The role of yeast DNA 3'-phosphatase Tpp1 and rad1/Rad10 endonuclease in processing spontaneous and induced base lesions.

Tpp1 is a DNA 3'-phosphatase in Saccharomyces cerevisiae that is believed to act during strand break repair. It is homologous to one domain of mammalian polynucleotide kinase/3'-phosphatase. Unlike in yeast, we found that Tpp1 could confer resistance to methylmethane sulfonate when expressed in bacteria that lack abasic endonuclease/3'-phosphodiesterase function. This species difference was due to the absence of delta-lyase activity in S. cerevisiae, since expression of bacterial Fpg conferred Tpp1-dependent resistance to methylmethane sulfonate in yeast lacking the abasic endonucleases Apn1 and Apn2. In contrast, beta-only lyases increased methylmethane sulfonate sensitivity independently of Tpp1, which was explained by the inability of Tpp1 to cleave 3' alpha,beta-unsaturated aldehydes. In parallel experiments, mutations of TPP1 and RAD1, encoding part of the Rad1/Rad10 3'-flap endonuclease, caused synthetic growth defects in yeast strains lacking Apn1. In contrast, Fpg expression led to a partial rescue of apn1 apn2 rad1 synthetic lethality by converting lesions into Tpp1-cleavable 3'-phosphates. The collected experiments reveal a profound toxicity of strand breaks with irreparable 3' blocking lesions, and extend the function of the Rad1/Rad10 salvage pathway to 3'-phosphates. They further demonstrate a role for Tpp1 in repairing endogenously created 3'-phosphates. The source of these phosphates remains enigmatic, however, because apn1 tpp1 rad1 slow growth could be correlated with neither the presence of a yeast delta-lyase, the activity of the 3'-phosphate-generating enzyme Tdp1, nor levels of endogenous oxidation.

DNA 3Ј-phosphatase is a more recently discovered BER enzyme. We have identified and begun to characterize Tpp1, the S. cerevisiae phosphatase (24,25). Tpp1 is homologous to one portion of the bifunctional enzyme polynucleotide kinase/3Ј phosphatase (PNKP) found in Schizosaccharomyces pombe and metazoans (26 -29). It is also homologous to a family of plant enzymes, including characterized Zea mays and Arabidopsis thaliana counterparts (30,31). Bacteria, in contrast, lack a related DNA 3Ј-phosphatase. Like the plant enzymes, Tpp1 is devoid of a polynucleotide kinase domain, and we have provided functional evidence that budding yeast in fact do not possess this activity (24). Thus, only the 3Ј-phosphatase function of this enzyme family is consistently conserved, and only in eukaryotes. Tpp1 is functionally redundant with yeast Apn1 and Apn2, which are themselves capable of removing 3Ј-phosphates (17,18). A deficiency of 3Ј-phosphate repair is only detected when APN1 and TPP1 both are mutated, with the greatest deficiency caused by simultaneous apn1 apn2 tpp1 mutation (25). Unlike Apn1 and Apn2, however, Tpp1 activity appears restricted to 3Ј-phosphates (25).
Here, we have used a variety of methods to explore the cellular function of Tpp1. We find that Tpp1 can function in bacteria to cleave 3Ј-phosphates left by ␦-elimination, but that budding yeast lack a ␦-lyase to account for the conservation of Tpp1. Tpp1 functions in the repair of endogenous DNA damage, but this effect is not realized until a redundant Rad1/ Rad10 salvage pathway is also disabled. This and other results demonstrate that strand breaks with persistent irreparable blocking lesions are highly cytotoxic. The source of the requisite endogenously created 3Ј-phosphates remains enigmatic, but does not appear related to ROS.

EXPERIMENTAL PROCEDURES
Reagents-Unless otherwise specified, chemicals were obtained from Sigma Chemical, restriction and DNA modifying enzymes from New England Biolabs or Roche Applied Science, and oligonucleotides from Operon Technologies or Invitrogen. Nfo, Nth, and human PNKP (hPNKP) enzymes were as previously described (29).
Bacterial and Yeast Strains-Bacterial strains were BW528 (⌬(xthpnc), nfo1::kan) and its parent AB1157, kindly provided by Dr. B. Weiss (Emory, Atlanta, GA). Yeast strains YW636 (SOD1) and YW1094 (sod1) were created by the Yeast Genome Deletion Project (32) and were obtained from Research Genetics. All other yeast strains were isogenic derivatives of YW465 (21); their genotypes are listed in Table I. Gene disruptions were created by one-step PCR-mediated gene replacement (33). The can1⌬::Myc-Fpg and can1⌬::Fpg chromosomal alleles, expressing Myc-Fpg and Fpg from the ADH1 promoter, respectively, were created by replacement of CAN1 with PCR products made from pMyc-Fpg and pFpg (see below). Transformants were selected by replica plating 1-day YPD plates to plates containing 50 g/ml canavanine. All gene replacements were verified by PCR; can1⌬::Fpg was also sequenced. Standard techniques of mating and tetrad dissection were used to create the further strain derivatives used in these studies.
Media and Growth Conditions-Bacteria were grown in Luria broth at 37°C unless otherwise indicated. Yeast were grown at 30°C, with shaking at 280 rpm for liquid cultures. Routine yeast media were either YPD (1% yeast extract, 2% peptone, 2% dextrose, 40 g/ml adenine) or synthetic defined medium (1.7 g/liter yeast nitrogen base (Difco), 5 g/liter ammonium sulfate, 2% dextrose) supplemented with the appropriate complete synthetic medium dropout mix (QBiogene; adenine concentration was 10 g/ml for plates and 40 g/ml for liquid cultures). When indicated, 2% ethanol plus 3% glycerol was substituted for dextrose (YPEG). Anaerobic growth was in sealed jars containing an AnaeroGen anaerobic pack (Camp Micro Inc.) on YPD supplemented with 0.5% Tween 80 and 20 g/ml ergosterol (34). When indicated, 25 mM ascorbic acid or 25 mM N-acetylcysteine were added as free radical scavengers.
Fpg and Nth Expression Plasmids-pMyc-Fpg and pMyc-Nth, which express the bacterial fpg and nth genes in yeast with single aminoterminal Myc tags, respectively, were made by gap repair cloning into pTW367 as previously described (forward primer, 5Ј-ACCATGGCGTC-CGAGCAAAAGCTCATTTCTGAAGAGGACTTGCGC; reverse primer, 5Ј-TTTATGTAACGTTATAGATATGAAGGATTTCATTCGTCTGTCG-AC) (36). pFpg, the pTW367 derivative that expresses untagged Fpg, was cloned similarly except that a different forward primer was used whose 5Ј-tail caused deletion of the Myc tag contained in the vector (5Ј-CAATATTTCAAGCTATACCAAGCATACAATAAGCTTCTCACC). Yeast strains bearing successful gap repair constructs were verified by PCR. In all cases, multiple PCR-positive transformants were tested for MMS sensitivity.
H 2 O 2 and MMS Sensitivity-Bacteria were transformed with the appropriate plasmid by the CaCl 2 method. Cultures of BW528 strains bearing the respective plasmids were grown in LB medium supplemented with 100 g/ml ampicillin. Overnight cultures were diluted 1:10 in the same medium, grown for 2.5 h at 30°C, and then the absorbance at 600 nm (OD 600 ) was adjusted to 0.9. The cultures were grown for a further 10 min before adding drugs. H 2 O 2 was added to a final concentration of 0.1%, samples were drawn at 0, 5, 10, and 15 min, serially diluted to 1 ϫ 10 Ϫ5 or 1 ϫ 10 Ϫ6 , and plated onto LB agar plates For determination of yeast MMS sensitivity, exponentially growing cells were exposed to varying concentrations of MMS in 50 mM potassium phosphate, pH 7.5, at 30°C for 30 min as previously described (25). Survival was determined by plating serial dilutions to YPD (or appropriate dropout plates in the case of yeast strains with plasmids), with colonies counted after 3 days growth at 30°C.
Primer Extension Assay-Exponentially growing cells in 5 ml of LB were either untreated or treated with 25 mM H 2 O 2 for 1 h at 37°C. Cells were harvested, washed twice with M9 buffer, and the cell pellet stored frozen for 1 h at Ϫ80°C. Extraction of the chromosomal DNA was performed as previously described (37). To measure the incorporation of Tubes were mixed and placed on ice for 10 min. The samples were processed on a 12-hole filtration apparatus (Millipore, Bedford, MA) using GF/C circle filters (Whatman). The trapped DNA was washed three times with 3 ml of 0.1 M sodium pyrophosphate, briefly rinsed with ethanol, air-dried, and counted with 5 ml of scintillation fluid (BCS, Amersham Biosciences). In the case of chromosomal DNA pretreated with either Nfo or hPNKP, 20 ng of the purified enzyme was incubated with the DNA for 20 min at 37°C in 10 l of reaction buffer (25 mM Hepes-KOH, pH 7.6; 50 mM KCL, 5 mM MgCl 2 , 1 mg/ml bovine serum albumin). The enzyme was heat-inactivated at 70°C for 3 min.
Enzyme Activity Assays-GST-Tpp1 and GST-Apn1 were purified from S. cerevisiae as previously described (24). Duplex oligonucleotides bearing a uracil, or a 3Ј-phosphate at a nick, were labeled and annealed as previously described (24,25). The uracil was converted to a singlebase gap with a 3Ј-aldehyde by treatment with uracil-DNA glycosylase (Udg, 0.01 units/5 pmol oligonucleotide) and Nth (28 ng/pmol oligonucleotide) in Udg buffer (NEB) for 1 h at 37°C. Activity assays contained 50 fmol of DNA substrate and appropriate amounts of enzyme in a reaction volume of 10 l such that the final buffer was 50 mM Tris-HCl (pH 7.5), 100 mM NaCl, 10 mM MgCl 2 , 1 mM dithiothreitol, 50 g/ml bovine serum albumin. Following incubation at 30°C for 10 min, the reaction was stopped with formamide/EDTA loading buffer, and samples were electrophoresed on 7 M urea, 12% polyacrylamide gels followed by autoradiography.

Tpp1 Can Participate in Repair of H 2 O 2 -and MMS-induced
Lesions When Expressed in Bacteria-E. coli strain BW528 bears mutations in both the nfo and xth abasic endonuclease genes and is consequently hypersensitive to agents that induce abasic sites or strand breaks. Because this strain has proven useful for the study of a number of heterologously expressed BER proteins (29,30,37), we examined whether a GST-Tpp1 fusion protein could complement the BW528 mutations. In our first assay, the bacterial chromosome is damaged in vivo with H 2 O 2 and subsequently isolated and used as a template for primer extension by DNA polymerase I in vitro (37). Only if 3Ј blocking lesions are removed at H 2 O 2 -induced strand breaks can [ 3 H]dTMP be incorporated. Thus, DNA isolated from the treated wild-type strain AB1157 supported primer extension without pretreatment ( Fig. 2A), while DNA isolated from BW528 supported extension only when it was pretreated with enzymes such as Nfo (data not shown, but see Tpp1-D35A mutation, Fig. 2C and below). In contrast, DNA from BW528 expressing GST-Tpp1 supported substantial primer extension without pretreatment (Fig. 2B). The amount of [ 3 H]dTMP incorporation could be increased to a minor extent by pretreatment of this DNA with Nfo, but not by pretreatment with hPNKP. Thus, GST-Tpp1 expressed in vivo was able to remove most, but not all, H 2 O 2 -induced 3Ј blocking lesions. Since H 2 O 2 leaves principally, but not exclusively, 3Ј phosphates (38, 39) (see Fig. 1), these data support the interpretation that the activities of Tpp1 and hPNKP are similar and restricted to 3Ј-phosphates.
Examining the survival of bacteria treated with H 2 O 2 gave similar results, in that GST-Tpp1 could largely, but not completely, rescue survival of BW528 as compared with wild type (Fig. 3A). This result parallels previous experiments in yeast showing that Tpp1 confers H 2 O 2 resistance (25). We were initially surprised to see that GST-Tpp1 could also confer partial BW528 resistance to MMS, however (Fig. 3B). This was unexpected since MMS leads principally to the formation of abasic sites through the action of DNA glycosylases (see Fig. 1), not 3Ј-phosphates. Indeed, apn1 apn2 tpp1 triple mutant yeast are no more sensitive to MMS than apn1 apn2 double mutant yeast (25). In the following experiments we sought to determine the basis of the differing effects of Tpp1 expression on MMS sensitivity in yeast and bacteria. Overlapping Function of Tpp1 and Rad1/Rad10 at 3Ј-Phosphates Tpp1 Cannot Remove 3Ј-Aldehydes-One explanation for the bacterial result would be that Tpp1 in fact has a 3Ј-phosphodiesterase activity, beyond 3Ј-phosphatase, that is able to cleave a subset of blocking lesions created by MMS treatment. We have previously shown that Tpp1 can not cleave abasic sites or 3Ј-phosphoglycolates, nor is it an exonuclease (25). An untested high frequency 3Ј-blocking lesion was the ␣,␤-unsaturated aldehyde resulting from ␤-elimination at a deoxyribose. As shown in Fig. 4, GST-Tpp1, unlike GST-Apn1, showed no activity at this lesion, even at enzyme concentrations 40-fold higher than those required to completely cleave a similar amount of 3Ј-phosphates. It is therefore highly likely that the beneficial effect of Tpp1 in bacteria is explainable purely by its 3Ј-phosphatase activity.
Fpg, but Not Nth, Confers Tpp1-dependent Phenotypes in apn1 apn2 Yeast-Our preferred hypothesis for the unexpected rescue of BW528 MMS resistance by GST-Tpp1 relates to the secondary metabolism of abasic sites (see Fig. 1). Specifically, action of the bacterial ␦-lyases Fpg and Nei at MMS-induced abasic sites would produce 3Ј-phosphates that are a suitable substrate for GST-Tpp1. In contrast, yeast are not known to express a ␦-lyase, although the recent unanticipated demonstration of Nei-related enzymes in humans makes this claim uncertain (13,14). We reasoned that if the difference between the bacterial and yeast MMS experiments was indeed a result of ␦-lyase expression, then yeast should become more like bacteria if a ␦-lyase was expressed ectopically.
Our goal was to express the E. coli fpg gene in S. cerevisiae. This experiment was complicated by the fact that the aminoterminal proline of Fpg acts as the Schiff base nucleophile during the sequential ␤and ␦-elimination reactions (40). Consequently, even a two residue (Gly-Met) N-terminal extension impairs Fpg activity by more than 100-fold (41). The extent to which the obligatory initiator methionine would be removed in yeast to leave an N-terminal proline could not be predicted, but fpg is known to complement ogg1 mutation in S. cerevisiae, indicating that some function is retained (42). We attempted to express Fpg both with and without an amino-terminal Myc tag. Fig. 5A and Table II show the interesting result that only very poorly growing colonies were obtained when untagged Fpg was expressed in apn1 apn2 tpp1 yeast. In contrast, Fpg was entirely tolerated by apn1 apn2 yeast. Although these results largely precluded meaningful drug sensitivity testing with untagged Fpg, they already demonstrated that Fpg could confer a Tpp1-dependent phenotype in yeast, presumably by cleaving spontaneously arising DNA lesions (see more below).
Myc-Fpg could be expressed in any yeast strain (Fig. 5A), indicating that it is less active than untagged Fpg, as expected. We reasoned that Myc-Fpg might nonetheless retain sufficient lyase activity to satisfy our initial goal of conferring Tpp1-dependent MMS resistance on yeast. This was indeed the case (Fig.  5B). apn1 apn2 and apn1 apn2 tpp1 yeast were themselves highly and equivalently sensitive to MMS, as previously demonstrated (25). Expression of Myc-Fpg in the apn1 apn2 background led to a measurable increase in resistance, although well below wild-type even at the low (0.01-0.1%) MMS concentrations used (compare with the apn2 strain in Fig. 5B). This effect was entirely dependent on Tpp1, the only remaining 3Ј phosphatase in the cells, demonstrating that it was due to Myc-Fpg-dependent ␦-elimination. In fact, Myc-Fpg dramatically increased the MMS sensitivity of apn1 apn2 tpp1 yeast, such that cultures were effectively sterilized by a 30 min treatment with the extremely low MMS concentration of 0.025%. Tpp1-dependent MMS resistance was also seen when Myc-Nei was heterologously expressed. 2 The implication of the above results is that Myc-Fpg was able Synthetic oligonucleotides bearing a nick with a 3Ј-phosphate (P) or an internal uracil (U) were prepared by radiolabeling with 32 P (asterisk) and annealing. The U was subsequently converted to a nick with a 3Ј ␣,␤-unsaturated aldehyde (CHO) by combined treatment with Udg and Nth. The top and bottom of an autoradiogram of a sequencing gel is shown; the middle portion with no bands has been removed for space. The identity of the 22-mer aldehyde product, which migrates as a closely-spaced doublet, is verified by the requirement for both Udg and Nth. Subsequent incubation with varying amounts of purified enzymes for 10 min at 30°C showed that GST-Apn1 (380, 1300, and 3600 fmol) but not GST-Tpp1 (10, 20, 40, and 400 fmol) could remove the aldehyde. For comparison, 10 fmol of GST-Tpp1 and 380 fmol of GST-Apn1 were shown in the same experiment to completely remove a 3Ј-phosphate. GST-Apn1 can additionally remove the 3Ј-nucleotide to form a 21-OH product as previously described (24). M indicates marker lanes with labeled synthetic product oligonucleotides.
to convert less toxic abasic lesions to strand breaks that became highly toxic in the absence of enzymes able to remove 3Јphosphate blocking lesions. This would predict that (over)expression of a ␤-only lyase would lead to an increase in MMS killing regardless of the presence of Tpp1, since Tpp1 cannot remove 3Ј-aldehydes (see above). Fig. 5C shows that Myc-Nth did indeed lead to an equivalent increase in MMS sensitivity in both apn1 apn2 and apn1 apn2 tpp1 yeast. A similar pattern was observed when the yeast glycosylases Ogg1, Ntg1, or Ntg2 were overexpressed as Myc fusion proteins from the strong constitutive ADH1 promoter. 2 These results provide clear in vivo corroboration of the lesion specificity of Tpp1.
Tpp1 Amino Acids Asp-35 and Asp-37 Are Essential for Its in Vivo Function-In addition to conferring H 2 O 2 and MMS resistance in bacteria (Fig. 3), plasmid-expressed Tpp1 reduced MMS sensitivity of apn1 apn2 tpp1 Myc-Fpg yeast more than 1500-fold (Fig. 5D). To validate dependence of these assays on Tpp1 catalytic function, we examined mutations of two universally conserved Tpp1 aspartate residues, Asp-35 and Asp-37. Based on homology with the L-haloacid dehalogenase superfamily of hydrolases, these residues are believed to participate in catalysis by forming a covalent bond between the phosphateleaving group and the first aspartate (24). Mutation of either aspartate to alanine (D35A and D37A) completely abolished Tpp1-dependent repair of H 2 O 2 -induced lesions ( Fig. 2 and data not shown) and resistance to both H 2 O 2 and MMS in bacteria (Fig. 3), as well as resistance to MMS in Myc-Fpgexpressing yeast (Fig. 5D). GST-Tpp1-D35A and GST-Tpp1-D37A were also completely inactive in vitro. 2 Synthetic Growth Defects Caused by apn1, apn2, tpp1, and rad1 Mutant Combinations-We have previously seen that apn1 apn2 tpp1 yeast grow similarly to wild-type but die when the recombination gene RAD52 is deleted, and that apn1 tpp1 rad52 and apn1 apn2 rad52 yeast grow extremely poorly (25). This suggested a recombination-dependent mechanism for removal of 3Ј blocking lesions. To explore this hypothesis further, we performed analogous tetrad dissection experiments with the RAD1 gene, which encodes part of the Rad1/Rad10 endonuclease that acts to trim 3Ј-ends during recombination (20). As shown in Fig. 6, apn1 tpp1 rad1 strains grew very poorly, similar to apn1 tpp1 rad52. This effect was not due to the NER deficiency of rad1 mutants (19) since rad2 and rad14 NER mutants did not share the tpp1 synthetic phenotype. 2 Strikingly, apn1 apn2 rad1 strains showed an even more severe phenotype of synthetic lethality, similar to a recent report (23). Microscopic examination of inferred apn1 apn2 rad1 microcolonies showed them to be typically comprised of 10 to ϳ50 large, budded, and misshapen cells (Table II and data not shown), characteristic of persistent irreparable DNA damage. When tpp1 mutation was added this phenotype became even more severe, such that apn1 apn2 tpp1 rad1 cells rarely formed microcolonies of more than 10 cells. The collected results demonstrate that while the synthetic interaction with rad1 mutation is most severe for apn1 apn2 mutation, the phenotype extends to tpp1 mutation, which is informative because Tpp1 is a pure DNA 3Ј-phosphatase.
Partial Suppression of the apn1 apn2 rad1 Growth Defect by Fpg-The most likely explanation for apn1 apn2 rad1 synthetic lethality is that spontaneously arising abasic sites either persist or are converted to strand breaks with 3Ј-aldehyde blocking lesions by ␤-lyases (see "Discussion"). In the absence of a ␦-lyase or abasic endonuclease/3Ј-phosphodiesterase, these lesions would persist and require Rad1/Rad10 for removal. A prediction of this model is that ectopic expression of a ␦-lyase should rescue apn1 apn2 rad1 lethality in a Tpp1-dependent fashion. Indeed, apn1 apn2 rad1 yeast expressing untagged Fpg did form visible colonies, but not when TPP1 was also mutated ( Table II). The rescue was partial, however, since these colonies were very small and failed to continue growing. This may be because only a small portion of the Fpg expressed in yeast is made competent for ␦-elimination by removal of the amino-terminal methionine. Alternatively, apn1 apn2 rad1 lethality may depend on functions in addition to processing of abasic sites and 3Ј-aldehydes.
Suppression of apn1 tpp1 rad1 Synthetic Slow Growth-Extending the above logic, endogenous DNA damage must also explain the tpp1 rad1 synthetic interaction, and this damage must include 3Ј-phosphates. There are three known mechanisms by which 3Ј-phosphates are generated in growing cells. First is through the action of ␦-lyases as discussed, but the above experiments functionally demonstrate that S. cerevisiae does not possess a ␦-lyase (see "Discussion").
Second, we and others have described a pathway in which the enzyme Tdp1 removes the DNA topoisomerase Top1 when the latter is irreversibly and covalently bound to a DNA 3Јterminus (21,25,43,44). Tdp1 leaves a 3Ј-phosphate (43) that is removed by the combined action of Tpp1, Apn1, and Apn2 (see Fig. 1) (25). The slow growth phenotype of tdp1 rad1 yeast demonstrates that the Tdp1 pathway is active to some extent even in the absence of Top1 inhibitors (21). This tdp1 rad1 slow growth is much less severe than even the apn1 tpp1 rad1 phenotype, however, making it unlikely that the Top1 pathway contributes substantially to the tpp1 rad1 synthetic interaction. Indeed, mutation of neither TDP1 nor TOP1 led to a measurable suppression of the apn1 tpp1 rad1 growth defect (Table II).
Third, endogenous reactive oxygen species can lead directly to the formation of strand breaks with 3Ј-phosphates in the same manner as exogenous H 2 O 2 . We therefore compared the growth of apn1 tpp1 rad1 and apn1 tpp1 yeast in aerobically and anaerobically grown streaks (Fig. 7). The profound growth difference of these strains was preserved on aerobic re-streak- a The size of the colonies formed after growth of tetrad dissection plates was determined prior to genotyping according to the following scoring system: 5, Wild-type colony growth (e.g. Fig. 6, A3). 4, Intermediate between 5 and 3 (e.g. Fig. 6, B3). 3, Macrocolony visible at 3 days growth (e.g. Fig. 6, A5). 2, Microcolony (not macroscopically visible at 5 days), Ͼ10 cells. 1, Microcolony, 2-10 cells. 0, Single cell, i.e. no germination (not tabulated).
b These colonies were macroscopically visible at 3 days, but were consistently smaller than either apn1 tpp1 rad1 or can1::Fpg apn1 apn2 tpp1 yeast, and failed to continue growing with prolonged incubation time. 6. Synthetic interactions between apn1, apn2, tpp1, and rad1 mutation. The diploid strain YW851 (APN1/apn1 APN2/apn2 TPP1/tpp1 RAD1/rad1) was sporulated and dissected. Spore genotypes were determined by replica plating and are indicated in the table. Genotypes of inviable spores were inferred from the segregation patterns. The two inviable spores in the fourth tetrad are both apn1 apn2 rad1 mutants, however only one spore additionally carries the tpp1 mutation. Since this genotype cannot be assigned unambiguously, they are listed in the table as A4 or D4.

FIG.
ing, i.e. small spore colonies were not an artifact of slow germination. In contrast, anaerobic growth tended to normalize the appearance of the two strains. This initially suggested that endogenous oxidation is a significant contributor to apn1 tpp1 rad1 slow growth. However, when we attempted to verify this phenotype by additional rounds of restreaking, we found that cells in anaerobically grown colonies were nearly all dead, regardless of genotype and despite the inclusion of lipid nutrients essential for anaerobically grown S. cerevisiae (34). We therefore sought to corroborate the anaerobic findings by independent means, but without success. Growth in a reduced O 2 environment in CampyPaks (BBL) relieved the cell death observed in anaerobic chambers, but did nothing to suppress the apn1 tpp1 rad1 growth defect. 2 Furthermore, apn1 tpp1 rad1 growth was not improved by growth in the presence of the ROS scavengers ascorbic acid (a superoxide scavenger, Ref. 45) and N-acetylcysteine (a hydroxyl radical scavenger, Ref. 46) (Fig.  7). The slow growth of an ROS-sensitive superoxide dismutase mutant strain (sod1) was normalized on these same plates, demonstrating the efficacy of the scavenger treatments.
The majority of endogenous ROS are believed to arise through mitochondrial generation of superoxide during respiration, which is ultimately converted to the DNA-damaging hydroxyl radical (1,7,47). We therefore reasoned that increasing or decreasing mitochondrial respiration should exacerbate or alleviate apn1 tpp1 rad1 slow growth, respectively. This again did not prove to be the case. Disruption of the mitochondrial electron transport chain at an early step by mutation of the nuclear COQ3 gene blocks respiration, and so suppresses stationary-phase death of sod1 mutants (7). No such suppression of apn1 tpp1 rad1 slow growth was observed; all coq3 apn1 tpp1 rad1 spore colonies remained slow growing (colony size 3, Table II). In the complementary approach, growth on plates containing the non-fermentable carbon sources ethanol and glycerol forces respiratory metabolism. Remarkably, ethanolglycerol medium did not exacerbate apn1 tpp1 rad1 slow growth, but in fact completely suppressed it (Fig. 7). This phenotype was reversible because apn1 tpp1 rad1 colonies streaked from ethanol-glycerol back to dextrose once again showed severely impaired growth. 2 Finally, we observed spontaneously arising slow-growth suppressors in aerobic streaks of apn1 tpp1 rad1 yeast, but none of 18 tested proved to be respiratory (i.e. petite) mutants. The implications of these findings are addressed in "Discussion."

DISCUSSION
The Tpp1 Family of Enzymes is 3Ј-Phosphate-specific-It has previously been demonstrated that both human and plant DNA 3Ј-phosphatases are able to restore DNA damage resistance in bacteria lacking the abasic endonucleases Nfo and Xth (29,30). We have found this to be true for Tpp1 as well (Fig. 3). There are numerous implications of this finding. First, Tpp1 can clearly act at 3Ј-phosphate lesions in vivo independently of other yeast proteins. Further, the extent to which Tpp1 could restore polymerase priming capacity to H 2 O 2 -damaged DNA (Fig. 2) draws a parallel to hPNKP function. H 2 O 2 -induced damage is characterized predominantly by strand breaks with 3Ј-phosphates (38,39). Tpp1 and hPNKP each facilitated extension by DNA polymerase I by removing these phosphates, but neither completely restored priming capacity as compared with the multifunctional 3Ј-phosphodiesterase Nfo. This indicates that Nfo, but not Tpp1 or hPNKP, is able to cleave the minority of non-phosphate H 2 O 2 lesions as well. Supporting this interpretation, we have demonstrated that Tpp1 is inactive at a wide variety of non-phosphate strand lesions, including abasic sites, simple nicks, 3Ј-phosphoglycolates (25), and 3Ј-aldehydes (Fig. 4). We have seen a similar pattern with the S. pombe PNKP homologue (27). DNA 3Ј-phosphatases from yeast and human thus appear to have a similar and considerably more restricted function than the abasic endonucleases. This is consistent with the inferred catalytic mechanism of this enzyme family (24), which is supported by our findings that Tpp1 D35A and D37A mutants are catalytically inactive (Figs. 2, 3, and 5).
␦-Elimination Is Not Required for Genome Maintenance in S. cerevisiae-It was initially surprising that Tpp1 conferred bacterial resistance to MMS as well H 2 O 2 . This finding was best explained by the presence of lyases in bacteria that can create 3Ј-phosphates from MMS-induced abasic sites by consecutive ␤/␦ elimination (Fig. 1). We validated this hypothesis by demonstrating that even the crippled Myc-Fpg led to a clear and predictable Tpp1-dependent MMS resistance when expressed in yeast (Fig. 5). The complete lack of Tpp1-dependent MMS resistance in the parent strains thus demonstrates functionally that S. cerevisiae does not possess a detectable ␦ lyase. This conclusion is supported by the fact that the characteristic structural features of the Fpg/Nei glycosylase/lyase family are not shared by any hypothetical protein in S. cerevisiae, as determined by BLAST searches or by considering all open reading frames with a proline as the second residue (14). 2 It is thus apparent that yeast ordinarily maintain their genomes without the benefit of ␦-elimination, and must instead rely on the abasic endonucleases to remove 3Ј-aldehydes left by chemical or enzymatic ␤-elimination. This is important in light of the recent unexpected discovery of functional homologues of Fpg/Nei in many eukaryotes, including humans and the fungus Candida albicans (13,14).
The paradox is this: why do bacteria possess ␦-lyases but no pure 3Ј-phosphatase, while yeast possess a 3Ј-phosphatase but no ␦-lyase? At a minimum, this paradox makes clear that Tpp1 FIG. 7. Suppression of the growth defect of apn1 tpp1 rad1 yeast by altering the growth conditions. Numerous large (apn1 tpp1, colony size 5; see Table II) and small (apn1 tpp1 rad1, colony size 3) macrocolonies were picked from tetrad dissection plates, streaked, and grown on the following media for the indicated number of days: YPD, 2 days; YPD plus 0.5% Tween 80 and 20 g/ml ergosterol, grown anaerobically for 3 days; YPD plus 25 mM ascorbic acid, 2 days; YPD plus 25 mM n-acetyl cysteine, 2 days; YP medium with ethanol-glycerol as the carbon source, 5 days. Representative streaks are shown. Isogenic SOD1 wild-type and sod1 mutant strains were streaked to these same media for comparison. 5-Fold lower concentrations of ascorbate and N-acetylcysteine gave equivalent results for all strains.
does not exist to cleave 3Ј-phosphate lesions left by ␦-elimination. Indeed, in bacteria this need is satisfied by Nfo and Xth. The question remains why Tpp1 is ever necessary, or preferred, in yeast cells that already possess Apn1 and Apn2.
A General Rad1/Rad10-dependent Salvage Pathway for Repair of 3Ј-Blocked Strand Breaks-Our results support the conclusion that Tpp1 participates in the repair of endogenous DNA damage, but in a fashion redundant with other 3Ј repair mechanisms. This was not initially obvious because even apn1 apn2 tpp1 cells grow at a normal rate. However, the addition of either rad52 (25) or rad1 (Fig. 6, Table II) mutation leads to lethality in this background, suggesting that Rad52 and Rad1/ Rad10 act in a redundant pathway that repairs persistent strand breaks. These results parallel similar recent findings. For example, Rad1/Rad10 participates in repair of strand breaks when Top1 peptide fragments are covalently bound to the 3Ј terminus, but only in the absence of Tdp1, the enzyme that normally removes these peptides (21,22). In addition, the apn1 apn2 rad1 mutation combination is lethal due to absence of Rad1/Rad10-dependent processing of 3Ј-aldehyde lesions that persist in apn1 apn2 cells (23). Analysis of tpp1 phenotypes is necessarily more complex, because depriving cells of 3Ј-phosphatase function requires apn1 and apn2 mutation and thus concomitant loss of abasic endonuclease/phosphodiesterase activity. Nonetheless, tpp1 mutation causes an incremental growth defect in both apn1 rad1 and apn1 apn2 rad1 backgrounds ( Fig. 6 and Table II). The lesion specificity of Tpp1 dictates that this effect must be due to failed repair of endogenously generated 3Ј-phosphates, and in turn that 3Ј-phosphates can be repaired by the Rad1/Rad10 pathway. The capacity of this pathway is clearly limited, however, since it cannot protect apn1 apn2 tpp1 cells from higher levels of H 2 O 2induced damage (25). Taken together, we argue that Rad1/ Rad10-dependent repair of 3Ј-phosphates is truly a salvage pathway not normally active in cells expressing Tpp1. Whether there is any normal function of this Rad1/Rad10 pathway of strand break repair remains to be determined.
Sources of Endogenously Generated 3Ј-Phosphates-Surprisingly, there appears to be little correlation between the endogenous mitochondrial ROS burden and the apn1 tpp1 rad1 growth defect. Anaerobic growth did normalize the growth of apn1 tpp1 rad1 strains (Fig. 7), but reducing mitochondrial ROS production by coq3 mutation did not (Table II). Instead, attempting to increase mitochondrial ROS production by growth on non-fermentable carbon sources antithetically suppressed apn1 tpp1 rad1 slow growth (Fig. 7). It is possible that the induction of the oxidative stress response that occurs during respiratory growth (48) more than compensates for the increased oxidative load. Exogenous ROS scavengers were ineffective in suppressing apn1 tpp1 rad1 slow growth, however, arguing against this hypothesis (Fig. 7).
A different explanation would be that the anaerobic environment and non-fermentable carbon sources each suppress the apn1 tpp1 rad1 growth defect independently of oxidation. Slow growth might simply allow more time for repair. Alternatively, faster replication and/or growth on glucose might itself result in increased production of 3Ј-phosphates. Given that S. cerevisiae does not possess a ␦-lyase, the enzyme Tdp1 is the only remaining known non-oxidative source of endogenous 3Ј-phosphates in this organism (43). However, while Top1 damage might be exacerbated by rapid replication, the Top1-Tdp1 pathway apparently does not lead to enough basal damage to account for the apn1 tpp1 rad1 slow growth phenotype (Table II). We must therefore consider the possibility that there is an as yet unidentified intracellular source of 3Ј-phosphates. This would most likely be an enzyme, perhaps one under carbon source regulation. ␦-Lyases and Tdp1 give clear precedent for the notion that a DNA metabolic enzyme might leave a 3Јphosphate in need of further repair.
Persistent Strand Breaks Are Especially Cytotoxic-Xiao et al. (11) have demonstrated that metabolism of modified bases to abasic sites is necessary to achieve cell killing and mutagenesis by agents such as MMS. A consistent implication of our studies is that it is secondary strand breaks, rather than abasic sites themselves, that are most cytotoxic. Guillet et al. (23) recently came to a similar conclusion. They demonstrated that the lethality of apn1 apn2 rad1 yeast was suppressed by the combined mutation of the ␤-only lyases expressed in S. cerevisiae, OGG1, NTG1, and NTG2, which suggested that it is 3Ј-aldehydes that are least well tolerated. We have extensively employed the complementary approach of expressing glycosylase/lyases to interconvert lesions to predictable endpoints. Overexpression of ␤-lyases, including Nth, Ogg1, Ntg1, and Ntg2, enhanced the already substantial MMS sensitivity of apn1 apn2 yeast (Fig. 5). 2 In contrast, even Fpg weakened by an amino-terminal fusion was able to reduce this MMS sensitivity, presumably by converting toxic 3Ј-aldehydes into Tpp1reparable 3Ј-phosphates (Fig. 5). Indeed, Myc-Fpg-expressing apn1 apn2 tpp1 yeast were exquisitely MMS hypersensitive. Similarly, untagged Fpg was able to partially suppress apn1 apn2 rad1 lethality (Table II), but conversion of even endogenous levels of base lesions to 3Ј-phosphates by Fpg overwhelmed the capacity of Rad1/Rad10 in the absence of 3Јphosphatases ( Fig. 5 and Table II).
We conclude that it is not strand breaks themselves that are toxic, but rather their persistence due to irreparable 3Ј blocking lesions. Replication would thus lead to the formation of doublestrand breaks still bearing the blocking lesion, as we have modeled in detail for camptothecin-induced damage (21). Interestingly, this overall pattern is apparently not restricted to 3Ј blocking lesions. It is the lyase activity of mammalian DNA polymerase ␤, which removes 5Ј-dRP blocking lesions, that is most critical during BER (49). In lyase-deficient mutants, 5Јblocked strand breaks persist and are thought to lead to a similar catastrophe during replication.