Loading of the Nonhomologous End Joining Factor, Ku, on Protein-occluded DNA Ends*

The nonhomologous end joining pathway for DNA double strand break repair requires Ku to bind DNA ends and subsequently recruit other nonhomologous end joining factors, including the DNA-dependent protein kinase catalytic subunit and the XRCC4-Ligase IV complex, to the break site. Ku loads at a break by threading the DNA ends through a circular channel in its structure. This binding mechanism explains both the high specificity of Ku for ends and its ability to translocate along DNA once loaded. However, DNA in cells is typically coated with other proteins (e.g. histones), which might be expected to block the ability of Ku to load in this manner. Here we address how the nature of a protein obstruction dictates how Ku interacts with a DNA end. Ku is unable to access the ends within an important intermediate in V(D)J recombination (a complex of RAG proteins bound to cleaved recombination targeting signals), but Ku readily displaces the linker histone, H1, from DNA. Ku also retains physiological affinity for nucleosome-associated ends. Loading onto nucleosome-associated ends still occurs by threading the end through its channel, but rather than displacing the nucleosome, Ku peels as much as 50 bp of DNA away from the histone octamer surface. We suggest a model where Ku utilizes an unusual characteristic of its three-dimensional structure to recognize certain protein-occluded ends without the extensive remodeling of chromatin structure required by other DNA repair pathways.

DNA double strand breaks (DSBs) 2 pose a threat to genomic integrity. Nonhomologous end joining (NHEJ) is a major pathway for repair of exogenously introduced DSBs in mammals and is the only efficient way to repair DSB intermediates in V(D)J recombination (1). NHEJ requires the DNA binding heterodimer Ku to first recognize broken DNA ends (2)(3)(4)(5) and subsequently recruit the additional NHEJ factors necessary to complete repair (e.g. DNA-dependent protein kinase catalytic subunit (DNA-PKcs) (6), the XRCC4-LigaseIV complex (7,8), and polymerases (9,10)). Consequently, deficiency in Ku cripples NHEJ and leads to severe immunodeficiency (11,12) and cellular sensitivity to agents that cause DSBs (e.g. ionizing radiation) (13,14). The ability of Ku to recognize DNA ends requires the end to be inserted through a channel in its structure (5,15,16). This manner of binding makes Ku highly specific for DNA ends but requires that the ends must be accessible through 360°. In cells, DNA is generally coated with proteins (e.g. histones and other chromatin-associated proteins) that might be expected to block the ability of Ku to load on DNA ends and thus impair NHEJ.
Here we have shown that Ku displays a variety of responses to protein obstructions at DNA ends. Ku is unable to bind a class of RAG (recombination activating gene) protein-bound DNA ends generated during V(D)J recombination. In contrast, Ku can displace certain proteins (e.g. histone H1) from DNA and retains the ability to load on nucleosome-associated ends by peeling up to 50 bp of DNA away from the nucleosome surface. We therefore suggest Ku may be specifically suited to loading on protein-obstructed DNA ends, ultimately facilitating the recruitment of the NHEJ machinery without requiring extensive remodeling of chromatin structure at broken ends.

EXPERIMENTAL PROCEDURES
Purified Proteins-Bulk histones were purified from 5 liters of HeLa cells as described previously (17). Briefly, a nuclear pellet was isolated and homogenized, and genomic DNA was sheared by sonication. The clarified supernatant was then applied to a hydroxyapatite column equilibrated in 0.6 M NaCl. The column was washed extensively with the equilibration buffer to deplete linker histones and non-histone proteins, and fractions primarily containing core histones were eluted with 2 M NaCl. This eluent was then applied to a S200 gel filtration column equilibrated in 2 M NaCl to further deplete linker histones and ensure the proper stoichiometry of the core histone octamer. Histone H1 was a gift from Dr. Yi Zhang. Recombinant core RAG1 and RAG2 maltose-binding protein fusions and recombinant Ku were obtained as previously described in Refs. 18 and 7, respectively.
Nucleosomes-As described previously (21), nucleosomes were reconstituted on positioning sequences by salt dialysis using a DNA to histone ratio determined empirically for each DNA preparation such that reconstitutions contained less than 5% free DNA. The quality of reconstitutions was monitored by EMSA. Nucleosomes used for footprinting were also purified by preparative electrophoresis using a Mini-prep cell (Bio-Rad) (22). Native mononucleosomes were obtained by resuspending HeLa nuclei in 10 ml of micrococcal nuclease digestion buffer (10 mM Tris, pH 7.5, 5 mM MgCl 2 , 5 mM CaCl 2 , 0.1 mM phenylmethylsulfonyl fluoride, and 0.5 mM dithiothreitol) and digestion with 0.02 units/l of micrococcal nuclease (USB) at 37°C for 8 min (17). After addition of NaCl to 0.6 M, the suspension was dounced to extract soluble chromatin and subsequently pelleted. The resulting supernatant was separated on a 5-30% sucrose gradient. As mononucleosomes derived from this digestion are rich in linker histones (17), we excluded mononucleosome-containing fractions and further digested pooled oligonucleosome-containing fractions with 1.5 units/l micrococcal nuclease at 0°C for 40 min to generate core mononucleosomes. Mononucleosomes were further purified on a second sucrose gradient.
In Fig. 4 Analysis of the Protein Composition of Ku-Nucleosome Complexes-Large-scale 50-l EMSA reactions were generated by incubating 80 nM nucleosome or free DNA with either 60, 120, or 240 nM Ku in our standard EMSA buffer at 25°C for 10 min. Reactions were then incubated for 5 min at 37°C before separating the formed complexes as described above. Ku-nucleosome and Ku-DNA complexes were visualized by ethidium bromide staining and subsequently excised. Complexes were then electro-eluted from the gel and their protein components concentrated by trichloroacetic acid precipitation. The relative amounts of Ku and histone H3 in each of the excised complexes were determined by semi-quantitative Western analysis probing with a polyclonal rabbit antibody raised against native, recombinant human Ku and a polyclonal antibody against histone H3 (Ab1791; Abcam) using fluorescent detection and a Typhoon imager (GE Healthcare). Western blots were quantified using ImageQuantTL (GE Healthcare), and the ratio of Ku70 (Ku80 is overexposed) to H3 was determined for nucleosome-containing complexes (Fig. 5B, species IV, V, and VI).
Hydroxyl Radical Footprinting-Footprinting experiments utilize an asymmetrically positioned nucleosome substrate containing a single 40-bp linker DNA (24). The DNA end of this linker was labeled with a bodipy 630 fluorophore (Integrated DNA Technologies) and blocked with a biotin-streptavidin complex (Fig. 6A). Footprinting reactions were conducted in 15 l of 30 mM Tris, pH 8.0, and 0.1 mM EDTA by incubating 80 nM nucleosome with 120 or 240 nM Ku at 25°C for 10 min. Reactions were then incubated at 37°C for 5 min before being placed on ice. Nucleosomes were twice treated at 4°C with 3.5 l of 3% hydrogen peroxide, 3.5 l of a 2-mM ammonium iron(II) sulfate and 4 mM EDTA mixture, and 3.5 l of 20 mM sodium ascorbate at 5-min intervals before reactions were stopped by addition of 5 l of 400 mM thiourea (24). Reactions were acidified by addition of 5 l of 3 M sodium acetate, and nicked DNA was purified through a minElute reaction clean-up column (Qiagen). Eluted DNA was electrophoresed on an 8% urea-PAGE sequencing gel at 1800 V and 40 watts for either 1.5 (Fig. 6B) or 4 h (Fig. 6C).
Protein Modeling-The SPOCK modeling program (25) was used to dock the crystal structures of the core nucleosome (Protein Data Bank accession 1AOI) and Ku bound to DNA (accession 1JEY) by multiple alignments of the phosphodiester backbone of residues 2-12 in chain C of 1JEY with the backbone of 10 consecutive residues in chain I of 1AOI. Iterative alignments were made, advancing target phosphate positions in the nucleosome in single nucleotide steps, until steric clashes were minimized to generate the docked structure model shown in Fig. 8, where residues 2-12 in chain C of 1JEY are aligned with residues 10 -20 in chain I of 1AOI.

RESULTS
Ku Displaces Histone H1 from DNA Ends-Binding of protein to DNA fragments has been shown in vitro to inhibit the ability of XRCC4-Ligase IV to join DNA ends (26,27). Interestingly, Ku and DNA-PKcs are required to relieve the specific inhibition caused by the presence of the linker histone H1 (27), suggesting Ku may play roles in recognizing H1-occluded DNA ends and making these ends accessible to DNA-PKcs and XRCC4-ligase IV. To address this possibility, we incubated a radiolabeled 31-bp DNA duplex with a large excess (500-fold) of the linker histone. This generates H1-DNA complexes with heterogeneous mobility (Fig. 1, lane 11), including aggregates that remain in the well (species III). Addition of Ku to these reactions results in progressive redistribution of the H1-bound DNA into two species (species V and VI). Species V is heterogenous, and its mobility can be reduced by antibodies to either H1 or Ku, indicative of the presence of at least one molecule of both. Species VI is more abundant at higher concentrations of Ku and is consistent with DNA where Ku has mostly evicted H1; the mobility of a large proportion of this species resists addition of the ␣-H1 antibody, and its mobility is equivalent to DNA saturated by Ku only (compare lanes 8 -10 to 15-17). Eviction of H1 occurs even when the Ku concentration is onefiftieth the concentration of H1. Ku is thus remarkably effective at clearing DNA of histone H1.
Ku Threads on Nucleosome-associated DNA Ends-We next addressed whether the ability of Ku to load on H1-occluded ends could be generalized to other physiologically relevant protein-DNA complexes. The nucleosome is the protein occlusion that Ku is most likely to encounter near a DSB produced by exogenous DNA-damaging agents (e.g. ionizing radiation). We therefore addressed whether Ku could load on DNA ends in the context of "linker-less" mononucleosomes purified from bulk cellular chromatin (Fig. 2, species I). To obtain these core nucleosomes, oligonucleosome fragments lacking linker histones were first purified by sucrose gradient and subsequently subjected to exhaustive micrococcal nuclease digestion, generating a mononucleosome containing 147 bp of DNA (17) 3 . We then assessed whether Ku could bind these mononucleosomes by EMSA. Consistent with previous reports (5,28), incubation of Ku with either native or reconstituted mononucleosomes shifts the mobility of the nucleosome in a manner dependent on Ku concentration (Fig. 2, species II and III), arguing Ku binds to nucleosome-associated ends.
Given the way Ku loads on naked DNA ends (threading of ends through a central channel), Ku must either have altered how it binds DNA ends or loading of Ku involves some form of nucleosome remodeling. We employed a series of homogeneously positioned mononucleosome substrates (e.g. Figs. 2 and 3) and a previously described end-blocking strategy (15,29) to address this issue in greater detail. Ku can bind if one but not both ends of a naked DNA fragment are blocked by terminal biotin-streptavidin complexes (Fig. 3A). We therefore reconstituted singly positioned mononucleosomes containing biotin 3 S. A. Roberts and D. A. Ramsden, unpublished data.   1-4), reconstituted core mononucleosomes ( lanes 5-8), or 147-bp naked DNA (lanes marked D) were incubated with 50 nM Ku (lanes 2, 6, and D), 100 nM Ku (lanes 3 and 7), or 200 nM Ku (lanes 4 and 8) for 10 min at 25°C. The resulting complexes were separated by gel electrophoresis and visualized by SYBR Green staining. Roman numerals indicate various Ku-bound species as described under "Results." appended to one or both DNA ends. Importantly, as with naked DNA, Ku requires at least one unblocked end before it can shift the equivalent nucleosome substrate (Fig. 3B). Antibodies to Ku further reduced the mobility of candidate Ku-bound nucleosome species, confirming Ku is stably retained in these species (Fig. 3C). Ku thus loads on nucleosome-associated DNA ends the same way it loads on naked DNA, by threading DNA ends through its channel.
The singly blocked nucleosome substrate was then used to determine the extent to which nucleosome association reduces the affinity of Ku for DNA ends. In accordance with previous estimates (2,23), Ku displays an apparent K d of 0.34 Ϯ 0.02 nM for the first molecule binding a naked DNA end (Fig. 4A). However, when confronted with ends on the surface of a nucleosome, Ku binds with an apparent K d of 6.0 Ϯ 0.4 nM for the first Ku bound (Fig. 4B). Ku is thus somewhat less able to load on nucleosome-associated ends but nevertheless retains an affinity for nucleosomeassociated DNA ends that is comparable with the affinity of other DNA-binding proteins for naked DNA (e.g. ϳ3 nM for HMG1 and ϳ1 M for Rad51) (30 -32).
Once bound, Ku can translocate internally on linear DNA, allowing successive molecules of Ku to load on the same DNA end (33). In EMSA experiments, this appears as "ladder" of distinct species of reduced mobility, each with an additional molecule of Ku (Fig. 3A and Refs. 5 and 33). Multiple species are similarly observed upon addition of Ku to nucleosome-associated ends (Fig. 3B, lanes 10 and 11,  Fig. 5A, lane 10). We purified these species and determined their protein complement by semi-quantitative Western analysis (Fig. 5B). Ku and a representative histone (H3) are present in each species as appropriate. Additionally, the ratio of Ku to histone in species VI is 2.1 times that of species V, consistent with the presence of one and two molecules of Ku per nucleosome-associated end in the two respective species. Significantly, while each molecule of Ku loads on naked DNA ends essentially independently of prior molecules loaded (Fig. 4A and Ref. 34), the nucleosome strongly resists loading of a second molecule of Ku (K d of second molecule Ͼ25 nM; Fig. 4B).
Ku Peels DNA Ends from the Nucleosome Surface-The ability of Ku to bind nucleosomeassociated DNA ends and even translocate internally to some extent indicates Ku must alter nucleosome structure in some manner. We considered three possibilities: Ku could reposition the histone octamer away from the DNA end ("pushing"), Ku could leave the nucleosome in its initial position and "peel" the DNA end away from the octamer surface, or Ku could evict a subset of core histones near the DNA end. To address these possibilities, we generated a 187-bp substrate with the nucleosome positioned at one DNA end, leaving the other DNA end spaced 40 bps from the histone octamer (e.g. as in Ref. 24). This end distal to the nucleosome was also blocked with a biotin-streptavidin complex, thereby forcing Ku to load from the nucleosome-associated end (Fig. 6A). Hydroxyl radical footprinting of this substrate shows a 10-bp phasing indicative of a Addition of Ku to our footprinting reactions has a general quenching effect, resulting in an overall reduction of hydroxyl radical cleavage (Fig. 6B). To account for this, we included a control substrate where both DNA ends are blocked with biotin-streptavidin complexes (Fig. 6, B and C, lanes 4). Our data show that even at high concentrations of Ku, there are no major changes in hydroxyl radical sensitivity of the linker DNA (0 -40 bps away from the label) or in the pattern of the phasing over most of the nucleosome (40 to ϳ140 bps from the label) (Fig. 6,   B and C, compare lanes 3 and 4). We therefore conclude that in the majority of molecules, loading of Ku does not alter the translational location of the nucleosome on this DNA fragment.
However, increasing concentrations of Ku result in a correlating loss of phases near the accessible end, consistent with Ku peeling the DNA end away from the surface of the nucleosome. As expected, the phasing pattern of the doubly blocked nucleosome in this region is resistant to addition of Ku (Fig. 6C, compare lanes 3 and 4). When Ku is present at 3-fold excess over the nucleosome, loss of phasing extends over the first 40 bp, with some protection observed as much as 50 bp (one-third of the length of the nucleosome) away from the end. Parallel experiments indicate the majority of nucleosome substrate possesses more than one molecule of Ku under these conditions (Fig. 5,  lane 10), arguing a molecule of Ku is able to load and translocate internally 40 -50 bp, allowing a second molecule to load. Thus, Ku displays the unique ability to thread on nucleosome-associated DNA ends, gaining access to the DNA end by a peeling mechanism.
It is possible that the nucleosome accommodates loading of Ku through displacement of some subset of core histones. To address this possibility, we first loaded Ku onto nucleosomeassociated ends and then added a large excess of linear compet-   1-5) or nucleosome substrates (lanes 6 -10) were incubated with 60 nM (lanes 3 and 8), 120 nM (lanes 4 and 9), or 240 nM (lanes 5 and 10) Ku at 25°C for 10 min. Reactions were then incubated at 37°C for 5 min before separating the formed complexes by polyacrylamide gel electrophoresis. Complexes were visualized by ethidium bromide straining, and the species of interest were excised as indicated with boxes. B, in lanes 4 -9, the protein complements of species gel-purified in panel A (labeled I-VI) were analyzed by Western blotting using ␣-Ku and ␣-histone H3 antibodies (as indicated to the right of the panel) after electro-elution, precipitation, and SDS-PAGE. In lanes 1-3 (Standards), 80, 40, and 20 pmol of Ku or nucleosome (e.g. H3 dimer) were loaded directly to verify the Western is semi-quantitative.
itor DNA to both remove Ku and ensure potentially displaced histones cannot be reincorporated into the nucleosome. Removal of Ku from the nucleosome restores the nucleosome to its original mobility (Fig. 6D), arguing against eviction of histones.
Ku Fails to Bind RAG-sequestered Signal Ends-Ku must also recognize and promote the joining of protein-occluded double strand break intermediates during V(D)J recombination, a genome rearrangement required for assembly of the mature antigen-specific receptor genes of the mammalian immune system. V(D)J recombination requires one each of the two types of targeting signals (12-RS and 23-RS) and is initiated when the RAG1 and RAG2 proteins bind this pair of signals and cleave the flanking chromosomal DNA. Importantly, RAG proteins remain in a complex with the paired signals after cleavage (paired SEC), and this complex is sufficient to block signal end joining in both extract- (35)(36)(37) and purified protein-based (26) in vitro assays. Nevertheless, signal end intermediates in V(D)J recombination are efficiently joined together by NHEJ in cells. We therefore utilized an EMSA to test whether Ku could directly load on RAG-bound signal ends.
We generated an SEC in vitro (see Fig. 7A) by incubating purified RAG proteins and HMG1 (high mobility group protein 1) with recombination signal-containing oligonucleotide duplexes. A stable SEC (species IV) requires RAG1, RAG2, HMG1, and both 12-RS and 23-RS to be present; this species is inefficiently formed when one of the signal sequences is omitted (Fig. 7B, compare species III and IV, lanes 3 and 4) or if one of the signals is substituted with an oligonucleotide duplex composed of irrelevant sequence. 3 We used a pair of RS-containing oligonucleotide duplexes to generate the SEC instead of a continuous DNA fragment terminating in signals ends (as is typically generated in vivo) because this keeps the complex sufficiently small to be resolved by EMSA. However, these duplexes possess an end distal to the site of cleavage that is anomalously accessible. We therefore selectively blocked access to these ends as described above by appending biotin to the appropriate ends of the oligonucleotides when synthesized and including streptavidin in these reactions (Fig. 7, A and B, species V).
Addition of Ku to SEC mixtures generates a new species (species VI) with increased mobility relative to the streptavidin-blocked SEC (species V). However, this probably reflects loading of Ku onto a minor population of incompletely formed SEC (evident as species with heterogeneous mobility in lanes 4 through 7) because Ku changes neither the mobility nor the intensity of accurately formed SEC (Fig.  7B, compare species V in lane 5 to species V in lane 8). Furthermore, the mobility of species V was significantly reduced by an antibody to the maltose-binding protein tag on the RAG proteins, but not an antibody directed against Ku, excluding the possibility that Ku was present in the SEC, but our EMSA was unable to resolve this species. Accurately formed SEC similarly resisted loading of Ku when incubated for longer times (1 h), at higher temperature (37°C), with increased salt (150 mM KCl), using either Mg 2ϩ or Ca 2ϩ , with higher concentrations of Ku (250 nM), or when accompanied by XRCC4-ligase IV, DNA-PKcs, and ATP. 3 Therefore, we conclude that Ku, either alone or together with core NHEJ factors, is probably insufficient to recognize and promote joining of DNA ends in the context of the SEC.

DISCUSSION
In vitro, at least 50 bps of naked DNA are required to assemble a functional ligation complex at a DNA end (38), but DNA double strand breaks in cells are typically occluded by nucleosomes, linker histones, and other proteins involved in chromatin structure. Therefore, the ability of Ku to load onto such ends is likely a major determinant of the efficiency of cellular NHEJ given the critical place of Ku in break recognition and subsequent nucleation of the NHEJ complex.
The unusual structure of Ku, which requires DNA ends to be threaded through a circular protein channel, might argue it is poorly suited for loading on protein-occluded ends. Consistent with this argument, we show the RAG proteins bound to signal end intermediates in V(D)J recombination strongly resist the ability of Ku to recognize and load on these DNA ends. Nevertheless, NHEJ must resolve these intermediates in cells to maintain intact receptor loci in certain contexts, and it may be critical in reducing RAG-mediated transposition activity (39). The prior disassembly of the SEC by factors extrinsic to the core NHEJ machinery and the SEC (e.g. by proteolysis) (40 -42) may thus be the limiting step in resolution of this class of proteinoccluded ends.
In contrast, Ku appears readily able to displace the more weakly bound linker histone H1. Because H1 occlusion is sufficient to block activity of the NHEJ ligase (ligase IV) in vitro (27), this argues that the ability of Ku to recognize broken ends occluded by H1 may be the important first step in allowing NHEJ to act on breaks generated in linker regions of chromatin. However, we note linker histones may be bound more tightly in the context of higher order chromatin; thus, this question should be re-addressed when it becomes possible to accurately recapitulate higher order chromatin structures in vitro.
Surprisingly, Ku can also recognize DNA ends on the surface of nucleosomes and does so by the same mechanism it uses for naked DNA, threading DNA ends through its central channel. Loading of Ku can be accompanied by peeling as much as onethird of the nucleosomal DNA away from the histone octamer. However, unlike traditional chromatin remodeling, the nucleosome structure (i.e. octamer composition or translational position) can remain otherwise largely unperturbed and energy from ATP hydrolysis is not required. Instead, Ku presumably takes advantage of transient dissociations of DNA ("breathing") as it enters and leaves the nucleosome (43). The crystal structures of Ku bound to a DNA end (16) and the nucleosome core particle (44) also suggest Ku may be particularly well suited for loading on nucleosome-associated DNA ends as the two structures can be docked with only slight alteration of the DNA path (Fig. 8). We propose that Ku's narrow ␤-strand bridge portion of its DNA binding channel can act as a wedge between the DNA end and the histone octamer surface, allowing Ku to pass DNA ends through its channel while minimally disrupting histone-DNA interactions. Once bound, Ku may then translocate internally, most likely by being pushed by DNA-PKcs (29), XRCC4-Ligase IV (38), or a second molecule of Ku.
The ability of Ku to load on chromatinized ends and translocate inward may provide cells with several benefits. As suggested above, Ku alone may be sufficient for recognition of chromatinized ends and activation of NHEJ, allowing this pathway to proceed without always requiring chromatin remodeling. Alternatively, loading of Ku may be a critical first step in directing other factors to perform a limited remodeling of chromatin at ends. In yeast, Ku interacts with the SWI/SNF family remodeling complex, RSC (45,46), and the ability of Ku to peel DNA ends from the nucleosome surface could help orient the direction of the activity of this complex such that nucleosomes are pushed away from break sites.
NHEJ thus may repair double strand breaks with minimal or no remodeling of flanking chromatin. In contrast, the other major double strand break repair pathway, homologous recom- FIGURE 7. The DNA binding activity of Ku at RAG bound signal ends. A, a signal end complex is defined here as RAG1, RAG2, and HMG1 bound to twooligonucleotide DNA duplexes, one containing the 12-RS and the other containing the 23-RS. A streptavidin-biotin complex was used to block the ends of the DNA duplexes distal to the site of cleavage. B, approximately 10 ng/l of purified RAG1 and RAG2 and 425 nM HMG1 were incubated with 0.4 nM radiolabeled 23-RS-containing and 10 nM 12-RS-containing DNA fragments at 37°C for 10 min. 5 M streptavidin was added at 25°C for 5 min prior to addition of 25 nM Ku. Antibody supershifts required an additional 10-min room temperature incubation step with either 0. bination, is associated with the removal of nucleosomes within several kilobases of the break site (47). This difference may help rationalize why NHEJ is the preferred repair pathway in differentiated cells (48,49), where significant disruption of chromatin state could lead to inappropriate gene activation.