Srs2 promotes synthesis-dependent strand annealing by disrupting DNA polymerase δ-extending D-loops

Synthesis-dependent strand annealing (SDSA) is the preferred mode of homologous recombination in somatic cells leading to an obligatory non-crossover outcome, thus avoiding the potential for chromosomal rearrangements and loss of heterozygosity. Genetic analysis identified the Srs2 helicase as a prime candidate to promote SDSA. Here, we demonstrate that Srs2 disrupts D-loops in an ATP-dependent fashion and with a distinct polarity. Specifically, we partly reconstitute the SDSA pathway using Rad51, Rad54, RPA, RFC, DNA Polymerase δ with different forms of PCNA. Consistent with genetic data showing the requirement for SUMO and PCNA binding for the SDSA role of Srs2, Srs2 displays a slight but significant preference to disrupt extending D-loops over unextended D-loops when SUMOylated PCNA is present, compared to unmodified PCNA or monoubiquitinated PCNA. Our data establish a biochemical mechanism for the role of Srs2 in crossover suppression by promoting SDSA through disruption of extended D-loops. DOI: http://dx.doi.org/10.7554/eLife.22195.001


Introduction
Loss of heterozygosity (LOH) can be an important contributor to carcinogenesis and a principal mechanism leading to LOH is crossover formation by homologous recombination (HR) (Knudson, 2001;Stern, 1936). HR is a high-fidelity DNA repair pathway for DNA double-strand breaks (DSB) and interstrand cross-links (Heyer et al., 2010). HR is also involved in the recovery of stalled or collapsed replication forks, as well as in the bypass of damage in the DNA template. In general, there are two types of HR repair products, crossovers and non-crossovers, depending on whether the arms of sister chromatids or homologs are exchanged or not, respectively. Crossovers between non-allelic DNA regions lead to chromosomal aberrations including deletions, inversions, and translocations. Non-crossover is the preferred outcome of HR in somatic cells avoiding potential loss of heterozygosity and possible chromosome rearrangements. In contrast, during meiosis HR is designed to generate at least one crossover per homolog to facilitate accurate chromosome segregation during the first meiotic division (Hunter, 2015). The D-loop is the joint molecule produced by the Rad51-DNA filament after homology search and DNA strand invasion, key steps of the HR process (Heyer et al., 2010). DNA polymerases, in particular DNA polymerase d, extend the invading 3'-OH in the D-loop (McVey et al., 2016). The extended D-loop is a critical intermediate in determining crossover/non-crossover outcome. In the double Holliday junction (dHJ) pathway, where both DSB ends are engaged to form a joint molecule with two Holliday junctions (HJs), both crossover and noncrossover products can be formed by nucleolytic resolution, while enzymatic dissolution leads to non-crossover products only. In break-induced replication, recombination-associated DNA synthesis extends for the rest of the chromosome arm directly causing LOH. During Synthesis-Dependent Strand Annealing (SDSA), the newly extended D-loop is disrupted to produce exclusively non-crossover products, thus avoiding the possibility for LOH. SDSA is the preferred HR pathway in somatic cells and is likely dynamically regulated to ensure faithful DNA repair while avoiding LOH and chromosome rearrangements.
Multiple enzymes can act on D-loops either to abort recombination (anti-recombination) or influence the repair outcome (crossover/non-crossover) (Heyer et al., 2010). The first DNA strand invasion product is the nascent D-loop and reversion of this HR intermediate represents a mechanism of anti-recombination. Several proteins have been shown to disrupt protein-free, nascent D-loops in vitro, including Sgs1 and its human homolog BLM (Bachrati et al., 2006;Fasching et al., 2015;van Brabant et al., 2000). Rad54 is required for D-loop formation in yeast and stimulates D-loop formation by the human RAD51 protein (Ceballos and Heyer, 2011). Both yeast and human Rad54 can also dismantle D-loops made with short invading ssDNAs in reconstituted in vitro reactions containing Rad51 and RPA. This is especially true for invading ssDNA lacking flanking dsDNA regions such as found in a resected DSB (Bugreev et al., 2007a;Wright and Heyer, 2014). Moreover, Top3-Rmi1, a type1A topoisomerase that associates with Sgs1 (human BLM-TOPOIIIa-RMI1/2 is the homologous complex) efficiently dissolves Rad51-mediated nascent D-loops in reconstituted reactions, an activity that is shared by the human TOPOIIIa-RMI1/2 complex (Fasching et al., 2015) and appears to be physiologically relevant (Kaur et al., 2015;Tang et al., 2015). Finally, human RTEL-1 can reverse RAD51-mediated D-loops and may act in the specific context of telomeres in dismantling T-loops (Vannier et al., 2012).
Although quite similar in DNA structure, the extended D-loop represents a very different target, and dismantling this HR intermediate is pro-recombinogenic affecting the crossover/non-crossover outcome. Three major helicases have been genetically implicated in this process in the budding yeast Saccharomyces cerevisiae: Sgs1, Mph1, and Srs2. Current evidence suggests that all three act in distinct but possibly partially overlapping pathways (Ira et al., 2003;Mitchel et al., 2013;Prakash et al., 2009). These helicases represent the prototypes for the homologous or analogous activities in all eukaryotes including the human proteins BLM (yeast Sgs1), FANCM (yeast Mph1), and FANCJ, FBH1, PARI, RECQ1, RECQ5, RTEL-1 (proteins with analogous activities to Srs2) (Heyer et al., 2010;Sung and Klein, 2006). Sgs1 (human BLM) is involved at multiple stages in HR as part of a complex with Top3-Rmi1 (human TOPOIIIa-RMI1/2) including DSB end-resection and dHJ dissolution (Bizard and Hickson, 2014;Cejka et al., 2010;Niu et al., 2010). Mutations in Sgs1 increase the crossover/non-crossover ratio, and this effect is explained by the role of Sgs1 in dHJ dissolution, as a defect in dHJ dissolution provides opportunity for nucleolytic resolution of dHJs and crossover formation (Ira et al., 2003). Mph1 (human FANCM) is a 3'À5' DNA helicase required for recombination-dependent error-free bypass of DNA damage during DNA replication (Prakash et al., 2005;Schürer et al., 2004). Genetic evidence shows that Mph1 contributes to the SDSA pathway, and the Mph1 protein is able to disrupt nascent and extended D-loops during in vitro reconstituted reactions (Prakash et al., 2009;Sebesta et al., 2011).
The SRS2 gene was originally discovered as a suppressor of a rad6 mutation (SRS2=Suppressor of RAD Six) and equally suppresses a defect in the RAD18 gene, which encode the Rad6-Rad18 ubiquitin E3 ligase that targets PCNA for mono-ubiquitination to favor tranlesion DNA synthesis (Hoege et al., 2002;Lawrence and Christensen, 1979). Srs2 is a member of the SF1 helicase family and translocates on ssDNA with a 3' to 5' polarity using the energy from ATP hydrolysis (Rong and Klein, 1993). The Srs2 ATPase activity is required for both its activities in HR (anti-recombination, pro-SDSA; see below). The Srs2 helicase represents the prototypic anti-recombinase dismantling the Rad51-ssDNA filament, which performs the central HR reactions of homology search and DNA strand invasion (Krejci et al., 2003;Veaute et al., 2003). In addition, accumulating genetic evidence also reveals a positive role of Srs2 in HR, specifically in the formation of non-crossover products by the SDSA pathway (Aylon et al., 2003;Dupaigne et al., 2008;Ira et al., 2003;Mitchel et al., 2013;Miura et al., 2013;Robert et al., 2006;Saponaro et al., 2010). In an HO-induced ectopic recombination assay, srs2D cells show a significant decrease in the level of non-crossover products; however, the level of crossovers remains almost unchanged, leading to an increase of percentages of crossovers among the total products (Ira et al., 2003). Similarly, in a spontaneous ectopic recombination assay, srs2D cells show a four-fold increase in crossover formation such that almost half of the recombination products are crossovers (Robert et al., 2006). Post-translational modifications of PCNA are involved, as suggested by the genetic analysis of PCNA mutants defective in ubiquitylation/sumoylation, of the SUMO ligase Siz1, and the ubiquitin-conjugating enzyme Rad6, all of which shows increase in crossover levels in a plasmid-based SDSA assay (Miura et al., 2013). The involvement of SUMO-PCNA in crossover suppression is also confirmed in other ectopic recombination systems (Burkovics et al., 2013;Robert et al., 2006). In sum, these data provide compelling evidence for an additional, pro-recombinogenic role of Srs2. It was proposed that Srs2 disrupts extended D-loops to favor annealing with the second end during SDSA to generate non-crossover products (Ira et al., 2003;Robert et al., 2006).
While the mechanism of anti-recombination by Srs2 is well established, the mechanisms involved in its pro-recombination (SDSA) role and its function in adaptation/recovery remain to be determined (Macris and Sung, 2005;Marini and Krejci, 2010;Vaze et al., 2002). Biochemical characterization of Srs2 demonstrated that it unwinds various synthetic replication and recombination intermediates including replication forks, 3 0 -and 5 0 -flaps, Holliday junction (HJ), and nicked HJ, but not D-loops in reconstituted recombination reactions Marini and Krejci, 2012;Prakash et al., 2009;Sebesta et al., 2011). It was proposed that Srs2 competes with Pol d for PCNA binding to inhibit HR-associated DNA synthesis, and by this SDSA, in an ATP-independent mechanism (Burkovics et al., 2013). However, later genetic work showed that the Srs2 ATPase activity is essential for its SDSA activity (Kolesar et al., 2016;Miura et al., 2013), suggesting that competition for PCNA binding could be an aspect but not the underlying fundamental mechanism. The apparent lack of D-loop disruption activity of Srs2 also suggested a second possible mechanism for its SDSA activity by dissociating Rad51 from the second end of the DSB to enable annealing with extended invading strand from the disrupted D-loop (Marini and Krejci, 2010;Mitchel et al., 2013).
Here, we report that Srs2 is capable in vitro to disrupt Rad51/Rad54-produced D-loop structures mimicking physiological length and structures, providing biochemical evidence for a specific SDSA mechanism by Srs2. The SDSA pathway was partially reconstituted with RPA, Rad51, Rad54, RFC, DNA Polymerase d and three different but fully modified forms of PCNA. Consistent with the genetic data, we demonstrate that Srs2 disrupts D-loops in an ATP-dependent fashion with a slight but significant preference for extending D-loops over unextended D-loops when SUMO-PCNA is present, compared to unmodified PCNA or monoubiquitinated PCNA (Ubi-PCNA). Our data provide a plausible mechanism of Srs2 activity in SDSA and assign a specific role of PCNA sumoylation during HR.

Results
Srs2 disrupts D-loops mimicking physiological length in a time-and concentration-dependent manner S. cerevisiae Rad51 relies on Rad54 to generate D-loop structures. Typically in reconstituted systems, short ssDNAs (~35-100 nts) are used for Rad51 filament formation and DNA strand invasion, however the D loops products in such reactions suffer from instability due to Rad54's ability to dismantle these short D loops efficiently (Bugreev et al., 2007b;Wright and Heyer, 2014). This complicates data analysis and was found to be unsuitable to explore the D-loop disruption activity of Srs2 (Burkovics et al., 2013;Prakash et al., 2009). Thus, there is a need to generate stable D-loop structures in reconstituted reactions. To this end, we utilized a series of long ssDNA substrates, which have been designed to resemble physiological-length Rad51 filaments that produce relatively stable D-loop structure (Wright and Heyer, 2014). An ssDNA with 607 nucleotides (named '607' thereafter) of homology to a supercoiled dsDNA donor is able to form D-loops with high yield (~50%) and stability over the time course of a D-loop reaction ( Figure 1A,B). In addition, we carefully staged the reaction by adding Srs2 10 min after the initiation of D-loop formation by Rad54 and dsDNA substrate, which allows accumulation of maximal D-loop levels (Wright and Heyer, 2014). This protocol avoids the complication of overlaying the phase of D-loop formation with D-loop dissociation catalyzed by Srs2. Previous attempts to identify D-loop disruption activity by Srs2 used reaction schemes where Srs2 was added immediately after or simultaneously with Rad54 and dsDNA substrate (Burkovics et al., 2013;Prakash et al., 2009). The combination of staged reactions and long ssDNA substrates enabled us to identify robust dissociation of D-loops by Srs2 in a time-and concentration-dependent manner ( Figure 1B,C). As previously observed (Wright and Heyer, 2014), 607 can invade multiple dsDNA molecules to produce D-loops, where one ssDNA invaded up to three duplex DNA molecules ( Figure 1B). All D-loop species from single invasion to multiple invasions were disassembled almost completely by 25 nM Srs2 in a 25 min reaction ( Figure 1C). At intermediate concentrations (5 nM in Figure 1B) it appears as if multiple invasion products are more sensitive to Srs2. However, multiple invasions are processed into single invasions before complete disruption, making it impossible to assess whether multiple invasions or single invasions are preferentially disrupted. As little as 2.5 nM Srs2 showed significant D-loop disruption in these reactions that contained 1 nM invading ssDNA, showing that Srs2 is highly effective in D-loop disruption. D-loop disruption by Srs2 is not dependent on the presence of the single-stranded DNA binding protein RPA (Figure 1-figure supplement 1). We conclude that Srs2 is capable of disrupting nascent D-loops, and below we show directly that Srs2 is also capable of disrupting D-loops that are being extended by DNA polymerase d, the presumed in vivo substrate in the SDSA pathway.

The translocase/helicase activity of Srs2 is required for D-loop disruption
In plasmid-based and chromosomal assays, the ATPase activity of Srs2 is required for normal levels of non-crossover formation (Kolesar et al., 2016;Miura et al., 2013). To determine whether the ability to translocate or unwind dsDNA is required for Srs2 to disassemble D-loops, we purified an ATPase-defective Srs2 point mutant carrying a single amino acid substitution in the Walker-A motif (Srs2-K41S). Srs2-K41S displays no detectable ATP hydrolysis ( Figure 1D), while wild type Srs2 shows strong, ssDNA-stimulated ATP hydrolysis. As expected, the motor/helicase activity of Srs2 was required to dissociate D-loops ( Figure 1E), since the addition of Srs2-K41S did not alter D-loop level. The Srs2-K41S mutant protein appears to fold properly as it purifies like the wild type protein (not shown) and interacts with Rad54 like wild type Srs2 ( Figure 3E), suggesting that it is the lack of and additional titrations. Absolute initial D-loop yields were between 50% and 75%. (D) Srs2-K41S is deficient in ATP hydrolysis. Rates of ATP hydrolysis were determined for 5 nM wild type Srs2 (WT) or catalytic-deficient Srs2-K41S (K41S) in D-loop reaction buffer containing either 0 or 100 mM NaCl, with or without 10 mM jX174 ssDNA as cofactor. (E) Time course of D-loop disruption by 5 nM of either wild type Srs2 (WT) or catalytic-deficient Srs2-K41S (K41S). Plotted are means ± standard deviation from n = 3. Absolute initial D-loop yields were between 44% and 72%. (F) Induced sister chromatid recombination assay. The normalized (wild type = 100%) induced recombination frequencies are shown at 0.24 mg/mL 4-NQO as means ± standard deviations from three experiments for wild type, srs24, and srs2-K41A (see Table 1). The frequency of spontaneous G418 R cells was subtracted from the induced recombination frequencies. The full 4-NQO dose response is shown in Figure 1-figure supplement 2. DOI: 10.7554/eLife.22195.002 The following source data and figure supplements are available for figure 1: Source data 1. Source data for Figure 1C. DOI: 10.7554/eLife.22195.003 Source data 2. Source data for Figure 1D. DOI: 10.7554/eLife.22195.004 Source data 1. Source data for Figure 1E. DOI: 10.7554/eLife.22195.005 Source data 4. Source data for Figure 1F and ATPase and not incorrect protein folding that leads to the D-loop disruption defect. Moreover, the Srs2 motor activity is required for 4-Nitroquinine-N-oxide (4-NQO)-induced recombination ( Figure 1F), an assay that reports on Rad51-dependent sister-chromatid recombination (Ede et al., 2011). As expected, deletion of RAD51 is epistatic to deletion of SRS2 for induced recombination, and induced recombination is strongly Rad51-dependent ( Figure 1 Additionally, mutation of SRS2 increases the spontaneous recombination rate in this assay system and this increase is abolished by a rad51 mutation ( Figure 1-figure supplement 2D). This result is consistent with previous observations that show a hyper-rec effect for SRS2 deletion mutations in spontaneous recombination (Palladino and Klein, 1992). Finally, the Srs2 motor activity is also required for anti-recombination in vivo as indicated by the suppression of the severe sensitivity of rad184 strain towards UV and methyl methanesulfonate (MMS) by the srs2-K41R and srs2-K41A mutations ( Figure 1-figure supplement 2E). rad184 suppression by these SRS2 Walker A box mutations was not evident in a different assay of UV-sensitivity (Burkovics et al., 2013). Together these results show that the translocase/ATPase activity of Srs2 is required for its two essential functions in HR, Rad51 filament disruption (anti-recombination) and D-loop disruption (pro-SDSA/anticrossover). These results are consistent with previous findings that the Srs2 motor activity is essential for its anti-recombination activity (Krejci et al., 2004) and pro-SDSA function (Kolesar et al., 2016;Miura et al., 2013).

Srs2 is capable of disrupting different D-loop structures
During long-range DSB resection, Sgs1/Dna2 or Exo1, degrade the 5' strand to produce 3' tailed ssDNA spanning hundreds of nucleotides, which represents the in vivo substrate for Rad51 filament formation (Mimitou and Symington, 2008;Zhu et al., 2008). After showing that Srs2 efficiently disrupts D-loops, we wanted to test further whether Srs2 shows any substrate preference towards D-loops produced from ssDNA with zero (607), one (98ds-607, 3'-tailed), or two (98ds-607-78ds) terminal duplex heterologies mimicking different cellular repair scenarios of DSB repair (3'-tailed) or gap repair ( Figure 2A). These D-loops with terminal heterologies add another layer of stability against Rad54 disruption in addition to increased ssDNA length that enables longer heteroduplex (hDNA) to form (Wright and Heyer, 2014). The yield of D-loop was slightly different with all the three substrates, but Srs2 dismantled all D-loops efficiently ( Figure 2B,C). After normalizing the D-loop level according to the initial yield, it became clear that Srs2 displays a significant preference toward D-loops made with 3'-tailed invading DNA (98ds-607) ( Figure 2D). This preference is confirmed by further experiments determining the preferred orientation of D-loop disruption (see below, Figure 4). We conclude that Srs2 is capable of disrupting different nascent D-loop structures with modest preference for D-loops containing 3'-tailed invading DNA, the mimic for physiological DSB repair.
Srs2 disrupts Rad51/Rad54 produced D-loops more effectively than protein-free D-loops Next, we wanted to discern the specificity of Srs2 towards D-loops as a DNA intermediate and identify any effects of the proteins required for D-loop formation on D-loop disruption. To this end, we generated protein-free D-loops and compared the disruption by Srs2 with D-loops formed in reconstituted reactions with the cognate Rad51 and Rad54 proteins. First, we used Rad51/Rad54 to generate 607-based D-loop structure at high yield conditions and purified protein-free D-loops by removing all proteins. Then, we tested whether these protein-free D-loops can serve as substrates for Srs2. The reactions contained about 85% of protein-free D-loops, and the results are plotted as normalized D-loops with the initial level set at 100% to allow comparison. As shown in Figure 3B, 15 nM Srs2 disrupts only about 35% of the available protein-free D-loops in 40 min, while Srs2 at similar enzyme/DNA ratios disrupts almost all D-loops at 25 min in the Rad51/Rad54 reconstituted reaction ( Figures 1C, 25 nM >95%, 5 nM~70%). Using this data ( Figure 1C, Figure 3B), we estimate an about ten-fold difference in the disruption between protein-free D-loops and D-loops in the reconstituted reaction. Thus, although Srs2 recognizes and loads onto D-loop as a protein-free DNA structure, the sheer presence of Rad51 and Rad54 enables more effective disruption by Srs2. Srs2 interacts with Rad51 physically through its C-terminal interaction site , and this protein-protein interaction has been demonstrated to trigger ATP turnover and dissociation of and Srs2 is sensitive to increasing ionic strength. GST-Rad54 and Srs2 were formed in buffer containing 0, 100, 175, 250, or 400 mM NaCl before pulldown. (E) Srs2-K41S interacts with Rad54 with similar salt sensitivity, compared to wild type Srs2. Both wild type Srs2 and Srs2-K41S were allowed to form complex with GST-Rad54 in buffer containing 0, 250, and 500 mM NaCl before pulldown. Plotted are means ± standard deviation from n = 3. In (D) and (E), 31.3 nM GST-Rad54 or 313 nM GST (GE Healthcare) were incubated with 31.3 nM of Srs2 or Srs2-K41S in the same buffer containing indicated amount of NaCl for 1 hr before pulldown. Pulldown Srs2 amount in buffer containing 0 mM NaCl was normalized to 100 (D) or 1 (E). DOI: 10.7554/eLife.22195.012 The following source data is available for figure 3: Source data 1. Source data for Figure 3B. DOI: 10.7554/eLife.22195.013 Source data 2. Source data for Figure 3E. DOI: 10.7554/eLife.22195.014 Rad51 from ssDNA (Antony et al., 2009). Only amino acids 783-859 of the Rad51 protein interaction site of Srs2 were found to be required for non-crossover formation in vivo but not amino acids 860-998 (Miura et al., 2013). In the absence of more defined Rad51 interaction mutations, it is difficult to assess whether Rad51 interaction is required for SDSA. Rad54 removes Rad51 at the 3' end of the invading strand during heteroduplex formation to clear the stage for DNA polymerases Wright and Heyer, 2014), suggesting that Rad51 may not be in a position to interact with Srs2 during D-loop disruption. Therefore, we entertained the idea that Srs2 might interact with Rad54 and discovered a direct and robust physical interaction between Srs2 and Rad54 ( Figure 3C,D). The Srs2-Rad54 interaction might be involved in the observed preference to disrupt D-loops in reconstituted reactions using the cognate proteins. We conclude that interactions with Rad51 and Rad54, but not RPA (see Figure 1-figure supplement 1), potentially play an important role in Srs2-mediated D-loop disruption.
Srs2 prefers to disrupt D-loops where the 3'-end is part of the heteroduplex DNA To determine whether Srs2 has a preferred polarity in D-loop disruption, we used a previously characterized hDNA digestion assay (Wright and Heyer, 2014). Formation and decrease of hDNA regions were assessed by the ability of the invading strand to be digested by restriction enzymes once the cut site has been made double-stranded by hDNA formation in the D-loop ( Figure 4A). The 3 kb plasmid donor will allow incorporation of about 200 nucleotides of the invading strand into hDNA, limited by the number of negative supercoils . For D-loops formed by the invasion of 607, this means that the heteroduplex region will not contain the MluI and BstXI sites (362 nucleotides apart) simultaneously. The observation that the sum of the MluI (5') and BstXI (3') digestion products matches exactly the amount of D-loop products ( Figure 4C,D) independently confirms that both restriction sites never cohabitate, otherwise a sum of greater than 100% of the D-loop amount would be expected. Thus, there were two major species of D-loop, MluI-incorporated D-loop (30% yield) and BstXI-incorporated D-loop (42% yield) ( Figure 4A,E). The addition of Srs2 to an ongoing Rad51/Rad54-reconstituted D-loop reaction revealed a marked preference for disruption of D-loops in which the 3' end-proximal region (BstX1 cleavage is 111 nt from the 3' end) had been incorporated into hDNA. The 3' proximal/BstXI-incorporated D-loops show an immediate and rapid decline from 42 to 10 percent in the~30 s it took to add the enzyme and stop the reaction ( Figure 4E). In contrast, the 5' proximal MluI -incorporated D-loops showed a slower time course of disruption that did not reach a final level until 15 min ( Figure 4E). As is the case in the absence of Srs2, the D-loop levels closely follow the sum of 3' and 5' proximal digestion products, and the cleavage levels never decrease for one site while increasing for the other ( Figure 4D), indicating that Srs2 action does not result in extensive branch migration of the hDNA region before complete hDNA disruption is accomplished. We conclude that Srs2 prefers disrupting D-loops, which include the 3' region of the invading DNA in the heteroduplex DNA. This is in congruence with the data of Figure 2B-D, where ds98-607 D loops are preferentially disrupted by Srs2, followed by 607 and then gapped ssDNA. This follows the same rank order of 3' proximal site incorporation preference that was previously mapped by the hDNA incorporation assay for these same three substrates under identical conditions (Wright and Heyer, 2014). Further, this may reflect the need for Srs2 to disrupt 3' heteroduplex to promote the D-loop reversal step of SDSA.

Srs2 disrupts D-loops extended by DNA polymerase d in the presence of PCNA/RFC
The presumed D-loop substrate for Srs2 disruption during SDSA is the extended D-loop that is engaged by DNA polymerase d with PCNA/RFC, as Pol d is the major polymerase active in HR-associated first-end DNA synthesis extending the invading strand McVey et al., 2016;Sebesta et al., 2011). To present Srs2 with this substrate, we partly reconstituted the SDSA pathway in vitro using Rad51, Rad54, RPA, PCNA, RFC, and DNA polymerase d and assessed the ability and efficiency of Srs2 in disrupting extending D-loops ( Figure 5).   topological block to DNA synthesis when attempting to extend the hDNA region. Therefore, a conventional 100-mer ssDNA was used to allow clear visual difference between extended and unextended D-loops (see Figure 5B). The reaction was staged to recapitulate the flow of the SDSA pathway in vivo to maximize D-loop synthesis before the addition of Srs2 ( Figure 5A). Rad51 and Rad54 were allowed to produce high levels of D-loop before the addition of RFC/PCNA and Pol d. Srs2 was added after efficient D-loop extension. To minimize fluctuations, the newly extended D-loops were split at the end into buffers containing various amount of Srs2 to reach a final concentration of 0, 2.5, 7.5, or 25 nM, and products analyzed ( Figure 5A). At the 25 min time point, i.e. 40 min after Pol d addition, almost all D-loops had been extended and the yield of extended D-loop plateaued at 30% ( Figure 5B,C). Addition of Srs2 led to a significant decrease in extended D-loops compared to the control without Srs2 ( Figure 5C). 25 nM Srs2 decreases extended D-loop level from 30% to 7.6%, and correspondingly increases collapsed extended D-loops ( Figure 5C). These data demonstrate that Srs2 dissociates Pold-extending D-loops.
To corroborate this important conclusion, we devised an independent experimental design using a-32 P-dCTP incorporation to specifically radiolabel D-loops upon extension by Pold ( Figure 5D).
Using 25 nM of Srs2 (25 nM) essentially all extended D-loops were disrupted within the first five minutes, ( Figure 5D,E, Figure 5-figure supplement 1). We conclude that Srs2 is capable of disrupting extended D-loops in the presence of RFC, PCNA, and Pold, providing a potential mechanism for how Srs2 promotes SDSA and favors non-crossover formation. We estimate that disruption of RFC/PCNA/ Pol d extending D-loop ( Figure 5C) is at least twice faster than disruption of D-loops with the 607-based invading DNA ( Figure 1C). Thus, replication machinery-covered D-loops serve as better substrates for Srs2, suggesting a positive recruitment of Srs2 through protein-protein interaction, which we examined in more below.

Srs2 exhibits slight preference for extending D-loop using SUMO-PCNA
Previous studies have established that SUMO-PCNA exhibits higher affinity to Srs2 than unmodified or ubiquitinated PCNA (Papouli et al., 2005;Pfander et al., 2005), leading to the model that SUMO-PCNA recruits Srs2 to its site of action. Genetic analyses showed that the protein interaction of Srs2 with SUMO-PCNA is required for SDSA in vivo, as siz14 and pol30-K164R mutant strains shared the same defects as srs24 strain in SDSA (Le Miura et al., 2013). This suggests that SUMO-PCNA and its interaction with Srs2 is required for SDSA and predicts that posttranslational modifications of PCNA affect the efficiency of Srs2 in D-loop disruption. Hence, we evaluated the effect of PCNA modification by SUMO and ubiquitin on Srs2-mediated D-loop disruption of Pol d extending D-loops. Split forms of wild type PCNA (referred as WT PCNA thereafter), monoubiquitylated-PCNA (Ubi-PCNA), and SUMOylated-PCNA (SUMO-PCNA) were utilized, where either ubiquitin or SUMO was fused in frame to the C-terminal peptide to produce large quantities of Ubi-PCNA and SUMO-PCNA with uniform modifications (Freudenthal et al., 2011(Freudenthal et al., , 2010. Crystal structures of split form Ubi-PCNA and SUMO-PCNA revealed that ubiquitin and SUMO moieties were localized on the back side of the PCNA trimeric ring without changing its conformation, and indeed these split forms of PCNA function properly at both the cellular and biochemical level (Freudenthal et al., 2011(Freudenthal et al., , 2010. We reconstituted the in vitro SDSA assay with WT PCNA, Ubi-PCNA, and SUMO-PCNA separately and monitored whether Srs2 D-loop disruption activity is affected. These split forms of PCNA have been reported to support normal cell replication in yeast and stimulate the ability of pol h or Pol d to incorporate nucleotide opposite an abasic site (Freudenthal et al., 2011(Freudenthal et al., , 2010. Consistent with these reports, we did not detect a difference among different species of PCNA in recruiting Pol d to nascent D-loops for new DNA synthesis and D-loop extension in the absence of Srs2 ( Figure 5E, Figure 6, Figure 6-figure supplement 1, Figure 4 continued the two other sites that it is not mutually exclusive for incorporation into a fraction of the 5' or 3' proximal hDNA products. Plotted are means ± standard deviation from n = 3. DOI: 10.7554/eLife.22195.015 The following source data is available for figure 4: Source data 1. Source data for Figure 4.   To detect a potential preference for a form of PCNA (SUMO, ubiquitin, unmodified) we moved away from optimized reaction conditions. As shown in Figure 6A, we lowered the temperature from 30˚C to 25˚C to slow down RFC/PCNA/Pol d, shortened the Pol d extension time to avoid accumulating fully extended D-loops, and titrated a lower range of Srs2 concentrations. The results show a slight, but significant, preference to disrupt SUMO-PCNA extending D-loops over WT PCNA-or Ubi-PCNA extending D-loops at a limiting concentration of Srs2 (7.5 nM) (Figure 6). High concentrations of Srs2 (25 nM) disrupt extended D-loops with all species of PCNA equally well in this experimental setup (Figure 6-figure supplement 1). To obtain more precise quantitation, we repeated this reaction scheme with 7.5 nM Srs2 using end-labeled invading DNA substrate. As shown in Figure 7, Figure 7-figure supplement 1, the total D-loop levels were almost identical across different species of PCNA, all showing a decline from~42% to~32% after 25 min incubation with Srs2, compared to an increase from~40% to~48% in total D-loops without Srs2. Importantly, the level of extended D-loops was lower while the unextended D-loops were, if anything, higher when comparing SUMO-PCNA vs. WT PCNA or Ubi-PCNA in the presence of Srs2 ( Figure 7D,E). The ratio of extended D-loops over unextended D-loops reflects this difference, revealing that Srs2 prefers to disrupt extended D-loops over unextended D-loops when SUMO-PCNA is present ( Figure 7F). When D-loop extension was prevented by the omission of dNTPs, no significant difference in D-loop disruption by Srs2 was noted between reactions containing PCNA, Ubi-PCNA, or SUMO-PCNA (Figure 7-figure supplement 2).
To test the idea whether PCNA actively recruits Srs2 onto newly extended D-loops, we also compared the Srs2 disruption efficiency when the replication machinery was only loaded at the nascent D-loops without new DNA synthesis because of the absence of dNTP to the condition when no replication machinery was loaded. We only detected a slight difference between the conditions when SUMO-PCNA/RFC/Pol d was present or not (Figure 7-figure supplement 3). From these experiments we conclude, that Srs2 preferentially disrupts D-loops that are actively extending by DNA Polymerase d using SUMO-PCNA.

Discussion
Genetic observations that Srs2 suppresses crossover formation and promotes the SDSA pathway have suggested a biochemical role for Srs2 to disrupt D-loops and to channel the recombination intermediates into the SDSA non-crossover pathway Ira et al., 2003;Le Breton et al., 2008;Robert et al., 2006). Early biochemical studies on Srs2 either did not employ the D-loop reaction or staged the reaction to focus on the anti-recombination activity of Srs2 (Krejci et al., 2003;Veaute et al., 2003). Several later studies failed to observe D-loop α-32 P-dCTP incorporation Figure 6. Slight preference of Srs2-mediated disruption for PCNA-SUMO containing reconstituted reactions compared to unmodified or ubiquitinated PCNA. (A) Extended D-loop disruption assay using a-32 P-dCTP incorporation. Fully unlabeled and homologous 100mer ssDNA was used as substrates, and a-32 P-dCTP was included in the buffer. 0.2 mM Rad51 was incubated with 6 nM ssDNA substrates (Rad51/nt = 1:3) for 10 min at 30˚C. RPA (165 nM) was added for another 10 min incubation before the addition of 84 nM Rad54 and 7 nM supercoiled plasmid dsDNA (21 mM bp). After 2 min, 20 nM Figure 6 continued on next page disruption activity by Srs2 in reconstituted in vitro reactions although using significantly higher concentrations of Srs2 (up to 167 nM) than used here (Prakash et al., 2009;Sebesta et al., 2011). We have identified several factors that explain this apparent discrepancy of the results reported here with these studies. First, the traditional short oligonucleotide-based D-loops undergo significant instability caused by Rad54 (Bugreev et al., 2007a;Wright and Heyer, 2014), making it difficult to identify an effect of Srs2 on D-loop stability. Moreover, the effect of Srs2 in disrupting D-loops becomes more apparent at later time points of Srs2 incubation, in particular in the reconstituted reactions with PCNA/RFC and Pold (Figures 1, 5-7, Figure 7-figure supplements 2 and 3). These differences in protocol explain why no disruption was observed when Srs2 was added 1 min after plasmid dsDNA addition and the reaction time was only 8 min (Prakash et al., 2009) or only 5 min (Sebesta et al., 2011). The D-loop disassembling activity of Srs2 was much more evident, when we used D-loops made of long ssDNA with physiological-relevant length and structure, which minimize Rad54-mediated D-loop instability (Figures 1 and 2). Another previous study showed that the Srs2 C-terminal truncation containing amino acids 1-898 was able to disrupt protein-free D-loops assembled from oligonucleotides (Marini and Krejci, 2012). This result showed that Srs2 can disrupt model D-loops, but left open the critical question, whether Srs2 can act in the context of the ongoing reaction, where proteins are bound to the HR-intermediates, in particular in extended D-loops which engage PCNA/RFC and DNA Polymerase d. For example, Sgs1 disrupts protein-free-D-loops, but Sgs1 is not able to disrupt D-loops in reconstituted reactions (Fasching et al., 2015). Importantly, we were able to partly reconstitute the SDSA pathway and show that Srs2 efficiently disrupts D-loops in an ATP-dependent fashion (Figures 1 and 5). These results are consistent with the genetic requirement for the Srs2 ATPase activity for SDSA (Kolesar et al., 2016;Miura et al., 2013). These findings provide a plausible mechanism for the SDSA function of Srs2 as discussed below (Figure 8).

Mechanisms of synthesis-dependent strand annealing: A model for Srs2
SDSA appears to be the primary HR pathway in somatic cells and includes a critical step of dissociating the extended D-loop to allow annealing of the newly synthesized strand with the processed second end of the DSB (Figure 8). Mph1 has been implicated in this step by genetic and biochemical experiments (Prakash et al., 2009;Sebesta et al., 2011) and genetic experiments strongly suggested that other enzymes, specifically Srs2, play a partially overlapping role (Mitchel et al., 2013;Prakash et al., 2009). Concomitant with DNA strand invasion by the Rad51 filament, Rad54 is required to remove Rad51 during hDNA formation to allow access to PCNA loading by RFC to enable DNA synthesis by DNA Polymerase d (Li et al., 2013. We speculate that sumoylation of PCNA during HR-associated DNA synthesis increases the likelihood of Srs2 recruitment to the extending D-loop, where it disengages the polymerase from PCNA, as Srs2 has been shown to compete with Pol d for PCNA binding (Burkovics et al., 2013). Srs2 contains a PCNA interaction motif (PIP box) and a SUMO interaction motif (SIM) providing the basis for the avid interaction between Srs2 and SUMO-PCNA (Armstrong et al., 2012;Papouli et al., 2005;Pfander et al., 2005). This enhanced recognition between Srs2 and SUMO-PCNA has classically been viewed as an effective means to eliminate inappropriate recombination events through Rad51 filament disruption  (Papouli et al., 2005;Pfander et al., 2005;Stelter and Ulrich, 2003). Based on the biochemical preference of Srs2 to disrupt D-loops that are extended using SUMO-PCNA (Figures 6 and 7), we suggest that PCNA sumoylation is also critical for the tuning of the SDSA function of Srs2. This is consistent with the genetic data showing a requirement for the ATPase, the PIP box and the SIM for the pro-SDSA function of Srs2 (Burkovics et al., 2013;Kolesar et al., 2016;Le Breton et al., 2008;Miura et al., 2013;Robert et al., 2006). Additionally, ablation of Siz1-mediated sumoylation of PCNA restores the formation of damage-induced and Rad51-dependent X-shaped sister chromatid junctions in an sgs14 rad184 strain background (Branzei et al., 2008), which suggests a positive role of SUMO-PCNA in processing such HR intermediates. It is possible, but remains to be demonstrated, that SUMO-PCNA stays associated with the D-loop and Srs2 during D-loop disruption (see Figure 8). The orientation of Srs2 in disrupting the D-loop in Figure 8 by translocating 3'À5' on the invading strand is consistent with the known polarity of Srs2 (Rong and Klein, 1993) and the preference of Srs2 for D-loops with an embedded 3'-end (Figures 2 and 4). Srs2 exhibits a significant preference for disrupting D-loops in the cognate reconstituted reactions compared to protein-free DNA intermediates (Figures 1 and 3), suggesting that this activity is of biological relevance and likely related to the physical interactions of Srs2 with Rad51 Seong et al., 2009) and/or Rad54 (Figure 3). Indeed, Srs2 stimulates ATP hydrolysis activity of Rad51 and triggers the dissociation of Rad51 clusters from ssDNA (Antony et al., 2009). The same principle could be utilized to dislodge residual Rad51 and Rad54 from newly extended D-loops, which could further destabilize hDNA by removing these potentially stabilizing protein factors. It was reported that Rad51 bound to adjacent duplex DNA stimulates the helicase activity of Srs2 .
As depicted in Figure 8 and proposed for Srs2 and RECQ5 (Mitchel et al., 2013;Schwendener et al., 2010), it is possible that Srs2 also acts on the second end during SDSA dissociating Rad51 to enable reannealing, as Rad51 inhibits the Rad52 annealing reaction (Wu et al., 2008). This mechanism would be complementary to Srs2 disrupting D-loops, as Rad51 displacement alone appears to be insufficient to explain the SDSA role of Srs2. Analysis of the srs2-R1 mutant showed that the interaction of Srs2 with SUMO-PCNA is not required for its anti-recombination activity but the mutant showed the increase in crossovers indicative of a defect in SDSA . In fact, the resected DSB is not a substrate for PCNA loading by RFC, as it has the opposite polarity (5' junction) of the native RFC substrate (3'-junction) (Majka and Burgers, 2004), and it appears unlikely that PCNA would be found on resected DSBs. In sum, Srs2 and Mph1 appear to provide two distinct pathways to process extended D-loops and shunt HR to non-crossover outcome and SDSA. The genetic profiles of both mutants show a clear difference: mph1 mutants shift non-crossover to crossover events without effect on repair efficiency, whereas srs2 mutants show a decrease in repair efficiency and loss of non-crossover products (Mitchel et al., 2013;Prakash et al., 2009;Ira et al., 2003). The synthetic negative interaction between both mutants suggests partial overlap between both pathways (see Figure 8-figure supplement 1) for a discussion why there are so many proteins involved in D-loop disruption). What  (Burkovics et al., 2013). 2) Srs2 translocates in a 3' fi 5' direction to dissociate the hDNA incorporated in the D-loop (this work). After the newly extended 3' invading end is dissociated it is available to anneal with the second end to generate non-crossover products. 3) Srs2 may also dissociate Rad51 from the second end to allow single-strand annealing in analogy to RECQ5 (Mitchel et al., 2013;Schwendener et al., 2010). DOI: 10.7554/eLife.22195.031

Rad54
The following figure supplement is available for figure 8: differentiates both proteins is their physical and functional interaction with PCNA, specifically SUMO-PCNA. We suggest that Srs2 is specialized to disrupt D-loops extended with SUMO-PCNA, and that in its absence such extended D-loops become a potentially lethal intermediate.

D-loop disruption assay
In short, Rad51-cataylzed D-loops were made before the addition of Srs2. First, Rad51 (0.2 mM) was incubated with various 32 P-labeled ssDNA substrates (1 nM molecule, 0.61 mM in nucleotides, or 0.61-0.78 mM in nucleotides/base pairs) for 10 min at 30˚C in a D-loop buffer containing 35 mM Trisacetate pH 7.5, 100 mM NaCl, 7 mM magnesium-acetate, 2 mM ATP, 1 mM DTT, 0.25 mg/mL BSA, 20 mM phosphocreatine and 100 mg/mL phosphocreatine kinase. RPA (33 nM) was then added and incubated for another 10 min at 30˚C. Rad54 (84 nM) and supercoiled plasmid dsDNA (7 nM molecule, 21 mM bp) were added and incubated for another 10 min at 30˚C. Indicated amounts of Srs2 were added and incubated at 30˚C. Equal volume of enzyme mixture with different amount of Srs2 and storage buffer was added to the reaction to deliver the final concentration of Srs2 as indicated. Samples were taken out at indicated time points, deproteinized, and separated on 1% agarose gels. Gels were dried and analyzed by a Storm phosphorimager. All the bands were quantified through densitometry using ImageQuant.

Protein-free D-loop disruption assay
Rad51/Rad54-catalyzed D-loops were prepared using 607 substrate as described above and further purified with a G25 spin column to remove SDS and protease K. The yield was about 78% as determined by densitometry. The protein-free D-loop (0.78 nM 607 substrate in total) was incubated with 15 nM Srs2 or buffer control in the same D-loop reaction buffer containing 35 mM Tris-acetate pH 7.5, 100 mM NaCl, 7 mM magnesium-acetate, 2 mM ATP, 1 mM DTT, 0.25 mg/mL BSA, 20 mM phosphocreatine and 100 mg/mL phosphocreatine kinase. Samples were taken and analyzed as describe above.

Extended D-loop disruption assay
Concentrations of different proteins and incubating temperature are indicated in figures and texts. In general, Rad51 (0.2 mM) was incubated with 32 P-labeled 100mer substrate (6.1 nM molecule, 0.61 mM in nucleotides) for 10 min at 30˚C in a D-loop buffer containing 35 mM Tris-acetate pH 7.5, 100 mM NaCl, 7 mM magnesium-acetate, 2 mM ATP, 1 mM DTT, 0.25 mg/mL BSA, 20 mM phosphocreatine, 100 mg/mL phosphocreatine kinase, and 100 mM each dATP, dGTP, dCTP, and dTTP. RPA (165 nM) was then added and incubated for another 10 min at 30˚C. Rad54 (84 nM) and supercoiled plasmid dsDNA (7 nM molecule, 21 mM bp) were added and incubated for another 2 min at either 30˚C or 25˚C, as specified. RFC (20 nM) and PCNA (20 nM trimeric form) were added and incubated for another 2 min at either 30˚C or 25˚C, as specified. Pol d (7 nM) was added and incubated for another 6 or 15 min at either 30˚C or 25˚C, as specified. Srs2 (0, 2.5, 5, 7.5 or 25 nM) was added and incubated at either 30˚C or 25˚C, as specified. For a-32 P-dCTP incorporation, 100 mM each of dATP, dGTP, dTTP, and only 0.22 mM of dCTP, spiked with a-32 P-dCTP, were included in the reaction buffer instead. Samples were taken out at the indicated time points, deproteinized, and separated on 1% agarose gels. Gels were dried and analyzed by a Storm phosphorimager. All the bands were quantified through densitometry using ImageQuant.

MMS and UV sensitivity assays
Serial dilution survival assay was carried out as published (Liu et al., 2011). Indicated strains (Table 1) were used to inoculate 5 mL YPD cultures and grown overnight at 30˚C. Cultures were then diluted to OD 600 nm = 2.0 with six sequential 5-fold dilutions. 2 ml of each culture dilution was plated on YPD with or without MMS (0.02%, 0.04%) or UV exposure (1 J/m 2 ) and incubated for 2 days at 30˚C before imaging.

Determination of damage induced recombination frequencies
The assay was carried out exactly as described before (Ede et al., 2011). Briefly, overnight cultures of different strains ( Table 1) were diluted into 6 ml of 1 Â 10 7 cells/mL and agitated at 30˚C for 1.5 hr before the addition of a-factor to a final concentration of 10 mg/mL. After 1.5 hr, more a-factor was added into the culture to reach a final concentration of 13.3 mg/mL and incubated for one more hour before harvesting. Cells were resuspended and incubated with SC-ura media containing 0, 0.04, 0.08, 0.16, 0.24 mg/mL 4-Nitroquinine-N-oxide (4-NQO) for 105 min at 30˚C with agitation. By this time cells have traversed S-phase and were collected, washed, and plated onto YPD to determine viable titer and G418 plates to determine the G418 R recombinants. Spontaneous rates were measured as before (Ede et al., 2011) using the method of the median (Lea and Coulson, 1949).