Efficient Processing of DNA Ends during Yeast Nonhomologous End Joining

Repair of DNA double strand breaks by nonhomologous end joining (NHEJ) requires enzymatic processing beyond simple ligation when the terminal bases are damaged or not fully compatible. We transformed yeast with a series of linearized plasmids to examine the role of Pol4 (Pol IV, DNA polymerase β) in repair at a variety of end configurations. Mutation of POL4 did not impair DNA polymerase-independent religation of fully compatible ends and led to at most a 2-fold reduction in the frequency of joins that require only DNA polymerization. In contrast, the frequency of joins that also required removal of a 5′- or 3′-terminal mismatch was markedly reduced in pol4 (but not rev3,exo1, apn1, or rad1) yeast. In a chromosomal double strand break assay, pol4 mutation conferred a marked increase in sensitivity to HO endonuclease in arad52 background, due primarily to loss of an NHEJ event that anneals with a 3′-terminal mismatch. The NHEJ activity of Pol4 was dependent on its nucleotidyl transferase function, as well as its unique amino terminus. Paradoxically, in vitro analyses with oligonucleotide substrates demonstrated that although Pol4 fills gaps with displacement of mismatched but not matched 5′ termini, it lacks both 5′- and 3′-terminal nuclease activities. Pol4 is thus specifically recruited to perform gap-filling in an NHEJ pathway that must also involve as yet unidentified nucleases.

Repair of DNA double strand breaks by nonhomologous end joining (NHEJ) requires enzymatic processing beyond simple ligation when the terminal bases are damaged or not fully compatible. We transformed yeast with a series of linearized plasmids to examine the role of Pol4 (Pol IV, DNA polymerase ␤) in repair at a variety of end configurations. Mutation of POL4 did not impair DNA polymerase-independent religation of fully compatible ends and led to at most a 2-fold reduction in the frequency of joins that require only DNA polymerization. In contrast, the frequency of joins that also required removal of a 5-or 3-terminal mismatch was markedly reduced in pol4 (but not rev3, exo1, apn1, or rad1) yeast. In a chromosomal double strand break assay, pol4 mutation conferred a marked increase in sensitivity to HO endonuclease in a rad52 background, due primarily to loss of an NHEJ event that anneals with a 3-terminal mismatch. The NHEJ activity of Pol4 was dependent on its nucleotidyl transferase function, as well as its unique amino terminus. Paradoxically, in vitro analyses with oligonucleotide substrates demonstrated that although Pol4 fills gaps with displacement of mismatched but not matched 5 termini, it lacks both 5-and 3-terminal nuclease activities. Pol4 is thus specifically recruited to perform gap-filling in an NHEJ pathway that must also involve as yet unidentified nucleases.
Eukaryotes have evolved two distinct enzymatic pathways of DNA double strand break repair (DSBR) 1 to maintain genomic integrity (1)(2)(3)(4)(5)(6): recombinational DSBR (rDSBR) and nonhomologous end joining (NHEJ). Unlike rDSBR, NHEJ proceeds in a template-independent fashion, with simple rejoining of broken ends. For this reason, NHEJ is more dependent on the DNA ends themselves, because many DSBs will have incom-patible or damaged bases that could potentially block joining. Although it is well established that the NHEJ apparatus can resolve such obstacles through mechanisms such as terminal microhomology usage, the molecular basis of this ability is not understood.
Early steps in the NHEJ reaction mechanism presumably involve formation of a stable joining complex that tolerates base incompatibilities. The Ku 70/86 heterodimer is implicated in this by virtue of its end binding function (7)(8)(9), as is the complex of Sir2-Sir3-Sir4 via interaction with Hdf1 (Ku70) in yeast (10). The Rad50-Mre11-Xrs2 complex is also required for NHEJ in yeast, probably at an early stage, but its precise role is unclear (8,11,12). In mammalian cells, the DNA-dependent protein kinase and poly(ADP-ribose) polymerase may serve a structural role, transmit cell cycle signals, activate the joining machinery, or a combination of these (13,14). The most certain functional assignment is DNA ligation, which is clearly mediated by DNA ligase IV in both mammalian and yeast cells (15)(16)(17)(18)(19), as part of a complex that includes at least the XRCC4 protein or its yeast homologue Lif1 (20 -22). It is not clear whether XRCC4/DNA ligase IV is also involved in assembly of the joining complex, however.
Least is known about the proteins that contribute processing activities during NHEJ (i.e. polymerization and nucleolysis). Mre11 has a 3Ј-5Ј nuclease activity that can promote joining of ends in vitro, but the contribution of this activity to NHEJ in vivo remains to be established (12,(23)(24)(25)(26). Six nuclear DNA polymerases have been described in eukaryotic cells (27)(28)(29)(30). DNA polymerase (Pol) ␣ (designated Pol I in yeast, where POL1 is the gene for the polymerase subunit), Pol ⑀ (Pol II, POL2), and Pol ␦ (Pol III, POL3) together catalyze the essential functions of DNA replication. Pol ␦ and Pol ⑀ are also involved in certain DNA repair events, notably nucleotide excision repair. Pol (REV3) and Pol (RAD30, 31) mediate different forms of translesion bypass synthesis in yeast. Pol ␤ (Pol IV, POL4) is a 39-kDa monomeric polymerase in vertebrates that mediates base excision repair (BER) (32)(33)(34). Of these polymerases, Pol ␤ has features that might suggest its involvement in NHEJ as well as BER, including low processivity and a preference for short strand gaps. Yeast Pol4 shares these biochemical properties, but has an additional 30 kDa of sequence of undetermined function and is not required for BER (35)(36)(37). Rather, pol4 yeast exhibit reduced spore viability with abnormally high levels of intragenic meiotic recombination and persistent meiotic DSBs (36,38). These phenotypes might be explained by impaired NHEJ.
We have previously used a color-based plasmid transformation assay to document that the yeast NHEJ pathway can create processed joins at compatible restriction site ends (17).
Here, we extend the versatility of our assay to examine in detail the handling of more complex end configurations, with an emphasis on defining the role of Pol4 in NHEJ, if any. We find that like mammalian cells, yeast can efficiently execute remarkably complex NHEJ reactions, making extensive use of even very limited base pairing (microhomologies) between protruding single-stranded ends. Pol4 is indeed involved in this, a finding confirmed using chromosomal cutting assays. Surprisingly, although nucleotidyl transfer is essential to Pol4's NHEJ function, the pol4 phenotype cannot be explained by this activity alone. Rather, Pol4 is stringently required only for joins that necessitate removal of 5Ј-or 3Ј-terminal mismatches. Despite this, in vitro studies using gapped oligonucleotide substrates confirm the absence of nuclease activities in Pol4. We discuss the implications of these potentially paradoxical findings in the context of the multiple pathways of NHEJ.

EXPERIMENTAL PROCEDURES
Reagents-Except as noted, components for yeast media were from United States Biologicals, chemical reagents were from Sigma, nucleotides were from Amersham Pharmacia Biotech, DNA modifying enzymes were from New England Biolabs, and oligonucleotides were synthesized by either the Protein and Nucleic Acid Chemistry Laboratory or Jeffrey Milbrandt laboratory at the Washington University School of Medicine.
Construction and Maintenance of Yeast Strains-The genotypes of all strains are shown in Table I. All are haploid and closely related to the strain YW112, a derivative of YPH499 (17,39). Complete deletions of the POL4, REV3, APN1, and RAD1 coding sequences were made by PCR-mediated HIS3 gene replacement as described (40). apn1⌬::HIS3 and rad1⌬::HIS3 alleles were additionally verified by documenting increased sensitivity to methylmethane sulfonate and ultraviolet radiation, respectively. The rad52/pol4 double mutant strain YW153 was made from YW144 and YW130 by mating and sporulation. Yeast medium was either YPD, or minimal medium (41) supplemented with the appropriate nutrient dropout mix (BIO101, adenine concentration ϭ 10 g/ml) and carbon source (2% glucose or 2% raffinose plus 2% galactose). Adenine was added to YPD and liquid cultures in minimal medium to 40 g/ml. Yeast were grown at 30°C for all experiments.
Construction of Plasmids-pES16 and pES26 were described previously (17). New derivatives were constructed using a three-way ligation strategy. Upstream PCR products were made with a common forward primer that annealed at the beginning of the ADE2 promoter and that was tailed with a NotI site (OW632, 5Ј-GAGGCGGCCGCGGATTCAT-GCTTATGGGTTAG) and specifically tailed reverse primers that spanned the start codon (common sequence 5Ј-ATCCATACTTGATTGTT-TTGTCCG). Downstream PCR products were made with specifically tailed forward primers that annealed just after the start codon (common sequence 5Ј-TCTAGAACAGTTGGTATATTAGG) and a common reverse primer that spanned the BglII site in the ADE2 gene (OW633, 5Ј-CCGTTAACAGATCTCACAATC). These were digested with a restriction enzyme common to both specific primers and either NotI or BglII, as appropriate, allowing simultaneous ligation into NotI-Bg-lII-digested vector. Plasmids pM1 and pMB0 were constructed using pES16 as the vector backbone, noting that the polylinker BamHI site was deleted by the preparative NotI-BglII digestion. pXB2 was constructed using a pES16 derivative in which the polylinker XhoI and SalI sites had been fused and destroyed (pTW287). pSK3 was constructed using a pES16 derivative in which the polylinker KpnI site had been destroyed by digestion with the 5Ј isoschizomer Asp-718 followed by end-filling and religation (pTW288). The sequences inserted between the 2nd and 3rd codons of the ADE2 gene were (restriction sites are in bold, excisable stop codons are underlined): pM1, 5Ј-AACGCGT; pXB2, 5Ј-AACTCGAGTAACTAGCTGACGGATCC; pMB0, 5Ј-ACGCGTTA-ACTAGCTGACGGATCC; pSK3, 5Ј-AAAGCATGCTAACTAGCTGAC-GGTACC.
Pol4 expression plasmids were derived from the URA3/CEN/ARS plasmid pTW268, which contains the CDC9 coding sequence fused to glutathione S-transferase under the control of the ADH1 promoter (17). First, the glutathione S-transferase coding sequence was replaced with a HindIII-BamHI-tailed PCR product made from pMS127b (42) that encoded a His 9 -Myc 3 epitope tag, to create pTW283. Next, the CDC9 coding sequence was replaced with a BamHI-SalI-tailed PCR fragment made from total Saccharomyces cerevisiae genomic DNA (Novagen) that encoded the Pol4 protein from the second to the last amino acid, to create pTW284. Unlike the published POL4 sequence, our POL4 PCR fragment did not contain an internal BamHI site (38), but rather had a silent mutation of codon 303 from GAT to GAC. To create the LEU2selectable derivative pTW285, the ADH1-HM-POL4-containing Asp-718-NotI fragment of pTW284 was ligated into SmaI-NotI-digested pRS315 (39) after blunting of the Asp-718 5Ј overhang. To create the LYS2-selectable derivative pTW286, the XhoI-NotI fragment of pTW285 was ligated into SalI-NotI-digested pRS317.
The plasmid expressing the ⌬1-61 amino-terminal deletion (pTW301) was created by replacing the BamHI-SalI fragment of pTW300 (pTW285 with the polylinker SalI destroyed) with a PCR fragment corresponding to the truncated Pol4 sequence. The D367E and K247R/K248R point mutations were created by a gap repair strategy. First, a three-way BamHI-XbaI-SalI ligation of PCR products was used to create an internal ⌬247-369 Pol4 deletion in pTW300 with XbaI and SmaI sites at the deletion junction (pTW304). YW144 was then transformed with SmaI-digested pTW304 and PCR products that spanned the ⌬247-369 deletion and that included degenerate bases in the primer region. Plasmids were recovered from Leu ϩ transformants, the gap repair region sequenced to verify the presence of targeted but not unexpected mutations (D367E, pTW305; K247R/K248R, pTW306), and retransformed into YW144 for functional analysis. Wild-type isolates expressed functional Pol4, ensuring that pTW304 was not cryptically mutated.
Plasmid Transformation Assay-Plasmids were prepared by severalfold overdigestion with the appropriate combination of restriction enzymes, followed by phenol/chloroform extraction and ethanol precipitation. DNAs were examined by agarose gel electrophoresis with ethidium bromide staining to ensure complete digestion and equivalent concentrations in parallel preparations. More than 95% of MluI-, BamHI-, and MluI-BamHI-digested pMB0 and SphI-, KpnI-, and SphI-KpnI-digested pSK3 could be religated in vitro, as determined by treatment with T4 DNA ligase followed by agarose gel electrophoresis. Partially end filled pM1 was prepared by an additional incubation of 5 g of plasmid with 0.   (41) and sequenced with primer OW563 (5Ј-GGCAGGAGAATTTTCAGCATC, a reverse primer 114-base pairs downstream of the ADE2 start codon). Colony PCR was performed by touching a plastic pipette tip to a fresh yeast streak (Ͻ24 h) and inoculating 40 l of PCR buffer (10 mM Tris, pH 9.2, 50 mM KCl, 2.5 mM MgCl 2 , 400 M each dNTP) containing 0.625 units Taq polymerase (Promega) and 50 pmol of each primer. After amplification (1 cycle of 94°C for 4 min, 40 cycles of 94°C for 30 s, 55°C for 30 s, 72°C for 45 s, and 1 cycle of 72°C for 5 min) products were electrophoresed on a 1.25% agarose gel with ethidium bromide visualization. Negative and positive controls were included that had been verified by sequencing, because faint false bands were occasionally observed but readily distinguished from true positives as exemplified by the controls. Join rates were calculated as the fraction of colonies positive for the join multiplied by the normalized transformation rate for the relevant colony color. Join-FIG. 1. The plasmid transformation assay. A, pES16 includes the URA3 gene for plasmid selection and the ADE2 gene for color-based indication of join type. pES26 is pES16 with an excisable polyterminator inserted at the BglII site (17). Other substrates were derived from pES16 by engineering novel restriction site(s) and frameshifting nucleotide(s) just after the first two codons of the ADE2 gene. In this way, a given linearized plasmid will yield ADE2 (white) yeast after transformation only when the digested restriction site(s) are joined intracellularly by a mechanism that (re)establishes the correct reading frame. Plasmid names reflect the combination of inserted sites and the number of frameshifting nucleotides, for example pBX2 contains BamHI and XhoI sites with a reading frame of ϩ2 relative to ADE2. B, pES26 transformation data are plotted as a ratio of the colony count obtained with BglII-digested plasmid over the colony count obtained in a parallel transformation with undigested plasmid, a measure of the relative rate of simple religation at the BglII ends. Each point represents an independent transformation experiment.

FIG. 2. NHEJ via gapped alignments.
A, diagram of the substrate ends of MluI-digested pM1 and BamHI-XhoI-digested pXB2. Restriction site nucleotides are shown in boldface type, and encoded ADE2 amino acids are indicated above the second nucleotide of the corresponding codon. The reading frame of religated ends relative to the ADE2 coding sequence is indicated in parentheses. The arrows indicate the positions that were filled during incubation with Taq polymerase and dCTP. B, pM1 transformation data are plotted as a ratio of the colony count obtained with MluI-digested plasmid over the colony count obtained in a parallel transformation with BglII-digested plasmid. Two points are plotted for each strain for each independent transformation experiment; open circles represent white colonies (E), and filled circles represent red colonies (q). C, pXB2 transformation data are plotted as a ratio of the colony count obtained with BamHI-XhoI-digested plasmid over the colony count obtained in a parallel transformation with BamHI-digested plasmid. Complementation of the pol4 mutation with pTW285 (HM-Pol4) is denoted pol4 ϩ POL4. D, joins occurring at overhanging MluI and BamHI-XhoI ends. The number in parentheses indicates the number of nucleotides added or lost relative to a simple religation, which in turn predicts the colony color. Undigested pM1 and pXB2 yield red colonies. A line is shown between nucleotides basedpaired in a terminal microhomology. Join rates are the average normalized transformation rate for the appropriate colony color from parts B and C multiplied by the fraction of those colonies positive for the given join type by sequencing and/or PCR. ND, not determined. specific primer sets coupled OW620 (5Ј-CTTGACTAGCGCACTACCAG, a forward primer just near the 5Ј-end of the ADE2 fragment) with the following reverse primers: XB(ϩ2), OW580 (5Ј-CAACTGTTCTAGAGG-ATCGAG); MB(ϩ1), OW621 (5Ј-CAACTGTTCTAGAGGATCGT); HO Endonuclease Sensitivity Assay-Saturated overnight cultures in pGAL-HO-selective glucose minimal medium were washed and diluted in water, and cells were plated to pGAL-HO-selective minimal medium. Glucose plates were incubated 3 days and raffinose-galactose plates 5 days. PCR diagnosis of MAT join types was performed essentially as described (11).
Structure-based Sequence Alignment-First, the entire Pol4 sequence was submitted to the web-based Swiss Model utility (45) for mapping onto the hPol ␤ polypeptide contained in a gapped DNA co-crystal (PDB ID 1BPX, see Ref. 46). Pol4 residues 208 -405 were aligned. Next, Pol4 residues 1-207 and 406 -582 were submitted for mapping individually, which resulted in the further alignment of residues 464 -549. Because the alignment of 464 -477 conflicted with a more optimal alignment within the 208 -405 segment, the former was disregarded. Finally, the optimize mode was used to additionally align residues 555-574.
Ni-NTA Fractionation and Enzymatic Analyses-YW144 was transformed with either pRS315, pTW300, pTW301, or pTW305, and 500-ml cultures grown to an A 600 of ϳ1.0 in plasmid-selective minimal medium, yielding a cell mass after centrifugal harvest of ϳ0.9 g. Cell pellets were washed with and resuspended in 2-pellet volumes of ice-cold buffer A (10 mM Tris, pH 7.5, 20 mM imidazole, 0.5 M KCl, 5 mM MgCl 2 , 1 mM 2-mercaptoethanol, 10% glycerol, 1 mM phenylmethylsulfonyl fluoride, 2 g/ml aprotinin, 1 g/ml pepstatin A, 1 g/ml leupeptin). 3-pellet volumes of glass beads were added, and the cells disrupted by vortexing. Buffer A was added to 9 ml, the lysate brought to 0.1% Nonidet P-40, and proteins extracted by rocking 30 min at 4°C. After clearing by centrifugation at 12,000 ϫ g for 15 min, 0.35 ml of packed volume Ni-NTA beads (Qiagen) equilibrated to buffer A were added, and batch binding allowed to proceed with rocking for 1 h at 4°C. The slurry was poured into a disposable mini-column (Bio-Rad), the resin washed with 10 ϫ 1 ml of buffer A and proteins eluted with 2 ϫ 250 l of buffer A, 500 mM imidazole, 50 mM KCl. Eluates were dialyzed against 2 ϫ 1 liter of buffer B (10 mM Tris, pH 7.5, 50 mM KCl, 5 mM MgCl 2 , 1 mM dithiothreitol, 10% glycerol, 0.1 mM phenylmethylsulfonyl fluoride), flash frozen in 50-l aliquots, and stored at Ϫ70°C.
For activity assays, labeled strands (5 pmol) were phosphorylated using 5 units of T4 polynucleotide kinase in 20 l of the supplied buffer with 50 uCi [␥-32 P]ATP (NEN Life Science Products) for 15 min at 37°C, followed by addition of ATP to 1 mM and incubation for an additional 15 min. Unlabeled strands (10 pmol) were phosphorylated using only 1 mM ATP. Reactions were stopped by addition of EDTA to 20 mM, template, proximal and distal strand reactions were mixed, placed in boiling water and allowed to cool slowly to room temperature. Annealed probes were purified over Nick columns (Amersham Pharmacia Biotech). Assays were performed by adding an 8-l reaction mixture to 2 l of Ni-NTA eluate or buffer B and mixing by pipetting up and down, such that the final reaction contained 12.5 fmol of oligonucleotide substrate in 50 mM HEPES, pH 7.9, 100 mM NaCl, 10 mM MgCl 2 , 1 mM dithiothreitol, 100 M each dNTP. Klenow and exo Ϫ Klenow (0.1 units  Fig. 2A. B, pSK3 transformation data are plotted as a ratio of the colony count obtained with SphI-KpnI-digested plasmid over the colony count obtained in a parallel transformation with KpnI-digested plasmid (open circles, white colonies; filled circles, red colonies). C, joins predicted to occur at SphI-KpnI-digested ends, as in Fig. 2D. The maximum join rate for SK(ϩ1) and SK(Ϫ1) joins in the pol4 strain was calculated using the upper limit of the 95% confidence interval from the binomial distribution for 0 of 29 events. Undigested pSK3 yields red colonies. ND, not determined. in 2 l of buffer B) served as controls. After 60 min at 30°C, the reactions were stopped by addition of 40 l of sequencing dye and heating to 90°C for 5 min. 5 l (ϳ5,000 cpm) was then electrophoresed on a 20% sequencing gel, which was exposed wet to a PhosphorImager screen for 3 h before imaging and quantitation.

RESULTS
The Plasmid Transformation Assay-In this assay, test plasmids are linearized in vitro with restriction enzymes and transformed into yeast, where recircularization is required for plasmid maintenance. pES16 (Fig. 1A) contains the URA3 gene to select for transformants and the CEN/ARS sequence for plasmid maintenance, which also prevents recovery of integrated plasmids, because this results in a lethal dicentric chromosome. DSBs are introduced into the color indicator gene ADE2 at a variety of restriction sites. First, we have previously used BglII-digested pES16 and pES26 to show that yeast DNA ligase IV (Dnl4, also called Lig4) catalyzes NHEJ ligation (17). Second, we have created a series of new pES16 variants in which unique restriction sites have been inserted after the second codon of the ADE2 gene. Essentially any combination of restriction sites (and therefore encoded amino acids) can be introduced at this location without affecting Ade2 function. Different NHEJ events yield different reading frames of the ADE2 gene, however, and therefore different color colonies (specifically, ADE2 yeast are white, ade2 yeast are red). Joins are illustrated as the intermediate alignment structures inferred from the sequences of the final products, assuming no prior end degradation (Figs. 2-5, see Discussion). Fully compatible overhangs can be joined by simple religation, i.e. requiring only DNA ligase. Repair events that proceed via partial end annealing (i.e. microhomology usage) are of three types. Gap joins anneal in a fashion that requires fill-in synthesis by a DNA polymerase. Flap joins anneal in a fashion that extrudes excess bases that must be removed. Mixed joins show characteristics of both gap and flap joins. Because all yeast used in this study bear a complete chromosomal deletion of ADE2 (17), non-NHEJ transformation events arise only from the small fraction of undigested plasmid and rare macrodeletions.
Pol4 Is Not Required for Simple Religation-In the plasmid transformation assay, the relative rate of plasmid DSBR is typically revealed by normalizing the transformation rate obtained with linearized plasmid to that obtained in parallel with uncut plasmid. A subset of colonies are next analyzed to determine the mechanism of plasmid recircularization, either by sequencing of recovered plasmids or colony PCR, with the relative rate of a given repair event calculated by multiplying the normalized transformation rate by the fraction of colonies positive for the event. In the experiments below, we wanted to specifically examine the effects of polymerase gene deletions on end processing by normalizing instead to parallel transformations with linearized plasmids that are joined by simple religation. To validate this approach, we first transformed all mutant yeast used herein with BglII-digested pES26 and normalized to undigested plasmid. Unlike dnl4 yeast, pol4 and all other mutants showed essentially normal rates of simple religation relative to the wild type strain (Fig. 1B), demonstrating that they do not have a generalized NHEJ defect. Importantly, dnl4 yeast were found to be deficient in all joins discussed below, verifying that they are all bona fide NHEJ events (data not shown). It is not possible to meaningfully represent the dnl4 data in figures normalized to simple religation, however, because all modes of NHEJ are grossly deficient in dnl4 mutants.
pol4 Mutants Show a Minor Defect in Gap Joining-The 4-base 5Ј overhang generated by the restriction enzyme MluI is a repeat of the dinucleotide CG, which makes it useful for exploring the partial alignment of ends. We hypothesized that MluI ends would be joined in yeast by competitive mechanisms including simple religation and 2-base gap joins, where the CG repeats align in different registers (Fig. 2, A and C). Because it is this alignment register that determines the final join sequence, joins throughout are designated with a letter(s) indicating the restriction site end(s) and a number in parentheses indicating the reading frameshift relative to a simple religation. Thus, M(0) indicates religation of MluI ends, M(ϩ2) is the MluI gap join, and so on. As expected, the fraction of processed joins was markedly higher for MluI ends (ϳ20%, Fig. 2) than we had previously observed for BglII ends (ϳ1%, see Ref. 17). The majority of this increase was due to the M(ϩ2) gap join, i.e. white colonies with pM1. Mutation of POL4 again did not significantly affect M(0) simple religation but did slightly and reproducibly reduce the rate of polymerization-dependent M(ϩ2) joining (2-fold, p ϭ 0.03 from the Student's t test applied to the ranges of transformation data points shown in Fig. 2B). Plasmid overexpression of Pol4 rescued the mutant strain, whereas rev3 mutants had rates of M(ϩ2) formation identical FIG. 5. Increased sensitivity of pol4 yeast to HO endonuclease expression in a rad52 background. A, diagram of the substrate ends of MAT␣ chromosomal DNA after cleavage in vivo by HO. As with the plasmid ends, boldface type is used to highlight the terminal nucleotides, even though the HO recognition sequence is much larger. B, yeast were transformed first with either pRS317 (indicated as ADH-POL4 (Ϫ)) or its derivative expressing HM-Pol4 under the control of the ADH1 promoter (pTW286, indicated as ADH-POL4 (ϩ)), and subsequently with the GAL1-regulated HO-expressing plasmid pGAL-HO. Sensitivity to induction of HO was determined by plating to medium selective for both plasmids that contained raffinose-galactose as the carbon source. Data are plotted as the percent survival compared with parallel platings to medium-containing glucose. Each point represents a separate experiment performed with an independent pGAL-HO transformant. C, HO end join designations are given in a manner consistent with the plasmid joins, as well as originally described by Moore and Haber (11). The inferred alignments are drawn as in Fig. 2D, and, unlike Moore and Haber (11), assume no prior end degradation.
to the wild type strain, verifying the specificity of the pol4 phenotype (data not shown). Thus, although MluI-digested pM1 did have a high frequency of gap joining that revealed a reproducible pol4 effect, the magnitude of this effect was small, suggesting a redundancy of polymerase action.
We next sought to eliminate competition with simple religation in an attempt to increase the dependence on Pol4. MluIdigested pM1 was partially end filled with dCTP, yielding a plasmid with a 3-base 5Ј overhang that can anneal using the same base pairing as M(ϩ2) but now with only single nucleotide gaps. Fig. 2B shows that this treatment largely blocked red colony formation (i.e. M(0) joining) as expected, and correspondingly increased the rate of white colony formation (i.e. M(ϩ2) joining) about 2-fold. The Pol4 dependence was similar to that seen with untreated ends, however. BamHI-XhoI-digested pXB2 presents "incompatible" ends that can only anneal as 2-base gap and flap joins (Fig. 2). The XB(ϩ2) gap join formed at 60% of the rate of BamHI simple religation, demonstrating that joining via partially annealed intermediates can be quite efficient. Despite this, mutation of pol4 caused only a very slight reduction in the frequency of XB(ϩ2) gap joining (1.4-fold, p ϭ 0.01).
pol4 Mutants Are Markedly Deficient in Mixed Joining-We next tested substrates with only single nucleotide microhomologies in an attempt to reveal a greater Pol4 dependence by destabilizing the annealed intermediate. Figs. 3 and 4 present two such substrates: pMB0 juxtaposes the incompatible 5Ј overhangs MluI-BamHI, and pSK3 the incompatible 3Ј overhangs SphI-KpnI. As illustrated, four alignments are predicted for each: a 3-base gap, a 3-base flap (5Ј versus 3Ј), and two mixed joins. The 3-base gap joins, i.e. MB(ϩ3) and SK(ϩ3), were initially of greatest interest, because they require only DNA polymerization. With pMB0, white colonies bearing MB(ϩ3) did form in the wild type strain at 1.5% of the rate of BamHI simple religation, but this join showed only the same 2-fold decrease in the pol4 mutant as the 2-base gap joins (p ϭ 0.003, Fig. 3). It became clear that the red colonies revealed a more striking result, however. MB(ϩ1) formed even more efficiently than MB(ϩ3) at 6.5% of the rate of simple religation, even though the MB(ϩ1) annealed intermediate includes a 2-base 5Ј-terminal mismatch. Further, MB(ϩ1) joining was markedly impaired in the pol4 mutant (28-fold). We did not detect macrodeletion, MB(Ϫ3), or MB(Ϫ1) events in the wild type strain. The remaining red colonies contained intact pMB0, arising from either single or undigested plasmid (see "Experimental Procedures" for details regarding stop codons that are excised during double digestion). pSK3 revealed a different pattern from pMB0 in that the SK(ϩ3) 3-base gap join was not detected. Rather, white colonies resulted from SK(Ϫ3) flap joining, at 2.6% of the rate of KpnI simple religation (Fig. 4). pSK3 was similar to pMB0, however, in that mixed joins (red colonies) predominated, despite the presence of 3Ј-terminal mismatches. SK(Ϫ1) was favored, but both SK(Ϫ1) and SK(ϩ1) were detected and together formed at 10% of the rate of simple religation. Most dramatically, SK(Ϫ1) joining was more than 95-fold reduced in the pol4 mutant, and the combination of the mixed joins was more than 120-fold reduced. The few red colonies obtained with the pol4 mutant were a mixture of intact plasmid (44%) and rare macrodeletion (44%) and microdeletion (9%) events. For both pMB0 and pSK3, mixed joining was REV3-independent and restored by plasmid-expressed His 9 -Myc 3 -Pol4 (HM-Pol4), demonstrating the specificity of the phenomenon for the pol4⌬ allele (Figs.  3 and 4).
pol4 Yeast Are Deficient in Chromosomal Mixed Joining-We next sought to verify the role of Pol4 in mixed joining at a chromosomal rather than a plasmid DSB. Expression of HO endonuclease (HO) in yeast that bear a rad52 mutation is largely lethal, because this prevents homologous repair of the DSB created by HO at MAT (11,47). Rare cells (ϳ0.1%) escape the effects of HO, however, predominantly by repair events that create HO-resistant (i.e. nonrecleavable) MAT alleles. The two most common HO-resistant MAT alleles, called ϩCA and ϪACA by Moore and Haber (11), are readily modeled as a mixed join with a 3Ј-terminal mismatch (HO(ϩ2)) and a 3-base flap (HO(Ϫ3)), respectively (Fig. 5, A and C). It seemed likely that the same mechanism was responsible for both these and plasmid NHEJ, because each is Rad50/Mre11-dependent (8,11). Indeed, hdf1/rad52 and dnl4/rad52 yeast each showed a more than 10-fold lower survival than rad52 yeast after induction of HO expression from the GAL1 promoter ( Fig. 5B and not shown) (8). MAT␣/hdf1/rad52 HO survivors all retained the ␣ mating type, indicative of escape by inactivation of HO with an intact MAT␣ allele, implying a complete absence of NHEJ at MAT DSBs in hdf1 yeast (Table II). A pol4/rad52 strain also showed a 10-fold reduction in HO survival compared with rad52 yeast with essentially complete loss of HO(ϩ2) joining (Ͼ70-fold), although a substantial fraction of MAT␣/pol4/rad52 HO survivors became sterile as a result of HO(Ϫ3) joins and lower frequency events. Overexpression of HM-Pol4 from a plasmid corrected the pol4 but not the hdf1 mutant phenotypes (Fig. 5B). This pattern is similar to the plasmid results, except for the greater effect of pol4 mutation on 3Ј flap joining in the chromosomal assay (8-fold for HO(Ϫ3) versus 2-fold for SK(Ϫ3)).

TABLE II
Survival, mating type, and join rates among HO-expressing rad52 strains rad52 (YW130), rad52/pol4 (YW153), and rad52/hdf1 (YW132) yeast strains carrying pGAL-HO were plated to plasmid-selective minimal medium containing raffinose-galactose, and the fraction surviving was determined relative to a parallel plating to medium containing glucose (mean Ϯ S.D. at least four independent experiments). When possible, independent sterile isolates were tested for the presence of the HO(ϩ2) and HO(Ϫ3) events by join-specific PCR; joins designated "other" were negative in both assays. Join rates reflect the fraction of cells plated to raffinose-galactose that gave rise to colonies with the indicated join type. Mutational Analysis of Pol4 -The simplest interpretation of the strong Pol4 dependence of MB(ϩ1), SK(Ϫ1), SK(ϩ1), and HO(ϩ2) mixed joins is that Pol4 itself removes terminal mismatches. To examine the Pol4 protein in detail to see if such a gap-dependent nuclease activity could be discovered, we performed a structure-based sequence alignment of Pol4 with the hPol ␤ polypeptide resolved in a co-crystal with gapped DNA (Fig. 6A, and Ref. 46). Pol ␤ binds to both the 3Ј and 5Ј termini of short nucleotide gaps as a critical part of its function (34,46,48). At the 3Ј terminus, the highly conserved "fingers," "palm," and "thumb" polymerase subdomains cooperate to bind and catalyze nucleotidyl transfer, with the three universal aspartic acid residues coordinating the incoming Mg 2ϩ -dNTP. At the 5Ј terminus, the "8 kDa" lyase domain binds and catalyzes re- Guided by this analysis, we made a series of focused HM-Pol4 mutations and tested them for their ability to complement joining of MluI-BamHI-digested pMB0 and SphI-KpnI-digested pSK3 by pol4 yeast (Fig. 6, B-D). First, the D367E palm mutation behaved similarly to the null allele with loss of mixed joining, consistent with loss of nucleotidyl transfer via disrupted geometry of Mg 2ϩ -dNTP coordination. Second, we mu- tated the lysine residues in the pocket where Schiff base formation should occur, to test whether Pol4 cleaves 5Ј mismatches by a ␤-elimination reaction similar to removal of a 5Ј-deoxyribose phosphate. The protein, K247R/K248R, did not demonstrate the specific loss of MB(ϩ1) joining that would be expected with loss of a 5Ј nuclease, but rather showed a slight reduction in all joins, consistent with weakened binding to the 5Ј terminus (K247A and K248A mutations gave similar results, not shown). Third, removal of even a small part of the unique Pol4 amino terminus in ⌬1-61 led to complete inactivation of Pol4 (similar results were obtained with ⌬1-145 and ⌬1-205 mutants, not shown). Although this result demonstrates the importance of the unique sequence of Pol4, it unfortunately does not provide any information regarding the function of this domain.
Pol4 Lacks Nuclease Activity in Vitro on Gapped Substrates-We next used Ni-NTA-agarose to partially purify HM-Pol4 and its mutant derivatives and examined their biochemical activities on a series of gapped oligonucleotide substrates (Fig. 7). Ni-NTA fractionation was not sufficient to yield pure protein, but no contaminating polymerase activities were detected in the vector-only control fraction (Fig. 7C, lane 7). Also, no Pol4-dependent bands were copurified in this single step that might be candidates for a Pol4-associated nuclease (Fig.  7B). HM-Pol4 (but not D367E) filled a 2-nt gap and stopped (Fig. 7C, lane 6), unlike the displacement synthesis seen with Klenow (lane 5), similar to published results (35). Interestingly, ⌬1-61 activity on gapped substrate (lane 9) was similar to the reduced HM-Pol4 activity in the absence of distal strand (lane 11), which suggests that this mutant may have impaired gap recognition, explaining its in vivo phenotype. When a 5Ј-terminal mismatch was present on the distal strand, the resulting 3-nt gap was filled efficiently, but again with no continued displacement into base paired positions (lane 12). The 5Ј-terminal mismatch was not cleaved during this reaction, which would be evident as a specific reduction of probe counts when the mismatched distal strand was labeled, although this was complicated by a low level phosphatase activity in all Ni-NTA fractions (Table III). A 3Ј-terminal mismatch on the proximal strand significantly impaired extension by HM-Pol4 and exo Ϫ Klenow, but not by Klenow (Fig. 7C, lanes 13-15). The limited remaining extension by HM-Pol4 is best explained by simple incorporation of the mismatch as opposed to removal and resynthesis, because it did not depend on dGTP (lane 17). Further, prevention of polymerization by the D367E mutation or by removal of dNTPs revealed no proximal strand shortening (lanes 16 and 20). Collectively, then, we find Pol4 to be a gap-filling polymerase, but with no detectable gap-dependent nuclease activity.
Exo1, Apn1, and Rad1 Do not Provide Pol4-associated Nuclease Activity-Finally, we have tested three genes as candidates for providing the essential Pol4-associated nuclease activities. Exo1, a major mitotic 5Ј nuclease in yeast cells (49), was not required for MB(ϩ1) joining (Fig. 3B). The apurinicapyrimidinic endonuclease interacts functionally with Pol ␤ during BER, cleaving on the 5Ј side of an abasic site (33). It was possible that a similar interaction is utilized in NHEJ, with the yeast homologue Apn1 providing 3Ј nuclease function (50). Alternatively, this might be provided by the Rad1-Rad10 complex, similar to cleavage of nonhomology tracts, i.e. 3Ј flaps, during rDSBR (51). apn1 and rad1 mutants made SK(Ϫ1)/ SK(ϩ1) joins at wild-type rates, however (Fig. 4B). DISCUSSION A Pol4-dependent Pathway of NHEJ-In this study, we demonstrate that yeast are capable of efficient NHEJ at incompatible ends via intermediates that make extensive use of terminal microhomology, as required at DSBs that are created by genotoxic agents and believed to lead to chromosomal translocation. Further, both plasmid and chromosomal assays show conclusively that Pol4 is one of the processing enzymes recruited for this purpose. This was established only by the subset of mixed joins that required resolution of terminal mismatches, however, especially at 3Ј overhangs (see Fig. 8  tants, respectively, but M(ϩ2), XB(ϩ2), and MB(ϩ3) gap joins were at best 2-fold reduced. Even the 3Ј flap join HO(Ϫ3), which would not be predicted to require DNA polymerization, was 8-fold reduced in a pol4 mutant. These large pol4 effects cannot be explained by a secondary or nonspecific effect due to loss of Pol4 protein, because the D367E nucleotidyl transfer point mutation impaired mixed joining similarly to the pol4⌬ mutation (Fig. 6). Rather, there is a catalytic requirement for Pol4. Further, 5Ј-and 3Ј-terminal mismatches must have been removed during joining. Simple incorporation of mismatched nucleotides (as seen in vitro) was not possible in some cases, because they outnumbered the adjacent gap length (e.g. SK(Ϫ1), Fig. 4). In the other cases, incorporation would yield a mixture of join sequences, but only one was recovered (e.g. MB(ϩ1), Fig. 3). Thus, Pol4 is required to provide efficient gap-filling polymerization at mixed joins in conjunction with the removal of 5Ј-and 3Ј-terminal mismatches.
Two hypotheses might explain this unique requirement of Pol4 for mixed joining. The simplest is that Pol4 itself provides 5Ј and 3Ј exonuclease activities. Even though early experiments with radiolabeled polynucleotide substrates did not reveal these (36,37), it was important to address the potential for gap-dependent nuclease functions in Pol4 by several means. Genetic assays were done with mutations of putative nuclease domains suggested by a structure-based sequence alignment of Pol4 with hPol ␤. Deletion of the uncharacterized unique amino terminus of Pol4 (⌬1-61) was uninformative in this regard, because this mutation led to a pattern indistinguishable from the null allele. Point mutation of the lysines predicted to be present in the lyase domain (K247R/K248R) slightly impaired join recovery overall, but did not reveal a specific pattern of joining that would support a nuclease function (Fig. 6). Biochemical analyses were performed with partially purified Pol4 and mutant derivatives. As predicted by the structural comparison and by previous studies (35), Pol4 exhibited the gapfilling synthesis that would be required at NHEJ complexes, i.e. a recognition of both 5Ј and 3Ј termini and displacement of only mismatched distal bases. However, we detected no Pol4dependent 5Ј or 3Ј nuclease activities on gapped substrates (Fig. 7, Table III). Although it is possible that oligonucleotide substrates are insufficient to reveal an existing Pol4 NHEJ nuclease activity, the available data strongly argue that Pol4 is not a nuclease.
The alternative hypothesis is that Pol4 interacts, either directly or indirectly, with the critical terminal mismatch-resolving nuclease(s) at the NHEJ active site. As neither we nor others observed a protein or nuclease activity co-purifying with Pol4 (36,37), this is more likely to be a transient interaction than a stable one. Among 5Ј nucleases, Exo1 is not required (Fig. 3), but in a previous report we have shown that the yeast flap endonuclease Rad27 is involved in processing a subset of 5Ј flaps during NHEJ (52). Among 3Ј nucleases, Apn1 and Rad1 are also not required (Fig. 4), indicating that processed NHEJ is distinct from BER and rDSBR in yeast. Several candidates remain. The proofreading exonuclease of Pol ␦ is used in part for processing of short 3Ј nonhomology flaps during rDSBR (51), although it is unlikely that another polymerase would provide a nuclease activity in NHEJ but rely on Pol4 for nucleotidyl transfer. Rnc1 (also called Nud1) is an endo-exonuclease involved in DSBR, but rnc1 mutation leads to increased survival after HO expression in a rad52 background (53). Most intriguingly, Mre11 is known to possess a Mn 2ϩ -dependent 3Ј-5Ј nuclease activity that can promote end joining in a limited in vitro system (24). This activity seems at odds with the strong reduction of 5Ј strand degradation in rad50 and mre11 mutants (12), but is entirely consistent with a role in NHEJ end processing. Unfortunately, mre11⌬ yeast are entirely deficient in NHEJ, but recently described Mre11 phosphodiesterase mutants will be interesting to study in our plasmid assay (23,25,26).
A unifying model that incorporates interaction with a nuclease with the concept of polymerase redundancy states that Pol4 is required for only one of the potentially several NHEJ pathways in yeast cells. A first more limited pathway can join ends only via simple religation or gapped intermediates, but is more promiscuous with regards to polymerase usage, explaining the limited effect of the pol4 mutation on gap joining. A second more versatile pathway can join a greater variety of end configurations, such as mixed joins, presumably by recruitment of additional processing activities. This pathway is also more selective, however, and will only utilize Pol4. There is precedent for the presence of multiple NHEJ pathways in yeast. Moore and Haber (11) examined NHEJ events at HO-cut MAT DNA, similar to Fig. 5, and showed that HO(ϩ2) joins predominate in S/G2 and are dependent on RAD50/MRE11. HO(Ϫ3) joins are less cell cycle dependent, and less impaired by loss of RAD50/MRE11, suggesting that they proceed through a different NHEJ pathway. Our pol4 data are very similar to the rad50 pattern observed by Moore and Haber, suggesting Pol4 recruitment as part of the molecular basis for the Rad50/Mre11-dependent NHEJ pathway dominant in S/G2, which in turn further implicates Mre11 as the missing 3Ј nuclease activity.
End Processing in Higher Eukaryotes-Unlike initial observations regarding DNA ligase IV (17,21), it is uncertain whether our Pol4 results indicate a role for Pol ␤ in NHEJ in higher eukaryotes. hPol ␤ mediates BER, whereas Pol4 is dispensable for this function (54). Further, the yeast-specific portions of Pol4 may be required for specific recruitment to the NHEJ active site. Finally, Sobol et al. (32) have examined Pol ␤-deficient fibroblasts and found no increase in IR sensitivity, indicative of an intact NHEJ mechanism. Despite this, the end alignment and processing we observed in yeast is very similar to that observed for vertebrate NHEJ both in vivo (55)(56)(57) and in vitro (58,59). Indeed, NHEJ in Xenopus egg extracts is inhibited by ddNTPs but not aphidicolin (58), a pattern characteristic of Pol ␤.

TABLE III
Pol4 Ni-NTA fractions contain a contaminating phosphatase but no specific 5Ј nuclease Labeled oligonucleotides and Ni-NTA fractions were incubated and electrophoresed similar to Fig. 7C and quantitated using a PhosphorImager. The fraction of lane counts extended to longer forms was calculated as a measure of the nucleotidyl transferase activity. The fraction of total input counts remaining in each lane (in comparison with probe-only control lanes) reflects 5Ј-phosphatase or 5Ј-exonuclease activity. Toward Understanding the Concerted NHEJ Mechanism-In the model suggested by this study and many others, DSB recognition leads to formation of a complex on each DNA end that presumably involves at least the Ku proteins and probably the Sir and Rad50-Mre11 complexes. Ends are brought together, and base pairing drives formation of a stable alignment structure that recruits enzymes for resolution of strand discontinuities and ultimately nick ligation. Importantly, our data indicate that the yeast NHEJ apparatus does not attempt to judge the correctness of an alignment register by biasing against discontinuous alignments by a means other than their relative thermodynamic stability. For example, the alignment gaps in M(ϩ2) are not frankly inhibitory, because this join formed once for every five M(0) joins (Fig. 2). Indeed, when only gap joining is possible it can be nearly as efficient as simple religation, as seen with XB(ϩ2) (Fig. 2).
The most tenuous aspect of the above model is the ordering of events. The alignments diagrammed in the figures are inferred from the join sequences based on the assumption that alignment occurs before processing. From a technical standpoint, we have verified that plasmid ends are ligation-competent in vitro, and therefore base loss must occur in vivo (see "Experimental Procedures"). Limited in vivo degradation of ends prior to joining could explain the observed join types but would provide no explanation for the differential effects of pol4 mutation. Further evidence that base removal occurs after alignment is provided by our finding that FEN-1 is involved in NHEJ, as discussed elsewhere (52). A final issue is the exten- sive and efficient 5Ј degradation seen in yeast and known to precede rDSBR events. The nature of the balance between 5Ј degradation and end preservation is uncertain, because Rad50 and Ku are both required for NHEJ and yet the former promotes end degradation, whereas the latter stabilizes ends (60). We cannot rule out the possibility that extensive 5Ј degradation occurs before NHEJ of 3Ј overhangs, because the 3Ј-terminal bases would still be available for pairing. Even limited 5Ј strand degradation would prevent occurrence of all joins observed at 5Ј overhangs, however. It will be of great interest to examine this dynamic relationship between NHEJ, rDSBR, and the degradation and polarity of ends in more detail.