and 5 m M EDTA for 9 h at 25 "C before spreading as described previously (Inman and Schnos, 1970).
To block spontaneous branch migration, some strand exchange reaction mixtures were cross-linked with AMT prior to deproteinization. Aliquots (20 p1) of the strand exchange reaction at 60 min (or shorter periods of time for the experiment in Fig. 7) were immediately mixed with AMT (30 pg ml-l, final concentration), incubated at 25 "C for 3 min, and irradiated with long wave UV light for 4 min at 25 "C (Umlauf et al., 1990). The W light was generated with two 15-watt fluorescent black light tubes. Samples were placed 8 cm below the light source. Intensity of UV light ranged between 2 and 4 milliwattskm' at 365 nm (as measured with a Blak-Ray 5-221 long wave UV monitor). A glass microscope slide was placed over the samples to filter out UV radiation below 320 nm. RecA protein and SSB were removed and the samples dialyzed as described above.
Deatment of Electron Microscopic Data-There are three types of information gained from the electron microscopic experiments.
1. We wanted to confirm that deproteinized reaction mixtures contained recombinational intermediates of the expected type. For the substrates shown in Fig. 1, we would expect intermediates to consist of a circular region (containing ssDNA and dsDNA segments) with separate dsDNA and ssDNA tails. The dsDNA segment within the circular region represents the DNA that has undergone exchange, whereas the dsDNA tail is that part of the dsDNA substrate which has not or cannot be exchanged. For examples, see Fig. 5.
2. We wished to estimate the proportion of the duplex linear substrates which was involved in DNA strand exchange reactions leading to intermediates. This was determined by counting the intermediates and unreacted linear duplex substrate molecules found in a representative sample derived from a reaction. Complex recombinational events involving more than two DNA substrate molecules and events that were interpreted to arise from broken or nicked substrates (the latter produce low levels of gapped circular and more complex products in the reactions) were ignored in these estimates. In some experiments, the grids from different reactions were assigned an undescriptive identification code by S. K. J. and provided in a random sequence to R. B. I. without further identification. The molecules on them were subsequently counted "blind" by R. B. I.
3. Finally, it was important to estimate the extent of DNA strand exchange (the length of hybrid DNA generated) for a representative sample of the intermediates. Because of the large number of samples, obtaining an accurate measurement of significant numbers of intermediates in all of the samples was impractical. Therefore, the length of the exchanged region was estimated by eye in many experiments. As an example, in Fig. 5B, the ratio of the exchanged region (length of the double-stranded segment within the circle) to the unexchanged region (length of the double-stranded tail) was judged to be about 1:4.5. For a duplex linear DNA substrate of 8.3 kbp, this would mean that DNA strand exchange had progressed about 1.5 kbp. For convenience, each linear duplex was divided into eight segments; judgments about the length of the exchanged region are given in 1-kbp segments for the 8.3-kbp molecules shown in Figs. 4 and 7, and in 0.21-or 0.27-kbp segments for the 1.7-and 2.2-kbp molecules shown in Fig. 6. Because of the way in which these data were accumulated, the first segment (starting at 0 kbp) is 50% wider than all of the others for all of the duplex substrates shown in Figs. 4, 6, and 7; the counts in the first segment have therefore been reduced proportionally so that the height of the bar reflects a constant percent of intermediates per incremental length (e.g. in Fig. 4, the height of the bars reflect the percent of intermediatedkbp).
The areas underneath the bars are an accurate reflection of the number of intermediates observed in each segment.
These judgments were checked in two ways. First, two grids each from two samples (corresponding to those shown in Fig. 4, E and D ) were counted blind as described above. The four sets of data were then compared by the , $ two-way contingency test at the 95% confidence level. The two data sets corresponding to sample B were not significantly different ( p = 0.531, and the same was true for the data sets corresponding to sample D ( p = 0.38). On the other hand, when the two data sets corresponding to B were compared with those from D, there was a difference ( p = 0.0002). The numbers of molecules counted in these data sets were 32 and 20 for the samples of Fig. 4 B , and 29 and 30 for the samples of Fig. 40. Second, measurements were made in two samples corresponding to Fig. 4, C and D, and these more accurate data were converted to the less accurate judgment scale shown in Fig. 4 and 6 . 2 two-way contingency tests of data in Fig. 4 showed that there was no significant difference ( p = 0.27 (C) and 0.22 (D)). Measurements were made on 34 and 32 molecules, respectively. Photography and computer-assisted length measurements of DNA molecules were performed as described previously (Littlewood and Inman, 1982).

RESULTS
Experimental Design-In the following discussion, the ends of the duplex DNA substrate are identified as 5' or 3' with respect to the strand that is identical to the circular ssDNA substrate. To provide information on the directionality of branch migration, duplex DNA substrates in which homology was restricted to the 5' or 3' end were used in strand exchange reactions with circular ssDNA (Fig. 1) Fig. 1). Reaction conditions are described under "Materials and Methods" and the legend to Table I. Time points in each set of reactions were 0,5,15,30, and 60 min, respectively, as indicated by the numbers above the gel lanes in reactions a d .
at one end reflects both the polarity of branch migration and the intrinsic stability of the joint molecules after RecA protein is removed. The strand exchange reactions were analyzed by agarose gel electrophoresis and by electron microscopy. Utilization of AMT cross-linking in some experiments permitted us to correct for loss of joint molecules due to spontaneous branch migration after removal of RecA.

I n the Presence of ATP, DNA Pairing Intermediates Are Extended and Preferentially Stabilized at the 5' End of the Duplex
DNA-Although pairing occurs at either end of the duplex DNA, the number ofjoint molecules observed when homology is restricted to the 5' end of the dsDNA is much greater than that observed with homology restricted to the 3' end. This is shown in Fig. 2, reactions a and c (3' homology) as compared with reactions b and d (5' homology). The bias for intermediate formation at the 5' end was similar for reactions involving long or short regions of homology in the duplex and is consistent with the established 5' to 3' direction of strand exchange in the presence ofATP hydrolysis. The intermediates formed on the 5' end migrate in a discrete band, as opposed to a diffuse set of bands observed for intermediates at the 3' end. As shown below, the discrete band in the joint molecules formed with homology restricted to the 5' end is a direct result of unidirectional DNA strand exchange, which extends the hybrid DNA to the end of the homologous region.
The results obtained by agarose gel electrophoresis were confirmed by direct observation of the pairing intermediates by electron microscopy (Table I). In the presence of ATP, intermediates were observed preferentially at the 5' end with both sets of DNA substrates.
There Is an Apparent Substrate-dependent End Preference for Joint Molecule Formation in the Absence of ATP Hydrolysis-Strand exchange reactions involving ATPyS are shown in Fig.  3. In reactions involving long (7 kbp) regions of homology, pairing intermediates appeared to form more efficiently at the 5' end as compared with the 3' end (Fig. 3, reactions b versus a ) . However, the observed bias for intermediate formation was reversed in strand exchange reactions involving duplex DNA molecules with short (about 1 kbp) regions of homology (Fig. 3,

reactions c versus d ).
The results were again reinforced by electron microscopy. In the presence of ATPyS, joint molecules appeared to form more efficiently at the 5' end of the duplex than at the 3' end when long (7 kbp) regions of homology were present; the apparent end bias was reversed with short homologous regions in the duplex DNA (Table I).

AMT Cross-linking of the Paired Intermediates Reveals That the Apparent End Bias Observed in the ATPyS Reactions Is
Artifactual-We carried out additional reactions in the presence of ATPyS, with the same substrates and under conditions identical to those described above. Just prior to termination of the reaction and removal of RecA protein and SSB, the DNA in these reactions was cross-linked with a photoreactive reagent (AMT) that cross-links pyrimidines in opposite DNA strands ( Table I). The cross-linking conditions used result in a crosslinking density of approximately one AMT cross-linW220 bp of DNA (Bedale et al., 1991). This treatment prevented branch migration during the deproteinization and dialysis steps that precede spreading the DNA for electron microscopy.
AMT cross-linking increased the fraction of the DNA substrates found in joint molecules for both long and short regions of homology whether it was positioned at the 5' or 3' end of the linear duplex. More significantly, the cross-linking abolished the end bias observed in the ATPyS reactions (Table I). Therefore, the apparent bias in favor of one end or the other in joint molecule formation with ATPyS is an artifact brought about by spontaneous branch migration following deproteinization. Since the bias is distinctly different with two different sets of duplex DNA substrates, the branch migration appears to be TAn1. E I Quantitation of DNA strand exchange intermediates by electron microscopy p~ RecA protein and 2.0 p~ SSB. ATPyS reactions had 6.0 PM RecA protein and 0.9 p~ SSB. All reactions contained 10 p~ ssDNA, and 11.4, 2.4, Reactions were performed and prepared for electron microscopy as described under "Materials and Methods." ATP-containing reactions had 6.6 and 3.0 p~ dsDNA for the 8.3-, 1.7-, and 2.2-kbp duplex DNA substrates, respectively. Other reaction conditions are described under "Materials and Methods." Molecules were counted, and the numbers of joint molecules are expressed as the percentage of the total duplex DNA substrates associated with a ssDNA molecule. The error limits shown reflect statistical counting errors and are not meant to indicate estimated error arising from repeated experiments. N is the number of DNA molecules containing duplex DNA counted in each experiment and represents the sum of the duplex DNA substrate molecules and the intermediates in which the duplex is paired with a ssDNA. Uncross-linked samples are those that were deproteinized, dialyzed, and spread as described under "Materials and Methods." The cross-linked samples were treated with AMT as described prior to deproteinization. DNA substrates are labeled as in Fig. 1. ND denotes a single reaction for which we were unable to generate meaningful data. AMT treatment of this particular sample resulted in unexpected multistranded DNA species in more than five separate experiments. This phenomenon is under investigation.

Duplex DNA substrate
Uncross-linked Cross-linked Joint molecules (%) N Joint molecules ( 0 ) N Nucleotide Long 3' homology (7.2 kbp) Long 5' homology (7.2 kbp) Short 3' homology (0.7 kbp) Short 5' homology (1.2 kbp) Long 3' homology (7.2 kbp) Long 5' homology (7.2 kbp) Short 3' homology (0.7 kbp) Short 5' homology (1.  Table I. selective. We do not understand the bias differences for the two substrate sets. The results obtained with ATP were essentially unaffected by AMT cross-linking. There was still a strong bias for joint molecule formation at the 5' end with either a long or short homologous region in the duplex substrate (Table I).

ATP Hydrolysis Facilitates Branch Migration and Permits Extensive DNA Strand Exchange-Even
when ATP is hydrolyzed, DNApairing can occur anywhere within the duplex DNA substrate, forming an intermediate that can contain thousands of bp of paired or hybrid DNA. With the subsequent strand exchange proceeding 5' to 3', ATP hydrolysis will extend the hybrid region when the joint is formed on the 5' end of the duplex and eliminate it when it is formed on the 3' end (Cox and Lehman, 1981;Dutreix et al., 1991). This would not only result in a selective accumulation of joint molecules on the 5' end, it should also be evident in the length of the hybrid DNA in the joints that are formed. Joints formed on the 5' end should be extended to the boundary of the heterology; 3' joints should be fewer in number, vary in length, and be much shorter (on average) than 5' joints.
The effect ofATP hydrolysis on the extent of strand exchange was quantified by reacting dsDNA, with a long (7 kbp) region of homology limited to one end of the duplex, with circular ssDNA. The extent of DNA strand exchange was determined by visualizing pairing intermediates by electron microscopy and esti- mating the length of the hybrid DNA as described under "Materials and Methods" (Fig. 4 ). When ATP was hydrolyzed, DNA strand exchange proceeded to the homologyheterology boundary ( 7 khp) in the majority ofjoint molecules formed on the 5' end (Fig. 4A ). This is in contrast to the random distrihution of the length of hyhrid regions seen with long homology at the 3' end of the duplex (Fig. 4R ). The data in Fig. 4 were obtained for joint molecules that were not cross-linked. When AMT was used to cross-link the joint molecules prior to deproteinization, virtually all of the joints formed at the 5' end had hybrid DNA extending to the homologyheterology houndaw. The small numher of molecules with shorter hyhrid regions seen in Fig. 4A was eliminated (data not shown).
In ATPyS-mediated reactions. joint molecules form effciently at either end of the duplex, but only 2 khp or less of hybrid DNA is formed at either end of the duplex DNA before branch migration is blocked in the majority of pairing intermediates (Fig. 4, C and D). The results were the same when the reactions were extended to 90 min or when the paired intermediates were cross-linked (data not shown ).
Two representative pairing intermediates formed in ATPmediated reactions with long (7.2 khp) homology restricted to the 5' end are shown in Fig. 5.4. The extensive region of duplex DNA encompassing the circles and the long single-stranded tails reflect a DNA strand exchange reaction that has proceeded to the homologyheterology boundary. Two typical joint molecules formed in the presence ofATPyS are shown in Fig. 5, R and C. The short region of dsDSA within the circlr represents the exchanged DNA, and the long duplex tail depicts the unreacted DNA within the duplex substrate.
A similar analysis was camed out for strand exchange renctions involving duplex DNA with short regions of homolorn in the presence of ATP or ATPyS, using electron microscopy Fig.  6). These reactions were cross-linked with AMT prior to removal of the RecA protein and SSR. The paired intermediates were visualized by electron microscopy. and the Iennh of the exchanged regions was estimated as he fort^. In ATP-mcdiated reactions, intermediates were still found prcBferentinlly at the 5' end of the duplex DNA. More than 9Hr; of the duplex DNA substrates were found in joint molecules when homolorn was limited to the 5' end; only about lGr; of thc. substrate duplexes were paired when the homology was at the 3' end (Tnblv I I . The effect of ATP hydrolysis on directional DNA strand exchange and the extent of hybrid DNA formation is less rvidcmt when a short region of homology is used, howevcr. The I m n h of hyhrid regions in the joint molecules ranged up to the homology/ heterology boundary in both 5' or 3' pairing intcrmcdiates (Fig.  6, A and 13). although a sharper peak at thr homology1 heterology houndnw is seen on the 5' end.
In the presence of ATPy!!. there was no preference for joint formation at either end of the duplex DNAs with short regions of homology (Tahle 11, and DNA strand exchange produced hybrid regions ranging up to the homologyheterolo~ houndnT ( Fig. 6, C and D I. Therefore, formation of short rvgions of hybrid DNA proceeds efficiently and without a dirwtional bias in the ahsence of ATP hydrolysis.

RccA-mrdiatrd DNA Strand Exrhnngr in thr Prrsrnrr nf
ATPyS Is Ridirrctionnl-The data in Fig. 4 and Table. I suggest that the strand exchange reaction is hidirectional when ATI'yS replaces ATP. However. the 60-min time points used in these experiments could obscure a substantial kinetic bias in the reaction. In addition. the conclusion of hidirrctionnlity depends on the assumption that joint molecules are initiated at n free end and that strand exchange then proceeds toward thc. middle of the duplex DNA. We carried out experiments to quantify more closely the early stages of the reaction with electron microscopy. The work of Menetski ct nl. ( 19901 indicates that the DNA strand exchange reaction reaches equilibrium within a few minutes in the presence of ATPy!!. IVe concentrated our efforts on the first minutes of reaction to capture prcequilihrium conditions and identify the initiation point of DNA strand exchange. DNA strand exchange reactions were again carried out in the presence ofATPyS with suhstrates n and h of Fig. 1 , except that aliquots were removed and cross-linked with AMT a t 1 , 2, 3 , 5. 10, and 20 min. The samples were visualized by electron microscopy, and a subset of these was examined in detnil as noted below. We detected no bias in the numhers of joint molecules formed at either end of the duplex DNA, even after 1 min o f reaction (data not shown); in this respect. the results are similar to those presented in Tahle I.
The data descrihed helow focus only on intermediates in which one single-stranded circle is paired with one linear duplex DNA. Three types of paired complexes wcw ohsicwed. In the first type, the two DNA substrates were apparently linked at one point, which could he a t a n y point nlong the linenr duplex. These complexes prohahly reflect the initial pairing interaction between the DNA substrates prior to net strand exchange, as documented in numerous studies. The point contacts were not ohserved in the ahsence of cross-linking:. and The observed strand displacement extended to one end of the ends ( p = 1.0 x and 1.0 x for the 5' and 3' ends, duplex substrate in all of these molecules. The third category of respectively). The apparently larger increase observed on the 5' paired complexes consisted of more comp!icated species, which end does not necessarily reflect a bias favoring 5' to 3' branch we presume arose from broken DNA substrates. These repre-migration, since the average increase in the length of exsented, on average, 6% of the paired complexes in the samples changed DNA was modest in both cases. Much of the change (2-3% in many samples) and were ignored in our analysis.
The progression of the strand exchange reaction was similar when the homology was restricted to either the 5' or the 3' end of the dsDNA. Within the 1st min of all reactions, the predominant DNA complexes (73 and 64% of the total paired molecules, on the 5' and 3' ends, respectively) were point contacts. The intermediates in which strand exchange had commenced (25 and 34% of the total on the 5' and 3' ends, respectively) exhibited a short region (less than 1 kbp) of exchanged DNA which was limited to the end of the duplex DNA molecule. There were no unambiguous instances in which DNA strand exchange had initiated from a n unnicked internal site within the dsDNA since no displaced single-stranded loops (D-loops) were observed within the duplex DNA.
As the reaction progressed, the number of complexes with occurred within the first 3 min. The lack of a more dramatic change is consistent with the rapid rate a t which these reactions come to equilibrium in the presence of ATPyS. When pairing occurs near a free DNA end, there appears to be a "burst" of DNA strand exchange which can result in rapid formation of a n exchanged region of 1-2 kbp. Further branch migration is very slow when ATPyS replaces ATP. Nevertheless, the increase in the length of exchanged DNA as the ATPyS reaction progressed indicates that a branch point in a joint molecule can migrate toward the middle of the dsDNAmolecule regardless from which end it initiates and confirms that DNA strand exchange in the presence of ATPyS is bidirectional.

DISCUSSION
ATP hydrolysis is required to render the DNA strand expoint contacts declined in number, whereas the numbers of change reaction unidirectional. When ATP is hydrolyzed, many joint molecules increased. At 20 min, fewer of the complexed more joint molecules are observed on the 5' end of the duplex molecules (18% on the 5' end and 28% on the 3' end) were point substrate, where they are stabilized by unidirectional strand contacts, and more (75 and 70% on the 5' and 3' ends, respec-exchange, than on the 3' end. When ATPyS is present, the end tively) were joint molecules. The absence of internal D-loops strongly indicates that net DNA strand exchange initiates only from a free end of the duplex DNA.
As the ATPyS reactions progressed, the length of hybrid DNA within the intermediates increased as shown in Fig. 7. Between 1 and 20 min of reaction, the fraction ofjoint molecules in which DNA strand exchange had progressed more than 1.5 kbp from a n end increased from 12 to 52% with homology on the 5' end and from 10 to 27% on the 3' end. As determined by a 2 two-way contingency test at the 95% confidence level, the increase in the length of exchanged DNA was significant for both bias is no longer observed. Careful analysis of the reaction indicates that DNA strand exchange with ATPyS is bidirectional. ATP hydrolysis also permits the formation of much more extended regions of hybrid DNA than are formed with ATPyS.
Gels such as those in Figs. 2 and 3 have been used to determine the polarity of reported DNA strand exchange reactions promoted by a wide range of prokaryotic and eukaryotic proteins. An analysis utilizing agarose gels alone can clearly be misleading, however. The effect of ATP hydrolysis in rendering the DNA strand exchange reaction unidirectional is most evident only when (a) a long (several kbp) region of homologous See "Materials and Methods" for details.

I -
DNA is present in the duplex DNA substrate; ( b ) spontaneous branch migration in joint molecules is blocked by cross-linking; and (c) the length of hybrid or heteroduplex DNA formed is considered and is part of the analysis. As noted previously by Radding and colleagues (Dutreix et al., 19911, i t can be especially important to include an assay to assess directly the length or extent of hybrid DNA formation. Our results indicate that another factor, a spontaneous branch migration process that may affect some deproteinized joint molecules more than others, can also produce artifacts in the standard gel assays. We do not yet know why some joint molecules formed with ATPyS are selectively destabilized when they are not crosslinked prior to deproteinization. One earlier study (Rosselli and Stasiak, 1990) found that DNA strand exchange with ATPyS was unidirectional, whereas another did not (Konforti and Davis, 1992). Both studies based their conclusions on the results of agarose gel assays for joint molecule formation; spontaneous branch migration was not controlled, and the extent of hybrid DNA formation was not measured directly. Our results support the conclusion of the latter study. A corollary is that evidence derived from agarose gel assays alone should be regarded as insufficient to establish the polarity of a DNA strand exchange reaction.
Another aspect of the RecA-mediated DNA strand exchange reaction in the presence of ATPyS, its kinetics, deserves some additional comment. There appears to be a rapid, or burst phase of this reaction, during which most of the net strand exchange occurs. After the first 2 or 3 min, additional DNA strand exchange occurs very slowly, consistent with the results reported by Menetski et al. (1990). The only explanation for this phenomenon suggested to date is that continued strand exchange after the burst may be blocked by discontinuities in the RecA filament (Menetski et al., 1990). An alternative explanation can be found in a reexamination of the ATP-mediated process. Kahn and Radding (1984) reported previously a burst phase in the ATP-mediated DNA strand exchange reaction. We have recently confirmed this result and found that 1 kbp or more of hybrid DNA may be formed by RecA nucleoprotein filaments with ATP during the burst phase.3 The early steps of DNA strand exchange may therefore be very similar with ATP ' W. Bedale and M. M. Cox, unpublished results. or ATPyS. Two DNA molecules are aligned, and if a free end is available, a burst of strand exchange occurs in which 1-2 kbp of hybrid DNA are formed. Extension of the exchanged hybrid DNA beyond the initial 1-2 kbp, however, is highly dependent on ATP hydrolysis. Some DNA strand exchange occurs without ATP hydrolysis, and the simple conclusion often drawn is that ATP hydrolysis and DNA strand exchange are unlinked. A RecA nucleoprotein filament has an intrinsic capacity to take up a homologous duplex DNA and promote DNA strand exchange. In the absence of ATP hydrolysis, however, DNA strand exchange is bidirectional, limited in extent, does not bypass structural barriers in the DNA substrates, and does not accommodate four DNA strands. Although not required for DNA strand exchange, ATP hydrolysis clearly confers new properties on the reaction, some of which are likely to be critical for Real function in DNA repair. When ATP is hydrolyzed, DNA strand exchange: (a) is unidirectional; ( b ) generates much more extensive regions of hybrid DNA, ( c ) readily bypasses substantial structural barriers in the DNA (Kim et al., 1992a;Rosselli and Stasiak, 1991); and ( d ) accommodates four DNA strands (Kim et al., 1992b). The overall conclusion we draw from these observations is that a major function of RecA-mediated ATP hydrolysis is to direct and transform the intrinsic DNA strand exchange activity of RecA filaments. ATP hydrolysis is coupled to DNA strand exchange, not as a requirement for the basic strand exchange process, but as an augmentation that greatly enhances the effectiveness of the reaction in recombinational DNA repair. technical assistance. We also thank Drs. Subodh Jain and Murray Clay-