Strains used are listed in Table 1. Standard yeast protocols were used for strain generation, growth, medium preparation and DNA damage sensitivity assays. As siz1Δ siz2Δ results in amplification of 2-micron plasmids (49), strains with siz1Δ siz2Δ mutations were cured of the plasmid as described (50). Spot assays were performed as described previously (7). Briefly, log phase cells were diluted 10-fold or 3-fold and spotted onto YPD (Yeast extract-Peptone-Dextrose) media with or without CPT, or irradiated with UV. Plates were incubated at 30°C and photographed after 24–72 h.

These were performed as described previously (7). In brief, cells were lysed by bead beating under denaturing conditions and TAP-tagged proteins were immunoprecipitated using immunoglobulin G (IgG)-Sepharose. These were then washed and eluted with loading dye, followed by sodium dodecylsulphate-polyacrylamide gel electrophoresis (SDS-PAGE) and western blotting with antibodies against SUMO (22) and the protein A portion of the TAP tag (Sigma-Aldrich). Damage-induced sumoylation was assayed by subjecting log-phase cells to 0.03–0.3% methylmethane sulfonate (MMS, Sigma-Aldrich), 50 ug/ml CPT (Sigma-Aldrich), or 0.2 M hydroxyurea (HU, US Biologicals) for 2 h, or UV irradiation using UV Stratalinker 1800 (Stratagene). For Figure 1A, UV and MMS doses were chosen based on comparable survival posttreatment as shown in Supplementary Figure S1A–B. Co-immunoprecipitation experiments were performed as described previously (51).

Chromosomal and plasmid-based SSA assays were performed as described (46). In brief, for chromosomal assay, log-phase cells grown in YP-Glycerol were plated on either YPD or YP-Galactose media. Only the latter induces HO expression, and thus double -strand break formation and SSA repair. Colonies were counted after incubation at 30°C for 3–4 days. SSA survival quantification was done by dividing the number of viable colonies on YP-galactose plates by that on YPD and multiplying this factor by 100. For plasmid-based assay, cells were transformed with either Bsu36I- or mock-digested pNSU208 plasmid, plated on media lacking leucine and incubated at 30°C for 3–4 days. Percent survival estimating SSA efficiency was calculated by dividing number of viable colonies of the Bsu36I-digested transformants by those of mock-digested ones.

The Escherichia coli strain Rosetta(DE3)pLysS was transformed with a bicistronic plasmid (gift from Dr. Steve Brill (52)) expressing (His)₆-Rad1 and Rad10, or (His)₆-Rad1-K32R and Rad10. The Rad1-K32R mutant was generated using site-directed mutagenesis. Protein expression was induced by 0.1 mM IPTG (Isopropyl β-D-1-thiogalactopyranoside) at 16°C overnight. A total of 9 g of cell paste was sonicated in 40 ml breakage buffer (50 mM Tris-HCl, pH 7.5, 10% sucrose, 2 mM ethylenediaminetetraacetic acid (EDTA), 200 mM KCl, 0.01% NP40, 1 mM β-mercaptoethanol and protease inhibitor cocktail containing pepstatin, aprotinin, benzamidine, chymostatin and leupeptin). The lysate was clarified by centrifugation (100k g, 4°C, 1 h), and the supernatant was incubated with 1 ml of HIS-Select nickel affinity gel (Sigma) for 1 h at 4°C. The beads were washed with 12 ml of buffer K (20 mM K₂HPO₄, 10% glycerol, 0.5 mM EDTA) containing 150 mM KCl. The bound proteins were eluted with buffer containing 50 mM KCl and imidazole (from 50 to 1000 mM). Pooled fractions containing Rad1-Rad10 (150–500 mM imidazole) were applied to a 1-ml Heparin column (GE Healthcare) followed by elution using 8 ml gradient of 275–1000 mM KCl in buffer K. The peak fractions of Rad1-Rad10 at ∼500 mM KCl elution were pooled, loaded onto a 0.5-ml MonoQ column (GE Healthcare) and eluted using a 5-ml gradient of 275–1000 mM KCl in buffer K. The Rad1-Rad10 fractions were concentrated to 400 μl in a VivaSpin-2 concentrator, and fractionated on a 23-ml Sephacryl S400 column (GE Healthcare) in K buffer containing 300 mM KCl. Peak Rad1-Rad10 fractions were concentrated to 2 μg/μl. To determine the oligomeric status of Rad1-Rad10, a 23 ml Sephacryl S400 column was eluted with buffer K containing 300 mM KCl (0.11 ml/min flow rate), and 0.35 ml fractions were collected. The indicated fractions were separated by SDS-PAGE and detected by western blotting using α-Rad1 antibody (Santa Cruz).

Purification of GST-Aos1/Uba2, His-Ubc9, His-Flag-Smt3, His-Flag-Smt3-KR, His-Siz1 (1–465) and Siz2 and the sumoylation assay were performed as previously (53). In brief, the 10 μl reaction contained 150 nM Aos1/Uba2, 0.5 μM Ubc9, 4.3 μM Smt3 or Smt3-KR, 0.4 μM Rad1-Rad10, 1 mM adenosine triphosphate (ATP) and buffer S1 (100 mM Tris-HCl pH 7.5, 10 mM MgCl₂) (53). In the indicated cases, 10–100 nM Siz1 or 10–100 nM Siz2 was added to the reaction.

All substrates were prepared as described (54). Synthetic oligonucleotides were purchased from VBC Biotech; sequences are available upon request.

Rad1-Rad10 was incubated with fluorescently labeled DNA substrates (4 nM) in 10 μl buffer D (40 mM Tris-HCl, pH 7.5, 50 mM KCl, 1 mM DTT and 100 μg/ml bovine serum albumin) at 37°C for 10 min. The reactions were stopped by adding loading buffer (60% glycerol, 10 mM Tris, pH 7.4 and 60 mM EDTA). Samples were separated on native polyacrylamide gel (10%) in 0.5×TBE buffer (40 mM Tris-HCl, 20 mM boric acid, 2 mM EDTA, pH 7.5). DNA was visualized by a scanner FLA9000 Starion (Fujifilm) and quantified using MultiGauge software (Fujifilm).

Rad1-Rad10 was incubated with fluorescently labeled DNA substrate (4 nM) in 10 μl buffer R (50 mM Tris-HCl, pH 7.5, 1 mM MgCl₂, 1 mM DTT and 100 μg/ml bovine serum albumin). The reaction mixtures were incubated at 30°C for 30 min and deproteinized by adding 0.1% SDS and 500 μg/ml of proteinase K for 5 min at 37°C. Samples were resolved on a 10% native polyacrylamide gel in TBE buffer. The fluorescent DNA species were visualized as in EMSA.

Recent proteomic and biochemical screens have identified Rad1 as one of the sumoylation targets during the DNA damage response (7,8,20). We confirmed Rad1 sumoylation in our W303 strain background using endogenously expressed and functional TAP-tagged Rad1 (Figure 1A, and data not shown). The sumoylated form of Rad1 was detected as a single band above the unmodified protein band on western blots when immunoprecipitated Rad1 was examined with anti-SUMO antibody (Figure 1A). Rad1 sumoylation was observed under normal growth conditions and an increase was seen upon exposure to UV, the alkylation agent MMS and the Top1 poison CPT (Figure 1A and Supplementary Figure S1C). Treatment with the ribonucleotide reductase inhibitor HU that is not known to require Rad1-mediated repair had no effect (Figure 1A). We note that even upon UV, MMS and CPT exposure, only a small proportion of Rad1 was sumoylated, as the modified Rad1 band was barely detectable on blots probed with anti-tag antibody after short exposures. We conclude that a small fraction of Rad1 is mono-sumoylated and this modification is enhanced by UV, MMS and CPT treatment.

The low level of Rad1 sumoylation and its DNA damage inducibility raised the possibility that Rad1 is sumoylated only in specific situations, such as when the protein is actively engaged in DNA repair. We addressed this possibility by examining whether Rad1 sumoylation depended on its recruitment to lesions. Under UV treatment, Rad14 recruits Rad1-Rad10 to DNA lesions by physical interaction (38). We found that rad14Δ greatly reduced the level of Rad1 sumoylation (Figure 1B). We also examined whether Saw1, a Rad1 recruitment factor in SSA (46,47), influences its sumoylation. Since SSA mutants are most sensitive to MMS treatment, we conducted the test in MMS and found that saw1Δ also reduced Rad1 sumoylation (Figure 1C). Residual sumoylation in rad14Δ and saw1Δ cells likely reflects Rad1 recruitment to DNA lesions by the remaining factor. Indeed, rad14Δ saw1Δ double mutants exhibited further reduction of Rad1 sumoylation than either single mutant (Figure 1D).

As Rad14 and Saw1 both recruit and form complexes with Rad1-Rad10, the observed Rad1 sumoylation decrease could be due to either impaired recruitment or impaired complex formation in their absence. To discern if Rad1-Rad10 recruitment per se is important for Rad1 sumoylation, we examined more upstream NER and SSA proteins that are necessary for Rad1-Rad10 recruitment but do not physically interact with the nuclease. In NER, Rad1 recruitment to UV lesions requires the lesion recognition factor Rad4, which does not bind to Rad1-Rad10 (55). We found that rad4Δ, like rad14Δ, reduced Rad1 sumoylation (Figure 1B), suggesting that it is the presence of Rad1 at DNA lesions that is essential for its sumoylation. Similarly, in mutants of the SSA factor Rad52, which is required for formation of 3′ flaps and Rad1 recruitment but does not physically interact with Rad1 (44), we also detected reduced Rad1 sumoylation, as in saw1Δ cells (Figure 1D). Also as in saw1Δ cells, the residual sumoylation in rad52Δ was largely dependent on Rad14, as the rad52Δ rad14Δ double mutant showed a further decrease in Rad1 sumoylation (Figure 1D), consistent with both NER and SSA contributing to the repair of MMS lesions. Taken together, the reduction in Rad1 sumoylation in mutants lacking two NER proteins, Rad4 and Rad14, and two SSA proteins, Saw1 and Rad52, suggests that the initiation of repair and subsequent recruitment of Rad1-Rad10 to DNA lesions are required for Rad1 sumoylation.

The above results suggest that Rad1 sumoylation occurs after it is recruited to DNA lesions. Next we addressed whether DNA lesion recruitment is sufficient to induce Rad1 sumoylation, using the slx4Δ mutant. Slx4 is required for Rad1 nucleolytic cleavage but not its recruitment to 3′ flaps during SSA (46,47,56). In slx4Δ cells, Rad1 is recruited to SSA lesions but cannot initiate cleavage (46,47). We found that slx4Δ reduced Rad1 sumoylation similarly to saw1Δ (Figure 1C). This effect suggests that Rad1 sumoylation requires an Slx4-dependent step after DNA lesion association.

To test if nucleolytic cleavage is required for sumoylation, we used two nuclease dead Rad1 mutants. The D825A mutant (Rad1-nd1) lacks a conserved catalytic residue and is trapped on DNA (46), and the D869A mutant (Rad1-nd2) corresponds to D720A in XPF that shows no nuclease activity in vitro (42,57). Interestingly, both catalytic site mutations increased Rad1 sumoylation, as relative levels of SUMO-Rad1 over unmodified forms were higher in rad1-nd1 and –nd2 cells than in wild-type (Figure 1E and F. Moreover, both mutants showed additional sumoylated Rad1 species (Figure 1E and F. The increased sumoylation still depended on Rad1 recruitment to DNA lesions as it was reduced in mutants of upstream NER and SSA factors, including Rad14, Saw1 and Rad52 (Figure 1G). This result, in conjunction with our other data, suggests that Rad1 sumoylation occurs at lesion sites and requires an Slx4-dependent step but not Rad1-Rad10 nucleolytic activity.

Several aspects of DNA repair are regulated by the DNA damage checkpoint (58,59). However, we have previously shown that the DNA damage-induced sumoylation of numerous proteins is checkpoint-independent, indicating that this sumoylation response is largely separable from the phosphorylation-based checkpoint pathway (7,60). To understand if Rad1 sumoylation follows this rule or is an exception, we analyzed its sumoylation in the absence of the main checkpoint kinase Mec1. As shown in Figure 2A, the level of Rad1 sumoylation was the same in mec1Δ cells as in wild-type after treatment with UV or MMS. Thus, we conclude that damage-induced sumoylation of Rad1, as for many other repair proteins, is not dependent on Mec1.

Yeast, like most organisms, contains only one SUMO E2, but multiple mitotic SUMO E3 ligases. Yeast E3 ligases include the homologous Siz1 and Siz2 proteins, and the more distant Mms21. The Siz enzymes, and sometimes Mms21, show redundancy in sumoylation of various substrates (61,62). To understand whether Rad1 sumoylation is dependent on a particular E3 or can be carried out by redundant E3s, we examined single and double E3 mutants in vivo. We found that none of the single E3 mutants affected Rad1 sumoylation level (Figure 2B). For combinatorial E3 mutants, only the siz1Δ siz2Δ double mutant significantly decreased Rad1 sumoylation, while combining the siz single mutations with mms21–11, which lacks the Mms21 ligase domain, resulted in smaller decreases (Figure 2C and D. The redundancy of the two Siz proteins is consistent with the observation that the siz1Δ siz2Δ double mutant, but neither single mutant, is sensitive to UV and MMS ((7,20) and data not shown). Thus, the three E3s are redundant in sumoylating Rad1 in vivo, with Siz1 and Siz2 playing prominent roles.

To test Siz-dependent sumoylation in vitro, we purified recombinant Rad1-Rad10, as well as SUMO machinery proteins, including SUMO (Smt3), E1 (Aos1/Uba2), E2 (Ubc9) and the Siz E3s, as previously described ((53), Figure 2E and data not shown). As shown in Figure 2F, Siz1 or Siz2 can efficiently sumoylate Rad1 in the presence of ATP. In both cases, a single Rad1 sumoylation band was detected by antibodies against Rad1 or SUMO on western blots. These observations are consistent with our in vivo findings, indicating that Rad1 can be sumoylated by both Siz ligases. We note that in vitro sumoylation requirements are more relaxed than in vivo, likely due to the high concentration of sumoylation machinery components (63–65). Consistent with this, the presence of DNA and/or Saw1 did not appear to affect Rad1 sumoylation (data not shown).

Despite the less stringent requirements for Rad1 sumoylation in vitro, the similar sumoylation patterns and the same Siz E3 dependence in vitro and in vivo suggested that in vitro sumoylated Rad1 can be informative for determining the sumoylation site using mass spectrometry. This experiment identified lysine 32 as the sumoylation site with high confidence (P < 0.05, Supplementary Figure S2). K32 lies within a sumoylation consensus motif (ψKxE/D, where ψ is a hydrophobic amino acid, (66,67)). This residue in the N-terminal region of the protein is far from Rad1's nuclease domain and the helix–hairpin–helix (HhH) domain involved in interaction with Rad10 and DNA (Figure 3A). The Rad1 N-terminal domain has not been studied in detail, though the same region in the human homolog, XPF, contributes to DNA interaction (68). Thus, the Rad1 sumoylation site localizes to a region likely involved in DNA binding.

To verify that K32 is indeed the bona fide sumoylated lysine, we first purified a recombinant K32R mutant of the Rad1 protein and subjected it to in vitro sumoylation. As shown in Figure 3B, wild-type Rad1, but not the K32R mutant, can be sumoylated in the presence of ATP. We then tested if K32R also abolishes sumoylation in vivo, by introducing the K32R mutation at the endogenous RAD1 locus and assaying sumoylation. rad1-K32R eliminated Rad1 sumoylation under both UV and MMS conditions (Figure 3C), confirming that K32 is the Rad1 sumoylation site in vivo.

Next, we examined how lack of Rad1 sumoylation affects cell survival in the presence of genotoxins. We first queried the UV sensitivity of cells expressing the nonsumoylatable rad1-K32R allele using both spotting and plate-out assays. In both assays, blocking Rad1 sumoylation decreased resistance to high doses of UV (Figure 3D and E, with an estimated reduction of ∼20% in the survival of rad1-K32R cells compared to wild-type cells (P < 0.05).

We did not detect increased sensitivity of rad1-K32R to other types of DNA damaging agents, including CPT, MMS and HU (data not shown). As Rad1 acts in a back-up pathway for CPT repair and its sumoylation is induced by CPT treatment ((42); Figure 1A), we examined if rad1-K32R affects resistance to this drug when other proteins involved in this repair are absent. Like rad1Δ but to a less degree, rad1-K32R sensitized cells lacking Tdp1, a phosphodiesterase that processes the majority of CPT lesions ((42,69); Figure 3F). As reported previously, rad1Δ did not sensitize cells lacking Apn2, an endonuclease that functions in base excision and CPT repair, as well as in processing blocked DNA ends with Rad1 (43,70,71); unexpectedly, rad1-K32R showed sensitization (Figure 3F). It is possible that unsumoylated Rad1 prevents downstream or other repair in the absence of Apn2 (see Discussion). In both tdp1Δ and apn2Δ backgrounds, rad1-K32R sensitization was observed only at high CPT doses (Figure 3F).

The manifestation of rad1-K32R's effects at high drug doses suggests that sumoylation influences aspects of Rad1 function that become more critical when the lesion burden is large. To test this idea, we examined how rad1-K32R affects the repair of a single double-strand break via SSA. Rad1-Rad10 cleaves at the ds-ssDNA junction in this process as in NER and CPT repair. If the above idea were correct, one would expect that rad1-K32R is proficient for repairing a single break. We used two well-established SSA assays where a double-strand break is generated either on a plasmid or at a chromosomal locus by the HO endonuclease (46). As the HO cut site is flanked by two complementary sequences, repair of the break is mediated by SSA. We found that rad1-K32R was proficient for SSA repair of both plasmid and chromosomal breaks, based on survival rates after HO induction (Figure 3G-H). Taken together, the phenotype of rad1-K32R supports the notion that Rad1 sumoylation becomes more important when large numbers of DNA lesions need to be repaired.

To gain a molecular understanding of the effect of Rad1 sumoylation and of the reason underlying the rad1-K32R phenotype, we examined Rad1 protein properties in several ways. First, rad1-K32R did not affect Rad1 protein levels under normal and genotoxin conditions (Figure 4A; data not shown). Second, Rad1-K32R behaved similarly to wild-type Rad1 in DNA binding assays (Supplementary Figure S3A and B). Third, rad1-K32R did not affect interaction with Rad10, as wild-type and mutant Rad1 proteins exhibited similar levels of Rad10 association in vitro and in vivo (Figures 2E and 4B). Additionally, the gel filtration profiles of unmodified and sumoylated Rad1 in complex with Rad10 were similar, with sumoylated Rad1 being eluted in the same fractions as the unmodified protein (Figure 4C). Thus, we conclude that the sumoylation status of Rad1 does not affect Rad10 interaction. We note that the gel filtration profiles of Rad1-Rad10 proteins are consistent with a dimer of heterodimers with an apparent molecular mass of 300 kDa (Figure 4C), though further work is necessary to gain a detailed understanding of this potential oligomerization.

Next, we examined Rad1 protein interactions important for its recruitment and/or nuclease activity. First, we tested Rad1 interaction with Rad14 after UV treatment, as NER is critical for UV repair. No difference was seen between Rad1 and Rad1-K32R in their ability to pull down Rad14 protein after UV treatment (Figure 4D). Similar results were obtained when wild-type Rad1 and the K32R mutant were assessed for Saw1 or Slx4 interactions (Figure 4E and F. Note that the Slx4 band shift after MMS treatment was seen as previously reported and is due to phosphorylation of the protein (72). The reason for lower levels of Slx4 pulled down after MMS treatment is unclear, though it could be due to either reduced Slx4-Rad1 interaction or less efficient detection. Nevertheless, the Rad1-K32R mutant behaved similarly to the wild-type protein in its ability to interact with Slx4 in this assay (Figure 4F). Taken together, we conclude that Rad1-K32R is proficient for interactions with Rad14, Saw1 and Slx4, suggesting that sumoylation of Rad1 does not cause major changes in protein–protein interactions.

To test the effect of sumoylation on Rad1 nuclease activity, we compared the activity of equal amounts of wild-type Rad1-Rad10 with the K32R variant complex on Y-form substrates. The fluorescently labeled DNA cleavage product migrated faster than the reactant on gels and the mutant complex exhibited activity similar to its wild-type counterpart (Supplementary Figure S3C). We also optimized our in vitro sumoylation system such that ∼80% of Rad1 was sumoylated (Figure 5A). When we compared equal amounts of sumoylated and unmodified Rad1 in these cleavage assays, again, no difference was detected (Figure 5B). As control reactions that contain only the sumoylation reaction proteins did not show any nuclease activity (Supplementary Figure S4A), the cleaved product was produced by sumoylated Rad1, rather than other components of the sumoylation reaction. We note that the presence of 20% unmodified Rad1 may exclude the detection of small effects of sumoylation. Together, these results show that neither the lack of Rad1 sumoylation nor using a protein prep containing predominantly sumoylated Rad1 affects nuclease activity.

Our in vivo data suggest that Rad1 sumoylation occurs after DNA lesion recruitment and that the modification becomes more important when dealing with large numbers of DNA lesions. Both observations suggest that sumoylation may aid the release of Rad1 from the product after nuclease cleavage. Thus, we examined whether sumoylation of Rad1 favors DNA dissociation.

We tested the interaction of unmodified and sumoylated Rad1-Rad10 with both Y-form DNA and dsDNA, which mimic the substrate and product of the nucleolysis reaction, respectively. We found that unmodified Rad1-Rad10 binds to both dsDNA and Y-forms (Figure 5C and D. Although Rad1-Rad10 is thought to be released from DNA after cleavage simply by intrinsic differences in the enzyme's affinity for different forms of DNA, we found that Rad1-Rad10 interaction with dsDNA is only slightly weaker than that with Y-forms. This result suggests that additional regulation exists to enable efficient dissociation of Rad1-Rad10 from dsDNA. Consistent with this, the excised oligomer in NER is released at a faster rate in vivo than in vitro (73), implying the existence of regulatory mechanisms specifically for achieving shorter turnaround times in NER in vivo.

When sumoylated Rad1 obtained as above was assayed, we detected a reproducible decrease in affinity for both Y-forms and dsDNA, compared with unmodified Rad1 (Figure 5C–E). This is not due to the sumoylation reaction components, as this mixture did not show DNA association (Supplementary Figure S4B and C). In addition, the effect of sumoylated Rad1 is not mimicked by Rad1-K32R-Rad10 in combination with free SUMO (Supplementary Figure S4D and E), suggesting that only the Rad1-conjugated form of SUMO exerts the observed effect on DNA association. We also found that sumoylated Rad1 exhibited less binding to 5′ overhang, but not ssDNA, compared to unmodified Rad1 (Supplementary Figure S4F–H). It is conceivable that the observed sumoylation-mediated reductions in affinity for the product of the cleavage reaction enable efficient disengagement of Rad1-Rad10 from DNA postcleavage.