Accurate in Vitro End Joining of a DNA Double Strand Break with Partially Cohesive 3′-Overhangs and 3′-Phosphoglycolate Termini

To examine determinants of fidelity in DNA end joining, a substrate containing a model of a staggered free radical-mediated double-strand break, with cohesive phosphoglycolate-terminated 3′-overhangs and a one-base gap in each strand, was constructed. In extracts of Xenopus eggs, human lymphoblastoid cells, hamster CHO-K1 cells, and a Chinese hamster ovary (CHO) derivative lacking the catalytic subunit of DNA-dependent protein kinase (DNA-PKcs), the predominant end joining product was that corresponding to accurate restoration of the original sequence. In extracts of the Ku-deficient CHO derivative xrs6, a shorter product, consistent with 3′ → 5′ resection before ligation, was formed. Similar results were seen for a substrate with 5′-overhangs and recessed 3′-phosphoglycolate ends. Supplementation of the xrs6 extracts with purified Ku restored accurate end joining. In Xenopus and human extracts, but not in hamster extracts, gap filling and ligation were blocked by wortmannin, consistent with a requirement for DNA-PKcs activity. The results suggest a Ku-dependent pathway, regulated by DNA-PKcs, that can accurately restore the original DNA sequence at sites of free radical-mediated double-strand breaks, by protecting DNA termini from degradation and maintaining the alignment of short partial complementarities during gap filling and ligation.

Cells deficient in any of the three components of DNA-dependent protein kinase DNA-PK 1 (i.e. the catalytic subunit DNA-PKcs or either subunit of the DNA end binding heterodimer Ku) are partially deficient in the rejoining of double strand breaks (DSBs) (1)(2)(3)(4). The precise roles of the individual subunits in the repair process are not known, but possible functions have been suggested based on their known biochemical properties. Ku has been proposed to align DNA ends and thus promote ligation (5)(6)(7), to protect the ends from degradation (8), and/or to unwind duplex DNA ends and thus expose microhomologies that could be used for splicing the ends together (1,9). DNA-PKcs has been proposed to regulate accessibility of DNA ends to processing (10), perhaps by promoting its own dissociation following autophosphorylation (11) and/or by allowing translocation of Ku from the ends into the interior of the DNA (10,12,13).
Ionizing radiation is a major environmental source of DSBs (14 -16). Radiation-induced DSBs are formed by fragmentation of deoxyribose, typically leaving in each strand a one-base gap with 5Ј-phosphate and either 3Ј-phosphate or 3Ј-phosphoglycolate (PG) termini (14,15,17). Such DSBs present to repair systems a more complex substrate than simple restriction enzyme cuts, potentially increasing the possibility of errors during rejoining. In this report, we describe how synthetic substrates containing mimics of radiation-induced breaks, with defined geometry and chemical structure (see Figs. 1 and 2), are processed in several in vitro end joining systems. The results suggest a Ku-dependent repair process that can accurately restore the original DNA sequence at sites of these complex DSBs, through the use of very short residual complementarities at the ends of the break.
A plasmid substrate containing a site-specific DNA double strand break with 2-base cohesive 5Ј-overhangs and recessed 3Ј-PG termini (Fig. 2C) was prepared by ligation of 3Ј-PG oligomers into larger 5Јoverhangs formed by controlled digestion of plasmid pSV56 with the 3Ј35Ј exonuclease activity of T4 polynucleotide kinase, as described (10,23). A similar construct, but bearing a PG-terminated -ACG 3Јoverhang was similarly constructed by ligation of slightly longer oligomers (pGCCGGACGCGACG* and 5Ј-32 P-labeled pCGAGGAACGC-GACG*, where the asterisk indicates a 3Ј-PG) into the same prepared vector (Fig. 1). The structure of the 3Ј-PG oligomers was verified by Fourier transform electrospray ionization mass spectrometry; all species detected could be ascribed to isotopes or salts of the oligomer or to the internal standard, with no detectable contaminants (24).
As described previously (23), the 5Ј-overhang construct was purged of any residual molecules having 3Ј-hydroxyl termini by treatment with T7 DNA polymerase in the absence of dNTPs. However, in the case of the substrate with a 3Ј-overhang, repeated attempts to digest residual 3Ј-hydroxyl species using T7 polymerase unexpectedly resulted in apparent formation of a substrate with a 3Ј-hydroxyl blunt end. Therefore, this digestion step was eliminated from the construction, and in order to show that particular repair products were derived from constructs that had incorporated 3Ј-overhangs into both ends, parallel experiments were performed with a construct into which a 3Ј-PG oligonucleotide had been ligated at one end only (Fig. 2B). Ku protein was purified as described previously (25).
End Joining Reactions-Reactions with whole-cell extracts of CHO-K1 cells and various CHO derivatives contained 8 l of extract (briefly dialyzed against 50 mM morpholinoethanesulfonate-NaOH, pH 7.5, 40 mM KCl, 10 mM MgCl 2 , 5 mM ␤-mercaptoethanol and with a protein concentration of 12.5 mg/ml) plus 10 ng of substrate in a total volume of 10 l. Samples were incubated at 25°C, usually for 6 h (8). For experiments with Xenopus egg extract, 20-l reactions contained 16 l of extract (in extraction buffer) and 10 ng of DNA substrate; reactions were incubated at 13°C, usually for 6 h (26). Reactions with whole-cell extracts of human GM00558B cells contained 50 mM triethanolamine-HCl, pH 7.5, 0.5 mM magnesium acetate, 60 mM potassium acetate, 2 mM ATP, 1 mM DTT, 100 g/ml bovine serum albumin plus 10 ng substrate and whole-cell extract at a final protein concentration of 3 mg/ml and were incubated at 37°C for 6 h (21). Wortmannin in Me 2 SO was added to some samples to a final concentration of 10 M. Some samples also contained various dNTPs and/or ddNTPs (50 M each), as indicated.
Analysis of Repair Intermediates and Products-Following the end joining reaction, all samples were deproteinized with proteinase K and phenol, and nucleic acids were precipitated, treated with BstXI and XhoI, denatured, and subjected to electrophoresis on 20% polyacrylamide gels as described previously (27). Wet gels were frozen and exposed on PhosphorImager screens for 1-2 days at Ϫ20°C. Phosphor images were analyzed with ImageQuant 3.3 software (Molecular Dynamics). Poorly resolved peaks were quantitated using the Gaussian deconvolution algorithm of PeakFit 4.06 (AISN Software).
For analysis of topological forms, deproteinized DNA samples were electrophoresed on 0.6% agarose gels at 6 V/cm, affording clear separation of the nicked circular and linear dimer end joining products. Gels were dried under vacuum for phosphorimaging.
Aliquots of some deproteinized samples (one-twentieth of the total sample) were transfected into DH5␣-competent cells (Max Efficiency, Life Technologies, Inc.) as prescribed by the vendor. Individual colonies were expanded in liquid culture, and isolated plasmid was sequenced in the region corresponding to the rejoined DSB by the fluorescent-dideoxy method using an ABI Prism automated sequencer. The sequencing primer was TATGCTTGCTGTGCTTACTG, which lies 80 base pairs from the DSB site.

Extracts of CHO-K1 Cells Accurately Rejoin DSBs with
Terminally Modified 3Ј-Overhangs and 1-Base Gaps-Radiation-induced DNA DSBs are formed when multiple radicals, emanating from a single ionization track, attack and fragment closely opposed deoxyribose moieties in both strands (28). To examine the repair of such lesions by end joining, a model DSB was constructed (Fig. 1). Each end of the break had a PG-terminated 3-base 3Ј-overhang with the sequence -ACG*. Thus, this substrate mimics the break that would result from free radical-mediated cleavage at each T in the self-complementary sequence ACGT. Accurate repair of such a break ( Fig.  2A) would require annealing of the CG sequences in each overhang, removal of PG, fill-in of the 1-base gap in each strand (presumably using the 3Ј-overhang from the other end of the break as a template), and finally ligation. In order that processing of the break could be followed, it was labeled with 32 P 14 bases from one of the 3Ј termini.
This substrate was incubated in CHO-K1 whole-cell extracts and then cut with restriction enzymes on each end to release short fragments, which were then analyzed on sequencing gels. As shown in Fig. 3A, the CHO-K1 extracts were able to rejoin such breaks, and the predominant end-joined product was a 43-base fragment, corresponding to accurate rejoining by annealing, gap filling, and ligation, as described above (Fig. 2). Analysis of repair joints in individual plasmid clones recovered from the repair reaction confirmed the sequence of this repair joint as well as its predominance among repair products (Table  I). In theory, such a product could only be formed from the head-to-tail joining of two ends, each of which had the -ACG* overhang. To test this requirement, parallel experiments were performed with a substrate having an -ACG* overhang ligated into the XhoI side of the substrate only (Fig. 2B). The opposite end of this plasmid would have a 10-base 5Ј-overhang, formed as an intermediate in the construction (see Fig. 1) (23). As predicted, this "one-sided" construct did not yield any 43-base repair product, but instead yielded a mixture of end-joined products with fragment lengths of 42, 41, 40, 39, 37, and 35 bases (Fig. 3B). The 39-base product would correspond to removal of the 3Ј-overhang and fill-in of the 10-base 5Ј-overhang, followed by direct ligation of the resulting blunt ends. The 40-, 41-, and 42-base products would correspond to retention of 1, 2, and 3 bases of the single 3Ј-overhang, plus all bases of the 5Ј-overhang (Fig. 2B), as was confirmed by sequencing clones of repaired plasmid (Table I). The 37-and 35-base products would correspond to annealing of 2 and 4 bases, respectively, of the CGCG sequences on opposite sides of the break, presumably preceded by a 5-or 7-base 3Ј resection of the end with a 3Ј-overhang (Fig. 2B). Since the 3Ј-overhang construct will necessarily contain some one-sided construct due to incomplete ligation of the PG oligomers into the vector, some or all of the inaccurate joins may be derived from those contaminants; thus, the proportion of accurate joins for the full 3Ј-overhang construct is probably even higher than is experimentally observed (as much as 70% is some experiments) and may approach 100%. The higher apparent incidence of inaccurate repair events indicated by the sequencing data (Table I) is expected, since these data will include repair products from plasmids into which only the unlabeled oligomer had been ligated, but such products will not be visible in the gel assays (Fig. 3).
As a further verification that the 43-base product was indeed the result of single-base gap filling, the reactions were performed with various combinations of dNTPs (Fig. 3A). As expected, only dTTP was essential for generation of the 43-base product, supporting the proposal that it was formed by gap filling opposite the template adenine in each -ACG overhang. The absence of dATP had little effect on the product distribution but, curiously, reduced overall end joining efficiency. The complete absence of any dNTPs appeared to accelerate 3Ј resection in extracts from all of the cell lines. Replacement of dTTP with ddTTP promoted formation of a 16-base species consistent with the expected trapping of an intermediate in which the one-base gap had been filled in with ddTTP, preventing the final ligation (Fig. 3E).
Thus, the presence of overhangs with a partial complementarity of as little as 2 bases was sufficient to permit accurate end joining with restoration of the original sequence, while lack of such complementarity led to multiple repair products corresponding to various degrees of 3Ј-resection.
Accurate End Joining Requires Ku but Not DNA-PKcs-Ku is a heterodimeric DNA-binding protein that has been implicated in DSB repair in vivo (1-4) and has been found to promote end-to-end DNA association (5,7,29) and to stimulate DSB rejoining by mammalian DNA ligases in vitro (29,30). To assess the possible effects of Ku on the efficiency and accuracy of end joining, parallel experiments were carried out using extracts of the Ku-defective CHO derivative xrs6 (31,32). As shown in Fig. 3A and Table I, there was no detectable formation of the accurately joined, 43-base repair product in xrs6 extracts. Instead, the predominant product was the 35-base product, consistent with 7-base 3Ј resection in each strand and annealing of the CGCG microhomology (see Fig. 2). In addition, there appeared to be more extensive 3Ј processing in the xrs6 than in the CHO-K1 extract. For xrs6 extracts, no unprocessed 3Ј-PG ends remained, and more than half of the radiolabel had been lost entirely, presumably due to resection at least several bases into the duplex region of the DNA. Likewise, sequencing revealed formation, in the xrs6 extracts, of several repair joints with even larger deletions at the break site (Table I). Sequencing also confirmed the predominance of the product with apparent splicing at the CGCG microhomology as well as the absence of any accurate repair joints.
To assess whether the differences between CHO-K1 and xrs6 extracts were due to presence or absence of Ku or to some ancillary difference between the cell lines, extracts were also prepared from an xrs6 derivative cell line that had been stably transfected with a hamster Ku80 gene (33). Unlike xrs6 extracts, these xrs6-Ku80 extracts were competent to form the accurately joined 43-base repair product ( Fig. 3A and Table I).
Moreover, the addition of purified human Ku also restored accurate repair in the xrs6 extracts, promoting formation of the 43-base product while suppressing formation of the 35-base product (Fig. 3A). However, whereas the accuracy of end joining was restored to near that of CHO-K1 extracts, the efficiency was much lower, with only 0.2% joining in the Ku-supplemented xrs6 extracts versus 7% joining in CHO-K1. The reason for this lower efficiency is not known, but it could reflect differences in Ku phosphorylation state, differences between the human and hamster proteins, or secondary effects of Ku deficiency on other proteins in the cell. Nevertheless, taken together, the results strongly suggest a specific and apparently absolute requirement for Ku in the accurate end joining of 3Ј-staggered, free radical-mediated DSBs.
Similar experiments were performed with a substrate having 2-base-recessed 3Ј-PG ends and cohesive 2-base 5Ј-overhangs. With this substrate (which, unlike an actual free radical-mediated DSB, would not have a 1-base gap in each strand when the ends were annealed), the presence of Ku likewise suppressed 3Ј 3 5Ј resection and promoted accurate end joining, although the requirement was not absolute (Fig.  3C). For CHO-K1 and xrs6-Ku80 extracts, 89 and 88% of the head-to-tail joins at 6 h incubation resulted in the accurate, 37-base product, consistent with annealing of the existing 2-base 5Ј-overhangs. The xrs6 extract yielded primarily the 35-base product (80%), consistent with a 2-base 3Ј resection followed by annealing of the CGCG microhomology (see Fig. 2), but it did yield some 37-base product (20%). The addition of purified human Ku to the xrs6 extracts restored accurate end joining (85% 37-base product), although the extent of end joining was only about half that seen with the CHO-K1 extract. All of the extracts yielded only very small amounts of the 18-base (simple annealing) and 16-base (2-base resection) head-to-head joining products, suggesting that recircularization was preferred over intermolecular dimerization. Again, there was more end processing (PG removal, fill-in, and, at longer times, loss of radiolabel) in xrs6 than in CHO-K1 extracts. Thus, either Ku transfection or direct Ku addition restored accurate end joining of the 5Ј-overhang substrate to the xrs6 extracts.
To assess whether end joining required DNA-PKcs, effects of the DNA-PKcs inhibitor wortmannin were examined. Me 2 SO solvent (2%) alone reduced both the efficiency and the accuracy of end joining. Wortmannin, at a concentration sufficient to FIG. 2. End joining substrates and repair products. A, the 3Ј-PG(•) 3Ј-overhang substrate mimics the DSB that would result from free radical-mediated deoxyribose fragmentation on a 3-base 3Ј stagger. Accurate repair (43-base product) would require alignment-based fill-in prior to ligation. B, the one-sided 3Ј-overhang substrate was included as a control to assess the effect of incomplete ligation in construction of the 3Ј-overhang substrate. In theory, it cannot form any 43base product, but it could form a 42-base product by fill-in of the 10-base 5Ј-overhang, abutting of the ends, and continuation of fill-in using the 3Ј-overhang as a template; 41-, 40-, and 39-base products would result from removal of 1, 2, and 3 bases, respectively, from the 3Ј-overhang prior to fill-in. C, the 5Ј-overhang substrate lacks the 1-base gaps but is otherwise similar to a 2-base-5Ј-staggered free radical-mediated DSB. All three substrates can, by various degrees of 3Ј-resection and/or fill-in, be converted to an intermediate with self-complementary CGCG 5Ј-overhangs, the alignment of which would result in the 35-base repair product.
completely suppress DNA-PK activity (10,27,34), had no additional effect on formation of the 35-and 37-base end joining products and produced only a modest reduction (less than 2-fold) in the formation of the 43-base product (Fig. 3D). To assess whether DNA-PKcs might play an essential role independent of its kinase activity, experiments were also performed with extracts of the CHO derivative XR-C1, which lacks detectable DNA-PKcs protein (18). Unlike xrs6 extracts, these extracts produced significant amounts of the 43-base end joining product (Fig. 3E), confirming that DNA-PKcs is not essential for accurate end joining in this system.
End Joining in Extracts of Human Cells and Xenopus Eggs Shows Similar Specificity-In extracts of Xenopus eggs, end joining of both simple restriction cuts and 3Ј-PG-terminated DSB constructs can be completely blocked by DNA-PK inhibitors such as wortmannin and LY294002 (10,27); end joining is also wortmannin-sensitive in at least some types of human cell extracts (21). However, wortmannin had little or no effect on end joining of PG-terminated substrates in CHO or xrs6 extracts. In an attempt to determine the basis for these differences and examine the possible role of DNA-PK in joining of the 3Ј-overhang PG substrate, similar experiments were performed in Xenopus egg extracts and in whole-cell extracts of human lymphoblastoid cells.
In the Xenopus extracts, the product distribution was nearly identical to that seen in CHO-K1 extracts (Fig. 4A). The 43base product was most prominent, but there was also significant formation of the 42-base and several shorter products. As in CHO-K1 extracts, the one-sided 3Ј-overhang construct yielded only the 42-base and shorter products, confirming that the 43-base product was formed by annealing of two -ACG overhangs. Unlike the mammalian extracts, Xenopus extracts also yielded small quantities of 29 -34-base products, which would correspond to partial or complete loss of the 10-base 5Ј-overhang (see Fig. 2B).
In Xenopus egg extracts, formation of all end-joined products was efficiently blocked by 10 M wortmannin (Ͼ20-fold reduction in end-joined products; Fig. 4), a concentration sufficient to completely suppress DNA-PK activity (27). Because joining was less sensitive to Me 2 SO in Xenopus extracts than in the mammalian extracts (possibly due to the lower temperature), the wortmannin effect was unambiguous. In addition, wort-

FIG. 3. End joining in hamster cell extracts and the effect of Ku deficiency. A-C, the internally labeled (*) 3Ј-PG-terminated (•) 3Ј-overhang (A)
, one-sided 3Ј-overhang (B), and 5Ј-overhang (C) substrates described in the legend to Fig. 2 were incubated for 6 h (A and B) or for 1, 6, or 16 h (C) with extracts of CHO-K1, Ku-deficient xrs6, or Ku-transfected xrs6-Ku80 (xrs-Ku) cells and then cut near each end with XhoI and BstXI (see Fig. 2) for analysis of end processing and end joining. The 43-and 37-base products reflect accurate head-to-tail end joining (recircularization) of the 3Ј-overhang and 5Ј-overhang substrates, respectively; the analogous head-to-head end joining products would be 24 and 18 bases. In some cases, purified Ku protein (0.1 or 0.4 g) was added as indicated, with incubation for 6 h. All reactions contained a 50 M concentration of each dNTP, except as indicated. In A and B, the 15PG band corresponds to the initial 3Ј-PG-terminated substrate. PG removal yields the 15-mer, and 3Ј resection yields shorter species, with the 12-mer corresponding to a blunt end. In A, a portion of the gel is reproduced at 10ϫ contrast to show weak bands, and above that, a replicate experiment is shown in which the gel was exposed to x-ray film to improve resolution of closely spaced bands. (The dTTP-depleted sample was lost from the first experiment.) In C, the 10PG band represents the unprocessed recessed 3Ј-PG terminus, and the 12-mer represents fill-in to a blunt end. D, effect of 10 M wortmannin on end joining; conditions and substrates are the same as in A and C. E, the 3Ј-PG-terminated 3Ј-overhang substrate was incubated for 6 h in extracts of CHO-K1 cells or of DNA-PKcsdeficient XR-C1 cells and analyzed as in A. The size markers (M) are 5Ј-end-labeled oligomers of the indicated lengths and with the same sequence as the expected end joining products or processing intermediates. DMSO, Me 2 SO. mannin prevented the formation of a species migrating in the position expected for a 3Ј-hydroxyl 16-mer (Fig. 4, A and B). This species could be an intermediate corresponding to fill-in of the expected 1-base gap in the labeled strand of the annealed ends, prior to ligation. However, as noted above, a 16-base resection-dependent head-to-head joining product could also be formed. To distinguish between these possibilities, ddNTPs were added in an attempt to trap the extension product prior to the final ligation step (Fig. 4B). The addition of either ddTTP or all four ddNTPs largely prevented formation of the 43-base end joining product and increased the intensity of the 16-base fragment, while ddATP had little effect. These results strongly suggest that the 16-base species was indeed the filled-in but unligated intermediate and that fill-in was blocked by wortmannin, presumably as a result of inhibition of the kinase activity of DNA-PK (27), which can regulate accessibility of DNA ends to enzymatic processing (12). In contrast to a recessed 3Ј-PG terminus (10), there was significant removal of PG from the 3Ј-overhang despite the presence of wortmannin.
In human cell extracts, a slightly different pattern of end joining was seen (Fig. 5). The 3Ј-overhang substrate yielded primarily the accurate 43-base end joining product, along with variable amounts of the 35-and 37-base products, but none of the 39 -42-base products. The 5Ј-overhang substrate yielded exclusively the accurately joined 37-base product and none of the 35-base resection-dependent product (not shown). In addition, human extracts yielded much more of the 24-, 18-, and 16-base head-to-head end joining products than did hamster and Xenopus extracts, suggesting that there was little or no preference for recircularization over intermolecular end-to-end dimerization. Indeed, analysis on agarose gels indicated that, as reported previously (21), the human cell extracts yielded exclusively intermolecular products, with no detectable recircularization (Fig. 5C). As in the Xenopus extracts, formation of all end joining products was blocked by wortmannin, while substitution of ddTTP for dTTP prevented formation of the accurate 43-and 24-base end joining products and trapped the filled-in but unligated intermediate (16-base fragment in Fig.  5B). Although end joining in human cell extracts was significantly inhibited (about 2-fold) by Me 2 SO, the inhibition by wortmannin was much greater, ϳ20-fold.

DISCUSSION
The capacity of mammalian cells to join mismatched DNA ends based on short partial complementarities in single strand overhangs was first noticed in the repair joints formed when linearized plasmids with mismatched restriction ends were recircularized upon transfection into monkey CV-1 cells (35). In some of these repair joints, partial complementarities in 3Јoverhangs had apparently been used for alignment, despite the a The sequence is shown in terms of the top strand in Fig. 2. The vertical line indicates the transition between sequences derived from opposite ends of the break, when it can be determined unambiguously from the joint sequence. Bold letters show the CGCG micronomology.
b Predicted length of the top strand of the fragment that would be produced by BstXI/XhoI cleavage. c One joint also had an additional 4-bp deletion CGCGTCCGGCA 3 CGCGGCA, in the right-hand side of the joint, where the underline indicates a direct repeat that might have been used for splicing. d One joint also had an inserted A, giving AAACGCG at the left-hand side of the joint. e The repair joint was AACGCGACG. . .CGTCGCGTCCGGCAA 3 AACGCGACAA, 12 bp shorter than the accurate product, where ". . ." indicates the break in the initial substrate.

FIG. 4. End joining in Xenopus egg extracts.
A, the various substrates were incubated in extracts at 13°C for 6 h and then cut with XhoI and BstXI. B, same as A, except that wortmannin (10 M) and various ddNTPs were present, all samples contained 2% Me 2 SO (DMSO), and some samples were cut with XhoI alone so that end processing intermediates and head-to-head joining products, but not head-to-tail recircularization products, were detected. See Fig. 2 and the legend to Fig. 3 for description of labeled products and processing intermediates.
fact that such alignment would initially leave single strand gaps in the aligned ends. This phenomenon implied a quite remarkable process, termed "postrepair ligation" (Fig. 2A), wherein the putative repair complex must have held the two ends in juxtaposition and maintained base pairing of a 2-or 3-base complementarity in the overhangs while at the same time allowing a polymerase to fill in the gaps to produce a ligatable substrate. Later in vitro experiments using Xenopus egg extracts (36,37) confirmed that such processing was indeed possible, and thus an "alignment factor" was proposed but not identified.
While such an alignment mechanism has been invoked primarily to explain the microhomologies typically found at sequence junctions of various illegitimate recombination events (38,39), the same mechanism would be ideally suited for the accurate repair of DSBs induced by ionizing radiation and other free radical-based genotoxins. Radiation-induced breaks are the result of clustered free radical-mediated sugar fragmentation (28), and therefore are likely to be randomly staggered and to have cohesive overhangs of one to several bases, along with blocked 3Ј termini (14,15,17), a one-base gap in each strand, and possible ancillary base damage in the overhangs (40). Thus, the use of partial complementarities for alignment of the overhangs during gap filling and ligation would be essential for accurate restoration of the original sequence (Fig. 2). The same requirement would apply to repair of breaks induced by enediyne antibiotics, which involve similar sugar fragmentation on a defined 2-or 3-base 3Ј stagger (41,42).
The possibility that Ku might be the long sought "alignment factor" was suggested by its capacity to promote DNA end-toend association (5)(6)(7) and to enhance the joining of either blunt or cohesive DSBs by mammalian DNA ligases (29,30). The strongest evidence, however, comes from recent analysis of repair joints formed in CHO-K1 and xrs6 extracts during the recircularization of substrates with mismatched restriction ends (8). These data imply a critical role for Ku in preserving 3Ј-overhangs during end joining. Repair joints requiring alignment-based fill-in (see Fig. 2) prior to ligation, in particular, were only formed in extracts from cells expressing functional Ku.
The present results imply a role for Ku in the accurate end joining of staggered free radical-mediated DSBs. For a break formed on a 3-base 3Ј stagger (giving a 2-base complementarity and a 1-base gap in each strand), a majority, perhaps nearly all, of the end joining events in CHO-K1 extracts reflected complete restoration of the original sequence, whereas in Ku-deficient xrs6 extracts, no accurate joins were detected. Accurate end joining could be restored by the addition of highly purified (25) human Ku to the xrs6 extracts, clearly implicating Ku (rather than, for example, other proteins that might have become either destabilized or overexpressed in Ku-deficient cells) as the critical factor in enforcing the fidelity of end joining. Analogous experiments implicate Ku in preserving the accuracy of joining of a substrate with cohesive 5Ј-overhangs as well, although in this case the requirement for Ku was not absolute. Intriguingly, however, whereas for the 5Ј-overhang substrate both the accuracy and the efficiency of joining in the Ku-supplemented extract was comparable with that seen in CHO-K1 extract, for the 3Ј-overhang substrate the efficiency was reduced by more than 30-fold. These results suggest qualitative differences in the precise function of Ku in the joining of the two types of breaks, such that one process is more sensitive to as yet undefined differences between the endogenous hamster and exogenous human Ku.
The internal labeling of the 3Ј-PG substrates allows intermediate processing of the DNA ends to be analyzed at singlenucleotide resolution, directly demonstrating that, as inferred from analysis of repair joints (8), 3Ј-end processing and resection are more extensive in Ku-deficient extracts. Nevertheless, the amount of free 3Ј-hydroxyl-terminated -ACG overhangs available for fill-in and ligation, particularly at short incubation times (data not shown), was comparable in xrs6 and CHO-K1 extracts, yet formation of accurately joined 43-base products in the xrs6 extracts was undetectable, at least 50-fold lower than in CHO-K1 (Fig. 3A). This result implies a specific requirement for Ku in alignment during gap filling and/or ligation, in addition to its apparent role in protecting ends from 3Ј resection. The enzyme that initiates processing by removing the 3Ј-PG is not known, but it is probably not the apurinic/ apyrimidinic endonuclease Ape1/Hap1; although Ape1 is the only mammalian enzyme thus far identified that is capable of removing 3Ј-PGs, it has no detectable activity toward PGs on 3Ј-overhangs (43,44). Other activities for removing blocking groups from 3Ј termini have been detected in cell extracts but have not been identified or cloned (45,46).
Our data, like those derived from substrates with mismatched restriction ends (8), do not support models in which Ku or DNA-PK catalyzes melting of DNA ends, resulting in exposure and annealing of microhomologies within double strand regions (1,9). On the contrary, end joining events consistent with such annealing were much more frequent in Kudeficient xrs6 extracts and were completely suppressed by the addition of purified Ku (Fig. 3A). Moreover, in the Xenopus FIG. 5. End joining in human cell extracts. A, the various substrates were incubated at 37°C for 6 h, in the presence of 10 M wortmannin and/or 2.5% Me 2 SO (DMSO) and then cut with XhoI and BstXI. B, same as A, except that certain dNTPs were replaced with ddNTPs as indicated. C, same as A except that substrates and products were analyzed on an agarose gel without restriction cleavage; for clarity, the top portion of the marker lane is shown at 5ϫ reduced contrast. system, it has been shown that such joining is efficiently catalyzed by extract fractions devoid of Ku and DNA-PKcs (47). Recent in vitro studies (48) suggest that the Mre11 complex is a much more likely candidate for catalyzing end joining based on microhomologies within duplex regions near DNA ends.
DNA-PKcs-deficient XR-C1 extracts, unlike Ku-deficient xrs6 extracts, yielded substantial quantities of accurate 43base repair products. Moreover, the DNA-PK inhibitor wortmannin had little effect on end joining in CHO-K1 extracts. These results suggest that only Ku and not DNA-PKcs is essential for accurate end joining in vitro. However, in human and Xenopus extracts, end joining of the same substrates is qualitatively similar but is wortmannin-sensitive. Although the wortmannin data are to some extent complicated by the effects of Me 2 SO alone (a finding that emphasizes the need for concurrent solvent controls whenever this inhibitor is used), we consistently find nearly complete suppression of end joining by wortmannin and other DNA-PK inhibitors in these extracts. While the disparity in wortmannin sensitivity could be due to technical factors such as the differences in preparation of the various extracts, it could reflect the levels of DNA-PKcs. DNA-PK activity is typically 50 -100-fold lower in hamster and other rodent cell extracts than in human cell or Xenopus egg extracts (49 -52), and electrophoresis mobility shift assays as well as Western blot assays are consistent with this difference being due to a much lower level of DNA-PKcs protein in rodent cells (53), including CHO cells. 2 Thus, a hypothesis consistent with all of the available data is that although DNA-PKcs is not required for accurate end joining in vitro, sufficient quantities of it produce an inhibitory effect that can be alleviated by activation of its kinase activity.
Recent chemical cross-linking studies (54), as well as physical-chemical data on the binding and activation of Ku and DNA-PKcs (55,56), suggest that when Ku recruits DNA-PKcs to a DNA end, DNA-PKcs then occupies the extreme end of the DNA and displaces Ku ϳ10 base pairs into the interior. The DNA end is probably threaded into a channel in DNA-PKcs (revealed by electron crystallography (57)) that is open at both ends and can accommodate ϳ12 base pairs of duplex DNA. It is also known that Ku can promote DNA end-to-end association and that both Ku and DNA-PKcs, when present, often remain bound at the end-to-end junction (5,7). Thus, the present data could be explained by a model of in vitro end joining wherein there is an absolute requirement for proper positioning of Ku such that it bridges the two DNA ends being joined, in order for PG removal, alignment-based fill-in, and ligation to proceed (see Fig. 2A). In the Xenopus and human extracts, Ku may be temporarily displaced from the ends and into the interior of DNA by DNA-PKcs, which may still maintain end-to-end association but not permit end processing. As has been proposed previously (12), DNA-PK-catalyzed phosphorylation (of DNA-PKcs, Ku, or some other protein in the complex) might then promote either DNA-PKcs dissociation or sliding of the whole complex along the aligned DNA ends, thus allowing Ku to move to its proper position bridging the ends. Phosphorylation of both Ku70 and Ku80 during end joining has been detected in Xenopus egg extracts (52), although in similar experiments we have detected only Ku80 phosphorylation. 3 Thus, in this model, the role of DNA-PKcs is primarily regulatory, initially displacing Ku and thus controlling the timing of end processing through specific phosphorylations, the details of which remain to be defined.
The apparent dispensability of DNA-PKcs in accurate end joining in CHO-K1 extracts does not necessarily imply that it plays no important role in end joining in intact rodent cells. Indeed, at least some rodent cells lacking DNA-PKcs exhibit the same radiosensitivity and DSB repair deficiency as those lacking Ku (4). One possible explanation for this disparity is that DNA-PKcs may play a role in bringing broken DNA ends together, a task that might be much more difficult and complex in intact cells than in in vitro assays, where free ends are relatively abundant. Consistent with this proposal, DNA-PKcs has been reported to promote intermolecular ligation by ligase IV plus XRCC4, while inhibiting intramolecular ligation (58). However, the fact that intramolecular recircularization dominates end joining in Xenopus extracts but does not occur at all in human extracts, despite an apparent abundance of DNA-PKcs in both (49 -52), remains to be explained. Alternatively, recent in situ fluorescence labeling studies suggest that DNA-PK may be largely or at least partly responsible for the phosphorylation of histone H2AX, an event that occurs within minutes of irradiation (59). Localization of phosphorylated H2AX at putative DSB repair foci precedes the colocalization of Brca1, hRad51, and/or the hMre11-hRad50-NBS complex at the same foci, suggesting that H2AX may recruit these DSB repair factors to sites of multiple or difficultto-repair breaks. Thus, DNA-PK holoenzyme may play a critical role in initiating a complex choreography that serves to optimize repair by assigning broken ends to specific repair pathways (60 -62), some of which probably do not utilize Ku for end alignment. Indeed, a defect in DNA-PK activation rather than in end alignment could be the primary reason for the radiosensitivity of Ku-defective cells. The availability of a stringent reconstitution assay (Fig. 3A) may facilitate the identification of Ku mutants differentially defective in each of these two functions and thus allow assessment of their relative importance in DSB repair efficiency, repair fidelity, and radiosensitivity. The assay might also be useful in determining what domains and biochemical properties of Ku70 and Ku80 are required for particular steps in the alignment-based end joining pathway.