hUNG2 Is the Major Repair Enzyme for Removal of Uracil from U:A Matches, U:G Mismatches, and U in Single-stranded DNA, with hSMUG1 as a Broad Specificity Backup*

hUNG2 and hSMUG1 are the only known glycosylases that may remove uracil from both double- and single-stranded DNA in nuclear chromatin, but their relative contribution to base excision repair remains elusive. The present study demonstrates that both enzymes are strongly stimulated by physiological concentrations of Mg2+ , at which the activity of hUNG2 is 2–3 orders of magnitude higher than of hSMUG1. Moreover, Mg2+ increases the preference of hUNG2 toward uracil in ssDNA nearly 40-fold. APE1 has a strong stimulatory effect on hSMUG1 against dsU, apparently because of enhanced dissociation of hSMUG1 from AP sites in dsDNA. hSMUG1 also has a broader substrate specificity than hUNG2, including 5-hydroxymethyluracil and 3,N 4-ethenocytosine. hUNG2 is excluded from, whereas hSMUG1 accumulates in, nucleoli in living cells. In contrast, only hUNG2 accumulates in replication foci in the S-phase. hUNG2 in nuclear extracts initiates base excision repair of plasmids containing either U:A and U:G in vitro. Moreover, an additional but delayed repair of the U:G plasmid is observed that is not inhibited by neutralizing antibodies against hUNG2 or hSMUG1. We propose a model in which hUNG2 is responsible for both prereplicative removal of deaminated cytosine and postreplicative removal of misincorporated uracil at the replication fork. We also provide evidence that hUNG2 is the major enzyme for removal of deaminated cytosine outside of replication foci, with hSMUG1 acting as a broad specificity backup.

Uracil in DNA can be introduced via two mechanisms, deamination of cytosine and misincorporation of dUMP during replication. Deamination of cytosine has been calculated from measured deamination rates to occur at a rate of 100 -500 per human cell/day (1,2) to yield mutagenic U:G mispairs. Uracil may also appear as a consequence of misincorporation of dUMP instead of dTMP during replication, resulting in a U:A base pair. The latter is not miscoding, but may produce cytotoxic and mutagenic AP site intermediates during repair. In organisms containing 5-methylcytosine in their genomes, deamination of 5-methylcytosine furthermore leads to T:G mismatches. All living organisms express uracil-DNA glycosylases (UDGs) 1 that prevent cytotoxic and mutagenic effects of the above lesions. UDGs remove uracil (and sometimes other damaged bases or thymine) from the deoxyribose and thus initiate a multistep base excision repair (BER) pathway, eventually restoring the correct DNA sequence. After removal of uracil by an UDG and cleavage of the resulting abasic site by AP endonuclease (APE1/APE2), the BER pathway splits into two branches (reviewed in Ref. 3). The presumed major track is the shortpatch pathway. It uses the 5Ј-deoxyribophosphodiesterase activity of DNA polymerase ␤ to cleave 3Ј of the abasic site, thus releasing deoxyribose-5-phosphate. Then pol ␤ inserts C or T, depending on the template base. Finally, DNA ligase III seals the nick, perhaps aided by the scaffold protein XRCC1. The alternative long-patch pathway largely uses replication proteins and may take place in replication foci (4). This pathway requires pol ⑀ and/or ␦, as well as the trimeric sliding clamp and polymerase processivity factor proliferating cell nuclear antigen (PCNA) and the clamp loader replication factor C (RFC). Repair synthesis is stimulated by pol ␤, which may be important in the first step of polymerization. The structure-specific endonuclease FEN1 removes the 2-8-nucleotide displaced "flap" of DNA, and DNA ligase I seals the nick (3).
Mammalian cells contain at least four UDGs, of which three (UNG, SMUG1, and TDG) belong to the same protein superfamily, possess the same fold, and have probably evolved from a common ancestor (5). Of these, UNG appears to be quantitatively dominating as determined from activity assays using human cell-free extracts and U:A substrates (6). UNG belongs to the family of highly conserved UDGs typified by Escherichia coli Ung, and is present in a large number of eukaryotes, bacteria, and large eukaryotic DNA viruses (7). The human and mouse UNG genes encode both mitochondrial (UNG1) and nuclear (UNG2) forms of the enzyme by way of alternative promoter usage and mRNA splicing (8). The catalytic domain of hUNG has been extensively studied, and its structure and molecular mechanism of catalysis and specificity established (9 -12). The enzyme removes uracil in vitro in the order of preference ssU Ͼ U:G Ͼ U:A (6). Certain closely related bases formed from cytosine after ␥-irradiation or oxidative stress are also substrates, such as 5-hydroxyuracil, isodialuric acid, and alloxan (13). These are, however, excised at a very low rate compared with uracil. A second human UDG against both ssand dsU, hSMUG1 (single-strand selective monofunctional uracil-DNA glycosylase), was identified more recently by in vitro expression cloning (14). SMUG1 is not found in bacteria and yeast, but is present in higher eukaryotes. Although they probably share the same fold and motifs necessary for substrate binding and catalysis as hUNG, little homology exists between the two enzymes at the amino acid level (5). hSMUG1 is furthermore located in the nuclei, and Xenopus SMUG1 has a substrate preference similar to the catalytic hUNG domain (ssU Ͼ U:G Ͼ U:A), although its specific activity is considerably lower (14). Recently, Boorstein and co-workers (15) demonstrated that hSMUG1 is also active against 5-hydroxymethyluracil (HmU:A). HmU is formed in DNA by oxidative attack on the methyl group of thymine, thereby creating HmU:A. It is also the product of the deamination of 5-hydroxymethylcytosine, which may be formed via oxidation of 5-methylcytosine. The latter creates a HmU:G base pair, which would be mutagenic if left unrepaired. Although HmU:G substrates were not tested, the authors suggested the latter could be a biologically important substrate (15). This is also supported by the fact that transition mutation from 5mC:G to T:A is the most frequent substitution mutation in human cancer (16). The two last UDGs identified in human cells, TDG and MBD4 (MED1), are both mismatch-specific and have no activity against singlestranded substrates. TDG excises uracil and thymine from U:G and T:G mismatches, as well as 3,N 4 -ethenocytosine (⑀C) and 5-fluorouracil (5-FU) from double-stranded DNA, and may restore G:C base pairs at sites of cytosine or 5-methylcytosine deamination, or alkylation, respectively (17). MBD4, which does not belong to the same superfamily as the three other UDGs, acts on uracil, thymine, 5-FU, and ⑀C mispaired with guanine (18), as well as on 5-methylcytosine at hemimethylated DNA (19). The preferred substrates, however, are G:T mismatches at methylated or unmethylated CpG islands. Thus, MBD4, as well as TDG, may have a function in the correction of T:G mismatches originating from deamination of 5-methylcytosine.
Several lines of evidence indicate that nuclear UNG2 has a major role in postreplicative removal of misincorporated uracil in mammalian cells (4). The contribution of UNG2 to repair of deaminated cytosines has, however, been debated. Whereas bacterial and yeast ung Ϫ mutants display a mutator phenotype unable to repair deaminated cytosines (20,21), such a phenotype is not clearly observed in Ung Ϫ/Ϫ mice (22). Based on this finding, and comparison of kinetic parameters of the Xenopus SMUG1 (14) and the hUNG catalytic domain (6), it was suggested that in higher eukaryotes, the contribution of UNG2 to the excision of deaminated cytosines was reduced, and that this function might instead be provided by SMUG1 (23). In human cells, however, the UNG catalytic domain alone is not observed in the nucleus (24). Rather, the entire N-terminal regulatory domain remains attached to the core catalytic domain after nuclear translocation. Until now, little information has existed on the enzymatic properties of full-length hUNG2. A likely reason for this is the susceptibility of the enzyme to N-terminal proteolytic degradation during purification, mainly resulting in the core catalytic domain (25). To gain further insight in the functional properties of hUNG2 and hSMUG1 and their relative contribution to nuclear base excision repair, both proteins were purified after overexpression in E. coli. In depth biochemical characterization and analysis of subnuclear localization revealed previously unrecognized properties of both hUNG2 and hSMUG1 that may have important implications for their functions in nuclear BER in vivo.

EXPERIMENTAL PROCEDURES
Expression and Purification of Recombinant hUNG2-An NdeI site was introduced at the hUNG2 translation start codon, and a 947-bp NdeI/HpaI fragment encompassing the entire hUNG2 coding region was ligated into the new vector pJB658cop251kan. This expression vector was constructed by introducing a high copy number variant of the trfA gene (Arg-251 3 Met) and changing the resistance from ampicillin to kanamycin in the toluic acid-inducible, broad host vector pJB658 (26). The resultant construct, p658KanUng2 was introduced into E. coli BL21. Fermentation, preparation of crude extract, and the initial chromatographic steps were essentially as described for the hUNG catalytic domain (6) except that the culture was induced by 1 mM toluic acid (final) and allowed to grow for additionally 3 h at 30°C before harvesting. Furthermore, Complete ® mini (EDTA-free) (Roche) protease inhibitor tablets were included in the initial steps of the purification (1 tablet/10 ml during homogenization and 1 tablet/l ml in subsequent buffers). During size exclusion chromatography on Superdex 75 26/60 (Amersham Biosciences), buffer was changed to 20 mM HEPES-NaOH (pH 8.0), 100 mM NaCl, 1 mM DTT. Pooled active fractions were loaded onto a MonoS HR 5/5 column, and after washing to base-line absorbance, bound proteins were eluted using a linear NaCl gradient from 100 mM to 1 M NaCl in the same buffer. To remove partially N-terminally deleted species, the purest fractions were subjected to MonoS rechromatography and fractions containing the highest portion of apparently full-length protein were pooled. The purified protein was finally concentrated by ultrafiltration and snap-frozen in liquid N 2 prior to storage at Ϫ80°C.
Expression and Purification of Recombinant hSMUG-IMAGE clone identification no. 726197 containing the hSMUG1 cDNA in pT7T3D was cut using NdeI and BfaI, and the 1154-bp fragment encoding hSMUG1 was cloned into the NdeI site of pET11a (Invitrogen). The resultant vector was transformed into E. coli BL21-CodonPlus(DE3)-RIL (Stratagene) and bacterial cell mass for purification produced by fermentation. Expression was induced by 1 mM isopropyl-1-thio-␤-Dgalactopyranoside (final) at 37°C and the culture allowed to grow for additional 4 h prior to cell harvest. Preparation of the hSMUG1 crude extract was essentially as described for the hUNG catalytic domain (6), except that homogenization and protamine sulfate precipitation was performed in 20 mM HEPES-NaOH (pH 6.8), 50 mM NaCl, 0.1 mM EDTA, 1 mM DTT, and 1 tablet/10 ml Complete ® mini (EDTA-free) protease inhibitor mixture (buffer A). The protamine sulfate fraction was loaded onto a DEAE-Sephacel column (Amersham Biosciences, 5 ϫ 9 cm) coupled in series with a CM-Sepharose column (Amersham Biosciences, 5 ϫ 8.5 cm). After washing to base-line absorbance, the DEAE-Sephacel column was bypassed and adsorbed proteins were eluted in a 0.05-1 M NaCl gradient in buffer A. Active fractions were pooled and dialyzed against 20 mM HEPES-NaOH (pH 8), 10 mM NaCl, 1 mM DTT (buffer B) and applied to a UnoS-12 column (Bio-Rad). Adsorbed proteins were eluted in a 0.01-0.7 M linear NaCl gradient. Active fractions were pooled and concentrated by ultrafiltration to 5 ml and loaded onto a Superdex 75 HiLoad 26/60 column (Amersham Biosciences) preequilibrated with buffer B containing 100 mM NaCl, and eluted with the same buffer. Fractions containing hSMUG1 were verified by SDS-PAGE and pooled. After a 4-fold dilution in buffer B, hSMUG1 was loaded onto a poly(U)-Sepharose column (Amersham Biosciences, 1.6 ϫ 10 cm) preequilibrated with buffer B containing 2 mM EDTA. Adsorbed proteins were eluted in a 0.01-1 M linear NaCl gradient in the same buffer and the fractions containing hSMUG1 identified by SDS-PAGE and pooled. The poly(U)-Sepharose fraction was then dialyzed against buffer B, applied onto a MonoS HR 5/5 column, and eluted in a linear 0.01-1 M NaCl gradient in the same buffer. The purest hSMUG1 fractions were collected, and the remaining hSMUG1-containing fractions rechromatographed on MonoS as above. The pooled hSMUG1 fraction was apparently homogeneous as determined by SDS-PAGE and silver staining. The purified protein was finally concentrated by ultrafiltration and snap-frozen in liquid N 2 prior to storage at Ϫ80°C.
Neutralizing Antibodies against hUNG2 and hSMUG1-Polyclonal PU101 against the hUNG catalytic domain was prepared as described (6). Polyclonal PSM1 against hSMUG1 was prepared by subcutaneous injection of 66 g of purified, recombinant hSMUG1 in Freund's complete adjuvant at the neck and flank of New Zealand White rabbits. Subsequent immunizations with the same amount of hSMUG1 in incomplete Freund's adjuvant were performed as above at 3-week inter-hUNG2 and hSMUG1 in Base Excision Repair vals, and the final bleed 10 days after the last immunization. IgGfractions of PU101 and PSM1 were purified on protein A-Sepharose HiTrap columns (Amersham Biosciences) before being used in inhibition experiments.
Cell Culture and Preparation of Nuclear Extracts-Spontaneously transformed human keratinocytes HaCaT, colorectal carcinoma CX-1, and HeLa cells were cultured in Dulbecco's modified Eagle's medium with 10% fetal calf serum, 0.03% glutamine, and 0.1 mg/ml gentamicin at 5% CO 2 . Cells were harvested in the logarithmic growth phase by trypsinization, washed twice in ice-cold phosphate-buffered saline prior to isolation of nuclei. Subsequently, CX-1 and HeLa were washed once in ice-cold isotonic buffer (20 mM HEPES-NaOH (pH 7.8), 1 mM MgCl 2 , 5 mM KCl, 1 mM DTT, and 250 mM sucrose). The pellet was resuspended in hypotonic buffer (isotonic buffer without sucrose) containing Complete ® protease inhibitor mix (1 tablet/50 ml) and left on ice for 45 min to swell. Cells were lysed by 10 -20 strokes in a Dounce homogenizer (tight-fitting pestle), and nuclei pelleted by centrifugation at 600 ϫ g for 6 min. The pellets were resuspended in hypertonic buffer (hypotonic buffer with 0.5 M NaCl and 25% glycerol), and incubated at 4°C on a rotary shaker at high speed for 30 min to extract nuclear proteins. Nuclear debris was removed by centrifugation at 12,000 ϫ g for 15 min, protein concentration in the supernatant measured using the Bio-Rad protein assay (BSA as standard), and the nuclear extract aliquoted, snap-frozen in liquid N 2 , and stored at Ϫ80°C until use. Nuclei from HaCaT keratinocytes were prepared using the NuClear ® nuclear extraction kit (Sigma) according to the protocol form the manufacturer, using isotonic buffer and detergent. Serum starvation, [ 3 H]thymidine pulse labeling, and harvesting of HaCaT cells were as described (27). Polyclonal PSM1 antiserum was covalently coupled to magnetic protein A Dynabeads (Dynal, Norway) according to the instructions from the manufacturer. Cells (5 ϫ 10 5 ) from each time point were resuspended in 1.5 ml of lysis buffer (1ϫ Tris-buffered saline, 1 mg/ml BSA, 1% Triton X-100, 1 tablet/50 ml Complete ® maxiϩEDTA (Roche) protease inhibitors) and passed 10 times through a 21-gauge syringe needle. PSM1coupled magnetic beads (10 l) were added, and the extract incubated on a rotary shaker for 2 h at 4°C. After extensive washing, the beads were eluted by boiling in SDS loading buffer and subjected to SDS-PAGE. After electroblotting to polyvinylidene difluoride membranes, bands were visualized by standard Western analysis using PSM1 as primary antibodies, HRP swine ϫ rabbit secondary antibodies (Dako), and ECL (Amersham Biosciences).
Cloning and Expression of ECFP and EYFP Constructs-PCR was run on DNA from IMAGE clone cDNA_id 726197 encoding hSMUG1 in pT7T3D, using Pfu polymerase (Stratagene), a T7 primer, and the primer 5Ј-ATT TCA ACA GCA GTG GCA GC-3Ј. The PCR product was purified by QIAQuick PCR purification (Qiagen) and digested with EcoRI. The PCR fragment (EcoRI/blunt) encoding hSMUG1 was cloned into EcoRI/SmaI of pEGFP-N1 (Clontech). hSMUG1 and hUNG2 cDNA from pSMUG1-EGFP and pUNG2EGFP (4) were ligated into pEYFP-N1 (Clontech) by transferring the NheI/AgeI fragment into the corresponding sites in pEYFP-N1 to make SMUG1-EYFP and UNG2-EYFP, respectively. To make the C-terminal fusion of SMUG1 with EYFP, the EcoRI/BamHI fragment from SMUG1-EYFP was ligated with EcoRI/BamHI-digested pEYFP-C1 (Clontech). The resulting vector was digested with SacI, the 3Ј overhang removed by T4 DNA polymerase, and the vector re-ligated to make EYFP-SMUG1. The fusion constructs were verified to be in-frame by sequencing.
The construct carrying the hUNG2 promoter, pGL2-P A (27), was digested with HindIII/EcoRV to remove the luciferase gene. The Hin-dIII/DraI fragment from pUNG2-EYFP was then ligated with HindIII/ EcoRV-digested pGL2-P A . From this plasmid, the SacI/NotI fragment containing ProA-UNG2-EYFP was excised, and ligated into SacI/NotIdigested pEGFP-1 (Clontech), thus replacing the EGFP-coding sequence in the vector and giving pProA-UNG2-EYFP. ECFP-PCNA fusion constructs were prepared from pEGFP-PCNAL2 (28), which contains the SV40 NLS in front of PCNA in a C-terminal fusion with EGFP. The SV40 NLS was amplified by using PCR primers containing the sites for the restriction enzymes NheI and AgeI. After digestion with NheI and AgeI, the fragment was cloned into the corresponding sites in the pECFP-C1 vectors (Clontech). Next, the BsrGI (in EGFP)/XbaI (behind PCNA) fragment from the pEGFP-PCNAL2 construct was cloned into the BrsGI/XbaI sites of pECFP-C1 vector with the SV40 NLS N-terminally, giving pNLS-ECFP-PCNA (named pECFP-PCNA). HeLa cells were transiently and stably transfected by using calcium phosphate (Profection, Promega) according to the recommendations from the manufacturer. Stable transfectants of HeLa with pProA-UNG2-EYFP were further cloned by use of limited dilutions in 96microwell dishes in medium containing G418. The cells were examined in a Zeiss LSM 510 laser-scanning microscope equipped with a Plan-Apochromat 63ϫ/1.4 oil immersion objective. The 458-nm laser line was used for excitation of ECFP (detected at 480 nm Ͻ ECFP Ͻ 520 nm) and the 514-nm laser line for EYFP (detected at EYFP Ͼ 560 nm).
UDG Activity Assays-Unless otherwise stated, UDG activity was measured in 20 l of assay mixture containing (final) 20 mM Tris-HCl (pH 7.5), 10 mM NaCl, 1 mM EDTA, 1 mM DTT, 0.5 mg/ml bovine serum albumin, 1.8 M [ 3 H]dUMP-containing calf thymus DNA (specific activity 0.5 mCi/mol) and varying amounts of enzyme. The mixture was incubated 10 min at 30°C, and the amount of released uracil measured as described (29). Kinetic assays using 0.24 -21.5 M calf thymus substrate were performed in the buffer above. In separate experiments kinetic parameters were measured in the presence of either 7.5 mM MgCl 2 or 7.5 mM MgCl 2 and 0.44 ng/l APE1 in the same buffer (molar ratio of APE1/glycosylase at least 10/1). The amount of hUNG2 or hSMUG1 used in the kinetic assays was adjusted to ensure that less than 30% of the substrate was consumed, to ensure linearity of the assay. Kinetic parameters were calculated using the Enzpack for Windows version 1.4 software package (Biosoft) using the method of Wilkinson. Enzyme kinetic parameters were also analyzed using a panel of oligonucleotide substrates. These were prepared by 33 P-5Ј-end labeling of 19-mer PAGE-purified oligonucleotides containing U or HmU at a central position (U141, 5Ј-CATAAAGTGUAAAGCCTGG-3Ј; HmU141, 5Ј-CATAAAGTGHmUAAAGCCTGG-3Ј). To generate double-stranded substrates, the labeled strands were annealed to a 50% excess of the complementary strand containing either A or G opposite U or HmU. Varying substrate concentrations (0.1-15 M) were made by addition of identical non-labeled oligonucleotides. The kinetic assays (10 l total) were performed in the same buffer as for the [ 3 H]dUMP-containing calf thymus DNA in the presence of MgCl 2 . After 10 min of incubation at 30°C, the reactions were stopped by addition of 50 l of 1 M piperidine and heated at 90°C for 20 min to cleave at AP sites. Piperidine was removed by drying under vacuum, and the oligonucleotides redissolved in 65% formamide loading buffer and analyzed by denaturing PAGE (12% polyacrylamide, 7 M urea). Cleaved and uncleaved oligonucleotides were quantified by phosphorimaging. Kinetic parameters were calculated as above. In separate experiments, the activities of hUNG2 and hSMUG1 were directly compared by using ds oligonucleotides (20 nM final) having the same sequence as above, and containing U:G, HmU:G, 5-fluorouracil:A (FU:A), 3,N 4 -ethenocytosine:G (⑀C:G), or 5-hydroxyuracil:G (5-OHU:G). To detect activity against the weaker substrates, reactions contained varying amounts of hUNG2 or hSMUG1 (0 -10 ng/l final) and were incubated at 37°C for 30 in the presence of 7.5 mM Mg 2ϩ with or without APE1 (0.1 ng/l final). Inhibition of hSMUG1 and hUNG2 by polyclonal antibodies or oligonucleotides containing AP sites was monitored after incubation of the purified enzymes or cell extracts with inhibitor for 10 min on ice prior to the enzyme assays.
BER Assay-Double-stranded plasmid DNA substrates containing uracil at defined positions were prepared essentially as described (30). Briefly 20 g of ssDNA (pGEM-3zfϩ) was annealed to 4.2 g of a 22-mer complementary oligonucleotide containing uracil and synthesis of duplex DNA carried out in the presence of T4 DNA polymerase, T4 DNA ligase, and T4 gene 32 ssDNA-binding protein at 37°C for 2h. Closed circular DNA duplex molecules were purified by CsCl gradient centrifugation.
The base excision repair mixtures (50 l) contained (final) 40 mM HEPES-KOH (pH 7.8), 70 mM KCl, 5 mM MgCl 2, 0.5 mM DTT, 2 mM ATP, 20 M dATP, 20 M dGTP, 8 M dTTP or dCTP depending on the isotope used, 8 M phosphocreatine, 0.36 mg/ml BSA, 1 g/ml creatine phosphokinase, 40 Ci/ml [␣-32 P]dTTP or [␣-32 P]dCTP, 0.6 or 0.2 mg/ml total cell or nuclear extract, respectively (measured as protein), and 6 g/ml dsDNA template. The repair mixtures were incubated at 30°C for the given times, stopped by adding EDTA and RNase A, and further incubated with proteinase K (30 min) and SDS (10 min) at 37°C. The repair product was recovered by phenol/chloroform extraction and ethanol/salt precipitation. The DNA was digested using XbaI/HincII for short-patch and HincII/PstI for long-patch BER, and repair products were analyzed by electrophoresis in 12% denaturing polyacrylamide gels and phosphorimaging of the dried gels.
To specifically monitor the glycosylase and AP endonuclease steps in the BER-assays, 50-mer double stranded oligonucleotides were prepared that corresponded to the region of the BER plasmid substrates containing the U:A or U:G lesions (uracil-containing strand: 5Ј-TCGG-TACCCGGGGATCCTCTAGAGXYGACCTGCAGGCATGCAAGCTTG-AG-3Ј, with X representing uracil in the U:A oligo, and Y representing uracil in the U:G oligo). The duplexes were incubated in the presence of the same amounts of nuclear extract and in the same buffer as in the BER assay, except that dNTPs were replaced by ddCTP or ddTTP to arrest polymerization and ligation. The amount of AP sites or incised product was quantified as described above.

Purification of Recombinant hUNG2 and hSMUG1-We
have found full-length hUNG2 to be notoriously difficult to express and purify from various E. coli strains and budding yeast (Pichia) caused by N-terminal proteolysis. Apparently, this was caused by both in vivo proteolytic attack and degradation during purification (data not shown). The degradation problem was partially overcome by expressing the protein containing an N-terminal His tag. For functional analysis we decided to avoid histidine tags, however, because this may seriously affect the functional properties of some proteins (31). In the present system, the highest relative amount of fulllength hUNG2 at the time of cell harvest was observed using E. coli BL21 (ompT Ϫ , lon Ϫ ), in which ϳ50% of the protein appeared unprocessed. The presence of mixed protease-inhibitors during purification only partially prevented further loss of the intact protein, which generated enzyme species lacking one or a few of the N-terminal residues. These results indicate that the hUNG2 N-terminal regulatory region constitutes a distinct domain sensitive to proteolysis. This would also explain why previous attempts to purify UNG from mammalian cells in the absence of appropriate protease inhibitors have yielded essentially the catalytic domain (25). This is also corroborated by the abnormal chromatographic behavior of hUNG2. The full-length protein (34.6 kDa) eluted as a 52-kDa protein in size-exclusion chromatography, whereas the compact, spherical catalytic domain (25.5 kDa) eluted as a 20-kDa protein (data not shown). Furthermore, hUNG2 eluted over a broad NaCl range in MonoS chromatography, in contrast to the sharp peak observed with the catalytic domain alone. This may indicate that the hUNG2 N-terminal is less ordered than the catalytic domain, and may adopt alternative conformations.
The purified enzyme after repeated MonoS chromatography was apparently homogeneous as judged by SDS chromatography and silver staining. N-terminal sequencing, however, revealed that this fraction contained ϳ60% full-length hUNG2, and ϳ40% of UNG2⌬1 lacking the N-terminal methionine. This was judged adequate for biochemical characterization of the enzyme, and no further attempts were made to remove the 1-amino acid truncated form. Approximately 1 mg of purified, full-length hUNG2 was obtained from a 5-liter fermentation. hSMUG1 was expressed as a full-length protein. However, the hSMUG1 N terminus also proved to be highly susceptible to proteolytic attack even in the presence of mixed protease inhibitors, and virtually all hSMUG1 obtained after purification proved to lack the 16 N-terminal amino acids. The yield was ϳ5 mg of protein/5 liters of fermenter culture. No significant difference in enzymatic activity was observed with the ⌬16 form as compared with full-length hSMUG1 expressed using another vector system (pTYB12; data not shown). The former was, however, chosen for biochemical analysis because of the very low yield from the pTYB12 system.
Both hUNG2 and hSMUG1 Are Stimulated by Mg 2ϩ and APE1-To analyze the kinetic properties of hUNG2 and hSMUG1, we first established reaction conditions under which both activities could be directly compared. The pH profile of each enzyme was broad, with an optimum between pH 7.0 and 7.5 (data not shown). Furthermore, hUNG2 was stimulated ϳ2-fold by 50 mM NaCl (Fig. 1A), whereas hSMUG1 was ϳ30% stimulated at 70 mM NaCl (Fig. 1B). Both the pH profiles and NaCl optima are similar to those reported for the major UDG activity partially purified from HeLa cells (29). In the latter study, the authors also observed a weak stimulation of the activity by Mg 2ϩ . To our knowledge, this has not been reported for any other monofunctional DNA glycosylase, including UNG, and it was thus of interest to analyze whether this could be caused by stimulation of co-purifying hSMUG1 in their study. To our surprise, however, both enzymes were markedly stimulated by the presence of Mg 2ϩ . In the absence of monovalent cations, hUNG2 was stimulated nearly 10-fold in the presence of 10 mM MgCl 2 (Fig. 1A), whereas a near 2-fold stimulation was observed for hSMUG1 (Fig. 1B). This is in contrast to properties of the catalytic core form of hUNG, which is inhibited by Mg 2ϩ at all concentrations tested (data not shown). Thus the N-terminal 86 amino acids in hUNG2 are required for the observed Mg 2ϩ -stimulation. Because hSMUG1 and hUNG2 are both nuclear proteins, and the hUNG2 N-terminal region is retained after nuclear translocation (24,32), we decided to undertake an in-depth study of the kinetic effects of magnesium on both enzymes. An additional advantage of studies of effects of Mg 2ϩ on hUNG2 and hSMUG1 was that their properties could be analyzed in the presence of fully active APE1, which catalyzes the next step in the BER pathway. APE1 has been demonstrated to affect the catalytic rate of several DNA glycosylases (33,34) and requires Mg 2ϩ for full catalytic activity, but not for DNA binding (35).
The effects of Mg 2ϩ on single-stranded (ssU) and doublestranded (dsU) DNA substrates in the absence and presence of APE1 are shown in Fig. 1, C (hUNG2) and D (hSMUG1). In the absence of Mg 2ϩ , the activities of hUNG2 against ss-and ds- essentially identical. Moreover, a weak stimulation of the activity against dsU was observed in the presence of APE1. Increasing the Mg 2ϩ concentration resulted in a gradual increase in activity up to 6 and 10 mM Mg 2ϩ for dsU and ssU, respectively. In addition, the stimulatory effect by APE1 markedly increased for dsU in the presence of Mg 2ϩ (Fig. 1C). A somewhat different pattern was observed for hSMUG1 (Fig.  1D). In the absence of Mg 2ϩ , hSMUG1 had ϳ4-fold higher activity against ssU than dsU. The activity against both substrates increased with increasing Mg 2ϩ up to 6 mM, although the relative ssU/dsU activity did not change. As for hUNG2, the presence of APE1 had only a weak stimulatory effect with both substrates in the absence of Mg 2ϩ . However, in the presence of Mg 2ϩ , APE1 had a marked stimulatory effect. This was most pronounced for dsU, as the activity was stimulated ϳ10-fold by APE1 in the presence of 10 mM Mg 2ϩ .
In separate experiments, potential stimulation of hUNG2 and hSMUG1 by the second human AP endonuclease APE2 and the Flap endonuclease FEN1 was investigated. Neither of these proteins, however, had any effect on the glycosylase activities (data not shown).
Mg 2ϩ Strongly Increases the Affinity of hUNG2 for Uracil in Single-stranded DNA-To further analyze this complex pattern of stimulation, kinetic assays were performed both in the presence and absence of Mg 2ϩ and APE1. Based on the above, the reaction conditions were kept identical for hSMUG1 and hUNG2 (see "Experimental Procedures"). The results are shown in Table I. In the absence of Mg 2ϩ , hUNG2 showed only a marginally elevated specificity (k cat /K m ) for ssU over dsU (1.3-fold), whereas for hSMUG1 the corresponding ratio was 1.6. Thus the common term "single-strand selective monofunctional uracil-DNA glycosylase 1" could be justified for hSMUG1 under these conditions. In the presence of Mg 2ϩ , however, this picture dramatically changed as the ssU/dsU specificity ratio of hUNG2 increased ϳ40-fold. This was caused by a 127-fold increase in k cat /K m of hUNG2 against ssU, whereas some 3-fold increase was observed with dsU. Interestingly, with the dsDNA substrate Mg 2ϩ appeared to influence the catalytic turnover only. With the ssDNA substrate, however, Mg 2ϩ only affected the affinity (reduced K m ). Thus, the presence of physiological concentrations of Mg 2ϩ apparently turns hUNG2 into an efficient, single-strand selective enzyme ideally suited for identifying rare, deaminated cytosine residues residing in singlestranded regions of DNA. A similar effect was not observed for hSMUG1. Although the ssU/dsU specificity ratio increased from 1.6 to 3.8 (based on k cat /K m ) in the presence of Mg 2ϩ , this effect was solely caused by changes in k cat . In fact, the presence of Mg 2ϩ reduced the affinity of hSMUG1 toward both substrates.
When both Mg 2ϩ and APE1 were included in the assays, a further increase in k cat /K m was observed for hUNG2 against dsU, whereas k cat /K m against ssU was reduced by 50% compared with Mg 2ϩ alone. However, hUNG2 still maintained a 7.4-fold preference for ssU over dsU. For hSMUG1, the opposite effect was observed in the presence of APE1. The affinity and turnover against both ssU and dsU increased, but the increase was quantitatively higher with dsU. Thus, in the presence of both Mg 2ϩ and APE1, hSMUG1 displays a 2-fold preference for the double-stranded substrate.
AP Sites Inhibit hSMUG1 but Not hUNG2-AP sites have been shown to be a strong (micromolar range) competitive inhibitor of the catalytic hUNG domain, and this domain binds to AP sites more strongly and more rapidly than to uracilcontaining DNA (33). Such binding is also observed with several other DNA glycosylases (36,37). We postulated that binding to the product AP site could be of crucial importance to avoid mutagenic and cytotoxic effects of the AP site until the subsequent AP endonuclease arrived and ensured further processing of the damage site (33). Surprisingly, no inhibition by AP sites was observed with the full-length hUNG2 protein when compared with corresponding non-AP-containing oligonucleotides ( Fig. 2A). An entirely different effect was observed with hSMUG1. Whereas no specific inhibition was observed with the AP-containing ss-oligonucleotides, a strong inhibitory effect was observed with the AP-containing ds-oligonucleotides (Fig. 2B). Furthermore, the AP:G construct inhibited somewhat more strongly than AP:A. These results may explain the 15-fold increase in k cat /K m for hSMUG1 against dsU when APE1 is present (Table I). Because APE1 does not stimulate this activity in the absence of Mg 2ϩ (Fig. 1C), this likely occurs mainly by Mg 2ϩ -stimulated endonucleolytic cleavage of the AP site, whereby hSMUG1 product rebinding is blocked. hSMUG1 displacement from ds AP sites by APE1, similar to human hTDG (36) may, however, be a contributing factor, because a weak stimulation by APE1 is also observed in the absence of Mg 2ϩ (Fig. 1D) (note that APE1 binds to AP sites in the absence of Mg 2ϩ , but cleaves at a very reduced rate). Because singlestranded AP sites do not inhibit hSMUG1, such a mechanism cannot explain the observed 2-fold increase in k cat /K m of hSMUG1 against ssU by APE1. Active recruitment of hSMUG1 to the APE1⅐DNA or formation of a hSMUG1⅐APE1 complex prior to substrate binding could hypothetically contribute to this, but no such interaction has been observed in immunoprecipitation experiments (23). It is also less obvious why APE1 only stimulates the activity of hUNG2 against dsU. Because hUNG2 is neither inhibited by AP sites in ssDNA nor dsDNA, and because the major effect of APE1 is on the K m against both substrates, it is tempting to speculate that APE1 may directly  (15) recently demonstrated that recombinant hSMUG1 and SMUG1 purified from calf thymus were able to excise HmU from DNA (HmU:A) in addition to U, although a direct comparison between the two substrates was not performed. To investigate the relative efficiency of hUNG2 and hSMUG1 against different uracil analogs, various concentrations of both enzymes were incubated with a panel of doublestranded oligonucleotides containing U:G, FU:A, HmU:G, ⑀C:G (Fig. 3), and 5-OHU. The results indicated narrow substrate specificity for hUNG2 restricted to uracil and uracil analogs with minor structural modifications at the 5-position (substrate preference: U Ͼ Ͼ 5-FU). In contrast, much broader substrate specificity was observed for hSMUG1. In addition to previously reported U and HmU, the enzyme was also active against FU and ⑀C, with the order of preference U Ͼ HmU Ͼ Ͼ ⑀C Ͼ FU. Neither enzyme was able to excise 5-OHU under the present conditions (data not shown). These results indicate considerable structural differences in the active site of hSMUG1 compared with hUNG2. Thus, whereas pyrimidines having bulky substitutions at the 5-position do not have access to the hUNG2 active site, hSMUG1 is able to accommodate such substitutions at the 3-, 4-, and 5-positions.
Based on analysis of UDG activities from Ung Ϫ/Ϫ mice (23) and the low K m value (0.035 M) of Xenopus SMUG1 (14), it was suggested that SMUG1 evolved to counteract the mutagenic effects of deaminated cytosines, and that the relative contribution of UNG2 to the repair of this lesion was reduced in mammalian cells (23). The data presented in Fig. 3 indicate that hUNG2 removes U from U:G mismatches even more efficiently than hSMUG1 and is thus a candidate enzyme to counteract cytosine deamination. To further analyze this, the reaction kinetics of hUNG2 and hSMUG1 were analyzed using oligonucleotides containing A, G, or no base opposite the substrate (U of HmU) and in the presence of Mg 2ϩ (Table II). The results demonstrated that hUNG2 had a 300-fold higher k cat /K m than hSMUG1 against U:G mismatches, and that this was caused both by a lower K m and a higher k cat in the case of hUNG2. Furthermore, hUNG2 had a Ͼ100-fold higher k cat /K m than hSMUG1 against ssU. Although the results with short oligo-nucleotides are not directly comparable to the kinetic parameters obtained with [ 3 H]dUMP-containing calf thymus DNA (Table I), they clearly demonstrate that both U:G and ssU are very good substrates for hUNG2. The results furthermore indicate that HmU is a somewhat poorer substrate for hSMUG1 than uracil, independent of the opposing base.
hSMUG1 and hUNG2 Are Differently Localized within the Nucleus-We have previously demonstrated that hUNG2 is efficiently translocated to nuclei and that, in the S-phase, it is clearly accumulating in replication foci (4). This is in accordance with its function in postreplicative removal of misincorporated uracil. Because the above analyses were performed with fixed cells, we wanted to re-examine the translocation of hUNG2 in living cells. In cells transiently transfected with UNG2-EYFP (Fig. 4), a significant portion of hUNG2 was localized outside replication foci. To ensure that this was not a result of overexpression of the fusion protein to levels exceeding the cellular translocation capacity, the fusion protein was stably transfected under the control of the hUNG2 promoter. A similar result was observed in these cells. hUNG2 appeared to be evenly distributed in the nucleoplasm outside the S-phase, and a fraction of hUNG2 accumulated in replication foci in the S-phase. Furthermore, hUNG2 appeared to be excluded from nucleoli both in the S-phase and outside the S-phase. hSMUG1, however, appeared to be less strictly localized to nuclei, and a substantial fraction is distributed throughout the cytoplasm.

hUNG2 and hSMUG1 in Base Excision Repair
This was also the case when EYFP was fused to the C-terminal end of hSMUG1 (data not shown). Furthermore, hSMUG1 appeared to be especially abundant in the nucleoli both in replicating and non-replicating cells. The physical localization of the two proteins indicates that both hUNG2 and hSMUG1 are available for repair of deaminated cytosines outside replication foci, in accordance with their substrate preferences. Only hUNG2, however, appears to have a function in postreplicative repair in replication foci. Conversely, only hSMUG1 is observed in nucleoli. The latter observation is intriguing, and may reflect a yet unrecognized function of hSMUG1 in nucleoli. hUNG2 Is the Major Glycosylase for Removal of Both Deaminated Cytosine and Misincorporated Uracil in Nuclear Extracts-Nilsen et al. (23) recently provided evidence that SMUG1 represented a major glycosylase against U:G mismatches in ung-deficient mice. To analyze whether this was also so in human cells and to compare the hSMUG1 activity against oligonucleotides containing U:A and U:G, nuclear extracts from HaCaT, HeLa, and CX-1 cells were incubated with neutralizing antibodies to specifically inhibit hUNG2, hSMUG1, or both. In separate experiments the degree of inhibition and specificity of the anti-hUNG PU101 and anti-hSMUG1 PSM1 polyclonal antibodies were analyzed. Both antibodies proved to be specific, and essentially no cross-reactivity was observed with the two enzymes (Fig. 5A). The apparently higher concentration of antibody necessary to inhibit hSMUG1 was a result of the relatively high concentration of hSMUG1 needed to obtain measurable activity using [ 3 H]dUMP-calf thymus DNA as substrate. The latter was considered not to pose a problem when neutralizing nuclear extracts, as the number of SMUG1 molecules per cell in the present cell lines appears to be considerably lower than UNG2. Whereas hUNG2 is easily detected by Western analysis of cell-free extracts, we were not able to detect hSMUG1 without previous concentration by immunoprecipitation (data not shown). Inhibition of the recombinant proteins by the PBS2encoded inhibitor Ugi was also analyzed. Ugi proved to be a potent inhibitor of hUNG2. However, ϳ40% of the hSMUG1 activity was also inhibited at the concentrations of Ugi needed to fully inhibit hUNG2 (data not shown). Thus, to provide the highest specificity, neutralizing antibodies were used as inhibitors throughout the present study. The inhibited nuclear extracts were then incubated with 33 P-labeled U141A and U141G. To obtain measurable activities of hSMUG1, both the final concentration of nuclear protein (0.33 g/l) and the incubation time (1 h) were elevated. Even under these conditions, preincubation of the extracts with hUNG2-neutralizing PU101 inhibited Ͼ90% of the total activity against both substrates (Fig. 5B). This clearly indicates that hUNG2 is the major activity against both misincorporated uracil and deaminated cytosine in nuclear extracts, and supports our earlier finding that UNG proteins represent Ͼ98% of the activity against misincorporated uracil in HeLa total cell extracts (6). Furthermore, essentially all UDG activity in the extracts could be inhibited by preincubation with both PU101 and PSM1, indicating that hUNG2 and hSMUG1 together represent essentially all detectable UDG activity in human nuclei under these conditions. hSMUG1 Is the Major 5-Hydroxymethyluracil-DNA Glycosylase in Nuclear Cell Extracts-Several lines of evidence indicate that human cells contain more than one HmU-DNA glycosylase (15,38). To quantitate the contribution of hSMUG1 to the total HmU glycosylase activity in human nuclei, the CX-1 nuclear extract was incubated with a duplex oligonucleotide containing HmU:G, a substrate common to all reported HmU-DNA glycosylases. Preincubation of the nuclear extract with UNG-neutralizing PU101 did not affect the amount of excised HmU, whereas preincubation with SMUG1-neutralizing PSM1 inhibited essentially all HmU glycosylase activity (Fig. 6A). Thus, SMUG1 likely represents the major HmU glycosylase activity in the nuclear extracts. Alternatively, a HmU glycosylase antigenically similar to SMUG1 may be present in human cells. To investigate the latter possibility, total protein was extracted from synchronized HaCaT cells at different times after release, and immunoprecipitated using PSM1 covalently linked to magnetic Dynabeads (Fig. 6, B and C). SDS-PAGE and Western analysis demonstrated a single band of similar molecular mass as recombinant hSMUG1 in all the samples. This supports the theory that hSMUG1 is the major HmU-DNA glycosylase in the cells. Furthermore, the near constitutive expression of hSMUG1 in both quiescent cells and through the cell cycle supports the idea that hSMUG1 is not linked to the replication process. This is in contrast to the human UNG proteins, which are markedly up-regulated in the S-phase (27,39), and the demonstrated function of hUNG2 in postreplicative removal of misincorporated uracil.
In Vitro Short-patch BER of U:A Is Initiated by hUNG2, whereas Repair of U:G May Be Initiated by Alternative Mech-anisms-To investigate whether UNG2 and SMUG1 from human cells were able to initiate BER of uracil in vitro, nuclear extracts from HeLa and CX-1 cells were preincubated with PU101 and/or PSM1 neutralizing antibodies prior to addition to an in vitro reconstituted BER assay. As demonstrated in Fig.  7A, inhibition of nuclear hUNG2 by PU101 completely abolished BER of the U:A plasmid substrate, whereas inhibition of hSMUG1 did not affect the amount of ligated product. This apparently reflected the initial rate of uracil removal by the two enzymes, as demonstrated by excision of uracil from a 50-mer U:A-containing oligonucleotide having a sequence identical to that of the sequence flanking U:A in the BER substrate (Fig. 7B). In this oligonucleotide the anti-UNG antibody completely inhibited uracil excision. When a circular U:G plasmid was used as BER substrate, preincubation with neutralizing antibodies had little effect on the amount of repair product (Fig.  7A). This was entirely unexpected, because uracil excision from the corresponding 50-mer U:G oligonucleotide was completely inhibited by the anti-UNG antibody (Fig. 7B) 7. hUNG2 but not hSMUG1 is a major initiator of in vitro short-patch BER. A, nuclear extracts from HeLa and CX-1 cells were assayed for short-patch BER activity using a closed circular plasmid substrate containing a single U:A or U:G at a defined position. The extracts were preincubated in the presence or absence of neutralizing anti-UNG (PU101) and/or anti-SMUG1 (PSM1) as indicated and the BER reaction allowed to proceed for 60 min. B, the degree of inhibition of the glycosylase step was analyzed using 50 bp of 33 P-labeled oligonucleotides corresponding to the region encompassing uracil in the plasmid substrates in A, and under otherwise identical conditions except that dNTPs were replaced by ddCTP or ddTTP to arrest polymerization and ligation. C, time-course BER assay using CX-1 nuclear extracts. The extracts were preincubated with neutralizing antibodies as in A, and reactions incubated for 10, 30, or 60 min. Ab, antibody. identical experimental conditions. In all BER experiments above, at least 90% of the product resulted from short-patch repair incorporation of only one nucleotide, whereas less than 10% resulted from long-patch BER (data not shown). Thus, the non-inhibited U:G repair was likely not caused by nucleotide excision repair or mismatch repair. The latter was also verified by experiments using nuclear extracts from mismatch repairdeficient HCT116 colorectal cancer cells (ATCC CCL-247), in which U:G repair of the plasmid substrate was also observed in the presence of both neutralizing antibodies (data not shown). Moreover, the short-patch DNA-repair process did not appear to be dependent on PCNA loading by replication factor C, because exclusion of ATP from the reactions had no effect on the steps prior to ligation (data not shown). A time-course BER experiment indicated, however, that the degree of inhibition by the neutralizing antibodies varied with the incubation time of the BER reactions. At short incubation times, preincubation with anti-UNG antibodies essentially abolished BER, whereas at prolonged incubation times the level of BER product in the anti-UNG reactions approached that of the uninhibited reactions (Fig. 7C). This might indicate that repair of the U:G mismatch plasmid takes place by two distinct short-patch BER mechanisms. In the initial phase of the reaction, hUNG2-initiated short-patch BER dominates. In the late phase initiation takes place by a yet unidentified mechanism, which is not detected using U:G oligonucleotides as substrate. The possible mechanisms of this alternative mode of repair will be discussed below.

DISCUSSION
By purifying the recombinant nuclear isoforms of hUNG2 and hSMUG1 to homogeneity, we are now, for the first time, able to carefully determine the kinetic parameters of hUNG2 and hSMUG1 and to compare their properties directly. The present biochemical data strongly suggest that UNG2 has a broader function than mere postreplicative removal misincorporated uracil, and likely is the major nuclear enzyme for removal of deaminated cytosine in both double-stranded and single-stranded DNA. Given the unsurpassed efficiency of hUNG2 compared with hSMUG1 to remove uracil, it is reasonable to believe that SMUG1 serves another primary function in higher eukaryotes. hSMUG1 shares many of the characteristics of hUNG2. There are, however, clear differences, such as markedly lower turnover number, strong binding to AP sites, broader substrate specificity, and the accumulation of hSMUG1 in nucleoli but not in replication foci. Notably, the activity of hSMUG1 against HmU is nearly as high as against U, indicating that this might be the primary substrate for hSMUG1 in vivo. This is also supported by the phylogenetic distribution of SMUG1, which is linked to the use by organisms of 5-methylcytosine as a mediator of gene expression (40). The present work shows that recombinant hSMUG1 has a substrate preference for HmU in the order ssHmU Ͼ HmU:G Ͼ HmU:A (Table II). For the corresponding U-containing substrates, somewhat higher activities are observed, although the order of preference is retained. This preference for U-containing substrates is also found by Boorstein and co-workers (15). More recently, however, a HmU-DNA glycosylase having distinct, but overlapping, substrate specificity compared with hSMUG1 was partially purified from HeLa cells (38). Like hSMUG1, this enzyme excised both U and HmU opposite A and G, with a preference for U. Likewise, the enzyme had a weak activity against 5-fluorouracil. However, no activity was observed against single-strand substrates or 3,N 4 -ethenocytosine. This activity may be the same as previously observed by Radany et al. (41) in human glioma cells, as the latter had no activity against G:T mismatches and was strictly specific for paired uracil with a preference for U:G. Interestingly, we identified a second EST derived from the SMUG1 gene by data base search (NM_014311). This novel splice form is identical to SMUG1 in the 135 N-terminal amino acids, but utilizes a different reading frame of the SMUG1 3Ј-untranslated region to generate the 42 C-terminal amino acids. The biological function of this novel variant, if any, is not known. Two of three main motifs believed to be important for SMUG1 glycosylase activity (5), including the residues Asp-163 and Arg-243, are not present in the variant, suggesting that this protein may have activities distinct from SMUG1.
The finding that hUNG2 is a highly effective single-strand selective UDG in the presence of physiological concentrations of magnesium may indicate that this is an important function of hUNG2 in vivo. Transient single-strand regions occur frequently in chromatin both as a consequence of "breathing" in A:T-rich DNA, and in normal DNA metabolism such as replication, transcription, and recombination. Furthermore, because the rate of deamination of cytosine is more than 100-fold higher in ssDNA than in dsDNA (1), this would justify an efficient enzymatic scanning for uracil in ssDNA. This is substantiated by the finding that the mutation rate in Ung-deficient yeast cells was increased 2-fold by a high transcription rate (42). Efficient removal of deaminated cytosine from transcriptionally active ssDNA could also be important to avoid miscoding transcripts and synthesis of harmful proteins. The potential of generating such proteins was demonstrated for human UNG, as a single base pair substitution changed UNG from being a DNA repair enzyme to actually becoming a mutator enzyme (12). Notably, hUNG2 contains two RPA-binding motifs in the N-terminal regulatory domain (4), which may aid the recruitment to single-stranded regions of DNA.
In the present study we identified a BER-initiating activity that is apparently specific for long and/or circular U:G substrates, and that is not inhibited by neutralizing antibodies against hUNG or hSMUG1. Slow U:G-BER could in principle be initiated by the mismatch glycosylases TDG or MBD4. However, in the cell lines used in this study, hUNG2 and hSMUG1 represented all detectable UDG activity, even at high concentrations of nuclear protein and at prolonged incubation times. It is thus tempting to speculate that the observed repair is initiated by a fraction of hUNG2 residing within a protein complex specific for U:G mismatches, and that is not readily accessible to binding by the antibodies. This is also supported by findings in our laboratory that a fraction of UNG2 is not accessible for immunoprecipitation. Interestingly, a 180-kDa multiprotein complex isolated from bovine testis was able to perform complete BER of a U:G-containing oligonucleotide (43). The complex contained pol ␤, ligase I, AP endonuclease, and UDG, although the precise nature of the latter two proteins was not determined. The mere size of such a complex, if existent in human cells, may explain why the 50-mer U:G oligonucleotide is not a substrate. The precise identification of the components of such a complex must, however, await its isolation in sufficient quantities for analysis by, e.g., mass spectrometry.
As shown in Fig. 8, uracil and HmU in DNA may be present in different positions relative to a replication fork, and in addition the sequence context may vary. It seems likely that both the type of initiating UDG, as well as the BER subpathway in the subsequent steps, will depend on these factors. Our data indicate that the majority of HmU is excised by hSMUG1 outside of replication foci, and is thus likely processed by shortpatch BER. Deaminated cytosine present in dsDNA prior to replication, e.g. in the G 1 -phase, may in principle be removed by any of the four identified human uracil-DNA glycosylases.
Most likely, hTDG and hMBD4 mainly function in CpG contexts and in dsDNA only, whereas hSMUG1 and hUNG2 may operate in any sequence context, albeit with varying efficiency (6,44,45). SMUG1 was recently suggested to be a major enzyme in repair of uracil in U:G mismatches (23). However, our data strongly suggest that, even in repair of U:G mispairs resulting from cytosine deamination, hUNG2 may be a major player. hUNG2 has at least as low a K m as hSMUG1, and is present in replication foci as well as in the nucleoplasm. However, hUNG2 is essentially excluded from nucleoli, whereas hSMUG1 accumulates in nucleoli, suggesting that hSMUG1 may have a specialized role in uracil repair in nucleoli. Among the uracil-DNA glycosylases, only hUNG2 specifically accumulates in the replication foci during S-phase, and all experimental evidence suggests that UNG2 has an important role in the removal of misincorporated uracil in replication foci.
UNG2 is by far the most efficient enzyme for removal of uracil in ssDNA. The lack of AP site rebinding by hUNG2 should not pose a problem in ssDNA outside of replication foci because the double-helical DNA conformation is likely restored prior to replication, thus creating a substrate for APE1. Binding of APE1 then ensures that the downstream processing steps take place by stepwise recruitment of BER-factors and completion of repair prior to replication. One important problem is, however, the following. How are deaminated cytosines that either escape repair prior to replication, or is formed at the replication fork, repaired, if at all? If uracil escapes repair and directs incorporation of adenine, subsequent removal of uracil by hUNG2 and replacement by thymine would result in a G:C 3 A:T transition. This would be a poor strategy unless postreplicative A:U pairs resulting from misincorporation of A opposite a deaminated cytosine are processed differently from misincorporated U opposite A. At present, there is no experimental evidence supporting such differential postreplicative processing. However, our data demonstrate that hUNG2 very efficiently removes uracil from single-stranded DNA, and may thus generate an abasic site that blocks replication. In addi-tion, specific binding to RPA may boost recruitment of hUNG2 to single-stranded DNA in front of the replication fork. The stalled replication fork may recruit proteins required for fork regression and homologous recombination, which may be alternative mechanisms to short-patch repair and long-patch repair in the downstream steps subsequent to uracil removal. Fortunately, AP endonucleases require double-stranded DNA so the risk of creating a strand break (functionally a double-strand break) is small. Involvement of recombination in the repair of abasic sites has been documented for E. coli (46). Furthermore, induction of deamination of cytosine by NO . is strongly cytotoxic in E. coli recBCD cells deficient in recombination (47). Notably, disruption of the genes for Ung and Fpg in these cells led to enhanced survival, inferring that both Ung and Fpg create substrates for recombinational repair (48). The finding that recombination factors are required for processing of abasic sites in bacteria suggests that this may also be the case in mammalian cells, because this basic process is highly likely to be conserved. We therefore propose that uracil in singlestranded DNA at the replication fork is incised by hUNG2 and repaired by recombination or fork regression, which are both processes requiring recombination proteins. Note that, when deaminated cytosines occur in the single-stranded region in front of the replication fork, hUNG2 probably excises uracil and leaves an AP site that may stall the replication fork and induce recombination or fork regression. Because AP endonucleases are highly double-strand-specific, the risk of crating a strand break (functionally a double-strand break) is very low. AP sites at the replication fork are most likely repaired either by BER subsequent to fork regression, or by recombination using information from the sister chromatid (which is now double-strand in this short region). Alternatively, AP sites are bypassed by translesion synthesis (TLS).