Two nicking enzyme systems specific for mismatch-containing DNA in nuclear extracts from human cells.

We have identified two novel enzyme systems in human HeLa nuclear extracts that can nick at specific sites of DNA molecules with base mismatches, in addition to the T/G mismatch-specific nicking enzyme system (Wiebauer, K., and Jiricny, J. (1989) Nature 339, 234-236). One enzyme (called all-type) can nick all eight base mismatches with different efficiencies. The other (A/G-specific) nicks only DNA containing an A/G mismatch. The all-type enzyme can be separated from the T/G-specific and A/G-specific nicking enzymes by Bio-Rex 70 chromatography. Further purification on a DEAE-5PW column separated the A/G-specific nicking enzyme from the T/G-specific nicking enzyme. Therefore, at least three different enzyme systems are able to cleave mismatched DNA in HeLa nuclear extracts. The all-type and A/G-specific enzymes work at different optimal salt concentrations and cleave at different sites within the mismatched DNA. The all-type enzyme can only cleave at the first phosphodiester bond 5' to the mispaired bases. This enzyme shows nick disparity to only one DNA strand and may be involved in genetic recombination. The A/G-specific enzyme simultaneously makes incisions at the first phosphodiester bond both 5' and 3' to the mispaired adenine but not the guanine base. This enzyme may be involved in an A/G mismatch-specific repair similar to the Escherichia coli mutY (or micA)-dependent pathway.

G mismatch-specific repair has been identified in E. coli (10)(11)(12)(13) and S. typhimurium (14). This mutY (or micA)-dependent pathway (15) acts on A/G mismatches to restore C/G base pairs exclusively, and in conjunction with MutT protein, also can reduce C/G-to-A/T transversions (16). Specific binding and nicking to DNA fragments containing A/G mispairs have been identified in E. coli extracts (16). The mechanism of the mutY (or micA)-dependent repair involves the action of a DNA glycosylase (17) followed by a 2-nucleotide excision and subsequent resynthesis (16). ' Recent discoveries support a common evolution of mismatch repair machinery among diverse organisms. Protein sequences of MutL of S. typhimurium (18), HexB of Streptococcus pneumoniae (19), and P M S l of Saccharomyces cereuisiae (20) have conserved regions. Proteins with significant homology to the MutS protein of S. typhimurium were found in human and mouse tissue (21, 22). Also, a 100-kDa protein has been identified that binds A/C-, T/C-, and T/T-containing DNAs in human Raji cells (23). Thus, mammalian cells may use mechanisms similar to those found in prokaryotes to correct replication errors in favor of the parental strand (24). I n vitro repair systems directed by strand breaks have been established in nuclear extracts of HeLa and Drosophila cells (25). A specific repair system in human HeLa cells can repair deaminated 5-methylcytosines (26) and is equivalent to the T/G-specific pathway found in E. coli (6). Binding to and nicking of T/G-mismatch-containing DNA have been reported in nuclear extracts of HeLa cells (27,28). The nicking of T/G-containing DNA is mediated through a DNA glycosylase and an apurinic/apyrimidinic (AP)2 endonuclease reaction (28,29). In this paper, we describe two novel nicking enzymes in HeLa nuclear extracts; one can recognize all eight mispairs and the other can only recognize A/G mismatches. These two enzymes can be distinguished from each other and from the T/G-specific nicking enzyme (28) by column chromatography and substrate specificity.

EXPERIMENTAL PROCEDURES
DNA Preparations-Eight 116-mer oligonucleotides (four upper and four lower strands, Fig. 1) were synthesized by a MilliGen 7500 DNA synthesizer and purified from 8% sequencing gels. The bases at position 51 of the upper strand and position 70 (counted from the 5' end) of the lower strand vary by A, C, G, or T . Two complementary 116-mer oligonucleotides were annealed to generate a heteroduplex DNA containing a mismatched base a t position 51 (of the upper strand). The annealed duplexes were labeled at the 3' end on the upper or lower strand with a DNA polymerase Klenow fragment and [u-"'P]dCTP or [a-"'P]dATP, respectively (30). After 25 min at 25 "C, the synthesis was completed by adding all four unlabeled deoxynucle- ' The abbreviations used are: AP, apurinic/apyrimidinic; HEPES, 4-(2-hydroxyethyl)-l-piperazineethanesulfonic acid. oside 5"triphosphates and incubated for an additional 5 min. The resulting filled-in duplex DNA is 120 base pairs in length. Alternatively, the upper strand was labeled at its 5' end with T4 polynucleotide kinase and [T-~'P]ATP before annealing with the lower strand.

Identification of Base Mismatch-specific Endonucleases in
HeLa Cells-In order to identify human enzymes that can nick mismatch-containing DNA fragments, we employed an assay similar to the specific nicking near mismatched bases of E. coli A/G endonuclease (16). Synthetic double-stranded DNA fragments containing different mismatches at one particular position ( Fig. 1) were incubated with HeLa nuclear extracts and then fractionated on a denaturing sequencing gel. The two DNA strands in Fig. 1 were arbitrarily defined as upper and lower strands. Initially, nicking was assayed with DNA fragments containing T/G or A/G mismatches, using C/G-containing DNA as a control. In HeLa nuclear extracts, nicking activities could be detected to T/G-or A/Gcontaining DNA but not to homoduplex DNA (data not shown). Nicking of T/G-containing DNA has been shown to proceed through a DNA glycosylase-AP endonuclease pathway in nuclear extracts of HeLa cells (28,29). Our results suggested that HeLa nuclear extracts might contain other mismatch-specific nicking enzymes. Therefore, these nuclear extracts were fractionated using a Bio-Rex 70 column and assayed for the nicking activities to mismatch-containing DNA substrates. As shown in Fig. 2, fractions 60-70 (Fraction 11-A) were able to nick T/G-or A/G-containing DNA substrate at the proximity of the mismatched site but not to A/ Aor C/G-containing DNA substrate. Fractions 110-130 (Fraction 11-B) could nick A/A-, TIC-, and A/G-containing DNA but not homoduplex DNA. Fraction 11-A had higher activity to T/G-containing DNA than to A/G-containing DNA. However, T/G-containing DNA was a poor substrate for Fraction 11-B.
Further purification of Fraction 11-A on a DEAE-5PW column yielded two overlapping peaks of activity (Fig. 3). Peak fraction 28 showed nicking activity for T/G-containing DNA, and peak fraction 32 showed it for A/G-containing DNA. However, Fraction 11-B could not be separated into subpeaks by DEAE-5PW chromatography (data not shown). Therefore, at least three mismatch-specific endonucleases can be observed in HeLa nuclear extracts. A/G Mismatch-specific Endonuclease Is Present in HeLa Nuclear Extracts-When Fraction 11-A from the Bio-Rex 70 column was assayed with DNA substrates containing one of the eight mismatches, we found it could only nick DNA containing T/G or A/G mismatches. We suspected that Fraction 11-A might contain one T/G-specific nicking enzyme (or a DNA glycosylase and an AP endonuclease) as reported by Wiebauer and Jiricny (28) and an A/G-specific enzyme similar to the E. coli mutY (or micA)-dependent A/G-nicking enzyme (16, 17). To prove this hypothesis, Fraction 11-A was further purified by DEAE-5PW chromatography. This fraction could be separated into two peaks ( Fig. 3) with one T/Gspecific and one A/G-specific enzyme. When these two enzymes were further separated by a third chromatographic step (heparin-agarose), the A/G-and T/G-specific endonuclease activities did not overlap each other (data not shown). The T/G-specific enzyme was proven to be a DNA glycosylase (data not shown), the same enzyme identified by Wiebauer and Jiricny (28), and was not further characterized. Fig. 4  (lanes 1-9) shows the A/G-specific enzyme has no nicking activity for T/G, A/A, TIT, GIG, C/C, CIA, C/T, or C/Gcontaining DNA. Furthermore, this A/G-specific nicking enzyme could only nick the " A strand but not the "G" strand (Fig. 5). These properties are similar to the A/G-specific enzyme of E. coli.

HeLa Nuclear Extracts Contain a Novel Enzyme That Can
Nick All the Mismatch-containing DNA-Fractions 110-130 (Fraction 11-B) from the Bio-Rex 70 column could nick effectively all mismatch-containing DNA (Fig. 1) labeled at the 3' end of the upper strand (Fig. 4, lanes 10-18). The nicking efficiency of mismatch-containing DNA as determined by densitometry was in the following order: C/C > A/A = C/A = C/T > A/G > G/G > T/T > T/G. However, no specific nicking was detected by using mismatch-containing DNA labeled at the 3' end of the lower strand (data not shown). The nicking at the mismatched site with broad substrate specificity and strand disparity are the unique characteristics of this "all-type" enzyme. There is no enzyme yet identified in prokaryotes equivalent to this human enzyme.
Requirement of the AIG-specific and All-type Nicking Enzymes-Both A/G-specific and all-type nicking enzymes did not require M$+ and ATP for cleavage, but activity was slightly enhanced by adding Zn2+ (data not shown). The concentration of NaCl dramatically affected the activity of the all-type nicking enzyme (Fig. 6). The nicking activities for T/G-specific, A/G-specific, and all-type nicking enzymes were decreased to 66, 21, and 096, respectively, when they were assayed in buffer containing 80 mM NaCl compared with no NaCl. Salt concentration may have an effect on mismatch conformation or kinetic parameters governing the formation of protein-DNA complexes.

Incision Sites of the A/G-specific and All-type Nicking En-
zymes-We have used the DNA substrates labeled at different ends and different DNA strands to determine the cleavage sites of the A/G-specific and all-type nicking enzymes. The denatured cleavage products were separated on a sequencing column was further purified by heparin-agarose chromatography to generate Fraction IV, which was then assayed in lanes [1][2][3][4][5][6][7][8][9]. Fraction 11-B from a Bio-Rex 70 column was the enzyme used in lanes 10-18. DNA substrates containing a different mismatch were assayed with the enzyme fractions as described in the legend to Fig. 2. gel in parallel with a sequencing ladder generated by the Maxam and Gilbert chemical method (31). As shown in Fig.  7a, the cleavage product of all-type enzyme on A/G-containing DNA ran at the same position of the T"' band of the sequencing ladder generated from C/G-containing DNA labeled at the 3' end of the upper strand. We conclude that the all-type enzyme cleaves 5' to the mispaired adenine. The cleavage product of T/G-specific enzyme ran at the same position of the C"' band (sequencing ladder from C/G-containing DNA fragment), This incision site at the 3' side of mispaired thymine is consistent with the result of Wiebauer and Jiricny (28). The A/G-specific enzyme cleaved at the same site as the T/G-specific enzyme (i.e. at the first phosphodiester bond 3' to the mispaired base, data not shown). Using DNA substrates labeled at the 5' end of the upper strand for A/G-specific nicking enzyme, a band migrating between A"' and G" could be detected on a sequencing gel (Fig. 76). According to the chemistry of the Maxam and Gilbert method, the site generated by the endonuclease was assigned between T"-A"' and probably contains a 3'-hydroxyl group. Thus, the cleavage site was mapped to the first phosphodiester bond 5' to the mispaired adenine. However, for the all-type nicking enzyme, when using DNA substrates labeled at the 5' end of the upper  Fig. 1) was labeled (presented as *) at the 3' end on the upper or lower strand with Klenow fragment of DNA polymerase I and [n-:"P]dCTP or [n-:"P]dATP, respectively. A/G-specific enzyme only nicked on the "A" but not the "G" strand. A nick on the 3' end-labeled upper strand gave a 69-nucleotide fragment, whereas a nick on the lower strand generated a 50-nucleotide band.  G (lanes 1-4) or T/G (lanes 5-6) mismatch were labeled at the 3' end of the upper strand and were incubated with A/G-specific (fraction 32 eluted from DEAE-5PW column, lanes 1 and 2), all-type (fraction 130 eluted from a Bio-Rex 70 column, lanes 3 and 4 ) , or T/G-specific (fraction 28 eluted from a DEAE-5PW column, lanes 5 and 6 ) nicking enzyme. The reactions were carried out in 20 mM Tris-HCI (pH 7.6), 10 p M ZnC12, 1 mM dithiothreitol, 1 mM EDTA, and 2. 9% glycerol (lanes 1, 3, and 5) or containing 80 mM NaCl in addition (lanes 2, 4 and 6 ) . The fraction 32 eluted from DEAE-5PW column was concentrated by Centricon 3 (Amicon) centrifugation. strand, no fragment could be found. The reason for this is not clear yet. Data from Fig. 7 are summarized in Fig. 8.

DISCUSSION
In this paper, we describe the preliminary characterization of two enzyme systems from human HeLa cells that recognize and nick mismatch-containing DNA fragments. The enzyme systems described here may involve more than one protein.
One enzyme system is specific for DNA containing an A/G mismatch, and the other can nick DNA containing one of eight possible base mismatches. These two enzyme systems can be separated by chromatography from the T/G-specific nicking enzyme system, which was shown to consist of a DNA glycosylase and an AP endonuclease (28,29).
Although we currently lack evidence demonstrating that  (28). The two nicking sites of the A/G-specific endonulcease were determined by 3' and 5' end-labeled A/G mismatch-containing DNA. There is no detectable incision a t the "G" strand. The all-type mismatch endonuclease can nick all eight mismatched bases at the first phosphodiester bond 5' to the mispaired base on the upper strand.
the HeLa A/G-specific nicking is mediated by DNA glycosylase-AP endonuclease, the human A/G-specific enzyme is similar to the E. coli MutY DNA glycosylase and AP endonuclease system involved in A/G-specific repair (10)(11)(12)(13). The high specificity to A/G mismatches and nicking to the "A" but not "G" strand are common for both enzyme systems. Both enzymes have no requirement for M$+ or ATP. Our results suggest that higher eukaryotes have A/G-specific repair pathways similar to those identified in bacteria (11,13). As in bacteria, this pathway may be involved in correcting replication errors to prevent C/G-to-A/T transversions. This is another highly conserved DNA mismatch repair pathway. T/G mismatch repair in human cells (28) appears similar to the very short patch repair system of E. coli (6). The nickdirected repair reactions in Drosophila and human cells (25) resemble the methyl-directed system of E. coli and S. typhi-murium and the hex pathway of S. pneumoniae (3)(4)(5) (13), but the T/G mismatch is the weakest substrate for the HeLa all-type repair enzyme. C/C mispair is repaired poorly in the E. coli methyldirected (11,13) and HeLa terminus-directed reactions (25) but is nicked very well by the HeLa all-type repair enzyme. The unique property of the HeLa all-type nicking enzyme is its strand disparity. With respect to the DNA fragment shown in Fig. 1, the enzyme only nicked the upper strand, and no incision on the lower strand could be detected. This strand specificity is not directed by strand breaks or methylation because unmodified linear DNA substrates are used in these experiments. Preliminary data suggest that the neighboring DNA sequences influence the di~parity.~ As far as we are aware, this type of enzyme has not been described in any organisms. One unsolved problem for this enzyme is that no nicking product could be observed by using DNA fragments labeled at the 5' end of the cutting strands. There may be a contamination of a 5'-phosphorylase or a 5'-to 3"exonuclease that degrades the nicking product. After specific nicking at the 5' side of the mismatched base, a 3'-to 5'-exonuclease also may act from this point and degrade the 5"labeled product. Another possibility involves a mismatch-specific exonuclease that acts from the 5' end toward the mismatched site with reaction stopping just before the mispair. Further purification and characterization are needed to address this question.
The human all-type mismatch nicking activity is functionally homologous to the resolvases from bacteriophage T4 (33, 34), yeast (35-37), calf thymus (38), and human (39) in two aspects. Both enzyme systems make an incision (or incisions) near the mismatched site or Holliday junction point, and cleavage occurs in one orientation depending on the neighboring sequences. In some respects, the Holliday junction may be viewed as two heteroduplex DNA molecules, each with one mismatched base pair. Some resolvases are active on heteroduplex loops (37, 40). However, the action of resolvase requires M$+, which is not essential for the human all-type mismatch repair enzyme. The incision sites were also different for both enzyme systems. The human all-type repair enzyme nicks at the first phosphodiester bond 5' to the side of the mispaired base. The nicking sites of most resolvases, except yeast Endo X1 (35) and T7 endonuclease I 3' side of the junction point or loop. While the T/G-specific nicking activity may be involved in repairing deaminated 5-methylcytosines and the A/G-specific nicking may be involved in preventing C/G-to-A/T transversions, the function of human all-type nicking enzyme is not known. It may be involved in the gene conversion during genetic recombination. Reciprocal and unequal mitotic recombination between nonidentical repeated sequences generates heteroduplex DNA. Gene conversion may play a role in sequence homogenization or diversification. Mismatch repair in heteroduplex DNA formed from the V regions or pseudo-V genes of the immunoglobulin genes could generate antibody diversity (42, 43).