DNA loop repair by Escherichia coli cell extracts.

The nick-directed DNA repair efficiency of a set of M13mp18-derived heteroduplexes containing 8-, 12-, 16-, 22-, 27-, 45-, and 429-nucleotide loops was determined by in vitro assay. Unpaired nucleotides of each heteroduplex reside within overlapping recognition sites for two restriction endonucleases, permitting independent evaluation of repair occurring on either DNA strand. Our results show that a strand break located either 3' or 5' to the loop is sufficient to direct heterology repair to the nicked strand in Escherichia coli extracts. Strand-specific repair by this system requires Mg2+ and the four dNTPs and is equally efficient on insertions and deletions. This activity is distinct from the MutHLS mismatch repair pathway. Strand specificity and repair efficiency are largely independent of the GATC methylation state of the DNA and presence of the products of mismatch repair genes mutH, mutL, and mutS. This study provides evidence for a loop repair pathway in E. coli that is distinct from conventional mismatch repair.

DNA mispairs can occur within the DNA helix as a consequence of DNA biosynthetic errors or as a result of recombinational strand transfer between nonidentical sequences (1)(2)(3)(4). Such pairing errors may take the form of base-base mismatches or loops, in which one strand contains one or more unpaired nucleotides. Strand-specific correction of base-base and loop mismatches produced during DNA biosynthesis plays an important role in mutation avoidance (2,5,6), and mismatch repair within the recombination heteroduplex has been implicated in gene conversion (3,4,7,8).
Base-base mispairs are subject to strand-specific correction by the mismatch repair system of both prokaryotes and eukaryotes (5,6,9,10), but action of this system on loop mispairs is limited to fairly small heterologies. The Escherichia coli mismatch repair pathway will correct loops up to about 7 unpaired nucleotides, but larger heterologies are poorly processed by this system (11)(12)(13)(14). A similar specificity is characteristic of the human mismatch repair system, which can correct loops up to about 10 unpaired nucleotides (15)(16)(17).
There is evidence that eukaryotes can rectify large unpaired heterologies by a pathway distinct from the mismatch repair system. Available evidence suggests that specificities of the eukaryotic mismatch repair and large loop repair systems partially overlap. When transformed into Saccharomyces cerevi-siae, plasmids harboring 8-or 12-base heterologies undergo loop correction (18), and repair is partially reduced but not eliminated by inactivation in the mismatch repair gene PMS1 (19). Corrette-Bennett et al. (20,21) found that heteroduplexes containing loops of 16,27, and 216 bases were repaired both in vivo and in vitro by mismatch repair-deficient yeast strains.
Transformation of monkey cells with heteroduplex DNAs containing unpaired single-stranded loops has indicated that mammalian cells can rectify such structures (22)(23)(24)(25). In vitro experiments have also indicated that human cells possess a system distinct from the mismatch repair for processing heteroduplexes with large heterologies. Umar et al. (16) found that incised heteroduplexes containing 8-and 16-nucleotide loops were repaired in nick-directed fashion in extracts of a mismatch repair-deficient cell line. Littman et al. (26) further demonstrated that nicked heteroduplexes containing 12-, 27-, 62-and 216-nucleotide loops are processed in a strand-specific manner and are independent of the mismatch repair system.
Prokaryotes also possess a mechanism that mediates repair of large loops. Using transfection assay, Dohet et al. (11) demonstrated repair of a bacteriophage heteroduplex containing an 800-nucleotide unpaired insertion heterology by a pathway that was independent of mutH, mutL, and mutS gene function. By contrast, Carraway and Marinus (13) failed to detect repair of large heterologies upon transformation of covalently closed circular plasmid heteroduplexes into E. coli. However, if repair is elicited by a nearby mismatch, large loops will undergo correction. Although strand breaks are known to be sufficient for efficient correction by conventional mismatch repair systems (6), a potential activating role for strand discontinuities in large heterology repair in E. coli has not been reported.
To test this possibility, we have constructed a set of heteroduplexes containing unpaired loops of 8 -429 nucleotides, evaluated the dependence of repair on presence and placement of a single-strand break, and tested the possibility that strand placement of a genetic heterology may confer asymmetry on the rectification process.
Construction of Heteroduplex DNA-Heteroduplex DNA substrates were constructed essentially under the conditions of Lu et al. (28). GATC-methylated M13LR derivative replicative form DNA (1 mg) was linearized with BglI or EcoRI and mixed with a 4-fold molar excess of unmethylated viral DNA, followed by alkaline denaturation and annealing. After isolation by hydroxyapatite chromatography, double strand linear homoduplex DNA was removed by treatment with ATPdependent DNase as described (29). The open circular heteroduplex was purified by Sephadex G-200 (Sigma) chromatography and benzoylated naphthylated DEAE-cellulose (Sigma) chromatography in 10 mM Tris-HCl (pH 7.6), 1 M NaCl, 1 mM EDTA (14). The strand break generated by EcoRI was about 49 bp 3Ј to the heterology for the loop size greater than 22 nucleotides and about 70 bases 3Ј to the smaller loops. A nick generated by BglI was about 150 bp 5Ј to the loops. The several-bp variation in the nick heterology distance for the different heteroduplexes is a consequence of the presence of different restriction site markers in the different heterologies.
Covalently closed DNA was ligated with E. coli DNA ligase in the presence of ethidium bromide (96 mmol of dye/mol of nucleotide) and isolated by equilibrium centrifugation in CsCl/ethidium bromide (28). Substrates containing fully methylated GATC sites were prepared by treatment of hemimethylated heteroduplex with Dam methylase as described (28).
Substrates used in this study are summarized in Table I and Fig. 1. By pairing different M13LR insertion derivatives, heteroduplexes containing base-base mismatch or distinct site-specific insertions/deletions were constructed. Extrahelical segments can be located within either viral or complementary strand.
Repair Assays-Growth of cells and preparation of cell extracts were as described by Lu et al. (28). Repair in concentrated E. coli lysate was carried out in 10-l reactions containing 0.02 M Tris-HCl (pH 7.6); 5 mM MgCl 2 ; bovine serum albumin at 50 g/ml; 1 mM ATP; 0.1 mM each dATP, dGTP, dTTP, and dCTP; and 0.1 g (21 fmol) of heteroduplex DNA. The optimal concentration of E. coli extracts was 7.5-10 mg of protein/ml. After incubation at 37°C for 1 h, reactions were terminated by adding 30 l of 25 mM EDTA (pH 8.0), and DNA was purified by phenol extraction and ethanol precipitation. The DNA was then analyzed by restriction endonuclease digestion and agarose gel electrophoresis. DNA products were quantitated after ethidium stain using a gel documentation CCD camera (UVP Ltd.) (14,30).

Construction of Heteroduplexes in Which the Large DNA Heterologies Reside in a Similar Sequence Environment-
Starting from phage M13mp18, we have prepared a set of M13LR derivatives (14) that contain extra nucleotides within the polylinker region located between the single EcoRI and HindIII sites of M13mp18. This set of M13LR derivatives permits construction of heteroduplexes representing base pair mismatches and 8-, 12-, 16-, 22-, 27-, 45-, and 429-nucleotide insertion/deletion heterologies ( Fig. 1 and Table I). The heterology in each of these heteroduplexes is located within a similar environment. Moreover, as shown in Fig. 1, each insertion/ deletion mispair is located within overlapping restriction endonuclease recognition sites, permitting independent evaluation of correction on either DNA strand. Digestion of the heteroduplex DNA with AlwNI ( Fig. 1) and the indicator restriction endonuclease, whose recognition site is blocked by the presence of the heterology (14), will yield a 7.2-kb fragment only. Similar digestion of DNA in which the recognition sequence has been restored by repair will yield 4.1-and 3.1-kb fragments. All substrates used here were refractory to the digestion by the indicator restriction endonucleases in the absence of repair (14) (data not shown).
GATC methylation state dictates the strand specificity of E. coli MutHLS dependent mismatch repair pathway (6). Thus, we also constructed large loop substrates of different methylation states to evaluate the effect of this system. The M13mp18 DNA contains six d(GATC) sequences, which are targets for Dam methylation (Fig. 1). The level of methylation is more difficult to determine for single-strand viral DNA than for double-strand replicative form DNA, which can be analyzed by digestion with DpnI and MboI. Hence, in this study, the unmethylated V-strand of hemimethylated heteroduplex was prepared from the purified phage grown in dam Ϫ E. coli strain, and the methylated Cϩ strand was prepared from the RF DNA grown in Dam methylase-proficient strain. The fully methylated substrates were also prepared and analyzed.
Although the E. coli mismatch repair system is able to repair small insertion/deletion mismatches in a methyl-directed, strand-specific manner (14), the activity of this system on such structures appears to be restricted to heterologies containing less than about seven unpaired nucleotides (14). Although strand breaks are known to be required for efficient correction by eukaryotic loop repair systems (6), a potential activation role for strand discontinuities in large heterology repair in E. coli has not been tested. In order to clarify the nature of this reaction, we have constructed a set of circular heteroduplexes in which the nick was located at different positions to evaluate the potential role of strand-specific single-strand breaks in the processing of such structures. We designate these heteroduplexes according to the DNA strand containing the unpaired segment, the number of unpaired nucleotides in the heterology, and placement of the strand break 3Ј or 5Ј to the heterology as viewed along the shorter path joining the two DNA sites in the circular DNA.

Nick-directed Repair of Large DNA Heterologies in Vitro-
Heteroduplexes containing large insertion/deletion heterologies and a site-specific strand break were tested for repair in E. coli cell-free extracts. Table II compares the efficiency of correction of all heteroduplexes as scored by restriction endonuclease digestion, and Fig. 2 illustrates typical results obtained with the restriction assay.
It is known that a single-strand break located either 3Ј or 5Ј to a mispair is sufficient to provide strand specificity for MutHLSdependent mismatch repair (6). In order to avoid MutHLS-dependent activity complicating our analysis, we choose to test extracts from a mismatch repair-deficient RK1517 mutS strain. As shown in Table II, unpaired heterologies of 22 nucleotides located in the complementary strand or viral strand of a covalently closed circular heteroduplexes were subject to limited processing in both mismatch-proficient and -deficient cell extracts (Table II, Table  II). In reactions containing the MutS-deficient extract, repair of these DNAs displayed a substantial bias (4 -40-fold) toward the incised DNA strand, although significant repair was detected on the closed DNA strand in some cases (Table II; 3Ј-V27, 3Ј-V429, 5Ј-V45, and 5Ј-V429). A G-T mismatch was used as a positive control for methyl-directed mismatch repair. As shown in Table II, the efficiency of nick-directed correction of these large heterologies was 50 -180% of that observed for a 3Ј-G-T mismatch on the unmethylated strand when extracts were derived from the methyl-directed repair-proficient NM522 cells.
Whereas all of the 3Ј-and 5Ј-heteroduplexes used in this study contained the strand break in the complementary strand, the substrates included several with the unpaired heterology present in either the complementary or viral strand (Table II).
When the loop was present in the C strand, processing of the 3Јor 5Ј-heteroduplex produced the deletion repair product, whereas the presence of the loop in the V strand yielded the insertion repair product. Consequently, the strand-specific asymmetry observed for repair of 5Ј-heteroduplexes cannot be attributed to the simple presence of an unpaired segment in a particular DNA strand. Rather, this effect must be due to the strand break. This conclusion is also consistent with the finding that unpaired heterologies of C22 and V22 are processed only at a basal level and without evident strand bias when present in closed circular DNAs (Table II). Thus, although basal processing of unpaired heterologies does occur, the presence of a nick substantially increases the efficiency of the reaction and confers strand specificity on the process. These results clearly indicated that loop heterologies can be processed by a pathway distinct from MutHLS methyl-directed system.
Unpaired heterologies of different sizes were corrected in mutS cells with efficiencies that varied up to 3-fold (Table II; IIIЈC-8 versus 3ЈC-27). In order to clarify the nature of this substrate specificity, we further tested this set in mismatch repair-proficient NM522 cells. As shown in Table II, nick-directed repair of these heterologies in NM522 extracts occurred at levels similar to those observed with extracts of mutS cells (Table II, compare Cϩ entries of NM522 and RK1517). However, significant repair was detected on the unmethylated closed DNA strand in some cases (Table II, VϪ entries of NM522). The variable level of repair observed on the unmethylated strand with NM522 could be due in part to the methyldirected reaction. Additional evidence supporting this view is described below.
Low Level in Vitro Correction of Large Loops Is MutS-and Methylation State-dependent-Although repair of these large loop DNAs displayed a substantial bias (up to 12-fold) toward the incised DNA strand, significant repair was detected on the unmethylated closed DNA strand, in some cases with NM522 extracts (see Table II, VϪ entries of V45 and V429). Moreover, almost all the loop repair that occurred on the unmethylated closed strand in NM522 extracts was slightly higher than that observed with extracts derived from mutS cells (Table II, VϪ entries of NM522 and RK1517). This observation suggested that a heteroduplex containing a large loop may provoke some level of methyl-directed repair. To ascertain that part of large loop repair was truly methyl-directed, we tested a subset of heteroduplex substrates in the fully methylated state (Table  III). Since methylated d(GATC) sequences are resistant to nicking by MutH endonucleases (31), methyl-directed correction of mismatched bases and loops should be greatly reduced under these conditions. The results obtained with fully methylated Vϩ/Cϩ heteroduplexes (Table III) are consistent with this prediction. The repair levels of G-T mismatch on methylated viral strands in wild-type extracts were much lower than unmethylated counterparts in Table II. Repair on methylated viral strands of C45, V45, and V429 heteroduplexes by NM522 extracts were also reduced to the levels observed with hemimethylated DNAs in MutS-deficient extracts. Thus, we attribute the low, but detectable, level of repair that occurs on the unmethylated strand of substrates containing a large heterology to the methyl-directed pathway.
Requirements for Nick-directed Loop Repair in E. coli Extracts-The only exogenous cofactors required for in vitro loop repair by E. coli extracts are Mg 2ϩ and four dNTPs. Omission of any of these components resulted in substantial decrease in repair of the several 3Ј-heteroduplexes tested (Table IV). The addition of exogenous ATP was not required in loop repair (Table IV). This observation further distinguishes large loop repair from the conventional mismatch repair system, since HindIII ATP is essential for the MutHLS pathway (6). However, we cannot exclude the possibility that the dATP used for repair resynthesis may substitute for ATP required in the reaction. Genetic Independence of Nick-directed Loop Repair from Mismatch Repair-Methyl-directed mismatch repair is dependent on the product of mutH, mutL, and mutS genes. In addition, MutHLS have been implicated in the repair of 1-7-base deletion heteroduplexes (12,14). Consequently, cell extracts were prepared from E. coli strains that were defective in each of these gene products and were assayed with large nucleotide insertion/deletion heteroduplexes described above. As shown in Table II, correction of large nucleotide heterologies is independent of the presence of the mutS gene product in crude extracts. Large loop repair is also independent of the mismatch genes mutL and mutH, as deduced from assays using extracts from a mutant strain (Table V). Correction was assayed using the nicked C12, V22, C45, and V429 substrates. Loop repair on nicked strand occurred at similar levels in extracts from mutL and mutH mutants and the wild-type control. Thus, in our system there is no defect in nick-directed loop repair activity associated with mutH, mutL, or mutS mutations.
Nick-stimulated Repair Efficiency Is Independent of the Distance Separating the Loop and the Strand Break-If a nick is used as an entry point for helicase 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 repair patches associated with loop correction. In order to assess dependence of the reaction on location of the strand break, we further prepared modified C22, V22, and C45 substrates in which a single-strand break was placed 655 or 3154 bp 5Ј to the loop. Heteroduplexes in the covalently closed circular form were also prepared as controls. As summarized in Fig. 3, un-FIG. 2. Repair of heteroduplex containing a mismatch or loop heterologies in E. coli extracts. Repair reactions with E. coli strains RK1517 (mutS) and NM522 (mismatch repair-proficient) cell-free extracts were performed as described under "Experimental Procedures." DNA products were digested with AlwN1 and the appropriate restriction endonuclease (Table I)   paired heterologies of 22 or 45 nucleotides were subjected to only low level processing in E. coli extracts when present in a covalently closed circular DNA. A strand break located as far as 3154 bp 5Ј to a mispair is sufficient to provoke the loop repair (Fig. 3, V22 and C45). This observation suggests that E. coli loop repair can function even under conditions of large nick heterology separation distance. Thus, although basal processing of unpaired heterologies does occur (Table II and Fig. 3, ccc entries), the presence of a strand break substantially increases the efficiency of the reaction and confers strand specificity on the process. The low levels of strand-independent reaction that we observe with covalently closed circular heteroduplexes could be the consequence of events directed by a strand break produced by endogenous endonucleases present in the extracts. The same activity may produce the low level of nick-independent repair on the closed strand of the mutS reaction (Table II). However, with increasing length of heterology, we have occasionally observed higher levels of reaction on heteroduplexes that are independent of MutS and strand break ( Table II, VϪ entries of  V27, C45, V45, and V429 in RK1517). Consequently, it is possible that large loops may be directly recognized and processed in a nick-independent fashion by other activities. Repair in these cases may be mediated by mechanisms such as the pathway described by Fishel et al. (32). DISCUSSION The major finding of this study is that repair of loops in E. coli extracts can be very efficient. In vitro assays with loops of 8,12,16,22,27,45, and 429 nucleotides show a high level of correction for all substrates. Under these assay conditions, the efficiency of loop repair was comparable with that of methyldirected mismatch repair. This extent of correction indicates that bacteria have substantial capacity to correct loop size well beyond the range of mismatch repair (14). The other significant conclusion from this study is that in vitro loop repair in bacteria shows close similarity to loop repair activities identified in extracts of yeast (20) and mammalian cells (16,24,26). The similarities of loop repair in prokaryotes and eukaryotes (20,26) and the independence these systems with respect to mismatch repair components suggest conservation of function, although the activities responsible for large heterology repair have not been established in any organism.
Large heterologies were subject to limited processing when present in covalently closed circular heteroduplexes, but repair was enhanced substantially by a strand break placed 3Ј or 5Ј to the loop, and in this case, rectification was highly biased to the incised strand. This dependence on nick but not Dam methylation state also distinguishes the reaction from mismatch repair, which can be directed to the unmethylated strand by a hemimethylated d(GATC) site. The two pathways are also dissimilar, as judged by the nature of reaction requirement (Table IV).
Transformation of E. coli cells with plasmid heteroduplexes by Carraway and Marinus (13) has previously indicated that large loops are not rectified in E. coli cells. In contrast, we have observed a significant degree of nick-dependent processing of large loops in cell extracts of E. coli. There are several potential explanations for the different observations. Whereas our heteroduplexes contained a strand break that activates large heterology repair, the transformation studies relied on covalently closed, circular DNAs, which we have found to be relatively weak substrates for large loop repair. The results of Carraway and Marinus may also have been complicated by strand loss effects that have been observed during transformation of E. coli with heteroduplex DNAs (13).
Biochemical experiments in eukaryotic cells clearly indicate that a 5Ј nick can be utilized to direct loop repair to the discontinuous strand. This preference is strong in human cell extracts (26) and also occurs to a lesser extent in yeast extracts (20). The loop repair activity in E. coli extracts demonstrates a similar preference. However, the interpretation of 5Ј-substrate reaction in E. coli extracts may be complicated by DNA polymerase I-mediated nick translation activity (Tables II and V, Cϩ reactions of 5Ј-G-T). Although we cannot exclude this possibility, it is important to note that we have observed variable repair efficiencies for different heterologies under conditions where each is directed by the same 5Ј-strand break. Furthermore, we have demonstrated effective heterology repair when the nick directing the reaction is located 3Ј to the loop. This observation excludes the possibility of nick translation activity, since such an activity would depend on a 5Ј nick location. It therefore seems highly unlikely that the correction events described above are a mere consequence of nonspecific nick translation. Effective correction of 3Ј-circles in E. coli cells also suggests that bacterial nick-directed loop repair has broader substrate specificity than that observed in the human pathway (26). The demonstration of strand-specific repair of loops in E. coli extracts raises questions concerning the function of this reaction. Although such a system may function in the processing of recombination heteroduplexes, the strand specificity of the reaction suggests a role in correction of insertion/deletion heterologies that arise by DNA misalignment events during replication (1,33). The observation of Fig. 3 that the efficiency of E. coli loop repair is independent of the distance separating the loop and the strand break suggests that the nick may act as a strand signal and not as a free end for excision and resynthesis. A nick as a strand signal and activator in heteroduplex corrections is not without precedent. Varlet et al. (34) demonstrated that DNA strand breaks act as signals rather than excision points in Xenopus mismatch repair. Furthermore, the experiments on the mechanism of lagging strand replication by E. coli DNA polymerase III holoenzyme demonstrated that the repli-case cycles from one DNA site to another via preassembled DNA sliding clamps. It was suggested that the clamp, left on the DNA at the internal DNA termini (e.g. those of the Okazaki fragments) may be harnessed by other machineries coordinated with chromosome replication (e.g. the repair and recombination systems) and used as a signal for the newly synthesized strand (35,36). Therefore, editing of DNA replication and recombination processes by the loop repair components could be accomplished using strand discontinuities as strand discrimination signals.
Together, mismatch repair and loop repair provide complementary correction of loops that range from one to hundreds of nucleotides. There appears to be overlap between pathways for certain loops. Since we have found that repair of 45-and 429-nucleotide loops are partially dependent on MutS and methylation state, we propose that the sequence of the loop may adopt a configuration suitable for MutS recognition. These results imply that mismatch repair has, under certain circumstances, some activity on loops as large as 429 bases. The overlapping substrate specificities of mismatch and loop repair systems may serve to ensure coverage of a wide spectrum of possible insertion/deletion heterologies. Isolation of the activities involved in the pathway and the identification of the corresponding structural genes should serve to further clarify the roles of this system.  3. Loop repair efficiency and nick positions. Heteroduplex repair of the C DNA strand was determined as described under "Experimental Procedures," and the reaction (10 l) contained 75 g of NM522 cell extracts and 21 fmol of circular heteroduplex DNA with the indicated heterology. Heteroduplexes contained a single-strand break on the C strand at the EcoRI site 49 nucleotides 3Ј to the heterology (3Ј-49), at the BglI site 150 nucleotides 5Ј to the heterology (5Ј-150), at the BglII site 655 nucleotides 5Ј to the heterology (5Ј-655), or at the AlwNI site 3154 nucleotides 5Ј to the heterology (5Ј-3154). Covalently closed circular (ccc) heteroduplexes were also tested. Heteroduplexes of C22 (black bar), V22 (white bar), and C45 (gray bar) were tested in this experiment. The error bars represent one S.D. from three determinations.