Slow extension of the invading DNA strand in a D-loop formed by RecA-mediated homologous recombination may enhance recognition of DNA homology

DNA recombination resulting from RecA-mediated strand exchange aided by RecBCD proteins often enables accurate repair of DNA double-strand breaks. However, the process of recombinational repair between short DNA regions of accidental similarity can lead to fatal genomic rearrangements. Previous studies have probed how effectively RecA discriminates against interactions involving a short similar sequence that is embedded in otherwise dissimilar sequences but have not yielded fully conclusive results. Here, we present results of in vitro experiments with fluorescent probes strategically located on the interacting DNA fragments used for recombination. Our findings suggest that DNA synthesis increases the stability of the recombination products. Fluorescence measurements can also probe the homology dependence of the extension of invading DNA strands in D-loops formed by RecA-mediated strand exchange. We examined the slow extension of the invading strand in a D-loop by DNA polymerase (Pol) IV and the more rapid extension by DNA polymerase LF-Bsu. We found that when DNA Pol IV extends the invading strand in a D-loop formed by RecA-mediated strand exchange, the extension afforded by 82 bp of homology is significantly longer than the extension on 50 bp of homology. In contrast, the extension of the invading strand in D-loops by DNA LF-Bsu Pol is similar for intermediates with ≥50 bp of homology. These results suggest that fatal genomic rearrangements due to the recombination of small regions of accidental homology may be reduced if RecA-mediated strand exchange is immediately followed by DNA synthesis by a slow polymerase.

During repair of breaks in double-stranded DNA (dsDNA), RecA-mediated recombination can create fatal rearrangements if the repair joins regions of accidental homology. In this work, we consider the possibility that the progression of DNA synthe-sis triggered by RecA-mediated recombination plays a role in rejecting such rearrangements. Fig. 1A, panel i, illustrates a double-strand break in the dsDNA with red backbones. On each side of the break, protein interactions form a singlestranded DNA (ssDNA) 2 with a 3Ј tail (see Fig. 1A, panel ii) (1). RecA subsequently binds to each ssDNA, forming two ssDNA-RecA filaments (1,2). If the RecBCD pathway is followed, there is a gap between the sequences at the 3Ј ends of the two filaments (3). To complete DNA repair, this gap must be filled by DNA synthesis (1).
The two ssDNA-RecA filaments catalyze alignment of the ssDNA with homologous sequences in dsDNA. If sufficient local homology is present, the base pairing of the complementary strand in the dsDNA is transferred from its original partner in the dsDNA to the invading ssDNA in the ssDNA-RecA filament. That transfer of base pairing is referred to as strand exchange. Strand exchange creates a heteroduplex dsDNA product in which the invading and complementary strands are base-paired (see Fig. 1A, panel iii) (1). After strand exchange, the complementary strand's original base pairing partner, the outgoing strand, remains unpaired, creating a D-loop (4). In the D-loop, the heteroduplex is surrounded by homoduplex dsDNA in which the outgoing and complementary strands remain base-paired. Once the heteroduplex product nears the 3Ј end of the invading ssDNA, a DNA polymerase can extend the D-loop by adding bases to the 3Ј end of the invading strand using the complementary strand as a template (Fig. 1A, panel iv) (1).
The extension of the invading strand in the D-loop depends on an interaction among a DNA polymerase, the 3Ј end of the invading ssDNA, and the complementary strand in the dsDNA. The required geometry between the 3Ј end of the invading ssDNA and the complementary strand can be established by heteroduplex strand exchange products near the 3Ј end of the invading ssDNA (Fig. 1A, panel iv). Because heteroduplex formation depends on the homology between the invading and complementary strands (5)(6)(7)(8), synthesis triggered by RecA-mediated homologous recombination (9) also depends on the homology between the invading and complementary strands. In the presence of synthesis, L, the total length of the dsDNA in which the invading and complementary strands are basepaired, is the sum of the length of the heteroduplex and the S bases in the synthesized dsDNA (see Fig. 1).
In vivo studies (10 -12) have measured the dependence of the frequency of recombination on N, the number of contiguous homologous bp in a short region of homology. Those studies did not detect recombination when N was less than 20 bp (10 -12). The measured frequency of recombination increased exponentially as N increased from 20 to 75 bp but only approximately linearly for N larger than ϳ75 bp (11). Remarkably, the frequency of recombination for a sequence with N ϭ 75 is ϳ10 times larger than the frequency for a sequence with N ϭ 50 (11).
In vitro, the stability of strand exchange products increases with N until approaching an asymptotic value at N ϭ ϳ20 (5,7,8,13). In the absence of ATP hydrolysis, heteroduplex products with N Ն 20 are very stable (13). This is not surprising because RecA is an ATPase (14), so hydrolysis occurs in vivo (15). With ATP hydrolysis, in vitro heteroduplex yield continues to increase slowly from N ϭ 20 to N ϭ 75, but even for N ϭ 75 it does not represent complete product formation (13). Thus, the steep N dependence of in vivo incorporation (10 -12) is not consistent with in vitro measurements of the N dependence of heteroduplex product lifetimes (8) or equilibrium product formation (13). Importantly, in vitro measurements of the influence of N on strand exchange products have not included DNA synthesis (5-8, 13, 16, 17).
If the RecBCD pathway is followed, strand exchange products are reversible until sufficient DNA synthesis has extended the invading strands in D-loops (13,18,19). Thus, the probability of incorporating an N-bp region of accidental homology into a genome may be influenced by the extension of the invading strand in the D-loop that is triggered by strand exchange.
Previous in vitro experiments have shown that N affects the triggering of DNA synthesis by RecA-mediated strand exchange (19). The previous studies indicated that heteroduplex products rarely trigger DNA synthesis unless the number of homologous bases is N Ն 20 (19). However, that previous work did not consider how DNA synthesis that extends the invading strand in a D-loop affects the yield of heteroduplex products or how N influences the progress of invading strand extension after the extension is triggered by the formation of a heteroduplex product (19).
In this work, we extend that study to determine whether DNA synthesis that extends the invading strand in a D-loop affects the yield of heteroduplex products. We also probe how N influences the extension of the invading strand in the D-loop after DNA synthesis is initiated by RecA-mediated strand exchange. Many DNA polymerases can potentially perform DNA synthesis. In this work, we considered two different polymerases, Bacillus subtilis DNA polymerase, large fragment (DNA LF-Bsu Pol) and DNA polymerase (Pol) IV.
DNA Pol IV is up-regulated by the SOS response that follows formation of a double-strand break among other insults (20), and DNA Pol IV is known to extend invading strands in D-loops formed by RecA-mediated homologous recombination (9). Thus, we chose to consider how DNA Pol IV extends the invading strand in D-loops formed by RecA-mediated recombination in vitro.
When DNA synthesis does not require strand displacement, DNA Pol IV synthesis is inefficient (21), whereas synthesis by DNA LF-Bsu is rapid and processive (22). DNA LF-Bsu Pol is a well-characterized, commercially available protein that also provides rapid strand displacement DNA synthesis in vitro; therefore, we chose to compare invading strand extension performed by DNA Pol IV with the invading strand extension performed by DNA LF-Bsu Pol.
For either polymerase, we found that extension of the invading strand in a D-loop increases the yield of heteroduplex strand exchange products. Importantly, when the extension of the invading strand is limited to Յ15 nt, the fluorescence signals for both polymerases are similar. In contrast, when the invading strand can be extended by 50 or more nt, DNA LF-Bsu Pol extends the invading strand more than DNA Pol IV. We also found that the homology dependence of the extension of the invading strand in the D-loop by DNA Pol IV is quite different from the homology dependence of the extension by DNA LF-Bsu Pol. Finally, we present a model that could explain these differences in homology dependence and consider implications of the model for genomic rearrangements.

Results
We used FRET to study both strand exchange and DNA synthesis that extends the invading strand in a D-loop formed by strand exchange. As in previous studies, a fluorescein label is placed on one of the dsDNA strands, and a rhodamine label is placed on the other strand (7,13). Thus, initially rhodamine quenches the fluorescein label. Because the fluorophores are positioned on opposite strands in the dsDNA, the quenching becomes less efficient as the separation between the dsDNA strands increases. Fig. 1B illustrates how heteroduplex formation can increase fluorescein emission if a FRET pair is placed within the region of the dsDNA that is homologous to the invading strand (6,7). Thus, we monitored heteroduplex formation using FRET probes positioned within the region of the dsDNA that is homologous to the invading strand. Previous in vitro experiments studied the N-dependent stability of heteroduplex products in the absence of any DNA polymerase. If heteroduplex stability is influenced by the extension of the invading strand in a D-loop, then in vitro studies that do not include DNA polymerase may not capture the N probability of permanently incorporating regions of accidental homology into genomes. Thus, the influence of DNA polymerase could contribute to the large discrepancy between the N dependence of the in vivo (10 -12) results and N dependence of in vitro (8,13) results that did not include any DNA polymerase.
If extension of the invading strand in a D-loop enhances heteroduplex stability, then that extension should also enhance the yield of heteroduplex strand exchange products. Fig. 2A shows a schematic of experiments designed to test whether the extension of the invading strand in a D-loop increases the yield of heteroduplex products. It is a particular example of the general scheme illustrated in Fig. 1B.
All of the experimental results were obtained using the same 180-bp dsDNA. The dsDNA interacted with 98-nt ssDNA-RecA filaments that were long enough (98 nt) to allow for com-

D-loop extension rates influence homology recognition
plete filament formation. Because the dsDNA was always the same, we controlled N, the number of contiguous bp at the 3Ј end of the invading strand, by changing the sequences of the 98-nt invading strands. The remaining bases in the invading strand were not sequence-matched to the corresponding bases in the complementary strand. In Fig. 2A, the matching sequences are indicated by the green regions in the dsDNA. The ends of the homologous regions are indicated by the green arrows.
The fluorophores are 58 (rhodamine) and 57 bp (fluorescein) from the end of the dsDNA, so D fluor ϳ 58 bp is the separation between the fluorescent labels and the end of the dsDNA. The 3Ј end of the ssDNA-RecA filaments is 50 bp from the end of the dsDNA. We will refer to that separation as D init3Ј , and we define ⌬L ϭ D fluor Ϫ D init3Ј . Fig. 2A shows that ⌬L ϭ 8 bp, so when ⌬L Ͼ 0, the labels are positioned within the region of the dsDNA containing the N homologous bases (the homologous region).
In these experiments, we measure the number of counts per second recorded by a fluorometer. We then calculate ⌬F(t), the difference between the number of counts obtained at time t ϭ 0 when the invading ssDNA is heterologous to the dsDNA (N ϭ 0) and the number of counts measured at a time t.  (Fig. S1). For both polymerases, when filaments are prepared in dATP, the fluorescence signal is larger than the fluorescence signal obtained when the filaments are prepared with ATP ( Fig. 2, C and E). Importantly, experiments using 180-bp dsDNA indicate that in either buffer the increase in fluorescence for DNA LF-Bsu Pol is much larger than the increase in fluorescence for DNA Pol IV.
The differences between the results for the two polymerases might be a consequence of differences in the initiation of the extension of the invading strand in the D-loop. To test whether the two polymerases initiate invading strand extension at different rates, we performed experiments in which the homoduplex tail is preserved for both polymerases because the polymerases only extend the invading strand by 1 base. Fig. S2 shows results of experiments with fluorophores that are 11 (rhodamine) and 9 bp (fluorescein) from the end of the dsDNA, and the 3Ј end of the initiating strand is 11 bp from the end of the dsDNA. Rhodamine modifies a "T" base, and for this 90-bp

D-loop extension rates influence homology recognition
dsDNA in the presence of dATP only 1 nucleotide can be incorporated (23). Fig. S2 shows that either polymerase can extend the invading strand by binding dATP to the rhodamine-labeled nucleotide. Furthermore, the fluorescence signals are similar for both polymerases.
To test whether the extensions for the two polymerases are different when the invading strand can be extended by more than 15 nt, we conducted additional experiments. Fig. 1C illustrates how the extension of the invading strand in a D-loop can increase fluorescein emission if a FRET pair is placed beyond the 3Ј end of the invading strand. In the absence of synthesis, the invading strand does not include any bases in this region; therefore, in the absence of synthesis, strand exchange cannot form a heteroduplex product in this region. Fig. 3A is a particular example of the general scheme shown in Fig. 1C. All of the experiments shown in Fig. 3 used the same 180-bp dsDNA that was used in Fig. 2, but N ϭ 82 for all of the data shown in Fig. 3. We varied the position of the 3Ј end of the ssDNA-RecA filaments so that D init3Ј ϭ 62, 66, 73, 77, or 88 bp from the end of the dsDNA (Fig. 3). Each D init3Ј corresponds to a different ⌬L, where ⌬L is the separation between the fluorophores and the 3Ј end of the ssDNA. By comparing results for different ⌬L values, we can probe the progress of the extension of the invading strand in the D-loop created by the heteroduplex product. In Fig. 3A, In the presence of either polymerase, the ⌬F(t) fluorescence signals for the heterologous ssDNA-RecA filaments are not negligible. The ⌬F(t) curves for DNA Pol IV and DNA LF-Bsu Pol are shown in Fig. S5, A and B, respectively. For each DNA polymerase, the same ⌬F(t) curve for heterologous ssDNA applies to all of the ⌬L values; therefore, consistent with previous work (19), in Fig. 3 for each ⌬L value we show a graph of ⌬⌬F(t), the difference between the ⌬F(t) curve for N ϭ 82, and the ⌬F(t) curve for a heterologous filament.   Because Fig. 3 shows that during 1800 s the extension of the invading strand by DNA LF-Bsu Pol can progress to at least 30 bp, we pursued additional experiments to probe how much further the extension of the invading strand might progress. Fig.  S6A shows a schematic of an experimental design in which the 3Ј ends of the invading strands are 62 and 77 bp from the end of the dsDNA with D fluor ϭ 11, yielding ⌬L ϭ Ϫ51 and ⌬L ϭ Ϫ66, respectively. Fig. S6, B and C, show ⌬F(t) and ⌬⌬F(t) curves, respectively, for the system shown schematically in Fig. S6A. The data in Fig. S6 indicate that synthesis by DNA LF-Bsu Pol can sometimes extend beyond 51 bp, although the ⌬F(t) curve for ⌬L ϭ Ϫ66 just barely exceeds the results for the heterologous filament. These results are consistent with previous measurements, suggesting that the processivity of DNA LF-Bsu Pol is 1-55 nt (22). In sum, the results in Figs. 2 and 3 suggest that DNA LF-Bsu Pol extends the invading strand in a D-loop more than DNA Pol IV and that heteroduplex product formation increases as the extension of the invading strand increases.

D-loop extension rates influence homology recognition
In vivo, dsDNA synthesis will almost never reach the end of a dsDNA, so in what follows we will highlight results with D init3Ј Ͼ ϳ55 bp. We chose D init3Ј Ն 55 because Fig. 3 and Fig. S6 indicate that D init3Ј ϳ 55 is sufficiently large that during our observation time RecA-triggered synthesis by DNA Pol IV will never reach the end of the dsDNA and synthesis by DNA LF-Bsu Pol will almost never reach the end of the dsDNA.
Having shown that heteroduplex product formation and the extension of the invading strand in a D-loop differ for LF-Bsu Pol and DNA Pol IV, we now consider whether the homology dependence of the extension of the invading strand in a D-loop is different for the two polymerases. Fig. 4, A-C, show the schematic representations for a system with D fluor ϭ 58 bp and D init3Ј ϭ 59, 62, and 66 bp, respectively. Those D init3Ј values correspond to ⌬L ϭ Ϫ1, Ϫ4, and Ϫ8 bp, respectively.
The ⌬F(t) curves obtained with DNA Pol IV are shown in Figs. S7, A-C, for ⌬L ϭ Ϫ1, ⌬L ϭ Ϫ4, and ⌬L ϭ Ϫ8, respectively. In each figure, the magenta, green, and blue curves show the results for N ϭ 20, 50, and 82, respectively. Fig. 4, D-F, show the ⌬⌬F(t) curves obtained with DNA Pol IV and all dNTPs for ⌬L ϭ Ϫ1, ⌬L ϭ Ϫ4, and ⌬L ϭ Ϫ8, respectively. The data shown have been averaged over 20 s. Data for ⌬L ϭ Ϫ2 is shown in Fig.  S8. For the data with ⌬L ϭ Ϫ1 and ⌬L ϭ Ϫ2, the results for N ϭ 82 and N ϭ 50 are very similar, but for ⌬L ϭ Ϫ4 and ⌬L ϭ Ϫ8,

D-loop extension rates influence homology recognition
the results for N ϭ 82 and N ϭ 50 differ. Importantly, for ⌬L ϭ Ϫ8, the N ϭ 50 signal is indistinguishable from the signal for N ϭ 20 or the signal for the heterologous filament. Thus, the N dependence of synthesis increases strongly as we monitor fluorescence at positions more distant from the 3Ј end of the filament. Fig. 4 shows that the extension of the invading strand in a D-loop by DNA Pol IV depends strongly on N. To determine whether this strong N dependence applies to all polymerases, we repeated the experiments using DNA LF-Bsu Pol. Fig. 5A shows ⌬⌬F(t) obtained with DNA LF-Bsu Pol for ⌬L ϭ Ϫ1. These data are analogous to the results for DNA Pol IV that are shown in Fig. 4D, so the schematic shown in Fig. 4A also applies to Fig. 5A. For both polymerases, the signals for N ϭ 50 and N ϭ 82 are similar when ⌬L ϭ Ϫ1; therefore, when ⌬L ϭ Ϫ1, neither polymerase distinguishes between the two N values. Fig. 5B shows the schematic for ⌬L ϭ Ϫ48 bp, and Fig. 5C shows the resulting ⌬⌬F(t) curves. Fig. S9 shows the ⌬F(t) results corresponding to ⌬⌬F(t) curves shown in Fig. 5. For DNA LF-Bsu Pol, the ⌬⌬F(t) curves for ⌬L ϭ Ϫ1 and ⌬L ϭ Ϫ48 (Fig. 5, A and C) are quite different, suggesting that for DNA LF-Bsu Pol not all of the strand displacement synthesis that reaches ⌬L ϭ Ϫ1 also reaches ⌬L ϭ Ϫ48; however, in both cases, the result for N ϭ 82 is the same as the result for N ϭ 50. Thus, the results shown in Fig. 5 indicate that even when the extension of the invading strand in a D-loop is as large as 50 bp, the probability that DNA LF-Bsu Pol will extend the invading strand is insensitive to N for N Ն 50. In contrast, Fig. 4 shows that for DNA extension as small as 4 bp, the probability that DNA Pol IV will extend the DNA is sensitive to N even when N Ն 50.
To provide graphical representations of the differences between the N dependences of the synthesis performed by the two polymerases and to offer some comparison with in vivo results, we quantified the N dependence of the DNA synthesis based on the fluorescence signals that we observed after 1800 s. In particular, we considered interactions with ⌬L Ͻ 0, and we defined the ⌬⌬F(1800) ratio [N] as the ratio of the ⌬⌬F(1800) value for a given N to the ⌬⌬F(1800) value for N ϭ 82. Thus, by definition, the ⌬⌬F(1800) ratio [82] ϭ 1. Fig. 6A shows ⌬⌬F(1800) ratio [N] versus N for the experimental results using DNA Pol IV that are shown in Fig. 4 and Fig. S8. The red, orange, blue, and purple curves correspond to ⌬L ϭ Ϫ1, Ϫ2, Ϫ4, and Ϫ8 bp, respectively. Fig. 6A shows that ⌬⌬F(1800) ratio [50] ϳ 1 when ⌬L ϭ Ϫ1 or Ϫ2, but ⌬⌬F(1800)ratio [50] decreases as ⌬L becomes more negative. To compare the in vitro results with in vivo results, we have presented the in vivo results in the same graph.
In vivo experiments measured the probability that a region of accidental homology containing N homologous bases will be incorporated in a genome. To compare our in vitro results with results of those in vivo studies (11), we determined the probability that a sequence including N ϭ 82 bp would be incorporated into a genome. We then normalized the in vivo results for

D-loop extension rates influence homology recognition
all other N values by dividing the measured probability for a particular N value to the probability for N ϭ 82 (green diamonds) (11). Although the red and orange lines depart strongly from the green diamonds, the purple curve is similar to the green diamond result, indicating that for ⌬L ϭ Ϫ8, the ⌬⌬F(1800) ratio [82] versus N is consistent with the N dependence of the in vivo ratios. Thus, for DNA Pol IV, the N dependence of DNA extension by a few bp is not consistent with in vivo results, but the N dependence of extending DNA by ϳ8 bases is.
In contrast to the results for DNA Pol IV shown in Fig. 6A, Fig. 6B indicates that for DNA LF-Bsu Pol, the N dependence of the extension of the invading strand in a D-loop is not consistent with in vivo results even when the synthesis extends over ϳ50 bp. Fig. 6C combines the results shown in Fig. 6, A and B, but each curve corresponds to a given N value, and the x axis represents the ͉⌬L͉ value. Thus, by following a single curve, one can visualize how the N-dependent discrimination changes as the separation between the fluorescent labels and the 3Ј end of the invading strand increases. The results for DNA LF-Bsu Pol are all straight lines, indicating that the N dependence is insensitive to ͉⌬L͉ for ⌬L Ͻ 0. By definition, for N ϭ 82, the curve for DNA Pol IV is also a straight line; however, for lower N values, the curves for DNA Pol IV decrease rapidly as the separation between the fluorescent labels and the 3Ј end of the invading strand increases.

Discussion
In this work, we studied the in vitro extension of invading strands in D-loops formed by RecA-mediated recombination and compared results for DNA LF-Bsu Pol with results for DNA Pol IV. Our observations suggest that DNA synthesis by either polymerase stabilizes heteroduplex products (Figs. 2 and S1) and that both polymerases initiate invading strand synthesis at a similar rate (Figs. S1 and S2); however, during our 1800-s observation time, DNA LF-Bsu Pol extends the invading strand more than DNA Pol IV if the invading strand can be extended by more than 15 nt (Fig. 3). Furthermore, for DNA LF-Bsu Pol, the rapid extension of the invading strand is similar for regions of accidental homology that include over Ͼ50 bp (Figs. 5 and 6). In contrast, for DNA Pol IV, regions of accidental homology that include 82 bp trigger more extensive invading strand extension than regions of accidental homology that include only 50 bp (Figs. 4 and 6). These differences in the homology dependence of invading strand extension are consistent with a

D-loop extension rates influence homology recognition
model suggesting that slow addition of successive bases to the invading strand provides additional kinetic proofreading steps that help discriminate against regions of accidental homology extending over Ͼ50 bp, whereas rapid extension does not offer the same discrimination (Fig. 7). Thus, slow initial extension of the invading strand in D-loops formed by RecA-mediated recombination may play an important role in suppressing genomic rearrangement associated with heteroduplex products formed by regions of accidental homology that include more than 50 bp.
Figs. 2 and S1 suggest that the extension of the invading strand in D-loops triggered by RecA-mediated recombination stabilizes heteroduplex products. Thus, in vitro measurements of the homology dependence of heteroduplex stability performed in the absence of DNA synthesis may not capture the homology dependence governing incorporation of a region of accidental homology into a genome. Fig. 1 shows that if the heteroduplex exists, then extending the invading strand increases L, the total length of the dsDNA in which the complementary and invading strands are paired. This increase in L could underlie the synthesis-dependent increase in heteroduplex yield that is shown in Figs. 2 and S1 as we discuss in the following. Figs. S1 and S2 show that for the 90-bp dsDNA the heteroduplex yields for both polymerases are almost identical. We propose the results are very similar because both polymerases initiate synthesis at a similar rate, and synthesis by either polymerase can remove the ϳ15-bp homoduplex tail at the end of the 90-bp dsDNA. As a result, the L values for the two polymerases are the same for the conditions in Figs. S1 and S2 that limit invading strand extension, S, to Ͻ15 bp. In contrast, Fig. 3 shows that for the 180-bp dsDNA, con-struct S frequently exceeds 15 bp when the invading strand is extended by DNA LF-Bsu Pol; therefore, L values for DNA LF-Bsu Pol are much larger than for DNA Pol IV. Fig. 2 shows that for the 180-bp dsDNA, the heteroduplex yields for DNA LF-Bsu Pol are much greater than for DNA Pol IV. Thus, we propose that Fig. 2 shows different results for the two polymerases because heteroduplex stability increases with L. Furthermore, we speculate that this general result can be extended to other polymerases or to more complex systems.
Double-strand break repairs that erroneously incorporate regions of accidental homology into genomes pose a serious threat to bacteria because bacterial genomes contain many long repeated sequences (19) that can produce genomic rearrangement if regions of accidental homology create strand exchange products (24). Previous work has suggested that RecA-mediated homologous recombination uses multistep kinetic proofreading to discriminate against strand exchange products involving accidental homology (25)(26)(27)(28). The stringency offered by a kinetic proofreading system can be enhanced by adding subsequent kinetic proofreading steps (29). In what follows, we will discuss the differences between the results for the two polymerases and consider implications of those differences for in vivo incorporation of sequences with N contiguous homologous bp. Fig. 6 suggests for DNA LF-Bsu Pol that the S values for N ϭ 50 and N ϭ 82 are the same, but for DNA Pol IV, the S value for N ϭ 82 is larger than the S value for N ϭ 50. Interestingly, Fig. 6 indicates that the homology dependence that governs 8 bp of synthesis by DNA Pol IV in vitro is similar to the homology dependence that governs incorporation of sequences into genomes. is so rapid that every heteroduplex that includes N Ͼ N min and reaches the 3Ј end of the invading strand creates an irreversible product by triggering DNA synthesis that progresses forever even if the heteroduplex partially reverses on the 5Ј side of the invading strand. Thus, once the D-loop forms, it never completely collapses. This system cannot distinguish between N values ՆN min . If N Ͻ N min , then S remains less than S min , so the D-loop can completely collapse. Thus, the system can distinguish between N values if N Ͻ N min . C, case in which S is always less than S min . Thus, the D-loop can always completely collapse, and extension is always influenced by the behavior of the heteroduplex because synthesis proceeds slowly enough that the heteroduplex product reverses before the polymerase synthesizes S min bases. The black arrow pointing to the lower left indicates a case in which only 1 base is synthesized before the D-loop collapses. The arrow pointing to the lower right shows a case in which a second base can be synthesized before the D-loop collapses. The time required for the heteroduplex to collapse depends on N, so S increases with N. After the D-loop collapses, restarting synthesis may require formation of another heteroduplex. Fig. 7 illustrates a very simple model that could explain this result. The model assumes that the stability of heteroduplex products depends on N. Consistent with previous experimental results (19), we assume that extension of the invading strand in a D-loop can only be triggered if the heteroduplex extends very close to the 3Ј end of the invading strand. The model also assumes that until S exceeds S min the extension of the invading strand in the D-loop is promoted by the continued presence of a heteroduplex product; however, after S exceeds S min , the extension of the invading strand in the D-loop no longer depends on the presence of a heteroduplex product. Fig. 7A represents the case in which there is no polymerization at all. This was the situation in previous in vitro studies of the N-dependent stability of heteroduplex products (5)(6)(7)(8). This case is not relevant in vivo. Fig. 7B illustrates a case in which extension of the invading strand in the D-loop is so rapid that S can exceed S min when heteroduplex products include N Ͼ N min . Those N min bases are required for the heteroduplex to live long enough for D-loop extension to exceed S min . For smaller N values, the heteroduplex lifetime is so short that S does not exceed S min even though the extension of the invading strand is very rapid. Thus, extension of the invading strand in the D-loop remains sensitive to N when N Ͻ N min , but is insensitive to N for N Ͼ N min . Furthermore, for N Ͼ N min , DNA synthesis would continue forever, so for N Ͼ N min the fluorescence would be independent of ⌬L.

D-loop extension rates influence homology recognition
If these conditions applied to our experiments, then for all N Ͼ N min the fluorescence curves shown in Figs. 4 and 5 should be independent of ⌬L and approach the asymptotic value corresponding to complete product formation. This is not true for either polymerase; however, for DNA LF-Bsu Pol, the stringency of the strand exchange is independent of ⌬L for N Ն 50. This suggests that although synthesis by DNA LF-Bsu Pol does not continue forever, the extension of the invading strand in the D-loop is insensitive to N for N Ն 50, consistent with the model's predictions for rapid extension of the invading strand in a D-loop if N min Ͻ 50. Fig. 7C represents an intermediate case between Fig. 7, A and B. In this case, during the observation time the increase in S is so slow that S always remains less than S min . As a result, synthesis cannot progress unless a heteroduplex product is present. In addition, if each initiation of synthesis is considered a step in kinetic proofreading, then a distributive polymerase provides additional kinetic proofreading steps. If this were true, the number of kinetic proofreading steps required to reach the fluorophores would increase as ⌬L becomes more negative. Thus, as ⌬L becomes more negative, the ⌬F(t) signals for lower N values should decrease more rapidly than the ⌬F(t) signals for higher N values. This proposal is consistent with the results for DNA Pol IV that are shown in Fig. 4.
Thus, the experiments suggest that it is advantageous for RecA-mediated homologous recombination to be immediately followed by strand displacement DNA synthesis by an especially inefficient polymerase rather than synthesis by a polymerase that can efficiently extend the invading strand in the D-loop. This could explain the previously puzzling in vivo results hinting that DNA Pol IV usually performs the synthesis triggered by RecA-mediated homologous recombination (30) even though DNA Pol IV synthesis is inefficient and error-prone (21). This suggestion is also consistent with in vitro studies showing that DNA Pol IV can extend D-loops formed by RecA, but DNA Pol V cannot (9). Importantly, the error rate for DNA Pol IV is Ͻ10 Ϫ2 -10 Ϫ3 (21). Thus, that error rate may not be very deleterious if only ϳ8 bp must be synthesized before DNA Pol IV is replaced by a DNA polymerase capable of rapidly performing strand displacement synthesis of more than 1 kb of DNA that is typically removed when the RecBCD pathway is followed (3).
In this work, we have only considered in vitro interactions that include ssDNA-RecA filaments, dsDNA, and either DNA LF-Bsu Pol or DNA Pol IV. We do not argue that the in vivo interaction necessarily involves DNA Pol IV, and we do not discount the possibility that other proteins play important roles in double-strand break repairs that follow the RecBCD pathway; however, slow extension of the invading strand in D-loops may provide a previously unsuspected stringency-enhancing intermediate step between RecA-mediated homologous recombination and the rapid and accurate synthesis required to complete the invading strand bases removed by RecBCD.

Strand exchange and DNA synthesis assay
To recreate DNA recombination and synthesis reactions in vitro, we mixed 0.06 M 98-nt ssDNA-RecA filament with 0.06 M labeled dsDNA and finally added DNA polymerase. We used 90-or 180-bp dsDNAs and ssDNAs of varying lengths of homology (from the heterologous N ϭ 0 up to N ϭ 98) and a total length of 98 nt. For polymerase, we used either 1 M Escherichia coli DNA Pol IV (obtained using Pol IV overproducer plasmids (31, 32)) or 5 units of DNA LF-Bsu Pol (New England Biolabs; 5000 units/ml). DNA Pol IV reactions were done in RecA buffer containing 0.1 mg/ml BSA, 2 mM dATP, and 0.4 mM dNTPs, whereas DNA LF-Bsu Pol experiments were performed in RecA buffer containing 1 mM ATP and 30 mM NaCl.
The ssDNA-RecA filaments were prepared by mixing the 0.06 M ssDNA with 2 M RecA (New England Biolabs), 1 mM ATP or dATP, 10 units/ml pyruvate kinase, and 3 mM phosphoenolpyruvate (ATP regeneration system) and with 0.2 M single-stranded binding protein, all in RecA buffer (70 mM Tris-HCl, 10 mM MgCl 2 , and 5 mM DTT, pH 7.6). This mixture was incubated at 37°C for 10 min.

Preparation of dsDNA samples
The short 90-bp labeled dsDNA was obtained by heating a mixture of the corresponding oligonucleotides to 90°C and then cooled back to 40°C at each 1°C per step and equilibrated for 1 min at each step. The fluorescein emission was followed at 518 nm at each temperature step using excitation at 493 nm.
To prepare the long construct, 180-bp labeled dsDNA, a 90-nt ssDNA containing an internal rhodamine label on base 58 (or base 11) from the 5Ј end and a 5Ј end-phosphorylated oligonucleotide (82 bases) containing an internal fluorescein label (position 57 or 9 from the 3Ј end) were first annealed together. Then, a 90-base 5Ј end-phosphorylated oligonucleotide was annealed with a 98-base oligonucleotide to produce another dsDNA but without labels. The two dsDNAs (labeled and unlabeled) were annealed and ligated overnight at 16°C with T4

D-loop extension rates influence homology recognition
DNA ligase in ligase reaction buffer (50 mM Tris, 10 mM MgCl 2 , 1 mM ATP, and 10 mM DTT, pH 7.5; New England Biolabs). The 180-bp construct was purified by running a 3% agarose gel in Tris borate/EDTA buffer for 2 h (6 V/cm). The 180-bp band was visualized with a midrange UV transilluminator and cut. Finally, the dsDNA was extracted from the agarose using a Nucleospin kit (Machery and Nagel, Bethlehem, PA). The oligonucleotide sequences used to prepare the 90-and 180-bp dsD-NAs as well as the 98-nt ssDNA for the ssDNA-RecA filaments are all listed in Tables S1-S3.

FRET measurements
To measure the fluorescence signal of these strand exchange reactions, the ssDNA-RecA filament, dsDNA, dNTPs, and polymerase mixture was placed in a quartz cuvette. Using Fluor-Essence spectroscopy software and the automated FluoroMax spectrofluorometer (Horiba, Edison, NJ), the emission of the fluorescein label was read by using a 2-nm slit and 493-nm excitation wavelength. The fluorescence emission in counts per second (cps) was detected at 518 nm with a 2-nm slit. The reaction was run for 30 min with emission measurements every 1 s and an integration time of 0.5 s. Temperature was kept constant at 37°C.
To calibrate for the formation of free ssDNA outgoing strand, measurements of dilutions of the corresponding ssDNA containing fluorescein were performed. The fluorescence signal in cps was considered equivalent to 100% yield of free outgoing ssDNA formation. The signals at 1800 s in all of the experiments were compared with the signal for 100% yield to calculate the corresponding yields.