Excision of 5-Halogenated Uracils by Human Thymine DNA Glycosylase

Thymine DNA glycosylase (TDG) excises thymine from G·T mispairs and removes a variety of damaged bases (X) with a preference for lesions in a CpG·X context. We recently reported that human TDG rapidly excises 5-halogenated uracils, exhibiting much greater activity for CpG·FU, CpG·ClU, and CpG·BrU than for CpG·T. Here we examine the effects of altering the CpG context on the excision activity for U, T, FU, ClU, and BrU. We show that the maximal activity (kmax) for G·X substrates depends significantly on the 5′ base pair. For example, kmax decreases by 6-, 11-, and 82-fold for TpG·ClU, GpG·ClU, and ApG·ClU, respectively, as compared with CpG·ClU. For the other G·X substrates, the 5′-neighbor effects have a similar trend but vary in magnitude. The activity for G·FU, G·ClU, and G·BrU, with any 5′-flanking pair, meets and in most cases significantly exceeds the CpG·T activity. Strikingly, human TDG activity is reduced 102.3–104.3-fold for A·X relative to G·X pairs and reduced further for A·X pairs with a 5′ pair other than C·G. The effect of altering the 5′ pair and/or the opposing base (G·X versus A·X) is greater for substrates that are larger (bromodeoxyuridine, dT) or have a more stable N-glycosidic bond (such as dT). The largest CpG context effects are observed for the excision of thymine. The potential role played by human TDG in the cytotoxic effects of ClU and BrU incorporation into DNA, which can occur under inflammatory conditions and in the cytotoxicity of FU, a widely used anticancer agent, are discussed.

The nucleobases in DNA are subject to continuous chemical modification, generating a broad range of mutagenic and cytotoxic lesions that can lead to cancer and other diseases (1,2). To counteract this inevitable damage, the cellular machinery includes systems for DNA repair (3). Damage occurring to the nucleobases is the purview of base excision repair, a pathway that is initiated by a damage-specific DNA glycosylase. These enzymes find damaged or mismatched bases within the vast expanse of normal DNA and catalyze the cleavage of the base-sugar (N-glycosidic) bond, producing an abasic or apurinic/ apyrimidinic (AP) 2 site in the DNA. The repair process is continued by follow-on base excision repair enzymes.
Human thymine DNA glycosylase (hTDG) was discovered as an enzyme that removes thymine from G⅐T and uracil from G⅐U mispairs in DNA (4,5). In vertebrates, G⅐T mispairs arise from replication errors, which are handled by the mismatch repair pathway or from the deamination of 5-methylcytosine to T (6,7). Because cytosine methylation occurs at CpG dinucleotides (8,9), G⅐T mispairs caused by 5-methylcytosine deamination are found at CpG sites. It has been shown that hTDG is most active for G⅐T mispairs with a 5Ј C⅐G pair, suggesting that a predominant biological role of the enzyme is to initiate the repair of CpG⅐T lesions (10,11). DNA methylation at CpG plays a fundamental role in many cellular processes, including transcriptional regulation and the silencing of repetitive genetic elements (8,9). Suggesting a biological imperative to maintain the integrity of CpG sites, another human DNA glycosylase exhibits specificity for G⅐T mispairs at CpG sites; methyl binding domain IV (MBD4) (12)(13)(14)(15).
In addition to its CpG⅐T activity, hTDG has been shown to remove a variety of damaged bases (5, 16 -19), most of which are shown in Fig. 1. We recently identified several new hTDG substrates (20), including 5-chlorouracil (ClU), 5-iodouracil (IU), 5-flourocytosine (FC), and 5-bromocytosine (BrC) (the activity is weak for IU, FC, and BrC and is probably not biologically relevant). The ability of hTDG to remove a broad range of damaged nucleobases is consistent in its relatively large and nonspecific active site (21,22). Yet despite its substrate promiscuity, hTDG exhibits exceedingly weak activity for the excision of cytosine and 5-methylcytosine (11,16,20). We recently showed that for a broad range of C5-substituted uracil and cytosine bases, hTDG specificity depends on substrate reactivity (i.e. the stability of the scissile C-N bond) rather than the selective recognition of substrates in the active site (20). Moreover, we showed that specificity against the excision of cytosine from the huge excess of normal G⅐C pairs in DNA is largely explained by the very low reactivity of dC rather than the inability of hTDG to flip cytosine into its active site (20). Consistent * This work was supported by National Institutes of Health Grant R01-GM72711 (to A. C. D.) and the University of Maryland Marlene and Stewart Greenebaum Cancer Center. 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.
with this catalytic mechanism and the enhanced reactivity of 5-halogenated dU substrates, we found that hTDG rapidly excises FU, ClU, and BrU from CpG sites (20). Indeed, compared with CpG⅐T, the activity is 920-fold greater for CpG⅐FU, 550-fold greater for CpG⅐ClU, and 53-fold greater for CpG⅐BrU (20).
The robust activity observed for CpG⅐ClU and CpG⅐BrU suggests that hTDG may also have significant activity for removing ClU and BrU from DNA contexts other than CpG, raising the possibility that hTDG could play a role in the mutagenic and cytotoxic effects associated with ClU and BrU incorporation into DNA (23,24). These lesions can arise in DNA when the 5-chloro-dUTP or 5-bromo-dUTP pools become elevated, which can be promoted by the activity of peroxidases during inflammation (25,26). The very strong hTDG activity for CpG⅐FU substrates is also of interest because FU has been used for decades to treat many types of cancer (27). The mechanism of FU cytotoxicity is thought to involve multiple pathways, including a repetitive cycle of U and FU incorporation into DNA followed by the excision of these bases by a DNA glycosylase, increasing the burden abasic sites and leading to DNA strand breaks (27). Thus, hTDG could potentially be involved in the cytotoxicity of FU, as suggested by a report that inactivation of TDG in fission yeast and in mouse embryonic fibroblasts diminishes the sensitivity of these cells to FU treatment (28).
To further examine these possibilities, it is important to determine the activity of hTDG for removing FU, ClU, and BrU from DNA contexts other than CpG, because the incorporation of these bases into DNA or their presence in the template strand can be expected to give predominantly A⅐X pairs but also some G⅐X pairs and with no significant preference for a CpG sequence context. Although previous studies have examined the effect of altering the 5Ј-flanking pair on hTDG activity for G⅐T and G⅐⑀C (10,29), quantitative studies have not been reported for the many other hTDG substrates. Moreover, previous studies (and our findings here) indicate that the effect of the 5Ј-flanking pair depends strongly on the nature of the target base (29), so the results for G⅐T substrates do not necessarily predict the 5Ј-neighbor effects for other substrates. In addition, the effect of pairing the target base with adenine rather than guanine (i.e. A⅐X versus G⅐X) has not been rigorously examined for substrates other than U (10,30), and the effect of altering the 5Ј-flanking pair for A⅐X substrates is completely unexplored.
Here, we use single turnover kinetics experiments to compare the activity of hTDG (k max ) for substrates that contain a G⅐X lesion with various 5Ј-flanking base pairs, i.e. CpG⅐X, TpG⅐X, GpG⅐X, and ApG⅐X, where X represents FU, ClU, BrU, U, or T. We also examine the effect of pairing the target base with adenine rather than guanine (i.e. CpA⅐X versus CpG⅐X). Finally, we examine the combined effect of pairing the target base with adenine and altering the 5Ј-flanking base pair using CpA⅐X, TpA⅐X, GpA⅐X, and ApA⅐X substrates. These studies provide the relative activity of hTDG for the excision of U, FU, ClU, and BrU from DNA contexts other than CpG, i.e. those in which they might be expected to arise in vivo. In addition, by systematically altering the CpG context for a series of target bases, our findings illuminate the catalytic role of the putative interactions that hTDG forms with the opposing guanine and with the 5Ј C⅐G base pair.

EXPERIMENTAL PROCEDURES
DNA Synthesis and Purification-Duplex DNA substrates were hybridized in 10 mM Tris, pH 8.0, 0.1 M NaCl, and 0.1 mM EDTA by rapid heating to 80°C and slow cooling to room temperature. Single-strand DNA oligonucleotides were synthesized at the Biopolymer Genomics Core Facility, University of Maryland, Baltimore and at the Keck Foundation Biotechnology Resource Laboratory of Yale University. The 5-chlorodeoxyuridine phosphoramidite was obtained from ChemGenes Corp. (Wilmington, MA). Oligonucleotides were purified by anion exchange HPLC using a Zorbax Oligo column (Agilent Technologies), desalted by gel filtration using pre-packed Sephadex G25 columns (GE Healthcare), and stored at Ϫ20°C. Oligonucleotide purity was verified by analytical anion-exchange HPLC under denaturing (pH 12) conditions using a DNAPac PA200 column (Dionex Corp.), as described previously (20). Oligonucleotides were quantified by absorbance (260 nm) using the pairwise extinction coefficients, calculated as described (31).
Expression and Purification of hTDG-Escherichia coli BL21(DE3) cells (Stratagene) were transformed with a pET-28based expression plasmid for human TDG, 410-amino acids (32). Expression cells were grown in Luria broth (typically 2 liter) at 37°C until A 600 ϭ 0.8, the temperature was reduced to 15°C, and hTDG expression was induced with 0.25 mM isopropyl ␤-D-thiogalactoside and continued for about 15 h. Cells were harvested by centrifugation and suspended in ϳ25 ml of lysis buffer (0.05 M sodium phosphate, pH 8.0, 0.3 M NaCl, 0.02 M imidazole, 0.01 M ␤-mercaptoethanol) with 1 mg/ml lysozyme and a protease inhibitor mixture (Roche Applied Science). The cell suspension was frozen on dry ice, thawed, and incubated on ice for 30 min with stirring followed by an addi-tional 30 min with DNase (Novagen). The lysate was cleared by centrifugation and incubated with 4 ml of nickel-nitrilotriacetic acid metal affinity resin (Qiagen) for 1 h at 4°C. The lysate-resin mix was placed in a gravity-flow column, washed with 30 ml of lysis buffer containing 1 M NaCl and 20 mM imidazole followed by 30 ml of lysis buffer containing 20 mM imidazole, and hTDG was eluted with lysis buffer containing 150 mM imidazole. hTDG was purified further using an SP-Sepharose HP column (GE Healthcare) with buffers IE-A (25 mM Tris, pH 7.5, 75 mM NaCl, 1 mM dithiothreitol, 0.2 mM EDTA, 1% glycerol) and IE-B (IE-A with 1 M NaCl) and a gradient of 5-20% IE-B over 60 min at 2.5 ml/min. hTDG was further purified using a Q-Sepharose HP column (GE Healthcare) with the same IE-A and IE-B buffers and a gradient of 0 -100% IE-B over 60 min at 2.5 ml/min. The purity of hTDG was Ͼ99% as judged by SDS-PAGE stained with Coomassie. Purified hTDG was dialyzed overnight versus storage buffer (20 mM HEPES 7.5, 0.1 M NaCl, 1 mM dithiothreitol, 0.5 mM EDTA, 1% glycerol), concentrated to about 0.1 mM, flash-frozen in small aliquots, and stored at Ϫ80°C. The concentration of hTDG was determined by absorbance using ⑀ 280 ϭ 31.5 mM Ϫ1 cm Ϫ1 (33).
Single Turnover Kinetics-Because hTDG is strongly inhibited by its abasic DNA product (10), the rate constant obtained from steady-state kinetics experiments (k cat ) is dominated by product release and is not useful for comparing hTDG activity for various substrates. Here, we use single turnover kinetics under saturating enzyme conditions to obtain rate constants (k max ) that are not impacted by product release or the association of enzyme and substrate and thereby reflect the maximal enzymatic activity for a given substrate. To ensure that the observed rate constants represent the maximal value (i.e. k obs Ϸ k max ), single turnover experiments were collected using a large excess of enzyme over substrate and with an enzyme concentration that is more than 100-fold higher the K D ϭ 41 nM reported as the apparent binding affinity of hTDG for DNA containing a CpG⅐T mispair (29). To confirm saturating enzyme conditions, experiments were in some cases conducted with two or more hTDG concentrations, typically 5 and 10 M, providing rate constants that were equivalent within experimental uncertainty. Substrate DNA concentrations were 500 nM unless noted otherwise. Single turnover reactions were performed either manually or using a rapid chemical quenchedflow instrument (RQF-3, Kintek Corp.). The reactions were conducted at 22°C in HEMN.1 buffer (20 mM HEPES, pH 7.50, 0.2 mM EDTA, 2.5 mM MgCl 2 , 0.1 M NaCl) with 0.1 mg/ml bovine serum albumin, quenched with 50% (v:v) 0.3 M NaOH, 0.03 M EDTA, and heated for 15 min at 85°C to induce cleavage of the DNA backbone at AP sites. The extent of product formation was analyzed by HPLC, as described below. Rate constants were determined by fitting the single turnover data to a single exponential equation using nonlinear regression with Grafit 5 (34). In most cases, the reactions proceeded to full completion, except those that are very slow (i.e. k max Ͻ ϳ1 ϫ 10 Ϫ4 min Ϫ1 ).
HPLC Assay for Monitoring hTDG Activity-We recently developed a HPLC assay for monitoring hTDG activity (20). Samples taken during a kinetics experiment contain a mixture of substrate and products that is comprised of four oligonucleotides; that is, the full-length target strand and its complement and two shorter strands resulting from alkaline cleavage of the nascent abasic strand. These strands are resolved by anion exchange HPLC using denaturing (pH 12.0) conditions with a DNAPac PA200 column (Dionex Corp.). The alkaline conditions serve to suppress hybridization of ssDNA during chromatography and have the added benefit of resolving strands that are of the same length but differ in the number of thymine and guanine bases, which are negatively charged at pH 12. The elution buffer is 0.02 M sodium phosphate pH 12.0 containing either 0.03 M NaClO 4 (buffer A) or 0.50 M NaClO 4 (buffer B). The oligonucleotides are detected by absorbance (260 nm), and the fraction product (F) is determined from the integrated peak areas for the target strand (A S ) and product strands (A P1 and A P2 ) using the equation F ϭ (A P1 ϩ A P2 )/(A S ϩ A P1 ϩ A P2 ). The determination of fraction product using this assay is reproducible to within 1%, as determined from multiple analyses of identical samples.

RESULTS
In a recent study we determined the activity of hTDG (k max ) for a series of 5-substituted uracil and cytosine substrates in which the target base was placed in a CpG context (20). We found that k max is much higher for CpG⅐FU, CpG⅐ClU, and CpG⅐BrU than for CpG⅐T, suggesting that hTDG may have significant activity for FU, ClU, and BrU in DNA contexts other than CpG. Here, we examine the effect of altering the CpG context on the activity of hTDG for the excision of five different target bases (X ϭ T, U, FU, ClU, and BrU) using a series of substrates as shown in Fig. 2. Thus, one set of substrates examines the effect of altering the 5Ј-flanking pair on G⅐X activity (i.e. YpG⅐X). Another set examines the effect of pairing the target base with adenine rather than guanine while preserving the 5Ј C⅐G pair (CpA⅐X versus CpG⅐X). A final set examines the effect of both, pairing the target base with adenine and altering the 5Ј-flanking pair (YpA⅐X).
Single Turnover Kinetics-Like many DNA glycosylases, hTDG is strongly inhibited by one of its reaction products, aba- sic DNA (10,14,(35)(36)(37). Indeed, previous studies show that under limiting enzyme conditions, the turnover of hTDG is exceedingly slow after it converts one molar equivalent of G⅐T (or G⅐U) substrate to G⅐AP product (10,38,39). Thus, the rate constant obtained from steady-state kinetics, k cat , is dominated by product release and cannot provide a meaningful comparison of activity for different substrates (Fig. 3). In contrast, single turnover kinetics conducted under saturating enzyme conditions provide a rate constant (k max ) that is not impacted by product release or the association of enzyme and substrate and, therefore, reflects the maximal activity for a given substrate (Fig. 3). For the hTDG reaction, k max reflects the rate constant for the chemical step (k chem ) and is also influenced by the equilibrium constant for base flipping (K flip ). In the base-flipping step, the target nucleotide flips out of the DNA duplex and into the active site, a process that likely involves a conformational change in hTDG, as observed for uracil DNA glycosylase (40). Thus, differences in k max that result from alterations to the CpG context reflect a change in k chem and/or K flip .
Effect of the 5Ј Base Pair on G⅐X Activity-We determined the effect of varying the 5Ј neighboring base pair on hTDG activity (k max ) for G⅐FU, G⅐ClU, G⅐BrU, G⅐U, and G⅐T using the YpG⅐X series of substrates (Fig. 2). The results are given in Table 1 and Fig. 4. Previous studies showed that hTDG activity for G⅐T sub- FIGURE 3. Minimal kinetic mechanism for hTDG. Shown is a minimal kinetic mechanism for hTDG, including the steps that comprise k cat , which is obtained from steady-state kinetics, and k max , as obtained from single turnover kinetics with a saturating enzyme concentration (used here). Association of hTDG (E) and DNA substrate (D) forms the initial collision complex, E⅐D, and base-flipping (K flip ) gives the reactive enzyme-substrate complex, E⅐ B D (where B D is base-flipped DNA). Base flipping likely involves a conformational change for hTDG, which is not explicitly shown. The chemical step (k chem ) involves cleavage of the base-sugar (N-glycosidic) bond and the addition of the water nucleophile, producing the ternary product complex, E⅐B⅐apD (B is the nucleobase, apD is abasic DNA). Release of the excised base likely precedes the dissociation of abasic DNA, which is known to be very slow. As shown, k max is influenced by K flip and k chem , where k max ϭ k chem (K flip /1 ϩ K flip ). a The rate constants (k max ) reflect the maximal enzymatic activity of hTDG for a given substrate, as determined using single turnover kinetics experiments with saturating enzyme conditions. b -Fold change relative to YpG⅐X gives the effect of pairing the target base (X) with adenine rather than guanine for a given 5Ј base pair (Y), i.e. the rate of TpA⅐U relative to TpG⅐U. c ND, Not determined. SEPTEMBER 21, 2007 • VOLUME 282 • NUMBER 38

JOURNAL OF BIOLOGICAL CHEMISTRY 27581
strates depends strongly on the 5Ј-flanking pair, with relative activity of CpG⅐T Ͼ Ͼ TpG⅐T Ͼ GpG⅐T Ͼ ApG⅐T (10,11). We find a similar trend here; compared with CpG⅐T, k max is reduced by 37-, 96-, and 582-fold for TpG⅐T, GpG⅐T, and ApG⅐T, respectively. The influence of the 5Ј neighbor is much smaller for G⅐U activity; compared with CpG⅐U, k max is decreased by 3.3-, 2.9-, and 22-fold for TpG⅐U, GpG⅐U, and ApG⅐U, respectively. The 5Ј-neighbor effect is also small for G⅐FU activity; compared with CpG⅐FU, k max decreases by merely 1.8-, 1.6-, and 11-fold for TpG⅐FU, GpG⅐FU, and ApG⅐FU, respectively (Fig. 4B). The 5Ј-neighbor effects are much larger for G⅐ClU activity; compared with CpG⅐ClU, k max is decreased by 6-, 11-, and 82-fold for TpG⅐ClU, GpG⅐ClU, and ApG⅐ClU, respectively (Fig. 4C). The results are similar for G⅐BrU; compared with CpG⅐BrU, k max is decreased by 9-, 26-, and 75-fold for TpG⅐BrU, GpG⅐BrU, and ApG⅐BrU, respectively.
It is important to note that the activity is significantly greater for all of the G⅐FU and G⅐ClU substrates than for CpG⅐T, which is generally considered the most biologically relevant substrate for hTDG (28,29). Indeed, k max is 75-920-fold greater for the G⅐FU substrates and 7-550-fold greater for the G⅐ClU substrates relative to CpG⅐T. In addition, the activity for all of the G⅐U and G⅐BrU substrates exceeds the CpG⅐T activity except for ApG⅐U and ApG⅐BrU, which are merely 2-and 1.4fold slower, respectively.
Activity Is Greatly Diminished for A⅐X Relative to G⅐X Substrates-It has been reported that hTDG does not remove thymine from A⅐T pairs (10,11,41), in keeping with the biological imperative to minimize the activity against undamaged DNA. We, therefore, wondered to what extent the robust hTDG activity observed for G⅐FU, G⅐ClU, and G⅐BrU is diminished when the 5-halouracil bases are paired with adenine rather than guanine. This is relevant because 5-halogenated dU is incorporated opposite template dA and vice versa during replication (24). We examined the effect of pairing the target base with adenine rather than guanine while preserving the 5Ј C⅐G pair using the CpA⅐X substrates (Fig. 2). The data are listed in Table 1 and displayed in Fig. 5. Under the experimental conditions used here, i.e. high concentrations of substrate (500 nM) and enzyme (5 M), we were able to measure the exceedingly weak activity for thymine excision from the CpA⅐T substrate, k max ϭ 1.3 ϫ 10 Ϫ5 min Ϫ1 . This is a striking 17,600-fold lower than the activity for CpG⅐T. In contrast, a much smaller 795fold decrease is observed for CpA⅐U compared with CpG⅐U, consistent with a previously reported 600-fold difference (30). Similarly, a small 187-fold decrease is observed for CpA⅐FU relative to CpG⅐FU. For the excision of ClU, which is similar to T in terms of its steric and electrostatic properties (20), we find a much larger 5460-fold decrease for CpA⅐ClU versus CpG⅐ClU. A significantly larger 14,870-fold decrease is  . hTDG activity is greatly reduced for A⅐X relative to G⅐X substrates. The hTDG activity (log k max ) for CpG⅐X versus CpA⅐X substrates is shown. The change in log k max is given. observed for CpA⅐BrU relative to CpG⅐BrU, approaching the difference in activity for CpA⅐T versus CpG⅐T. Consistent with our results is a previous report that hTDG excision of 5-hydroxymethyluracil (hmU) is 10 4.0 -fold slower for CpA⅐hmU compared with CpG⅐hmU (30). Our results show that for the substrate bases examined here, the effect on k max of pairing the target base with A rather than G is much larger than the effect altering the 5Ј-flanking pair for the G⅐X substrates.
Effect of the 5Ј Base Pair on A⅐X Activity-We also examined the effect of altering the 5Ј-flanking pair on hTDG activity for A⅐X pairs using the YpA⅐X substrates (Fig. 2). The data are listed in Table 1 and shown in Fig. 6. Generally, these effects follow the same trend observed for the 5Ј-neighbor effect on G⅐X activity, where the relative activity is CpA⅐X Ͼ TpA⅐X Ͼ GpA⅐X Ͼ ApA⅐X. We find a significant 5Ј-neighbor effect for A⅐U activity; k max decreases by 4-, 62-, and 104-fold for TpA⅐U, GpA⅐U, and ApA⅐U, respectively, as compared with the CpA⅐U activity. The 5Ј-neighbor effects are larger for A⅐FU pairs; k max decreases by 5-, 78-, and 171-fold for TpA⅐U, GpA⅐U, and ApA⅐U, respectively, relative to CpA⅐FU. The effects are larger still for A⅐ClU; k max decreases by 5-, 257-, and 436-fold for TpA⅐ClU, GpA⅐ClU, and ApA⅐ClU, respectively, compared with CpA⅐ClU. Somewhat smaller effects are observed for A⅐BrU; k max decreases by 5-, 65-, and 103-fold for TpA⅐BrU, GpA⅐BrU, and ApA⅐BrU relative to CpA⅐BrU. We note that of the A⅐X substrates examined, only CpA⅐FU and TpA⅐FU exhibit greater activity than observed for CpG⅐T (Fig. 6).

DISCUSSION
Role of the 5Ј C⅐G Pair on G⅐X Activity-The effect of altering the 5Ј C⅐G pair on G⅐X activity was examined using the YpG⅐X series of substrates (Fig. 2), and the results are given in Table 1 and Fig. 4A. Our findings indicate that the magnitude of the 5Ј-neighbor effect depends on the size of the nucleobase and on substrate reactivity, i.e. the stability of the C-N bond that is cleaved by hTDG. For the uracil analogues examined here, size depends on the C5 substituent, with nucleobase volume varying as U Ͻ FU Ͻ Ͻ ClU Ϸ T Ͻ BrU (20). The substrate reactivity (N-glycosidic bond stability) also depends on the C5 substituent. Thus, for the hTDG-catalyzed and non-catalyzed reactions, the rate depends on the leaving ability of the departing nucleobase (20,(42)(43)(44). The electron withdrawing halogens enhance the leaving ability of uracil, whereas the electron donating methyl group suppresses it. Accordingly, substrate reactivity varies as: CldU Ն BrdUrdՆ FdU Ͼ Ͼ dU Ͼ dT (20). As we consider the effects of altering the CpG context, we note that differences in k max reflect a change in the equilibria for the base flipping step (K flip ) and/or the rate of the chemical step (k chem ), as shown in Fig. 3.
The dependence of the 5Ј-neighbor effect on substrate reactivity is illustrated by comparing the effects for G⅐ClU, ranging from 6-to 82-fold, with the effects for G⅐T, which range from 37-to 582-fold. As we have noted previously, the size and electrostatic properties of T and ClU are quite similar (20), suggesting that T and ClU are not substantially discriminated by the relatively large and nonspecific hTDG active site (21,22). In addition, melting studies showed that G⅐T and G⅐ClU base pairs have essentially the same stability in DNA, indicating that ClU and T have a similar propensity for flipping spontaneously out of the duplex (20). This suggests that the steeper dependence of k max on the 5Ј flanking pair for G⅐T relative to G⅐ClU is not due to substantial effects on the base-flipping step. The most significant difference is that CldU is much more reactive than dT, which suggests that altering the 5Ј C⅐G pair elicits a greater effect on k chem for G⅐T than for G⅐ClU pairs. Thus, interactions with the 5Ј C⅐G pair may induce a conformational change in hTDG and/or the DNA substrate that stabilizes the transition state, and this effect appears to be greater for less reactive substrates.
The dependence of the 5Ј-neighbor effect on substrate size is illustrated by comparing the effects for G⅐FU, 2-11-fold, with G⅐BrU, 9 -75-fold. The reactivity of FdU and BrdUrd is about the same as is the stability of G⅐FU and G⅐BrU base pairs in DNA; however, BrU is substantially larger than FU (20). Although the hTDG active site is accommodating, we found previously that k max decreases for substrates with large C5 substituents, including BrU, 5-iodouracil, and 5-bromocytosine (20). Thus, our results suggest that interactions with the 5Ј C⅐G pair serve to increase K flip or stabilize the transition state of the reaction and that the effect is greater for substrates with a larger C5-group (BrdUrd and dT). A previous study found that the apparent K D is nearly the same for CpG⅐T and TpG⅐T substrates, suggesting that recognition of the 5Ј C⅐G pair contributes largely to stabilizing the transition state (increasing k chem ) rather than promoting base flipping (29). However, additional studies are required to establish which steps of the hTDG reaction are affected by recognition of the 5Ј C⅐G pair.
Effect of Pairing the Substrate Base with A Rather than G-Because hTDG excises a normal base, thymine, from G⅐T mispairs, it has an effective mechanism for avoiding the excision of T from the huge excess of A⅐T pairs in DNA (10,11). The specificity against A⅐T may involve H-bond interactions that select for guanine, as observed in a crystal structure of the related mismatch uracil DNA glycosylase from E. coli (eMUG) (45). Here, we have quantitatively established the specificity of hTDG for excising bases from G⅐X versus A⅐X pairs, and we find that it is strikingly large. The difference in k max ranges from 187-fold for CpA⅐FU versus CpG⅐FU to 17,600-fold for CpA⅐T versus CpG⅐T (Table 1, Fig. 5). Our findings indicate that these large differences depend on the size and reactivity (C-N bond stability) of the nucleotide substrate. The much larger effect for BrdUrd(10 4.2 -fold) relative to FdU (10 2.3 -fold) is likely due to the larger size of BrdUrd because these substrates have similar reactivity (20). The larger effect for dU (10 2.9 -fold) compared with FdU is probably attributable to the significantly greater reactivity of FdU, since dU and FdU have relatively similar steric and electrostatic properties (20). Thus, our results suggest that the putative interactions formed with the mismatched guanine serve to promote base-flipping (increase K flip ) and/or stabilize the transition state (increase k chem ) and that disrupting these interactions has a greater effect for larger and less reactive substrates (i.e. dT). Additional mechanistic and/or structural studies are needed to determine how hTDG selects for G⅐X over A⅐X pairs and which step(s) of the reaction is involved.
It also of interest to consider the difference in activity for A⅐X versus G⅐X for substrates in which the 5Ј-flanking pair is not C⅐G. These effects are listed in Table 1 under "Fold change relative to YpG⅐X." For example, considering U excision, the decrease in k max is 10 3.0 -fold for TpA⅐U versus TpG⅐U, 10 4.2fold for GpA⅐U versus GpG⅐U, and 10 3.6 -fold for ApA⅐U versus ApG⅐U as compared with 10 2.9 -fold for CpA⅐U versus CpG⅐U. The effects for the FU substrates have the same trend but are smaller in magnitude, whereas the effects for the ClU and BrU substrates have a similar trend and are larger in magnitude. Overall, the results indicate that the difference in activity for A⅐X versus G⅐X substrates is similar when the 5Ј-flanking pair is C⅐G or T⅐A, becomes larger for a 5Ј A⅐T pair, and is maximal for a 5Ј G⅐C pair.
Effect of the 5Ј Base Pair on A⅐X Activity-The effects of altering the 5Ј C⅐G pair for A⅐X substrates are substantial and are larger than those observed for the G⅐X substrates (Fig. 6, Table  1). For example, the effects are 4 -104-fold for A⅐U activity as compared with 3-22-fold for G⅐U pairs. Likewise, the 5Ј-neighbor effects are greater for A⅐FU versus G⅐FU and for A⅐ClU versus G⅐ClU activity, whereas the effects are nearly equivalent for A⅐BrU versus G⅐BrU activity. Thus, our results do not support a previous suggestion that specificity for CpG⅐T involves cooperativity between the interactions formed with the mismatched guanine and the 5Ј-flanking C⅐G pair (10). If the interactions were cooperative, one would expect that the effects of altering the 5Ј C⅐G pair would be smaller for A⅐X relative to G⅐X substrates because the interactions formed with the mismatched guanine would be disrupted for the A⅐X substrates.
Combined Effect of Pairing the Target Base with Adenine and Altering the 5Ј C⅐G Pair-It is illuminating to consider the effect on k max of replacing the opposing guanine with adenine and altering the 5Ј-C⅐G pair. The combined effects are very large, as shown in the lower region of Table 1 under the heading "Fold change relative to CpG⅐X." For example, compared with the activity for CpG⅐U, k max is reduced by 10 3.5 -, 10 4.7 -, and 10 4.7fold for TpA⅐U, GpA⅐U, and ApA⅐U, respectively. Similar reductions in activity of 10 2.9 -10 4.5 -fold are seen for the FU substrates. The effects are even more striking for the larger bases, ranging from 10 4.4 -to 10 6.4 -fold for ClU and 10 4.8 -to 10 6.2 -fold for BrU. The huge effects observed for ClU and BrU suggest that the specificity of hTDG for CpG⅐T lesions over normal A⅐T pairs ranges in magnitude from 10 4.3 -to 10 6.4 -fold or more, corresponding to 6 -8.7 kcal/mol.
Implications for CpG⅐T Specificity-Taken together our findings indicate that disrupting the interactions that hTDG forms with the 5Ј C⅐G pair and the mismatched guanine have a greater effect for substrates with a large C5-substituent and/or a more stable N-glycosidic bond. Thus, the relatively large size and low reactivity of dT likely explains the large (37-625-fold) 5Ј-neighbor effect on k max and the huge (10 4.3 -fold) difference in activity for CpA⅐T versus CpG⅐T. Indeed, the smaller (3-22fold) 5Ј-neighbor effect for G⅐U activity and the much smaller difference between CpA⅐U and CpG⅐U (10 2.9 -fold) is likely explained by the substantially smaller size and enhanced reactivity of dU relative to dT. These differences in activity would seem to be consistent with the requirement of the cell to restrict thymine excision to thymines that arise by 5-methylcytosine deamination at CpG sites, whereas uracil can be removed wherever it arises in DNA.
Biomedical Relevance of FU Activity-Our finding of very strong hTDG activity for G⅐FU lesions with any 5Ј base pair, some 75-920-fold greater than CpG⅐T activity, may be relevant to the cytotoxicity of FU, which has been used for decades to treat cancer (27). The anticancer effect of FU is thought to involve multiple pathways, including the incorporation of U and FU into DNA followed by their excision by a DNA glycosylase, leading to a cycle of incorporation and excision, an increased level of abasic sites, DNA fragmentation, and cell death (24,27). Consistent with a potential role for TDG in this process, it was reported that TDG inactivation diminishes the sensitivity of fission yeast and mouse embryo fibroblasts to FU treatment and leads to a decrease in FU-induced DNA strand breaks (28). This likely reflects TDG activity against G⅐FU lesions because its A⅐FU activity is relatively weak. Other DNA glycosylases that could potentially elicit a similar effect by removing misincorporated U and/or FU include uracil DNA glycosylase (UNG2), SMUG1, and MBD4. The removal of U probably involves UNG2 and SMUG1 (46) because TDG and MBD4 are much less efficient against U (Table 1) (14). Although UNG2 has the highest activity for U removal, its activity is much lower for FU (47)(48)(49), and UNG2 does not remove FU from DNA in mouse embryo fibroblasts (49). Although hMBD4 has significant activity for CpG⅐FU lesions, the effect of the 5Ј base pair on G⅐FU activity is unknown for this CpG specific enzyme, and it is inactive against A⅐FU pairs (50). Recent studies report that MBD4 inactivation decreases the sensitivity of mice to FU treatment, although this may reflect a loss of an MBD4-mediated apoptotic response to DNA damage rather than FU excision (51).
The toxicity of FU may also depend on its presence in DNA due to mutagenesis or perturbations of protein-DNA interactions as was suggested by a report that FU excision by SMUG1 protects mouse embryo fibroblasts against FU-induced toxicity (49). The potential for mutagenesis is significant; FU incorporation yields mostly A⅐FU pairs but also some G⅐FU pairs, which can cause G⅐C 3 A⅐T mutations (52,53). Thus, repair of G⅐FU lesions for which G is in the parental strand would be protective. However, because G can be incorporated opposite template FU, replication of A⅐FU pairs can give G⅐FU lesions, leading to A⅐T 3 G⅐C mutations (52,53). It is important to note that base excision repair processing of these G⅐FU lesions would facilitate the A⅐T 3 G⅐C transition rather than protect against it. Perhaps the protective role observed for SMUG1 (49) reflects the repair of initial A⅐FU lesions (i.e. FU in the daughter strand). The reported effect of hTDG in enhancing the sensitivity of mouse embryo fibroblasts to FU treatment may reflect hTDG activity against G⅐FU lesions in which FU is in the parental strand or the production of abasic sites as discussed above.
Biological Relevance of BrU and ClU Activity-We find that hTDG exhibits robust activity for G⅐ClU and G⅐BrU pairs, which meets and in most cases exceeds the activity for CpG⅐T (Fig. 4). Our findings may have important biological implications because it has been shown that ClU and BrU arise in DNA due to oxidative processes associated with inflammation (25,26), leading to mutagenic, genotoxic, and cytotoxic effects (23,24). As an element of host defense, peroxidases produce hypochlorous and hypobromous acid, which promote the halogenation of pyrimidines (at C5), leading to 5-chloro-dUTP and 5-bromo-dUTP (25, 26, 54 -58). These dTTP analogues are incorporated into DNA, giving predominantly A⅐ClU and A⅐BrU pairs and some G⅐ClU and G⅐BrU lesions (53,59), which can cause G⅐C 3 A⅐T mutations. The initial A⅐ClU and A⅐BrU pairs are also mutagenic, because incorporation of G opposite template ClU or BrU gives G⅐ClU or G⅐BrU and eventually A⅐T 3 G⅐C transitions (53,59,60). Until recently, it did not appear that any human enzymes could remove ClU or BrU from DNA. It had been shown that UNG2 and SMUG1 do not remove ClU or BrU likely due to steric hindrance in the active site (25,(61)(62)(63). Our findings raise the possibility that hTDG is active against G⅐ClU or G⅐BrU lesions that arise in vivo, and a recent study suggests that hMBD4 may also be active against these lesions (50). However, A⅐ClU and A⅐BrU lesions are poor substrates for hTDG (Table 1) and hMBD4 (50), and these lesions may persist in DNA. The effect of excising ClU or BrU may depend on whether the halogenated base is misincorporated or resides in the template strand. The excision of ClU or BrU that was misincorporated opposite template G could protect against G⅐C 3 A⅐T mutations. On the other hand, repair of G⅐ClU and G⅐BrU lesions that have arisen from replication of A⅐ClU or A⅐BrU (i.e. template ClU or BrU) will facilitate an A⅐T 3 G⅐C transition rather than protect against it. In addition to these potential effects on ClU-and BrU-induced mutagenesis, hTDG could potentially contribute to a repetitive cycle of ClU and BrU misincorporation and excision, leading to abasic sites and DNA strand breaks, similar to the potential role played by hTDG in FU toxicity, as discussed above.
It has been shown that sister-chromatid exchange increases with the amount of CldU and BrdUrd present in replicated DNA (64,65). Although the mechanism of ClUand BrU-induced sister-chromatid exchange has not been established, evidence suggests that one mechanism involves the formation of single-strand breaks that originate from abasic sites (23,24,66). The generally accepted model is that these abasic sites arise from dehalogenation of ClU or BrU followed by excision of the resulting uracil by UNG2 (66). Our findings suggest an alternative mechanism, that the abasic sites originate from the direct excision of ClU and BrU by TDG (and potentially MBD4). It has also been observed that CldU induces 3-5 times more sister-chromatid exchange than BrdUrd when these dT analogs are incorporated at equivalent level in DNA (24,64). Our findings offer a potential explanation; the ϳ10-fold higher activity of hTDG for G⅐ClU over G⅐BrU could generate more abasic sites given a similar level of ClU and BrU incorporation into DNA. Thus, our findings and recent observations for MBD4 raise the possibility that these enzymes might contribute to the mechanism of CldU-and BrdUrd-induced sister-chromatid exchange.