Opposing roles for DNA structure-specific proteins Rad1, Msh2, Msh3, and Sgs1 in yeast gene targeting

Experimental system

As previously described, we used restriction site polymorphisms in the two homology regions flanking a selectable marker to monitor formation of heteroduplex between corresponding regions of the targeting DNA and the targeted chromosomal locus during TGR (Langston and Symington, 2004) (Figure 1). The restriction enzyme recognition sites were incorporated into 29 bp palindromic inserts that avoid mismatch repair by forming stable, internally complementary loops when paired with DNA lacking the insertion (Nag et al, 1989). Genomic DNA was prepared from isolated single colonies identified as transformants resulting from TGR. Each flanking homology region was analyzed by RFLP analysis after PCR amplification to determine if it contained the restriction enzyme recognition site from the targeting DNA, the chromosomal DNA, or both, the latter indicating a sectored colony resulting from unrepaired hDNA (see Materials and methods for details). Transformants that exhibited hDNA in both flanking homology regions were analyzed further to determine if the same (cis; Figure 2, right) or different (trans; Figure 2, left) strands of targeting DNA at each end interacted with the chromosomal DNA.

We previously examined hDNA formation in wild-type cells during targeted replacement of a chromosomal trp1∷URA3 site by TRP1 (Figure 1A) and a chromosomal met17∷ADE2 site (where ADE2 is under control of the MET17 promoter) with MET17 (Figure 1B). Of the 44 products of TGR that had hDNA in both flanking homology regions, 43 were in the trans configuration (diagrammed in Figure 2A), consistent with a model in which a different strand of targeting DNA at each end invades the chromosome as the dominant mechanism for TGR initiation in wild-type cells (Langston and Symington, 2004).

The mechanism and frequency of TGR are conspicuously altered in the absence of the MSH2 and RAD1 genes

One of the assumptions underlying our experimental system is that palindromic sequence polymorphisms in the flanking homology regions of the targeting DNA escape mismatch repair when hDNA is formed with the native chromosomal DNA. To confirm the validity of this assumption, we analyzed hDNA formation during TGR in strains deleted of MSH2, a homolog of Escherichia coli MutS and a key component of the yeast mismatch repair machinery (Reenan and Kolodner, 1992). Consistent with previous studies (Saparbaev et al, 1996), the frequency of TGR was reduced at both MET17 (decreased four-fold, P=0.01; Figure 3A) and TRP1 (decreased three-fold, P=0.004; Figure 3B) in the msh2 strain. If hDNA in the flanking homology regions was undergoing mismatch repair, we would have expected to see an increased frequency of hDNA in msh2 mutants. To our surprise, we found that the proportion of hDNA was reduced in the msh2 mutant at both MET17 (44/95, or 46% in msh2 versus 65/94, or 69% for the isogenic MSH2 strain, P=0.002; Figure 4A) and TRP1 (43/94, or 46% in msh2 versus 70/103, or 68% for MSH2, P=0.002; Figure 4B). More importantly, we observed that hDNA was altered significantly in the configuration of flanking markers, with eight out of 12 transformants (six of seven at TRP1 and two of five at MET17) with the cis configuration in msh2 versus one cis out of a total of 18 in the MSH2 controls (P=0.0006; Figure 4, bottom). The reduced frequency of TGR and the considerable reduction in the number of transformants with hDNA in the trans configuration in the absence of Msh2 suggests that this protein is important for facilitating the type of TGR found in wild-type cells. The shift to the cis configuration of flanking markers suggests that Msh2 also plays a role in restricting an alternate mechanism of TGR.

The structure-specific endonuclease encoded by the ERCC1 and XPF genes is indispensable for TGR in mammalian cells (Niedernhofer et al, 2001). To investigate the role of the corresponding endonuclease in yeast, we examined TGR in strains deleted of RAD1, the homolog of XPF. In accordance with previously published results (Schiestl et al, 1994; Symington et al, 2000), we observed a modest decline in TGR frequency in rad1 mutants, reduced three-fold at TRP1 (P=0.001; Figure 3B) and two-fold at MET17 (P=0.06; Figure 3A) compared to wild-type controls. However, the proportion of transformants with hDNA was radically reduced in rad1 mutants, with only 6/93 (6%) at TRP1 and 15/94 (16%) at MET17 showing hDNA formation in either flanking homology region compared to 70/103 (68%, P<0.0001) and 65/94 (69%, P<0.0001) in the corresponding RAD1 strains (Figure 4). In rad1 strains, only four transformants, all at MET17, formed hDNA at both ends, and all four of these were in the trans configuration (Figure 4A, bottom), suggesting that a small portion of two-end invasion intermediates can still be processed in the absence of Rad1. This alternative processing does not appear to be mediated by the Rad1-related protein Mus81, as the frequency of TGR in a strain mutated for both MUS81 and RAD1 was reduced only slightly compared to the rad1 single mutant (Figure 3A).

The role of Msh2 in TGR is mediated by its interaction with Msh3 but not Msh6

Msh2 forms separate complexes with Msh3 and Msh6 to process different types of intermediates in mismatch repair and HR (Johnson et al, 1996; Marsischky et al, 1996; Saparbaev et al, 1996). To determine if either of these proteins is involved in gene targeting, we analyzed TGR frequency and hDNA formation at MET17 in msh3 and msh6 strains. As expected, deletion of MSH6, which is specific for base–base or very small insertion–deletion mispairs, had no effect on either the frequency (Figure 3A) or the mechanism (Figure 4A) of TGR. By contrast, deletion of MSH3, which is specific for larger insertion/deletion mispairs, reduced the frequency of TGR two-fold when compared with either wild type (P=0.12) or msh6 (P=0.03; Figure 3A) and altered the mechanics of hDNA formation in a manner similar to that of msh2, with 32/95 (34%) transformants showing hDNA in one or both flanking homology regions compared to 44/95 (46%, P=0.10) in msh2, 60/91 (66%, P<0.0001) in msh6, and 65/94 (69%, P<0.0001) in the wild-type MSH3 strain (Figure 4A). Analysis of transformants with hDNA at both flanking markers showed 10/12 in the trans configuration in the msh6 mutant while 4/4 were in the cis configuration in the msh3 mutant (P=0.008, Figure 4A, bottom), confirming that the role of Msh2 in TGR is mediated by its interaction with Msh3, not Msh6.

Msh2:Msh3 and Rad1 operate together to prevent single-strand assimilation

Because the hDNA patterns observed in the msh2 and msh3 mutants were considerably different from those in rad1, we sought to determine whether Msh2:Msh3 and Rad1 affect the mechanism of TGR by the same or different pathways. To do so, we repeated the TGR experiments at MET17 in an msh2 rad1 double mutant. Differences in the frequency of TGR between the double mutant and the respective single mutants were not significant, but hDNA formation in the msh2 rad1 strain closely resembled that of the rad1 single mutant and was significantly different from the msh2 single mutant. In all, 7/96 (7%) transformants showed hDNA in one or both flanking homology regions in the msh2 rad1 double mutant compared to 15/94 (16%) in the rad1 single mutant (P=0.07) and 44/95 (46%) in the msh2 single mutant (P<0.0001) (Figure 4A). As discussed more fully below, these findings are best explained by a model in which Msh2:Msh3 detect the formation of a large unpaired bubble during TGR and direct nicking of the targeting DNA by Rad1, leading to preferential repair of the bubble mismatch in favor of the chromosomal sequences (Figure 2B).

Patterns of flanking markers support the model of an alternate mechanism for TGR in the absence of RAD1

Previously, it was shown that TGR in wild-type yeast occurs by invasion of separate strands of targeting DNA at the two ends, followed by distinct, apparently uncoordinated processing events at the 5′ and 3′ ends of a given strand (Langston and Symington, 2004) (Figure 2A). A model to account for TGR in the absence of Rad1 is that the process frequently occurs by assimilation of a single strand of targeting DNA into a long, unrepaired hDNA tract encompassing portions of both flanking homology regions (Figure 2D). A detailed analysis of MET17 transformants in rad1 and msh2 rad1 shows that the majority of transformants incorporated the flanking marker from one end of the targeting DNA but not the other (Figure 5). By contrast, at TRP1, more than two-thirds of transformants incorporated both flanking markers in the rad1 mutant (P<0.0001). The flanking markers on the MET17 targeting fragment are much farther apart (2.2 kbp; Figure 1B; depicted by circles in Figure 2) than those on the TRP1 targeting fragment (0.9 kbp; Figure 1A; depicted by boxes in Figure 2) and are less likely to be found in a common hDNA tract. These results strongly support a model for TGR in the absence of RAD1 in which strand invasion by one 3′ end of the targeting DNA is accompanied by heteroduplex extension sufficient to incorporate some or all of the 5′ flanking sequences from the same strand, as shown on the right side of Figure 2. Sequences more distant from the invading end are less likely to be incorporated into the single hDNA tract, especially if the opposite, 5′ end of the invading strand has undergone nucleolytic processing analogous to that at a DSB (depicted by dotted lines in Figure 2).

RAD1 is not required for correction of a point mutation by gene targeting

In TGR, the two ends of the linear targeting fragment are normally identical in sequence to the chromosomal regions flanking the targeted gene, but the gene itself is replaced by a selectable marker. Thus, after strand invasion, base pairing between the targeting DNA and the targeted gene can extend only as far as the homology/heterology boundary represented by the selectable marker. To explain the requirement for Ercc1 during TGR in mammalian cells, Niedernhofer et al (2001) suggested that the Ercc1:Xpf endonuclease cleaves at the junction of homologous and heterologous sequences formed by strand invasion and branch migration (Niedernhofer et al, 2001). If this model is correct, deletion of ERCC1 or XPF (RAD10 or RAD1) should have no effect when the targeting DNA and the targeted gene are nearly identical.

To test this hypothesis, we examined the role of RAD1 during gene targeting of a chromosomal frameshift mutation, met17-sna, using the MET17 targeting fragment (Figure 1C). Like TGR, this type of gene targeting is initiated by separate strand invasions at the two ends in wild-type cells (Langston and Symington, 2004). As shown in Figure 6, deletion of RAD1 had no significant effect on hDNA formation at met17-sna (P=0.37), and all eight transformants recovered from the rad1 strain with hDNA in both flanking homology regions were in the trans configuration. Furthermore, deletion of RAD1 had no effect on targeting frequency at met17-sna (data not shown), all of which suggest that correction of small point mutations by gene targeting occurs in a RAD1-independent manner. These results differ greatly from those during TGR at met17∷ADE2 and trp1∷URA3, where deletion of RAD1 had a significant effect on both hDNA formation and targeting frequency (Figures 3 and 4), suggesting that RAD1 is only required during TGR when replacing a segment of chromosomal DNA with a large heterologous sequence.

The frequency of TGR increases significantly in the absence of the SGS1 gene with no apparent change in the mechanism of TGR

The RecQ family helicase Sgs1 was recently shown to suppress crossover recombination during DSB repair in yeast, possibly through a Holliday junction dissolution mechanism (Ira et al, 2003). With separate strand invasion at each end as the initiating event of TGR, integration of recombinant DNA into the genome requires covalent joining of both the invading and noninvading strands of targeting DNA to the chromosomal DNA. In principle, one strand at each end could be connected through a concerted breakage and rejoining reaction like that postulated to occur during resolution of a Holliday junction. To determine whether Sgs1 plays a role in reversing branched intermediates formed during TGR, we determined the frequency of TGR at the MET17 locus in the presence and absence of the SGS1 gene. In the sgs1 strain, TGR was almost six-fold higher than in the SGS1 control (P=0.0005), showing that the suppression of mitotic crossover recombination by Sgs1 extends to TGR (Figure 3A).

To determine if this increase was attributable to an altered mechanism of TGR, we examined hDNA formation at the MET17 locus in the sgs1 strain. There was a modest although statistically significant reduction in hDNA formation, with 47 of 94 transformants (50%) showing hDNA at either end in the sgs1 strain compared to 65/94 (69%) in SGS1 (P=0.01; Figure 4A). However, among 14 transformants in the sgs1 strain that showed hDNA in both flanking homology regions, subclone analysis showed that all 14 were in the trans configuration, as in the SGS1 control (Figure 4A, bottom), suggesting that TGR is still initiated by separate strand invasions at the two ends of the targeting DNA in the absence of Sgs1. Thus, the increase in TGR frequency seen in the strain deleted of sgs1 does not appear to be attributable to use of an alternate TGR pathway.

Msh2:Msh3 and Rad1 form the boundary between two distinct mechanisms of TGR

If single-strand assimilation occurs during TGR in wild-type cells, it is extremely unproductive because, among 44 transformants with hDNA in both flanking homology regions, only one was in the cis configuration (Langston and Symington, 2004). The best evidence that this mechanism is suppressed during TGR in wild-type cells comes from msh2 and msh3 mismatch repair mutants, where we find a significant shift to the cis configuration of flanking markers (Figure 4, bottom). Assimilation of both flanking homology regions from a single strand of targeting DNA into a long hDNA tract with the homologous chromosomal locus results in a large bubble mismatch between the selectable marker on the targeting DNA and the targeted gene on the chromosome, as shown on the right-hand side of Figure 2. In wild-type cells, we suggest that Msh2 and Msh3 direct repair of the bubble mismatch in favor of the chromosomal sequences, so this pathway does not lead to productive TGR (Figure 2B). In msh2 and msh3 mutants, by contrast, the bubble mismatch is still repaired, but the bias in favor of chromosomal sequences is relaxed (Figure 2C), resulting in more transformants with the cis configuration of flanking markers (Figure 4, bottom). The fact that TGR frequency declines in msh2 and msh3 strains even though these mutations open up a secondary pathway for TGR strongly suggests that Msh2 and Msh3 facilitate targeted integration by the two-end invasion pathway in wild-type yeast (Figure 2, left) while simultaneously inhibiting the single-strand assimilation pathway.

Our data suggest a similar dual role for Rad1, but the effect of rad1 mutation is a highly significant reduction in the proportion of flanking marker hDNA rather than a shift to the cis configuration (Figure 4). Combined with a two- to three-fold reduction in TGR frequency in rad1 mutants (Figure 3), the four- to 10-fold reduction in hDNA proportion is consistent with a significantly reduced ability to complete targeted integration by the two-end invasion pathway in the absence of Rad1. As with msh2 and msh3, we suggest that the bulk of TGR in rad1 mutants occurs by single-strand assimilation; however, unlike msh2 and msh3, rad1 mutants are unable to repair large bubble mismatches formed by the single-strand assimilation pathway (Figure 2D). Thus, after the next round of DNA replication and cell division, one daughter cell fails to grow because it lacks the selectable gene on the targeting DNA leading to the extremely low levels of hDNA (and the lack of cis transformants) observed in rad1 (Figure 4). The failure of rad1 mutants to repair large bubble mismatches has previously been shown during meiotic recombination (Kirkpatrick and Petes, 1997; Kearney et al, 2001), but this is the first evidence for a similar role in large loop repair during mitotic recombination.

Although other explanations are possible for the observed reduction in hDNA, several results support the hypothesis that deletion of RAD1 shifts the balance of TGR from the two-end invasion pathway to the single-strand assimilation pathway. First, as shown in Figure 6, the defect in hDNA formation in rad1 mutants is specific to TGR with a heterologous selectable marker and not a general defect in heteroduplex formation. Second, as indicated by the boxes and circles in Figure 2, the single-strand assimilation model for TGR predicts that markers near the homology/heterology boundary are more likely to be found in a common hDNA tract, and the data presented for rad1 mutants in Figure 5 support this expectation. Finally, we previously examined repair of a single base-pair insertion by gene targeting and found that PMS1 prevents single-strand assimilation in this context. As with the rad1 mutants presented here, deletion of PMS1 was accompanied by a reduction in flanking marker hDNA, consistent with the model presented in Figure 2D (Langston and Symington, 2004).

How do Msh2:Msh3 and Rad1 facilitate TGR in wild-type cells?

The dual activities of Msh2:Msh3 and Rad1 in TGR are strongly reminiscent of the role these same factors play in other types of recombination involving interrupted homology sequences. For example, recombination between directly repeated sequences by single-strand annealing (SSA) leaves unpaired 3′-ended sequences that are stabilized by Msh2:Msh3 and cleaved by Rad1:Rad10 to permit completion of recombination (Schiestl and Prakash, 1988, 1990; Saparbaev et al, 1996). Rad1 and Msh2:Msh3 are also required for the removal of terminal heterologies during Rad51-dependent strand invasion (Fishman-Lobell and Haber, 1992; Sugawara et al, 1997). Similarly, in TGR, annealing between the targeting DNA and the targeted chromosomal locus is limited to the flanking homology sequences and pairing does not occur between the selectable marker and the targeted gene. Rad1 is not required when correcting a single base-pair insertion mutation by gene targeting (Figure 6), so clearly the requirement for Rad1 in TGR is related to the interrupted homology, but its exact role is unclear.

Based on the known substrate specificities of the Rad1:Rad10 and Ercc1/Xpf endonucleases (Bardwell et al, 1994; Davies et al, 1995; de Laat et al, 1998), there are various ways that Rad1 may be acting to facilitate TGR, as shown in Figure 7. Although data presented here support the hypothesis of Niedernhofer et al (2001), that these enzymes act at the homology/heterology border, replacement of the targeted locus requires cleavage at all four positions indicated in Figure 7A or additional processing by unidentified factors to remove the intervening chromosomal DNA. An alternative hypothesis, presented in Figure 7B, involves pairing of the D-loop with the noninvading strand of targeting DNA, resolution of the Holliday junction, and cleavage of the D-loop by Rad1:Rad10, as shown. While an enzyme for processing Holliday junctions has yet to be identified in eukaryotes, a resolvase activity with specificity similar to that of E. coli RuvC has been demonstrated in fractionated extracts of mammalian cells (Constantinou et al, 2001, 2002), and its ultimate identification should allow us to examine its role in TGR. An alternate pathway for gene targeting that is independent of the need for Holliday junction resolution has been suggested involving strand invasion and duplication of the entire donor chromosome onto the two ends of the targeting fragment (Morrow et al, 1997). It is unclear how this pathway would contribute to TGR because the targeted locus is not altered by break-induced replication (BIR). Furthermore, RAD51 is required for BIR (Davis and Symington, 2004) but is less important for gene targeting (Schiestl et al, 1994) and rad1 mutants are defective for TGR but not for BIR (A Davis and LS Symington, unpublished data).

Finally, as shown in Figure 7C, homologous integration of one end of targeting DNA into the chromosome creates two chromosome fragments with directly repeated sequences at both free ends, as in SSA. Repair could proceed by SSA using the overlapping homologies present on the two chromosome fragments. Because the regions of homology are interrupted, a branched structure would be formed at the homology/heterology boundary requiring the Rad1:Rad10 nuclease for resolution. Thus, Rad1 might not be required for the initial integration but for later steps in repairing the resulting DSB.

Deletion of Sgs1 increases TGR frequency by the two-end invasion pathway

Sgs1 was recently shown to suppress crossovers during DSB repair, possibly by a Holliday junction dissolution mechanism (Ira et al, 2003), and it also cooperates with Msh2 and other mismatch repair factors to inhibit recombination between diverged sequences (Spell and Jinks-Robertson, 2004; Sugawara et al, 2004). In the present study, deletion of SGS1 led to a significant increase in the frequency of gene replacement (Figure 3A) without a major alteration in the mechanism of TGR (Figure 4A), which is the opposite of msh2 mutants. Thus, it is likely that Sgs1 does not inhibit TGR as part of a heteroduplex rejection complex but by unwinding recombination structures formed in the homologous flanking regions. Whether these structures are D-loops or Holliday junctions remains to be determined, as both intermediates are substrates for the mammalian RecQ homolog BLM (Karow et al, 2000; van Brabant et al, 2000). Because a similar increase in TGR was previously identified in human and chicken cells deficient in BLM (Wang et al, 2000; Traverso et al, 2003) and WRN (Imamura et al, 2002), it appears that deletion or inhibition of RecQ helicases may be a general technique for improving gene targeting efficiency that can be applied to a wide variety of cell types. These results also confirm strong similarities between yeast and mammals in the early steps of TGR.

We now have a developing picture of the means by which linear targeting DNA containing a selectable marker is precisely integrated into the yeast chromosome at a position determined by the flanking homology sequences on the targeting DNA. It is clear that TGR is initiated by strand invasion into homologous sequences by at least one end of the targeting fragment and that this strand invasion is actively opposed or unwound by Sgs1. In most cases, heteroduplex extension of the initial strand invasion intermediate is blocked by the heterology between the selectable marker and the targeted locus, creating a requirement for Msh2:Msh3 and Rad1:Rad10 for further processing, although the exact nature of this processing is unclear. The heterology block also necessitates a second strand invasion by the opposite end of the targeting fragment to complete TGR by this process, known as two-end invasion/resolution (Figure 2, left).

In some cases, the hDNA tract initiated by the first strand invasion extends far enough to include the opposite flanking homology sequence, incorporating the selectable marker and the targeted locus into a large bubble heteroduplex (Figure 2, right). This process, known as single-strand assimilation, is normally reversed by the concerted repair activity of Msh2:Msh3 and Rad1 (Figure 2B) and thus is rarely observed in wild-type cells. Formation and repair of large insertion/deletion mispairs have also been demonstrated in mammalian cells (Raynard and Baker, 2002), but repair of a preformed 216-nucleotide insertion mispair in mammalian cell extracts was independent of Ercc1/Xpf (Littman et al, 1999), which might explain why this pathway is unavailable for TGR in Ercc1^−/− ES cells.

Yeast strains and plasmids

The S. cerevisiae strains used for this study are derivatives of haploid strain W303-1A (Thomas and Rothstein, 1989) with the corrected RAD5 allele (Fan et al, 1996) and are listed in Table I. Parental strains used for gene targeting assays, LSY697, LSY698, and LSY1103, containing the met17-sna, met17∷ADE2, and trp1∷URA3 alleles, respectively, were described previously (Bartsch et al, 2000; Langston and Symington, 2004). Strains containing msh2∷HIS3, msh3∷HIS3, msh6∷TRP1, and mus81∷kanMX6 alleles were made by PCR-mediated gene disruption (Baudin et al, 1993) of LSY1103 or LSY698, and correct integration was confirmed by PCR or Southern blot. Strains containing msh2∷URA3 were derived from crosses to HKY592-11A (a gift from Hannah Klein, New York University School of Medicine, New York). LSY1683 was derived from a cross between LSY698 and W3643-38A (a gift from Rodney Rothstein, Columbia University, New York).

Plasmids pLL113 (ARS1⁺) and pLL147 used in gene targeting assays were described previously (Langston and Symington, 2004). The TRP1 linear targeting fragment containing diagnostic restriction site polymorphisms in the flanking homology regions was released from pLL113 by digestion with restriction endonucleases SspI and BsaBI. The MET17 linear targeting fragment, also containing flanking restriction site polymorphisms, was released from pLL147 by SpeI digestion. Uncut transformation control plasmids pSB110 and pRS424, containing the MET17 and TRP1 genes, respectively, were described previously (Sikorski and Hieter, 1989; Bartsch et al, 2000).

Determination of transformation frequency and analysis of hDNA

Transformation of competent yeast cells was carried out by the lithium acetate method using 100–300 ng of linear targeting fragment DNA or 50–100 ng of uncut plasmid control DNA along with 50 μg of salmon sperm carrier DNA. Transformed cells were plated onto synthetic complete (SC) medium lacking Trp (SC−Trp) or Met (SC−Met), as appropriate. To exclude transformants arising by circularization of the linear targeting fragment or random integration, Trp⁺ colonies were replicated to medium containing 5-fluoroorotic acid (5-FOA) to assure TGR of the chromosomal trp1∷URA3 marker. Similarly, Met⁺ colonies were replicated to medium lacking both Met and Ade (SC−Met−Ade) to assure TGR of the chromosomal met17∷ADE2 marker.

To determine the frequency of TGR for each strain, colony-forming units (CFU) per 100 ng of transformed linear targeting DNA were compared to CFU/1 ng for the corresponding uncut plasmid for each strain (Figure 3) to normalize for differing transformation competence among strains. Transformation efficiencies at TRP1 were approximately 10-fold lower than those at MET17, reflecting normal variation in transformation efficiencies with different selectable markers and at different locations in the yeast genome. Except where noted, each transformation was performed at least three times and the means and standard deviations from all trials are shown in Figure 3. Statistical comparisons of mean TGR efficiencies between strains were performed using the unpaired t-test. P-values<0.05 were considered statistically significant.

Analysis of hDNA was previously described in detail (Langston and Symington, 2004). Briefly, genomic DNA from ∼96 whole colonies arising from TGR of each strain was amplified by PCR and digested separately with restriction enzymes corresponding to the restriction site on the targeting vector DNA or the site on the chromosomal DNA. PCR products that digested to completion with the enzyme specific to the restriction site on the targeting vector DNA or the site on the chromosomal DNA were scored as ‘vector site only' or ‘chromosomal site only', respectively. PCR products that showed partial digestion with both enzymes were scored as hDNA. For Figure 4 (top), if either flanking homology region showed hDNA, the colony is shown as ‘hDNA either end'; if neither flanking region showed hDNA but at least one region contained vector sequences, the colony is represented as ‘vector site(s), no hDNA'; if both flanking regions showed only chromosomal sequences, the colony is represented as ‘chromosomal sites only'. To compare hDNA formation, differences between strains in the number of hDNA and non-hDNA transformants were treated as categorical variables and analyzed by Fisher's two-tailed exact test. P-values<0.05 were considered statistically significant. Colonies that showed hDNA in both flanking homology regions were subcloned and analyzed by PCR and restriction digestion to determine if hDNA formed between the same strand (cis) or different strands (trans) of the targeting DNA at the two ends, and the results are shown at the bottom of Figure 4.