Homologous recombination repair intermediates promote efficient de novo telomere addition at DNA double-strand breaks

Abstract The healing of broken chromosomes by de novo telomere addition, while a normal developmental process in some organisms, has the potential to cause extensive loss of heterozygosity, genetic disease, or cell death. However, it is unclear how de novo telomere addition (dnTA) is regulated at DNA double-strand breaks (DSBs). Here, using a non-essential minichromosome in fission yeast, we identify roles for the HR factors Rqh1 helicase, in concert with Rad55, in suppressing dnTA at or near a DSB. We find the frequency of dnTA in rqh1Δ rad55Δ cells is reduced following loss of Exo1, Swi5 or Rad51. Strikingly, in the absence of the distal homologous chromosome arm dnTA is further increased, with nearly half of the breaks being healed in rqh1Δ rad55Δ or rqh1Δ exo1Δ cells. These findings provide new insights into the genetic context of highly efficient dnTA within HR intermediates, and how such events are normally suppressed to maintain genome stability.


INTRODUCTION
DNA double-strand breaks (DSBs) are potentially lethal lesions if unrepaired, and their misrepair can give rise to chromosomal rearrangements, a hallmark of cancer cells (1,2). To maintain both viability and genome stability in response to such lesions cells have evolved two types of DSB repair pathways: non-homologous end joining (NHEJ) and homologous recombination (HR). During classic nonhomologous end joining (C-NHEJ), the broken ends are bound by the Ku70/Ku80 heterodimer, and following the removal of damaged bases, are ligated together through the activity of Ligase 4 (Lig4) (reviewed in (3)). During HR repair, homologous sequences within a chromatid or chromosome are used as a template for accurate repair. HR repair is initiated by nucleolytic resection of the 5 end to generate a 3 single-stranded DNA (ssDNA) overhang. Resection is a two-step process, which is initiated by the MRN complex (comprised of Mre11-Rad50-Nbs1 in Schizosaccharomyces pombe (Sp) and in Homo sapiens (Hs)), and CtIP resulting in partly resected intermediates. During the second step, Exo1 together with Rqh1 Sp (BLM Hs ) facilitates extensive resection (4-6) (reviewed in (7)). The 3 ssDNA overhang is bound by Replication Protein A (RPA), which facilitates binding of the mediator Rad52 ,Sp and removal of secondary structures (8,9). Rad52 Sp , together with the auxiliary heterodimers Rad55 Sp -Rad57 Sp or Swi5 Sp -Sfr1 Sp mediate the loading of the RecA homologue, Rad51 Sp onto the ssDNA overhang to create a Rad51 nucleoprotein filament. This structure facilitates a homology search and strand exchange between the broken end and the homologous sequence to form a displacement-loop (D-loop) structure (10)(11)(12)(13). Following DNA synthesis the invading strand can be expelled by BLM and RecQL5 in mammalian cells, thus facilitating synthesis-dependent strand annealing (SDSA). Alternatively, second-end capture and ligation can result in a double-Holliday junction structure, which can be resolved with or without crossovers, a process involving Yen1, Mus81 Sp -Eme1 Sp , or dissolved through the activities of BLM-Top3 (reviewed in (14)).
Consistent with multiple roles in HR-dependent DSB repair, the RecQ family of DNA helicases plays a key role in maintaining genome stability in all organisms (15). A hallmark of BLM mutations in human cells is increased levels of sister chromatid and inter-homolog exchanges (16). In fission yeast, loss of the BLM orthologue, Rqh1, results in increased genome instability and sensitivity to DNA damaging agents (17,18). Rqh1 is an ATP-dependent 3 to 5 he-licase, in which the N-terminus interacts with Top3 (19,20). Rqh1 has been implicated in a variety of processes including HR, both before and after Rad51 filament formation (19,21,22); suppressing mitotic crossovers and promoting meiotic crossovers (23)(24)(25); suppressing inappropriate recombination following S phase arrest (17,18); facilitating the repair of collapsed replication forks (26)(27)(28); intra-S checkpoint function (29); efficient chromosome segregation (30) and telomere maintenance (31,32). A role for Rqh1 has also been identified in regulating HR-dependent Alternative Lengthening of Telomeres (ALT) pathway in the absence of Taz1 in fission yeast (33).
While normally repaired by the NHEJ or HR pathways, broken chromosome ends can sometimes be 'healed' as a result of telomeric capture or de novo telomere addition (dnTA). While dnTA is part of the normal developmental process in unicellular ciliates, chromosome healing in mammalian cells is associated with terminal deletions and genetic disease (34,35). Indeed, chromosome healing of a break within the body of a chromosome would be expected to result in extensive loss of heterozygosity (LOH) or potentially cell death through loss of genetic material centromeredistal to the break site. Accordingly, dnTA is not normally observed in response to ionizing radiation (IR) or enzymeinduced DSBs in yeasts or mammals (36)(37)(38), and may reflect the absence of telomeric seed sequences close to the break site, low levels or inhibition of telomerase, or competition with DSB repair pathways (39)(40)(41).
Here, we have investigated the relationship between DSB repair and loss of heterozygosity arising through dnTA. By introducing a site-specific DSB into a non-essential minichromosome, Ch 16 , we have uncovered a critical role for Rqh1 helicase, together with Rad55 in suppressing chromosome healing through dnTA at break sites. Further analysis suggests that stabilized HR intermediates are efficient substrates for dnTA.

Yeast strains, media and genetic methods
All S. pombe strains were cultured, manipulated, and stored as previously described (42). A list of strain genotypes can be found in Supplementary Table S1.

DSB assay
The DSB assay using Ch 16 -MGH was performed as previously described (41,43). The minichromosome (Ch 16 ) is a mitotically and meiotically stable 530 kb chromosomal element derived from ChIII (44). The DSB assay was performed at 25 • C for strains containing the cold-sensitive mutant pfh1-R20 cs (45) and appropriate comparison strains as indicated in the tables. The colony percentage undergoing NHEJ/SCR (ade + G418 R his + ), GC (ade + G418 S his + ), Ch 16 -MGH loss (ade − G418 S his − ), or LOH (ade + G418 S his − ) were calculated. LOH in this context refers to events which retain the ade + marker but have lost the G418 S marker. It is not possible to distinguish genetically between minichromosome loss and other rearrangements resulting in ade − G418 S his − colonies, such as isochromomosome formation, using Ch 16 -MGH so this population is collectively termed here 'Ch 16 loss'. Each experiment was performed three times using independently derived strains for all mutants tested. More than 1000 colonies were scored for each time point. Mean ± SEM values were obtained from triplicate strains. Differences were deemed significant if Pvalues obtained using Student's t test were ≤0.05.

Pulsed field gel electrophoresis (PFGE)
The PFGE protocols used in this study have been previously described (42). For higher resolution separation of Ch 16 -MGH, a 1.2% chromosomal grade agarose gel was used under the following conditions: 4 V/cm 112 • angle with a switch time of 1 min. Samples were separated for 48 h in 1× Tris-acetate-EDTA at 14 • C.

PCR assay for de novo telomere addition
Up to 20 randomly chosen ade + G418 s his − (LOH) colonies from each genetic background indicated were screened for telomeric sequence distal to the MATa break site as described. PCR amplification with primers targeted to the rad21 gene (5 -GATTTAAACCTGGATTTGGGC-3 ) and telomeric repeats (5 -CTGTAACCGTGTAACCGTAAC-3 ) was performed, followed by digestion with MfeI, yielding a distinct 300 bp band in telomere-positive strains.

Gene targeting
Plasmid pJK148 (48) was linearized with NdeI restriction, and transformed into the strains indicated the using Lithium Acetate protocol (47), and the number of Leu + transformants determined for each strain. The gene targeting efficiency was adjusted according to the relative transformation efficiencies of each strain, as determined using a circular pREP81X (49) as a control.

Rqh1 suppresses loss of heterozygosity in rad55Δ
To investigate the role of Rqh1 in genome stability, we examined the relationship between Rqh1 and other DNA recombination genes in the cellular response to DSBs. We found that deletion of rqh1 + in a rad55Δ background  Table 1. resulted in a significant increase in the IR-sensitivity of rad55Δ ( Figure 1A). To investigate this further, we examined the relationship between rad55 and rqh1Δ deletion mutants using a site-specific DSB assay. Using this assay, different repair and misrepair events can be quantitated by determining genetic marker loss following HO endonuclease induction of a site-specific DSB at the MATa site inserted within a non-essential minichromosome, Ch 16 -MGH, derived from chromosome III ( Figure 1B). Ch 16 -MGH carries an ade6-M216 heteroallele which complements the ade6-M210 heteroallele on ChIII to confer an ade + phenotype through intragenic complementation (50). Following HO endonuclease expression from a thiaminerepressible nmt promoter (rep81X-HO) DSB induction can result in a variety of outcomes: DSB repair through NHEJ or sister chromatid recombination (SCR) in which a broken chromatid uses its unbroken sister chromatid as a repair template; failed DSB repair with loss of the minichromosome; gene conversion using the homology of chromosome III; extensive loss of heterozygosity, resulting through break-induced non-reciprocal translocations or partial loss of heterozygosity ( Figure 1B) (41).
Surprisingly, HO endonuclease-induced cleavage at the MATa site in an rqh1Δ rad55Δ double mutant resulted in a striking increase in LOH (27.3%, P < 0.001) compared to wild type (1.7%). This increase in LOH was associated with significantly increased NHEJ/SCR (46.1%, P < 0.001) and significantly reduced GC (3.3%, P < 0.001) compared to wild type, while Ch 16 loss (18.6%) was similar to both single mutant and wild-type backgrounds ( Figure 1C: Table 1). These findings indicate that Rqh1 suppresses LOH in a rad55Δ background. No loss of viability was observed in these strains following DSB induction (Supplementary Figure S1).

Rqh1 suppresses de novo telomere addition in a rad55Δ background
To identify the mechanism of break-induced LOH, the chromosomes of 21 LOH colonies from an rqh1Δ rad55Δ background were examined by pulsed-field gel electrophoresis (PFGE). While endogenous chromosomes I and II derived from these LOH colonies remained unchanged, crossovers were sometimes observed (9.5% of LOH colonies) between ChIII and the homologous minichromosome, Ch 16 (Figure 2A, lane 4). Importantly, minichromosomes from the remaining 90.5% of the LOH colonies appeared truncated, as determined by highresolution separation of chromosomal DNA by PFGE ( Figure 2B). As break-induced LOH retained the ade6 marker ∼25kb centromere proximal to the break site (Figure 1A), this raised the possibility that Ch 16 truncations resulted from dnTA, as was previously observed in a rad55Δ background (41). This possibility was examined by colony PCR amplification using primers annealing to rad21 + (centromere proximal to the MATa break site) and a telomere specific primer. A PCR product of 300 bp following digestion with MfeI (a restriction site just upstream of the MATa site) was scored positive for dnTA. Sequence analysis of the PCR products indicated the presence of ∼300 bp of G 2-5 TTACA 0-1 repeats, consistent with dnTA at, or very close to, the break site in 13 of the LOH colonies tested ( Figure 2C). In 6 of the remaining colonies, telomeres were added ∼9-19 kb centromere proximal to the break site. In full, 24.7% of colonies underwent dnTA in rqh1Δ rad55Δ strains, which equated to a 1450-fold increase in dnTA compared to wild type (0.017%) ( Figure 2D, Table 1).

Suppression of de novo telomere addition requires Rqh1 helicase activity
To determine whether Rqh1 required its helicase activity to suppress dnTA, we introduced a helicase-dead mutation rqh1-K547A (19,51) into a rad55Δ background. In this strain, levels of break-induced LOH and dnTA resembled those observed in the rqh1Δ rad55Δ strain ( Figure 3A, Table 1) suggesting Rqh1 helicase activity is required to suppress dnTA in a rad55Δ background.
Rqh1 has been shown to function in a complex with Top3 (19,20,52). As the top3Δ strain is non-viable (19,52), we tested the requirement of the Top3 interaction in suppressing dnTA using an rqh1ΔN1-322 mutant which has lost the Top3 binding domain (20) and is expressed at similar levels to the wild-type Rqh1 (20). DSB-induced LOH in the rad55Δrqh1ΔN1-322 mutant was significantly higher (11.9%, P<0.001) than that observed in rad55Δ (1.8%), but less than the rqh1Δ rad55Δ strain (27.3%; Figure 3A; Table 1). This effect could be attributed to a requirement for Rqh1-Top3 interaction in suppressing break-induced LOH in a rad55Δ background or to partial loss of Rqh1 helicase activity in the rqh1ΔN1-322 mutant.
To determine whether other helicases could function similarly to Rqh1, we introduced a deletion of srs2 + or a cold-sensitive allele of pfh1 + , pfh1-R20 (pfh1 cs ) (45) into a rad55Δ background and examined levels of dnTA. Srs2 is implicated in regulation of HR where it antagonizes the activity of the Rad55-Rad57 heterodimer (53)(54)(55). In contrast to the rqh1Δ mutant, the srs2Δ mutant failed to significantly increase levels of break-induced LOH in a rad55Δ background ( Figure 3A; Table 1). The S. cerevisiae Pfh1 homologue, Pif1 has been identified as a suppressor of dnTA and gross chromosomal rearrangements (56,57). The rad55Δ pfh1 cs strain showed a modest yet significant increase in LOH (6.4%) compared to wild-type background at semi-permissive temperature (0.8%, P = 0.013), and a significant increase in comparison to rad55Δ LOH levels (P = 0.027), at semi-permissive temperature ( Figure 3A; Table 1). Therefore, Pfh1 can suppress LOH arising predominantly from dnTA, in accordance with the described role in S. cerevisiae. However, in our assay, Rqh1 helicase clearly plays a more prominent role in suppressing dnTA than Srs2 or Pfh1.

Rqh1 functions with early acting HR proteins to suppress dnTA
Next we wished to examine the potential role of other HR factors in suppressing dnTA in an rqh1Δ background. As Exo1 functions early in HR during DSB resection, we examined the relationship between Rqh1 and Exo1 (5,6,58). Deletion of exo1 + did not significantly alter levels of breakinduced LOH compared to wild type (41). However, a striking increase in levels of break-induced LOH (20.7%) was 26. observed in an rqh1 exo1 double mutant, 60% (12 of 20 examined colonies) of which was due to dnTA ( Figure  3B, Table 1). Break-induced marker loss after deletion of exo1 + in an rqh1Δ or rqh1Δrad55Δ background was also determined ( Table 1). Break-induced LOH was significantly reduced in the rqh1Δ rad55Δ exo1Δ mutant compared to the rqh1Δ rad55Δ mutant (P<0.001) and no dnTA products were obtained upon further analysis (Table 1). This requirement for exo1 + in facilitating dnTA in the rqh1Δ rad55Δ background is in accordance with Exo1-dependent endresection facilitating dnTA, as previously proposed (41). We were unable to test the role of Rad52 in suppressing dnTA as the rqh1Δrad52Δ strain was extremely sick, consistent with previously reported findings (19). As Rad57 forms a heterodimer with Rad55 (59), we examined gene marker loss in a rad57Δ background. The resultant marker loss profile was similar to that of rad55 strains (Table 1). Following DSB induction in an rqh1Δ rad57Δ background, 7.7% of colonies exhibited extensive LOH (P = 0.011 compared to a rad57Δ single mutant), out of which 50% of the double mutant had undergone dnTA (Table 1). Thus, Rqh1 can suppress LOH in a rad57Δ background, albeit not to the same extent as in a rad55Δ background.
DSB induction within Ch 16 in a rad51Δ background has previously been shown to result in a higher proportion of minichromosome loss, demonstrating a failure to repair the DSB (41,42). DSB induction in a rqh1Δ rad51Δ background resulted in reduced levels (0.39%) of LOH colonies compared to wild type (1.7%; P = 0.007) and rqh1Δ (1.5%; P = 0.051), indicating that, in contrast to a rad55Δ back-ground, Rqh1 does not suppress break-induced LOH significantly in a rad51 background ( Figure 3B; Table 1). Instead, a significant increase in NHEJ/SCR (76.8%) was observed in an rqh1Δ rad51Δ background compared to that of rad51 (35.9%, (41)), indicating that DSBs in an rqh1 rad51 double mutant are still competent for HRindependent repair, even though HR is severely impaired. These observations are consistent with an early role for Rqh1 in HR, as described for Sgs1 and BLM in budding yeast and human cells, respectively (4-6,58).

A critical role for Rad51 loading in facilitating efficient dnTA
Following DSB induction, dnTA may occur before or after Rad51-dependent strand invasion. If dnTA resulted immediately following resection, then this event should be Rad51-independent. In an rqh1Δ rad55Δ rad51 triple mu- Graph depicting analysis of dnTA and GC levels following over-expression of either an empty vector (pIRT3) or Rad51 (pIRT3-rad51) in an rqh1Δ rad55Δ background (Table 2).
We also tested the rad55Δ swi5Δ double mutant. Marker loss in a rad55Δ swi5Δ strain was very similar to that in a rad51 strain, resulting in high levels of Ch 16 loss (66%), consistent with failed Rad51 loading. Levels of NHEJ/SCR colonies were also reduced in the rad55Δ swi5Δ background (30%) compared to rad55Δ (62%) or swi5Δ (57%) single mutants, consistent with this population arising through HR-dependent SCR. Further, levels of LOH through dnTA (1.9%) in rad55Δ swi5Δ strain were equivalent to that of rad55Δ strain alone (1.4%: P = 0.743: Table 2).
We have previously demonstrated that Rad51 overexpression (OPrad51) reduced levels of dnTA in a rad55Δ background (41), consistent with competition between the Rad51 recombinase and telomerase for resected ends. We therefore tested whether Rad51 overexpression could similarly reduce levels of dnTA observed in an rqh1Δ rad55Δ background by introducing pIRT3-rad51 (60). OPrad51 resulted in significantly increased levels of GC (3.3%, P = 0.05), and SCR (88.3%, P = 0.03), and significantly reduced levels of Ch16 loss (3.9%, P = 0.04) and LOH (4.45%, P = 0.04), and therefore reduced levels of dnTA, in an rqh1Δ rad55Δ background compared to vector alone ( Figure 3D; Table 2). However, whilst rad55Δ, rqh1Δ and rad55Δ rqh1Δ are exquisitely sensitive to radiation in a rad51 background (Supplementary Figure S2), overexpression of Rad51 does not significantly rescue radiation sensitivity in these mutants (Supplementary Figure S3). Together, these data identify a critical role for Rad51 recombinase levels in facilitating dnTA in an rqh1Δ rad55Δ background. The mean ± SE from at least three independent experiments with three individual strains are shown. % dnTA was calculated by multiplying the fraction of dnTA positive colonies identified from the 20 ade+ G418S/HygS colonies examined (indicated in brackets) by the % LOH. The P-value for rad55 rqh1+ pIRT3-rad51 is 0.042 relative to rad55 rqh1+ pIRT3. Wt values presented as previously described (Cullen et al., 2007) The MRN complex promotes SCR and partially suppresses dnTA in a rad55Δ background We wished to examine whether deletion of other early acting HR genes could suppress dnTA in a rad55Δ background. We previously reported that deleting exo1 + in a rad55Δ background resulted in reduced levels of dnTA compared to rad55Δ strains (41). Here we further tested the effect of deleting the MRN complex in a rad55Δ background. Deletion of mre11 + , rad50 + or nbs1 + in a rad55Δ background resulted in a striking reduction in levels of NHEJ/SCR colonies and increased levels of Ch 16 loss compared to rad55Δ (Table 3). These results resemble those of another HR mutant, rad55Δ rad51Δ (42), and are consistent with the break-induced ade + G418 R his + population in a rad55Δ background resulting from HR-dependent SCR following break-induction. Levels of LOH colonies arising through dnTA were modestly increased in rad55Δ mre11Δ (4.8%), rad55Δ rad50Δ (4.5%), and rad55Δ nbs1Δ (2.4%) compared to rad55Δ (1.4%) or the respective individual MRN deletion mutants (Table 3). Thus, while the MRN complex is important for SCR, it performs only a minor role in suppressing dnTA in a rad55Δ background compared to Rqh1. Thus, Rqh1 plays a functionally distinct role from Exo1 and the MRN complex in suppressing dnTA in rad55Δ strains.

Striking levels of dnTA at a DSB lacking a homologous distal chromosome arm
The above findings indicate that efficient dnTA is associated with HR intermediates. To test this further, we asked whether dnTA would be further increased under circumstances in which post-synaptic second-end capture was abrogated. To address this, the (130 kb) homologous arm centromere-distal to the break site was replaced by a construct containing the 1.8 kb MATa target sequence/G418resistant marker and a 1 kb synthetic telomere, TASTel fragment containing 700 bp of subtelomeric DNA (TAS) and 300 bp of telomeric DNA (Tel) ( Figure 4A) (62). We found DSB induction in Ch 16 -MGTASTel in a wildtype background resulted in 79.3% of the colonies becoming ade − G418 S , consistent with very high levels of unrepaired breaks leading to chromosome loss or other undetectable rearrangements; 17.4% remained ade + G418 R , consistent with NHEJ or SCR; and 3.3% became ade + G418 S , having undergone LOH (Table 4). Further PCR analysis of 20 individually isolated ade + G418 S colonies failed to detect dnTA ( Figure 4B; Table 4). Deletion of rqh1 + , exo1 + or rad55 + each resulted in increased levels of NHEJ/SCR and reduced Ch 16 loss, as was observed in Ch 16 -MGH. This was associated with modest increases in LOH and dnTA with 13% dnTA noted in a rad55Δ background ( Figure 4B; Table 4). Remarkably, DSB induction in an rqh1Δ rad55Δ background resulted in 53% of the colonies becoming ade + G418 S , corresponding to 45.2% dnTA ( Figure 4B; Table  4). Similarly, following DSB induction in an rqh1Δ exo1Δ background, 51% of the colonies became ade + G418 S which corresponded to 41.4% dnTA.
To test whether the increased levels of dnTA resulted from loss of the second homologous chromosome arm, or from proximity to the TASTel synthetic telomere sequence, an additional minichromosome was constructed in which the TASTel sequence was integrated distal to the MATa site of Ch 16 -MG, (in the same locus as Ch 16 -MGH), but retaining the distal arm of the minichromosome, to form Ch 16 -MG(TASTel)Ch ( Figure 4C). Surprisingly, DSB induction in a wild-type strain containing Ch 16 -MG(TASTel)Ch resulted in 76% Ch 16 loss or extensive LOH; while 2% of the colonies underwent LOH or GC, and dnTA was not detected (Table 5). Although we cannot distinguish LOH or GC colonies, the levels of ade + G418 S colonies (combining LOH and GC) in Ch 16 (Table 1). (C) Schematic of the Ch 16 -MGTASTelCh minichromosome. Minichromosome whose features are described in (A); ChIII as described in Figure 1B. (D) Histogram of percentage break-induced LOH arising through dnTA (gray) or other (white) in wild-type (TH8597), rqh1Δ exo1Δ double mutant (TH8598) and rqh1Δ rad55Δ double mutant (TH8708) strains following HO-endonuclease induction for 48 h (Table 5).  The mean ± SE from at least three independent experiments with three individual strains are shown. % dnTA was calculated by multiplying the fraction of dnTA positive colonies identified from the 20 ade+ G418S colonies examined (indicated in brackets) by the % LOH.
reduced GC observed in Ch 16 -MG(TASTel)Ch strain presumably reflects the reduced homology with ChIII due to the addition of the non-homologous TASTel cassette. Following DSB induction in an rqh1Δ rad55Δ background, 33% of colonies were ade + G418 S , corresponding to 27% dnTA ( Figure 4D; Table 5). This was again significantly reduced compared to 45.2% dnTA (P = 0.0142) observed using Ch 16 -MGTASTel, but was very similar to levels observed using Ch 16 -MGH (25%). Following DSB induction within an rqh1Δ exo1Δ background, in which GC was abrogated, 31% of the colonies were ade + G418 S , corresponding to 26% dnTA ( Figure 4D; Table 5). This was significantly reduced compared to that observed using Ch 16 -MGTASTel (41.4% P = 0.04), but was greater than levels observed using Ch 16 -MGH (12% P>0.05). Thus, dnTA was significantly further increased in the context of a 'one-armed' break in either rqh1Δ rad55Δ or rqh1Δ exo1Δ backgrounds. While in the rqh1Δ exo1Δ background, proximity of the DSB site to telomeric sequence could contribute to high dnTA levels, the striking dnTA levels observed in an rqh1Δ rad55Δ background are consistent with disrupting post-synaptic HR events.

DISCUSSION
Here we describe roles for the BLM homologue, Rqh1 helicase and the Rad51 paralog, Rad55, in both facilitating homologous recombination and in suppressing chromosome healing at a break site in fission yeast. We find Rqh1 helicase, together with either Rad55 or Exo1 suppresses dnTA. Further, we find that a DSB lacking a homologous distal chromosome arm undergoes highly efficient dnTA in these genetic backgrounds. Together these findings indicate that chromosome healing can occur highly efficiently within HR intermediates. Here we consider the mechanisms by which these events occur in the absence of these genes and the implications for genome stability.

Suppressing chromosome healing
Our study identifies an independent role for the HR proteins Rqh1, together with Rad55 or Exo1 in suppressing dnTA, with a striking increase in dnTA being observed in rqh1Δ rad55Δ and to a lesser extent rqh1Δ exo1Δ backgrounds, compared to wild type. As extensive resection requires both Rqh1 and Exo1 (4-6), these findings are consistent with partially resected ends acting as efficient substrates for dnTA (37,38). Loss of both Rad55 and Rqh1 may also facilitate presynaptic dnTA either through reduced resection or through altering the structure of the Rad51 nucleofilament so that it is more conducive to dnTA ( Figure 5) (63)(64)(65). That overexpression of Rad51 in an rqh1Δ rad55Δ background led to significantly reduced levels of dnTA and significantly elevated levels of both GC and SCR suggests that Rad55 suppresses dnTA through facilitating efficient Rad51 assembly. These findings are broadly consistent with a role for HR in preventing dnTA through competition for resected ends (41). However, RecQ helicase activity is also required for branch migration, to displace non-allelic recombination, and to prevent the formation of double-Holliday junctions (15), and loss of these post-synaptic functions may also  facilitate dnTA. Here, loss of Rqh1 helicase activity may lead to stabilization of the invading strand in a D-loop, which may facilitate dnTA ( Figure 5). Consistent with this, dnTA was occasionally associated with crossover events between Ch 16 and ChIII in rqh1Δ rad55Δ ( Figure 3A). Loss of Rad55 may also contribute to post-synaptic dnTA by promoting Srs2-dependent exclusion of the non-homologous MATa site from the D-loop, which may now act as a landing pad for telomerase. Accordingly, postsynaptic roles have also been assigned to the Rad51 paralogues, which are the human counterparts to Rad55 and Rad57 (66,67). In this respect, the Rad51L3-XRCC2 complex physically interacts with and stimulates BLM to disrupt Holliday junctions in vitro (67). The observation that a DSB lacking a homologous distal chromosome arm significantly further increased dnTA levels in rqh1Δ rad55Δ or rqh1Δ exo1Δ backgrounds is consistent with a model in which the post-synaptic HR events of second end capture or strand annealing compete with dnTA ( Figure 5).

Determinants of chromosome healing
We found Rad51 to be required for efficient dnTA. This was unexpected as efficient Rad51 loading suppresses dnTA. Accordingly, dnTA levels were significantly reduced in a rad51 rqh1 rad55Δ strain compared to an rqh1Δ rad55Δ background. Similarly, preventing Rad51 loading by simultaneously disrupting both Rad55-Rad57 and Swi5-Sfr1 heterodimers significantly reduced dnTA in swi5Δ rad55Δ rqh1-K547A compared to a rad55Δ rqh1-K547A background. Taken together, our data are consistent with the hypothesis that Rad55-Rad57 and Swi5-Sfr1 have distinct roles in Rad51 assembly (68). It has been shown that Rad51-foci form less efficiently in Swi5-Sfr1 compared to a Rad55-Rad57 mutant (11) whereas Rad55-Rad57 subtly organizes the Rad51-nucleofilament structure (55). Therefore, Swi5-Srf1 is required to stabilize Rad51, thus promoting dnTA, whereas Rad55-Rad57 is required to modulate Rad51 structure, which promotes GC and suppresses dnTA. As Rad51 over expression suppressed dnTA and promoted GC in an rqh1Δ rad55Δ background, this suggests the Rad51 nucleofilament structure is likely to play a critical role in defining the fate of broken chromosome ends. Rad51 may potentially promote dnTA presynaptically in this respect, through assembling a nucleofilament structure to which telomerase may preferentially bind in the absence of Rad55 and Rqh1. Rad51 may also promote dnTA postsynaptically, through facilitating D-loop formation, which in the absence of Rqh1 or Rad55 facilitates dnTA at nonhomologous ends extruded from the D-loop ( Figure 5). As D-loops are structurally analogous to T-loops (69), our findings suggest a structural context through which T-loops may promote telomerase activity.
We found Exo1 to be required for dnTA in an rqh1Δ rad55Δ background. As efficient dnTA was observed in an rqh1Δ exo1Δ background this indicates that Exo1 is not required for telomere addition. However, in S. cerevisiae Sae2/MRX and Sgs1 activities are required to allow Exo1 access, which contributes to telomere end processing and elongation (70). We speculate that further loss of Exo1-dependent resection in an rqh1Δ rad55Δ background fails to generate sufficient ssDNA necessary to facilitate dnTA. Such resection may facilitate Rad51 binding as indicated above. Reduced dnTA is associated with increased NHEJ/SCR following Exo1 deletion in an rqh1Δ rad55Δ background. However, further studies will be required to elucidate the precise role of Exo1 in this context.

Mechanisms of telomere addition
The finding that efficient dnTA was observed at or near the MATa site was unexpected, as this region lacks the canonical GGTTACA S. pombe telomeric repeat sequence (61). Studies in S. cerevisiae have shown dnTA was restricted to very short regions of homology to the telomerase guide RNA that were likely to facilitate annealing of such RNA (71,72). Thus, efficient dnTA observed in our study may result from recognition of degenerate telomeric sequences by guide RNA or other telomere recruitment factors. Alternatively, telomere recruitment may be achieved in a sequenceindependent manner through interaction between ssDNA binding factors and telomerase (73).

Chromosome healing and genome instability
Our findings indicate that dnTA has the capacity to stabilize broken chromosomes. However, such a role comes at the price of potentially extensive loss of genetic material centromere-distal to the break site. While dnTA is predicted to result in loss of viability in a haploid setting, such extensive LOH in a diploid or polyploidy cells may be tolerated. Thus, dnTA may provide an important back-up mechanism to rescue broken chromosomes, thus facilitating cell survival.