S. cerevisiae Cells Can Grow without the Pds5 Cohesin Subunit

ABSTRACT During DNA replication, the newly created sister chromatids are held together until their separation at anaphase. The cohesin complex is in charge of creating and maintaining sister chromatid cohesion (SCC) in all eukaryotes. In Saccharomyces cerevisiae cells, cohesin is composed of two elongated proteins, Smc1 and Smc3, bridged by the kleisin Mcd1/Scc1. The latter also acts as a scaffold for three additional proteins, Scc3/Irr1, Wpl1/Rad61, and Pds5. Although the HEAT-repeat protein Pds5 is essential for cohesion, its precise function is still debated. Deletion of the ELG1 gene, encoding a PCNA unloader, can partially suppress the temperature-sensitive pds5-1 allele, but not a complete deletion of PDS5. We carried out a genetic screen for high-copy-number suppressors and another for spontaneously arising mutants, allowing the survival of a pds5Δ elg1Δ strain. Our results show that cells remain viable in the absence of Pds5 provided that there is both an elevation in the level of Mcd1 (which can be due to mutations in the CLN2 gene, encoding a G1 cyclin), and an increase in the level of SUMO-modified PCNA on chromatin (caused by lack of PCNA unloading in elg1Δ mutants). The elevated SUMO-PCNA levels increase the recruitment of the Srs2 helicase, which evicts Rad51 molecules from the moving fork, creating single-stranded DNA (ssDNA) regions that serve as sites for increased cohesin loading and SCC establishment. Thus, our results delineate a double role for Pds5 in protecting the cohesin ring and interacting with the DNA replication machinery.

advantage of the pds5-1 mutant, which can grow at the permissive temperature of 25°C, but does not grow at temperatures higher than 34°C (6,20,29). Previous studies revealed that a deletion of the ELG1 PCNA unloader suppresses the temperature sensitivity of pds5-1 mutant cells, allowing them to grow at higher temperatures (29). We confirmed this result (data not shown) and tried to test whether the lack of Elg1 could also suppress a total deletion of PDS5. We created a pds5D elg1D double mutant strain kept alive by the presence of a URA3-marked centromeric plasmid carrying the PDS5 gene. This strain, however, was unable to form colonies on 5-fluoroorotic acid (5-FOA) plates, which select for Ura 2 cells that have lost the covering plasmid (see Fig. S1A in the supplemental material). We thus conclude that whereas the deletion of ELG1 can suppress the pds5-1 temperature-sensitive allele, which may still carry some residual Pds5 protein at high temperature, it cannot rescue the complete lack of Pds5 protein.
To better understand the interactions between Pds5 and Elg1, we performed two independent genetic screens looking for the suppressors of the pds5D elg1D double mutant. We looked for high-copy-number suppressors on the first screen, whereas in the second screen, we searched for spontaneous mutations in the genome that allowed the pds5D elg1D strain to survive without the covering plasmid.
Pds5 ensures cell viability by enhancing the amount of Mcd1 in cohesin complexes. In our high-copy-number suppressor screen, we transformed a pds5D elg1D strain kept alive by the presence of a covering URA3 PDS5 TRP1 plasmid with a yeast genomic library overexpressed from a 2m plasmid marked with a LEU2 marker (the Yeast Genomic Tiling Collection) (31) (Fig. 1A). We searched for colonies able to grow in the absence of the covering plasmid. Since 5-FOA-resistant colonies could also arise from mutations in the URA3 gene carried on the plasmid, we identified Leu 1 5-FOA r (Ura 2 ) Trp 2 colonies, and isolated their library LEU2-marked plasmid (Fig. 1A).
Out of the 80 Leu 1 Ura 2 Trp 2 colonies obtained, 53 plasmids carried the genomic fragment carrying the PDS5 gene, confirming the validity of our approach. Twenty-one additional plasmids carried a DNA fragment containing the MCD1 gene. Mcd1 is one of the four core subunits of the cohesin complex. We further confirmed these results by transforming the cells with a subclone carrying only the MCD1 gene. Figure 1B shows that overexpression of MCD1 suppressed the lethality of pds5D in the absence of ELG1, but not in its presence.
To further understand the mechanism of this suppression, we selected different mutants of Mcd1 and observed their potential to rescue the lethality of pds5D and pds5D elg1D cells. We hypothesized that deletion of ELG1 may elicit the DNA damage-dependent, Chk1-dependent cohesion establishment pathway, which requires acetylation of Mcd1 at lysines 84 and 210 (32). If this proposition was true, then overexpression of the mcd1-RR allele (no acetylation possible) should not suppress, whereas overexpression of the mcd1-QQ allele (mimicking constant acetylation) should suppress the pds5D elg1D cells. However, both alleles were equally able to rescue the lethality of pds5D elg1D, suggesting that the rescue is independent of the DNA damage-mediated pathway (Fig. 1C). Furthermore, the deletion of the CHK1 gene did not affect the suppression provided by Mcd1 overexpression (data not shown).
Overexpression of Mcd1 could be titrating an interacting protein; alternatively, it might be required to increase the levels of active cohesin. We thus introduced MCD1 alleles unable to interact with cohesin (mcd1-F528R and mcd1-L532R) (33) or, as a control, an allele that does not interact with Pds5 (mcd1-V137K) (34). Figure 1D shows that only overexpression of the mcd1 alleles that could be incorporated into the cohesin complex allowed the pds5D elg1D double mutant to grow on 5-FOA plates, ruling out a titration effect. The overproduction of different Mcd1 alleles was also confirmed by Western blot in pds5D and pds5D elg1D double mutant background (Fig. S1B). Thus, increased levels of Mcd1 at chromatin allow the pds5D elg1D mutant to grow. The fact that overexpression of Mcd1 cannot suppress the single pds5D mutant but efficiently suppresses the double pds5D elg1D suggests that in the absence of Pds5, two independent changes are necessary: on the one hand, an elevation of Mcd1 levels, and on the other hand, something that the absence of ELG1 is providing. Each of these two changes is by itself insufficient to allow pds5D strains to grow.
Spontaneous mutations in the G 1 cyclin CLN2 ensure cell viability of pds5D elg1D double mutant. In our second screen, we looked for spontaneous mutants that allow the pds5D elg1D double mutant strain to lose its covering plasmid. We plated a large number of yeast cells on 5-FOA plates in several batches and looked for colonies that grew on 5-FOA plates and were Leu 2 . We confirmed that these colonies had lost the covering plasmid and performed whole-genome sequencing to identify the suppressor mutations in the genome (Fig. 1E).
Out of the 40 independent 5-FOA-resistant Leu 2 mutants that lost their covering plasmid, 23 carried de novo mutations in the CLN2 gene. Most of the mutations were nonsense, frameshift, or indel mutations that inactivated the gene (Fig. S1C). The CLN2 gene encodes a G 1 cyclin that is necessary for the transition between G 1 and S phases. In order to test these results, we made a genomic deletion of CLN2 gene in the pds5D elg1D background. As expected, the strain carrying the triple deletion pds5D elg1D cln2D grew well on 5-FOA plates, suggesting that the CLN2 deletion suppresses the lethality of the pds5D elg1D strain (Fig. 1F). (B) 5-fold serial dilutions of cells harboring either empty vector or high-copy-number vectors overexpressing MCD1 or PDS5 in addition to the covering plasmid (carrying the URA3 and PDS5 genes); (C, D) spot assay with 5-fold serial dilutions of cells harboring either empty vector or high-copy-number plasmids overexpressing MCD1 with different mutations at specified residues in addition to the covering plasmid (carrying the URA3 and PDS5 genes). (E) Experimental regimen of a screen looking for the spontaneous suppressor mutants able to grow in the complete absence of PDS5 and ELG1; (F) spot assay with 5-fold serial dilutions of the pds5D background strains carrying specified gene deletions on Ura 2 and 5-FOA plates. All mentioned strains carry a Pds5 covering plasmid (carrying the URA3 selection marker).
A second G 1 /S cyclin gene, CLN1, has 57% sequence identity (72% in the N-terminal region) to the CLN2 gene (35) and is expressed with similar timing, attaining maximal expression during the G 1 /S transition (36). Therefore, both CLN1 and CLN2 genes are considered functionally redundant (37). Figure 1F, however, shows that a deletion of CLN1 could not suppress the lethality of the pds5D elg1D double mutant strain. As in the case of MCD1 overexpression, the deletion of CLN2 only allows growth of the pds5D strain if ELG1 is deleted too, confirming the existence of two different pathways that need to be modified to allow life in the absence of Pds5.
Pds5 counteracts mechanisms that limit Mcd1 levels in cells. Based on the results from our genetic screens, our working hypothesis was that the deletion of CLN2 mimics the overexpression of MCD1, increasing its protein level. In the following experiments, we used an auxin (3-indole acetic acid [IAA])-inducible degron (AID) in order to be able to degrade Pds5 conditionally. The AID-PDS5 strain grew normally and showed no cohesion or cell cycle defects. Adding auxin to the medium leads to the rapid degradation of Pds5 ( Fig. S2A and B). We arrested the cells in the cell cycle at the M phase with nocodazole and treated them with auxin for 2 h. As expected from previous studies (20), there is a significant decrease in the level of Mcd1 protein in the AID-PDS5 strain compared with the untagged strain in the presence of auxin (wild type [WT] versus AID-PDS5, P = 0.02) ( Fig. 2A and B; Fig. S2C). AID-PDS5 elg1D and AID-PDS5 cln2D strains treated with auxin showed a decrease of Mcd1 protein although its level was higher than that of the AID-PDS strain (AID-PDS5 versus AID-PDS5 elg1D P = 0.036; AID-PDS5 versus AID-PDS5 cln2D P = 0.016). Mcd1 levels, however, were improved in the AID-PDS5 elg1D cln2D strain in the presence of auxin (AID-PDS5 versus AID-PDS5 elg1D cln2D, P = 0.005) ( Fig. 2A and B). To follow the kinetics of Mcd1 protein in the absence of Pds5, we induced the degradation of Pds5 by adding auxin to mid-log cultures and then measured the level of Mcd1 every 20 min. Following Pds5 degradation, the Mcd1 protein levels significantly dropped in the AID-PDS5 strain and in the single elg1D and cln2D mutants. In contrast, we observed a much slower kinetic of Mcd1 reduction in the AID-PDS5 elg1D cln2D mutant, which retained more than half of the Mcd1 protein levels after 2 h of auxin addition ( Fig. 2C to F). We conclude that only the concomitant deletion of ELG1 and CLN2 can restore enough Mcd1 to allow cell growth without Pds5.
CLN2 deletion leads to overexpression of the Mcd1 gene. The high level of Mcd1 could be due to increased gene expression or to protein stabilization. To test whether the deletion of both ELG1 and CLN2 prevented Mcd1 degradation, we measured the half-life of Mcd1 in the presence of cycloheximide (CHX), which inhibits global protein synthesis. No significant difference in the rate of degradation was found between AID-PDS5 and AID-PDS5 elg1D cln2D strains in the presence or absence of auxin ( Fig. S3A to F). Therefore, the increased levels of Mcd1 in the AID-PDS5 elg1D cln2D strain are not due to the increased stability of the Mcd1 protein. We thus hypothesized that the higher Mcd1 levels would be a consequence of increased Mcd1 transcription. To test this hypothesis, we constructed a plasmid vector carrying short-lived green fluorescent protein (GFP) under the control of the MCD1 promoter and a mCherry gene under the control of a constitutive ADH1 promoter, which serves as an internal plasmid copy number control (Fig. 3A). We introduced this plasmid into the different AID-PDS5 strains, and using a flow cytometer, we measured the mean fluorescence intensity (MFI) for GFP and mCherry. We observe that the GFP/mCherry mean fluorescent intensity (MFI) ratio is significantly higher in AID-PDS5 elg1D cln2D and AID-PDS5 cln2D strains compared to AID-PDS5 in the absence or presence of auxin (Fig. 3B). To validate the results from flow cytometry, we did a Western blot analysis to observe the GFP protein levels in different strains carrying the reporter plasmid. In agreement with the earlier experiment, we observe a significant increase in the GFP protein levels in the AID-PDS5 elg1D cln2D and AID-PDS5 cln2D strains ( Fig. 3C and D).
Next, we wanted to understand how deletion of CLN2 results in hypertranscription of the MCD1 gene. Cln2 is a G 1 cyclin that promotes MBF (Mlu1 cell cycle box binding factor)dependent transcription of many DNA replication and repair-associated genes during the G 1 /S-phase transition (38). These genes contain distinct DNA binding domains for the MBF complex in their promoter (MCB motifs). The MCD1 promoter contains two putative MCB motifs. Simultaneous deletion of both MCB motifs from the MCD1 promoter completely abolished the GFP expression of all strains ( Fig. 3E and F). These results show that the increased transcription of MCD1 observed in cln2D cells is dependent on the MBF complex. Thus, the deletion of CLN2 hyperactivates the MBF complex. Our results are consistent with previous studies, which also observed a high transcription of the MBF regulon in a cln1D cln2D strain background (39,40).
Simultaneous deletion of CLN2 and ELG1 restores SCC to cells lacking Pds5. In the absence of Pds5, yeast cells die due to SCC defects. These cells are defective both in the establishment and maintenance of cohesion (6,41). Similarly, elg1D strains were shown to be slightly defective SCC and exhibit increased levels of premature sister chromatid separation (28), although it was unclear whether the defect resides in the establishment or the maintenance of the cohesion. The simultaneous deletion of ELG1 and CLN2 provides robust growth in the absence of Pds5. To test whether SCC was also restored, we used the two-dot GFP assay (42). In this assay, an array of Lac operators is inserted in the chromosomal arms, recognized by a Lac repressor-GFP fusion protein. The binding of LacI-GFP to chromosomal arms can be observed under the fluorescence microscope as a bright dot in living yeast cells. When sister chromatids are adequately aligned by cohesion, only a single dot is seen, whereas two dots are observed in cells exhibiting premature separation (42). We carried out a cohesion assay by synchronizing the cells in G 1 with a-factor, then releasing the cells into the cell cycle in the presence of auxin and nocodazole ( Fig. 4A and B). This assay mainly measures the cells' ability to establish functional cohesin molecules at the beginning of the S-phase. Under these conditions, the AID-PDS5 strain exhibited more than 40% of cells with double dots, consistent with previous reports (20,41). Deletion of ELG1 or CLN2 reduced the number of cells with premature sister chromatid separation, and the number was significantly further reduced in the AID-PDS5 elg1D cln2D strain (P = 0.021), indicative of an additive effect of the elg1D and cln2D mutations. As expected, no precocious chromatid separation was detected when auxin was omitted from the assay.
SCC is established during DNA replication in S phase and maintained until anaphase. To test for SCC maintenance, cells were synchronized in early mitosis with nocodazole (after establishing cohesion) and maintained for 2 h in the presence of auxin and nocodazole ( Fig. 4C and D). The AID-PDS5 strain exhibited a substantial maintenance defect: close to 60% of the cells exhibited two dots, consistent with previous reports (22). In this assay, the deletion of ELG1 had only a minor effect, reducing the number of two-dot cells to ;40%. In contrast, the AID-PDS5 cln2D strain strongly reduced the number of cells with two dots, not significantly changed in the AID-PDS5 elg1D cln2D strain (t test, P = 0.022). Our results thus point at two different roles of the CLN2 and ELG1 in sister chromatid cohesion: whereas both of them affect the establishment by separate pathways (and thus the mutants show additivity), the elg1D mutant plays only a small role once the sister chromatid cohesion has been established, whereas cln2D affects maintenance too. Both mutations are required for full viability (Fig. 1).
Elg1 contributes to the suppression by accumulating more PCNA on chromatin. The absence of Elg1 causes an accumulation of PCNA on the chromatin (26,43). This increased level of PCNA is held responsible for most genome instability phenotypes exhibited by elg1D strains (44). To understand the function of Elg1 in SCC, we compared pds5D cln2D strains carrying URA3 PDS5 covering plasmids, bearing different ELG1 alleles in their genomes. The ability of the different alleles to provide Elg1 function was assayed by plating on 5-FOA plates (Fig. 5). Whereas cells carrying an empty vector can lose their covering plasmid and grow on 5-FOA plates, the presence of the WT ELG1 gene prevents growth, confirming our previous observations (Fig. 5A). We observe that mutations in the ELG1 Walker A motif, alleles with reduced ability to unload and recycle PCNA, such as elg1-TT386,7DD, elg1-sim1TT386,387DD (44), Walker B mutant elg1-DVD to -KVK, and the Walker A/Walker B double mutants (45) were unable to complement the ELG1 deletion, and grew on 5-FOA plates. In contrast, mutations that do not greatly affect PCNA unloading, such as the elg1-KK343,344AA allele, fully complemented the Elg1 defect and thus were unable to allow growth on 5-FOA plates. A good correlation was observed between the degree of sensitivity to methyl methanesulfonate (MMS) (which reflects the amount of PCNA on the chromatin [44]) and the ability to lose the covering plasmid (Fig. 5A). Moreover, PCNA variants that spontaneously disassemble from the chromatin such as pol30-D150E, -E143K, or -S152P (46), suppress the sensitivity of pds5D elg1D cln2D strains to MMS and prevent growth on 5-FOA (Fig. 5B), indicating that the effect conferred by the deletion of ELG1 is due to the increased levels of PCNA on chromatin. PCNA acts as a binding platform for the cohesin acetyltransferase EcoI (16). Therefore, a simple hypothesis to explain the increased SCC in elg1D strains is that high levels of PCNA accumulation on chromatin caused by the ELG1 deletion might elevate the chromatin levels of EcoI protein. To test this possibility, we monitored EcoI's overall chromatin abundance. We observe that although the elg1D strain has higher levels of PCNA on chromatin, a corresponding increase in EcoI abundance is not observed (Fig. 5C and D).
Suppression of Pds5 depletion suggests that cohesin function is limited by Elg1-dependent removal of SUMOylated PCNA from DNA. The posttranslational modifications of PCNA play an essential role in genome stability by coordinating several replication-coupled DNA damage tolerance pathways. When a replisome encounters a DNA lesion on a template strand, it may undergo modifications to activate a specific DNA damage bypass pathway (reviewed in reference 23). The Rad6/Rad18-dependent PCNA monoubiquitination at the K164 residue results in recruitment of an error-prone TLS (translesion synthesis polymerase) which adds more or less random bases at the damage site, allowing its bypass. The Rad5-dependent polyubiquitination at the K164 residue promotes an error-free template switch pathway (47). Similarly, PCNA SUMOylation at K127 and K164 by the SUMO ligase Siz1 recruits the helicase Srs2, which acts as a local antirecombination factor (48).
In order to test whether PCNA modification plays any role in the suppression via elg1D, we mutated the conserved lysine residues K164 and K127 to the unmodifiable residue arginine in the background of pds5D elg1D cln2D. Interestingly, we find that the PCNA mutations pol30-K164R and pol30-KK127,164RR both prevent plasmid loss and render cells inviable on FOA plates (Fig. 6A). These results suggest that elg1D contributes to suppression by accumulating modified PCNA on chromatin. Next, we asked which kind of PCNA modification (SUMOylation or ubiquitination) is essential for promoting cohesion via elg1D. Deleting RAD18 or RAD5 in the pds5D elg1D cln2D background renders these strains susceptible to the DNA-damaging agent MMS; however, the lack of these factors did not affect the growth of yeast cells on FOA plates. In contrast, the deletion of the SUMO ligase Siz1 in the pds5D elg1D cln2D background abolished the rescue, and cells could not grow on FOA plates (Fig. 6B). Therefore, we conclude that elg1D promotes cohesion by accumulating SUMOylated PCNA on the chromatin.
Suppression of Pds5 depletion suggests that cohesin function is limited by Srs2-dependent removal of Rad51. Srs2 is a helicase that inhibits homologous recombination by stripping Rad51 filaments from the single-stranded DNA (ssDNA) (49). Srs2 binds to SUMOylated PCNA, and we have shown that elg1D strains accumulate a high level of both SUMOylated PCNA and Srs2 on chromatin (26). Based on this information, we deleted SRS2 in the pds5D elg1D cln2D background and found that indeed pds5D elg1D cln2D srs2D strains are unable to lose the covering PDS5 plasmid and are inviable on FOA plates.
Moreover, we could rescue this quadruple mutant by deleting the RAD51 gene, encoding an ssDNA binding protein involved in homologous recombination and substrate of Srs2 (Fig. 6C). Therefore, in summary, we have found that elg1D promotes cohesion in the absence of Pds5 by accumulating SUMOylated PCNA on chromatin, thus promoting Srs2 activity to remove Rad51 filaments from ssDNA. We propose that by removing the Rad51 nucleoprotein, Srs2 generates ssDNA, which allows the deposition of cohesin molecules to establish sister chromatid cohesion when Pds5 is not present.

DISCUSSION
Sister chromatid cohesion plays a fundamental role in cell division by ensuring faithful chromosome segregation. The establishment of sister chromatid cohesion is intimately linked to DNA replication, and many bona fide replication factors have been shown to be essential for cohesion establishment (13,50,51). In this study, we aimed to explore the genetic interactions between the PCNA unloader Elg1 and the cohesin accessory subunit Pds5. Although previous work showed that the deletion of ELG1 could allow a temperature-sensitive pds5-1 strain to grow at higher temperatures (22), the mechanistic details of this genetic interaction were not well understood.
Our genetic screens show that cells can retain SCC and viability in the absence of Pds5, if the essential functions provided by this protein are supplied by two alternative routes. We show that Pds5 protein is critical to protect cohesin function that is limited by Cln2-dependent inhibition of the MCD1 transcription at the G 1 /S transition. We also show that the loss of cohesion caused by Pds5 deficiency can be partially suppressed by ectopic overexpression of MCD1 or by deletion of CLN2 (Fig. 1). Our results indicate that cln2D enhances cohesin function by promoting MBF activity, and thus MCD1 cell cycle-dependent transcription at the G 1 /S transition. Thus, the set point for cellular cohesin function is below its potential capacity because of limiting MCD1 transcription early in the cell cycle. The notion that Mcd1 transcription limits cohesin function to suboptimal levels has precedent in recent studies of Ewing sarcoma (52). These studies demonstrated that EWS-FLS1 fusion, a key determinant of this cancer, causes replicative stress and cellular senescence. The acquisition of an extra copy of the RAD21 (human ortholog of MCD1) dampens this stress and increases cell proliferation. Thus, also in these cells the level of Rad21 expression is suboptimal for addressing replicative stress (52). The existence of a suboptimal set point for MCD1 transcription for cohesion and DNA repair infers optimal levels may have counteracting deleterious effectsfor example, inhibiting chromosome segregation or cohesin-independent pathways of DNA repair. Indeed, artificially limiting the Mcd1 levels by a quantized reduction (QR) approach affects the chromosome condensation, repetitive DNA stability, and DNA repair in yeast (53). While previous studies have not revealed phenotypes for cells overexpressing MCD1, our study suggests that a more comprehensive characterization of chromosome segregation, DNA repair, and transcription in these cells is warranted.
Thus, our work helps delineate the molecular roles played by the Pds5 cohesin accessory factor.
Pds5 is a cohesin stabilizer during S phase. Cells lacking Pds5 protein exhibit high levels of premature separation of sister chromatids, which eventually jeopardize the chromosomal segregation program and result in cell death (6,20,34) (Fig. 3). Previous work showed that deletion of the SUMO E3 ligase Siz2 can rescue the temperature sensitivity and cohesion defects of the pds5-1 temperature-sensitive strain by protecting the cohesin subunit Mcd1 from SUMO-dependent degradation (20). These results imply that Pds5 exerts a protective effect, and in its absence, Mcd1 is degraded, leading to the disintegration of cohesin complexes and to premature sister separation. However, overexpression of Mcd1 from high-copy-number plasmids or by deleting the G 1 cyclin CLN2 was not sufficient to restore viability to cells completely lacking the Pds5 protein (Fig. 1). These results suggest that Pds5 plays several different roles in SCC. Our unbiased genetic screens help delineate them.
By using a degron allele of PDS5, we demonstrate that indeed, Mcd1 is quickly degraded following the auxin-induced degradation of Pds5, resulting in cell death. In contrast, we find that in the background of elg1D cln2D, Mcd1 protein no longer follows the sharp degradation kinetics associated with auxin-induced Pds5 degradation (Fig. 2). Thus, decoupling the dependence of Mcd1 protein on Pds5 for its stability renders the pds5D elg1D cln2D strain viable. Altogether, our results show that Pds5 provides essential protection to the cohesin complex. Recently it was observed that conditional degradation of Pds5 adversely affects the loop extrusion activity of a cohesin complex (54). The loop extrusion function of Pds5 is linked to its cohesin stabilization activity (55,56). The observation that the pds5D elg1D cln2D strain has sufficient cohesion ( Fig. 3) suggests that these cells stabilize cohesin complex in the absence of Pds5. In the future, it will be interesting to observe the cohesin's loop extrusion activity in the elg1D cln2D background.
The G 1 cyclin CLN2 as a novel suppressor of Pds5. In budding yeast, three G 1 cyclin genes, CLN1, CLN2, and CLN3, are critical for starting the cell cycle and entry into subsequent cell cycle phases (57). These cyclins associate with the cell-cycle-dependent kinase Cdc28 in a spatial and temporal manner to regulate the global gene expression. The Cln3 cyclin works upstream and is essential for the start of the cell cycle (58), where it activates the SBF (Swi4 cell cycle box binding factor) and MBF transcription complexes. Cln1 and Cln2, on the other hand, are mainly involved in the G 1 /S transition and are believed to play functionally redundant roles (37).
We show that deletion of CLN2, but not CLN1, provides viability to a pds5D elg1D strain (Fig. 1). This result provides strong evidence that Cln1 and Cln2 are functionally distinct. The effect of cln2D is not due to increased stability of the Mcd1 protein, but rather to increased transcription of the MCD1 gene by the MBF complex in the absence of CLN2 (Fig. 4). G 1 cyclins Cln1 and Cln2 play a vital role in generating a phospho-degron on Sic1 protein, which is a potent S-phase inhibitor (59). The deletion of CLN2 delays the entry into S phase, prolonging the transcription period of MCD1 and leading to an accumulation of its product. Thus, cln2D, similar to the high-copy-number plasmid carrying the MCD1 gene, rescues Pds5 deletion by providing an adequate amount of Mcd1 to compensate for its higher turnover in the absence of Pds5. These results establish an essential role of Pds5 in protecting Mcd1 at the G 1 /S boundary to ensure proper SCC.
ELG1 deletion promotes cohesion via SUMO-PCNA. In the absence of ELG1, cells accumulate PCNA on chromatin, both unmodified and SUMOylated (26). In the two-dot assays, the deletion of ELG1 showed its effect mainly during SCC establishment and had only a minor effect during SCC maintenance (Fig. 3). By using different elg1 alleles, we show that the ability of the different alleles to confer viability to a pds5D cln2D strain is negatively correlated with their sensitivity to DNA-damaging agents (Fig. 5A), reflecting their ability to unload PCNA from chromatin (44). Moreover, mutations in PCNA that lead to their spontaneous disassembly from chromatin (46) completely abolished the suppressive effect produced by deleting ELG1. Taken together, these results show that the suppression of pds5D cln2D is due to higher PCNA levels on the chromatin in the absence of the Elg1 PCNA unloader. The EcoI acetyltransferase binds PCNA, directly linking cohesion establishment to DNA replication (16). A simple model for the effect of deleting ELG1 on the suppression of pds5D would therefore be through increased recruitment of the EcoI acetyltransferase. Unexpectedly, although high levels of PCNA on chromatin were observed in elg1D, the EcoI levels on chromatin were not affected ( Fig. 5C and D), ruling out this simple explanation. However, despite the lack of increase in EcoI protein abundance at the fork, the level of EcoI-dependent Smc3 acetylation is elevated in elg1D mutants (60).
SUMOylated PCNA recruits Srs2 to evict Rad51 from chromatin. Srs2 is a DNA helicase that evicts Rad51 filaments from the ssDNA and performs pro-and antirecombination roles during DNA replication (61,62). Srs2 is recruited to chromatin by binding to SUMOylated PCNA (26) and has previously been shown to affect SCC (51). Our results show that Srs2 plays a central role in the procohesion phenotype conferred by elg1D. Mutations that preclude SUMOylation of PCNA, or deletion of the SRS2 gene itself, abolished the suppressive effect of elg1D and led to inviability of pds5D elg1D cln2D cells. Consistent with the known function of Srs2 function, the viability of a pds5D elg1D cln2D srs2D strain could be restored by deleting the RAD51 gene, demonstrating that the role of elg1D is to recruit Srs2 in order to evict Rad51 from the chromatin (Fig. 6C).
What could be the consequence of Rad51 eviction? One possible explanation is that eviction of Rad51 exposes ssDNA, and this is interpreted as a local DNA damage signal which may induce EcoI activity and cohesion. This could be in principle the role played by Pds5 during S phase. Importantly, this proposed mechanism is different from the known Chk1-dependent pathway in which DNA damage induces cohesion through acetylation of Mcd1 at lysines 84 and 210 (49) (Fig. 1D). Similarly, a complete deletion of CHK1 had no effect on the viability of a pds5D elg1D cln2D strain and did not prevent suppression of a pds5D elg1D strain by overexpression of Mcd1 (data not shown).
An alternative possibility is that Rad51 eviction allows the coupling between DNA replication and SCC establishment. Elegant biochemical assays by the Uhlmann's lab recently established that cohesin can be loaded onto double-stranded DNA (dsDNA), but second-strand entrapment requires ssDNA (63). They therefore suggested a model in which cohesin is loaded onto the dsDNA present on the leading strand at the moving fork, followed by entrapment of ssDNA at the lagging strand, which is then stabilized by further DNA synthesis (63). Thus, a stretch of protein-free ssDNA becomes essential for cohesion establishment. The ssDNA gaps left by Rad51's eviction could thus allow more cohesion establishment in elg1D. Smc3 acetylation is a hallmark of stably established cohesion, and Smc3 acetylation protein levels are used as a proxy to monitor the extent of cohesion establishment during DNA replication (14). Consistent with our model, the elg1D mutant has a higher level of Smc3 acetylation than the wild type (60), suggesting that the absence of Elg1 promotes increased cohesion establishment, provided that an ample enough amount of Mcd1 protein is available.
A model for the roles of Pds5 and the suppression of pds5D by elg1D cln2D. Our results delineate two essential roles for Pds5 in SCC: it protects the integrity of cohesin by preventing Mcd1 degradation, and it is involved in the activation of Smc3 acetylation by EcoI. These two roles take place during S phase and coordinate DNA replication with SCC.
Pds5 is necessary in order to protect the Mcd1 protein from SUMOylation and STUbL-dependent degradation (20,21,64). Deletion of both CLN2 and ELG1, or overexpression of MCD1 from a plasmid, contribute to increase Mcd1 levels. Whereas the first deletion increases MBFdependent transcription of the MCD1 gene (Fig. 4), ELG1 deletion may indirectly ensure higher levels of cohesive cohesin, in which, after EcoI activity, Mcd1 may become resistant to degradation. However, the increase in the Mcd1 protein level is not sufficient to provide SCC in the absence of Pds5 (Fig. 7). The second role for Pds5 occurs during DNA replication and involves the activation of EcoI activity, required for stabilizing cohesin on the chromatin. This second activity can be supplied by a deletion of ELG1, provided enough Mcd1 is present. As we have shown, increased SUMO-PCNA on the chromatin allows increased cohesin loading and establishment by recruiting the Srs2 helicase to evict Rad51 (Fig. 6). The increased SCC establishment explains the ability of elg1D to rescue the temperature sensitivity of both pds5-1 and eco1-1 strains (22,65) and is consistent with higher Smc3 acetylation levels (60) of elg1D mutants. Just increasing the rate of establishment, however, is not enough, if the level of Mcd1 is kept low due to its deprotection by the absence of Pds5. Only a combination of higher Mcd1 levels (provided by cln2D or by MCD1 overexpression), together with the increased Rad51 eviction (indirectly caused by ELG1 deletion) ensure a robust SCC in the total absence of Pds5 (Fig. 7). In summary, our results thus provide novel insights on the function of the accessory cohesin subunit Pds5 in SCC.

MATERIALS AND METHODS
Yeast strains and media. All yeast strains used in this study are of the A364A background. The genotypes of the strains used are shown in Table 1. Yeast extract-peptone-dextrose (YPD) medium was prepared with a ready-to-use mixture (Formedium). Synthetic complete (SC) minimal medium was prepared with 2% dextrose (Formedium), yeast nitrogen base without amino acids (Difco), and all necessary amino acids. Two percent agar (Difco) was added for solid medium. Auxin (3-indole acetic acid [IAA]) (Sigma-Aldrich; catalogue no. I3705) was added to SC minimal medium with a 300 mM final concentration in dimethyl sulfoxide (DMSO). 5-FOA is synthetic defined (SD) medium with all amino acids and nucleobases, but only 50 mg of uracil and 0.8 g of 5-fluoroortic acid (5-FOA) were used per liter of medium.
Cell cycle arrest. For experiments requiring cell cycle arrest, cells were grown at 30°C in SC medium until mid-log phase (optical density at 600 nm [OD 600 ] of 0.6) and incubated with nocodazole (Sigma-Aldrich; catalogue no. M1404) (15 mg/mL) for G 2 /M arrest or a-factor (Sigma-Aldrich; catalogue no. T6901) (50 ng/mL) for G 1 arrest. Both incubation times were of 2-h duration. The figure legends mention all cell cycle arrest experiment details.
Yeast spot assays. Cells were grown to saturation in SC medium at 30°C, diluted to an OD 600 of 1, and then plated in 5-fold serial dilutions. Cells were incubated on plates at 30°C for 3 to 5 days. Ten microliters from each appropriate dilution was then spotted onto the respective plates.
Yeast genetic screen for the suppressors of pds5D elg1D. For the high-copy-number suppressor screen, the yeast cells were transformed with the entire Prelich collection, consisting of over 1,500 plasmids containing a unique clone of a segment of the yeast S. cerevisiae genome. The plasmids used in this study are listed in Table 2. The cells were plated on 5-FOA plates to lose the Pds5 covering plasmid. The colonies that grew on 5-FOA were confirmed for the loss of covering plasmid followed by plasmid isolation and sequencing. The library was constructed by partially digesting prototrophic yeast genomic DNA with MboI and subcloning   (Continued on next page) it into the BamHI sites of the Escherichia coli-yeast shuttle vector pGP564. The proteins are untagged and expressed from their endogenous wild-type promoter. The pGP564 shuttle vector contains the LEU2 selectable marker and 2m plasmid sequences necessary to maintain a high copy number in yeast. The average insert size in this library is approximately 10 kb, with each insert containing an average of 4 to 5 genes.
For the spontaneous suppressor screen, the cells carrying a double deletion of PDS5 and ELG1 and a URA3 PDS5 LEU2 covering plasmid were plated on 5-FOA plates. Cells that grew on 5-FOA and were also Leu 2 (i.e., lost the covering plasmid) were subjected to whole-genome sequencing to find suppressor mutations in the genome.
Whole-genome sequencing of yeast strains. Sequencing libraries were constructed for each strain from whole-genome DNA, using a small-volume Nextera (Illumina.com) tagmentation protocol (66). Unique combinations of Nextera dual-index adapters were used for each sample, and all samples were multiplexed onto one Illumina HiSeq 2000 lane. Sequencing was performed at the Stanford Center for Genomics and Personalized Medicine using 2 Â 101-bp paired-end read technology. Variant calling was carried out using CLC Genomics Workbench v8.5 (Qiagen.com).
Cohesion analysis using the LacO-LacI system. We monitored the cohesion establishment and maintenance using the LacO-LacI system. Briefly, cells carrying tandem LacO repeats integrated at LYS4, located 470 kb from CEN4, and a GFP-LacI fusion was used. For establishment experiments, cells were grown at 30°C in SC minimal medium until mid-log phase (OD 600 of 0.6) and then incubated with a-factor (50 ng/mL) for G 1 arrest for 2 h. For depletion of AID-Pds5, auxin was added (300 mM) simultaneously. After this incubation, cells were washed three times in YPD (30°C) containing 0.1 mg/mL pronase E (Sigma-Aldrich; catalogue no. P5147), resuspended in SC minimal medium containing nocodazole (15 mg/mL), and then incubated at 30°C for 2 h to early mitosis arrest while cohesion disjunction was analyzed every 20 min. For maintenance experiments, cells were grown at 30°C in SC minimal medium until mid-log phase (OD 600 of 0.6) and then incubated with nocodazole (15 mg/mL) for 2 h. After this incubation, auxin was added (300 mM) for the depletion of AID-Pds5 proteins together with nocodazole (15 mg/mL) for 2 h at 30°C, while cohesion disjunction was analyzed every 20 min. Images were acquired with an EVO FL microscope (Thermo Fisher Scientific; catalogue no. AMF4300) equipped with the GFP Light Cube (470/22nm excitation and 510/42-nm emission) (Thermo Fisher Scientific; catalogue no. AMEP4651).
Data availability. Whole-genome sequences were uploaded to the NIH SRA database under project no. PRJNA742489.

SUPPLEMENTAL MATERIAL
Supplemental material is available online only.