Unifying the DNA End-processing Roles of the Artemis Nuclease

Background: Artemis is a nuclease that is necessary for hairpin opening in V(D)J recombination. Results: Artemis action on blunt DNA ends is dependent on DNA sequence (breathing) and Ku. Conclusion: Breathing of blunt DNA ends into a transient ss/dsDNA boundary is needed for Artemis action. Significance: Unification of Artemis nuclease action explains the features of NHEJ and V(D)J recombination. Artemis is a member of the metallo-β-lactamase protein family of nucleases. It is essential in vertebrates because, during V(D)J recombination, the RAG complex generates hairpins when it creates the double strand breaks at V, D, and J segments, and Artemis is required to open the hairpins so that they can be joined. Artemis is a diverse endo- and exonuclease, and creating a unified model for its wide range of nuclease properties has been challenging. Here we show that Artemis resects iteratively into blunt DNA ends with an efficiency that reflects the AT-richness of the DNA end. GC-rich ends are not cut by Artemis alone because of a requirement for DNA end breathing (and confirmed using fixed pseudo-Y structures). All DNA ends are cut when both the DNA-dependent protein kinase catalytic subunit and Ku accompany Artemis but not when Ku is omitted. These are the first biochemical data demonstrating a Ku dependence of Artemis action on DNA ends of any configuration. The action of Artemis at blunt DNA ends is slower than at overhangs, consistent with a requirement for a slow DNA end breathing step preceding the cut. The AT sequence dependence, the order of strand cutting, the length of the cuts, and the Ku-dependence of Artemis action at blunt ends can be reconciled with the other nucleolytic properties of both Artemis and Artemis·DNA-PKcs in a model incorporating DNA end breathing of blunt ends to form transient single to double strand boundaries that have structural similarities to hairpins and fixed 5′ and 3′ overhangs.

In dividing mammalian cells, there are an estimated 10 DNA double-stranded breaks (DSBs) 2 per day per cell (1)(2)(3). These pathological DSBs arise from ionizing radiation, oxygen free radicals, replication errors, and inadvertent cleavage by nuclear enzymes (4). Many of these pathological breaks and the physiological breaks in V(D)J recombination require the nonhomologous DNA end-joining (NHEJ) nuclease Artemis to repair the DNA. In NHEJ, the DSB is first recognized by the Ku heterodimer (Ku70/80), which acts as a "tool belt" to which other NHEJ proteins are recruited as needed. When DNA resection is required, a DNA-dependent protein kinase catalytic subunit (DNA-PKcs) is recruited in complex with Artemis. DNA-PKcs undergoes autophosphorylation and activates Artemis endonuclease activity (5,6). Activated Artemis then gains the ability to cut many DNA substrates at the single to double strand DNA boundaries (7) (Fig. 1).
Artemis is in the metallo-␤-lactamase family of nucleases, characterized by conserved metallo-␤-lactamase and ␤-CASP domains. This family of nucleases is able to hydrolyze DNA or RNA in various configurations (8). Artemis alone has intrinsic 5Ј exonuclease activity on single-stranded DNA (ssDNA) (9). On duplex DNA, Artemis, in complex with DNA-PKcs, has endonuclease activity on the 5Ј and 3Ј DNA overhangs that are often created by pathological DNA breaks and on DNA hairpins that are formed during V(D)J recombination. Therefore, patients lacking Artemis suffer from severe combined immunodeficiency because of a defect in antibody formation (10).
We have determined biochemically that Artemis, in complex with DNA-PKcs, resects 5Ј and 3Ј DNA overhangs to create DNA end structures that are able to be ligated by the DNA ligase IV-XRCC4 complex (11,12). At 5Ј overhangs, Artemis cuts directly at the ss/dsDNA boundary. However, when processing 3Ј overhangs and DNA hairpins, it preferentially leaves a 4-nt overhang (7). Perfect DNA hairpins have sterically constrained tight turns and, therefore, have ss/ds boundaries at the last two base pairs (4 nt) (13). This may make them structurally similar to DNA overhangs. From these observations, we proposed that Artemis activity on duplex DNA can be unified under a model in which Artemis⅐DNA-PKcs binds to the ss/ds-DNA boundary to occupy 4 nt along the single-stranded segment. This binding is followed by preferentially nicking the 3Ј side of those 4 nt. Although the current model is extensive, it lacks information on whether Artemis can cut blunt-ended DNA that can be generated by chemotherapeutic agents, free radicals, or ionizing radiation (14).
Here we explore the action of Artemis at blunt DNA ends, which appears substantially weaker than its action at ss/dsDNA boundaries. We found that resection at blunt end depends on the DNA sequence of the duplex end, with AT-rich DNA ends being resected by Artemis alone but GC ends hardly at all. When Artemis is accompanied by DNA-PKcs, it can act at all blunt ends, but in a Ku-dependent manner. This is the first documentation showing that Ku can modulate Artemis activity. The sequence dependence and other features of Artemis action permit the unification of all nucleolytic properties of Artemis into a single model for its action at ss/dsDNA boundaries.

Experimental Procedures
Oligonucleotides and DNA Substrates-The Oligonucleotides used in this study were synthesized by Integrated DNA Technologies, Inc. (San Diego, CA). The oligonucleotides were purified using denaturing PAGE, and the concentrations were determined spectrophotometrically. 5Ј end labeling of DNA substrates was done with [␥-32 P]ATP (3000 Ci/mmol) (PerkinElmer Life Sciences) and T4 polynucleotide kinase (PNK) (New England Biolabs) according to the instructions of the manufacturer. Unincorporated radioisotope was removed by using Sephadex G-25 spin columns (Enzymax). To create dsDNA, the unlabeled complementary strand was added, and the sample was boiled for 5 min to inactivate T4 PNK. The DNA was allowed to slowly cool to room temperature and then incubated overnight at 4°C to promote proper base pairing of the complementary strand.
3Ј end labeling was performed by fill-in synthesis of [␣-32 P]dNTP (3000 Ci/mmol) (PerkinElmer Life Sciences) onto preannealed dsDNA with a 1-nt 5Ј overhang with the Klenow fragment of DNA polymerase I (3Ј 3 5Ј exo Ϫ ) (New England Biolabs) to create a blunt-ended substrate. Unincorporated radioisotope was removed as stated above.
Biochemical Nuclease Assay-The in vitro DNA nuclease assays were performed in a volume of 10 l with a buffer composition of 25 mM Tris-HCl (pH 8.0), 75 mM KCl, 10 mM MgCl 2 , and 1 mM DTT. The reactions consisted of 20 nM of 32 P-labeled DNA incubated with 0.5 mM ATP (or 0.5 mM AMP-PNP), 50 nM Ku, 50 nM DNA-PKcs, and 50 nM Artemis (or 50 nM ARM14) at 37°C for 30 min unless specified otherwise. The time course assays were performed in a volume of 12 or 15 l. The ladders were created by incubating 40 nM 32 P-labeled ssDNA with 0.5 milliunits/ml snake venom phosphodiesterase I (Sigma-Aldrich, catalog no. P3243) for 15 min at 37°C. All reactions were stopped by the addition of an equal volume of 98% formamide, heated to 95°C for 5 min, placed on ice for 5 min, and analyzed on a 12 or 14% denaturing PAGE. The gels were then dried and exposed to a phosphorimaging screen overnight. The screen was scanned, and quantification was performed in Quantity One (Bio-Rad).

Artemis Can Resect Blunt DNA Ends But at a Much Slower
Rate Than at DNA Overhangs-In Fig. 1, we show a summary of major Artemis substrates and preferred positions of Artemis cutting on those substrates. The open thick arrows indicate the putative orientation and position of the Artemis enzyme, and the red arrows indicate the preferred cutting sites. Artemis is known to have endonuclease activity on DNA hairpins and on 5Ј and 3Ј overhangs of duplex DNA, which is dependent on DNA-PKcs activity ( Fig. 1, top row). Previous biochemical data have indicated that Artemis preferentially cuts on the 3Ј side of a 4-nt ssDNA segment directly adjacent to a ss/dsDNA boundary (7,11). The other known Artemis substrates listed in Fig. 1 (center and bottom rows) are physiologically relevant because they may arise from replication errors or by inadvertent action by nuclear enzymes. Nevertheless, we found that Artemis is able to cut these substrates on the ssDNA portion nearest the ss/dsDNA boundaries. We then wondered what would occur at blunt DNA ends when such ends may be able breathe open to form a transient pseudo-Y structure (Fig. 1, center row and center column).
We initially performed a time course experiment with Artemis and DNA-PKcs on 9-nt 5Ј overhang, blunt-ended, and ssDNA substrates (Fig. 2). Predictably, the predominant cut on the 5Ј overhang substrate is at 9 nt, producing a blunt-ended product ( Fig. 2A, lanes 1-6). The 34-nt ssDNA substrate (which matched the sequence of the top strand of the duplex portion of the other substrates) produced random endonucleolytic cuts ( Fig. 2A, lanes 13-18). In addition, the apparent intense band at ϳ18 nt may be explained by the palindromic 5Ј-TGCA-3Ј on the 11th-14th nt, allowing the substrate to selfanneal. The 18th nt is then 4 nt on the 3Ј side of the substrate, resulting in a more intense band. In contrast, the 34-bp bluntended substrate produced a distribution of cuts 1-6 nt into the duplex DNA end ( Fig. 2A, lanes 7-12).
The cut products were then quantified as a percentage of the total substrate and plotted as a function of time (Fig. 2B). The rate of activity was determined by calculating the slope from the average of two entirely independent experiments (product percent per minute of reaction). (Several additional very similar experiments further supported these observations.) The 5Ј overhang substrate was processed ϳ3-fold faster than bluntended DNA and 2-fold faster than ssDNA (Fig. 2C). Although it is known that Artemis has endonuclease activity on ssDNA (6), it was surprising to observe that the activity was only 2-fold slower than at overhangs. Unlike homopolymeric ssDNA substrates, we suspect that the ssDNA substrate used here may have self-annealed to produce partial duplex DNA structures.
Artemis Endonuclease Activity on Blunt DNA Ends Is Strictly Dependent on DNA-PKcs and ATP and Is Modulated by Ku-As mentioned, Artemis has intrinsic 5Ј exonuclease activity and DNA-PKcs-dependent endonuclease activity. Therefore, we wondered whether this activity also applied to blunt-ended DNA. We incubated a 5Ј-radiolabeled 34-bp dsDNA substrate flanked with dTs in a reaction with Ku, DNA-PKcs, Artemis, and ARM14 (catalytically inactive Artemis H115A ) with or without ATP for 30 min at 37° (Fig. 3). Intrinsic Artemis 5Ј exonuclease activity was able to resect the first 5Ј nucleotide in the absence of DNA-PKcs and ATP (Fig. 3, lanes 2, 4, and 7). Fur-FIGURE 1. Optimal substrates for Artemis have ss/dsDNA boundaries that direct Artemis to the point of preferred nuclease action. Shown are the known major substrates for Artemis (7,11). The putative orientation and position of the Artemis enzyme are indicated by the open thick arrows, consistent with the preference of Artemis for ss/dsDNA boundaries and its propensity to cut on the 3Ј side of an ϳ4-nt ssDNA region at this boundary. The preferred cutting sites of Artemis are noted with red arrows. We propose that weaker activity may be seen on variants of these structures when conditions favor DNA breathing into one of the forms shown here. In addition to the substrates shown, Artemis has 5Ј exonuclease activity at 5Ј protruding DNA ends (9), which is the only activity that does not rely on DNA-PKcs (under Mg 2ϩ conditions) (see text). Artemis also has very weak endonuclease activity on ssDNA, which is stimulated by DNA-PKcs and can be slightly stronger for ssDNA that can self-anneal or fold back at internal positions because of self-complementarity (6). thermore, Artemis 5Ј endonuclease activity predominately cut 1-6 nt from the 5Ј end in the presence of DNA-PKcs (Fig. 3, lane 7 versus lane 9), and this activity was dependent on ATP (Fig. 3

, lane 4 versus lane 9).
Interestingly, Ku modulated Artemis endonuclease activity at blunt ends by restricting the cuts to 2 and 4 nt instead of the 1-to 6-nt distribution observed in Ku-independent reactions (Fig. 3, lane 9 versus lane 10, 5Ј end resection). The addition of Ku also stimulated 3Ј endonuclease activity by enabling Artemis to cut up to 9 nt into the 3Ј end (Fig. 3, lane 9 versus lane 10, 3Ј end resection). This 3Ј activity remained DNA-PKcs-dependent (Fig. 3, lane 8 versus lane 10, 3Ј end resection). Furthermore, the observed activities are not due to contaminating nucleases because the catalytically inactive Artemis point mutant, ARM14, which was purified in the same manner, did not show any 5Ј or 3Ј activity (Fig. 3, lanes 10 and 11). We also confirmed that, similar to Artemis endonuclease activity on overhangs and ssDNA (5, 6), Artemis phosphorylation is not required for its endonuclease activity on blunt-ended DNA (data not shown), although autophosphorylated DNA-PKcs is required for all ends.
Titration Reveals That DNA-PKcs Can Be Inhibitory at High Concentrations-Next we examined the effect of the amount of DNA-PKcs on the observed activity of Artemis. We generated a 5Ј-labeled, 34-bp, blunt-ended DNA by annealing 5Ј-labeled ssDNA to its unlabeled complementary strand containing a 5Ј biotin (Fig. 4). This substrate was then incubated along with streptavidin, which binds biotin tightly, to prevent/inhibit Artemis action on one end of the dsDNA to examine the activity on the labeled end only. Using a blunt-ended substrate flanked with four dAs, we titrated the amount of DNA-PKcs in the reaction from 2 to 105 nM (Fig. 4). Ku and DNA-PKcs alone did not have any detectable levels of nuclease activity (Fig. 4,  lanes 2 and 3). Artemis activity was maximal at a 1:1 molar ratio with DNA-PKcs (Fig. 4, lanes 6 -9).
Furthermore, Artemis activity was stimulated ϳ2to 3-fold in the presence of Ku (Fig. 4, lanes 6 -9 versus lanes 10 -13). Interestingly, increasing the molar ratio of DNA-PKcs to Arte-  OCTOBER 2, 2015 • VOLUME 290 • NUMBER 40 mis to 2:1 reduced Artemis activity by ϳ50% (Fig. 4, lane 8 versus lane 9 and lane 12 versus lane 13). These data support the view that Artemis acts in a complex with DNA-PKcs at a 1:1 molar ratio and that increasing the concentration of DNA-PKcs may block the DNA ends from Artemis activity.

Artemis and Ku at DNA Ends
The Blunt End DNA Sequence Determines the Extent to Which Ku and DNA-PKcs Stimulate the 5Ј Nuclease Action of Artemis-We then asked whether the observed Ku stimulation is sequence-dependent at 5Ј blunt ends. We generated 5Ј-labeled, 34-bp, blunt-ended DNA with four dAs (Fig. 5A), four dTs (Fig. 5B), four dGs (Fig. 5C), or four dCs (Fig. 5D) by annealing 5Ј-labeled ssDNA with unlabeled 5Ј biotinylated complementary strands. The substrate was then incubated with proteins as indicated, along with ATP and streptavidin. Artemis 5Ј exonuclease activity was observed with substrates that are flanked by AT but not GC base pairs at the blunt DNA ends  [5][6][7][8]. We hypothesized that the weak base pairing in AT-rich ends compared with GC-rich ends allowed the DNA end to breathe and form transient ss/dsDNA boundaries to provide a substrate for Artemis. To examine the sequence-dependent variation further, we simulated breathing by creating a fixed 2-nt pseudo-Y structure with dA or dG unpaired ssDNA overhangs (Fig. 6). DNA with 5Ј dGs was processed by Artemis when forced into a pseudo-Y structure (Fig. 6, lanes 9 -12 versus lanes [13][14][15][16]. Interestingly, Y-structure DNA with 5Ј dAs was also stimulated 2-to 3-fold (Fig. 6, lanes 1-4 versus lanes 5-8).
These data indicate that DNA breathing may allow AT-rich blunt DNA to be cleaved by the Artemis 5Ј exonuclease activity in the absence of other proteins (DNA-PKcs and Ku).

Artemis 3Ј Endonuclease Resection Does Not Show Substantial Sequence Dependence but Is Stimulated by DNA-PKcs and
Ku-We then asked whether the sequence-dependent activity is also observed on the 3Ј end of blunt-ended DNA. 3Ј-labeled, 34-bp, blunt-ended DNA flanked by four dAs (Fig. 7A), four dTs (Fig. 7B), four dGs (Fig. 7C), or four dCs (Fig. 7D) was generated by fill-in synthesis with the Klenow fragment of DNA polymerase I (3Ј 3 5Ј exo Ϫ ) and [␣-32 P]dNTP. The substrate was then incubated with Ku, DNA-PKcs, and Artemis, as indicated, in the presence of ATP and streptavidin. The unlabeled complementary strand contained an unlabeled 5Ј phosphate and a 3Ј biotin to block Artemis action on the 3Ј end. 3Ј exonuclease activity was observed only on the 3Ј dA DNA and slightly on 3Ј dT, dG, and dC DNA (Fig. 7, A-D, lane 2). Interestingly, Ku did not block this 3Ј exonuclease activity (Fig. 7 Interestingly, these results are in contrast to Artemis activity observed on the 5Ј end. For one, the breathing requirement was not sequence-dependent on the 3Ј end. The mild activity stimulation by DNA-PKcs on the 3Ј end, in which only up to 8% of the substrate was cut after 60 min (Fig. 7, A-D, lane 8), was similar to the stimulation observed on GC-rich 5Ј ends (Figs. 5,  C and D, lane 12, and 7, A-D, lane 12). Secondly, the addition of Ku stimulated Artemis activity 6-to 8-fold on the 3Ј end independent of sequence (Fig. 7, A-D, lane 8 versus lane 12), whereas Artemis activity on 5Ј ends was sequence-dependent (a 3-to 4-fold increase for AT-rich and a more than 10-fold increase for GC-rich).
5Ј Resection Occurs Prior to 3Ј Resection on Blunt DNA Ends-We determined that Artemis performs asymmetric resection for the 5Ј versus 3Ј strands. Resection on the 5Ј strand at AT-rich blunt ends occurred in a largely Ku-independent manner. GC-rich 5Ј ends were mildly stimulated by DNA-PKcs and stimulated strongly by Ku. Conversely, Artemis exhibited sequence-independent activity on 3Ј ends that required Ku for robust activity. Because it appeared that 5Ј end resection was more efficient, we asked whether 5Ј end processing was required prior to 3Ј end processing. We annealed a 5Ј-labeled, 34-nt oligo to its 3Ј biotinylated complementary oligo with or without 10 phosphorothioate bonds on the 5Ј end. A phosphorothioate bond, formed by the sulfurization of a nonbridging oxygen in the phosphodiester bond, suppresses the action of nucleases. We performed a time course of Artemis activity in the presence of DNA-PKcs and Ku. We observed that Artemis activity on the 3Ј end of dsDNA substrate was reduced significantly when the complementary strand was unable to be cut first (Fig. 8A, lanes 1-5 versus lanes 6 -10). Therefore, it appears that the 5Ј strand resection occurs first, always leaving a 3Ј ssDNA overhang that, by its ssDNA nature, does not have any breathing requirement. This allows the 3Ј overhang resection to be sequence-independent.
Artemis Resection Progresses Internally along the DNA with Time-Because we established that Artemis first acts on the 5Ј end prior to acting on the resulting 3Ј overhang, we wondered whether this process would repeat until the DNA was processed completely. We initially showed that a 34-bp, bluntended DNA could be processed up to 11 nt after 2 h (Fig. 8A,  lane 5), which would leave ϳ23 bp remaining. We then incubated a 73-bp DNA with Ku, DNA-PKcs, and Artemis for up to 160 min (Fig. 8B, lanes 1-6). The resulting gel showed that Artemis endonucleolytically resected up to 4 nt on the 5Ј end (Fig. 8B, lane 6, 5Ј end resection). Processing on the 3Ј end proceeded until ϳ25-30 nt remained (Fig. 8B, lane 6, 3Ј end  resection). Because we know that ϳ26 bp are required for maximal Ku-dependent DNA-PKcs activity (17), it is likely that Artemis resection proceeds until the DNA is long enough for one Ku⅐DNA-PKcs complex to remain bound to the DNA.

Discussion
Under physiological conditions (Artemis plus DNA-PKcs in Mg 2ϩ solutions), we found that resection at blunt DNA ends depends on the DNA sequence. However, when Ku is also present, the disparity is eliminated, and blunt ends of all sequences are resected efficiently. This is the first biochemical documentation showing that Ku can affect the nucleolytic behavior of Artemis. Further study of it has revealed major unifying features of the Artemis nuclease.
Unifying Kinetic Features for Artemis Action at DNA Ends-Previous work on Artemis has documented its action at ss/ds-DNA boundaries, specifically at hairpins and overhangs and gaps and flaps (7,11). One study also described endonuclease action by Artemis alone in Mg 2ϩ solutions at blunt DNA ends (18). But this latter finding was curious because no other laboratory has found endonuclease activity of Artemis alone under Mg 2ϩ conditions. We noted that numerous freeze and thaw cycles of stored Artemis enzyme preparations eventually cause such behavior, suggesting a denaturation-induced change in the enzyme that partially eliminates the requirement for DNA-PKcs. In addition, Artemis alone can cut overhangs in Mn 2ϩ solutions, suggesting a permanent conformational change of the enzyme in the presence of this divalent cation (6,9,19). For this reason, evaluation of Artemis under conditions where it is dependent on DNA-PKcs was necessary to consider its function under physiologic conditions at blunt DNA ends and to integrate this into the spectrum of Artemis nuclease activities.
Our finding of resection by Artemis alone in Mg 2ϩ solutions at AT-rich DNA ends but not at GC-rich DNA ends suggested that DNA end breathing is required for Artemis action. In other studies, we have documented the increased breathing of ATrich DNA ends compared with GC-rich ends (20). When we "forced" breathing here by creating a fixed, 2-nt Y-structure in which the 2 nt arms of the Y cannot anneal, we found that, indeed, the DNA sequence of the arms no longer matters because the ends are in a permanent ssDNA configuration.
A requirement for end breathing for Artemis to act on a blunt end explains why such end resection by Artemis is slower than overhang resection. The breathing step is required, and the end is only in the available configuration (ss/dsDNA) for cutting for a fraction of the time.
A Unifying Physical Model for Artemis Nuclease Activity at DNA Ends-On the basis of the findings here and shown previously, it is possible to formulate a model that encompasses all of the nuclease activities of Artemis. These include hairpin opening (2 nt 3Ј of the tip), 5Ј overhang cutting (primarily at the ss/dsDNA boundary), 3Ј overhang cutting (primarily 4 nt 3Ј of the ss/dsDNA boundary), blunt end resection (slower than ss/dsDNA boundary substrates), and cutting at all DNA struc- tures containing ss/dsDNA boundaries (gaps, flaps, heterologous loops, and Y-structures). We note that poly(dT) ssDNA can be cut by Artemis⅐DNA-PKcs but that this cutting is much slower than for DNA containing a ss/dsDNA boundary (6). The apparently efficient cutting of ssDNA shown in Fig. 2B is due to this ssDNA being a mixed sequence (not poly(dT)) and being capable of forming secondary structures.
For optimal activity, we have proposed previously that the Artemis enzyme recognizes ss/dsDNA boundaries and cuts on the 3Ј side of an ϳ4-nt ssDNA region at this boundary. This explains the hairpin opening position, which is not at the tip but is located 2 nt 3Ј of the tip (the turn in the perfect hairpin provides the 4 nt of ssDNA because it is largely unpaired) (11). This also explains why 5Ј and 3Ј overhangs yield different overhang products (blunt for 5Ј overhangs but 4 nt 3Ј overhangs for long 3Ј overhangs) (7,11). DNA bending by Artemis, and, more efficiently, by Artemis⅐DNA-PKcs, at a 5Ј or 3Ј overhang may generate a structure similar to a hairpin (21). Fig. 9 shows how these same principles apply to Artemis action at blunt DNA ends. A blunt end can breathe into a Y-structure with short ssDNA arms. Artemis then resects the 5Ј end first. The resulting 3Ј overhang is then bent by Artemis or diffuses independently into a transient hairpin-like bend that is cut by Artemis. Because the resulting 3Ј overhang may not be sufficiently long for efficient processing by Artemis, Ku is likely required to provide a scaffold for Artemis⅐DNA-PKcs to bind near the overhang. All of the structures in Fig. 1 are likely to require a similar bending of the ss/dsDNA boundary into a hairpin-like state (Fig. 10).
The 5Ј Exonuclease of Artemis-Even the 5Ј exonuclease of Artemis can be incorporated into this model. We have noted previously that the 5Ј exonuclease of Artemis is independent of DNA-PKcs at 5Ј overhangs (7). However, for recessed 5Ј overhangs or blunt ends, we find that AT-richness around such a non-protruding 5Ј end is necessary for it to be cut by Artemis exonuclease activity (data not shown), implying that the 5Ј exonuclease activity of Artemis also relies on ss/dsDNA transition structures in the substrate. Only the protruding 5Ј DNA ends are targets of the 5Ј exonuclease activity of Artemis without any effect on DNA sequence. This may arise because the protruding 5Ј end can insert directly into the active site of Artemis without any more substantial substrate positioning by Artemis that is required for it to process all other DNA ends.
The Role of DNA-PKcs and Ku in Artemis Interaction with Its Substrate-We believe that the role of DNA-PKcs is to change the conformation of Artemis to a form that can more easily configure DNA ends into an optimal hairpin-like substrate (Fig.  10). As pointed out previously, Artemis under Mn 2ϩ conditions does not require DNA-PKcs for any of the reactions shown in Fig. 1 (6, 9, 19).
DNA-PKcs and Ku might simply increase the length of time that Artemis remains bound so that, at the time of DNA breath- FIGURE 9. DNA end breathing model for Artemis action at blunt DNA ends. The diagrams summarize the action of Artemis at blunt DNA ends (in 10 mM MgCl 2 and 75 mM KCl). Consistent with the optimal substrate requirements of Artemis (Fig. 1), blunt DNA ends must first breathe to pseudo-Y structures before efficient nuclease action can begin. Artemis first acts on the 5Ј strand of the DNA end and then on the 3Ј strand (which, at that point, is a 3Ј overhang). The result of 5Ј and 3Ј cuts is a blunt or near-blunt end that is then subject to iterative rounds of the same process, consistent with our results. The red arrows indicate the preferred Artemis cut sites. The open thick arrows represent the proposed Artemis binding site and orientation on the DNA substrates.
ing into a transient ss/dsDNA boundary structure, Artemis can hydrolyze the phosphodiester bond. Artemis in vivo exists predominantly in a very salt-stable complex with DNA-PKcs, whereas Ku only forms a complex with Artemis⅐DNA-PKcs when a DNA end is present (1,11). DNA-PKcs autophosphorylation is required for DNA-PKcs to activate Artemis endonuclease activity (1). Ku improves the affinity of DNA-PKcs alone 100-fold, from 3 ϫ 10 Ϫ9 to 3 ϫ 10 Ϫ11 M (17), and we assume that this applies to the Artemis⅐DNA-PKcs complex as well. The Artemis⅐DNA-PKcs⅐Ku complex would be able to remain at a DNA end longer, giving Artemis more opportunity to cut.
Biological Implications of the Unified Model of Artemis Nuclease Action for NHEJ-The key in vivo roles of Artemis are diverse and have not been easy to reconcile with a single definition of the substrate configuration that is optimal for Artemis nuclease action. Among the key DNA end configurations at which Artemis is known to act are perfect DNA hairpins during the coding joint formation step of V(D)J recombination (11,23,24), repair of 10 -40% of ionizing radiation-induced DSBs (25), and repair of DSBs produced by interruption of topoisomerase II action (26). All of the structures for in vivo and in vitro action of Artemis have ss/ds DNA boundaries. Importantly, the boundary for hairpins is due to the imperfect base pairing at the hairpin tip (11).
However, the action of Artemis at blunt DNA ends (and in a DNA-PKcs-independent manner) complicated this simple model (18). Here we confirmed the action at blunt ends, but we show that, like the other endonucleolytic actions of Artemis, this too is dependent on DNA-PKcs. Moreover, the DNA endbreathing requirement unifies this substrate configuration with the other known in vivo and in vitro substrate configurations.
Biological Implications of the Unified Model of Artemis Nuclease Action for V(D)J Recombination-This study is likely to explain why GC-rich coding ends suffer less end resection FIGURE 10. Functional relationship between Artemis and ss/dsDNA ends. A, Artemis can only cut at DNA hairpins in Mn 2ϩ solutions or in Mg 2ϩ solutions that also contain DNA-PKcs and ATP. We propose that this allows the C-terminal portion (C) of Artemis to no longer suppress the endonuclease action by the active site of Artemis (N). The red arrow indicates the preferred Artemis cut site. B, we propose that 5Ј and 3Ј overhangs must adopt a folded structure similar to a hairpin to be cut by Artemis. Consistent with our previous proposals that Artemis recognizes ss/dsDNA boundaries and cuts on the 3Ј side of a 4-nt length within the ssDNA portion, the 5Ј overhangs are predominantly cut to initially generate a blunt DNA end, whereas the 3Ј overhangs are initially cut predominantly to generate a 4-nt 3Ј overhang. Of course, subsequent cuts into the DNA ends can occur in the manner shown in Figs. 1 and 9. The red arrows indicate the preferred Artemis cut sites. than AT-rich coding ends in human pre-B cell lines (27). GCrich coding ends will breathe less than AT-rich coding ends. This curious but well documented finding on the sequence dependence of coding end resection has gone without explanation for nearly two decades, but the findings in this study provide the first explanation for how DNA end resection at coding ends in V(D)J recombination is affected by DNA end sequence.
A second point of in vivo relevance concerns the difference in resection of coding ends versus signal ends in V(D)J recombination. It has been unclear why signal ends suffer end resection much less often than coding ends. Signal ends are blunt, and they suffer end resection in wild-type human lymphoid cells but only at 5% of all signal joints (28). In contrast, coding ends are opened (by Artemis⅐DNA-PKcs) to give 3Ј overhangs (usually 4 nt in length), and these coding ends suffer end resection in over 90% of coding joints in human pre-B cells (27). This disparity has never been explained, but one possible explanation is raised by our finding of a slower and weaker resection of Artemis at blunt DNA ends because of the requirement for DNA end breathing described here. Such an explanation would be consistent with the fact that end resection at signal ends decreases when Artemis is absent (29) and increases when the partner of Artemis, namely DNA-PKcs, is mutated-a point that has remained without any reasonable explanation until now (30).
A third point of possible in vivo relevance concerns our observation in human cells of inverted repeats at partially resected coding ends in the final coding joints formed in human pre-B cells (27). The Artemis resection in 2-to 6-nt increments at blunt ends, as in Fig. 2, may liberate these short oligonucleotides, allowing them to be religated via the single-strand ligation activity of the ligase IV complex (22). This would generate the short inverted repeats observed at partially resected coding ends (27), which have previously been difficult to explain mechanistically until now. Therefore, the findings here not only provide a unification of the enzymatic activities of an important structure-specific nuclease but also integrate several in vivo observations with this unified model.