Correction of Large Mispaired DNA Loops by Extracts ofSaccharomyces cerevisiae *

Single base mispairs and small loops are corrected by DNA mismatch repair, but little is known about the correction of large loops. In this paper, large loop repair was examined in nuclear extracts of yeast. Biochemical assays showed that repair activity occurred on loops of 16, 27, and 216 bases, whereas a G-T mispair and an 8-base loop were poorly corrected under these conditions. Two modes of loop repair were revealed by comparison of heteroduplexes that contained a site-specific nick or were covalently closed. A nick-stimulated repair mode directs correction to the discontinuous strand, regardless of which strand contains the loop. An alternative mode is nick-independent and preferentially removes the loop. Both outcomes of repair were largely eliminated when DNA replication was inhibited, suggesting a requirement for repair synthesis. Excision tracts of 100–200 nucleotides, spanning the position of the loop, were observed on each strand under conditions of limited DNA repair synthesis. Both repair modes were independent of the mismatch correction genes MSH2, MSH3,MLH1, and PMS1, as judged by activity in mutant extracts. Together the loop specificity and mutant results furnish evidence for a large loop repair pathway in yeast that is distinct from mismatch repair.

Single base mispairs and small loops are corrected by DNA mismatch repair, but little is known about the correction of large loops. In this paper, large loop repair was examined in nuclear extracts of yeast. Biochemical assays showed that repair activity occurred on loops of 16, 27, and 216 bases, whereas a G-T mispair and an 8-base loop were poorly corrected under these conditions. Two modes of loop repair were revealed by comparison of heteroduplexes that contained a site-specific nick or were covalently closed. A nick-stimulated repair mode directs correction to the discontinuous strand, regardless of which strand contains the loop. An alternative mode is nick-independent and preferentially removes the loop. Both outcomes of repair were largely eliminated when DNA replication was inhibited, suggesting a requirement for repair synthesis. Excision tracts of 100 -200 nucleotides, spanning the position of the loop, were observed on each strand under conditions of limited DNA repair synthesis. Both repair modes were independent of the mismatch correction genes MSH2, MSH3, MLH1, and PMS1, as judged by activity in mutant extracts. Together the loop specificity and mutant results furnish evidence for a large loop repair pathway in yeast that is distinct from mismatch repair.
DNA mispairs can be categorized as either single-base mispairs such as G-T or loops, in which one strand harbors extra nucleotides. Loops can arise in vivo as polymerase slippage events (1,2) within repeating elements or between duplicated sequences. These looped mispairs are precursors for insertion mutations if the loop resides on the newly replicated strand or for deletion mutations if the template strand contains the heterology. Thus one biological consequence of loop repair is to reduce mutation rates of insertions and deletions. Looped mispairs can also occur during genetic recombination. Processing of loop-containing recombination intermediates plays an important role in gene conversion of insertions and deletions (reviewed in Ref. 3) and provides a second, biologically significant role for loop repair. Given the richness of repeating elements and duplicated sequences in eukaryotes, genetic stability of these organisms is particularly dependent on the ability to repair loops of varying sizes. Small loops are processed by mismatch repair in bacteria, yeast, and mammalian systems (reviewed in Refs. 4 -7), but little is known about the processing of large loops. The definition of small and large is dependent on the experimental system, as described below.
In prokaryotes, small loops are substrates for mismatch correction, but large loop repair is absent or occurs inefficiently. In vivo experiments with Escherichia coli and pneumococcus indicate that loops up to 4 nt 1 are corrected by mismatch repair, whereas heterologies of 5 bases or more are usually poorly corrected (8 -11). If repair is elicited by a nearby mismatch, large loops will undergo correction (11). This co-correction of large loops suggests that they are poorly recognized in bacteria, rather than being inherently resistant to correction. In vitro experiments indicate a similar loop size spectrum for mismatch repair. MutS protein binds loops of 1-4 nt but not a 5-base loop (10). Mismatch repair-dependent correction in E. coli extracts is efficient for loops up to 7 nt but repair is very low for 8 -22-base heterologies (12). Thus loop repair in prokaryotes seems limited to small heterologies, and correction is dependent on mismatch repair. Exceptions have been reported in which large loops are processed at low levels in E. coli (13,14).
As summarized below, eukaryotic mismatch repair is active on loops up to about 12-13 nt. In contrast to bacteria, however, a number of experiments indicate that eukaryotes also repair large loops, a possibility that was suggested previously (5). Available evidence suggests that mismatch repair and large loop repair are distinct pathways but that their correction activities overlap partially for certain loop sizes. When transformed into Saccharomyces cerevisiae, plasmids harboring 8-or 12-base heterologies undergo loop correction (15). Repair is partially reduced but not eliminated by mutation in the mismatch repair gene PMS1 (16). Examination of the mutation rates of microsatellite and minisatellite repeats indicates that loops of 1-13 nt are efficiently repaired by mismatch repair, as mutations in MSH2, MSH3, MLH1, or PMS1 lead to increased mutation rates for these repeats (17)(18)(19)(20). In contrast, mutation rates of repeating elements of 16 or 20 bp are unaffected by mutations in MSH2, MSH3, or MSH6 (19). Correction of strand slippage events between small tandem repeats in the DNA polymerase pol3-t mutant also depends on the size of the loop and on the mismatch repair genes PMS1, MSH2, and MSH3; loops of 1 or 7 nt are efficiently processed by mismatch repair, whereas loops of 31 or 61 bases are not targets for this repair system (21). As an alternative to large loop repair, it has been suggested that recombinational repair may function to correct certain types of replicational errors arising in replication-impaired backgrounds such as pol3-t or rad27 (22)(23)(24). In addition to possible loop processing by double-strand break recombination, the evidence in this paper and another (25) provides direct evidence for repair of large loops. Certain alleles appear to undergo correction by an unusual, low efficiency process; transformation with a 38-base heterology arising from one such exceptional allele resulted in an apparent low level of repair that was reduced by mutations in MSH2 or MSH6 (26).
Large loop repair is also active during yeast meiosis. Most large (Ͼ15 bp) insertions or deletions undergo gene conversion at normal levels (27)(28)(29), consistent with correction of a looped intermediate (reviewed in Ref. 3). Looped intermediates can be visualized directly if they are capable of forming stable hairpin heteroduplexes during yeast meiosis (30,31). Such hairpins are poorly corrected in yeast (28,29), but co-correction is observed if a well repaired mismatch is near the hairpin (32). One loop repair activity was revealed by examination of gene conversion of a 26-bp non-palindromic insertion. Gene conversion (and, by inference, loop repair) was reduced in msh2 or rad1 mutants but was unaffected by rad2 or rad14 mutations (33). No other mismatch repair mutants were reported in this study.
Evidence from mammalian systems is also consistent with co-existence of small loop mismatch repair and a distinct large loop repair system. Heteroduplexes with loops up to 283 nt are efficiently repaired when transformed into mammalian cells (34 -36). Similar conclusions have been drawn from recombination experiments (37)(38)(39), although the loops tested were smaller. Microsatellite instability measurements and in vitro assays indicate that loops of 1-4 bases are substrates for mismatch repair (reviewed in Refs. 4 -6). Conversely, in vitro assays on loops of 8 or 16 nt were repaired independently of human MLH1 function (40). In another study (41), loops of 5-27 bases were corrected in extracts defective in human MSH2. Complementation experiments with purified hMutS␣ or hMutS␤ proteins indicated a reduced dependence on the purified proteins as the loop size was increased. These authors (41) attributed large loop repair to a pathway distinct from mismatch repair.
The evidence from yeast and mammalian systems provides indirect support for a large loop repair system that is independent of mismatch repair. In this paper we furnish biochemical evidence for large loop repair in yeast. Efficient in vitro correction was observed for loops of 16, 27, and 216 nt. Substrate specificity was indicated by little to no correction of an 8-base loop or a G-T mispair. Large loop repair required DNA synthesis and was independent of the mismatch repair genes MSH2, MSH3, PMS1, and MLH1. A recent paper by Modrich and colleagues (25) provides biochemical evidence for a similar function in extracts from human cells.

EXPERIMENTAL PROCEDURES
Reagents and Enzymes-Standard reagents, including molecular biology grade CsCl, were obtained from Sigma. Hydroxyapatite resin was a product of Bio-Rad. All restriction enzymes were obtained from New England Biolabs or Stratagene. Exonuclease V was from U. S. Biochemical Corp. Enzymatic reactions were performed as recommended by the manufacturers. Rad1 protein and Rad10 protein were the generous gifts of A. Tomkinson (University of Texas Health Science Center, San Antonio).
Heteroduplex Preparations-f1 phage MR1, MR3, MR9, MR11, MR24, MR30, and MR33 were kindly provided by P. Modrich (Duke University). DNA from these phage was used to create heteroduplex molecules as described below. The details of loop sequence and location are provided by Littman et al. (25). An additional phage variant with a 16-bp insertion in the EcoRI site of MR11 was created in our laboratory by insertion of a duplex oligonucleotide. The resulting phage, MR11ϩ16, contains the sequence 5Ј AATTGCTAGCAAGCTT 3Ј on the viral strand as confirmed by DNA sequencing. The underlined sequence encodes a site for NheI.
Both single-stranded and double-stranded DNA from the f1 phage were purified from E. coli strain JM101 (FЈ traD36 lacI q ⌬ (lacZ)M15 proA ϩ B ϩ /supE thi ⌬(lac-proAB)) using published procedures (42). Heteroduplexes were prepared by slight variations of the method of Lu et al. (43) as described (42,44). Briefly, double-stranded DNA harboring the desired sequence on the complementary (C) strand was linearized with Sau96I. In some cases, HincII was used for linearization. The linear product was mixed with a 10-fold excess (w/w) of single-strand circular DNA containing the viral (V) strand sequence. NaOH denaturation and subsequent neutralization resulted in heteroduplex formation. Two consecutive hydroxyapatite columns were used to remove the bulk of the excess single-strand DNA. Linear homoduplex contaminants were removed by treatment with exonuclease V (U. S. Biochemical Corp.). Heteroduplexes were further purified on one or two CsCl gradients in the presence of ethidium bromide. To create substrates that are covalently closed on both strands, the nicked heteroduplexes were treated with DNA ligase. The resulting covalently closed molecules were purified by agarose gel electrophoresis in the presence of 1 g/ml ethidium bromide. The DNA was subsequently released from the gel slice by treatment with ␤-agarase (New England Biolabs) according to the manufacturer's recommendations. Final heteroduplex preparations were Ն98% pure, as judged by agarose gel electrophoresis.
Nicked heteroduplexes prepared by the method described in the previous paragraph contain a nick 114 bp 5Ј to the loop. In one experiment, the nick was placed 797 bp 5Ј to the loop by cleavage of the double-stranded DNA with HincII instead of Sau96I. Heteroduplex formation and purification were performed the same way for both types of nicked substrate.
Heteroduplexes were created by combining the respective C and V strands from the following f1 phage: G/T, MR3 and MR1 (45) Yeast Strains-Yeast strains used in this study were either DY6 (MATa ura3-52 leu2 trp1 prb1-1122 pep4-3 prc1-407; from B. Jones, Carnegie-Mellon University via T. Hsieh, Duke University) or isogenic derivatives. Gene disruptions to yield msh2::Tn10LUK msh3::TRP1 and pms1⌬ mlh1::URA3 derivatives were performed by two rounds of single-step protocols (46), whereas the rad1::URA3 strain arose from a single round of disruption. Knock-out plasmids were kindly provided by R. Kolodner (University of California at San Diego) for MSH2, M. Liskay (Oregon Health Sciences University) for MLH1, and L. Prakash (University of Texas Health Sciences) for RAD1. Knock-out plasmids for PMS1 and MSH3 were created in our laboratory. All derivatives were confirmed by Southern blotting and by appropriate genetic tests.
Nuclear Extract Preparation-Nuclear extracts were prepared essentially as described by Wang et al. (47). Fifteen liters of cells were grown to an A 600 of approximately 2.0 and harvested, and spheroplasts were prepared by incubation with Zymolyase 100T (ICN Pharmaceuticals). Washed spheroplasts were lysed by homogenization with a Yamato LSC Homogenizer (model LH-41). Nuclei were isolated by differential centrifugation, and nuclear proteins were extracted by addition of NaCl to a final concentration of 0.2 M. Following removal of intact nuclei, protein was precipitated with ammonium sulfate at a final concentration of 0.35 mg/ml. Dialyzed protein was stored at Ϫ80°C. Protein concentration was determined by the method of Lowry et al. (48) after precipitation with trichloroacetic acid.
Loop Repair Assays-All loop repair assays were performed as described here unless otherwise noted. The 15-l reaction mixture contained 20 mM Hepes-KOH, pH 7.6, 1 mM glutathione, 1.5 mM ATP, 0.1 mM each dNTP, 0.05 mg/ml bovine serum albumin, and 24 fmol (0.1 g) of looped substrate. When repair to both strands was being evaluated, the reaction was doubled; the DNA was split in half prior to restriction digestion. In some experiments, dNTPs were omitted with or without the addition of dideoxy-NTPs (0.1 mM each). The repair reaction was initiated by the addition of 50 -100 g of yeast nuclear protein with subsequent incubation at 30°C for 60 min. In some cases 5 pmol of purified Rad1p (with or without 6 pmol of Rad10p) was added to the reaction mix at time 0. Quenching of the reaction was achieved by the addition of 30 l of 25 mM EDTA, pH 8.0. For negative controls, the EDTA solution was added prior to the nuclear extract, and the samples were placed on ice. Following incubation, the substrate was purified by phenol and ether extractions, ethanol precipitation, and drying under vacuum. Each sample was resuspended in 14 l of restriction buffer (1ϫ Buffer 2, 1ϫ acetylated bovine serum albumin; New England Biolabs). All restriction digests contained 6 units of Bsp106I (Stratagene) to linearize the DNA. Repair that led to removal of the loop was evaluated by addition of 5 units of EcoRI for substrates C 27 , V 27 , V 16 , and V 216 . Repair in favor of the loop was evaluated with 2.5 units of NheI. For the C 8 substrate, the corresponding digests utilized 3 units of XcmI or 6 units of XhoI. Repair of the G-T mispair was tested with HindIII or XhoI, as described (45). Following incubation at 37°C for 60 min, the restriction digestions were analyzed on a 1% agarose gel. Unless otherwise noted, the gels were stained in a 1:10,000 dilution of Vistra Green dye (Amersham Pharmacia Biotech), and densitometric analysis of repair efficiencies was performed using a Molecular Dynamics Storm 860 PhosphorImager using the Blue Fluorescence/Chemiluminescence mode. Repair activity was measured as DNA present at 3.1 ϩ 3.3-kb bands divided by the total heteroduplex recovered and then converted to femtomoles of loop repaired per h per mg of yeast nuclear protein. For conventional photographic purposes, the gel was subsequently stained in a solution of 1 g/ml ethidium bromide.
Analysis of Repair Intermediates Produced under Conditions of Limited Repair DNA Synthesis-The analysis of the excision tracts is based on published methods (25,49). The excision intermediates were trapped by the omission of exogenous dNTPs from the reaction, resulting in the inhibition of repair synthesis. In the analysis of the complementary strand, the amount of DNA was doubled (to 48 fmol) to increase the sensitivity of detection. Following restriction by SspI, the samples were separated by electrophoresis through a 6% denaturing polyacrylamide gel (45 mM Tris borate, pH 7.6, 1 mM EDTA, 8.3 M urea) and subsequently electrotransferred to a Hybond-N filter (Amersham Pharmacia Biotech) and UV cross-linked. DNA oligonucleotide probes were 5Ј-endlabeled and used to visualize regions of interest on both the viral and the complementary strands of the substrate. The sequences of the oligonucleotides used were as follows: oBL222 (ATTGTTCTGGATAT-TACCAG) corresponding to the MR11 viral strand nucleotides 5216 -5235 and oBL223 (CTGGTAATATCCAGAACAAT) and oBL224 (AT-TCGCGTTAAATTTTTGTT) corresponding to MR11 complementary strand nucleotides 5235-5216 and 5915-5896, respectively. Analysis was performed using a Storm 860 PhosphorImager (Molecular Dynamics).

Looped Heteroduplex Substrates Are Corrected in Yeast
Nuclear Extracts-Loop repair was examined using a biochemical assay. Loop-containing DNA molecules were created with defined size and location of the heterology (Fig. 1A). We use a nomenclature in which the letter C or V indicates the presence of the loop on the complementary or viral strand, respectively. The numeral indicates the loop size in nucleotides. Thus C 27 refers to a 27-base loop on the complementary strand (Fig. 1A). The two strands carry different restriction enzyme sites. For example, C 27 has an NheI site within the looped complementary strand but an EcoRI sequence on the unlooped viral strand. For V 27 , the loop containing the NheI site resides on the viral strand. The presence of the heterology prevents cleavage by either enzyme. If the loop undergoes correction upon incubation with yeast cellular proteins, repair can be assessed by the acquisition of restriction sensitivity. In our analysis, unrepaired DNA migrates as a 6.4-kb linear molecule, whereas repair products are observed at 3.3 and 3.1 kb. The precise sizes of the DNA bands vary slightly depending on the size of the loop. The strandedness of repair is also revealed by this analysis. For example, correction of C 27 to NheI sensitivity indicates repair in favor of the loop, whereas cleavage by EcoRI is specific for loop removal. Another cis-acting feature of these heteroduplexes is the presence or absence of a site-specific nick on the C strand, 114 bp 5Ј from the loop. These substrates are referred to as nicked or covalently closed, respectively.
Repair activity on nick-containing C 27 and V 27 heteroduplexes is demonstrated in Fig. 1B. The lanes marked 0 min are negative controls where repair was prevented by the addition of EDTA prior to addition of extract. Subsequent scoring for re-striction enzyme sensitivity showed little or no DNA migrating at the position of the repair products. In contrast, incubation for 60 min with yeast nuclear proteins resulted in significant repair of the 27-nt loop. Repair of the C 27 molecule to an EcoRI-sensitive form is shown in the 4th lane, indicating correction that removed the loop on the nick-containing strand. There was little if any repair on the closed, viral strand (to NheI sensitivity). Correction of the V 27 substrate led to substantial repair on both strands (last 2 lanes), a characteristic that is considered in more detail below. Repair of the C 27 and V 27 molecules was dependent on the protein concentration in the assay (data not shown).
Repair Activity Is Determined by Loop Size-Based on the results in Fig. 1, it seemed likely that large loop repair was occurring in yeast extracts. This activity was also specific for the loop size, as shown in Table I and Fig. 2. The specific activity of correction to EcoRI sensitivity was about equal for C 27 , V 27 , and V 216 ( Table I), suggesting that loops in this size range are readily repaired. As shown in Fig. 2, correction of V 16 (3rd and 4th lanes) also occurred at similar levels to V 27 (1st and 2nd lanes), whereas heteroduplexes containing a C 8 loop or a G-T mispair yielded little or no repair under these conditions (5th to 8th lanes). The small amount of repair of C 8 in the 5th lane of Fig. 2 is considered in more detail later. The general lack of repair for C 8 and G-T suggests specificity of the reaction for large loops but not smaller heterologies, with a cut-off somewhere between 8 and 16 nt. The negative results with the G-T mispair and the C 8 loop also argue against nonspecific reactions, such as random excision and resynthesis or simple loop clipping by endonucleases. Had either of these nonspecific pathways been operative, there should have been correction of the G-T mispair and/or the C 8 loop. To test if C 8 repair were being inhibited by a diffusible substance in the heteroduplex preparation, a mixing experiment was performed in which nicked V 27 and C 8 loop substrates were present together. V 27 heteroduplex correction still occurred readily in the presence of the C 8 loop (data not shown), indicating that no inhibitor was present.
Two Modes of Loop Repair Occur in Yeast Extracts-Based on parallels with mismatch repair (reviewed in Refs. 4 -6), it seemed reasonable to expect that in vitro loop repair might require a nicked substrate, with correction directed primarily to the strand containing the discontinuity. However, repair of the nicked V 16 , V 27 , and V 216 heteroduplexes resulted in substantial levels of correction on both strands ( Fig. 2 and Table I). This unexpected result led us to consider the possibility of two modes for repairing loops (Fig. 3). One mode is nick-stimulated and results in correction of the nick-containing strand. For nicked V 27 (Fig. 3A), the nick-stimulated mode would yield NheI-sensitive product. The other mode is nick-independent, with correction leading to removal of the loop and subsequent EcoRI sensitivity for V 27 . In the case of the nicked C 27 heteroduplex (Fig. 3C), both repair outcomes would contribute to accumulation of EcoRI-sensitive material, but no NheI-cleavable DNA would result. The results in Table I for the nicked C 27 , V 27 , and V 216 substrates are consistent with the predictions from this two-mode repair model. We also infer that loop repair by the nick-stimulated mode is directed to the discontinuous strand, regardless of which strand harbors the nick.
If the model in Fig. 3 is correct, covalently closed loop heteroduplexes should undergo correction in predictable ways since the nick-independent mode would still be operative but nick-stimulated repair would be eliminated. This prediction is borne out by results shown in Table I. Covalently closed C 27 substrate was still repaired at high levels, presumably by the nick-independent pathway (Fig. 3C). The covalently closed V 27 heteroduplex still showed the nick-independent activity (EcoRI sensitivity), but the nick-stimulated repair to NheI sensitivity was abolished as predicted in Fig. 3B. The extent of reaction at 60 min by the nick-independent mode was not detectably altered by the absence of the nick-stimulated mode (Table I; compare EcoRI-sensitive repair in covalently closed versus nicked V 27 and C 27 ). However, time course experiments for both V 27 and C 27 revealed that repair of the covalently closed substrate at 15-30 min of reaction was reduced about one-third to one-half compared with a nicked heteroduplex. An example for V 27 is shown in Fig. 4. Time course analysis of the V 27 substrate is particularly informative since the nick-independent removal of the loop can be evaluated separately (by EcoRI sensitivity) from nick-stimulated correction in favor of the loop (to NheI sensitivity). This apparent reduction in reaction rate for the covalently closed heteroduplexes is considered further under "Discussion." The two-mode model of loop repair sheds additional light on   27 , repair stimulated by the nick leads to correction on the discontinuous strand and concomitant sensitivity to NheI. In contrast, nick-independent correction on the looped V strand would yield EcoRI-sensitive products. In the case of covalently closed V 27 , nick-stimulated repair is prevented (indicated by the X) whereas the nick-independent removal of the loop can proceed. For nicked C 27 heteroduplex, EcoRI-sensitive product is generated by both modes of correction because the loop is removed in each case.
the relative repair efficiencies of the nicked C 8 and V 27 heteroduplexes (Fig. 2). For C 8 , as for C 27 , we assume that both outcomes of repair will contribute to loop removal (Fig. 3C), whereas V 27 correction is partitioned between the two possible products (Fig. 3A). Thus the repair efficiency of C 8 (Fig. 2) can be estimated at about 20% that of V 27 (32 units for C 8 divided by 156 total units for V 27 ) or V 16 (32 units/158 units).
Distance from the Nick to the Loop Affects Nick-stimulated Repair but not Nick-independent Correction-If a nick is used as an entry point for helicases and/or nucleases involved in nick-stimulated loop repair, then the location of the nick relative to the loop may provide information about the length of excision tracts associated with loop correction. In contrast, the nick-independent reaction should not be affected by the location of the nick. We prepared a modified V 27 substrate in which the nick was placed 797 bp 5Ј to the loop. Repair time courses were measured for V 27 molecules with nicks either 114 or 797 bp distant. As seen in Fig. 5, nick-independent loop removal to an EcoRI-sensitive product occurred at similar rates for both substrates, as expected. In contrast, nick-stimulated repair to the loop-containing, NheI-sensitive product was sensitive to the location of the nick. Repair occurred readily when the nick was 114 bp away, but correction of the heteroduplex with a nick 797 bp distant was reduced to levels near background. We infer that excision tracts associated with nick-stimulated loop repair are frequently less than 797 bp in yeast. Results with mammalian loop repair (25) suggest that about 50% of excision tracts in human cell extracts may reach 797 bp.
The Two Modes of Loop Repair Require DNA Synthesis-Loop repair is predicted to require DNA repair synthesis if the loop and neighboring bases from the duplex undergo excision and replacement. To test this idea, loop repair assays were performed under conditions of restricted replication. Omission of dNTPs with or without addition of chain-terminating dideoxy-NTPs reduced correction of the C 27 substrate by 62-90%. DNA running at the position of repair products was not a discrete species but rather was a partial smear (data not shown), consistent with the presence of heterogeneous, gapcontaining molecules. When nicked V 27 was used as a substrate, similar dideoxy-NTP inhibition was observed for both repair products. These results indicate that both modes of loop repair are inhibited when DNA synthesis is restricted.
Mapping experiments were performed to map the end points of excision tracts associated with loop repair (25,49). In this approach, DNA synthesis was inhibited by the omission of dNTPs. The DNA samples were subsequently cleaved with SspI to generate a 717-bp fragment spanning the position of the loop. Following denaturing gel electrophoresis, the DNA species were transferred to a filter and probed with strand-specific oligonucleotides. This indirect end-labeling method provides estimates of excision tract end points to within about 10 nt.
Excision tracts for the nick-stimulated repair reaction extend a total of about 170 nt, from the nick to approximately 50 nt beyond the loop (Fig. 6A, lane 2). A number of species are observed, including those with end points both 5Ј and 3Ј to the position of the loop. Comparison with a homoduplex control (lane 3) indicates that many of these intermediates are elicited by the presence of the loop, suggesting their association with nick-stimulated loop repair. When dNTPs are included in the reaction (lane 1), two prominent bands are observed. The upper band corresponds to the location of the nick and presumably reflects unrepaired and unligated material. The lower band corresponds to a position between the nick and the loop; the significance of this species is not yet clear. We note that the estimate of ϳ170-nt excision tracts is consistent with the distance dependence for the nick in this reaction (Fig. 5), in which a nick 797 nt from the loop showed little repair.
The nick-independent loop removal reaction also generates excised intermediates of about 100 -200 nt (Fig. 6, B and C). Excision toward the 5Ј end is clear in Fig. 6B, lane 5, where the majority of the DNA molecules have been shortened by about 100 nt from the position of the loop. Some shorter species are also observed, suggesting that excision tracts can extend up to perhaps 150 nt. In contrast, inclusion of dNTPs (lane 4) yields a limited number of products, the most prominent of which maps to a position just 5Ј to the loop. Association of these intermediates with loop removal is confirmed by the homoduplex control (lane 6), which showed almost no excised intermediates.
When the nick-independent reaction was hybridized to a probe on the 3Ј end of the SspI fragment (Fig. 6C), there was much less difference in the DNA species observed with or without dNTPs. Both lanes 7 and 8 show similar patterns and extents of excision intermediates, although some bands in the 2nd lane are somewhat more intense. The most prominent species maps to a position just 3Ј to the loop and presumably corresponds to the darkest band in Fig. 6B, lane 4. The homo- duplex substrate in lane 9 yielded little if any hybridizing DNA. We conclude from Fig. 6, B and C, that excision tracts associated with nick-independent loop removal are 100 -200 nt in length but are asymmetric with respect to the loop, with the majority of excision extending in the 5Ј direction. These experiments do not distinguish between nicking 5Ј to the loop followed by 5Ј to 3Ј excision or nicking 3Ј to the loop with associated 3Ј to 5Ј degradation.
Genetic Independence of Loop Repair from Mismatch Repair-Large loop repair is independent of the mismatch repair genes MSH2, MSH3, PMS1, and MLH1, as deduced from assays using extracts from an msh2 msh3 strain or from a pms1 mlh1 strain (Table II). Correction was assayed using the nicked V 27 substrate to allow measurement of both repair modes. Loop repair to both possible products occurred at similar levels in extracts from the mismatch repair mutants and the wild-type control. The repair activity in the mutants was consistently as high or higher than in the wild type. Similar results were obtained when nicked C 27 was the substrate (not shown). In our system there is no defect in loop repair activity associated with msh2 msh3 or pms1 mlh1 mutations. We did not test MSH6 because of the overwhelming literature evidence in yeast (19,50,51) that shows no role for this gene when loop size exceeds 1-2 nucleotides.
Loop repair was examined in an extract from a rad1 mutant for two reasons as follows: first to investigate the possibility that nucleotide excision repair factors might be operative on the looped substrates, and second to assess whether the in vitro repair activity might be similar or identical to the meiotic RAD1-dependent loop repair observed by Petes and colleagues (33). Both modes of loop repair in the rad1 extract were reduced to 34 -47% of wild-type levels, suggesting a partial dependence on RAD1. However, the defect in this extract was complex; the addition of purified Rad1 protein or of purified Rad1/Rad10 proteins increased the repair by only 3-11%, still well short of wild-type levels of correction. The protein amounts added (5 pmol of Rad1p with or without 6 pmol of Rad10p) were similar to those used to restore nucleotide excision repair to deficient extracts (47). Thus the role of RAD1 in loop repair remains unclear. The independence of loop repair activity from MSH2 function suggests that loop correction in extracts of dividing cells, described here, differs from the meiotic repair process described by Kirkpatrick and Petes (33) which was dependent on both MSH2 and RAD1. DISCUSSION Large loop repair activity in eukaryotes has been indirectly supported by the results of several laboratories. Our results and those of Modrich and colleagues (25) provide direct biochemical evidence for a distinct pathway of large loop repair. In yeast extracts, loops of 16, 27, and 216 nt were efficiently repaired. Correction of these heterologies was unaffected by mutations in the mismatch repair genes MSH2, MSH3, MLH1, or PMS1. There was no evidence for simple loop clipping or nonspecific processing of C 8 or G-T heteroduplexes, suggesting that correction of 16 -216-nt heterologies is a bona fide repair reaction. Excision tracts associated with loop repair are 100 -200 nt, in contrast to the ϳ500 -1500-nt gene conversion tracts thought to result from meiotic mismatch repair (32,52). These findings indicate an independent pathway for large loop repair, consistent with parallel results from human cells (40,41). Based on similar results in yeast and human cells (25), large loop repair may be a conserved activity in eukaryotes.
Together, mismatch repair and large loop repair provide complementary correction of loops that range from 1 to hundreds of nt. There appears to be overlap between pathways for intermediate size loops. Genetic evidence from yeast indicates that mismatch repair is active on loops of 1-13 nt but that its efficiency is reduced somewhat for 10 -13-nt loops (19). Our in vitro assays for large loop repair indicate the opposite trend; efficiency rises about 5-fold as the loop is increased from 8 to 16 or 27 nt (Fig. 2), similar to the trend in mismatch repairindependent correction in mammalian extracts (41). This specificity is consistent with the idea that both repair activities correct intermediate size loops, albeit with reduced efficiency. Overlapping repair of intermediate size loops by the two activities may thus ensure coverage of a wide spectrum of possible loop sizes. Evidence from mammalian cells (41) is also consistent with loop size overlap between mismatch repair and large loop repair.
The sequence of the heterology does not seem to play an important role in correction of the loops examined here. Aside from the NheI restriction site, the loop sequences were all  This finding further supports the idea that a large loop repair system is active on many different heterologies. One exception in yeast might be loops that form hairpin structures, which are not corrected during meiosis (28,29,31). Transformation experiments with mammalian cells had suggested the possibility of two modes of loop repair (35,36). Repair of 25-or 57-nt heterologies led to loop removal about twice as often as correction favoring the loop. Our results with V 216 , V 27 , and V 216 also showed a 1.5-2-fold preference for loop removal (Table I and Fig. 2). In the mammalian cell experiments (36), the effects of a nick close to (71 or 125 bp) or far from (Ͼ2 kb) the loop indicated that the nick provided a relatively modest effect toward the preference for loop removal. These authors suggested that the loop was the major signal for correction. It should be noted that the fate of the nick following transformation could not be monitored in these experiments. If ligation occurred soon after the heteroduplex entered the cell, one would expect a limited effect on strand selection. We find that there is a distinct difference between the two modes of loop repair in nicked versus closed substrates ( Table I), suggesting that the covalent state of the unrepaired heteroduplex remains largely unchanged under the experimental conditions used. Overall, our experiments support the findings of Weiss and Wilson (35,36) that favor the relative roles of nicks and loops in determining strand selectivity in repair.
If two repair modes are operative, are they due to the same or different correction systems? Experiments with the nicked V 27 substrate allowed the simultaneous measurement of nickstimulated and nick-independent repair. Both modes of correction responded similarly when DNA synthesis was inhibited, suggesting that excision and resynthesis steps are comparable. Extracts from mismatch repair mutants were active in both repair modes. The simplest explanation of these two results is that the two repair modes occur by the same system, but our results do not exclude the possibility of different systems. One distinction we found between the two loop correction modes was a reproducible lag in initial rates for nick-independent repair when the heteroduplex was covalently closed (Fig. 4). We hypothesize that this lag corresponds to an initial, rate-limiting endonucleolytic incision. Once the incision is made, loop removal proceeds at rates similar to or slightly higher than on the nicked heteroduplex. An alternative possibility is that nickstimulated correction might somehow increase repair by the nick-independent mode. Supercoiling cannot explain the time course results because the covalently closed molecules were topologically relaxed. We tested whether Rad1 protein might be necessary for the rate-limiting incision hypothesis. Rather than a specific deficiency in nick-independent correction, both repair modes were reduced to similar extents in rad1 extracts. The addition of purified Rad1p or Rad1p-Rad10p complex failed to restore wild-type levels of correction to rad1 extracts. The potential role of RAD1 remains unclear, and additional experimentation will be required to elucidate the mechanism(s) of loop repair. However, it seems likely that the loop repair activity we observe in extracts of dividing cells is distinct from meiotic loop processing that requires both RAD1 and MSH2 (33).