Rad51 recruitment and exclusion of non-homologous end joining during homologous recombination at a Tus/Ter mammalian replication fork barrier

Classical non-homologous end joining (C-NHEJ) and homologous recombination (HR) compete to repair mammalian chromosomal double strand breaks (DSBs). However, C-NHEJ has no impact on HR induced by DNA nicking enzymes. In this case, the replication fork is thought to convert the DNA nick into a one-ended DSB, which lacks a readily available partner for C-NHEJ. Whether C-NHEJ competes with HR at a non-enzymatic mammalian replication fork barrier (RFB) remains unknown. We previously showed that conservative “short tract” gene conversion (STGC) induced by a chromosomal Tus/Ter RFB is a product of bidirectional replication fork stalling. This finding raises the possibility that Tus/Ter-induced STGC proceeds via a two-ended DSB intermediate. If so, Tus/Ter-induced STGC might be subject to competition by C-NHEJ. However, in contrast to the DSB response, where genetic ablation of C-NHEJ stimulates HR, we report here that Tus/Ter-induced HR is unaffected by deletion of either of two C-NHEJ genes, Xrcc4 or Ku70. These results show that Tus/Ter-induced HR does not entail the formation of a two-ended DSB to which C-NHEJ has competitive access. We found no evidence that the alternative end-joining factor, DNA polymerase θ, competes with Tus/Ter-induced HR. We used chromatin-immunoprecipitation to compare Rad51 recruitment to a Tus/Ter RFB and to a neighboring site-specific DSB. Rad51 accumulation at Tus/Ter was more intense and more sustained than at a DSB. In contrast to the DSB response, Rad51 accumulation at Tus/Ter was restricted to within a few hundred base pairs of the RFB. Taken together, these findings suggest that the major DNA structures that bind Rad51 at a Tus/Ter RFB are not conventional DSBs. We propose that Rad51 acts as an “early responder” at stalled forks, binding single stranded daughter strand gaps on the arrested lagging strand, and that Rad51-mediated fork remodeling generates HR intermediates that are incapable of Ku binding and therefore invisible to the C-NHEJ machinery.


Introduction
The stalling of replication forks at sites of abnormal DNA structure, following collisions with transcription complexes or due to nucleotide pool depletion-collectively termed "replication stress"-is a significant contributor to genomic instability.Inherited mutations in genes that regulate the replication stress response cause a number of human diseases, ranging from developmental disorders to highly penetrant cancer predisposition syndromes [1][2][3][4][5].Replication stress is thought to be a near-universal phenomenon in tumorigenesis and some of the molecules that act upon the stalled fork are considered promising targets for cancer therapy [6].Replication fork stalling provokes a diverse set of cellular responses, including: stabilization of the stalled replisome; regulated replisome disassembly ("fork collapse"); protection of the fork from deleterious nucleolytic processing; remodeling of DNA structure at the stalled fork; and engagement of repair or "replication restart" [5,[7][8][9][10][11][12][13][14][15].The S phase checkpoint and the homologous recombination (HR) systems are intimately involved in coordinating these responses, collaborating to suppress deleterious genome rearrangements at the stalled fork [2,[16][17][18][19][20].However, the mechanisms governing this coordination remain poorly understood in mammalian cells.
DNA structure at the stalled fork is remodeled by topological stresses on the chromosome at the site of stalling and by the direct action of remodeling enzymes [5,12,21].The fork can be reversed to form a Holliday junction, generating a solitary DNA end which is extensively single stranded due to accompanying nascent lagging strand resection [20,22,23].Other forms of template switching can also occur in the vicinity of the stall site [18,24,25].Endonuclease-mediated fork breakage-either scheduled or unscheduled-can generate double strand breaks (DSBs), which might be either one-ended or two-ended [5,20].The DNA structures generated by fork remodeling presumably limit the repair pathways that can be engaged.Twoended DSBs can potentially be repaired by end joining mechanisms as well as by recombination [26,27].In contrast, a one-ended DSB or a solitary DNA end lacks a readily available ligation partner for end joining, and may preferentially engage break-induced replication [28,29].Consistent with this, HR induced by a two-ended chromosomal DSB is subject to competition by classical non-homologous end joining (C-NHEJ), whereas HR induced by a nicking enzyme ("nickase")-in which the replication fork converts the nick into a one-ended DSB-is unaffected by deletion of C-NHEJ genes [30][31][32].Thus, in mammalian cells, the susceptibility of HR to competition by C-NHEJ in a particular cellular context is a useful "probe" with which to analyze the DNA structural intermediates of HR.Since the stalled fork response entails the formation of diverse DNA structures and is not restricted to two-ended DSBs, repair pathway "choice" at a stalled fork may differ from that at a defined two-ended DSB.
Study of replication-coupled repair of a covalent DNA inter-strand crosslink (ICL) in Xenopus laevis egg extracts has revealed some of the fundamental steps of stalled fork processing and repair [2,20].The Fanconi anemia (FA)/BRCA pathway plays a key role in detecting and processing forks bidirectionally arrested at the ICL [33][34][35].The FANCD2/FANCI heterodimer orchestrates dual incisions of one of the sister chromatids on either side of the ICL.Importantly, efficient incision of the bidirectionally arrested forks is suppressed until the two opposing forks have each stalled at the ICL [36].The resulting two-ended DSB is repaired by HR-mediated sister chromatid recombination, in which the BRCA gene products play canonical roles in promoting Rad51 loading and strand exchange functions [37][38][39][40][41]. HR repair of such a two-ended DSB intermediate could, in principle, be subject to competition by C-NHEJ or other end joining systems.However, recent evidence of fork reversal during ICL repair suggests that at least one of the two DSB ends is extensively single stranded [23].
Competition between HR and C-NHEJ is not a major feature of DSB repair in yeast, since C-NHEJ is a relatively low-flux pathway.Additionally, the Fanconi anemia pathway in yeast is limited to evolutionarily conserved homologs of FANCM [42,43], suggesting that the innovation of FANCD2/FANCI-coordinated incision of bidirectionally arrested forks occurred relatively recently in evolution.Thus, although certain "core" elements of DSB repair and stalled fork metabolism are conserved between yeast and vertebrates, there are likely significant interspecies differences that remain to be fully defined.Studies in yeast, using non-enzymatic, locus-specific replication fork barriers (RFBs), show that stalled fork HR can mediate both conservative and deleterious repair, the latter including gross chromosomal rearrangements and more localized copy number changes at the site of stalling [14,18,19,24,[44][45][46][47][48].In contrast to the above-noted X. laevis ICL repair model, HR at an RTS1 RFB in Schizosaccharomyces pombe is not accompanied by evidence of DSB formation [19,24].Processing of the stalled fork in S. pombe may also trigger an aberrant form of "replication restart", a rad22 Rad52 -dependent process in which the restarted fork is prone to collapse [45].This aberrant fork restart mechanism is reminiscent of break-induced replication (BIR) in Saccharomyces cerevisiae, which is characteristically unstable and mutation-prone [49,50].Indeed, current models of aberrant replication restart in S. pombe invoke a migrating bubble mechanism equivalent to the mechanism of BIR in S. cerevisiae [49].Rad52-dependent pathways have also been implicated in stalled fork repair in mammalian cells [51][52][53].
To facilitate analysis of mammalian stalled fork metabolism and repair, we adapted the Escherichia coli Tus/Ter RFB for use in mammalian cells [54][55][56][57][58].A chromosomally integrated array of six 23 bp Ter sites mediates Tus-dependent, locus-specific replication fork stalling and HR on a mammalian chromosome, enabling direct quantitation of the repair products of mammalian replication fork stalling.We showed that conservative "short tract" gene conversion (STGC) at Tus/Ter is positively regulated by BRCA1, BRCA2, Rad51 and the Fanconi anemia pathway-consistent with the idea that STGC represents a physiological HR response to fork stalling [56,58].In contrast, "long tract" gene conversion (LTGC)-an error-prone HR outcome in which a replicative mechanism copies several kilobases from the partner sister chromatid-is suppressed by BRCA1 and appears to be Rad51-independent.We recently identified a novel product of stalled fork repair in primary mouse cells lacking the hereditary breast/ovarian cancer predisposition gene, Brca1: the formation of small (2-6 kb) non-homologous or microhomology-mediated tandem duplications (TDs) [58].Tus/Ter-induced TDs in Brca1 mutant cells are mediated by a replication restart-bypass mechanism, which is completed by Xrcc4-dependent C-NHEJ.This finding, together with previous observations, suggests that C-NHEJ can access DNA ends positioned close to the site of fork stalling [59,60].Notably, Tus/Ter-induced STGC is a product of bidirectional replication fork stalling [56].By analogy with the processing of forks bidirectionally arrested at an ICL in X. laevis, Tus/Terinduced STGC might entail the formation of a two-ended DSB intermediate and might therefore be subject to competition by C-NHEJ.To test this hypothesis, we have analyzed the impact of deletion of the C-NHEJ genes Xrcc4 and Ku70 on Tus/Ter-induced HR.

Impact of Xrcc4 deletion on Tus/Ter-induced HR
To determine whether C-NHEJ interacts with HR at Tus/Ter-stalled replication forks, we targeted a 6xTer-HR reporter as a single copy to the ROSA26 locus of mouse embryonic stem (mES) cells carrying biallelic conditional alleles of the C-NHEJ gene Xrcc4 (here termed "Xrcc4 fl/fl "), as described in Materials and Methods [56,61].The 6xTer-HR reporter contains an I-SceI target site adjacent to the 6xTer array (Fig 1A).Thus, transfection of Tus enables analysis of HR in the stalled fork response, while transfection of I-SceI in parallel samples enables analysis of DSB-induced HR.The reporter also contains elements to distinguish short tract gene conversions (STGC) from long tract gene conversions (LTGC), the latter being rare HR products in wild type cells [62,63].Although HR by either STGC or LTGC converts the cell to GFP + , LTGC additionally converts the cell to RFP + , by replicative duplication of an RFP cassette within the reporter (Fig 1A) [56,64].
We transduced a ROSA26-targeted Xrcc4 fl/fl 6xTer-HR reporter clone with adenovirallyencoded Cre recombinase and screened for derivative clones that had either lost (Xrcc4 Δ/Δ ) or retained (Xrcc4 fl/fl ) Xrcc4.Xrcc4 loss or retention was detected by PCR on genomic (g)DNA and was confirmed in a subset of clones by western blotting (Fig 1B and 1C).We studied HR in five independent Cre-treated Xrcc4 fl/fl 6xTer-HR reporter clones and five independent Cretreated Xrcc4 Δ/Δ 6xTer-HR reporter clones in response to either Tus or I-SceI-each transfected in parallel samples (see Materials and Methods).As expected, I-SceI-induced STGC and LTGC were elevated up to 4-fold in Xrcc4 Δ/Δ cells in comparison to Xrcc4 fl/fl cells (Fig 1D and  1E) [30].Interestingly, deletion of Xrcc4 stimulated STGC more strongly than LTGC; as a result, the proportion of I-SceI-induced HR events that resolved as LTGC was reduced from ~5% in Xrcc4 fl/fl cells to ~2-3% in Xrcc4 Δ/Δ cells (Fig 1E).The impact of Xrcc4 deletion on Tus/ Ter-induced HR was quite different.Tus/Ter-induced STGC was marginally reduced in Xrcc4 Δ/Δ cells in comparison to Xrcc4 fl/fl cells, while Tus/Ter-induced LTGC was unaffected by deletion of Xrcc4 (Fig 1D and 1E).These results suggest that the interaction between HR and C-NHEJ at a chromosomal DSB is not recapitulated in the regulation of HR at a stalled replication fork.
To determine whether the observed phenotypes are affected by re-expression of wtXrcc4, we used lentiviral transduction to express N-terminal influenza haemagglutinin (HA)-tagged wild type mouse (m)Xrcc4 in Xrcc4 Δ/Δ 6xTer-HR reporter clones #11 and #13 and in Xrcc4 fl/fl 6xTer-HR reporter clones #8 and #39.Briefly, we adapted the lentiviral vector pHIV-Zsgreen [65] by replacing the Zsgreen cDNA with a bicistronic cDNA encoding the enzyme nourseothricin (NTC) acetyl transferase (NAT) [66] fused via a self-cleaving T2A peptide to the human (h)CD52 antigen (S1A Fig) [67].Transient expression of the empty pHIV-NAT-CD52 vector in mouse ES cells produced strong cell surface staining of hCD52, as revealed by immunostaining using an anti hCD52-specific monoclonal antibody [68]  of mES cells with the empty pHIV-NAT-CD52 vector, followed by selection in NTC, generated pools of transduced cells that stained strongly and specifically with anti-hCD52, whereas transduction with pHIV-NAT (i.e., lacking hCD52 expression), followed by NTC selection, generated no CD52-specific cell surface signal (S1B Fig) .CD52 expression levels in pHIV-NAT-CD52-mXrcc4-transduced, NTC-selected mES cells were lower than in control empty vector (pHIV-NAT-CD52)-transduced controls, possibly reflecting constraints imposed by Xrcc4 expression from the multicistronic lentiviral expression cassette.Nonetheless, exogenous wtXrcc4 was overexpressed in comparison to endogenous Xrcc4, as revealed by RT-qPCR and by western blotting in lentivirally transduced Xrcc4 fl/fl cultures (Fig 2A and 2B).As expected, re-expression of wtXrcc4 complemented the sensitivity of Xrcc4 Δ/Δ cells to the radiomimetic drug phleomycin (Fig 2C).

(S1B Fig). Transduction
Xrcc4 Δ/Δ 6xTer-HR reporter cells transduced with pHIV-NAT-CD52-Xrcc4 and selected in NTC revealed suppression of I-SceI-induced HR to levels equivalent to that observed in isogenic Xrcc4 fl/fl 6xTer-HR reporter cells (Fig 2D).Indeed, I-SceI-induced STGC and LTGC were each restored to wild type levels and the ratio of LTGC:Total HR reverted from ~2% to ~4% in Xrcc4-transduced Xrcc4 Δ/Δ cells.Parallel cultures transduced with pHIV-NAT-CD52 empty vector and selected in NTC retained the original Xrcc4 Δ/Δ phenotype.These experiments confirm that Xrcc4 affects the balance between I-SceI-induced STGC and LTGC, suppressing STGC more strongly than LTGC.In contrast, all measures of Tus/Ter-induced HR were unaffected by re-expression of wtXrcc4 in Xrcc4 Δ/Δ cells (Fig 2D).To confirm these findings, and to minimize opportunities for cellular adaptation during complementation with wtXrcc4, we used transient transfection to restore expression of wtXrcc4 in Xrcc4 Δ/Δ cells.Consistent with the above-noted findings, transient Xrcc4 expression strongly suppressed I-SceIinduced HR in Xrcc4 Δ/Δ 6xTer-HR reporter cells, but had no significant impact on Tus/Terinduced STGC or LTGC in these cells (S2 Fig) .Taken together, these experiments show that Xrcc4 status has no impact on Tus/Ter-induced HR in mouse ES cells.
We showed previously that STGC at Tus/Ter-stalled forks is controlled by the HR proteins BRCA1, CtIP, BRCA2 and Rad51 and by the structure-specific nuclease scaffold SLX4 [56,58].In contrast, Tus/Ter-induced LTGC is suppressed by BRCA1 and is independent of BRCA2 or Rad51.We found that these relationships were unaffected by Xrcc4 status (Fig 3A).In the regulation of I-SceI-induced HR, we previously noted a specific role for BRCA1 and CtIP in suppressing an HR bias towards LTGC [64].In contrast, loss of BRCA2 or Rad51 had little impact on the LTGC/Total HR ratio in response to an I-SceI-induced DSB.We observed similar effects on I-SceI-induced HR in Xrcc4 Δ/Δ 6xTer-HR reporter cells (Fig 3B).Thus, although Xrcc4 deletion affects the ratio of LTGC:total HR in response to I-SceI, the interactions between HR mediators in execution of HR appear to be largely unaffected by loss of C-NHEJ.

Impact of DNA polymerase θ depletion on Tus/Ter-and I-SceI-induced HR
DNA polymerase θ, encoded by the POLQ gene, has been implicated in an alternative end joining (A-EJ) pathway and in the prevention of genomic instability at sites of replication fork stalling [69][70][71][72].Polθ has also been found to suppress DSB-induced HR in some cell types [73,74].We therefore asked whether Polθ interacts with HR in mouse ES cells, either at a Tus/Ter RFB or in DSB repair.Interestingly, siRNA-mediated depletion of Polθ modestly suppressed Tus/Ter-induced STGC in multiple clones, but in each case the effect failed reach statistical significance (Fig 4).Depletion of Polθ had no impact on I-SceI-induced HR either in wild type or Xrcc4 null cells.These findings raise the possibility that Polθ supports conservative STGC at stalled forks.They also suggest that the previously reported competition between Polθ and HR in DSB repair is not a feature of mouse ES cells [73,74].

Impact of Ku70 deletion on Tus/Ter-induced HR
The binding of the Ku70/Ku80 heterodimer to DNA ends is required for engagement of C-NHEJ [75].Ku has also been implicated in modulation of repair functions at forks stalled by the action of Topoisomerase I inhibitors, where one-ended breaks are thought to predominate [76,77].To determine whether Ku DNA end binding activity can influence Tus/Ter-induced HR independent of later steps of the C-NHEJ pathway, we targeted a single copy of the 6xTer-HR reporter to the ROSA26 locus of Ku70 -/-mES cells [78].Nine independent ROSA26-targeted Ku70 -/-6xTer-HR reporter clones revealed wild type levels of Tus/Ter-induced HR but greatly elevated levels of I-SceI-induced HR (Fig 5).To complement this phenotype, we cotransfected either Tus or I-SceI expression vectors with either empty vector or with a vector for expression of wt human KU70.Transient expression of wtKU70 suppressed I-SceI-induced HR and complemented phleomycin sensitivity of Ku70 -/-cells, as expected (Fig 6).In contrast, wtKU70 expression had no impact on Tus/Ter-induced HR (Fig 6).
In the processing of a conventional DSB, Ku binding to the DNA end is a barrier to DNA end resection.DNA end resection activity, initiated by CtIP and the Mre11 nuclease, can displace Ku from the DNA end, providing a mechanism by which the HR machinery can overcome the barrier formed by Ku DNA end binding [79].To further search for evidence of Ku interaction with stalled fork HR, we determined the impact of siRNA-mediated CtIP depletion on HR in Ku70 -/-cells either uncomplemented or transiently complemented with wtKU70.As previously reported, CtIP depletion reduced HR in response to Tus/Ter or to an I-SceI-mediated DSB [58], and this effect was observed in both uncomplemented and Ku70-complemented Ku70 -/-cells (Fig 7A and 7B).However, the proportional impact of CtIP depletion appeared less pronounced in uncomplemented I-SceI-transfected Ku70 -/-cells than in the same cells complemented with wtKU70 (Fig 7B).We quantified this effect by calculating, for each test group, the induced HR in cells that received siCtIP as a proportion of induced HR in cells that received the control siRNA directed to luciferase.Notably, for I-SceI-induced HR, this ratio was increased in uncomplemented Ku70 -/-cells in comparison to wtKU70-complemented cells (Fig 7C and 7D).In contrast, for Tus/Ter-induced HR, this ratio was unaffected by Ku70 status.We interpret these results as follows: at a DSB, Ku binding creates a barrier to end resection and CtIP plays a significant role in displacing Ku.This Ku-displacing role of CtIP is not required in Ku70 -/-cells, and the relative importance of CtIP in HR at a DSB in Ku70 -/-cells is correspondingly less.In contrast, at a Tus/Ter RFB, CtIP plays a significant role in HR that is fully independent of Ku70.Taken together with the above findings with regard to Xrcc4, the data indicate that C-NHEJ does not compete with HR at a mammalian Tus/Ter RFB.

Discussion
In contrast to HR induced by a chromosomal DSB, where C-NHEJ competes to repair the two-ended break, we show here that HR induced by a Tus/Ter RFB in mammalian cells is unaffected by the status of the C-NHEJ genes Xrcc4 or Ku70.This shows that the fundamental mechanisms of repair pathway "choice" at a stalled replication fork and a chromosomal DSB differ markedly.The simplest explanation of these findings is that HR at Tus/Ter does not entail formation of a two-ended DSB intermediate.We recently used High Throughput Translocation Sequencing (HTGTS) to study translocation-competent DNA lesions at Tus/Ter [58].In contrast to I-SceI-induced DSBs, where two-ended breaks predominate, the major lesions detected by HTGTS at Tus/Ter were solitary DNA ends.However, it is possible that twoended DSB intermediates of STGC arise at Tus/Ter but are not readily detected by HTGTS.
https://doi.org/10.1371/journal.pgen.1007486.g003Indeed, in the X. laevis model of replication-coupled ICL repair, temporally coordinated dual incisions of one sister chromatid generate a two-ended DSB intermediate.Bidirectional replication fork stalling is a critical step in this repair process, the arrival of both forks being required for replisome disassembly, asymmetrical fork reversal, nascent lagging strand resection and FANCD2/FANCI-coordinated incisions flanking the ICL [20,23,36].
A notable problem with the above-noted models, which invoke a conventional DSB intermediate, is their failure to account for the distinctive pattern of Rad51 accumulation we observe at Tus/Ter.We found that Rad51 accumulation at Tus/Ter is more intense, more sustained and more precisely localized than at a conventional DSB.These findings strongly suggest that the major DNA structures that recruit Rad51 to the Tus/Ter RFB are not conventional DSBs.We propose that Rad51 is recruited to non-DSB ssDNA structures at stalled forks and that the interaction of Rad51 with these structures accounts for the functional exclusion of C-NHEJ from stalled fork HR.A major trigger to Rad51 loading at Tus/Ter may be ssDNA gaps on the arrested lagging strand, present immediately adjacent to the Tus/Ter RFB (Fig 9B).Such ssDNA gaps would be present, albeit transiently, within a normally processive fork.However, fork stalling would render these same DNA structures abnormal, by virtue of their persistence.A static ssDNA signal at the site of stalling could provide a stable platform for the loading of Rad51.By this model, Rad51 might act as an "early responder" during stalled fork repair, as has been suggested previously [83].If Rad51 deposition were a scheduled, early response to fork stalling, this might explain the intensity and localization of the Rad51 signal we observe at Tus/Ter.
Rad51 supports fork reversal in mammalian cells in response to a variety of DNA damaging agents [83].Rad51-mediated template switching at the site of stalling could drive limited reversal of the collapsed fork.If initiated by Rad51-coated lagging strand gaps, this process would displace the unresected nascent leading strand as a 3' ssDNA tail (Fig 9B).Rapid coating of the displaced ssDNA tail by RPA and Rad51 could render it inaccessible to binding by Ku and, hence, "invisible" to the C-NHEJ pathway.The hypothetical limited fork reversal intermediate envisioned by this model might be subject to further processing, leading to more extensive fork reversal and potentially enabling HR initiation without formation of a DSB.Alternatively, incision of the cruciate structure of the reversed fork could liberate a one-ended DSB with a long 3' ssDNA tail formed by the displaced nascent leading strand.It is not yet clear whether STGC, one-way ANOVA test: siLUC vs siCtIP all, p<0.0001; siLUC vs siCtIP +EV, p = 0.0106; siLUC vs siCtIP +hKU70, p<0.0001; siLUC all, p<0.0001; siCtIP all, p<0.0001.I-SceI-induced LTGC, one-way ANOVA test: siLUC vs siCtIP all, p<0.0001; siLUC vs siCtIP +EV, p = 0.1503; siLUC vs siCtIP +hKU70, p = 0.1010; siLUC all, p = 0.0010; siCtIP all, p<0.0001.I-SceI-induced LTGC/(Total HR), one-way ANOVA test: siLUC vs siCtIP all, p<0.0001; siLUC vs siCtIP +EV, p<0.0001; siLUC vs siCtIP +hKU70, p<0.0001; siLUC all, p = 0.0147; siCtIP all, p<0.Tus/Ter-induced HR entails the formation of such a DSB intermediate.In summary, a template switch/fork reversal model of HR initiation satisfies two of the key findings reported here: first, the intense, distinctively localized recruitment of Rad51 to the Tus/Ter RFB; second, the functional exclusion of C-NHEJ during Tus/Ter-induced HR.This hypothetical model makes a number of additional predictions, which it will be relevant to test in future studies.An interesting feature of I-SceI-induced HR was revealed in this study.Specifically, although deletion of Xrcc4 elevated the frequencies of both STGC and LTGC, LTGC products as a proportion of all HR products were reduced from ~4% to ~2% in Xrcc4 null cells.Xrcc4 deletion did not perturb the fundamental relationships of I-SceI-induced HR control reported previously for BRCA1, CtIP, BRCA2 and Rad51 [64].This suggests that Xrcc4 loss influences the balance between STGC and LTGC via an HR-independent mechanism.I-SceI-induced LTGCs, generated by the HR reporter used here, can be considered a type of gap repair [26].Thus, I-SceI-induced LTGC might entail repair synthesis in one of two directions.The first would entail Rad51-mediated invasion of the misaligned GFP copy while the second would entail Rad51-mediated invasion of the correctly aligned, unbroken I-SceI site-containing GFP copy.(In the latter case, wtGFP would be generated by annealing at the point of SDSA termination.)In Xrcc4 Δ/Δ cells, the loss of high flux error-free religation of I-SceI-induced DSBs might increase the proportion of cells in which I-SceI sites on both sister chromatids are broken simultaneously.In such a circumstance, the second mechanism of LTGC noted above would be suppressed.This, in turn, could lead to the observed reduction in the proportion of I-SceI-induced HR events that resolve as LTGCs in Xrcc4 Δ/Δ cells.

Molecular biology and siRNAs
The 6xTer-HR reporters used were assembled using standard cloning methods described previously for the 6xTer-HR reporter (REF).Stable Ter-containing plasmids were generated and manipulated in JJC33 (Tus -) mutant strains of E. coli.All primers for conventional and quantitative PCR were purchased from Life Technologies.All plasmids used for mouse embryonic stem (ES) cell transfection and 293T cell transfections were prepared by endotoxin-free maxiprep (QIAGEN Sciences, Maryland, MD).siRNA SMARTpools were purchased from GE Healthcare/Dharmacon.

Lentivirus production and target mouse cell transduction
293T cells were propagated in standard DMEM media supplemented with 10% serum, glutamine and antibiotics.For lentivirus generation, 8 x 10 6 cells were seeded on 10 cm dishes and transfected 24 hours later with 5 μg pHIV, 4.45 μg psPAX2, and 0.55 μg pMD2G in antibioticfree media using Lipofectamine 2000 (Invitrogen).Media was replaced 24 hours later, and supernatant harvested every 12 hours between 48 and 72 hours after transfection and stored at 4˚C.Lentiviral particles were concentrated using Centricon Plus-70 filter devices (Millipore) per manufacturer's instructions.5 x 10 5 target mES cells were seeded per well in 6-well plate format, allowed to proliferate for 24 hours, transduced and placed under 100 μg/mL nourseothricin selection beginning 24 hours after transduction.

Cell staining
Live cells were prepared for measurement of cell surface expression of human CD52 as previously described.Cells were trypsinized and resuspended in FACS blocking buffer (PBS containing 1% BSA, 0.1% sodium-azide, and 5% heat-inactivated goat serum).Cells were stained for CD52 in blocking buffer: primary antibody, rat anti-hCD52 mAb YTH 34.

Competition assays
For Xrcc4 mutant cell competition experiments, 1.6 x 10 5 cells were co-transfected in suspension with 0.45 μg empty vector and either 50 ng empty vector or 50 ng GFP-expression plasmid using Lipofectamine 2000 (Invitrogen).For KU70 complementation cell competition experiments, 1.6 x 10 5 cells were co-transfected in suspension with 0.35 μg empty vector, 0.15 μg empty vector or hKU70-expression plasmid, and either 50 ng empty vector or 50 ng GFP-expression plasmid using Lipofectamine 2000 (Invitrogen).18 hours after transfection, cells were counted, mixed 5:1, uncolored vs. GFP + marked cells, and 5 x 10 4 cells plated in triplicate.6 hours after cell plating growth medium was replaced with media containing phleomycin (Sigma-aldrich, P9564).After two days incubation, GFP + frequencies were scored on a Beckman Coulter CytoFlex LX.Fold enrichment of cultures transiently co-transfected with GFP-expression plasmid normalized to 0 μg/mL phleomycin control.Plots represent the mean of triplicate samples from three independent experiments (n = 3).

Chromatin immunoprecipitation assays
24-48 parallel transfections of 1.6 x 10 5 cells were performed in suspension with 0. Data are presented as the mean calculated from three independent experiments (n = 3) normalized against untreated controls (empty vector or guide RNA controls) and control locus (beta-actin) using the 2 -ΔΔCT method as described previously [84].

Statistical methods
Data presented represents the arithmetic mean and error bars represent the standard error of the mean (s.e.m.) of between three (n = 3) and nine (n = 9) independent experiments (n values given in figure legends).Figure legends specify the number of replicates within each independent experiment (performed in duplicate) and the number of independent experiments (n) that were performed to generate the data presented.The arithmetic mean of samples collected for groups of independent experiments for repair frequency statistical analysis, was calculated and data points for each independent experiment used to calculate the mean and standard error of the mean (s.e.m.), calculated as standard deviation/ p n, (n indicates the number of independent experiments).Differences between sample pairs repair frequencies were analyzed by Student's two-tailed unpaired t-test, assuming unequal variance.One-way ANOVA statistical analysis of greater than three samples was performed when indicated.P-values are indicated in the figure legends.All statistics were performed using GraphPad Prism v6.0d software.

Fig 8 .
Fig 8. Distinct patterns of Rad51 recruitment to the Tus/Ter fork barrier and to a conventional DSB.A, Schematic of the 1xGFP 6xTer reporter.Location of telomere (TEL) and centromere (CEN) shown.Red half-arrow heads: primer pairs.Grey box: mutant GFP allele (6xTer-I-SceI-GFP).Orange triangle: 6xTer element array.Navy blue line: I-SceI endonuclease target site and I-SceI site-specific guide RNA target site.Primer pair positions indicated as distance between the proximal end of the closer primer sequence to proximal edge of the 6xTer array.B and C, Rad51 protein abundance at sites near the 6xTer-I-SceI-GFP allele in response to Tus/ Ter-induced replication fork stalling or DSB induction at 24 or 48 hours after transfection.Cells transiently transfected with empty vector (grey), pcDNA3β-myc NLS-Tus-F140A-3xHA (orange), pcDNA3β-myc NLS-I-SceI (royal blue), or co-transfected with spCas9 expression plasmid with control (white) or I-SceI site-specific (navy blue) guide RNA.Fold enrichment of Rad51 protein calculated as the mean 2 -ΔΔCT from three independent experiments (n = 3) normalized against untreated controls (empty vector or guide RNA controls) and β-Actin control locus.Error bars indicate the standard deviation of the ΔCT measurement calculated as the change in Ct value obtained from the proximal-Ter locus and that obtained from β-Actin control locus.https://doi.org/10.1371/journal.pgen.1007486.g008

Fig 9 .
Fig 9. Hypothetical models of Tus/Ter-induced HR.A, Conventional DSB intermediate model.Dual incision of bidirectionally arrested forks generates DNA ends that are processed for HR.Unknown mechanisms prevent Ku access to the DNA ends at the stalled fork.Dark blue: parental strands.Light blue: nascent strands.Half arrows indicate direction of nascent strand synthesis.Orange triangles: Tus/Ter RFB. Green circles: Rad51 monomers.B, Template switch/fork reversal model.Rad51 is loaded onto exposed ssDNA lagging strand daughter strand gaps at the arrested fork.Following replisome disassembly, Rad51 mediates fork remodeling via a template switch mechanism.This process displaces the 3' ssDNA end of the nascent leading strand, which is rapidly coated with RPA (not shown) followed by Rad51.The DNA end thus generated is incapable of binding Ku, excluding engagement of C-NHEJ.Further processing of the reversed fork may liberate the DNA end by more extensive fork reversal (not shown) and/or via incision of the 4-way reversed fork structure (red arrowhead).Although processing of the two opposing forks is depicted here as sequential, this model is also compatible with synchronous remodeling of both forks.Symbols as in panel A. Pale green circles, Rad51 monomers displaced from lagging strand.https://doi.org/10.1371/journal.pgen.1007486.g009 ΔΔCt = [Ct Ku70 -Ct Gapdh ]-[Ct siLUCIFERASE -Ct siGAPDH ]).Error-bars represent the standard deviation of the ΔCt value (SDEV = p