The purification of a mismatch-specific thymine-DNA glycosylase from HeLa cells.

G/T mispairs that arise in the DNA of higher eukaryotes as a result of spontaneous hydrolytic deamination of 5-methylcytosine to thymine must be corrected to G/C pairs. We describe here the purification to apparent homogeneity of the enzyme that initiates this repair process by excising the mispaired thymine from the hetero-duplex to generate an apyrimidinic site. The enzymatic activity could be attributed to a 55-kDa polypeptide, which was purified from extracts of HeLa cells by a combination of conventional and DNA-affinity chromatography. The enzyme is a mismatch-specific thymine-DNA N-glycosylase, capable of hydrolyzing the carbon-nitrogen bond between the sugar-phosphate backbone of the DNA and a mispaired thymine. In addition to the G/T, the enzyme can remove thymine also from C/T and T/T mispairs in the order G/T >> C/T > T/T. It has no detectable endonucleolytic activity on apyrimidinic sites and does not catalyze the removal of thymine from A/T pairs or from single-stranded DNA.

be at the molecular level, perturbations of the DNA methylation pattern lead to aberrations in cellular differentiation (see, for example, Jones and Taylor (1980)). In a recent series of experiments involving transgenic mice, Bestor and colleagues (Li et al., 1992) demonstrated that knocking out the DNA cytosine methylase gene leads to lethal defects during embryonal development, thus providing direct evidence for this hypothesis.
The loss of 5-methylcytosine through deamination results in a change in the DNA methylation pattern, which could be detrimental to the cell. In addition, the C+T mutation resulting from the deamination process in the body of a gene could alter the sequence of the encoded protein, which could have potentially devastating results. Thus, for example, the inactivation of the p53 tumor suppressor protein in a large proportion of bladder carcinomas has been attributed to deamination of 5-* The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked "advertisement" in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.
$ To whom all correspondence should be addressed.
methylcytosine (Rideout et al., 1990). In fact, this latter process has long been held responsible for a large proportion of C+T transition mutations, both in prokaryotes (Duncan and Miller 1980) and in higher eukaryotes, where it has been postulated to be the major cause of the conversion of CpG dinucleotides to TpG and CpA (Bird et al., 1979). On the basis of these data, it had been suggested that GPT mispairs associated with the deamination process are not repaired, at least not with high efficiency.
In contrast to these predictions, however, we showed that GPT mismatches are very efficiently corrected to GIC pairs in cultured mammalian cells (Brown and Jiricny, 19871, by a pathway apparently dedicated to this type of lesion and distinct from the general mismatch repair system employed in the correction of biosynthetic errors (Brown and Jiricny, 1988;Modrich, 1991). Later, using nuclear extracts of HeLa cells and synthetic oligonucleotide substrates, we demonstrated that the specific GPT to GIC repair event was initiated by a mismatchspecific thymine-DNA glycosylase (Wiebauer and Jiricny, 1990).
We now describe the purification of this activity, a 55-kDa protein, from HeLa cells.

EXPERIMENTAL PROCEDURES
All the reagents and solvents used in this work were of analytical grade purity. HeLa cells were purchased from Computer Cell Culture Center (Mons, Belgium), Phosphocellulose P11 was from Whatman, the other chromatography matrices, DEAE-Sepharosd Fast Flow, Mono--, and Hihad S-Sepharosd HP from Pharmacia. Streptavidin-derivatized Dynabeads were from Dynal AS., Biotin-16-dUTP from Boehringer Mannheim,[3HldlTP,and the Rainbow prestained protein molecular weight standards were from Amersham Corp. Low molecular weight protein standards were from Bio-Rad. The oligonucleotides were synthesized on an Applied Biosystems model 380B automated synthesizer and purified by polyacrylamide gel electrophoresis. Bovine serum albumin fraction V (BSA)' was from Life Technologies, Inc. Standard molecular biological manipulations (labeling of oligonucleotides, ethanol precipitation of DNA, polyacrylamide al. (1989).
gel electrophoresis, etc.) were carried out as described by Sambrook et Synthesis ofAffinity Mutrk-The G/U affinity matrix was prepared by annealing 14-mer oligonucleotides 5'-GATCCGTCGACCTG-3' and 5'-GATCCAGGTUGACG-3' in annealing buffer (10 nm Tris-HC1, pH 8.0, 10 m M MgCl2) as described (Jiricny et al. 1986). 200 pg of the annealed 14-mer duplexes were allowed to ligate end-to-end overnight at 12 "C in ligation buffer (25 mM Tris-HC1, pH 7.4, 5 m M MgClz, 5 m M dithiothreitol, 0.25 m~ spermidine, 1.25 mM hexamine cobalt chloride, 10 pg/ml BSA), 1 ~M A T P , and 10 p1 of T,-DNAligase (400 units@, New England Biolabs) in a total volume of 200 pl. After ligation, the DNA was recovered by ethanol precipitation and the dried DNA pellet was dissolved in 48 p1 of HzO. The 5'-overhangs were then filled-in with Sequenase version 2.0 (U. S. Biochemical Corp., 2  dATP, 0.15 PM [a-32PldATP (3000 Ci/mmol), and 0.1 m M Biotin-16-dUTP in a total volume of 200 pl for 10 min at 37 "C. Free nucleotides were removed using spin-column centrifugation with a 1-ml syringe filled with Sephadea G50 Superfine (Pharmacia). 2.8 ml of Dynabeads "280 derivatized with streptavidin were pretreated as directed by the manufacturer and preincubated once with HE buffer (25 rn Heped NaOH, pH 7.8, 1 rn EDTA, 1 m M dithiothreitol, 10% glycerol), 0.1 M NaC1, and 1 mg/ml BSA. After washing in the same buffer without BSA, they were incubated with the biotinylated DNA for 1 h at room temperature on a horizontal roller. They were then washed three times with the same buffer without BSA and stored at 4 "C.
Purification of the Thymine-DNA Glycosylase-The whole cell extracts were made from 15 60-g batches (a total of 900 g) of HeLa cells as follows. The frozen cells were quick-thawed in a 37 "C water bath and allowed to swell in three cell volumes of hypotonic buffer (25 ~l l~ Hepes/ NaOH, pH 7.8, 1 m M EDTA, 2 m M dithiothreitol, 1 m M phenylmethylsulfonyl fluoride, 0.5 m M spermidine, and 0.1 m M spermine) at 4 "C for 20 min. They were then homogenized in a glasdglass Dounce homogenizer (Bellco) with 20 strokes of a tightly fitting pestle. Glycerol was added to a final concentration of 20% (v/v), followed by a saturated and neutralized (NH&S04 solution (11 mVlOO ml of extract). The mixture was allowed to stir for 30 min, and the extract was cleared by centrifugation in a Beckman ultracentrifuge, using a Ti-70 rotor at 60,000 rpm for 90 min at 4 "C.
The cleared extract was diluted 1:4 with HE buffer and incubated batchwise with DEAE-Sepharosa Fast Flow (equilibrated in HE buffer containing 0.1 M NaCl; 10-15 mg of proteidml of matrix) for 1 h at 4 "C. The matrix was washed stepwise in a sintered glass funnel with two bed volumes each (collected separately) of HE buffer containing 0.1 and 0.5 M NaCl.
The flow-through and the first wash from the DEAE-Sepharose Fast Flow were directly incubated batchwise for 1 h at 4 "C with phospho-cellu~ose P11 (25-30 mg of protein/ml of matrix), which had been pretreated as directed by the manufacturer and equilibrated with HE buffer containing 0.1 M NaCl. The phosphocellulose was washed stepwise as before with three matrix volumes each (collected in 3 aliquots) of HE buffer containing 0.1, 0.3, and 0.5 M NaCl.
Following the phosphocellulose chromatography step, active fractions (0.3 M/3, 0.5/1, and 0.5 M/2 in Fig. la) equivalent to 120 g of HeLa cells were pooled, diluted 1:2 with HE buffer, and loaded on a HiLoad S-Sepharosd High Performance (HP) 26/10 column (HP-S, Pharmacia) with a peristaltic pump at 4 "C. All following steps were camed out at room temperature. The column was washed with HE buffer containing 0.25 M NaCl, and the proteins were eluted with a two-step gradient, a 150-ml gradient from 0.25 to 0.5 M NaCl followed by an 80-ml gradient from 0.5 to 1 M NaCl.
Active fractions from the HP-S FPLC (28)(29)(30)(31)(32)(33)(34)(35)(36)(37)(38)Fig. l b ) were diluted 1:4 with HE buffer and loaded on a Mono--FPLCB column (1 ml), equilibrated with HE buffer containing 0.1 M NaCl. Proteins were eluted with a 12.5-ml linear gradient from 0.1 to 0.5 M NaCl in HE buffer. Due to the large amount of protein loaded onto the column, the activity eluted as a broad peak. The active fractions were pooled and reloaded onto the same column, eluted with the same buffer gradient.
The active fractions (10-14) from all the Mono-Q columns were pooled, diluted 1:2 with HE buffer in a siliconized 50-ml Falcon tube, and incubated with 1 ml of streptavidin-Dynabeads (prepared as described above), equilibrated with HE buffer containing 0.1 M NaCl for l h at 4 "C. All magnetic beads were concentrated in one 1.7-ml MulTP tube (MulTi-Technology Inc.) using the magnetic particle concentrator for microtubes, supplied by the manufacturer. The beads were washed with three 200-pl volumes each of HE buffer containing 0.1,0.2,0.4, and 0.5 M NaC1, respectively. Active fractions were pooled, diluted 1:4 with HE buffer in a 4-ml siliconized tube, and incubated with 1.5 ml of streptavidin-Dynabeads, pretreated as described above. All the following steps were performed as described for the first affinity chromatography, except that 100-pl wash volumes were used.
Enzymatic Activity Assays-The enzymatic activity was monitored by means of a 'nicking assay," described previously (Wiebauer and Jiricny, 1989). The mismatch-containing 34-mer or 90-mer oligonucleotide duplexes were constructed as described earlier (Wiebauer and Jiricny, 1989;Hughes and Jiricny, 1992) (see also Scheme 1). The substrate duplexes were labeled at the 5'-end of the T oligonucleotide with 32P. The duplexes G/C, used either as markers or as controls, were labeled on the C strand, except in Fig. 6, where either strand was labeled.
The mismatch-containing substrate (40 fmol) was incubated with the chromatography fractions in binding buffer (25 m Hepes, pH 7.8, 0.5 m M EDTA, 0.01 m ZnCl,, 0.5 rn dithiothreitol), in a total volume of25 pl at the temperature and for the times indicated in the respective figure legends. Under these conditions, the excision of the mispaired thymine is in most cases accompanied by a cleavage of the labeled strand of the duplex at the 3'"side of the apyrimidinic ( A P ) site, presumably by a base-catalyzed p-elimination. After the addition of an equal volume of stop buffer (50 m M Tris-HC1, pH 7.5,25 m M EDTA, 2% SDS, 800 pg/ml proteinase K, 7.5 pg of yeast tRNA) and further incubation for 20 min at 37 "C, the DNA was recovered by ethanol precipitation and the sample was resuspended in formamide loading dye. The oligonucleotide fragments were separated by denaturing polyacrylamide gel electrophoresis in 1 x TBE buffer, and the labeled fragments were visualized by autoradiography. In the case of the affinity-purified fractions, only partial cleavage of the T strand was obtained. In these cases, following the ethanol precipitation step, the DNA was resuspended in 4 pl of 0.1 M NaOH and heated at 90 "C for 30 min; an equal volume of urea dye (8 M urea, 0.04% bromphenol blue, 0.04% xylene cyano1 FF in 1 x TBE) was then added, and the samples were loaded on the gels as described.
Band Shift Assay-The band shift assays were carried out essentially as described previously . 40 fmol of the 5'-32Plabeled 34-mer oligonucleotide were incubated with the relevant protein fraction in 1 x binding buffer for 30 min at 37 "C in a final volume of 20 pl, in either the presence (Fig. 2) or absence (Fig. 4b) of 100 ng of unspecific competitor (poly(dI.dC).poly(dI.dC)) (Pharmacia). 5 pl of 20% Ficoll were then added, and 5 pl of the reaction mix were loaded onto a 7% nondenaturing polyacrylamide gel made in TAE buffer (40 m M Tris acetate, pH 7.5, 1 m M EDTA). Electrophoresis was carried out at 10 V/cm for 70 min.
Zkitiated G / T 90-mer Duplex-This substrate consisted of a 90-mer G oligonucleotide annealed with a 90-mer T oligonucleotide (see Scheme l), where the mispaired thymine was labeled to a high specific activity with 3H. It was constructed as described earlier (Wiebauer and Jiricny, 1990). The specific activity of the duplex was 48.7 Ci/mmol. Specific Activity Measurements-In order to measure the specific activity of the enzyme preparations, the active fractions listed in Table  I were incubated in a mix containing 800 fmol of the 90-mer G/I3H1T in binding buffer and 20 pl of whole cell extract, DEAE-Sepharose, HP-S, and Mono-Q FPLC fractions or 4 pl of the 0.4/1 affinity column fraction (see Fig. 4u) in a total volume of 100 pl. 25-pl aliquots were removed after 10 min, 30 min, and 17 h and mixed with an equal volume of stop buffer, and the reaction was terminated by incubation at 37 "C for 20 min. The mixtures were diluted to 300 pl with water and applied directly on a 0.3-ml DEAE-Sepharosa Fast Flow column. The column was washed with 1 ml of water, the flow-through and the wash were combined, 5 ml of Ready SafeTM liquid scintillation mixture (Beckman) were added, and the mixture was counted for 5 min. As the oligonucleotides and free nucleotides remain bound to the ion-exchange column (data not shown), the amount of radioactivity contained in the flowthrough represents the amount of thymine liberated by the glycosylase (see Table I).
Renaturation of the Glycosylase Activity-The proteins were electrophoresed on discontinuous SDS-polyacrylamide gels (5% stacking gel, 10% separating gel) as described by Sambrook et al. (1989). The gel was loaded as follows: lanes 1 and 3, 10 pl of prestained Rainbow protein molecular weight marker (Amersham Corp.); lane 2, 10 pl of fraction 0.4/1 (Fig. 4u); lane 4,8 pl of Rainbow marker (diluted 1:lOO); lane 5 , 2 pl of fraction 0.4/1. After electrophoresis, lane 2 was cut into twelve 0.4-cm slices, and lanes 4 and 5 were silver-stained (Fig. 5a). The proteins were eluted from the gel slices overnight at 37 "C with vigorous shaking in 400 pl of elution buffer (1 x HE buffer, 10 m M NaCl, 2 m M dithiothreitol, 0.1% SDS, 0.1 mg/ml BSA) and were then precipitated with 4 volumes of ice-cold acetone. The dried protein pellet was dissolved in 25 pl of guanidinium buffer (1 x HE buffer, 50 n m NaCl, 2 mM dithiothreitol, 0.1 mgiml BSA, 6 M guanidinium HCl) and denatured for 30 min at room temperature. Renaturation was performed at 4 "C essentially as described by Hager and Burgess (1980); the protein solution was diluted 50-fold (1.25 ml) with the dilution buffer (as above but without guanidinium HCl). The fractions were concentrated to 40 pl in a Centricon 30 microconcentrator (Amicon Corp., prewashed with 0.5 ml of dilution buffer containing 0.5 mgiml BSA). 10 pl of the concentrated fractions were used in the enzyme activity assay as described above.
AP Endonuclease Activity Assays-To test the enzyme preparations for the presence of AP endonuclease activity, the nicking assays were camed out at pH 6.8. Following digestion with proteinase K and ethanol precipitation, the samples were either resuspended in 4 p1 of binding buffer (pH 6.8) or in 4 pl of 0.1 N NaOH and the latter were heated at 90 "C for 30 min. 4 pl of urea dye were then added to both assays, and The 34-mer corresponds to the fragment of the restriction endonucleases HincII, AccI, and SulI. 5'-End labeling of the C strand of the 34-mer G/C duplex, followed by the restriction digest using these three enzymes, generated the size marker shown in Fig. 4c,5b, and 6 (center lane). 5'-End labeling of the G strand of the 34-mer G/C duplex, followed by similar restriction digests, generated the size marker shown in Fig. 6 (right lane).
the samples were loaded on a denaturing polyacrylamide gel made and run in 1 x BBE buffer (90 m M Bis-Tris borate, pH 6.8, 2 m M EDTA).

RESULTS
Purification of the Mismatch-specific Thymine-DNA Glycosylase-During the purification, the enzymatic activity was monitored by the nicking assay. The specific activities of the respective fractions were estimated from the amounts of tritiated thymine liberated from the 90-mer oligonucleotide G/L3H1T (see "Experimental Procedures").
The starting material for the purification were whole cell extracts from 900 g of HeLa cells. These were used because, although the protein was found predominantly in the nuclear fraction in fresh cell extracts, in extracts from commerciallyavailable frozen cells, up to 50% of our activity appeared in the cytoplasmic fraction (data not shown).
The initial step in the purification scheme was a batchwise "filtration" chromatography of the extract on an anion-exchanger (DEAE-Sepharose Fast Flow), which removed approximately one half of the total proteins and most of the nucleic acids ( Table I). As can be seen in Fig. la, the GPT-processing enzyme was found in the flow-through and in the 0.1 M NaCl washes. Cation-exchange chromatography of the pooled active fractions on Phosphocellulose P11 afforded a 4.7-fold enrichment of the specific activity (Table I), with the enzyme eluting with 0.3-0.5 M salt (Fig. la). The following step, FPLC on a n HP-S-Sepharose column did not bring about a dramatic increase in specific activity (Table I) but offered other significant advantages. It removed most of the contaminating 3' + 5' exonucleases, the presence of which in the active phosphocellulose fractions could be witnessed by the apparent degradation of the cleaved product to shorter oligonucleotides (cf. Fig. 1, a  and b). Furthermore, it reduced the volume of the fractions by an order of magnitude, enabling us to use a small (1-ml) Mono-Q FPLC column. This step yielded the Gfl"processing activity in a very concentrated form in a total volume of 15 ml. The elution profiles of the HP-S and Mono-Q FPLC columns are shown in Fig. 1 ( b and c, respectively). The protein profiles of the activity-containing fractions are shown in Fig. 3.
During numerous attempts to purify the thymine-DNA glycosylase by conventional chromatography, we consistently found a number of polypeptides of various sizes co-purifying 3.0 with our activity. In particular, we tried to ensure that we could separate our activity from uracil-DNA glycosylase, which is present in the HeLa extracts in considerable amounts. For this reason, we routinely monitored both the thymine-and the uracil-DNA glycosylases by a nicking assay, using Gfl' and G/U duplexes, as well as a single-stranded uracil-containing oligonucleotide (data not shown). The results of these assays indicated that the two activities co-purified throughout the various purification procedures. However, we noted that the thymine-DNA glycosylase appeared to be a significantly slower acting enzyme than the uracil glycosylase, as witnessed by the long incubation times needed to visualize enzymatic activity in the nicking assays. We postulated that our enzyme may thus form proteinDNA complexes that are sufficiently long-lived to enable us to purify the protein by DNA affinity chromatography. To our knowledge, this approach has to date not been attempted in enzyme purification, due to the high turnover rates of most enzymes on DNA and to the inherently transient nature protein that could form relatively stable complexes with GPT but not APT or G/C heteroduplexes. The same fractions also gave a band-shift with a G/U duplex (Fig. 2). We discounted the possibility that this latter complex was due to uracil-DNA glycosylase, as no band-shift could be seen with the same G N oligonucleotide incubated with a bacterial uracil-DNA glycosylase under identical experimental conditions (data not shown).2 As the G/U-containing complex had similar electrophoretic mobility as that containing the G/T oligonucleotide (Fig. 2), we concluded that it was formed between the thymine-DNA glycosylase and the GPT or the G/U duplex, respectively.
Due to the fact that the binding to the G N substrate appeared noticeably stronger, it was this latter substrate that we decided to employ as our affinity matrix. Following two rounds of DNA affinity purification, only a single protein band was visible by silver staining on a SDS-polyacrylamide gel. It migrated with an apparent molecular mass of 55 kDa (Fig. 3, lanes A f l and  A f 2 ) . The 55-kDa Protein Is a Thymine-DNA Glycosylase-In order to ensure that the protein band eluting from the affinity matrix with 0.4-0.5 M NaCl represented the thymine-DNA glycosylase, we camed out a series of experiments shown in Fig. 4. In panel a we show the protein band pattern eluting from the affinity * No active purified HeLa uracil-DNA glycosylase was available for the control reaction.   ; 0.2/1, 0 . 4 / 1 , and 0.5/1, first washes ofthe affinity matrix with HE buffer containing 0.2. 0.4. and 0.5 M NaCl, respectively; 0.5 pl of the above fractions were incubated with 40 fmol of a 34-mer G/r oligonucleotide at 37 "C for 30 rnin in a total volume of 20 pl. 4 pl were then mixed with 1 pl of 20% Ficoll and loaded on a native 7% polyacrylamide gel. the autoradiogram of which is shown in h. The remainder was treated with NaOH as described under 'Experimental Procedures." and the oligonucleotide products were separated on a 20% denaturing polyacrylamide gel, shown in c. M , marker oligonucleotide G/C, labeled at the 5'-end of the C strand with n2P and digested with HincII, AccI, and Sal1 (see Scheme 1).
Additional evidence as to the identity of the GPT mismatch processing activity and the 55-kDa protein band comes from an elutiodrenaturation experiment shown in Fig. 5, which shows that the nicking activity and the 55-kDa protein band co-migrate in a denaturing SDS-polyacrylamide gel. These experiments thus provide convincing evidence that the 55-kDa protein represents the thymine-DNA glycosylase.
Substrate Specificity of the Thymine-DNA Glycosylase-As shown above (see also Wiebauer and Jiricny (199011, the enzyme catalyzes the removal of a mispaired thymine from GPT mismatches. We wanted to test whether its activity was restricted solely to this mispair or whether the enzyme was able to act also on other thymine-containing mismatches. To this end we constructed 34-mer heteroduplexes containing G/C, GPT, APT, TPT, and CPT base pairs and incubated them with the purified glycosylase preparation (fraction 0.4/1, Fig. 4u). As can be seen from Fig. 6, no processing of the Watson-Crick (G/C and APT) duplexes was observed, but the thymine-containing heteroduplexes were seen to be nicked at the site of the mispair. with the nicking efficiency decreasing in the order GPT >> CPT > TPT. It is unlikely that the latter two mispairs represent true substrates in vivo. Rather, it would seem more probable that under the conditions of the assay, i.e. where only the purified protein and the mispaired oligonucleotide are present, even a small structural deviation from a Watson-Crick base pair is sufficient for recognition by the enzyme.
The Thymine-DNA Glycosylase Lacks an Associated AP Endonuclease Activity-Our initial studies with HeLa nuclear extracts (Wiebauer and Jiricny, 1989) suggested that following the action of the thymine-DNA glycosylase, the baseless sugarphosphate residue was removed by a two-step excision process, which first "nicked" the DNA 3'-from the AP site by p-elimination and then removed the baseless sugar-phosphate by 3'-.5' exonucleolysis. As this mechanism of AP site processing is normally associated solely with AP lyases, i.e. enzymes possessing both a glycosylase and an AP endonuclease activity (see Weiss and Grossman (1987) for review), we decided to test whether this latter function was also associated with the purified thymine-DNA glycosylase preparations. Incubation of the GPT oligonucleotide with an active fraction, followed by separation of the fragments by a conventional denaturing polyacrylamide gel electrophoresis in 1 x TBE buffer, always resulted in the "nicking" of the substrate. This would suggest that protein fractions contained, in addition to the glycosylase, an AP endonuclease.
However, as bacterial uracil-DNA glycosylase, which is known to possess no AP endonuclease activity, could also be seen to produce a nick in a similar, GN-containing substrate (data not shown), we suspected that the cleavage of the sugar-phosphate backbone at the apyrimidinic site was an artifact of our analytical system. We thus tested the possibility that the observed p-elimination reaction was catalyzed by the conditions of the assay: high pH (8.3) and high temperatures (heating of sample in loading dye for 5 min at 95 "C). We therefore repeated the nicking assay experiments at a pH below 7, i.e. under conditions that do not favor j3-elimination. Indeed, as shown in Fig.   7 (lanes -NaOH). the labeled T strand of the GPT duplex remained mostly intact following incubation with the respective protein fractions and electrophoresis. The same substrates could be shown to contain AP sites by treatment with 0.1 N NaOH (Fig. 7, lanes +NaOH), which leads to a quantitative double p-elimination (2'-3' and 4'-5') reaction that first cleaves the DNA 3' from the depyrimidinated nucleotide and then removes the baseless sugar to leave a phosphate at the 3'-end of the labeled fragment (Maxam and Gilbert, 1977). This experiment therefore demonstrates that the purified mismatch-specific thymine-DNA glycosylase contains no intrinsic AP endonuclease activity.

DISCUSSION
In nuclear extracts from HeLa cells, a GPT mispair incorporated in a synthetic oligonucleotide duplex is addressed initially by a mismatch-specific thymine-DNA glycosylase, which excises the mispaired thymine to generate an apyrimidinic site opposite the guanine (Wiebauer and Jiricny, 1990). The exact mechanism of processing of this AP site was not clear. Our initial in vitro studies (Wiebauer and Jiricny, 1989) suggested that the DNA backbone was first cleaved at it. 9 3'-side by a process of p-elimination, following which a 3' -* 5' exonucleolytic step generated a single nucleotide gap, which was, in turn. filled in with a dCMP residue by polymerase+ to yield a G/C base pair. The remaining nick was then sealed by a DNA ligase (Wiebauer and Jiricny, 1990). Our present findings suggest that the kelimination at the AP site was an artifact of our assay system or that it may have been catalyzed by an unknown basic factods) present in the extracts, as the purified Procedures") were assayed by incubation with the 34-mer Gm oligonucleotide for 7 h at 37 "C. c, 34-mer Gm oligonucleotide incubated with the first affinity fraction prior to electrophoresis on an SDS-polyacrylamide gel; M, marker oligonucleotide GIC (labeled at the 5'-end of the C strand with :'*P) and digested with HincII, AccI, and SalI. thymine-DNA glycosylase does not possess any detectable endonucleolytic activity (Fig. 7).
The G R repair process thus closely resembles the classical base-excision repair pathway (Dianov et al., 1992) (see Weiss and Grossman (1987) and Lindahl (1993) for reviews). The concept of a glycosylase excising an unmodified DNA base was initially rather surprising, given that all DNA glycosylases characterized until recently were restricted in their substrate specificity to modified or damaged DNA bases (Lindahl, 1982;Sancar and Sancar, 1988). It now appears, however, that mismatch-specific DNA glycosylases may be limited neither to the Gfl' mispair nor to mammals. Other organisms have also evolved enzymes capable of acting independently of the repli-RG. 6. The glycosylase catalyzes the removal of thymine from all thymine-containing mispaire. 40 fmol of each 34-mer oligonucleotide duplex ( G / C * , GIT', A/T*. TIT'. CIT', T*lG. T'IT. and T * / C ) were incubated for 30 min a t 37 "C with 1 p1 of the active affinity fraction 0.411 (Fig. 4 u ) as described under 'Experimental Procedures." The asterisk indicates the "P-labeled strand. M, marker oligonucleotide GIC, labeled as denoted by asterisk and digested with HincII. AccI. and SafI. The figure shows an autoradiogram of a denaturing 20% polyacrylamide gel. The band migrating at the height of the Sal1 or AccI marker in the GIT* and T*IG lanes. respectively. representn a contaminating 33-mer in the oligonucleotide preparation, visible due to the gross overexposure of the autoradiogram.
* " +cleavage +products RG. 7. The thymineDNA glycoeylane 7 no intrinsic A F ' endonuclease activity. 40 fmol of the 90-mer ohgonucleotide Gm (labeled at the 5'-end of the T strand with :v2P) were incubated for 1 h at 37 "C at pH 6.8 with the active fractions from the Mono-(2 1.9 pg of protein) and DNAaffinity (0.5 pg of protein) ntages of purification. With no NaOH treatment, only a small amount of cleavage of the labeled strand was observed (fanes " Z O H ) . which is due to the spontaneous &elimination a t AP sites. Treatment with 0.1 x NaOH resulted in the quantitative cleavage at these sites (lanes +NaOH). (3 and Af. active fractions from the Mono-Q FPLC and the first DNA-affinity chromatography purification stage. (The latter (+NaOH) reaction product in shorter than the former (-NUOH) oligonucleotide by 1 baseless sugar residue and has an additional charge on the 3'-terminal phosphate. I t therefore migrates faster in polyacrylamide gels.) cation-associated mismatch repair machinery. Thus the MutY protein of Escherichia coli was shown to catalyze the excision of adenine from G/A and, to a lesser extent, A/C mispairs (Au et af., 1989;%ai-Wu et al., 1992), and a similar activity was also described in mammalian cells Weh et af.. 1991). Clearly. such glycosylases must have different requirements for substrate recognition, in that only bases in a mispair may be removed. This is a necessary constraint on the enzyme, which prevenb the loss of natural bases from Watson-Crick base pairs or single-stranded DNA. We proposed (Wiebauer and Jiricny, 1989) that the biological role of the thymine glycosylase is the correction of G/T mispairs arising from hydrolytic deamination of 5-methylcytosine. In the light of the above data, the enzyme apparently satisfies all the criteria required for this function, with the possible exception of one; SV40 transfection experiments (Brown and Jiricny, 1987) showed that it lacked specificity for GPT mismatches in the context of CpG dinucleotides, the sites of mammalian cytosine methylation. The thymine-DNA glycosylase could potentially address also GPT mispairs arising as biosynthetic errors and might thus interfere with the proper functioning of the replication-associated mismatch correction process by ignoring the strand bias required in the repair of biosynthetic errors (Modrich, 1991). However, a recent report by Ullah and Day (1993) suggests that the enzyme may have a preference for CpG-associated mispairs in uitro, which would further substantiate our hypothesis that its biological role is the correction of deamination-associated G/T mispairs.
One should not, however, discount the possibility of a competition between the replication-associated and the thymine glycosylase-mediated mismatch repair processes. Transfection experiments with mismatch-carrying SV40 heteroduplexes clearly showed that, unlike all other mismatches, the G/T mispairs were predominantly repaired by the latter pathway, but that a small fraction of the transfected G/T heteroduplexes (approximately 8%) appeared to have been addressed by a different mismatch correction system, which lacked the G/T + G/C directionality (Brown and Jiricny, 1987). We postulated (Jiricny, 1991) that these molecules may have contained random nicks in the circular DNA and were therefore addressed by the nick-directed long-patch mismatch repair pathway (Holmes et al., 1990;Thomas et al., 1991). These data would imply that these two mismatch correction systems can indeed compete for the same substrate. It could be argued, however, that both the SV40 transfection experiments and the in uitro mismatch correction assays represent artificial systems that may or may not mirror the situation in uiuo. We must not discount the possibility, for example, that in a cell the two pathways are temporally compartmentalized, in that they may not act in the same stage of the cell cycle. The recently reported isolation of a mutator cell line lacking a GPT-binding protein (Branch et al., 19931, which is presumably a component of the replicationassociated mismatch repair process (Hughes and Jiricny, 1992) ought to help us provide the answer to these questions.