Processing of 3′-Phosphoglycolate-terminated DNA Double Strand Breaks by Artemis Nuclease*

The Artemis nuclease is required for V(D)J recombination and for repair of an as yet undefined subset of radiation-induced DNA double strand breaks. To assess the possibility that Artemis acts on oxidatively modified double strand break termini, its activity toward model DNA substrates, bearing either 3′-hydroxyl or 3′-phosphoglycolate moieties, was examined. A 3′-phosphoglycolate had little effect on Artemis-mediated trimming of long 3′ overhangs (≥9 nucleotides), which were efficiently trimmed to 4–5 nucleotides. However, 3′-phosphoglycolates on overhangs of 4–5 bases promoted Artemis-mediated removal of a single 3′-terminal nucleotide, while at least 2 nucleotides were trimmed from identical hydroxyl-terminated substrates. Artemis also efficiently removed a single nucleotide from a phosphoglycolate-terminated 3-base 3′ overhang, while leaving an analogous hydroxyl-terminated overhang largely intact. Such removal was completely dependent on DNA-dependent protein kinase and ATP and was largely dependent on Ku, which markedly stimulated Artemis activity toward all 3′ overhangs. Together, these data suggest that efficient Artemis-mediated cleavage of 3′ overhangs requires a minimum of 2 nucleotides, or a nucleotide plus a phosphoglycolate, 3′ to the cleavage site, as well as 2 unpaired nucleotides 5′ to the cleavage site. Shorter 3′-phosphoglycolate-terminated overhangs and blunt ends were also processed by Artemis but much more slowly. Consistent with a role for Artemis in repair of terminally blocked double strand breaks in vivo, human cells lacking Artemis exhibited hypersensitivity to x-rays, bleomycin, and neocarzinostatin, which all induce 3′-phosphoglycolate-terminated double strand breaks.

nitrilotriacetic acid; Amersham Biosciences). Fractions containing Artemis were dialyzed into 50 mM HEPES, pH 7.5, 10% glycerol, 2 mM EDTA, 1 mM dithiothreitol, 0.01% Nonidet P-40, 20 g/ml phenylmethylsulfonyl fluoride, 1 g/ml aprotinin, pepstatin A, and leupeptin (HCB), containing 0.1 M NaCl. The affinity-purified protein was loaded onto a Mono Q column (Amersham Biosciences) and eluted with a linear gradient of 100 -500 mM NaCl in HCB. Aliquots of Artemis-containing fractions were snap-frozen and stored at Ϫ70°C. Protein from these aliquots was subjected to tandem mass spectrometry and identified as Artemis. Endonuclease and exonuclease activities were verified by standard biochemical assays, and aliquots were discarded after four or fewer freeze/thaw cycles. Ku70/80 was purified from insect cells coinfected with a mixture of recombinant baculovirus harboring the human KU70 and KU80 genes, and DNA-PKcs was purified from HeLa cells, both as described previously (11). All protein concentrations were determined by Bradford assays using bovine serum albumin as a protein standard (Bio-Rad), and protein purity was evaluated by Coomassie Blue-stained SDS-polyacrylamide gels. When necessary, proteins were diluted immediately before use in reaction buffer lacking ATP either on ice (Artemis, DNA-PKcs) or at 22°C (Ku).
DNA Substrates-Oligonucleotides were purchased from Qiagen or Integrated DNA Technologies. All labeled oligomers were purified by gel electrophoresis followed by reverse-phase HPLC. The yield of each oligomer was determined by integrating the absorbance at 260 nm recorded on an in-line UV monitor. To generate 3Ј-PG oligomers with the sequence CGAGG-AACGCG(A n )CG (0 Յ n Յ 4), 5Ј-32 P-end-labeled oligomers CGAGGAACGCG(A n )CGCCC were treated with bleomycin plus H 2 O 2 , and the desired 14 -17-base 3Ј-PG products were isolated from a sequencing gel and purified by HPLC (12). The terminal structure and purity of the 14-base (n ϭ 1) 3Ј-PG oligomer was verified by electrospray mass spectrometry and fragmentation-based chemical sequencing. Treatment of the other (n ϭ 0 and 2-4) oligomers with bleomycin likewise resulted in efficient site-specific cleavage at the -CGCCC site near the 3Ј end, as judged by the appearance of a prominent band with the expected mobility (supplemental Fig. 1). Upon extraction from the gel, each such oligomer eluted as a single well defined peak from reverse-phase HPLC. The terminal structure and purity of each putative 3Ј-PG oligomer was confirmed by quantitative conversion to the corresponding 3Ј-phosphate and 3Ј-hydroxyl oligomers by treatment with tyrosyl-DNA phosphodiesterase and polynucleotide kinase/phosphatase, which in every case induced the expected shifts in electrophoretic mobility (supplemental Fig. 1). An analogous oligomer (n ϭ 1) bearing a 3Ј-phosphotyrosyl terminus was purchased from Midland Certified Reagents. The 9-and 11-base 3Ј-PG oligomers CGAGG-AACG and CGAGGAACGCG were similarly prepared as described previously and their structures verified by mass spectrometry (12). The 36-base 3Ј-PG or 3Ј-hydroxyl oligomers were prepared by ligating the corresponding labeled 14-mers to the 22-mer GCCATGTACTTGGATGATCTAT in the presence of the complementary 20-mer GCGTTCCTCGATAGA-TCATC and were again gel/HPLC-purified. Partial duplexes were annealed in 10 mM Tris-HCl, pH 8, 0.1 M NaCl, 1 mM EDTA by heating a mixture of labeled oligomer and a 1.5-fold excess of unlabeled complement to 80°C followed by slow cooling to 10°C over a period of 3 h. Internally labeled plasmid substrates with various overhangs were constructed by ligating 3Ј-PG, 3Ј-phosphotyrosyl, or 3Ј-hydroxyl oligomers (9 -24 bases in length) into plasmids with an 11-base 5Ј overhang, as described (13). Each plasmid was gel-purified and eluted, and the concentration was determined from the 260 nm peak of the absorbance spectrum.
Nuclease Assays-Reaction mixtures (10 l) containing 25 mM Tris-HCl, pH 8, 25 mM NaCl, 10 mM MgCl 2 , 1 mM dithiothreitol, 0.25 mM ATP, 50 g/ml bovine serum albumin, and either 5 nM oligomeric substrate or 1 nM plasmid substrate were prepared at 22°C. In some cases, a blunt-ended doublestranded 35-mer (4 nM to 1 M) was also included (3). Ku was added, and the mixture was immediately vortexed and briefly centrifuged. DNA-PKcs was then added followed immediately by Artemis, and the reaction was mixed by pipeting and placed in a 37°C bath. Reactions with oligomeric substrates were stopped by addition of 10 l of formamide containing 20 mM EDTA, and the DNA was heat-denatured and analyzed on sequencing gels. Reactions containing plasmid substrates were stopped by addition of 20 l of 10 mM EDTA, 0.45 M sodium acetate, 100 g/ml tRNA, followed immediately by phenol extraction and ethanol precipitation. DNA was cut with AvaI (20 units, 6 h, 37°C) or Taq ␣ I (hereafter referred to as TaqI; 20 units, 4 h, 65°C) in 50 l of the buffer supplied by the vendor (New England Biolabs). In some cases 0.25 mM CoCl 2 and 0.1 mM ddGTP were subsequently added, and the samples were treated with 20 units of terminal deoxynucleotidyltransferase (New England Biolabs) for 1 h at 37°C. DNA was precipitated, dissolved in formamide, heat-denatured, and analyzed on sequencing gels. The AvaI and TaqI cleavage sites differ by one nucleotide; however, as expected these two enzymes always gave essentially identical results in parallel experiments.
Cell Lines and Proliferation Assays-Fibroblast lines were established from skin biopsies from SCIDA patients 04 and 05 and an unrelated immunologically normal individual (AK). Early passage cultures were immortalized by introduction of human telomerase reverse transcriptase cDNA using methods previously described (6,11). Cells were either irradiated using a Pantak x-ray generator operating at 320 kV/10 mA with 0.5 mm copper filtration or exposed to genotoxic drugs for 1 h, washed three times with PBS, and then labeled by the addition of fresh medium containing 10 g/ml bromodeoxyuridine (BrdUrd) (Sigma) for 24 -35 h. Cells were harvested, fixed, and stained using standard procedures, and cell cycle distribution and BrdUrd incorporation were analyzed with a Beckman-Coulter EPICS XL-MCL flow cytometer using XL Data Acquisition software and WinMDI 2.8 or FlowJo 8.1 software packages. The fraction of proliferating cells (F) was calculated by scoring the percent of intact cells staining positive for BrdUrd and normalizing to the untreated control of the same cell line (Ն10,000 events were scored for each point). To quantitatively compare the toxicity of each agent toward SCIDA and normal cells, relative toxicity at each radiation dose or drug concentration was calculated as ln(F1)/ln(F2), where F1 and F2 are the proliferating fractions, normalized to untreated controls, for SCIDA and normal cells, respectively. This parameter was relatively constant over a range of concentrations even in the case of bleomycin, which typically shows a distinct upward concavity in the response.

Trimming of Long 3Ј-PG-terminated Overhangs by Artemis-
DSBs induced by radiation and other free radical-based toxins commonly bear 3Ј-PG termini (7, 14 -16), which block polymerase and ligase activities as well as most human exonucleases. The Artemis nuclease, which reportedly shortens long 3Ј overhangs, could thereby resolve PG and other blocked termini on 3Ј overhangs of DSBs. To determine whether the endonucleolytic activity of Artemis was affected by the presence of PG termini, an internally labeled 3Ј-PG or 3Ј-hydroxyl-terminated 36-mer was annealed to 21-, 23-, and 27-base complementary strands to yield substrates with 15-, 13-, and 9-base 3Ј overhangs.
As expected, Artemis treatment of the 3Ј-hydroxyl substrates resulted in efficient trimming of overhangs ( Fig. 1A) (17). The 15-base (36/21 substrate) and 13-base (36/23 substrate) overhangs were each predominantly shortened to a 5-base overhang, yielding 26-and 28-base products, respectively. The 9-base overhang (36/27p substrate) was trimmed predominantly to a 4-base overhang (31-mer) rather than 5-base overhang. The PG-terminated 15-and 13-base overhangs were likewise trimmed predominantly to 5-base overhangs, whereas the 9-base overhang was trimmed to a 4-base overhang. These cleavage patterns and levels of activity are essentially identical to those obtained with the analogous 3Ј-hydroxyl overhangs (compare Fig. 1, A and B), indicating that PG moieties several bases from the cleavage site had little or no effect on the efficiency or specificity of Artemis-mediated DNA cleavage. In all cases, 38 -60% of the substrate was cleaved, with the major product accounting for 50 -66% of the total.
A substrate composed of a 19-mer annealed to the labeled 36-mer, with 19 bp of duplex DNA and a 17-base overhang, sustained at least 3-fold less Artemis-mediated cleavage than the other substrates and showed altered specificity (Fig. 1A). To determine whether this inefficiency was because of the 19-bp duplex being too short to accommodate the Ku-DNA-Artemis complex, or the 17-base overhang being too long to allow efficient cleavage, we prepared substrates having the same 23-bp duplex DNA and differing only in the length of the 3Ј singlestranded overhangs, 13, 19, and 25 bases. Artemis nuclease cleaved all three substrates, with comparable efficiency and essentially identical specificity, with the dominant 5-base overhang (28-mer) product accounting for 44 -52% of total cleavage in each case (Fig. 1C). A substrate having only 19 bp of duplex DNA (48/19) sustained 50% less cleavage, and the usual 5-base overhang specificity was lost, with an 11-base overhang (30mer) being the most frequent product (Fig. 1C).
Taken together, these data indicate that Artemis is capable of trimming 3Ј overhangs at least as long as 25 bases, with the incision site being predominantly 5 nucleotides from the double strand/single strand transition. As footprinting studies indicate that Ku plus DNA-PKcs protects ϳ28 bp at a DNA end (18), the apparent requirement for at least ϳ20 bp of duplex DNA for proper binding and alignment of the nuclease complex is not surprising.
Control reactions here and throughout reveal that Artemis, Ku, and DNA-PKcs protein preparations are free of significant contaminating nucleases. A Coomassie Blue-stained polyacrylamide gel showing the purity of preparations of DNA-PKcs, Ku70/80, and Artemis is shown (Fig. 1D).
DNA-PKcs and Ku Dependence of Artemis Activity on Long 3Ј Overhangs-Previous work with immunoconjugated Artemis immobilized on agarose beads suggested that Artemis nuclease cleaved long 3Ј overhangs to a length of ϳ5 bases, in a reaction that required DNA-PKcs but was unexpectedly independent of the presence or absence of Ku (3). To assess the DNA-PKcs and Ku dependence of DNA cleavage by purified soluble Artemis, a substrate having 23 bp of duplex DNA and a 13-base 3Ј-hydroxyl-terminated 3Ј overhang (36/23-mer, as above but labeled at The 27p oligomer was 5Ј-phosphorylated, but the other complementary strands were not. Note that the substrates in A and B are internally labeled 14 bases from the 3Ј end. Thus, the Artemis cleavage products, which bear both 5Ј-and 3Ј-hydroxyls, run a full nucleotide position more slowly than Maxam-Gilbert markers of the same length, which bear 5Ј-and 3Ј-phosphates; for example, the 31-base cleavage product from the 36/27p duplex comigrates with the 32-base marker generated by chemical cleavage of the G at position 33 in the sequence ( . . . CGCGACG). D, purity of proteins used in Artemis cleavage reactions. Recombinant Artemis (10 g), recombinant Ku (10 g), and DNA-PKcs (5 g) purified from HeLa cells were subjected to denaturing PAGE and stained with Coomassie Blue. Leftmost lane contains molecular size markers (kDa). FEBRUARY 9, 2007 • VOLUME 282 • NUMBER 6 the 5Ј end rather than internally) was reacted with Artemis plus DNA-PK component proteins ( Fig. 2A). As expected, cleavage of this substrate was completely dependent on Artemis, DNA-PKcs, and ATP and yielded a 5-base overhang as the predominant product (49% of total cleavage in Fig. 2A, lanes 3 and 8). A time course showed that the cleavage pattern was established within the first few minutes of the reaction (Fig. 2B), consistent with endo-rather than exonucleolytic cleavage. In addition, the similarity of the cleavage patterns between the end-labeled (Fig. 2) and internally labeled ( Fig. 1) substrates indicates that there was negligible degradation of the 5Ј termini of these substrates.

Artemis Processing of 3-Phosphoglycolate Termini
Although Ku was not strictly required for cleavage, the addition of purified Ku70/80 stimulated Artemis nuclease activity 4-fold, in a DNA-PKcs-and ATPdependent manner ( Fig Titrations with limiting Artemis revealed that Ku stimulated Artemis activity over a wide range of Artemis concentrations (Fig. 2E). In the presence of Ku, the concentration of Artemis required for a given level of cleavage was 10-fold lower than in its absence.
To generate substrates more similar to DSBs produced in vivo (i.e. having very long duplex regions and short overhangs), a 3Ј-resected plasmid with an 11-base 5Ј overhang was pre- pared, to which various 5Ј-labeled oligomers were ligated. Processing of 3Ј overhanging termini on these substrates was analyzed by cutting the treated plasmid with TaqI or AvaI to release the labeled oligomer from the end, as described previously (19). Consistent with data obtained using the oligomeric substrates, assays with a plasmid bearing a 13-base overhang revealed a significant stimulatory effect of Ku on Artemis nuclease activity over a range of Artemis concentrations (Fig. 2F), with a maximum stimulation of 6-fold when Artemis was limiting (1-10 nM). Titrations with limiting DNA-PKcs revealed that Ku stimulation of Artemis is DNA-PKcs concentration-dependent and that optimal nuclease activity required DNA-PKcs concentrations that approach molar equivalence with Artemis (Fig. 2G). Taken together, these data are most consistent with Ku stimulating Artemis activity through enhancing assembly or recruitment of Artemis/DNA-PKcs to DNA termini, thereby increasing nuclease activity.
Ku-dependent Serial Cleavage of 3Ј Overhangs-To examine in more detail the effect of Ku on the kinetics and specificity of cleavage, the same plasmid substrate used above was treated with Artemis in the presence or absence of Ku for various times, and the percentage of uncleaved substrate as well as of individual cleavage products was quantified (Fig. 3). Quantification of uncleaved substrate as a function of time reveals that the rate of Artemis/DNA-PKcs-mediated cleavage was three times faster in the presence of Ku than in its absence (Fig.  3A). As expected, bands corresponding to the trimmed overhangs were only detected when the internally labeled DNA end was released as an oligomer by treatment with TaqI, and nuclease activity was ATP-, DNA-PKcs-, and Artemisdependent (Fig. 3B). Notably, the addition of Ku also markedly altered the digestion pattern for the 3Ј overhang (Fig. 3B). In the absence of Ku, the predominant cleavage product was a 5-base overhang (40% of total cleavage), with significant amounts of 2-4-and 6-base overhang products also being apparent. In contrast, a 3-base overhang product consistently dominated the products in reactions containing Ku.
To investigate further this Kuinduced change in product distribution, reaction products were examined at various reaction times (Fig. 3C). In the presence of Ku, the proportion of 5-base overhang decreased from 56% of total cleavage at 2 min to only 6% at 60 min, whereas the proportion of 3-base overhang increased from 10 to 54%, becoming the dominant product by 60 min. Moreover, the appearance of the 3-base product was concomitant with the decrease in the 5-base product (Fig. 3D). These data are consistent with a two-step reaction, wherein the 5-base product is the substrate for a secondary nuclease cleavage yielding the 3-base product. Notably, in the absence of Ku this crossover pattern was not apparent (Fig. 3E), and the distribution of products was essentially invariant over the course of the reactions. In addition, this same effect was observed in titrations with limiting Ku, namely the appearance of the 3-base product accompanied by the loss of the 5-base product (as well as the 6-base product) as Ku concentrations increased (Fig. 3F). A more detailed curve-fitting analysis indicated that these kinetics could be accurately described by a two-step model, and that although Ku accelerated the initial generation of the 5-base Curve-fitting of these data to a two-step reaction indicated k 1 ϭ 0.11 min Ϫ1 and k 2 ϭ 0.09 min Ϫ1 in the presence of Ku, and k 1 ϭ 0.033 min Ϫ1 and k 2 ϳ0.005 min Ϫ1 the absence of Ku, where k 1 and k 2 are the rate constants for formation of the 5-base overhang and for its conversion to 3-base overhang, respectively (see supplemental Fig. 3). FEBRUARY 9, 2007 • VOLUME 282 • NUMBER 6 JOURNAL OF BIOLOGICAL CHEMISTRY 3551 overhang 3-fold, Ku increased the rate of its conversion to 3-base overhang by ϳ20-fold (supplemental Fig. 3).

Artemis Processing of 3-Phosphoglycolate Termini
3Ј-Phosphoglycolate Termini Alter Artemis Nuclease Specificity on Shorter Overhangs-3Ј-PG blocking groups did not alter Artemis/DNA-PK endonuclease activity when on long 3Ј overhangs distal to the cleavage site (Fig. 1). To determine the effect of 3Ј-PG closer to the site of nuclease activity, plasmid substrates with 3-6-base overhangs, bearing either 3Ј-PG or 3Ј-hydroxyl termini, were constructed (Fig. 4A) and subjected to treatment with Artemis/DNA-PK for 2-30 min. For a 6-base 3Ј-hydroxyl overhang substrate, little if any single-nucleotide trimming was observed; 2-base removal was predominant (60% of cleavage products at 5 min), with some 3-and 4-base removal also being evident (Fig. 4B). In contrast, the 3Ј-PG substrate showed, in addition to these products, subtle but detectable removal of a single 3Ј-PG nucleotide (Fig. 4C). However, this 5-base overhang product was evident only at early time points. At later times, the abundance of both the 5-base and 4-base overhangs diminished, with concomitant increases in the 3and 2-base overhangs (Fig. 4D). These results are consistent with a secondary cleavage that removes two additional bases, as seen above with the 13-base overhang substrate. The analogous hydroxyl-terminated substrate (Fig. 4B) also showed a time-dependent increase in the relative abundance of shorter reaction products, albeit less apparent than with the 3Ј-PG substrate. This is likely because the initial cleavage reaction proceeded more slowly (see supplemental Fig. 4).
The influence of 3Ј-PG on Artemis nuclease specificity was much more evident on substrate sets with 5-and 4-base overhangs (Fig. 5). For a 5-base 3Ј-hydroxyl overhang, two-nucleotide trimming was dominant (78% of cleavage products at 5 min), and single-nucleotide trimming was Ͻ5%, whereas for the analogous PG-terminated substrate single-nucleotide removal accounted for nearly half (48%) of the total initial cleavage (Fig. 5A). This single-nucleotide removal was 10-fold greater in the presence of Ku than in its absence, reflecting both an overall increase in cleavage and a change in cleavage specificity. The resulting 4-base overhang reached a maximum by 10 min and appeared to decrease slightly between 10 and 30 min, whereas shorter reaction products continued to accumulate. Just as with the 13-base overhang, this shift from longer to shorter overhangs at 10 -30 min was seen only in those reactions containing Ku, consistent with Ku-dependent secondary cleavage of a portion of the longest initial product; without Ku, the three cleavage products accumulated in parallel (Fig. 5A,  graphs). At longer times, there was little change in the overall cleavage pattern, although traces of shorter fragments (*), indicating trimming into duplex DNA, began to accumulate (Fig.  5A, top right). These data do not distinguish whether this late trimming was endo-or exonucleolytic, but subsequent experiments wherein a blunt-ended substrate was cleaved at the same sites (see Fig. 7) indicated the same dependence on activated DNA-PK.
A similar effect of PG termini on nuclease specificity was apparent with substrates having 4-base overhangs (Fig. 5B), i.e. almost exclusively (88%) 2-base trimming for the hydroxyl-terminated substrate, but approximately equal 1-base and 2-base trimming for the PG-terminated substrate. Unlike the longer overhangs, the hydroxyl-and PG-terminated 4-base overhang substrates did not show loss of longer reaction products or a significant change in product distribution over time, suggesting that the 2-and 3-base overhangs are poor substrates for secondary Artemis-mediated cleavage even in the presence of Ku. Inasmuch as there was almost no 3-base overhang produced from the hydroxyl-terminated 4-base substrate, the 3-base overhang generated from the 5-base PG-terminated substrate must have arisen by 2-nucleotide trimming of the initial substrate.
A titration with limiting Artemis (Fig. 5C) revealed that cleavage of the hydroxyl-terminated 5-base substrate by Artemis/DNA-PK shows the same 2-base removal specificity, as well as nearly complete Ku dependence, over a range of Artemis concentrations. When Artemis was limiting (5.6 nM), this cleavage was 17 times greater in the presence of Ku (26 Ϯ 2%, n ϭ 4) Internally labeled plasmid substrates bearing 6-base hydroxyl-or PG-terminated 3Ј overhangs were treated with Artemis and then cut with TaqI. Reaction conditions are the same as in Fig. 3. A, general structure of short 3Ј overhang substrates showing restriction sites used for analysis. B, cleavage of a 6-base 3Ј-hydroxyl overhang. C, cleavage of a 6-base 3Ј-PG overhang. Arrows above the sequence indicate initial cleavage sites. Numbers to the right indicate overhang length. D, abundance of 5-base (F), 4-base (f), 3-base (OE), and 2-base () overhang products was quantitated following treatment of the 3Ј-PG substrate. Error bars show range of values obtained in two independent experiments, when larger than the symbols. See supplemental Fig. 4 for additional kinetic data. than in its absence (1.5 Ϯ 0.1%, n ϭ 3), and increasing the Artemis concentration more than 20-fold did not fully compensate for the lack of Ku (Fig. 5C).
Finally, when substrates with 3-base overhangs were treated with Artemis, there was almost no processing of the hydroxylterminated substrate, whereas the PG-terminated substrate was processed almost exclusively (85-90%) by single-nucleotide removal, 90% of which was Ku-dependent (Fig. 6, A and B). An analogous phosphotyrosyl-terminated substrate, typical of strand breaks induced by topoisomerase I inhibitors (20), was likewise processed primarily by single-nucleotide removal, albeit 3-fold less efficiently (Fig. 6, A and B). To verify the terminal structure of the putative 2-base overhang product generated by removal of the terminal 3Ј-PG nucleotide, this product was subsequently treated with terminal transferase plus ddGTP, which produced the expected ϩ1 nucleotide shift, demonstrating that the Artemis reaction product has a 3Ј-hydroxyl rather than a 3Ј-PG terminus (Fig. 6C, lane 6). Moreover, both the putative 14-mer and the resulting 15-mer precisely comigrated with authentic markers of the predicted sequence. As expected, the initial 3Ј-PG substrate was unaffected by treatment with terminal transferase, confirming that it had a blocked 3Ј terminus (Fig. 6C, lane 5). As expected for Artemismediated cleavage, the band corresponding to the 2-base overhang product was only seen when Artemis, DNA-PKcs, and ATP were present in the reaction and when the substrate was subsequently cleaved with AvaI or TaqI (Fig. 6D).
Thus, Artemis nuclease trims short hydroxyl-terminated 3Ј overhangs predominantly by removal of two terminal nucleotides, whereas 3Ј-PG termini (and to a lesser extent 3Ј-phosphotyrosyl) alter this specificity, promoting trimming of a single nucleotide. For unmodified substrates, the minimum length  FEBRUARY 9, 2007 • VOLUME 282 • NUMBER 6 of overhang that will support efficient Artemis-mediated cleavage is 4 bases.

Artemis Processing of 3-Phosphoglycolate Termini
Slow Trimming of 3Ј-PG-terminated Blunt Ends-To further investigate the specificity of Artemis/DNA-PK, a series of plasmid substrates bearing a 3Ј-PG terminus on a 2-base overhang, a blunt end, or a 2-base recessed 3Ј end were generated and assayed for susceptibility to Artemis nuclease (Fig. 7A). Although all substrates were subject to some degree of digestion, there was a pronounced decrease in nuclease efficiency and specificity for PG-terminated substrates having overhangs of less than 3 bases. For example, a 30-min Artemis treatment of a substrate with a PG-terminated blunt end or 2-base 3Ј overhang yielded small amounts of multiple cleavage products, each of which accounted for Ͻ10% of the initial substrate (Fig. 7, A  and B). To assess the requirements, specificity, and extent of this processing, nuclease assays with the 3Ј-PG blunt-ended plasmid substrate were carried out for up to 2 h (Fig. 7B). Although control reactions with Ku and/or DNA-PKcs alone showed no significant trimming, those containing Artemis indicated time-dependent trimming corresponding to removal of 2-4 bases from the 3Ј-PG-terminated blunt ends (Fig. 7B).
Although this trimming proceeded about seven times more slowly than trimming of a 5-base 3Ј overhang, by 2 h nearly half (43 Ϯ 3%, n ϭ 3) of the blunt ends were trimmed (Fig. 7, C and   D) in an Artemis-, DNA-PKcs-, and ATP-dependent manner. The addition of Ku stimulated this activity 5-fold but did not markedly alter the product profile. Thus, this processing, although slower and having different cleavage specificity than that of 3Ј overhangs, exhibited the same protein and cofactor requirements. Similar results (not shown) were obtained with a PG-terminated 1-base 3Ј overhang.
To assess whether the unlabeled 5Ј-terminal strand was still intact following trimming of the 3Ј terminus, similar experiments were performed wherein the labeled oligomer was released by treatment with AvaI, which requires a double strand substrate. AvaI yielded essentially the same product profile as TaqI, which can cleave both single-and double-stranded DNA (Fig.  7E). This result implies that, at least for the majority of trimmed molecules, both strands were still intact at the AvaI site and that trimming of the 3Ј end was accompanied by either no 5Ј 3 3Ј exonucleolytic resection or trimming of at most a few (Ͻ10) nucleotides from the 5Ј terminus. Thus, Artemis plus DNA-PK is clearly able to process blunt 3Ј-PG ends, albeit more slowly than short overhangs, with removal of a few adjacent nucleotides.
Toxicity of 3Ј-Phosphoglycolate-terminated DSBs in Artemisdeficient Cells-The above data clearly show that Artemis can process 3Ј-PG-terminated DSBs. However, although hypersensitivity of Artemis-deficient mouse fibroblasts to bleomycin has been reported (21), a role for Artemis in processing of PGterminated DSBs in intact human cells has not been established. To address this question, we assayed the toxicity of two drugs that induce different types of 3Ј-PG-terminated DSBs in vivo. Bleomycin treatment gives rise to DSBs, nearly all of which have either blunt ends or single-base 5Ј overhangs, with 5Ј-phosphate and 3Ј-PG termini at both ends of the break. Neocarzinostatin-induced DSBs have at one end a 5Ј-phosphate and a 3Ј-phosphate on a 2-base 3Ј overhang; the opposite end has a 5Ј-aldehyde and either a 3Ј-PG (ϳ75%) or a 3Ј-phosphate (ϳ25%) on a 1-base 3Ј overhang (14,15). Following treatment of exponentially growing cells with each of these agents, bromodeoxyuridine incorporation into Ն10,000 cells was measured ϳ24 h after treatment, and from these data the proliferating fraction of cells was calculated (Fig. 8).
We previously showed that fibroblasts from Artemis-deficient SCIDA patients exhibit significantly elevated sensitivity to x-irradiation but a much lesser sensitivity to DSBs induced by experiments with each substrate, when larger than the symbols. C, terminal transferase-mediated extension of the 3Ј-hydroxyl terminus generated by Artemis-mediated trimming of a 3Ј-PG-terminated 3-base overhang. Following treatment with Artemis and AvaI as in A, samples were treated with terminal transferase (TdT) plus ddGTP. Lanes marked M contain 5Ј-end-labeled 14-or 15-base fragments of the expected sequence (TCGAGGAACGC-GAC and TCGAGGAACGCGACG, respectively). For the 3Ј-PG substrate, essentially identical results were obtained when TaqI was used instead of AvaI (data not shown). D, requirements for single-nucleotide trimming. The PG-terminated 3-base overhang was treated with Artemis plus DNA-PK for 30 min as in A, but various reaction components were omitted as indicated.
etoposide (6). Here we find that human telomerase reverse transcriptase-immortalized SCIDA fibroblasts show hypersensitivity to the DSB-inducing agent bleomycin, albeit not as great as their sensitivity to x-rays (Fig. 8A). Interestingly, DSBs produced by neocarzinostatin (ϳ75% of which bear an overhanging 3Ј-PG terminus at one end of the break) appear to be nearly as toxic to SCIDA cells as x-rays ( Fig. 8A and Table 1). Similarly, treatment with x-rays, neocarzinostatin, and bleomycin, all of which induce 3Ј-PG-terminated DSBs, results in at least 2-fold greater accumulation in G 2 /M in cycling SCIDA cells than in normal cells (Fig. 8B). Comparison of the relative toxicities of these three agents in multiple experiments reveals that the hypersensitivity of SCIDA cells is about the same for x-rays and neocarzinostatin but less for bleomycin (Table 1). Together, these data suggest that x-rays and radiomimetic drugs induce DSBs or other lesions that require Artemis for resolution prior to continued cellular proliferation.

DISCUSSION
Data presented here show that in the context of a variety of model DSB substrates, purified histidinetagged Artemis efficiently cleaves long 3Ј overhangs to a length of 4 -5 nucleotides, in a reaction dependent on ATP and catalytically active DNA-PKcs. Our reactions with soluble purified proteins also reveal significant stimulation of Artemis activity by Ku for all substrates tested (Figs. 2, 3, and 5-7). Inasmuch as Ku is normally required for efficient DNA-PK assembly and self-activation on DNA ends, except at very low ionic strength (22,23), and catalytically active DNA-PK is required to activate Artemis nuclease, it is expected that Ku would have a stimulatory effect on Artemis-mediated DNA cleavage. The reported Ku independence for immobilized Artemis activity (3) is likely because of the presence of excess DNA ends in those reactions, which in our hands almost completely suppresses the stimulatory effect of Ku (Fig. 2).
At low DNA terminus concentrations (1-5 nM), a Ku-mediated recruitment of Artemis/DNA-PKcs to DNA ends should markedly improve the reaction rate, as is observed (Fig. 2D; Fig. 3, A and C). At high DNA terminus concentrations (1 M), the effect of any increase in affinity will be at least partially abrogated by increased binding to unlabeled competitor DNA, and the increase in cleavage will be less pronounced, as seen in Fig. 2. The finding that unlabeled 35-bp duplex does not compete with labeled substrate in the absence Reaction conditions are the same as in Fig. 3. A, indicated 3Ј-PG (F) substrates were treated with Artemis plus DNA-PK for 30 min and cut with AvaI. B, blunt end 3Ј-PG substrate was treated with Artemis plus DNA-PKcs for the indicated times in the presence or absence of Ku and then cut with TaqI. C, time course for overall cleavage of the blunt and 5-base overhang 3Ј-PG substrates. The fraction of unprocessed substrate was calculated. D, time course and specificity of product formation for the blunt 3Ј-PG substrate. Abundance of products corresponding to the removal of 2 (F), 3 (f), or 4 (OE) bases was quantitated following treatment of the blunt-ended 3Ј-PG substrate with Artemis plus DNA-PK. Error bars show means Ϯ S.E. of three independent experiments when larger than the symbols. E, same as B, except that following incubation with Artemis, half of each sample was treated with AvaI and half was treated with TaqI.
of Ku, but is an effective competitor in the presence of Ku ( Fig.  2A), also suggests that the Artemis-DNA-PK complex has higher affinity for blunt ends in the presence of Ku than in its absence. Conversely, the finding that cleavage of very short overhangs is more strongly dependent on Ku than is cleavage of longer overhangs (compare Figs. 2F and 5C) is consistent with a model wherein contacts between Artemis and the long overhang can enhance affinity and thereby partially compensate for the lack of Ku. It is also notable that excess DNA ends, which should stimulate global DNA-PK kinase activity, do not stimulate Artemis nuclease activity (Fig. 2, A, C, and D). Rather, optimal nuclease activity is seen with the lowest concentrations of DNA ends, consistent with a model in which Ku and autophosphorylated DNA-PKcs form a stable complex with long 3Ј overhangs (24), so that only local activation of DNA-PK on a given DNA end is important for Artemis nuclease activation.
In the case of a long (13-base) overhang on a plasmid substrate, the presence of Ku not only stimulated the cleavage reaction but also markedly altered the product distribution (Fig. 3). The data are most consistent with a two-step process, wherein the initial cleavage removes the extended 3Ј single strand tail, leaving a 5-base overhang which, only in the presence of Ku, is subsequently trimmed by an additional 2 bases. Consistent with this idea, direct treatment of a 5-base overhang substrate with Artemis under the same conditions resulted almost exclusively in 2-base trimming to a 3-base overhang, in a reaction that was almost completely dependent on Ku (Fig.  5C). Other substrates (Figs. 4C and 5A) show, to varying degrees, similar progression from longer to shorter cleavage products with time, and although time-dependent changes in Artemis specificity cannot be formally ruled out, all of the data can be explained by the same two-step mechanism wherein a portion of the initial cleavage products are subsequently trimmed, 2 bases at a time, until the overhang length is reduced to 3 bases or less.
Although 3Ј-blocking PG moieties on long 3Ј overhangs have no discernible influence on Artemis endonuclease activity, the presence of 3Ј PG closer to the site of cleavage significantly alters Artemis nuclease specificity, promoting single-nucleotide removal from short overhangs. Most strikingly, Artemis trims a 3-base 3Ј-PG overhang almost exclusively by single-nucleotide removal, while leaving an analogous 3Ј-hydroxyl substrate largely intact (Fig. 6).
Based on these and other results (Figs. 1-7), the site of Artemis-mediated cleavage of 3Ј overhangs appears to be dependent on the following three criteria: 1) a requirement for either 2 nucleotides or a nucleotide plus a 3Ј-PG 3Ј to the cleavage site; 2) a requirement for at least 2 unpaired nucleotides 5Ј to the cleavage site; and 3) a preference for cleavage 4 -5 bases from the single strand/double strand transition. It is likely that for long overhangs, DNA-PK at the double strand/single strand transition positions the Artemis active site for cleavage 4 -5 bases from the transition. However, on shorter overhangs this positioning is preempted by the more stringent requirement for 2 nucleotides 3Ј to the cleavage site, thus shifting the cleavage site closer to the double strand/single strand transition. A 3Ј-PG can apparently partially substitute for the terminal nucleotide, resulting in single-nucleotide removal from short PG-terminated overhangs. Unmodified overhangs shorter than 4 bases, and PG-terminated overhangs shorter than 3 bases, are much less efficiently processed, likely due to the fact that the requirements for 2 bases 3Ј and 2 unpaired bases 5Ј to the lesion cannot be simultaneously satisfied. Nevertheless, very short overhangs, blunt ends, and even recessed ends bearing 3Ј-PG termini show significant Artemis-mediated processing (Fig. 7), albeit at a much slower rate than longer overhangs, and after 2 h the 5-base overhang also shows traces of trimming into duplex regions. Overall, the observed patterns of cleavage are consist- termini. An estimated 75% of neocarzinostatin-induced DSBs have the structure shown, whereas the remainder have a 3Ј-phosphate rather than an 3Ј-phosphoglycolate. Radiation-induced DSBs will have short 5Ј or 3Ј overhangs of varied length with predominantly 5Ј-phosphate termini and ϳ50% 3Ј-PG termini. B, cell cycle distribution of BrdUrd-labeled cells 29 h after treatment with 5 gray (Gy) x-rays, 10 nM neocarzinostatin, or 8 g/ml bleomycin. ent with "iterative" models of DSB repair (17), in which blocked or incompatible ends are subjected to sequential rounds of Artemis-mediated trimming, until a pair of ends amenable to patching and ligation is generated.
We and others have shown that the majority of radiationinduced DSBs can be rapidly rejoined even in the absence of Artemis (4 -6). However, at both 6 and 24 h after irradiation, Artemis-deficient cells exhibit ϳ2-fold more residual ␥-H2AX foci/nucleus than normal cells, suggesting that a subpopulation of DSBs in these cells is resistant to repair in the absence of Artemis (6). Pseudo-epistasis studies, using a combination of kinase inhibitors and genetic defects, suggest that repair of these breaks requires not only Artemis and DNA-PK but also the ataxia telangiectasia-mutated kinase as well as proteins typically found in DSB repair foci such as 53BP1 and the Mre11-Rad50-NBS1 complex (5). It is presently unclear what types of DSBs strictly require Artemis, but the finding that a greater proportion of ␣-particleinduced DSBs (ϳ20%) than of x-ray-induced DSBs (ϳ10%) require Artemis for repair led to the suggestion that chemical complexity of the breaks may be a factor (5). However, unlike radiation, the radiomimetic agents neocarzinostatin and bleomycin are not thought to induce any locally multiply damaged sites, such as DSBs accompanied by nearby base damage (14,15). Yet, SCIDA cells show hypersensitivity to neocarzinostatin and bleomycin as well as to x-rays, suggesting that 3Ј-PG and/or 5Ј-aldehyde termini alone at a DSB may represent relevant in vivo substrates for Artemis. Notably, the relative sensitivity of SCIDA cells to these agents does not correlate directly with the relative induction of PG termini. Although bleomycin induces DSBs with nearly 100% PG termini (14,15), these PGs are on blunt and 3Ј-recessed DSBs (Fig. 8) and so could be removed, albeit inefficiently, by Ape1. Conversely, although only ϳ75% of DSBs induced by neocarzinostatin bear PG termini, and then at only one end of the break (14,15,25), these PGs on 1-base 3Ј overhangs will be refractory to Ape1 (26), as will PG termini of any x-ray-induced DSBs that have 3Ј overhangs. Although radiation-induced DSBs are much more heterogeneous in both chemistry and geometry than those induced by neocarzinostatin, the fraction of DSB ends with protruding 3Ј-PG termini is predicted to be comparable, i.e. ϳ25% for radiation (the proportion of breaks with PG termini (ϳ50%) times the proportion with 3Ј overhangs (ϳ50%)) versus ϳ37% for neocarzinostatin (one end of ϳ75% of the breaks). Within experimental error, this prediction correlates with the similar sensitivity of SCIDA cells to these two agents. Alternative PG removal from the blunt and 3Ј-recessed bleomycin-induced DSBs by Ape1 could account for the milder sensitivity of SCIDA cells to bleomycin ( Table 1).
The only other enzyme known to be capable of processing 3Ј-PG termini is TDP1 (9), which has substrate preferences opposite those of Ape1, acting more efficiently on 3Ј overhangs. However, processing of PG termini by TDP1 is also relatively inefficient, at least compared with its canonical 3Ј-phosphotyrosyl substrate (9), and TDP1 mutant cells are at best only marginally sensitive to x-rays (10). Thus, PG termini on recessed 3Ј ends, blunt ends, and very short 3Ј overhangs are not particularly favorable substrates for any of the three candidate enzymes (Ape1, TDP1, and Artemis), and of the three, only Artemis has been shown to interact with and be stimulated by other components of the nonhomologous -joining pathway. Although in vitro studies (Fig. 7) indicate that Artemis trims PG-terminated blunt ends very slowly (1-2 h), these kinetics are consistent with the slow time course of Artemis-dependent DSB repair in intact cells (5). Overall, the moderate hypersensitivity of SCIDA cells to radiation and radiomimetic drugs is consistent with Artemis providing an important pathway, but not the sole pathway, for repair of PG-terminated DSBs.
Alternatively, rather than being strictly required for processing 3Ј-PG DSBs, a more critical function for Artemis may be the resolution of derivative structures that can arise from them. In principle, a 3Ј-PG terminus will block gap filling and ligation but allow 5Ј 3 3Ј resection by the Mre11-Rad50-NBS1 complex or other nucleases. Artemis, in the presence of DNA-PK, is capable of trimming any resulting blocked 3Ј overhangs while they are still relatively short, at which point gap filling and ligation can proceed. However, blocked 3Ј termini not resolved at this point may result in more extensive 5Ј resection. Such resection in the context of an inverted repeat may lead to the formation of hairpin-like termini via annealing of the 3Ј overhangs. This sequence of events would convert a fraction of 3Ј-blocked DSBs that were initially repairable by DNA-PK, TDP1, polynucleotide kinase/phosphatase, polymerase , and ligase IV (8) to hairpin structures, akin to V(D)J recombination intermediates, that strictly require Artemis for resolution. Thus, a failure to resolve 3Ј-PG structures at an early stage in the repair pathway could lead to an increased incidence of repair-resistant derivative structures, either the long 3Ј overhangs themselves (25) or resulting hairpins. The observed accumulation of cycling Artemis-deficient cells in G 2 /M is consistent with the evolution of unresolvable derivative DNA a Ratio of the toxicity of a given x-ray dose or drug concentration toward SCIDA cells to the toxicity of the same treatment in normal cells. Toxicity is defined as the change in ln(F), where F is the proliferating fraction of cells. b Mean Ϯ S.E. or range when n ϭ 2. n ϭ number of x-ray doses or drug concentrations analyzed. Treatments that reduced the proliferating fraction by less than 10% in normal cells were excluded. c Mean Ϯ S.E. of the average values obtained in the three independent experiments shown. d Significance of differences between SCIDA and normal cells by Student's t test. e The hypersensitivity of SCIDA 05 cells to bleomycin was less than its hypersensitivity to x-rays (p Ͻ 0.05) or neocarzinostatin (p Ͻ 0.03). For SCIDA 04 cells these differences were not significant (p ϳ 0.08).
structures in these cells. This mechanism would also account for the relative lack of hypersensitivity of SCIDA cells to etoposide (5,6), as etoposide-induced DSBs have 5Ј termini blocked with covalently bound topoisomerase II and would therefore not be subject to 5Ј resection. Although there is evidence that Artemis is phosphorylated by ataxia telangiectasia-mutated, it is unlikely that hypersensitivity is because of a defect in cell cycle checkpoints, as Artemis deficiency does not confer a significant checkpoint defect (6,27). Given that Artemis has no known activities other than DNA end-processing, and given that the radiomimetic compounds used here specifically induce DNA breaks and abasic sites almost exclusively, it seems unlikely that hypersensitivity is because of a defect wholly unrelated to DSB processing. Nevertheless, the possibility that efficient repair of some radiomimetic induced DSBs requires Artemis for reasons other than removal of chemically modified termini cannot be entirely excluded.
In summary, our results demonstrate that Artemis has the capability to carry out an essential step in repair of radiationinduced DNA damage, elimination of the 3Ј-PG blocking lesions commonly found at DSB termini. Although Artemis lacks the specificity to remove only the 3Ј-PG, it can nevertheless resolve this type of lesion by excising one or a few terminal nucleotides along with the contiguous 3Ј-PG moiety. This Artemis endonuclease activity is dependent on DNA-PKcs and stimulated by Ku, two core nonhomologous end-joining proteins. Moreover, like x-rays, agents that induce well characterized PG-terminated DSBs induce elevated toxicity in Artemisdeficient human cells. All of these findings are consistent with the proposal that the processing of PG-terminated DSBs, and/or derivative structures arising from them, is a biologically relevant activity of Artemis.