An archaeal endonuclease displays key properties of both eukaryal XPF-ERCC1 and Mus81.

Structure-specific nucleases of the XPF/Mus81 family function in several DNA recombination and repair pathways in eukaryotes, cleaving a variety of flap and branched DNA substrates. Mus81 and XPF are clearly related evolutionarily but differ markedly in their substrate specificity and protein partners. We demonstrate that the XPF endonuclease from Sulfolobus solfataricus, which is dependent on the sliding clamp proliferating cell nuclear antigen for activity, represents an ancestral form of the XPF/Mus81 family, with key properties in common with both enzymes. The archaeal XPF has a domain organization and sequence preference very similar to eukaryal XPF-ERCC1. However, the archaeal enzyme has a pronounced preference for Mus81-type substrates such as D loops, nicked four-way junctions, and 3' flaps. These all have in common a 5'-DNA end next to the cleavage site. The availability of the sliding clamp proliferating cell nuclear antigen may dictate the activity of Sulfolobus XPF in vivo.

Branched DNA structures are created in many circumstances, arising both as a direct consequence of DNA damage and as intermediates in cellular DNA recombination and repair pathways. These structures must be processed correctly by structurespecific nucleases prior to DNA replication. In eukaryotes, the related nucleases XPF-ERCC1 and Mus81-Eme1/Mms4 (abbreviated here to Mus81*) process a variety of DNA substrates with 3Ј branches and bubbles (see Table II for a definition of the DNA substrates). The cellular functions of these important enzymes are still being delineated, as is the molecular basis for their interaction with DNA substrates.
Most euryarchaea encode a clear homolog of eukaryal XPF, including the N-terminal helicase and C-terminal nuclease domains. The homolog from Pyrococcus furiosus, Hef, cleaves a variety of fork and flap structures that are substrates for eukaryal Mus81* (1). The crystal structure of the nuclease domain of Pyrococcus Hef unveiled the relationship of this domain with a superfamily of nucleases including the restriction enzymes, Holliday junction resolving enzymes Hjc and T7, endonuclease I, and other DNA repair enzymes (2). Crenarchaeal XPF homologs lack the N-terminal helicase domain altogether and consist only of the nuclease and HhH 2 domains (see Fig. 3). In this respect, they may represent the simplest and perhaps ancestral version of the XPF/Mus81 nuclease. We showed previously that the XPF hom-olog from Sulfolobus solfataricus (SsoXPF) 1 cuts flap or splayed duplex structures (3) and that its activity is crucially dependent on an association with the sliding clamp PCNA, via a C-terminal PCNA interaction motif (3). Here we report the detailed kinetic analysis of the substrate specificity of Sulfolobus XPF. We demonstrate key similarities with both the eukaryal XPF and Mus81* nucleases.

MATERIALS AND METHODS
Recombinant Proteins-S. solfataricus XPF and PCNA heterotrimer were expressed and purified as described previously (3,4).
DNA Substrates-The oligonucleotides used for substrates (Table I) were 5Ј-32 P-end-labeled and assembled into various structures (see Table II) by slow cooling from 85°C to room temperature for 3 h or overnight. Substrates were purified on a native 6% acrylamide gel followed by electro-elution and ethanol precipitation. Size markers (G and A) were prepared from labeled substrates using standard protocols.
Endonuclease Assays-Nuclease reactions (10 l) were assembled in 30 mM HEPES, pH 7.6, 5% glycerol, 40 mM KCl, 0.1 mg/ml bovine serum albumin, and 0.1 mg/ml calf thymus DNA with 80 nM DNA substrate and 1 M XPF-PCNA and equilibrated at 55°C. Cleavage was initiated by adding MgCl 2 to a final concentration of 10 mM, mixed briefly, and incubated at 55°C. 5-l aliquots were taken at selected time points and added to chilled stop solution (10 mM Tris-HCl, pH 8, 10 mM EDTA, 0.1 mg/ml calf thymus DNA) to terminate the reaction. DNA was ethanolprecipitated and analyzed on a denaturing polyacrylamide/urea/Trisborate-EDTA gel. Uncut substrate and products were quantified (Image Gauge, Fuji), and the reaction rate was obtained by linear regression. In the experiment, to determine the influence of a 5Ј-phospho group on activity, the PCNA-XPF concentration was reduced to 0.1 M to allow more accurate quantitation.

RESULTS
First, we determined the structure specificity and requirements of Sulfolobus XPF. To pin down the function of structure-specific nucleases, it is necessary to define their specificity for the different branched DNA structures that arise in vivo due to DNA damage and repair pathways. Quantitative data for eukaryal XPF-ERCC1 and Mus81* are not yet available. We have carried out a systematic quantitative analysis of the substrate preference of SsoXPF using a panel of branched DNA substrates (Table II). In each case, the sequence of the strand cleaved by XPF is the same, although its length varies in some cases.
In general terms, the order of substrate preference of SsoXPF matches that of the Mus81* family rather than the eukaryal XPF family. The best substrate is a 3Ј flap followed by a nicked duplex, nicked three-way junctions, and nicked fourway junctions. The efficient cleavage of a nicked duplex substrate demonstrates that no single-stranded flap is required, rather a discontinuity in the DNA duplex is enough to promote the nuclease activity. These DNA species are all the preferred substrates of Mus81* and share in common a 5Ј-DNA end near the branch point or point of discontinuity that has been shown to be essential in directing the activity of Mus81* (5). The 3Ј flap and nicked three-way substrates are analogs of a stalled and reversed replication fork, respectively (6). In contrast, a substrate analog of an active replication fork with a 3Ј-DNA end near the branch point is cut much more slowly, again in common with Mus81* (6). A splayed duplex species, the preferred substrate of eukaryotic XPF-ERCC1, is cleaved ϳ10-fold more slowly than a 3Ј flap substrate. Lastly, four-way junctions are very poor substrates for SsoXPF, cut more than 2 orders of magnitude more slowly than the preferred substrates.
The clear preference of SsoXPF for a 5Ј-DNA end near the cleavage site mirrors that reported previously for Mus81* from Saccharomyces cerevisiae and Schizosaccharomyces pombe (5,6). In vivo, most substrates of this type would have a phosphate group at this 5Ј position, although this was not present in the synthetic substrates used for Mus81*. To test whether phosphorylation of this terminus was important for the activity of SsoXPF, we compared the rates of cleavage with a nicked duplex substrate with and without a 5Ј-phosphate at the nick (Fig. 1). We found a measurable although modest increase in the cleavage rates of a phosphorylated substrate as compared with a non-phosphorylated one, suggesting that the archaeal XPF (and probably also Mus81*) does not require the presence of a 5Ј-phosphate to recognize this DNA end.
Recent studies of Mus81* suggest that it can play an important role in the generation of crossover recombination events by cleavage of D loop (displacement loop) substrates formed as a consequence of strand invasion after a chromosome sustains a double strand break (7,8). A D loop bubble is a poor substrate for SsoXPF in the absence of an invading 3Ј-ssDNA end (Fig.  2B, left). This structure is equivalent to an NER bubble, the favored substrate of eukaryal XPF-ERCC1 (9). Weak cleavage of the b75 strand is seen at the splayed duplex junction (   table I). The labeled strand is indicated with a circle at the 5Ј end, and cleavage sites are indicated with an arrow. All rates were measured under pseudo single turnover conditions as described under "Materials and Methods." Rates for the two fastest-cut substrates were estimated from initial time points, and no standard error is indicated. 2B, white arrows, rate 0.02 min Ϫ1 ). The z75 strand is also cut weakly at the equivalent site on the other side of the bubble (rate 0.054 min Ϫ1 ). Once the invading strand is present, however, SsoXPF is strongly activated to cut the b75 strand at a new position (Fig. 2B, a, and C, black arrows). The invading strand creates a good substrate for SsoXPF, as observed for Mus81*, probably stimulated by the 5Ј-DNA end paired with the invading strand ( Fig. 2A, 5Ј end of oligo r26 -50) (7). The rate of cleavage at this new position is at least 50-fold faster than those for the substrate without the invading strand. Cleavage at site b on the z75 strand (Fig. 2B, b, and C, gray arrows) is also stimulated by the presence of the invading duplex but is cut significantly more slowly than site a. Second end capture and DNA synthesis result in the formation of a 3Ј flap substrate (site c) that is the most preferred substrate for SsoXPF (Table II). This is the second site proposed by Osman et al. (7) for Mus81*. Cleavage at sites a and c, coupled with capture of the second strand, DNA synthesis, and strand exchange, would result in crossover recombination products (Fig.  2D). Any 3Ј or 5Ј flaps generated as a result of this process could be removed by XPF or Fen1. Cleavage at site a alone would dissociate a D loop and could be functionally relevant for resetting of collapsed replication forks (10). Every strand scission was mapped 3Ј to pyrimidine residues, confirming that although SsoXPF cuts preferentially at Mus81*-type substrates, it retains key characteristics of XPF-ERCC1. DISCUSSION In addition to the clear similarities in their domain organization (Fig. 3), the archaeal XPF has the same sequence preference for cleavage at pyrimidines as eukaryal XPF (3). In contrast, Mus81* has little or no sequence dependence (5). SsoXPF has a significant activity against splayed duplex substrates that are found at the 5Ј side of lesions repaired by the eukaryal XPF protein during NER. These substrates are cut very poorly by Mus81*. Taken together, these data suggest a close relationship between the eukaryal and archaeal XPF enzymes.
On the other hand, our kinetic analysis of the substrate preference of SsoXPF demonstrates clearly that the best substrates resemble stalled replication forks, D loops, and nicked Holliday junctions, with a 5Ј-DNA end near the branch point or strand discontinuity (Table II and Fig. 3), a feature in common with Mus81*. Cleavage of 3Ј flap substrates by Mus81-Mms4 is strongly stimulated by the presence of a 5Ј terminus within 4 nucleotides of the branch point (5). The highly efficient cleav-age of a nicked duplex substrate demonstrates that the archaeal XPF protein seeks out strand discontinuities that allow DNA deformation, flanked on either side by duplex DNA, rather than branched DNA substrates per se. This is also the case for Mus81*, which cleaves nicked duplexes efficiently to generate 5-nucleotide gapped duplexes (5). These substrates are cut very poorly by eukaryal XPF-ERCC1. D loop substrates are also cut by SsoXPF and Mus81* in a similar way, with strong stimulation observed when strand invasion occurs.
Therefore, the archaeal XPF clearly has the potential to act as both a Mus81-like flap endonuclease and an XPF-like repair endonuclease. Previous comparisons of the activity of Mus81-Mms4 with Rad1-Rad10 have come to the conclusion that the specificities of the two enzymes do not overlap. The presence of a 5Ј-DNA terminus within 4 nucleotides of the branch point results in strong stimulation of Mus81-Mms4 and reduces the activity of Rad1-Rad10 such that the two enzymes have at least a 100-fold preference for their specific substrates (5). SsoXPF bridges these two diverged proteins and likely represents the ancestral form of both enzymes.
Hyperthermophiles such as S. solfataricus suffer a higher frequency of DNA damage such as hydrolytic deamination, oxidation, and single strand breaks as a consequence of their higher growth temperatures (11). Many of these lesions are known to stall or collapse replication forks; for example, a single strand break is converted to a double strand break if replicated. Stalled and collapsed forks are a frequent event in bacteria and eukarya, which consequently have efficient mechanisms to recover from these potentially catastrophic events. One mechanism for fork rescue is via the formation of a Holliday junction, which can be resolved to allow replication restart (reviewed in Ref. 12). The substrate specificity we have described for SsoXPF is consistent with a role in the cleavage of stalled replication forks to generate substrates for recombination and/or resolution of early recombination intermediates (D loops), similar to Mus81*. In this regard, the requirement of SsoXPF for PCNA as an essential cofactor may be relevant as the target sites of the enzyme may be determined in part by PCNA loading. This is well illustrated by consideration of the D loop substrate (Fig. 2D). The recombination scheme requires extension by DNA replication of both the invading 3Ј strand (strand x50) and of the second DNA end after capture. This replication will result in loading of PCNA on both these duplexes. This may direct SsoXPF to cut at sites a and c, thus generating recombinogenic DNA ends. In contrast, cleavage at FIG. 3. Structure and evolution of the XPF/Mus81 nuclease family. The conserved nuclease (nuc) domain is found in many other nucleases involved in DNA restriction and repair pathways. In the XPF/Mus81 family, the nuclease domain is linked to a helix hairpin helix (HhH 2 ) DNA binding domain. The archaeal XPF enzymes are homodimers, whereas the eukaryotic enzyme has duplicated to form heterodimeric proteins, allowing subunit specialization, and to form nucleases with differing substrate specificity. The archaeal and Mus81 enzymes show a distinct preference for substrates with a 5Ј DNA terminus (red circle), whereas eukaryal XPF-ERCC1 has specialized for cleavage of splayed duplex or bubble substrates. site b could be disfavored in vivo as PCNA will not be loaded on this duplex.
The data presented here do not rule out the possibility that SsoXPF also functions in an archaeal NER pathway. The function of eukaryal XPF-ERCC1 is largely determined by the protein:protein interactions it makes. In NER, this includes specific interactions between ERCC1 and XPA (13) and between XPF and the C terminus of the helicase XPB, a component of the multiprotein complex TFIIH (14). Thus XPF-ERCC1 is recruited into the assembling repair complex and positioned to introduce a single strand cleavage 5Ј of a DNA lesion. XPF-ERCC1 also has roles in the resolution of DNA interstrand cross-links (15,16) and in targeted gene replacement (17). These functions link XPF-ERCC1 to recombinational pathways and necessitate recruitment of the enzyme into different DNA processing complexes from those formed during NER. Similarly, interactions of SsoXPF mediated by PCNA may help to define its role in vivo.