The SUMO Isopeptidase Ulp2p Is Required to Prevent Recombination-Induced Chromosome Segregation Lethality following DNA Replication Stress

SUMO conjugation is a key regulator of the cellular response to DNA replication stress, acting in part to control recombination at stalled DNA replication forks. Here we examine recombination-related phenotypes in yeast mutants defective for the SUMO de-conjugating/chain-editing enzyme Ulp2p. We find that spontaneous recombination is elevated in ulp2 strains and that recombination DNA repair is essential for ulp2 survival. In contrast to other SUMO pathway mutants, however, the frequency of spontaneous chromosome rearrangements is markedly reduced in ulp2 strains, and some types of rearrangements arising through recombination can apparently not be tolerated. In investigating the basis for this, we find DNA repair foci do not disassemble in ulp2 cells during recovery from the replication fork-blocking drug methyl methanesulfonate (MMS), corresponding with an accumulation of X-shaped recombination intermediates. ulp2 cells satisfy the DNA damage checkpoint during MMS recovery and commit to chromosome segregation with similar kinetics to wild-type cells. However, sister chromatids fail to disjoin, resulting in abortive chromosome segregation and cell lethality. This chromosome segregation defect can be rescued by overproducing the anti-recombinase Srs2p, indicating that recombination plays an underlying causal role in blocking chromatid separation. Overall, our results are consistent with a role for Ulp2p in preventing the formation of DNA lesions that must be repaired through recombination. At the same time, Ulp2p is also required to either suppress or resolve recombination-induced attachments between sister chromatids. These opposing defects may synergize to greatly increase the toxicity of DNA replication stress.


Introduction
As part of the DNA damage response, homologous recombination (HR), particularly template switch recombination through the post-replication DNA repair pathway (PRR), provides an important mechanism for restarting stalled replication forks and filling in un-replicated gaps in DNA (reviewed in [1,2]). These recombination events must be managed carefully, however. DNA strand exchange during HR, followed by re-initiating replication using the nascent sister chromatid as a template, can result in the formation of DNA linkages between daughter chromosomes. Failure to resolve these linkages, called sister chromatid junctions (SCJs), leads to chromosome breakage or aneuploidy, and may contribute to genome instability in many forms of cancer (reviewed in [3]).
A variety of studies implicate SUMO post-translational modification as an important regulator of HR in response to replication stress. Following activation of the SUMO precursor protein, SUMO modification is catalyzed by the E2 conjugating enzyme Ubc9p, which typically acts through one of several E3 ligases to covalently join SUMO moieties to lysine residues on substrate proteins (reviewed in [4]). One SUMO substrate that plays an especially prominent role in controlling HR at replication forks is Pol30p/PCNA, which is modified to recruit different activities to the replisome. During S phase, Ubc9p works through the E3 ligase Siz1p to sumoylate PCNA on K164 and K127 [5]. SUMO modified PCNA recruits the Srs2p helicase [6,7], which suppresses unscheduled HR by disassembling Rad51p nucleoprotein filaments [8][9][10]. Following replication fork stalling at MMSinduced DNA lesions, however, PRR proteins catalyze either mono-or poly-ubiquitinylation of PCNA K164 [5]. These modifications recruit trans-lesion bypass polymerases or induce template switching HR, respectively, providing alternative mechanisms to bypass the lesion and restart replication [5,[11][12][13]. The existence of additional SUMO substrates that control HR is suggested by the observations that mutations affecting both Ubc9p and the E3 ligase Mms21p, which is not required for PCNA sumoylation, confer sensitivity to the replication impeding drugs hydroxyurea (HU) and methyl methansulfonate (MMS) [5,[14][15][16][17][18][19]. Mms21p exists in a complex with two members of the structural maintenance of chromosomes family of proteins, Smc5p and Smc6p, which are also required for HU and MMS-resistance [15,16,[20][21][22]. Notably, in response to MMS, ubc9, mms21, smc5 and smc6 mutants show an accumulation of X-shaped DNA structures that are thought to represent either regressed forks-a possible intermediate in fork restart-or hemi-catenate SCJs [17,19,23]. In this sense, they resemble mutants defective for the Sgs1p/Top3p/Rmi1p complex, which, through concerted heli-case/topoisomerase activities, catalyzes the dissolution of hemicatenates and other DNA linkages [24][25][26][27]. These findings suggest complex roles for sumoylation in either preventing excessive/ improper HR at stalled replication forks and/or mediating the active dissolution of SCJs.
As with the forward SUMO pathway, SUMO de-conjugation is also required to tolerate replication stress. Budding yeast contains two members of the SENP/Ulp family of SUMO isopeptidases, Ulp1p and Ulp2p, which catalyze removal of SUMO [28][29][30]. Ulp1p is an essential enzyme that is preferentially localized to the nuclear pore [28][29][30][31][32], whereas Ulp2p is distributed throughout the nucleus [29,30,33]. Ulp2p (first identified as Smt4p; [34]) is not essential, but ulp2 mutants grow poorly and exhibit a complex assortment of phenotypes, including chromosome segregation and cell division defects. [29,30,[35][36][37][38][39][40][41][42]. Ulp1p and Ulp2p also mediate functions that promote SUMO modification. Ulp1p is required to cleave the SUMO precursor to expose a glycine residue necessary for conjugation [28], while Ulp2p possesses a chain editing activity that prevents formation of aberrantly poly-sumoylated substrates [43]. Poly-sumoylation has the potential to interfere with the functional role of SUMO addition. Moreover, recent evidence has revealed that some poly-sumoylated substrates are targeted for degradation by the SUMO-targeted Slx5p-Slx8p ubiquitin ligase [44][45][46]. Although Ulp1p and Ulp2p play different roles in the SUMO pathway, one trait shared by ulp1 and ulp2 strains is that both exhibit sensitivity to HU and MMS [29,30,43]. Previously, a ulp1-I615N mutant was shown to accumulate single-stranded gaps during DNA replication, to exhibit increased spontaneous recombination, and to become dependent on Srs2p and HR for viability, suggesting a role for Ulp1p in suppressing replication errors that induce HR [47]. Insight into the replication stress sensitivity of ulp2 mutants has come from the important finding that Ulp2p is required for cells to complete mitosis following DNA damage checkpoint arrest [41]. From this, de-sumoylation of Ulp2p substrates may be necessary to restart the chromosome segregation machinery once the checkpoint block to mitosis has been relieved [41,48]. But whether Ulp2p, like other components of the SUMO pathway, is also involved in controlling HR during DNA damage or replication stress has not yet been examined. In this study, we find that, following replication fork stalling by MMS, ulp2 mutants accumulate persistent recombination intermediates that are likely to correspond to SCJs. This mis-regulation is accompanied by a severe, recombination-dependent, block to chromosome segregation, revealing a critical role for Ulp2p in allowing sister chromatids to disjoin following HR DNA repair.

Recombination is elevated and essential in ulp2 mutants
We initially set out to determine if ulp2 mutants displayed a similar dependency on recombination as ulp1-I615N strains [47]. A ulp2 deletion mutant (ulp2D) was mated to rad52D, rad51D and rad6D strains. Rad51p and Rad52p are required for most forms of HR [2], while Rad6p controls trans-lesion synthesis and template switching PRR [1]. ulp2D rad52D, ulp2D rad51D and ulp2D rad6D double mutants were either not obtained or were obtained at lower than expected frequencies from these crosses (Table 1, Table 2,  Table 3). For rad52D, we examined this apparent synthetic lethality further by isolating ulp2D rad52D segregants harboring a wild type (WT) copy of RAD52 on a URA3 minichromosome (pRAD52). ulp2D rad52D/pRAD52 mutants grew weakly, if at all, on media containing 5-FOA, a drug that only allows growth if cells are capable of losing pRAD52 ( Figure 1A). Thus, Rad52p is essential for proliferation of ulp2D cells.
The essential role of Rad52p prompted us to examine whether HR was elevated in the absence of Ulp2p. Yeast cells exhibit a uniform nuclear distribution of fluorescent Rad52p-GFP in the absence of DNA damage ( Figure 1B, [49]), but Rad52p-GFP rapidly assembles into intra-nuclear foci during HR DNA repair [49]. We found that an average of 17% of ulp2D cells in midlogarithmic phase cultures displayed Rad52p-GFP foci, a significant increase (p = 0.0074) compared to less than 1% in WT cells. ( Figure 1B). As a second assay, we utilized a reporter in which recombination events between direct repeats on chromosome XV can be selected because they restore an intact HIS3 gene [50]. ulp2D cells exhibited a 4.7-fold increase in the median frequency of this form of recombination ( Figure 1C; p = 0.044), indicating spontaneous HR at this genomic locus is significantly increased in ulp2D mutants.
Ulp2 mutants display a reduced frequency of chromosome rearrangements DNA replication errors are potent inducers of HR and can initiate chromosome rearrangements [51,52]. Based on this, we used a yeast artificial chromosome (YAC) assay to examine the frequency of spontaneous gross chromosomal rearrangements (GCRs) in ulp2D cells ( [53]; Figure 2A). For comparison, we also measured GCR frequencies in ulp1-333, smt3-331 and ubc9-1 SUMO pathway mutants (SMT3 encodes the single SUMO isoform in budding yeast). Using the YAC system, we obtained median GCR frequencies of 252610 27 for WT cells, 5490610 27 for smt3-331 cells, 6109610 27 for ubc9-1 cells, and 2617610 27 for ulp1-333 cells ( Figure 2B), representing 22-, 24-, and 10-fold increases, respectively, compared to WT controls. In contrast, and counter to initial expectations, it proved difficult to recover spontaneous GCRs in ulp2D mutants, with a median GCR frequency of 56610 27 ( Figure 2C). This represents a significant (p = 0.025) 4.5-fold decrease compared to WT.

Author Summary
DNA damage, arising from environmental stress or errors in DNA metabolism, can interfere with DNA replication. Cells respond by using homologous recombination to bypass the damage, resulting in DNA strand linkages between the replicated chromosomes. It is crucial to undo these linkages so chromosomes can segregate properly. Previously, a regulatory mechanism known as SUMO modification was shown to be important in controlling recombination following replication interference by the DNA damaging agent MMS. We show that mutations in a yeast enzyme called Ulp2p, which reverses SUMO modification, increase recombination and impose a requirement for recombination to maintain survival. MMS-treated ulp2 mutants also accumulate recombination intermediates and fail to separate their chromosomes, leading to a permanent block to cell division. Further analysis suggests this block may not simply be due to a failure to resolve recombination intermediates, but may reflect a role for Ulp2p in undoing additional chromosome attachments that accompany recombination. In sum, our data indicate that cells defective for Ulp2p develop a love/hate relationship with recombination, requiring recombination for viability while failing to resolve chromosome attachments induced by recombination repair. Identification of Ulp2p substrates that ensure chromosome separation following recombination will shed light on how SUMO modification maintains genome stability.
To further monitor chromosome rearrangements we examined two circular dicentric minichromosomes. In one (p2XCEN direct ), two copies of a CEN sequence were oriented in a direct repeat configuration. In the other (p2XCEN invert ), the same CEN duplication was oriented as inverted repeats. Previous studies have shown that both direct and inverted repeat dicentrics can be efficiently transformed into yeast, and are initially retained through a combination of co-orientation of the two CENs on the spindle and non-disjunction following dicentric bridging [54,55]. During outgrowth, however, rearranged minichromosomes that have deleted one of the CENs accumulate. For direct CEN repeats these deletions tend to arise through loop out events, whereas inverted CEN repeats are resolved through more complex rearrangements. Consistent with this characterization, in WT transformants p2XCEN direct and p2XCEN invert exhibited similar mitotic stabilities to p1XCEN controls ( Figure 2D). Analysis of minichromosomes rescued from these cells revealed precise CEN1 excision for p2XCEN direct and a diversity of plasmid species for p2XCEN invert (not shown). In ulp2D mutants, p1XCEN was only retained in ,30% of the cells; this result is in keeping with previous studies showing reduced minichromosome stability in the absence of Ulp2p [29]. p2XCEN direct demonstrated a similar stability to p1XCEN ( Figure 2D), and underwent the same precise CEN deletions observed in WT (not shown). In contrast, p2XCEN invert proved extremely unstable, with less than 1% of ulp2D cells maintaining the mini-chromosome. These results suggest that some chromosome re-arrangements either fail to occur or cannot be tolerated in ulp2 mutants.

MMS-induced HR intermediates accumulate in ulp2D mutants
In order to more directly examine the consequences of HR in ulp2D mutants, we used MMS to induce recombination. As an initial experiment, we examined chromosome integrity following exposure to MMS by pulse-field gel electrophoresis. WT and ulp2D cells were arrested in G1, released into media containing 0.01% MMS for 2 hr, and then allowed to recover in MMS-free media. Following MMS treatment a lower molecular weight DNA smear was observed in both WT and ulp2D strains ( Figure 3A), reflecting MMS-induced chromosome breakage [17]. For both strains, a one hr recovery largely restored the normal chromosome banding pattern. This suggests Ulp2p is not obviously required for healing MMS-induced DNA breaks.
We next examined processing of MMS-induced DNA lesions. In the experiment shown in Figure 3B, WT cells and ulp2D mutants expressing RAD52-GFP were treated with 0.01% MMS and allowed to recover. After a 2 hr recovery, ,30% of WT cells accumulated Rad52p-GFP foci ( Figure 3B). By 6 hr, however, the percentage of cells with Rad52p-GPF foci had substantially declined and many cells were proceeding with the next round of cell division. In contrast, ulp2D mutants showed a much stronger accumulation of Rad52p-GFP foci, reaching a maximum of ,60% ( Figure 3B), and these foci tended to persist for the duration of the recovery period. We also examined Rad52p-GFP foci in ulp2D cells treated with 200 mM HU. HU does not normally induce Rad52p foci because the integrity of the replisome is maintained by the S phase checkpoint ( Figure 3B, [56]). HU treated ulp2D cells, however, exhibited a strong induction of Rad52p-GFP foci.
In response to MMS, proper regulation of HR is required to prevent X-shaped recombination intermediates from accumulating in the vicinity of origins of replication [17,19,23]. On twodimensional gels these structures migrate as a ''X-spike'' that is distinct from replication forks and bubbles [57,58]. To determine whether ulp2D mutants accumulated this type of HR intermediate, ulp2D cells, along with WT and sgs1D controls, were released from a G2/M nocodazole block and treated with 0.033% MMS for 3 hr as previously described [17]. Genomic DNAs were fractionated on two-dimensional gels, and probed with a DNA fragment corresponding to ARS305. A prominent X-spike signal was observed in sgs1D and ulp2D samples ( Figure 3D). Thus, Ulp2p deconjugating and/or chain editing activities are required to prevent accumulation of MMS-induced HR intermediates.

Interactions between Ulp2p and Sgs1p
Based on current evidence, Sgs1p is one SUMO target that could be connected to Ulp2p's role in HR. In particular, a recent  Table 3. Genetic interactions between ulp2D and rad6D mutants.
Cross between JBY238 and JBY285; ulp2D 6 rad6D; 22  Ulp2p Counteracts Recombination DNA Linkages study has shown that a single prominent SUMO species of Sgs1p accumulates after MMS exposure, and K621 has been identified as the acceptor lysine that is responsible for this modification [59]. We were able to confirm that treatment with 0.3% MMS resulted in a substantial fraction of Sgs1p-myc shifting into a reduced mobility species ( Figure 4A and Figure S1), and that a decreased amount of this form was observed following treatment with a lower MMS concentration (0.033%; Figure 4B, 4C). The appearance of this form was abolished in ubc9-1 strains ( Figure 4B) and a sgs1-K621R mutant ( Figure S1), indicating it is likely to correspond to the previously reported K621 conjugate. In ulp2D strains, however, a marked increase in this putative Sgs1p SUMO species was observed ( Figure 4B, 4C), which persisted for at least 3 hr after removal of MMS ( Figure 4C). In sum, these results suggest that sumoylation of Sgs1p is likely to be regulated by Ulp2p. If failure to properly control Sgs1p sumoylation was responsible for ulp2D HR defects, SUMO-resistant Sgs1p might ameliorate these phenotypes. We therefore examined whether a plasmid-born copy of the sgs1-K621R allele could prevent Rad52p foci accumulation. Following a two hr treatment with 0.010% MMS, however, no significant reduction in ulp2D sgs1-K621R cells displaying Rad52p-GFP foci was observed ( Figure 4D). Previous studies have shown that a form of Smt3 (smt3-3KR) that cannot form polymeric SUMO chains can rescue the HU and MMS sensitivity of ulp2 mutants [43], leading us to test whether smt3-3KR could prevent Rad52p foci accumulation. This proved to be the case, as smt3-3KR ulp2D double mutants did in fact show a substantial reduction in the accumulation of both spontaneous and MMS-induced Rad52p foci ( Figure 4F). Thus, proper SUMO chain editing through Ulp2p is likely to be important in controlling HR.
ulp2D mutants fail in chromosome segregation after exposure to MMS In our experiments, it was apparent that ulp2D cells frequently remained blocked in the cell cycle during recovery from MMS, similar to previous results examining ulp2 recovery following HU treatment and in response to an irreparable DNA double strand break [41]. We took four experimental approaches to investigate the basis for the apparent MMS recovery defect of ulp2D cells. First, phospho-activation of the Rad53p checkpoint kinase during the DNA damage response results in a series of slower migrating gel mobility variants [60], and collapse of these forms provides a means to assess silencing of the checkpoint. In WT cells, Rad53p phospho-variants almost completely disappeared during a 2-4 hr recovery after treatment with 0.01% MMS ( Figure 5A). A similar pattern was observed in ulp2D strains, although the accumulation and disappearance of shifted Rad53p appeared to be slightly delayed.
Second, we examined degradation of Pds1p/securin. Pds1p is a downstream target of the DNA damage checkpoint that is stabilized to block cohesin proteolysis and anaphase entry [61,62]. The kinetics of Pds1p degradation therefore provides a read-out of commitment to anaphase. In these experiments, we used the cdc14-1 allele to block Pds1p re-synthesis once cells recovered from the checkpoint. cdc14-1 PDS1-myc and cdc14-1 ulp2D PDS1-myc cells were treated with 0.001%, 0.005% and 0.01% MMS for 2 hr, allowed to recover at a cdc14-1 non- Figure 1. Spontaneous HR in ulp2D mutants. (A) rad52D and ulpD rad52D segregants harboring pRAD52-URA3 were derived from a cross between rad52D/pRAD52-URA3 (MLY031; pRAD52 is pVL191) and ulp2D (JBY238) haploid strains, cultured in the absence of selection for pRAD52-URA3, and a 10-fold dilution series was stamped onto Ura 2 /SC (growth requires pRAD52-URA3) and 5-FOA media (growth requires loss of pRAD52-URA3). Plates were cultured at 30uC and photographed after 3 days. (B) Aliquots of mid-log phase cultures of WT (MLY061) and ulp2D (MLY060) RAD52-GFP strains were fixed and scored to determine the percentage of unbudded and budded cells exhibiting distinct Rad52p-GFP foci. Micrographs (superimposed bright field and fluorescent images) depict dispersed nuclear Rad52p-GFP signal in WT cells and Rad52p-GFP foci in ulp2D mutants. Scale bar, 4 mm. Graph displays the average percentage of budded and un-budded cells with and without Rad52p foci determined from three separate experiments. Error bars, 6 one standard deviation. A total of 1,018 WT and 523 ulp2D cells were scored; 4 WT cells displayed distinct foci (0.4%). The p value (Student's t-test) for comparing total number of cells with foci from the WT and ulp2D datasets is 0.0074. (C) WT (MLY066; 8 cultures) and ulp2D (MLY067; 10 cultures) strains were constructed in which a plasmid insertion at the HIS3 locus is flanked by 414 bp direct repeats. Recombination events between the repeats can be selected because they restore HIS3 function. Box plot graphs display the median number of His + clones per 10 5 viable cells, 25 th and 75 th percentiles, and range of data (p = 0.044; Student's t-test). doi:10.1371/journal.pgen.1001355.g001 permissive temperature, and Pds1p-myc abundance was monitored over a 24 hr period. In cdc14-1 cells, Pds1p-myc degradation proceeded in a dose-dependent manner until 10 hr post-treatment ( Figure 5B, 5C). At this point, Pds1p started to increase in the 0.001% and 0.005% MMS cultures, probably reflecting leakage through the cdc14-1 arrest. These degradation kinetics were virtually indistinguishable in cdc14-1 ulp2D cells, although resynthesis of Pds1p was not observed ( Figure 5B, 5C). These results suggest that MMS treated ulp2D cells can terminate checkpoint signaling and commit to anaphase.
Third, we used micro-colony analysis to determine whether getting rid of the checkpoint relieved the restraint on cell division. Cells from MMS treated and untreated cultures were positioned on agar plates, and the appearance of cell bodies was examined over time. A budded yeast cell arrested at the DNA damage checkpoint consists of two cell bodies. If this cell completes mitosis and one of progeny cells sends forth a bud, the microcolony now contains three cell bodies, and the number of cell bodies increases exponentially with continued division. We found that an average of 68% of WT cells were able to form microcolonies containing $ 16 cell bodies within a 3 day period after transient exposure to MMS, indicating the majority recovered efficiently ( Figure 6). In comparison, even in the absence of MMS, 20% of ulp2D cells remained blocked at the 2-3 cell body stage. This lethality was strongly exacerbated by MMS treatment, with 64% of ulp2D cells failing to proliferate beyond 2-3 cell bodies. Inactivating the DNA damage checkpoint in rad9D ulp2D mutants, or both the DNA damage and S phase checkpoints in mec1D ulp2D mutants, did not relieve the ulp2D block to cell division ( Figure 6). ulp2 cells fail to maintain chromatid cohesion at centromeric regions during DNA damage checkpoint arrest [36,42], which could potentially activate the spindle assembly checkpoint (SAC). We therefore tested whether abolishing the SAC could restore ulp2D division. However, ,60% of MMS treated ulp2D mad2D mutants still remained blocked with 2-3 cell bodies ( Figure S2). We further generated a ulp2D rad9D mad2D triple mutant to abolish both DNA damage and SAC checkpoint responses. This triple mutant grew extremely poorly, and, following exposure to MMS, ,90% of the cells failed to recover ( Figure 6). Thus, MMS treated ulp2D mutants experience a terminal block to cell division even in the absence of pre-anaphase checkpoint controls.
Fourth, we examined mitotic progression in ulp2D cells by cytology and flow cytometry. Following a 2 hr treatment with 0.01% MMS, WT cells arrested at the DNA damage checkpoint typically showed short pre-anaphase spindles and an undivided mass of chromatin ( Figure 7A, 7B). Completion of mitosis during recovery was characterized by normal spindle extension and chromosome transmission. As monitored by DAPI staining and a Lac operator-GFP chromosome tag (TRP1-GFP), ,70% of cdc14-1 cells underwent chromosome separation and segregation during recovery ( Figure 8A, 8B), and FACS analysis indicated that many cells proceeded with additional rounds of cell division ( Figure S3). In contrast, many MMS treated ulp2D cells showed partial, incomplete spindle extension during recovery, accompanied by an apparent block to nuclear division ( Figure 7A, 7B). In some cells it was possible to visualize chromatin fibers that appeared to be pulled away from an undivided mass of chromatin ( Figure 7B iii; arrows). In others, chromosome separation appeared more complete, but chromatin was stretched to varying degrees along the spindle (Figure 7B iv). DAPI staining indicated less than 20% of cdc14-1 ulp2D cells successfully completed chromosome segregation ( Figure 8A). ,30% of cdc14-1 ulp2D cells underwent TRP1-GFP separation during recovery, but the separated foci largely failed to segregate ( Figure 8B). FACS analysis suggested that MMS treated ulp2D cells potentially tried to proceed with a second round of DNA replication following this block chromosome segregation, although the FACS profiles were quite heterogeneous and did not clearly resolve into a peak of cells with a 4N content of DNA ( Figure S3).
Since sgs1D and ulp2D mutants both accumulate HR intermediates that might be expected to link sister chromatids ( Figure 3C), we additionally examined chromosome segregation during MMS recovery in cdc14-1 sgs1D cells. Compared to the ulp2D defect, the fraction of MMS treated cdc14-1 sgs1D cells that could segregate their chromosomes to an extent necessary to form two distinct nuclear masses was only slightly reduced compared to cdc14-1 controls ( Figure 8A; see Figure S4 for a more complete description). Taken as a whole, these results allow us to conclude that, although they commit to anaphase, ulp2D mutants are unable to separate their chromosomes efficiently following MMS treatment. Furthermore, this non-disjunction defect appears more severe than that observed in a sgs1D strain.
Blocking HR restores chromosome segregation in MMStreated ulp2 mutants If defective HR in MMS treated ulp2D cells is causally linked to the chromosome separation defect that we observed in our experiments, blocking recombination should restore chromosome segregation. Given that HR is essential in ulp2D mutants ( Figure 1) our approach to test this was to overproduce (OP) the Srs2p helicase. In addition to antagonizing nucleoprotein filament assembly [8][9][10], Srs2p also appears to exert anti-recobinogenic activity by unwinding D-loop intermediates [63,64]. Srs2p OP should therefore be an effective way to short circuit early stages of HR. cdc14-1, cdc14-1 rad9D, cdc14-1 ulp2D and cdc14-1 rad9D ulp2D strains were transformed with a vector control or a high copy plasmid in which SRS2 was expressed under control of its endogenous promoter (pSRS2). The transformants were then treated with 0.01% MMS for 2 hr and allowed to recover at a cdc14-1 non-permissive temperature. Compared to vector controls, cdc14-1/pSRS2 cells remained blocked in a pre-anaphase configuration for the duration of the recovery period ( Figure 9A). This delay was abolished in cdc14-1 rad9D/pSRS2 transformants, suggesting Srs2p OP was able to prolong DNA damage checkpoint arrest. In the absence of Ulp2p, however, inactivating the checkpoint in the cdc14-1 rad9D ulp2D/vector strain was insufficient to allow cells to proceed with chromosome segregation (Figure 9A, 9B). Significantly, Srs2p OP demonstrated a remarkable ability to allow ulp2D strains to escape this mitotic block, with ,50% of cdc14-1 rad9D ulp2D/pSRS2 cells now segregating their chromosomes in a seemingly normal anaphase ( Figure 9A, 9B). Thus, Srs2p OP substantially relieves the block to chromosome separation in MMS treated ulp2D cells.

HR and genome stability in ulp2 mutants
One principal finding of this study is that, even in the absence of exogenous DNA replication stress, spontaneous recombination is increased in ulp2D cells. This conclusion is based on two observations. First, by genetic criteria, spontaneous recombination at a genomic location on chromosome XV is elevated in ulp2D strains. Second, ulp2D mutants also display an increase in the frequency of spontaneous Rad52p DNA repair foci. A similar increase in Rad52p foci has been observed in a number of other SUMO pathway mutants, and has been shown to be largely attributable to a requirement for sumoylation in preventing inappropriate recombination events involving the 2 mm circle, an endogenous plasmid found in most S. cerevisiae strains [65]. Since we have not directly examined the effect of the 2 mm circle on recombination in ulp2 mutants, destabilization of this extrachromosomal element may well contribute to the ulp2D increase in Rad52p foci. However, as the 2 mm circle is not required for S. cerevisiae growth, our finding that HR DNA repair becomes essential in ulp2D strains strongly suggests that Ulp2p acts to suppress the formation of genomic DNA lesions that must be repaired through recombination. Previous analyses of the SUMO pathway support this possibility. For example, SUMO conjugation-defective ubc9-1 mutants exhibit synthetic growth defects in the absence of HR and, at the non-permissive temperature, accumulate DNA structures that activate Rad53p [17]. Furthermore, as described in the Introduction, ulp1-I615N mutants also show increased HR and require HR for viability; in this case, the requirement for HR was shown to correspond with single-stranded DNA gaps arising during S phase [47]. It is striking that perturbations to Ulp1p and Ulp2p, which appear to target largely distinct sets of SUMO substrates [29], impose such seemingly similar dependencies on HR. Another observation that lends credence to the idea that Ulp2p suppresses recombinogenic DNA lesions is that ulp2D mutants greatly induce the formation of Rad52p foci following HU treatment. Such foci are not observed  [59]. (D) sgs1D (MLY200) and sgs1D ulp2D (MLY144) strains harboring RAD52-GFP were transformed with a vector control (pRS426; [84]), a plasmid expressing SGS1 (pSGS1; YCplac33/SGS1; [85]) or a plasmid expressing a sgs1-K621R mutant allele (psgs1-K621R; pJBN269). Three transformants for each strain/plasmid combination were treated with 0.01% MMS for 2 hr and allowed to recover in MMS-free media. After a 6 hr recovery, aliquots were fixed and scored to determine the percentage of cells with Rad52p-GFP foci. Graphs display the average of the three cultures, 6 one standard deviation. (E) Aliquots of logarithmic phase cultures of three WT (MLY061, JBY1815, JBY1816), smt3-3KR (JBY1817-1819), ulp2D (JBY1820-1822), and ulp2D smt3-KR (JBY1823-1825) strains harboring RAD52-GFP were fixed and evaluated as in (D) to determine the percentage of cells displaying spontaneous Rad52-GFP foci (spontaneous foci graph). In parallel, the same set of cells was also treated with 0.01% MMS, allowed to recover for 6 hr, and scored for persistence of Rad52-GFP foci (MMS-induced foci graph). doi:10.1371/journal.pgen.1001355.g004 in HU treated WT cells [56], consistent with an underlying replication problem in ulp2D mutants that is exacerbated by slowed fork progression.
In analyzing genome stability in ulp2D strains, we observed two interesting differences compared to other SUMO pathway mutants. First, whereas our data indicate that Rad6p-dependent PRR is essential in ulp2 mutants, mis-regulation of SUMO conjugation in ulp1-I165N rad18 [47], ubc9-1 rad18 [19], siz1 rad18 [11], pol30-K164R rad18 and pol30-K164R rad6 [5] mutants can actually compensate for defective PRR. One scenario that might account for this difference is if poly-sumoylation of a Ulp2p substrate(s) caused a distinct perturbation to DNA replication that was repaired through PRR-mediated HR. In keeping with this interpretation, we find that blocking poly-SUMO chain formation reduces the accumulation of both spontaneous and MMS-induced HR foci in ulp2D mutants.
A second apparent difference concerns the formation of GCRs. In contrast to smt3-331, ubc9-1 and ulp1-333 strains, where spontaneous GCRs are increased, ulp2D mutants show reduced GCRs. Formally, Ulp2p could promote GCR formation by stimulating error prone DNA repair. There is precedence for this, as a previous study found that, in the absence of template switch PRR, Siz1p-mediated sumoylation of PCNA was required to form certain types of GCRs [66]. Alternatively, Ulp2p could be required for cells that would give rise to GCRs to recover and propagate efficiently. Our observations with dicentric minichromosomes are consistent with the idea that repair events leading to some GCRs may not be tolerated in ulp2D strains. We were able to recover re-arranged dicentrics from ulp2D mutants when duplicated CEN sequences were present in a direct repeat configuration. Such deletions can occur through single-strand annealing, an intra-chromosomal form of recombination [67]. In contrast, CEN deletion GCRs were not recovered when the two CENs were oriented as inverted repeats. Recent studies have shown that faulty template switch PRR is frequently involved in initiating deletions between inverted repeats [68,69]. As discussed below, one possibility is that such recombination events are accompanied by formation of SCJs or other types of chromatid attachments that fail to be resolved in ulp2 cells.

Ulp2p prevents accumulation of HR intermediates
Our results led us to suspect that HR DNA repair, while required for viability, might at the same time be toxic to ulp2 cells, prompting us to examine processing of MMS-induced recombination events. From this analysis, one conclusion is that, similar to Ubc9p, Mms21p, Smc5p/Smc6p, and Sgs1p/Top3p [17,19], Ulp2p is required to prevent X-shaped DNA structures from accumulating at sites of replication fork stalling/collapse. We also find that, whereas Rad52p foci disappear during MMS recovery in WT cells, the incidence of these foci remains elevated in ulp2D strains, suggesting a possible role for Ulp2p in terminating recombination. Determining the molecular basis for how Ulp2p prevents accumulation of HR intermediates, and whether this Based on current information, Ulp2p could be connected to HR through a number of different SUMO substrates. First, Mms21pmediated sumoylation of unknown substrates, probably in conjunction with Smc5p/Smc6p [22,70], has been proposed to prevent excessive template switch recombination through PRR [19]. Alternatively, more recent evidence suggests Smc5p/Smc6p may instead act downstream of PRR to facilitate the dissolution of HR intermediates [71]. Second, Sgs1p is sumoylated under conditions when it is active in SCJ dissolution [17,59], although apparently through an Mms21-independent pathway [17]. Third, Ubc9p/Siz1p-controlled sumoylation of PCNA and recruitment of Srs2p may suppress PRR-independent recombination at replication forks [6,7,19]. Fourth, Srs2p has also been shown to be sumoylated, with poly-sumoylation being proposed to trigger Srs2p degradation through the Slx5p/Slx8p pathway [72]. Fifth, a fraction of Rad52p [73][74][75], and other HR proteins [76], are sumoylated in response to MMS, which may be involved in finetuning processing of broken DNA. Finally, a growing number of protein-protein interactions within HR foci have been found to be controlled by sumoylation (reviewed in [77]). As part of completion of repair, Ulp2p may catalyze the disassembly of these networks.
As a first step in placing Ulp2p in these pathways, we tested whether mis-regulation of Sgs1p sumoylation was connected to ulp2D HR defects. Overproduction of Ulp2p was recently shown to block Sgs1p sumoylation on K621 following MMS treatment [59], and, as we report here, MMS-induced sumoylation of Sgs1p is elevated in the absence of Ulp2p. It is therefore likely that Ulp2p acts as the SUMO deconjugating enzyme for Sgs1p. Despite this, short-circuiting Sgs1p sumoylation using the sgs1-K621R mutation   [86]), ulp2D mec1D GAP-RNR3 (MLY108), mad2D (JBY554), and ulp2D rad9D mad2D (MLY160) strains were split, and half the cultures was treated with 0.01% MMS for 2 hr. For both treated (+MMS) and untreated (2MMS) cultures, 50 large-budded cells (equivalent to 2 cell bodies) were micro-manipulated onto a grid pattern on a YPD plate, and the number of cell bodies in the resulting micro-colonies were monitored over the next 3 days. The graphs display the average of three separate experiments for each strain, 6 one standard deviation. Numbers on the +MMS graphs indicate the average percent of cells that remain blocked with 2-3 cell bodies. For the WT and ulp2D strains, the average percentage of cells capable of recovering to form microcolonies of $ 16 cell bodies is also indicated. doi:10.1371/journal.pgen.1001355.g006 did not reduce Rad52p foci accumulation in ulp2D cells, indicating mis-regulation of other Ulp2p substrates is likely to be involved in modulating HR.

HR and the ulp2 recovery defect
The failure of ulp2 mutants to resume cell division following DNA damage is one of the most intriguing aspects of the ulp2 phenotype. The first study to document this phenomenon showed that, following adaptation to a persistent DNA break, only a fraction of ulp2 cells were able to proceed with nuclear division, frequently accompanied by abnormally extended or broken mitotic spindles [41]. Inactivating the DNA damage checkpoint rescued this defect, suggesting a critical role for Ulp2p in reinitiating chromosome segregation following completion of the checkpoint response [41,48].
While our results are largely in accord with this study, we observed a potentially informative difference in the role of the checkpoint in manifesting the ulp2D recovery defect. During MMS recovery, ulp2D cells dephosphorylated Rad53p and degraded Pds1p on schedule, suggesting they were competent to silence the checkpoint and initiate anaphase. Despite this, sister chromatids failed to disjoin, resulting in a dramatic failure in chromosome segregation. OP of Srs2p, which antagonizes HR [8][9][10]63,64], was able to largely restore chromosome segregation. In addition to modulating nucleo-protein filament assembly, Srs2p has also been shown to be required for full activation of the DNA damage checkpoint and for recovery from DNA damage checkpoint arrest [78,79]. In our experiments, we observed that Srs2p OP greatly extended DNA damage checkpoint arrest in MMS treated WT cells. Based on the above considerations, this extended arrest could presumably reflect either mis-regulation of the checkpoint pathway, or, by interfering with HR DNA repair, elevated Srs2p could simply prolong normal checkpoint signaling. While the effects of Srs2 OP on checkpoint signaling and HR may be multifaceted, the key point we wish to emphasize here is that abolishing the DNA damage checkpoint (or the SAC) did not allow ulp2D cells to divide more times during recovery from MMS treatment. Furthermore, preventing DNA damage checkpoint arrest in MMS treated ulp2D rad9D cells was insufficient to relieve the block to chromosome separation; OP of Srs2 was also necessary. In sum, these findings strongly suggest that, following replication fork stalling by MMS, downstream events initiated through HR, rather than checkpoint arrest per se, appear to play a causal role in interfering with chromosome segregation.
A key question concerns how HR could have this effect. Perhaps the simplest idea is that unresolved SCJs block chromatid disjunction. Whether this is a sufficient explanation, however, is unclear. First, in the experiments examining ulp2 adaptation to a persistent, endonuclease-targeted DNA break, both chromatids would be expected to be cut, preventing HR strand exchange [41]. Thus, the only way in which DNA linkages could form between chromosomes in these cells would be if extensive resection during prolonged checkpoint arrest triggered illegitimate recombination events. Second, we show that MMS treated sgs1D mutants, which are clearly defective in the dissolution of SCJs [17,25,27], do not show as severe a block to chromosome separation as Ulp2pdeficient cells. This is consistent with a recent study that showed, from among a collection of helicase-, nuclease-, and topoisomerase-deficient mutants, only smc5, smc6 and mms21 strains showed chromosome segregation defects after a pulse of MMS delivered in G1 [71]. This suggests a role for Mms21p-mediated sumoylation and the Smc5p/Smc6p complex in resolving SCJs or other types of chromatid linkages outside the Sgs1p/Top3p pathway [71]. Along these lines, it is notable that Ulp2p has been implicated in multiple facets of chromatid separation, including controlling sumoylation of cohesin regulatory proteins [37,42], condensin [35,38], and DNA topoisomerase II [36,40]. Speculatively, following induction of HR, there may be an increased requirement for Ulp2p in the vicinity of DNA lesions, not only to prevent accumulation of HR intermediates, but also to complete replication, to disentangle DNA or to release protein-based forms of cohesion. Given the dramatic way in which the absence of Ulp2p potentiates the ability of replication toxins to block cell proliferation, a further understanding of the ulp2 recovery defect could lead to insights that are relevant to cancer treatment.

Materials and Methods
Yeast strains and culture All S. cerevisiae strains used in this study were derived from the W303-related CRY1 strain and are listed in Table S1. A description of how different genetic elements were introduced into the CRY1 background can be found in Text S1. For all experiments, cells were cultured in standard formulations of yeast Recombination frequency pLAY202 ( [50]; provided by A. Bailis, City of Hope National Medical Center, Duarte, CA) was linearized with BstXI and targeted to the HIS3 locus, placing a URA3 marker between partially duplicated HIS3 sequences. pLAY202 integrants were propagated in Ura 2 /SC media, and, following overnight incubation, cell density was quantified using a hemacytometer. Viable cell counts were determined by plating a defined number of cells onto YPD and counting the resulting colonies. Recombination events were selected by plating a larger number of cells onto His 2 /SC media, and replica plating colonies that arose onto 5-FOA. Colonies that reverted to a His + , Ura 2 phenotype were scored as recombinants.

Minichromosome loss
To monitor the mitotic stability of dicentric minichromosomes, p2XCEN direct (pJBN152; a YRp14-derived minichromosome containing two copies of a 1.7 kb CEN1 DNA fragment in a direct repeat configuration, see Text S1) and p2XCEN invert (pJBN151; similar to pJBN152 but with the CEN1 duplication oriented as an inverted repeat) were transformed into WT and ulp2D strains and compared to p1XCEN (YRp14/CEN1) controls. Transformants were inoculated into parallel YPD and Ura 2 /SC cultures and incubated at 30uC. After ,15 hr of outgrowth, appropriate dilutions were plated onto YPD and Ura 2 /SC media. Mitotic stability was calculated by dividing the number of Ura + colonies by the total number of colonies obtained on YPD.

Microscopy and flow cytometry
Cultures for microscopy were supplemented with 50 mg/ml adenine to quench auto-fluorescence. To visualize Rad52p-GFP and TRP1-GFP, cells were fixed in 1% formaldehyde for 1.5 min and washed into PBS. DAPI staining was performed using Vecta-Shield (Vector Laboratories) containing 10 mg/ml DAPI. TUB1-GFP and HHF2-YFP strains were visualized as live mounts. HHF2-YFP is typically propagated as a heterogyzous diploid (HHF2-YFP-HIS3/+) to minimize selective pressure for rearranged variants that lose the fluorescent marker. However, in order to compare the response of HHF2-YFP strains to MMS concentrations similar to those used in our other recovery experiments, we chose to examine HHF2-YFP haploid segregants that were generated on an experiment-by-experiment basis. This proved to allow propagation of haploid strains with robust Hhf2-YFP fluorescence. In all cases, cells were visualized on Nikon E-800 or Nikon Eclipse 80i microscopes equipped with florescence optics and 100X (1.4 NA) or 60X (1.4 NA) objectives. Rad52p-GFP foci were typically scored using a number 4 neutral density filter to minimize photobleaching. A Zeiss Axioskop 40 microscope equipped with a 25 mm diameter optical fiber dissection needle was used to micromanipulate yeast cells for microcolony analysis. FACS analysis was performed by staining ethanol fixed yeast cells with propidium iodide as previously described [82].
Pulse field gel electrophoresis 10 ml aliquots of OD 600 0.8 cultures were harvested by centrifugation and concentrated into 400 ml cell suspension buffer (10 mM Tris, 20 mM NaCl, 50 mM EDTA, pH 7.2). The cell suspension was warmed to 55uC and mixed with 400 ml 2% low melting temperature agarose (SeaKem) dissolved in TBE gel electrophoresis buffer (kept molten at 55uC) containing lyticase (Sigma L4025; final concentration 1 mg/ml). The cell suspension was transferred into molds and allowed to solidify to form plugs (4uC, 15 min). Plugs were pushed out into 50 ml conical tubes and incubated with 5 ml 1 mg/ml lyticase dissolved in 10 mM Tris, 50 mM EDTA, pH 7.2 for one hr at 37uC, followed by treatment 1 mg/ml Proteinase K (Sigma) dissolved in 100 mM EDTA, 0.2% Na Deoxycholate, 1% Na lauryl sarcosine, pH 8.0 at 50uC overnight. Plugs were washed (20 mM Tris, 50 mM EDTA, pH 8.0) 4 times one hour each and stored in wash buffer. Prior to electrophoresis, plugs were placed on a glass plate and trimmed to fit electrophoresis wells. Samples were then fractionated on 1% agarose gels in TBE using a Bio-Rad CHEF-DR II pulsed field electrophoresis system at 6V/cm for 22 hrs with a switch ramp time ramped from 50 to 90 sec at 14uC. Gels were stained with ethidium bromide (0.5 mg/ml, 15 min) prior to photography.

Two-dimensional gel analysis
Genomic DNA preparations and two-dimensional gel electrophoresis were performed according to detailed online methods available from the Brewer-Raghuraman laboratory: (http://fangman-brewer.genetics.washington.edu/DNA_prep. html) (http://fangman-brewer.genetics.washington.edu/2Dgel.html) In brief, cells were grown in 500 ml YPD until the cultures reached an OD 600 of 0.6. The cultures were synchronized in nocodazole for 2 hr, washed, and released into fresh YPD containing 0.033% MMS. After a 3 hr treatment, cells were harvested by centrifugation and stored in 5 ml of NIB buffer (17% glycerol, 50 mM MPOS free acid, 150 mM potassium acetate, 2 mM magnesium chloride, 150 mM spermine and 500 mM spermidine, pH 7.2). Cells were lysed by bead beating in NIB buffer, and genomic DNA was purified on cesium chloride density gradients. The resulting DNA samples were digested with HindIII and EcoRV. For first dimension separation, ,30 mg of digested DNA was loaded onto 0.35% agarose gels and fractionated at 22 volts for 42-48 hr at room temperature. Gel slices containing DNA in the 3-10 kb range were excised and positioned onto a 0.95% agarose gel. Electrophoresis in the second dimension was performed at 4uC at 80 volts for 17 hr at room temperature and 130 volts for another 1.5 hr. Following transfer to nylon membranes (Hybond-XL, GE Healthcare), samples were hybridized with a 280 bp ARS305 DNA fragment PCR amplified from genomic DNA using the following primers: 59-CTCCGTTTTTAGCCCCCCGTG-59-GATTGAGGCCACAGCAAGACCG The PCR product was radio-labeled (Megaprime DNA labeling system, GE Healthcare) and hybridized using Southern blot procedures as previously described [83].

Protein techniques
Protein extracts were prepared by mechanical beakage of cells in 20% TCA as previously described [36]. 6% SDS-PAGE gels were used to fractionate samples for analysis of Sgs1p-myc and Pds1p-myc, while 12% SDS-PAGE gels (acrylamide: bis = 30:0.39) were used to analyze phosphorylated species of Rad53p. a-myc (9E10, 1:1000, Covance), a-Rad53p (SC-6749, 1:2000, Santa Cruz), and HRP conjugated secondary (1:25,000; Jackson ImmunoResearch) antibodies were used for immunoblotting. Figure S1 Analysis of Sgs1p SUMO conjugation. A sgs1D mutant strain (MLY200) was transformed with plasmids encoding untagged SGS1 (no tag; YCplac33/SGS1), SGS1-18Xmyc (pJBN276), or the sgs1-K621R-18Xmyc mutant allele (pJBN277). Duplicate cultures of these transformants were treated with 0.03% MMS for 2 hrs or were maintained in MMS-free media, and protein samples were analyzed by a-myc immunoblotting. As observed previously [11], a shifted form of Sgs1p is observed in MMS-treated SGS1 transformants but not in cells transformed with the sgs1-K621R allele. Found at: doi:10.1371/journal.pgen.1001355.s001 (0.88 MB TIF) Figure S2 Microcolonly analysis of MMS recovery in ulp2D mad2D mutants. In conjunction with the experiments shown in Figure 6, mid-logarithmic phase cultures of a ulp2D mad2D strain (JBY733) were split. One half was treated with 0.01% MMS for 2 hr while the other half was cultured for an equivalent period in MMS-free media. For both treated (+MMS) and untreated (2MMS) cultures, 50 large-budded cells (2 cell bodies) were micro-manipulated onto a grid pattern on a YPD plate, and the number of cell bodies in the resulting micro-colonies were monitored over 3 days. The graphs display the average from three separate experiments, 6 one standard deviation. The average percent of MMS treated cells that remain blocked with 2-3 cell bodies is indicated. Found at: doi:10.1371/journal.pgen.1001355.s002 (0.31 MB TIF) Figure S3 FACS analysis of MMS treated ulp2D cells. WT (MLY061) and ulp2D (MLY060) strains were treated with 0.03% MMS for 2 hrs (22 to 0 timepoints) and allowed to recover in MMS-free media. At the indicated times, samples were withdrawn and DNA content was examined by flow cytometry. Although ulp2D cells are blocked for chromosome segregation, they show an apparent increase in DNA content, suggesting that following this mitotic catastrophe they may attempt an addition round of DNA replication. Found at: doi:10.1371/journal.pgen.1001355.s003 (0.81 MB TIF) Figure S4 Chromosome segregation in MMS treated cdc14-1 sgs1D cells. cdc14-1 sgs1D RAD52-GFP cells (JBY1808) were cultured at 23uC, treated with 0.01% MMS for 2 hr, and allowed to recover at 35uC to inactivate cdc14-1. After a 5 hr recovery, aliquots were fixed, stained with DAPI, and imaged to visualize DNA and Rad52p-GFP foci. Micrographs depict cells in which chromosome separation and nuclear segregation are largely complete, but with faint DAPI-staining fibers that appear to connect the nuclei. Persistent Rad52p-GFP foci are often observed in these cells. In some cases the Rad52p-GFP foci appear to colocalize on the DAPI-staining fibers, while in others the foci appear to have partitioned with the bulk of chromosomal DNA. BLM-deficient human cells (BLM is the human homologue of Sgs1) exposed to replication stress have been shown to accumulate ultra-fine sister chromatid bridges that are thought to arise from unresolved DNA linkages [12]. Speculatively, MMS treated sgs1D cells may form similar types of structures. Found at: doi:10.1371/journal.pgen.1001355.s004 (2.40 MB TIF)