The non-homologous end joining factor Ku orchestrates replication fork resection and fine-tunes Rad51-mediated fork restart

Replication requires Homologous Recombination (HR) to stabilize and restart terminally-arrested forks. HR-mediated fork processing requires single stranded DNA (ssDNA) gaps and not necessarily Double Strand Breaks. We used genetic and molecular assays to investigate fork-resection and restart at dysfunctional, unbroken forks in Schizosaccharomyces pombe. We found that fork-resection is a two-step process coordinated by the non-homologous end joining factor Ku. An initial resection mediated by MRN/Ctp1 removes Ku from terminally-arrested forks, generating ~ 110 bp sized gaps obligatory for subsequent Exo1-mediated long-range resection and replication restart. The lack of Ku results in slower fork restart, excessive resection, and impaired RPA recruitment. We propose that terminally-arrested forks undergo fork reversal, providing a single DNA end for Ku binding which primes RPA-coated ssDNA. We uncover an unprecedented role for Ku in orchestrating resection of unbroken forks and in fine-tuning HR-mediated replication restart. Ku orchestrates a two-steps DNA end-resection of terminally-arrested and unbroken forks MRN/Ctp1 removes Ku from terminally-arrested forks to initiate fork-resection a ~110 bp sized ssDNA gap is sufficient and necessary to promote fork restart. The lack of Ku decreases ssDNA RPA-coating, and slows down replication fork restart.


Introduction 57
At each cell division, ensuring the correct duplication and segregation of the genetic material is 58 crucial for maintaining genome integrity. Although mutational events contribute to genome 59 evolution, DNA lesions trigger genome instability often associated with human diseases such as 60 cancer, genomic disorders, aging and neurological dysfunctions 1 . 61 DNA replication constitutes a major peril for genome stability. This is particularly evident in the case 62 of oncogene-induced proliferation, which results in replication stress and faulty genome duplication, 63 contributing to acquire genetic instability in neoplasic lesions 2,3 . A key feature of replication stress is 64 the alteration of replication fork progression by numerous replication fork barriers (RFBs) that 65 threaten faithful DNA duplication 4 . These hurdles are both intrinsic and extrinsic to the cell. Among 66 them are non-histone protein-DNA complexes, the clashing of replication with transcription 67 machineries, DNA secondary structures, damaged DNA, and nucleotide depletion. 68 RFBs impact replisomes' functionality, and may result in replication forks stalling, requiring 69 stabilization by the S-phase checkpoint to ensure DNA synthesis resumption 5 . RFBs may also result in 70 dysfunctional and terminally-arrested forks, which lack their replication-competent state, and 71 necessitate additional mechanisms to resume DNA synthesis. Through nucleolytic cleavage, 72 terminally-arrested forks are converted into broken forks, exhibiting one ended DSB 6 . A DNA nick 73 directly converts an active fork into a broken fork, accompanied with a loss of some replisome 74 components 7 . Forks lacking their replication competence, and thus terminally-arrested, are often 75 referred to as collapsed forks, whether broken or not. 76 To counteract deleterious outcomes of replication stress, cells have evolved the Homologous 77 Recombination (HR) pathway which ensures the repair of DSBs and secures DNA replication 8 . HR is 78 initiated by the loading of the recombinase Rad51 onto ssDNA, with the assistance of mediators such 79 as yeast Rad52 and mammalian BRCA2. The Rad51 filament then promotes homology search and 80 strand invasion with an intact homologous DNA template. HR allows the repair of forks exhibiting 81 one ended DSBs likely through Break Induced Replication 9-12 . However, growing evidences point out 82 that DSBs are not a pre-requirement neither for the recruitment of HR factors at dysfunctional forks 83 nor for their restart 13-17 . 84 DSBs are also repaired by the non-homologous end joining (NHEJ) pathway which promotes the 85 direct ligation of DNA ends with limited or no end-resection 18 . While HR is active in S and G2-phase, 86 NHEJ is active throughout the cell cycle. A key component of NHEJ is the heterodimer Ku composed 87 of two subunits, Ku70 and Ku80, both required for the stability of the complex. Yeast Ku binds dsDNA 88 ends, inhibits end-resection, and allows the ligation of DSBs trough the recruitment of Ligase 4 19-21 . 89 Interestingly, Ku is also involved in the repair of replication-born DSBs, where it limits end-90 resection 21-24 and yeast Ku acts as a backup to promote cell survival upon replication stress 25,26 . 91 In most eukaryotes, DSBs resection is a two-step process 27,28 . An initial 5' to 3' nucleolytic processing, 92 limited to the vicinity of the DNA end, is mediated by the MRN (Mre11/Rad50/Nbs1) complex which 93 binds DSBs as an early sensor 29  range resection to occur. The lack of Ku results in a slower replication fork restart, characterized by 119 an extensive resection but surprisingly a reduced amount of RPA loading. Our data are consistent 120 with dysfunctional forks undergoing fork reversal, which provides a single DNA end for Ku binding. 121 We uncover an unprecedented role for Ku in orchestrating the resection of terminally-arrested, in 122 the absence of DSBs, and in fine-tuning replication restart.

Results 139
A genetic assay to investigate replication fork-resection and restart at a conditional RFB in fission 140 yeast 141 We exploited a previously described conditional RFB, named RTS1, to block a replisome in a polar 142 manner at a specific locus 45 (Fig. 1a). The blockage of the replication fork is mediated by the RTS1-143 bound protein Rtf1 whose expression is controlled through the nmt41 promoter repressed in the 144 presence of thiamine. The RTS1-RFB has been integrated at the natural ura4 locus, at which most 145 forks travel from the centromere-proximal origin towards the telomere. 16 hours after thiamine 146 removal, > 90% of forks travelling in the main replication direction are blocked at the RTS1-RFB 147 resulting in dysfunctional forks. These terminally-arrested forks are either restarted by HR or rescued 148 by the progression of opposite forks 17 . HR-mediated replication restart occurs in 20 minutes and is 149 initiated by a ssDNA gap, not a DSB, on which RPA, Rad52 and Rad51 are loaded 46-48 . 150 We previously reported that HR-mediated fork restart is associated with a non-processive DNA 151 synthesis, liable to replication slippage (RS) 49 . Both strands of restarted forks are synthetized by 152 Polymerase delta, reflecting a non-canonical replisome likely insensitive to the RFB 50 . We developed 153 a reporter assay consisting of an inactivated allele of ura4, ura4sd20, which contains a 20 nt 154 duplication flanked by 5 bp of micro-homology 49 . When the ura4sd20 allele is replicated by a 155 restarted fork, the non-processive DNA synthesis undergoes RS resulting in the deletion of the 156 duplication and the restoration of a functional ura4 + gene. The ura4sd20 allele was inserted either 157 downstream or upstream from the RTS1-RFB to generate the construct t-ura4sd20<ori or 158 t<ura4sd20-ori (t for telomere, < for the RTS1-RFB and its polarity), respectively (Fig. 1b). We showed 159 that RS occurring upstream and downstream from the RTS1-RFB are both consequences of restarted 160 forks 47 . To monitor the basal level of RS in different genetic backgrounds, we also generated a t-161 ura4sd20-ori construct, devoid of the RTS1-RFB. 162 The frequency of downstream RS was increased by 23 fold upon expression of Rtf1 (t-ura4sd20<ori, 163 compare Rft1 repressed to expressed situation, Supplementary Fig. 1a) and by 17 fold compared to 164 the strain devoid of RFB (t-ura4sd20-ori, Supplementary Fig. 1a). In rad51 or rad52 deleted strains, 165 the leakiness of Rft1 repression was more obvious as the RS frequency was slightly higher when Rft1 166 was repressed compared to the strain devoid of RFB ( Supplementary Fig. 1a, compare Rtf1 167 expression: -, RTS1: -, RFB activity: -with Rtf1 expression: -, RTS1: +, RFB: L). Thus, to obtain the true 168 occurrence of RS by the RTS1-RFB, independently of the genetic background, we subtracted the RS 169 frequency of the strain devoid of RFB from the frequency of the strain containing the t-ura4sd20<ori 170 construct, upon expression of Rtf1 (Supplementary Fig. 1a, red arrows).The frequency of RS induced 171 by the RTS1-RFB was of 49.6, 9.8 and 10.8 10 -5 , in wt, rad51-d and rad52-d strain, respectively 172 ( Supplementary Fig. 1b). We estimated that the lack of Rad51 or Rad52 results in 80 % of forks 173 irreversibly and terminally arrested at the RFB. 174 The frequency of upstream RS was increased by 3 fold in wt cells (compared to the strain devoid of 175 RFB) and this induction was abolished in rad52-d and rad51-d cells (Supplementary Fig. 1c-e). 176 Indeed, we reported that the DNA synthesis associated to Rad51-mediated fork restart occurs 177 occasionally upstream from the initial site of fork arrest, as a consequence of ssDNA gap formation 178 by Exo1 47 . 179 Thus, the RS assays associated to the RTS1-RFB allows the quantification of replication restart 180 efficiency and an indirect monitoring of fork-resection.
The long-range resection is performed by Exo1, but not Rqh1, and is dispensable to replication 182 restart. 183 HR-mediated fork restart at the RTS1-RFB is initiated by ssDNA gaps of  1kb in size formed upstream 184 from terminally-arrested forks 47 . Although Exo1 is the main nuclease responsible for the formation of 185 these gaps, the lack of Exo1 does not impair the efficiency of replication restart, suggesting that 186 additional nucleases are likely involved. Consistent with this, upstream RFB-induced RS was abolished 187 in exo1-d cells, whereas downstream RFB-induced RS was not affected (Fig. 1c). The DSB long-range 188 resection is mediated by two independent parallel pathways: Exo1 and Rqh1 Sgs1/BLM /Dna2 27,28 . We 189 applied the RS assays to rqh1-d cells. The lack of Rqh1 did not affect upstream and downstream  induced RS, even when exo1 was deleted, indicating that Rqh1 is not required for fork-resection and 191 restart (Fig. 1c). 192 We recently reported a novel method to monitor fork-resection by bi-dimensional gel 193 electrophoresis 51 (2DGE). We identified an intermediate emanating from the fork arrest signal and 194 descending towards the linear arc, indicative of a loss of mass and shape ( Fig. 1d-e, see red arrow). 195 Alkaline 2DGE showed that this tail signal corresponds to terminally-arrested forks containing newly 196 replicated strands undergoing resection 51 . Consistent with this, the tail signal was abolished in exo1-197 d ( Fig. 1e-f). We analysed fork-resection by 2DGE in rqh1-d cells. The tail signal was unaffected in 198 rqh1-d cells compared to wt cells, confirming that, in contrast to DSBs, Rqh1 is not part of a long-199 range resection pathway of terminally-arrested forks ( Fig. 1e-f). 200 To get a deeper analysis of ssDNA length generated upstream from the RTS1-RFB, we recently 201 developed a qPCR assay to directly monitor ssDNA, based on ssDNA being refractory to restriction 202 digestion 51,52 . As previously reported, in wt cells, ssDNA was enriched 110 bp and 450 bp upstream 203 from the RTS1-RFB but was undetectable at 1.8 Kb from the RFB. Yet, in the absence of Exo1, ssDNA 204 was still enriched at 110 bp but not at 450 bp (Fig. 1g). These data hint at the presence of additional 205 nucleases acting on terminally-arrested forks to generate short ssDNA gap and subsequent 206 replication restart. 207

MRN/Ctp1 promotes fork-resection and restart, independently of the Mre11 nuclease activity 208
We investigated the role of Rad50 and Ctp1 in fork-resection and restart. Compared to wt cells, 209 downstream RFB-induced RS was decreased by 1.8 and 3.4 fold in ctp1-d and rad50-d cells, 210 respectively ( Fig. 2a,  resection by 2DGE showed that the level of resected forks was severely and similarly decreased in 218 either rad50-d or ctp1-d cells (Fig. 2b-c). We concluded that MRN/Ctp1 act together to promote fork-219 resection and subsequent fork-restart. 220 We reported that Rad52 recruitment to the RTS1-RFB relies on Mre11, independently of its nuclease 221 activity. We analysed RFB-induced RS and fork-resection by 2DGE in cells expressing a nuclease-dead 222 form of Mre11 (Mre11-D65N), defective for both the endo-and exo-nuclease activities 53,54 . This mutant showed neither a defect in restarting replication forks nor a defect in resecting dysfunctional 224 forks blocked at the RTS1-RFB (Supplementary Fig. 2). Thus, MRN/Ctp1-dependent fork-resection 225 and restart requires an intact MRN complex but not the Mre11 nuclease activity. 226

MRN/Ctp1 promotes short ssDNA gaps to prime Exo1-mediated long-range fork resection. 227
In contrast to Exo1-mediated fork-resection, the MRN/Ctp1-dependent resection step is critical to 228 ensure an efficient restart of DNA synthesis. We tested if MRN/Ctp1 acts upstream Exo1 in 229 promoting the formation of ssDNA gaps. In the simultaneous absence of Exo1 and either Ctp1 or 230 Rad50, upstream and downstream RFB-induced RS was decreased to the same extent as the single 231 deletion of either ctp1 or rad50 (Fig. 2a). Fork-resection analysis by 2DGE revealed a similar defect in 232 the single ctp1-d mutant than in the double ctp1-d exo1-d mutant (Fig. 2b-c). These data argue that 233 the role of MRN/Ctp1 is not redundant with Exo1 function, and that MRN/Ctp1 acts upstream Exo1. 234 To clarify the role of these factors in ssDNA gaps formation, we monitored ssDNA by qPCR in ctp1-d 235 and ctp1-d exo1-d cells. ssDNA enrichment at both 110 bp and 450 bp was dependent on Ctp1. Thus, 236 short and large ssDNA gaps are dependent on the MRN/Ctp1 pathway (Fig. 2d). Furthermore, Exo1 237 was no longer required for ssDNA formation at 110 bp in the absence of Ctp1. As shown for DSBs 238 resection 27,28 , we propose that fork-resection is a two-step process: MRN/Ctp1 promotes the 239 formation of short ssDNA gaps of  110 bp in size which primes the Exo1-mediated long-range 240 resection; the initial resection being critical for replication fork restart, but not the extensive one. 241

Ku accumulates at terminally-arrested forks in the absence of MRN/Ctp1. 242
Despite the absence of DSBs at the RTS1-RFB, we tested the role of Ku in the resection of terminally-243 arrested forks. The deletion of pku70 on its own did not impact upstream RFB-induced RS, but 244 rescued the defect observed in rad50-d and ctp1-d cells (Fig. 3a). To confirm that this rescue occurs 245 at the step of fork-resection, we analysed resected-forks by 2DGE. The level of resected forks was 246 fully restored in ctp1-d pku70-d and rad50-d pku70-d cells (Fig. 3b-c). Thus, the lack of Ku bypasses 247 the requirement of Rad50/Ctp1-mediated initial resection of terminally-arrested forks. 248 We investigated the recruitment of Pku70 to the RTS1-RFB by ChIP-qPCR. In wt cells, we were unable 249 to detect any recruitment, which likely reflects that the binding of Ku to terminally-arrested forks is 250 momentary, as observed to DSBs 36 (Fig. 3d). In the absence of Rad50, Pku70 accumulated exclusively 251 110bp upstream from the RTS1-RFB, suggesting that Ku binds near or at the expected 3-branched 252 junction of the dysfunctional fork. Collectively, these data argue a role of MRN/Ctp1 in releasing Ku 253 from terminally-arrested forks. 254 The lack of Ku suppresses the sensitivity of ctp1-d and rad50-d cells not only to DSBs-inducing agents 255 but also to replication-blocking agents 19,22,23,36 . These data were interpreted as a role of Ku in binding 256 replication-born DSBs formed in the vicinity of stressed forks. As reported, we found that the 257 deletion of pku70 rescued partially the sensitivity of ctp1-d and rad50-d cells to very low doses of 258 camptothecin (CPT) and methylmethane sulfonate (MMS) (Supplementary Fig. 3). CPT is an inhibitor 259 of topoisomerase I which slows down replication fork progression while MMS is an alkylated agent 260 resulting in damaged replication forks. Both drugs do not induce detectable DSBs at very low 261 doses 55,56 . We tested whether similar genetic interactions were observed upon induction of the RTS1-262 RFB. Both ctp1-d and rad50-d cells showed a loss of viability upon induction of the RTS1-RFB, that is 263 rescued by deleting pku70 (Fig. 3e). Altogether, these data support that the inability of MRN/Ctp1 to 264 remove Ku from terminally-arrested forks is a lethal event.
In the absence of Ku, MRN/Ctp1 is no longer needed to promote ssDNA gap formation. We tested 267 whether fork-resection is dependent on Exo1 in the absence of Ku. Firstly, upstream RFB-induced RS 268 observed in the pku70-d ctp1-d double mutant was dependent on Exo1 (Fig. 4a). Secondly, 2DGE 269 analysis revealed that Exo1 is responsible for fork-resection occurring in ctp1-d pku70-d cells (Fig. 4b-270 c). Thirdly, as reported by Langerak et al. 36 , the suppressive effect of pku70 deletion on cpt1-d cells 271 sensitivity to CPT and MMS was dependent on Exo1 as the pku70-d cpt1-d exo1-d triple mutant 272 exhibited the same sensitivity as the single ctp1-d mutant (Supplementary Fig. 3). These data 273 establish that the bypass of initial fork-resection by the lack of Ku relies on Exo1. We propose that Ku 274 has an inhibitory effect on the Exo1-mediated long-range fork-resection, requiring a MRN/Ctp1 relief, 275 as proposed for DSBs and telomere resection. 276 In the absence of Ku, terminally-arrested forks are no longer resected in a two-step manner and 277 ssDNA gaps are directly formed by Exo1. Quantification of ssDNA by qPCR showed that ssDNA was 278 significantly more abundant at 1.8 Kb upstream from the RTS1-RFB in pku70-d cells compared to wt 279 cells (Fig. 4d). 2DGE analysis revealed that this extensive fork-resection relies on Exo1 (Fig. 4b-c). 280 Thus, in the absence of Ku, terminally-arrested forks are excessively resected trough the Exo1-281 mediated long-range resection. We propose that Ku orchestrates the formation of ssDNA gaps by 282 ensuring that fork-resection occurs in a two-step manner. 283

The lack of Ku slows-down replication restart independently of NHEJ. 284
To test whether the lack of Ku impairs HR-mediated fork restart, we applied the downstream RS 285 assay to pku70-d cells. RFB-induced RS was decreased by 2.2 fold compared to wt, indicating that  286 50 % of forks are difficult to restart (Fig. 5a). This cannot be explained by a lower expression of 287 Rad51, Rad52 and RPA (Supplementary Fig. 4a-b). We tested a possible involvement of Ligase 4, and 288 found no defect in upstream and downstream RFB-induced RS in lig4-d cells, showing that Ligase 4 is 289 neither required for fork-resection nor restart (Supplementary Fig. 4c). These data suggest that Ku is 290 involved in HR-mediated replication restart, independently of NHEJ. 291 We asked whether excessive fork-resection impairs replication restart. We analysed downstream 292 RFB-induced RS in the pku70-d exo1-d double mutant in which long-range resection is abolished, and 293 the pku70-d ctp1-d, and the pku70-d rad50-d double mutants, in which long-range resection still 294 occurs (Fig. 5 a-b, Supplementary Fig. 4d). The three strains behave as the single pku70-d strain, 295 with a  2 fold reduction in RFB-induced RS compared to wt (Fig. 5a). The pku70-d ctp1-d exo1-d 296 triple mutant showed up to 3 fold reduction in RFB-induced RS (Fig. 5a). Thus, the lack of Ku cripples 297 HR-mediated fork restart whatever the extent of fork resection and the length of ssDNA gaps. 298 We further investigated the requirement of Ku in fork-restart at the RTS1-RFB. We employed a strain 299 in which fork convergence from the distal side is minimized by the presence of 10 repeats of the 300 TER2/TER3 rDNA rRFBs (Fig. 5b). Unlike the RTS1-RFB, TER2/TER3 slows down fork progression 301 without inducing terminally-arrested forks and recruitment of HR factors 46 . Delaying the arrival of 302 opposite forks allows more time for the process of HR-mediated fork-restart to occur at the RTS1-303 RFB 50 . We monitored the level of downstream RFB-induced RS in wt and pku70-d and observed no 304 significant differences, even when the Exo1-long range resection was abolished (Fig. 5b). Thus, the 305 defect in RFB-induced RS is rescued by delaying the arrival of converging forks. This suggests that the 306 defect is a consequence of a slow HR-mediated fork restart process rather than an inability to restart 307 replication-forks in the absence of Ku. 308 We investigated the dynamics of RPA recruitment to the RTS1-RFB by fluorescence-based imaging in 310 living cells. We employed a strain in which a GFP-LacI-bound LacO array was integrated closed to the 311 the RTS1-RFB, and expressing the RPA subunit Ssb3 fused to mCherry 51 (Fig. 5c). In both wt and 312 pku70-d strains, we observed a similar increase in S-phase cells showing RPA recruitment to the 313 LacO-marked RFB (Fig 5c-d). Such recruitment did not occur in G2 cells. Thus, RPA is recruited to 314 resected forks in S-phase cells and then evicted once arrested forks have been restarted or rescued 315 by converging forks. Thus, despite slowing down of fork restart in the absence of Ku, this does not 316 result in an accumulation of resected and arrested forks in G2 cells, suggesting that dysfunctional 317 forks are ultimately rescued by the progression of opposite forks. 318 When performing cell imaging, we noticed that RPA foci were less bright in the absence of Ku. We 319 quantified the area and intensity of RPA foci recruited to the LacO-marked RFB. We found that the 320 area was not affected, whereas the intensity of RPA foci was decreased by half in pku70-d cells 321 compared to wt (Fig. 5e). These observations cannot be explained by a reduced amount of ssDNA in 322 the absence of Ku, as ssDNA gaps are  twice longer than in wt cells (Fig. 4d). These data suggest a 323 role for Ku in recruiting RPA onto ssDNA at terminally-arrested forks. In support of this, we observed 324 that Pku70-HA co-immunoprecipitates with two RPA subunits, Rad11-YFP and Ssb3-YFP, from protein 325 extracts treated with benzonase. Thus, as observed in budding yeast, Ku interacts with RPA, 326 independently of DNA and RNA 57 ( Fig. 5f and supplementary Fig. 5). Altogether, these data indicate a 327 role of Ku in ensuring RPA-coated ssDNA gaps at terminally-arrested forks to fine-tune HR-mediated 328 replication restart. Rad51-dependent processing of replication forks can occur independently of DSBs 14,17 . We 348 investigated the resection step of terminally-arrested, unbroken, forks that allows the exposure of 349 ssDNA, subsequent recruitment of RPA, and Rad51-mediated fork-restart 47 . We made unexpected 350 findings. First, the two-step model of DSB resection applies to unbroken forks. Second, the initial 351 fork-resection includes Ku eviction, suggesting that dysfunctional forks undergo fork-reversal, 352 providing a single dsDNA end for Ku binding. Third, the lack of Ku impairs efficient ssDNA coating by 353 RPA and slows down HR-mediated fork restart, independently of NHEJ. 354 Ku is recruited to terminally-arrested and unbroken forks. 355 Our genetic data establish that Ku inhibits Exo1-mediated long range resection of terminally-arrested 356 forks. MRN/Ctp1 counteracts this inhibition. Ku binds dsDNA ends with high affinity, and with poor 357 affinity for ssDNA 58 . Despite the lack of DSBs, we demonstrate that Ku is recruited to dysfunctional 358 and unbroken forks. These surprising findings favor a model in which terminally-arrested forks 359 undergo fork-reversal 55 (Fig. 6) independently of DSBs. We propose that the DNA end of reversed fork is recognized and processed 371 as a DSB, despite the absence of detectable DNA breakage (Fig. 6). 372 Thus, we propose that Ku eviction from terminally-arrested forks by MRN/Ctp1 is an essential step 392 for subsequent replication restart and cell viability. 393

MRN/Ctp1 ensures Ku removal to initiate fork-resection and restart
Ku coordinates the two-step process of fork-resection and fine-tunes Rad51-mediated fork restart. 394 Together with our previous work, our data establish that the resection of terminally-arrested and 395 unbroken forks occurs in two-steps which are coordinated by Ku. The initial MRN/Ctp1-dependent 396 resection promotes Ku eviction and generates 110 bp of ssDNA, essential to HR factors' recruitment 397 and subsequent fork-restart 47 . The second step is a long-range resection, mediated by Exo1, but not 398 Rqh1, generating up to 1kb of ssDNA which is not strictly required to the resumption of DNA 399 synthesis. Thus, a limited amount of ssDNA is sufficient to promote Rad51-mediated fork restart 400 whereas long-range resection may reinforce checkpoint activation. 401 In the absence of Ku, MRN/Ctp1 is no longer required to initiate fork-resection, which then relies 402 only on Exo1. As a consequence, terminally-arrested forks are extensively resected with an 403 accumulation of 2Kb of ssDNA. This supports that Exo1 inhibition by Ku takes place at the initial 404 resection step even with a functional MRN/Ctp1 pathway. Remarkably, the initial resection of DSBs 405 and telomeres is increased in the absence of Ku 21,38 . Thus, Ku plays an important role in ensuring that 406 fork-resection occurs in two-steps to avoid unnecessary extensive resection, which can be 407 detrimental to genome stability. 408 Previous works have proposed a NHEJ-independent role of Ku at replication-born DSBs, to channel 409 repair towards the HR pathway 22,23,36 . Fission yeast genetics indicate a role for Ku in the recovery 410 from replication stress and stabilizing replication forks 25,26 . An important finding we made is that the 411 lack of Ku slows down Rad51-mediated fork restart, accompanied with an extensive fork resection 412 and a reduced amount of RPA-coated ssDNA. These data contrast with Ku acting as a barrier against 413 RPA loading onto ssDNA during DSBs repair 36 , suggesting replicative-specific functions for Ku. RPA is a 414 critical factor to control end-resection, stability of ssDNA and subsequent recruitment of HR factors 62 . 415 As observed in budding yeast 57 , we report physical interactions between Ku and RPA, suggesting that 416 Ku facilitates RPA loading onto ssDNA. We propose that Ku fine-tunes Rad51-mediated fork restart 417 by priming RPA loading onto ssDNA and subsequent HR factors' recruitment. 418 Ku allows the recruitment of downstream NHEJ factors such as Ligase 4 to promote DSBs ligation. We 419 found no role for Ligase 4 in promoting fork-resection and restart. Our data rather suggest that NHEJ 420 inhibition enhances HR-mediated replication restart. Given the potential deleterious outcomes of 421 error-prone NHEJ events on genome stability, it is possible that the recruitment of additional NHEJ 422 factors is prevented to avoid unwanted ligation of the reversed arm at terminally-arrested forks. 423 Recent reports indicate that ssDNA stabilization by RPA influences repair pathway choice at DSB 424 between BIR and microhomology-mediated end joining (MMEJ) 63 Replication slippage assay with ura4-SD20 allele 494 Replication slippage using the ura4-SD20 allele was performed as previously described 49 . 5-FOA 495 resistant colonies were grown on uracil-containing plates with or without thiamine for 2 days at 30°C, 496 and then inoculated in uracil-containing EMM for 24h. Cells were diluted and plated on YE plates (for 497 survival counting) and on uracil-free plates containing thiamine to determine the reversion 498 frequency. Colonies were counted after 5 to 7 days of incubation at 30°C. Statistics were performed 499 using the non-parametric Mann and Whitney test. To avoid taking into account events that do not 500 represent the behavior of each mutant (such as suppressor or other additional spontaneous 501 mutations), outliers were not included in the statistical analysis or graphical representation. Outliers 502 were calculated according to the formulas: Superior Outlier>1.5x(Q3-Q1)+Q3, and Inferior 503 Outlier<1.5x(Q3-Q1)-Q1, where Q1 is the first quartile, and Q3 is the third quartile. 504

Cell viability 505
Cell viability assays were performed as previously described 17 . Cells were grown on supplemented 506 EMM without thiamine for 14 hours, then they were appropriately diluted and plated on EMM plates 507 with (RFB OFF) or without thiamine (RFB ON). Colonies were counted after 5-7 days incubation at 508 30°C and viability was calculated as the ratio of colonies growing in ON condition relative to those 509 growing in OFF condition. 510 Co-immunopercipitation 511 5.10 8 cells were harvested, washed in cold water and resuspended in 400 ml of EB buffer (50 mM  512 HEPES High salt, 50mM KOAc pH 7.5, 5 mM EGTA, 1% triton X-100, 1 mM PMSF, and protease inhibitors). Cell lysis was performed with a Precellys homogenizer. The lysate was treated with 514 250mU/µl of benzonase for 30min. After centrifugation, the supernatant was recovered and an 515 aliquot of 50 µl was saved as the INPUT control. To 300µl of lysate, 2 µl of anti-GFP (A11122 from Life 516 Technologies) antibody were added and incubated for 1h30 at 4°C on a wheel. Then, 20µl of 517 Dynabeads protein-G (Life Technologies) prewashed in PBS were added and then incubated at 4°C 518 overnight. Beads were then washed twice 10 min in EB buffer before migration on acrylamide gel for 519 analysis by Western blot. Ku70-HA, and Ssb3-YFP and Rad11-YFP were detected using anti-HA high 520 affinity antibody (clone 3F10, Roche), and using anti-GFP antibody (Roche, clones 7.1 and 13.1), 521 respectively. The supplementary Fig. 5 shows that Pku70-HA was slightly interacting in an unspecific 522 way with anti-GFP antibody. However, the intensity of Pku70-HA in the IP fraction was highly 523 increased We thank Fuyuki Ishikawa for the gift of the Pku70-HA strain and Tony Carr for the lig4-d strain. We 552 thank Vincent Geli for critical comments of the manuscript. We also thank the PICT-553 IBiSA@Orsay Imaging Facility of the Institut Curie. We thank Soraya Benazzouk and Yasmina Chekkal 554 for their technical assistance. This study was supported by grants from the Institut Curie, the CNRS, 555 the Fondation ARC, the Fondation Ligue (comité Essone), l'Agence Nationale de la Recherche ANR-14-556