MainTens of thousands of single-strand DNA breaks (SSBs) arise each day1,2. When a replication fork encounters an SSB, it generates a double-strand break (DSB)3,4,5,6,7,8. This process, termed ‘fork collapse’ (ref. 7), blocks DNA synthesis and occurs every cell cycle in humans9 in response to SSBs generated by topoisomerases4, transcription10, viral infection11, antiviral responses12, lagging-strand maturation13, abasic sites14, defective DNA repair15 and genome editing16. Unresolved collapsed forks are highly mutagenic and potentially oncogenic5,17. Fork collapse is central to many anticancer therapies18, including poly(ADP-ribose) polymerase (PARP) inhibitors1,19,20. Although multiple pathways can resolve collapsed forks5,8,21,22,23,24,25,26,27,28,29,30,31,32,33,34,35,36,37,38,39,40,41, how their usage is determined remains unclear. Addressing these questions is imperative given the profound implications of fork collapse for normal cellular functioning and human health.When a replication fork encounters an SSB, the fork can break to form a single-ended DSB (seDSB)4,5,6,21,22,27, leaving a sister DNA duplex with a gap that is rapidly filled and ligated6. In parallel, the replisome is lost through different mechanisms depending on which parental strand contains the SSB. A leading-strand template SSB (‘leading collapse’) causes rapid replisome dissociation6 from loss of the translocating strand (Extended Data Fig. 1a). A lagging-strand template SSB (‘lagging collapse’) causes either replisome unloading through the termination pathway (Extended Data Fig. 1b)6 or continued unwinding22 to form a double-ended DSB (deDSB) behind the fork (Extended Data Fig. 1c)21,22,23. deDSBs also form at leading SSBs21,22,23,42,43, likely through a converging fork5,22,27 (Extended Data Fig. 1d). Fork convergence at an SSB is reported to generate a deDSB43,44 but an seDSB intermediate has not been observed and extensive rereplication can occur44,45. Thus, we lack direct evidence that converging forks at an SSB generate an seDSB that converts to a deDSB.DSB ‘resolution’ at collapsed forks (removal of the free DNA end) depends critically on homologous recombination (HR)3,5,8,21,22,23,24,25,26. Resolution likely involves RAD51-dependent D-loop formation26, regardless of whether an seDSB or deDSB forms24. In contrast, replication-independent DSBs undergo both HR and end joining46,47,48. End joining may occur following fork collapse27,28,49,50,51,52 but its contribution remains unclear. Once a D-loop forms at an seDSB, it is capable of extensive DNA synthesis through ‘break-induced replication’ (BIR)53,54,55, which is well established in eukaryotes56 and may function analogously to bacterial replication restart57. However, several observations call into question BIR as a major restart mechanism; BIR is detected through error-prone repair events29,36 that may not represent most outcomes, key bacterial restart proteins are not conserved in eukaryotes21, BIR is strongly negatively regulated27 and human BIR proteins58,59 are dispensable for resolving DNA ends following fork collapse22. Direct analysis of BIR is complicated because interfering with seDSB resolution results in deDSB formation by fork convergence22. Hence, it remains unclear how effectively collapsed forks complete DNA synthesis, with or without fork convergence to generate a deDSB.To address these questions, we replicated ‘simple’ SSBs using Xenopus laevis egg extracts60,61, which mimic the nuclear proteome of human cells62, support many DNA repair pathways63 and recapitulate fork collapse mechanisms that function in human cells6,22. We found that either leading or lagging collapse induced seDSBs that were efficiently resolved by RAD51 to generate joint molecules arising from D-loops, as well as erroneous end-to-end fusions involving microhomology. Restart of DNA synthesis was not detected at single collapsed forks. seDSBs generated by leading but not lagging template SSBs also underwent extensive resection that disassembled the sister fork through ‘secondary collapse’. In contrast, semisynchronous convergent collapse generated deDSBs that efficiently completed DNA synthesis independently of RAD51 through annealing-dependent DSB repair, which generated precise deletions and templated insertions. We observed similar outcomes at SSBs generated by Cas9 nickase (nCas9; H840A). Overall, our data demonstrate that single and convergent collapsed forks can elicit distinct repair outcomes.ResultsLeading and lagging collapse elicit DSB resolution but not replication restartTo examine leading-strand fork collapse, we replicated pSSBLEAD (Fig. 1a and Extended Data Fig. 1e,f)6,60,64 in X. laevis egg extracts61. Replication without TetR permitted SSB religation6, yielding θ structures arising from converging forks stalled at the LacR barrier (Fig. 1b, lanes 1–5)60. TetR addition stabilized SSBs, converting θ intermediates into σ structures arising from fork collapse (Fig. 1b, lanes 6–8, and Fig. 1c)6, with collapse approaching 100% (Extended Data Fig. 1g–i). Total DNA synthesis was unaltered (Extended Data Fig. 1g), indicating normal replication before collapse.Fig. 1: Leading and lagging seDSBs are resolved but do not restart DNA synthesis.The alternative text for this image may have been generated using AI.Full size imagea, pSSBLEAD contains five tandem, site-specific and strand-specific SSBs generated by nicking endonucleases before enzyme removal. SSBs are flanked by tetO arrays bound by TetR to competitively inhibit religation without impairing replisome progression (Extended Data Fig. 1e). A downstream 1.6-kb lacO array bound by LacR blocks converging forks at a distance60 to isolate leading-strand collapse events. Plasmids harbor multiple sequence-nonspecific origins distributed throughout the backbone89,90. The tetO array is placed 320 bp from the LacR array, such that approximately 90% of events should involve a single fork encountering the SSB given the 2,684-bp backbone. pSSBLEAD was replicated using X. laevis egg extracts derived from replicating nuclei61 in the presence of TetR (+TetR) to stabilize the SSBs. TetR was omitted in the buffer control (+buffer), which allowed religation of SSBs before replication and prevented collapse (as described previously6). LacR was included to impede converging replication forks and ensure strand-specific fork collapse at a single fork. Nascent strands were radiolabeled by inclusion of [α-32P]dATP. A prereplicated internal control plasmid (scCMctrl) was included as a loading control for normalization of replication intermediates (Extended Data Fig. 1f). b, Samples from a were separated on an agarose gel and visualized by autoradiography (Extended Data Fig. 1g–l). c, Quantification of σ structures from b. Data are presented as the mean ± s.d. (n = 9 independent experiments). d, Quantification of HMw products from b. Data are presented as the mean ± s.d. (n = 9 independent experiments). e, Schematic depicting the assay for restart of DNA synthesis in the collapse region (Extended Data Fig. 1m,n). f, Purified DNA samples from t = 120 in b were digested with XmnI and SacI, then separated on an agarose gel and visualized by autoradiography. g, Quantification of collapse region DNA synthesis from f as in e. Signal was normalized to control fragment ‘Lin1’. Expected values for no repair account for replication efficiency and collapse efficiency (Extended Data Fig. 1n). Data presented are the mean ± s.d. (n = 9 independent experiments). Data were analyzed by one-way analysis of variance (ANOVA) and Dunnett’s multiple-comparison method. h, Schematic of pSSBLAG as for pSSBLEAD in a, but with SSBs positioned on the lagging-strand template. i, Samples from h were separated and visualized as in b (Extended Data Fig. 1o–r). j, Quantification of σ structures from i as in c. Data are presented as the mean ± s.d. (n = 3 independent experiments). k, Quantification of HMw products from i as in d. Data are presented as the mean ± s.d. (n = 3 independent experiments). l, Schematic of the assay for collapse region DNA synthesis as in e, but for lagging collapse. m, Purified DNA samples from t = 120 in i were digested and visualized as in f. n, Quantification of collapse region DNA synthesis from m as in g. Data are presented as the mean ± s.d. (n = 6 independent experiments). Data were analyzed by one-way ANOVA and Dunnett’s multiple-comparison method. Lin, linear; E–E, end–end joining; NS, not significant; nd, not determined.Source dataFollowing collapse, σ structures declined (Fig. 1c) and two products emerged. The major product was high-molecular-weight (HMw) species remaining in the well (Fig. 1b, lanes 9 and 10, and Fig. 1d), which resembled DSB repair products65,66. We also observed nicked and supercoiled circular monomers (nCMs and scCMs; Fig. 1b, lanes 9 and 10, and Extended Data Fig. 1j,k) that were much less abundant than HMw species (Extended Data Fig. 1l). Because CMs were absent from the control (Fig. 1b, lanes 1–5), they were unlikely to arise from displacement of the LacR barrier. Their existence suggested a mechanism that converts collapsed molecules into full-length circular plasmids (Fig. 3). Thus, seDSBs from leading collapse were efficiently converted into at least three products.seDSBs can restart replication5,26,27,28,29,30,31,32,33,34,35,36. To assess restart efficiency, we used restriction digests to measure nascent DNA synthesis around the leading SSB (collapse region DNA synthesis; Fig. 1e and Extended Data Fig. 1m,n). Collapse reduced synthesis ~4-fold in this region (Fig. 1f,g), matching levels expected if no restart had occurred (Fig. 1g). This reduction was not because of TetR blocking restart (Extended Data Fig. 2a–e) or interlinked repair intermediates (Extended Data Fig. 2f–h). Denaturing analysis confirmed the absence of restart and detected the broken end after 2 h (arm; Extended Data Fig. 2g), indicating a stable seDSB. Both native and denaturing analyses detected end-to-end fusions (Fig. 1f and Extended Data Fig. 2g). Increasing the distance between the collapse site and the LacR array did not promote restart (Extended Data Fig. 2i–r); a small increase in synthesis was attributable to fork convergence arising from multiple sequence-nonspecific origins (Extended Data Fig. 2q,r). Thus, seDSBs arising from leading collapse did not detectably restart DNA synthesis, regardless of distance from the LacR barrier.We next examined lagging seDSBs using pSSBLAG (Fig. 1h and Extended Data Fig. 1o). Lagging collapse also approached 100% efficiency (Extended Data Fig. 1p,q) and generated seDSBs (Fig. 1i,k) that were converted to HMw species (Fig. 1i,k and Extended Data Fig. 1r–t), as for leading collapse (Extended Data Fig. 1s,t). No DNA synthesis was detected in the collapse region (Fig. 1l–n) and apparent end-to-end fusions were detected (Fig. 1m), also as for leading collapse. However, several differences were noted; CMs were not detected (Fig. 1I), θ intermediates remained detectable, indicating less efficient collapse (Fig. 1i, lanes 6–10), and a more pronounced reduction in total DNA signal occurred (Extended Data Fig. 1o), suggesting increased degradation. By combining leading and lagging collapse data, we calculated that no more than 8.5% of collapse events led to restart (Extended Data Fig. 1u). Overall, both leading and lagging collapse generated seDSBs that were resolved to HMw species, with replication restart either rare (≤8.5%) or absent.Collapsed forks form D-loops and undergo end joiningTo test the role of RAD51 in resolving leading seDSBs, we induced fork collapse in extracts containing a wild-type BRCA2-derived peptide that inhibits RAD51 binding to chromatin (BRC4WT) or a mutant control (BRC4Mut, Fig. 2a)67,68. RAD51 inhibition greatly reduced HMw species (Fig. 2b, lanes 6–10, Fig. 2c and Extended Data Fig. 3a) and stabilized collapsed forks (Fig. 2b, lanes 6–10), indicating that RAD51 converts collapsed forks to HMw species. CMs increased upon RAD51 inhibition (Fig. 2b, lanes 8–10, and Extended Data Fig. 3b), demonstrating that RAD51 typically suppresses these products. RAD51 inhibition did not reduce collapse region DNA synthesis (Extended Data Fig. 3c,d), supporting the conclusion that seDSB resolution does not restart replication. HMw species were resolved by RuvC treatment (Fig. 2d), indicating they contain joint molecules69. These junctions are consistent with D-loops, rather than reversed forks or Holliday junctions, because they arise from RAD51-dependent repair of an seDSB. Thus, following leading collapse, RAD51 converts seDSBs to joint molecules arising from D-loops and suppresses formation of full-length molecules.Fig. 2: Collapsed forks undergo RAD51-dependent D-loop formation and end joining.The alternative text for this image may have been generated using AI.Full size imagea, Leading collapse was induced by replication of pSSBLEAD in the presence of TetR and LacR, with vehicle, BRC4WT peptide or BRC4Mut peptide. BRC4WT is a BRCA2-derived peptide that inhibits RAD51 binding to chromatin67,68. BRC4Mut is a mutant peptide that does not disrupt RAD51 function and served as a negative control67,68. b, Samples from a were separated on an agarose gel and visualized by autoradiography. c, Quantification of HMw products from b. Data are presented as the mean ± s.d. (n = 3 independent experiments). d, Products of leading collapse were purified and digested with RuvC. Digested samples were then separated on an agarose gel and visualized by autoradiography (Extended Data Fig. 3a–e) (n = 5 independent experiments). e, Lagging collapse was induced by replication of pSSBLAG as for pSSBLEAD in a. f, Samples from e were separated and visualized as in b. g, Quantification of HMw products from f as in c. Data are presented as the mean ± s.d. (n = 3 independent experiments). h, Products of lagging collapse were purified and digested with RuvC as in d (Extended Data Fig. 3f–i) (n = 2 independent experiments). i, Schematic depicting expected products of end-to-end fusion events. Restriction enzymes located at different distances from the collapse site were used to map end-to-end fusion products. Digestion of end-to-end fused molecules was expected to generate linear fragments twice the length of the distance between the restriction site and the SSB. j, Products of leading collapse were purified with phenol–chloroform extraction to recover total DNA products or using column purification to remove HMw species (Extended Data Fig. 3j,k). Purified DNA samples were restriction-digested, separated on an agarose gel and visualized by autoradiography. Colored arrows indicate end-to-end linear fragments indicated in i. The asterisk indicates collapsed arm fragments (Extended Data Fig. 3j–m) (n = 4 independent experiments). k, Mapping of palindromic reads from end-to-end fusion products to the tetO array. Each horizontal bar represents an individual palindromic read, with the green portion indicating the 5′ palindrome sequence and dashed lines indicating hairpin regions. Black vertical lines mark the junction point of each fusion. In total, 89% of reads (39/44) mapped to the vicinity of the tetO array and did not extend beyond it, consistent with inefficient restart of DNA synthesis from collapsed ends (Extended Data Fig. 4).Source dataRAD51 inhibition also reduced HMw species following lagging collapse (Fig. 2e–g and Extended Data Fig. 3f) and these were also RuvC sensitive (Fig. 2h). RAD51 inhibition did not reduce collapse region DNA synthesis (Extended Data Fig. 3g,h). Thus, lagging collapse also converts seDSBs to joint molecules arising from D-loops.End-to-end fusions followed leading collapse (Fig. 1f) and were suppressed by RAD51 inhibition (Extended Data Fig. 3c,e). End-to-end fusions also formed following lagging collapse but were not suppressed by RAD51 inhibition (Extended Data Fig. 3g,i). To confirm that HMw species contained end-to-end fusions, we used column purification to remove HMw products (Extended Data Fig. 3j–p). Restriction digests with enzymes located at different distances from the collapse site (Fig. 2i) produced fragments consistent with end-to-end fusions (Fig. 2j, lanes 1, 3, 5 and 7, and Extended Data Fig. 3q) that were greatly reduced upon HMw removal (Fig. 2j, lanes 2, 4, 6 and 8). These products appeared late (Extended Data Fig. 3l) and were insensitive to topoisomerase II (Extended Data Fig. 3m). Illumina sequencing of palindromic reads mapping to the vicinity of the tetO array (Fig. 2k) revealed approximately 2 bp of microhomology at each junction (Extended Data Fig. 4), indicating microhomology-mediated end joining (MMEJ). Only 11% of reads mapped downstream of the tetO array (Fig. 2k), supporting our assessment that restart following fork collapse was rare or absent. Overall, seDSBs from both leading and lagging collapse formed RAD51-dependent joint molecules and underwent end-to-end fusions involving MMEJ, with RAD51 promoting these fusions only for leading collapse.Leading SSBs cause secondary replication fork collapseLeading collapse yielded full-length CMs (Fig. 1b) that were suppressed by RAD51 (Fig. 2b) and absent in lagging collapse (Fig. 1i). We initially hypothesized that these arose from enhanced progression through the LacR barrier. However, restriction digests revealed that leading collapse did not induce detectable DNA synthesis within the lacO array (Fig. 3a,b,d) yet produced a ~3-fold increase in full-length molecules (Fig. 3b, lanes 1 and 2, and Fig. 3c). Thus, leading collapse produced full-length molecules without synthesis through the LacR barrier.Fig. 3: Leading seDSBs can be resolved by extensive nuclease digestion.The alternative text for this image may have been generated using AI.Full size imagea, Schematic depicting formation of full-length molecules by synthesis through the LacR array or independent of LacR array displacement. b, Leading collapse was induced as depicted in Fig. 1a. Samples from t = 120 min were then purified and analyzed by restriction digest using either AlwNI alone or SacI and KpnI combined. Digested samples were separated on an agarose gel and visualized by autoradiography. c, Quantification of full-length linear products arising from AlwNI digestion in b. Data are the mean ± s.d. (n = 3 independent experiments). Data were analyzed by unpaired two-sided t-test. ***P = 0.0002. d, Quantification of linearized lacO array fragments arising from SacI–KpnI digestion in b. Data are presented as the mean ± s.d. (n = 3 independent experiments). Data were analyzed by unpaired two-sided t-test. e, Leading collapse was induced as depicted in Fig. 1a in mock or MRE11-immunodepleted extracts. Schematic depicts generation of CMs by extensive resection of seDSBs, which would be blocked by MRE11 immunodepletion. f, Samples from e were separated on an agarose gel and visualized by autoradiography. σ* is a topoisomer of σ because the two species are converted to a single product by restriction digest (Extended Data Fig. 5b). σ species migrated more slowly over time in mock but not MRE11-immunodepleted conditions, indicating the slowed migration was because of resection—presumably because of secondary structures formed by exposed single-stranded DNA. g, Quantification of σ structures from f. Data are presented as the mean ± s.d. (n = 3 independent experiments). h, Quantification of nCMs plus scCMs from f. Data are presented as the mean ± s.d. (n = 3 independent experiments). i, Schematic comparing the predicted outcomes of secondary collapse for small (pSSBLEAD, 2,684-bp backbone) and large (pSSBLEAD-Large, 4684-bp backbone) plasmids. Increasing the backbone size positions the sister fork further from the collapse site, which is predicted to prevent resection from reaching the sister fork and, thus, abolish CM formation. j, Leading collapse was induced for pSSBLEAD and pSSBLEAD-Large as in Fig. 1a. Samples were separated on an agarose gel and visualized by autoradiography. k, Quantification of nCMs from j. Data are presented as the mean ± s.d. (n = 3 independent experiments).Source dataWe next hypothesized that extensive 5′–3′ resection removed the lagging-strand template from the sister fork, causing it to collapse6. This secondary collapse model predicts that blocking resection should abolish full-length molecules (Fig. 3e). MRE11 immunodepletion inhibited resection, stabilized collapsed forks (Fig. 3f,g and Extended Data Fig. 5a–c) and abolished formation of both HMw species and full-length molecules (CMs; Fig. 3f,h and Extended Data Fig. 5d,e). We also tested whether DNA polymerase activity counteracted resection during secondary collapse. Addition of aphidicolin following collapse caused a substantial increase in nCMs (Extended Data Fig. 5f–i), consistent with DNA polymerase activity guarding against resection. Increasing the size of the plasmid backbone to move the sister fork further away essentially eliminated nCMs (Fig. 3i–k). This result strongly supported the conclusion that CMs arise from resection past the sister fork. Thus, leading SSBs can trigger secondary collapse through extensive resection. Because these observations were made in X. laevis egg extracts on plasmid substrates, the frequency with which this process occurs in cells with linear chromosomes remains to be established (Discussion).Efficient end resolution and completion of DNA synthesis following convergent collapseWe next asked whether fork convergence at an SSB (‘convergent collapse’) could trigger deDSB formation and completion of DNA synthesis. We replicated pSSBLEAD without LacR to allow fork convergence (pSSB; Fig. 4a). Without TetR, SSBs were repaired and expected CM replication products formed (Fig. 4b)6,64. With TetR, DNA synthesis increased slightly (Extended Data Fig. 6a), possibly because of degradation and resynthesis of the parental strand. We detected linear molecules corresponding to deDSBs (Fig. 4b, lanes 6 and 7, and Fig. 4e), collapse approached 100% (Extended Data Fig. 6b) and HMw species formed as for single-fork collapse (Fig. 4c).Fig. 4: deDSBs arising from fork collapse readily complete DNA synthesis.The alternative text for this image may have been generated using AI.Full size imagea, pSSB was replicated using X. laevis egg extract in the presence of TetR (+TetR) to stabilize the SSBs. TetR was omitted in the buffer control (+buffer), which allowed religation of SSBs before replication. LacR was omitted from the reaction so that forks could converge upon the SSBs to elicit convergent collapse. Nascent strands were radiolabeled by inclusion of [α-32P]dATP. Fully replicated plasmid DNA (scCMctrl) served as a loading control. b, Samples from a were separated on an agarose gel and visualized by autoradiography (Extended Data Fig. 6a–c). c, Quantification of HMw products from b. Data are presented as the mean ± s.d. (n = 4 independent experiments). d, Quantification of σ structures from b. Data are presented as the mean ± s.d. (n = 4 independent experiments). e, Quantification of linears from b. Data are presented as the mean ± s.d. (n = 4 independent experiments). f, Schematic depicting the assay for completion of DNA synthesis in the collapse region. g, Purified DNA samples from t = 120 min in b were digested with XmnI and SacI, then separated on an agarose gel and visualized by autoradiography. h, Quantification of collapse region DNA synthesis from g as in f. Signal was normalized to control fragment Lin1. Expected values for no repair account for replication efficiency and collapse efficiency. Data are presented as the mean ± s.d. (n = 4 independent experiments). Data were analyzed by one-way ANOVA and Tukey’s multiple-comparison method. **P = 0.0012. i, Convergent collapse was induced by replication of pSSB in the presence of TetR and vehicle, BRC4WT or BRC4Mut as a negative control. j, Samples from i were separated on an agarose gel and visualized by autoradiography. k, Quantification of HMw products from j. Data are presented as the mean ± s.d. (n = 3 independent experiments).Source dataDuring convergent collapse, we also detected seDSBs (Fig. 4b, lanes 6 and 7) that peaked at 10 min and rapidly declined by 20 min (Fig. 4b,d), while deDSBs persisted from 10–20 min before declining (Fig. 4b,e). Thus, seDSBs resolved before deDSBs. seDSBs were ~2.5-fold lower during convergent collapse than during single-fork collapse (~20% in Fig. 4d versus ~50% in Fig. 1c,j), indicating rapid conversion to deDSBs by the converging fork. deDSBs were also ~3-fold less abundant than seDSBs (~6% in Fig. 4e versus ~20% in Fig. 4d), indicating that deDSBs were more readily resolved. Accordingly, deDSBs resolved by 30 min (Fig. 4e) while seDSBs from single-fork collapse did not resolve until after 60 min (Fig. 1c,j). These data provide direct evidence that convergent collapse initially forms an seDSB that the converging fork converts to a more rapidly resolved deDSB (Fig. 4a).Collapse region DNA synthesis was indistinguishable from the control (Fig. 4f–h) and HMw products contained only full-length molecules (Extended Data Fig. 6c–e). Thus, convergent collapse enabled efficient completion of DNA synthesis through DSB repair. Inspection of the collapse region revealed small deletion products (Fig. 4g) specific to convergent collapse; end-to-end fusions arising from single-fork collapse (Figs. 1f and 2i–k) were not detected (Fig. 4g).RAD51 inhibition decreased HMw species slightly but consistently (Fig. 4i,j, lanes 6–10, Fig. 4k and Extended Data Fig. 6f), suggesting a limited requirement for HR. It was possible that some seDSBs were rapidly resolved by RAD51 before fork convergence. To address this, we performed a fine time course (Extended Data Fig. 6g–k). We observed no difference in the rate of formation or resolution of seDSBs (Extended Data Fig. 6i), arguing against rapid HR-mediated resolution before fork convergence. The limited role for RAD51 was unlikely to reflect inefficient BRC peptide inhibition, as RAD51 immunodepletion produced the same result (Extended Data Fig. 7a–f). HMw products were not resolved by RuvC or topoisomerase treatment but were fully resolved by restriction digest (Extended Data Fig. 7g,h), indicating that they were concatemers formed by end ligation rather than D-loops. RAD51 inhibition did not affect collapse region DNA synthesis or deletion events (Extended Data Fig. 7i–k). Thus, convergent collapse led to efficient completion of DNA synthesis largely independently of HR, in contrast to other reports3,5,8,21,22,23,24,25,26 but consistent with the ability of HR-independent repair pathways to act at deDSBs27,28,49,50,51,52.Single and convergent forks elicit distinct outcomes following nCas9-induced collapseWe next tested whether our findings apply to CRISPR–Cas9 SSBs using H840A Cas9 nickase (nCas9), which primarily generates seDSBs22. We replicated a plasmid harboring multiple nCas9 target sites adjacent to a lacO array (pCRISPR; Fig. 5a), with or without LacR to directly compare leading and convergent collapse.Fig. 5: Analysis of nCas9-induced fork collapse.The alternative text for this image may have been generated using AI.Full size imagea, pCRISPR was replicated using X. laevis egg extract in the presence of nCas9 to generate SSBs. Nuclease-dead Cas9 (dCas9) did not generate SSBs and was used as a control. LacR was included to induce leading collapse or omitted to induce convergent collapse. Nascent strands were radiolabeled by inclusion of [α-32P]dATP. Fully replicated plasmid DNA (scCMctrl) served as a loading control. b, Samples from a were separated on an agarose gel and visualized by autoradiography. c, Quantification of σ structures in b. Data are presented as the mean ± s.d. (n = 3 independent experiments). d, Quantification of HMw products in b. Data are presented as the mean ± s.d. (n = 3 independent experiments). e, Schematic depicting the assay for restart or completion of DNA synthesis in the collapse region. Numbers indicate the expected DNA signal for pCRISPR (used above) followed by pCRISPR1X (used below). f, Purified DNA samples from t = 120 min in b were digested with XmnI and SacI, then separated on an agarose gel and visualized by autoradiography. g, Quantification of collapse region DNA synthesis from f as in e. Signal was normalized to control fragment Lin1. Expected values for no repair account for replication efficiency and collapse efficiency. Data are presented as the mean ± s.d. (n = 3 independent experiments). Data were analyzed by one-way ANOVA and Tukey’s multiple-comparison method. *P = 0.0112. h, Fork collapse was induced as in a using pCRISPR1X that contained only a single copy of the nCas9 target sequence. Purified DNA samples were digested with XmnI and SacI, then separated on an agarose gel and visualized by autoradiography (Extended Data Fig. 8t). i, Quantification of collapse region DNA synthesis from h as in e. Signal was normalized to control fragment Lin2. Expected values for no repair account for replication efficiency and collapse efficiency. Data are presented as the mean ± s.d. (n = 6 independent experiments). Data were analyzed by one-way ANOVA and Tukey’s multiple-comparison method. **P = 0.0057.Source datanCas9 induction of leading or convergent collapse generated seDSBs (Fig. 5b, lanes 6–10 and 16–20, and Fig. 5c) with ~80% collapse efficiency (Extended Data Fig. 8a–e), slightly lower than for TetR-stabilized SSBs (~90–100%; Extended Data Fig. 1i,q). The DNA structures closely resembled those formed at simple SSBs; leading collapse produced seDSBs converted to RAD51-dependent, RuvC-sensitive HMw species (Fig. 5b, lanes 1–10, Fig. 5c,d and Extended Data Fig. 8f–j,p,q), whereas convergent collapse produced fewer seDSBs that were converted to deDSBs and then HMw species (Fig. 5b, lanes 16–20, and Fig. 5c,d) that were only minimally affected by RAD51 inhibition (Extended Data Fig. 8k–p). nCas9 reduced CM mobility (Fig. 5b, lanes 17–20) by preventing the compensatory supercoiling that otherwise accompanies nucleosome removal during deproteinization70. Full-length products were generated following leading collapse (Fig. 5b, lanes 6–10) but not in the nuclease-dead control (Extended Data Fig. 8r) or following nCas9-induced lagging collapse (Extended Data Fig. 8s), indicating that secondary collapse occurred. As for collapse at simple SSBs, the frequency at which secondary collapse occurs in cells with linear chromosomes remains to be established (Discussion). Thus, the DNA structures generated by nCas9 collapse closely resembled those formed at simple SSBs.We next monitored DNA synthesis following nCas9-induced collapse (Fig. 5e). Leading collapse did not result in detectable collapse region synthesis (Fig. 5f, lanes 1 and 2, and Fig. 5g), as expected. However, nCas9-induced convergent collapse also did not result in detectable synthesis (Fig. 5f, lanes 3 and 4, and Fig. 5g), in contrast to convergent collapse at simple SSBs (Fig. 4f–h), despite efficient DSB resolution. Because Cas9 remains associated with DSBs71 and inhibits their repair72, the multiple nCas9 target sites in pCRISPR could have interfered with completion of DNA synthesis. In support of this interpretation, seDSBs persisted longer when induced by nCas9 (Fig. 5c) than in response to simple SSBs (Fig. 1c). Therefore, we induced leading and convergent collapse at a single nCas9 target site (Fig. 5h,i). Under these conditions, leading and convergent collapse was inefficient but detectable (Extended Data Fig. 8t–w). Collapse region DNA synthesis was detected for convergent collapse but not leading collapse (Fig. 5i, buffer versus LacR). These experiments were highly variable (Fig. 5i), suggesting complex and stochastic repair events. These results support the conclusion that convergent collapsed forks can efficiently complete DNA synthesis while single collapsed forks do not.Convergent collapse leads to resection-dependent DSB repairWe next tested whether resection was required for repair following convergent collapse. We examined fork collapse in mock and MRE11-immunodepleted extracts (Fig. 6a and Extended Data Fig. 9a,b). Resolution of seDSBs was unaffected by MRE11 depletion (Fig. 6b,c), consistent with conversion to deDSBs by fork convergence. In contrast, deDSBs persisted for essentially the entire time course (Fig. 6b,d) and HMw products were abolished (Fig. 6b,e). Persistence of deDSBs but unaltered seDSB resolution strongly supported our conclusion that most seDSBs are resolved by fork convergence rather than HR. The ratio of deDSBs to fully replicated molecules approached 1:1 (Extended Data Fig. 9c), as expected if collapse generated one deDSB and one fully replicated molecule that were both stable. When we directly examined repair, there was no detectable repair in MRE11-immunodepleted extracts (Fig. 6f–h). Thus, MRE11 immunodepletion abolished detectable repair following convergent collapse.Fig. 6: Convergent collapse requires resection for DSB repair.The alternative text for this image may have been generated using AI.Full size imagea, Schematic of convergent collapse in mock or MRE11-immunodepleted extracts. pSSB was replicated with TetR and [α-32P]dATP to allow fork convergence at stabilized SSBs. b, Samples from a were separated on an agarose gel and visualized by autoradiography. c, Quantification of σ structures from b. Data are presented as the mean ± s.d. (n = 4 independent experiments). d, Quantification of linear molecules (deDSBs) from b. Data are presented as the mean ± s.d. (n = 4 independent experiments). e, Quantification of HMw products from b. Data are presented as the mean ± s.d. (n = 4 independent experiments). f, Schematic of the assay for collapse region DNA synthesis following convergent collapse in mock or MRE11-immunodepleted extracts, as in Fig. 4f. g, Purified DNA samples from t = 120 min in b were digested with XmnI and SacI, separated on an agarose gel and visualized by autoradiography. h, Quantification of collapse region DNA synthesis from g as in f. Data are presented as the mean ± s.d. (n = 4 independent experiments). Data were analyzed by paired two-sided t-test. ***P = 0.0004. i, Schematic of convergent collapse in mock or CtIP-immunodepleted extracts as in a. j, Samples from i were separated on an agarose gel and visualized by autoradiography. k, Quantification of σ structures from j. Data are presented as the mean ± s.d. (n = 4 independent experiments). l, Quantification of linear molecules (deDSBs) from j. Data are presented as the mean ± s.d. (n = 4 independent experiments). m, Quantification of HMw products from j. Data are presented as the mean ± s.d. (n = 4 independent experiments). n, Schematic of the assay for collapse region DNA synthesis following convergent collapse in mock or CtIP-immunodepleted extracts as in f. o, Purified DNA samples from t = 120 min in j were digested and visualized as in g. p, Quantification of collapse region DNA synthesis from o as in h. Data are presented as the mean ± s.d. (n = 4 independent experiments). Data were analyzed by paired two-sided t-test (Extended Data Fig. 9). *P = 0.0150 and 0.0101.Source dataTo test whether this reflected a specific role for the early stages of resection, we immunodepleted CtIP (Fig. 6i and Extended Data Fig. 9a,d). The resolution of seDSBs was unaffected (Fig. 6j,k), as for MRE11 depletion. deDSBs persisted (Fig. 6j,l) and HMw products were reduced (Fig. 6j,m). However, in contrast to MRE11 depletion, deDSBs were gradually resolved (60–120 min; Fig. 6j,l), HMw products were eventually produced (60–120 min; Fig. 6j,m) and repair was only modestly diminished (Fig. 6n–p). Codepletion controls confirmed that MRE11 and CtIP immunodepletions targeted distinct components of the initial resection machinery (Extended Data Fig. 9a). Thus, the initial resection machinery is important for deDSB repair following convergent collapse, with MRE11 having a more crucial role than CtIP.Convergent collapse generates error-prone repair products distinct from single-fork collapseTo characterize repair products, we sequenced the outcome of convergent collapse (Fig. 7a–d and Extended Data Fig. 10a–c) compared to a no-replication control (Cdc7i). The major replication-dependent product involved precise deletions corresponding to removal of 1–5 tetO repeats (Fig. 7b,c,e). Furthermore, 77.3% of reads contained deletions, which exceeded the ~50% maximum expected if all repair events produced mutations. The excess deletions arose from replication-independent deletions (19.2% for Cdc7i; Fig. 7b), including preexisting deletions in the parental plasmid (8.7%; Extended Data Fig. 10d). Importantly, the 58.1% frequency of replication-dependent deletions was similar to the expected ~50% maximum. We also observed ~10% substitutions but these increased upon Cdc7i treatment (Fig. 7b), indicating that they were replication independent. Deletions were skewed toward larger sizes; we never observed deletion of all six repeats, indicating that the process was homology dependent (Fig. 7c). The precise nature of these junctions, involving exact removal of the 23-bp repeating unit, meant that we could not determine whether repair used the full 23 bp of homology or a shorter sequence within each repeat.Fig. 7: Convergent collapse generates error-prone repair products distinct from single-fork collapse.The alternative text for this image may have been generated using AI.Full size imagea, Schematic of convergent collapse repair products. Replication of pSSB in the presence of TetR (vehicle) generates deDSBs that undergo repair, producing accurate products and deletions. In the no-replication control (Cdc7i), SSBs remain as unrepaired nicks. b, Frequency of mutation classes following convergent collapse (vehicle) compared to the no-replication control (Cdc7i), determined by Illumina sequencing. Ins, insertions; Del, deletions; Sub, substitutions. Data are presented as the mean ± s.d. (n = 3 independent experiments). Data were analyzed by two-sided Welch’s t-test with Benjamini–Hochberg multiple-comparison correction. ***P = 2.47 × 10−5, 3.68 × 10−4, 5.91 × 10−5, 2.14 × 10−5 and 3.50 × 10−5. c, Frequency of precise tetO repeat deletions (1–5 repeats (Rep.)) from b. Data are presented as the mean ± s.d. (n = 3 independent experiments). Data were analyzed by two-sided Welch’s t-test with Benjamini–Hochberg multiple-comparison correction. **P = 5.38 × 10−4, ***P = 8.30 × 10−5 and 4.59 × 10−5 and ****P = 6.38 × 10−7 and 3.44 × 10−7. d, Frequency of templated insertions (full tetO repeat, 2–3 repeats and 7–8-bp tetO fragment) from b. Data are presented as the mean ± s.d. (n = 3 independent experiments). Data were analyzed by two-sided Welch’s t-test with Benjamini–Hochberg multiple-comparison correction (Extended Data Fig. 10). **P = 3.04 × 10−3 and 4.31 × 10−3. e, Schematic of deletion events involving removal of one or more full tetO repeats from the array. f, Schematics of insertion events: insertion of a full tetO repeat (left) and insertion of a tetO fragment (right). g, Schematic of convergent collapse in mock or MRE11-immunodepleted extracts. MRE11 depletion is predicted to block resection and prevent repair, stabilizing unrepaired deDSBs. h, Frequency of mutation classes following convergent collapse in mock or MRE11-immunodepleted extracts as in b (n = 2 independent experiments). i, Frequency of precise tetO repeat deletions from h as in c (n = 2 independent experiments). j, Frequency of templated insertions from h as in d (n = 2 independent experiments). k, Schematic of leading collapse repair products. Replication of pSSBLEAD in the presence of TetR and LacR (vehicle) generates seDSBs that form stable D-loops. In the no-replication control (Cdc7i), SSBs remain as unrepaired nicks. l, Frequency of mutation classes following leading collapse compared to the no-replication control, as in b. Data are the mean ± s.d. (n = 3 independent experiments). Data were analyzed by two-sided Welch’s t-test with Benjamini–Hochberg multiple-comparison correction. *P = 2.28 × 10−3 and 4.99 × 10−3. m, Frequency of precise tetO repeat deletions following leading collapse from l as in c. Data are presented as the mean ± s.d. (n = 3 independent experiments). Data were analyzed by two-sided Welch’s t-test with Benjamini–Hochberg multiple-comparison correction. *P = 4.59 × 10−3 and **P = 2.36 × 10−4. n, Frequency of templated insertions following leading collapse from l as in d. Data are presented as the mean ± s.d. (n = 3 independent experiments). Data were analyzed by two-sided Welch’s t-test with Benjamini–Hochberg multiple-comparison correction. In all panels, P values are listed in order of appearance, from left to right.Source dataTemplated insertions were also detected at low frequency, involving either one complete tetO repeat or a 7–8-bp tetO fragment at sites of 2–5-bp microhomology (Fig. 7d,f and Extended Data Fig. 10a–c,e). These insertions are characteristic of MMEJ73. Both deletions and insertions were diminished by MRE11 immunodepletion (Fig. 7g–j and Extended Data Fig. 10f–h). Combined with the resection dependence and RAD51 independence of convergent collapse repair (Figs. 4–6), these data indicate that repair proceeds through annealing-dependent DSB repair (single-strand annealing or alternative end joining)74.We next sequenced products from leading collapse (Fig. 7k–n) to test whether convergent collapse products were also detected in this setting. Leading collapse produced replication-dependent tetO deletions at greatly reduced frequency (~11.4%; Fig. 7l). Smaller deletions (1–3 repeats) were not detected and larger deletions (4–5 repeats) were diminished compared to convergent collapse (Fig. 7c,m). Templated insertions were not detected (Fig. 7n and Extended Data Fig. 10i–k). The lack of insertions and deletions was not because of these mechanisms forming palindromes, which represented only 0.023% ± 0.006% (mean ± s.d.) of total reads. This difference in mutation profiles demonstrates that semisynchronous convergent collapse produces repair outcomes distinct from single collapsed forks.DiscussionWe interrogated replication fork collapse in X. laevis egg extracts using both simple SSBs and nCas9 (H840A)-generated SSBs. DSBs were efficiently resolved regardless of whether a single fork collapsed at a leading-strand or lagging-strand SSB or whether convergent forks collapsed (Fig. 8a–c). We provide direct evidence that fork convergence at an SSB involves formation of an seDSB that is converted to a deDSB (Figs. 4 and 5), consistent with previous results21,22,23,27. Single collapsed forks formed stable joint molecules arising from D-loops without detectable restart of DNA synthesis (Fig. 8a,b), while semisynchronous convergent collapse led to efficient completion of DNA synthesis independently of RAD51, generating precise deletions and templated insertions consistent with MMEJ (Fig. 8c). These findings parallel recent work75, where fork collapse in bacteria engages HR to form D-loops that cannot efficiently restart DNA synthesis alone without replicative helicase reloading.Fig. 8: Models for repair outcomes following replication fork collapse at SSBs in X. laevis egg extracts.The alternative text for this image may have been generated using AI.Full size imagea, Leading collapse (single fork). An SSB on the leading-strand template generates a blunt seDSB upon fork encounter. Resection enables RAD51-dependent strand invasion to form a stable D-loop that does not support restart of DNA synthesis. RAD51 promotes end-to-end fusions at leading collapse (‘Discussion’). End-to-end fusions involving microhomology are an alternative, erroneous outcome. Both pathways result in HR-dependent DSB resolution without restart of DNA synthesis. b, Lagging collapse (single fork). An SSB on the lagging-strand template generates an seDSB with a 3′ overhang. The seDSB is similarly converted to a stable D-loop, although whether resection is required was not directly addressed in this study (dashed box). End-to-end fusions are also formed, possibly because RAD51 binding to the preresected 3′ overhang does not inhibit resection of the opposing strand as effectively as RAD51 binding to blunt ends. c, Convergent forks. When a converging fork arrives at an SSB following initial seDSB formation, a second DSB is generated to produce a deDSB. The deDSB undergoes resection and is repaired largely independently of RAD51 through end joining, enabling efficient completion of DNA synthesis. Templated insertions and precise deletions are error-prone outcomes of this repair process. d, Secondary collapse. In the context depicted here, one replisome has stalled near the sister fork (as in Fig. 3). Following leading collapse, extensive 5′–3′ resection of the seDSB extends past the sister fork, removing the lagging-strand template and triggering replisome unloading to generate a full-length molecule. Fork convergence from adjacent origins may then allow replication of the collapse region. D-loop formation at the seDSB inhibits secondary collapse by blocking resection. Whether replisome unloading occurs and whether fork convergence completes replication of the collapse region were not directly addressed in this study (dashed boxes). e, Negative regulation of deDSBs. The secondary collapse mechanism may explain why certain types of lagging-strand nicks (for example, nCas9 (H840A) in Fig. 5 and a previous study22 or Tet-nick in Fig. 1) do not generate deDSBs. Following deDSB formation at a lagging-strand nick, 5′–3′ resection could degrade the 5′ flap generated by unwinding past the nick, converting the deDSB into an seDSB. Continued resection past the sister fork triggers secondary collapse, leading to replisome unloading and fill-in synthesis to produce a full-length molecule. Whether replisome unloading occurs in this context was not directly addressed in this study (dashed box).Our data support the conclusion that the dominant outcome of single-fork collapse is a stable D-loop that does not restart DNA synthesis (Fig. 8a,b). Our inability to detect restart is unlikely to reflect synchrony or density, as high-density fork collapse in human cells does not noticeably alter the response22. Across DNA synthesis measurements (~9%; Extended Data Fig. 1u) and sequencing analysis (~11%; Fig. 2k), our data support an upper bound of ~10% restart likely because of low levels of convergent collapse (Extended Data Fig. 2n–r). Our inability to detect BIR is consistent with evidence that most fork collapse events involve deDSB formation21,22,23,42,43 and that BIR proteins do not appear to be major determinants of PARP inhibitor cytotoxicity76. BIR may be more critical when fork convergence is not possible, such as during mitotic DNA synthesis or alternative lengthening of telomeres37,38,39,40,41. However, we cannot exclude unique properties of our in vitro system.seDSBs formed RAD51-dependent joint molecules that persisted for hours (Figs. 1, 2 and 8a,b), while deDSBs arising from semisynchronous convergent collapse were repaired largely independently of RAD51 (Figs. 5 and 8c and Extended Data Fig. 7). Our in vitro findings differ from cellular findings that collapse repair depends critically on HR21,23,42. X. laevis egg extracts are proficient for HR (Fig. 1a,b)68 but induce convergent collapse with relatively high synchrony between the two collapse events (Fig. 4). Our favored explanation is that the synchrony with which convergent forks collapse has a role, with more synchronous collapse favoring HR-independent pathways. We consistently observed an approximately 10% reduction in HMw species following convergent collapse in the presence of BRC peptide (Extended Data Fig. 6f), indicating that HR can operate but is limited to a minor role. In fission yeast, the balance between BIR and end joining following fork collapse is shifted toward BIR when arrival of the converging fork is delayed27. Additionally, end joining is well documented following fork collapse at a trapped TOP1 complex28,50,51,52, where TOP1 negatively regulates collapse of single forks and favors synchronous collapse of converging forks43. Lastly, enzymatic deDSBs, involving synchronous formation of two DNA ends, can be repaired by end joining throughout the S phase46,47,48. It will be important to determine whether the timing of fork convergence influences DSB repair pathway choice.Convergent collapse produced two classes of error-prone products (Fig. 8c). Templated insertions (~1%; Fig. 7d) are characteristic of MMEJ73. The precise deletions (~99%; Fig. 7b) could arise from either MMEJ77 or SSA (single-strand annealing)78. SSA is generally thought to act on longer homologies but MMEJ can readily generate deletions exceeding 100 bp when canonical nonhomologous end joining (c-NHEJ) is unable to operate79. This condition applies in our system, where lagging collapse generates an extended 3′ overhang that should inhibit c-NHEJ. Therefore, we are unable to determine whether the deletions arose through MMEJ, SSA or both. However, SSA and MMEJ are increasingly recognized as a single mechanistic class distinguished primarily by homology length74,80,81,82; therefore, we conclude that the repair we observe is exclusively annealing-dependent DSB repair74.We are not aware of other examples of precise deletions arising from annealing-dependent DSB repair following fork collapse (Fig. 7). Similar deletions can arise following fork collapse in bacteria through a recombination-dependent mechanism83 and in human cells following enzymatic DSB induction35,84, demonstrating that collapse-driven deletions can occur in vivo. This annealing-dependent pathway could support programmed deletions in a chromosomal context and reduce faithful repair that otherwise limits DSB-based genome editing85.End-to-end fusions arising from single-fork collapse occurred at sites of microhomology (Fig. 2k and Extended Data Fig. 4) and were diminished by RAD51 inhibition for leading collapse (Extended Data Fig. 3e and Fig. 8a) but not for lagging collapse (Extended Data Fig. 3i and Fig. 8b). RAD51 can suppress MMEJ through its nonenzymatic DNA binding86. We speculate that RAD51 promotes these fusions through its dsDNA-binding activity, which may limit resection and facilitate end joining at the blunt DNA end generated by leading collapse. For lagging collapse, RAD51 would not be expected to influence end joining because the native 3′ overhang generated at the moment of collapse can directly anneal and undergo MMEJ without further exonucleolytic processing.Our data also reveal that leading collapse can trigger secondary collapse through extensive resection that disassembles the sister fork (Fig. 8d). We propose that 5′–3′ resection removes the lagging-strand template of the sister fork, triggering replisome removal through the termination pathway6. Secondary collapse is, thus, specific to leading collapse; lagging collapse 5′–3′ resection targets the nascent rather than parental strands. Secondary collapse disassembles the replication bubble, allowing converging forks from adjacent origins to replicate the region. Secondary collapse was abolished when the sister fork was moved further away (Fig. 3i–k) and enhanced by RAD51 inhibition (Fig. 2b), suggesting that it could operate when the sister fork stalls after origin firing (Fig. 3) and be promoted by HR deficiency22,87. This mechanism could also convert deDSBs arising from lagging collapse (as described previously21,22,23) into seDSBs (Fig. 8e) and may explain why deDSBs are not detected for certain types of lagging-strand nicks (for example, nCas9 (H840A) in Fig. 5 and a previous study22 or Tet-nick in Fig. 1). Importantly, our results demonstrate secondary collapse in X. laevis egg extracts but not in cells. A mechanistically similar process was recently described in bacteria88, supporting the idea that secondary collapse can also occur in cells. However, the frequency with which it operates in eukaryotic cells with linear chromosomes and multiple origins remains to be established.Overall, our findings establish that strand specificity of SSBs and fork convergence can determine which repair pathways operate, with implications for understanding genome instability downstream of replication stress, PARP inhibitor therapy and genome engineering.MethodsStatistics and reproducibilityExperimental repeats and statistical analyses are provided in figure legends. Where not specified, experiments were performed independently two times. Sequencing data were analyzed using SciPy (version 1.17.0) and statsmodels (version 0.14.6) in Python (version 3.12.10); all other analyses used GraphPad Prism (version 10). Significance is indicated in figures (NS, not significant). No statistical method was used to predetermine sample size. No data were excluded from the analyses. Experiments were not randomized. The investigators were not blinded to allocation during experiments and outcome assessment.X. laevis egg extractsX. laevis egg extracts (‘Xenopus egg extracts’) were prepared from WT X. laevis male and female frogs, as described previously91. Animal protocols were approved by the Vanderbilt Division of Animal Care and the Institutional Animal Care and Use Committee.Plasmid constructionCommercially available plasmid pET-28b(+) (Novagen), referred to as ‘ctrl’ (pJD1), served as an internal loading control in indicated experiments. Plasmids pSSBLEAD, pSSBLAG, pCRISPR, pSSBLEAD 686 bp and pSSBLEAD-Large are modified derivatives of pJD161. Briefly, pJD161 contains 50× tandem repeats of the lacO sequence that collectively form a ~1,600-bp lacO array. Details on pJD161 construction were published previously60. To create pSSBLEAD and pSSBLAG, blunt-ended DNA duplex JDD27 (Supplementary Table 1) was cloned into pJD161 that was linearized with PsiI (New England Biolabs (NEB)). JDD27 contained 5× tandem Nb.BsmI sites flanked by a single tetO sequence on either side. Blunt-end ligation allowed insertion of JDD27 in both forward and reverse orientations. Ligation products were transformed into DH5α cells and constructs were selected such that digestion with Nb.BsmI would nick either the top (pSSBLAG) or bottom (pSSBLEAD) strand.Plasmids pSSBLEAD 686 bp and pSSBLEAD-Large were constructed from pSSBLEAD. Plasmid pSSBLEAD-Large differs from pSSBLEAD by the addition of a 2.0-kb insert ‘IS1’ (Supplementary Table 1) placed 330 bp downstream of the lacO array, on the opposite side of the tetO array. pSSBLEAD 686 bp differs from pSSBLEAD by the addition of a 300 bp insert ‘IS2’ (Supplementary Table 1) placed 304 bp downstream of the tetO array, placing the location of this insert between the tetO and lacO arrays. Both plasmids pSSBLEAD 686 bp and pSSBLEAD-Large were constructed by GenScript.To create pCRISPR, DNA oligonucleotides JDO143 and JDO144 (Supplementary Table 1) were annealed and cloned into pJD161 that was linearized using PsiI (NEB). Constructs were chosen that contained JDO144 in the top strand and JDO143 in the bottom strand in a 5′–3′ orientation. pCRISPR contained 5× tandem repeats of the target sequences (5′-GGTTGAGTGTTGTTCCAGTT-3′) and (5′-AGATAGGGTTGAGTGTTGTT-3′), targeted by nCas9 (H840A) to generate SSBs in the leading and lagging strand, respectively. Plasmids and antibodies generated in this study are available from the corresponding author upon reasonable request.Protein purificationTetR and LacR were expressed in T7 Express Escherichia coli (NEB). His-tagged TetR was purified by Ni-NTA affinity chromatography and dialyzed. Biotinylated LacR was extracted from the insoluble cell fraction, subjected to polymin P precipitation and ammonium sulfate fractionation, purified on SoftLink monomeric avidin resin (Promega) and dialyzed, as described previously64.Preparation of damaged plasmid templatesTo generate SSB-containing plasmid (for pSSBLEAD, pSSBLAG, pSSBLEAD 686 bp and pSSBLead-Large), 30 μg of plasmid was digested with 75 U of Nb.BsmI (NEB) in 1× NEBuffer r3.1 (NEB) for 1 h 45 min at 37 °C. Nicked DNA was then resolved on a 0.9% agarose gel at 5 V cm−1 and stained with SYBR gold (Invitrogen). The nicked DNA band was excised using a blue-light transilluminator and purified by electroelution. Electroelution was performed by placing the excised gel slice in SnakeSkin dialysis tubing (3.5-kDa molecular weight cutoff (MWCO), 35-mm diameter, Thermo Scientific) containing 1 ml of 1× TBE supplemented with BSA (0.3 mg ml−1 final). The slice was electrophoresed at 5 V cm−1 for 1.5 h. The slice was discarded and purified DNA was dialyzed overnight at 4 °C in 1 L of 10 mM Tris pH 8.0. Dialyzed DNA was concentrated using Amicon centrifugal concentrators (100-kDa MWCO, 0.5 ml). DNA concentration was estimated by resolving purified SSB-containing plasmids alongside a DNA standard purified by extraction with phenol, chloroform and isoamyl alcohol (25:24:1) followed by ethanol precipitation (70% (v/v) final) in the presence of sodium acetate (270 mM final) and glycogen (1% final). SSB-containing plasmids were adjusted to 225 ng μl−1 and stored at −20 °C.To generate 1× or 5× SSBs in DNA using nCas9 (H840A), pCRISPR1X (pJD161) or pCRISPR was incubated during licensing with assembled CRISPR–Cas9 RNP, as described below.Assembly of CRISPR–Cas9 ribonucleoprotein (RNP) complexThe CRISPR–Cas9 RNP complex was assembled by incubation of guide RNA (gRNA) with either Alt-R S.p. dCas9 protein V3 (Integrated DNA Technologies (IDT)) or Alt-R S.p. Cas9 H840A nickase V3 (IDT). gRNA was prepared by mixing Alt-R CRISPR–Cas9 trans-activating CRISPR RNA (tracrRNA) and ATTO550 (IDT) with a tenfold molar excess of Alt-R CRISPR–Cas9 crRNA in nuclease-free duplex buffer (IDT). crRNAs scRNA1 (5′-GGUUGAGUGUUGUUCCAGUU-3′) and scRNA2 (5′-AACAACACUCAACCCUAUCU-3′) targeted leading and lagging strands, respectively. crRNA–tracrRNA mixtures were heated to 95 °C for 5 min and then cooled to room temperature for 1 h. gRNA was stored at −20 °C. Immediately before experiments, RNP was formed by incubating 25 μM Cas9 with 2 μM gRNA in 1× Cas9 dilution buffer (15 mM KCl and 3 mM HEPES, pH 7.5) in 2 μl for ≥ 20 min.DNA replication in Xenopus extractsControl plasmid (pJD1) was fully replicated in extract before fork collapse. Activated high-speed supernatant (HSS) was prepared by incubating HSS with an ATP-regenerating system (ARS; 20 mM phosphocreatine, 2 mM ATP and 5 ng μl−1 creatine phosphokinase) and nocodazole (3 ng μl−1) for 5 min at room temperature. pJD161 (0.1 volumes, 100 ng μl−1) was licensed in 0.9 volumes of activated HSS for 20 min at room temperature. Activated nucleoplasmic extract (NPE) was prepared by supplementing NPE with ARS, dithiothreitol (DTT; 2 mM final) and [α-32P]dATP (350 nM final). The NPE mix was diluted with 1× egg lysis buffer (ELB; 250 mM sucrose, 2.5 mM MgCl2 50 mM KCl and 10 mM HEPES, pH 7.7) to 45% NPE (v/v). Replication was initiated by adding 0.1 volume of licensing mix to 0.9 volumes of 45% NPE mix. After 40 min, 0.05 volumes of the reaction was added to 0.95 volumes of fresh, activated NPE mix (65% in 1× ELB). This control NPE mix was used in collapse experiments. Rereplication of pJD1 did not occur because NPE contains a high concentration of factors that inhibit origin licensing61.Fork collapse was induced by replicating SSB-containing plasmids in Xenopus egg extract. For experiments that used pSSBLEAD, pSSBLAG, pSSBLEAD 686 bp or pSSBLEAD-Large as template, 0.29 volumes of pSSB (225 ng μl−1) was incubated with 0.36 volumes of either TetR (100 μM) or TetR buffer and 0.36 volumes of either LacR (32 μM) or LacR buffer for 1 h at room temperature. These steps were performed immediately before licensing in activated HSS. Repressor-bound plasmid (0.2 volumes) was licensed in 0.8 volumes activated HSS for a final DNA concentration of ~13 ng μl−1 and then incubated for 20 min at room temperature. Replication of pSSB was initiated by addition of one volume of licensing mix to two volumes of control NPE mix. For TetR-stabilized reactions, 0.05 volumes of TetR (100 μM) was added to 0.95 volumes of control NPE mix before replication. Where indicated, BRC peptide92 (18 μM final) or Cdc7 inhibitor PHA-767491 (300 μM; Sigma-Aldrich) was added. Where indicated, aphidicolin (300 μM; Sigma-Aldrich) was added 22.5 min into reactions and tetracycline (100 μM final; Sigma-Aldrich) was added at 0 and 60 min.When pCRISPR or pJD161 was used, 0.57 volumes of plasmid (115 ng μl−1) was bound with 0.43 volumes of either LacR (32 μM) or LacR buffer for 1 h at room temperature. Assembled RNP was diluted 1.75-fold in H2O and 0.57 volumes of RNP was incubated with 0.43 volumes of LacR (32 μM) or LacR buffer for 1 h at room temperature. Repressor-bound plasmid (0.1 volumes) was licensed with 0.8 volumes of activated HSS for 20 min at room temperature. After 20 min, 0.1 volumes of RNP was added and licensing continued for another 10 min (~30 min in total). Replication was initiated by adding one of volume licensing mix to two of volumes control NPE mix.At indicated time points, reactions were sampled into either 20 volumes of replication stop solution (8 mM EDTA, 0.13% phosphoric acid, 10% Ficoll, 5% SDS, 0.2% bromophenol blue and 80 mM Tris pH 8.0), which stops replication reactions and serves as a DNA loading dye, or extraction stop solution (1% SDS, 25 mM EDTA and 50 mM Tris-HCl pH 7.5), which also stops reactions but is compatible with downstream processing (that is, DNA purification). Stopped reactions were treated with RNase A (182 ng μl−1 final) and then proteinase K (1.6 mg ml−1 final). Replication stop samples were analyzed by agarose gel electrophoresis at 5 V cm−1. Extraction stop samples were purified by either column (Monarch PCR and DNA cleanup kit, NEB) to remove HMw DNA species (as described previously93) or extraction with phenol and chloroform followed by ethanol precipitation (70% (v/v) final) in the presence of sodium acetate (270 mM final) and glycogen (1% final) to purify total DNA. Purified DNA samples were resuspended in 6 μl of 10 mM Tris-HCl (pH 8.0).Experiments conducted in Figs. 1–4, 6 and 7 and Extended Data Figs. 1–7, 9 and 10 used either pSSBLEAD or pSSBLAG as the template for replication. Experiments shown in Fig. 3i–k additionally used pSSBLEAD-Large as a template while the experiments shown in Extended Data Fig. 2i–r used pSSBLEAD 686 bp as a template. Experiments conducted in Fig. 5 and Extended Data Fig. 8 used pCRISPR or pJD161, alternatively called pCRISPR1X, as a template. Control plasmid was included as a loading control in all experiments with the following exceptions, for which it was omitted: Fig. 3e–h and Extended Data Figs. 3q, 5a–e, 6g–k and 7a–f.Antibodies and peptidesAntibodies targeting Xenopus MRE11 and RAD51 were raised by New England Peptide by immunizing rabbits with Ac-CDPFKKSGPSRRGRR-OH for MRE11 and polypeptide PP1 (Supplementary Table 1) for RAD51. For targeting Xenopus CtIP, commercial anti-CtIP antibody MAEB1072 (Sigma-Aldrich) was used. The BRC4 peptide comprises residues 1517–1551 of BRCA2 (ref. 68). Bacterial cultures carrying pGEX-4T-3 (WT or mutant*) were grown at 37 °C to an optical density at 600 nm of 0.5 and then protein expression was induced with IPTG for 3 h at 37 °C. Following cell lysis and clarification, the GST-tagged peptide was purified using glutathione affinity chromatography and recovered by batch elution. Lastly, the eluate was concentrated and subjected to buffer exchange using a centrifugal filter unit.Western blotting and immunodepletionProtein-A-coupled magnetic beads (Dynabeads protein A, 30 μg μl−1) were bound with 0.5 μg of control IgGs or anti-MRE11, anti-RAD51 or anti-CtIP antibody per 1 μl of magnetic beads. For each immunodepletion round, 1.29 volumes of antibody-bound beads were incubated with two volumes of HSS or one volume of NPE for 20 min at room temperature with end-over-end rotation. This was repeated twice for HSS and NPE, for a total of three rounds, as described previously94. Depleted extracts were isolated and used for either DNA replication or western blotting. For western analysis, the secondary antibody was horseradish peroxidase (HRP)-conjugated goat anti-rabbit antibody (Jackson Immunoresearch, 111-035-003) for MRE11 and RAD51 and goat anti-mouse for CtIP (Jackson Immunoresearch, 315-035-00). The following antibody dilutions were used for western blotting: MRE11, RAD51 and CtIP (1:5000) and HRP goat anti-rabbit or mouse (1:30,000). Images were acquired using Amersham Imager 600 (GE Healthcare).Analysis of replication and repair intermediatesTo monitor replication, samples that had been collected into replication stop were separated on a 1% agarose gel at 5 V cm−1. Radiolabeled DNA was detected by phosphorimaging, which measured the incorporation of radiolabeled nucleotides, and DNA signal was measured using ImageJ. To measure DNA synthesis, individual whole-lane signals were normalized to the loading control in each lane. Normalized signals were expressed relative to the maximum whole-lane signal across all time points and conditions. Because of nicking of the internal loading control by nCas9 (H840A), DNA synthesis was not normalized to the loading control for CRISPR–Cas9 experiments (Fig. 5 and Extended Data Fig. 8). DNA synthesis was also not normalized to loading control for the MRE11 depletion experiment in Fig. 3 because of omission of the loading control to more clearly visualize CMs. Additionally, DNA synthesis was not normalized to the control for depletion experiments in Fig. 6 because of the overlap of the control with linear molecules.The abundance of individual DNA species was expressed as a percentage of whole-lane signal (%). Collapse efficiency in single-fork conditions was calculated as 1 − (θ% in collapse conditions/θ% in control conditions) × 100 for each time point. The principle behind this formula is that replication fork structures are reduced in a manner that is directly proportional to the frequency of collapse because of conversion of θ to σ. For converging fork experiments in which plasmids pSSBLEAD, pSSBLAG or pCRISPR were used as a template, collapse efficiency was calculated as (((((nCM% + scCM%)/Lin%) − 1)/2) + 1)−1 × 100%. The principle behind this formula is that, for converging forks, each convergent collapse event should generate an nCM or scCM and a linear molecule while completion of DNA synthesis without collapse should generate two nCMs or scCMs. For converging fork experiments in which plasmid pJD161 (that is, pCRISPR1X) was used as a template, the aforementioned formula was not used to calculate collapse efficiency, as the linear band generated from collapse migrated at the same position as the supercoiled control plasmid. Collapse efficiency was instead calculated using a less direct approach. First, CMs were normalized to the control using CMnorm = (nCM% + scCM% in collapse conditions)/(maximum nCM% + scCM% in control conditions). Second, CMnorm was used to calculate collapse efficiency across all time points by applying the following formula: [1 − ((CMnorm − 0.5)/0.5))] × 100%. The principle behind this formula is that each collapse event should lead to loss of one CM of the two that would be formed if replication proceeded without collapse.Analysis of DNA synthesis in the collapse regionTo monitor synthesis in the collapse region, purified DNA samples were digested with 0.4 U per μl XmnI and 0.4 U per μl SacI in 1× rCutSmartBuffer (NEB) for 1 h at 37 °C. Digested products were then separated under either native or denaturing agarose gel conditions. For native separation, 6× native loading buffer (Ficoll, EDTA, SDS, bromophenol blue and Tris-HCl (1 M, pH 8)) was added to samples at 1× concentration before separation on a 1% agarose gel at 5 V cm−1. For separation under denaturing conditions, digests were stopped by addition of EDTA to a final concentration of 30 mM before the addition of 6× alkaline loading buffer (Ficoll, EDTA, xylene cyanol, bromocresol green and NaOH (10 N)) to a final concentration of 1×. Digests were then separated on a 1.5% alkaline gel at 1.5 V cm−1. The gel was then neutralized with gentle agitation in 7% trichloroacetic acid. Dried gels were imaged on the Amersham Typhoon Scanner (GE Healthcare).Synthesis in the collapse region was calculated using the formula (((a/b) in each sample condition/(average (a/b) in control—Tet buffer or vehicle conditions)) × 100%, where a is the signal of the collapsed fragment and b is the signal of pJD1 control fragment 1. For single-fork collapse experiments that used pSSBLEAD, pSSBLAG or pSSBLEAD 686 bp as a template, ‘expected no repair’ was calculated as (((747/2,212) × fraction of molecules that collapsed) + fraction of molecules that did not collapse) × DNA synthesis percentage within the collapse region in control conditions. In the above formula, the value (747/2,212) corresponds to the amount of nascent synthesis within the collapsed region, assuming no repair. To adjust the expected no repair for differences in replication efficiency, the calculated value was then multiplied by DNA synthesis at t = 120 in collapsed conditions/DNA synthesis at t = 120 in control conditions). For nCas9-induced single-fork collapse experiments that used template plasmids pCRISPR or pJD161, alternatively termed pCRISPR1X, the same calculations as above were used except that 724/2,336 and 625/1,944 were used, respectively, as the expected nascent synthesis within the collapsed region, assuming no repair of collapsed ends. For collapse under converging fork conditions, the same calculations were applied, except that 0.5 was used as the expected signal assuming no repair because the parental strand lacking a SSB was fully replicated. The expected no repair + conv. values given in Extended Data Fig. 2q,r were calculated by adding an additional 12.7% (for pSSBLEAD 386 bp) and 20.5% (for pSSBLEAD 686 bp) repair signal to expected no repair values. These percentages represent the signal beyond the expected no repair, corresponding to the repair signal that would be expected from forks converging on the SSB as the SSB is moved further away from the lacO array. For Extended Data Fig. 1u, the maximum undetectable restart was calculated by expressing the upper bound of the 95% confidence interval of the unpaired difference (TetR − expected no repair) as a percentage of the remaining synthesis capacity (100% − ENR mean). Supplementary Table 2 shows a worked example of how DNA synthesis in the collapse region was calculated.Enzymatic analysis of replication and repair intermediatesWhere indicated, purified DNA was subjected to the following enzymatic digestions: digestion with 0.8 U per μl AlwNI, XmnI or SapI in 1× rCutSmartBuffer (NEB) for 1 h at 37 °C, double digestion with 0.4 U per μl SacI and 0.4 U per μl KpnI in 1× rCutSmartBuffer (NEB) for 1 h at 37 °C, single digestion with 0.12 U per μl hTOP2α in 1× topoisomerase II assay buffer (Topogen) for 15 min at 37 °C or single digestion with 0.5 nM to 1.5 nM RuvC in 1× NEBuffer 2.1(NEB) supplemented with DTT (1 mM final) and Tris pH 8.0 (40 mM final) for 1 h at 37 °C. hTOP2α digestion was stopped by the addition of 0.5 volumes of TopSTOP solution (3% SDS and 2 mg ml−1 proteinase K). RuvC digestion was stopped by the addition of 0.2 volumes of RuvCSTOP solution (3% SDS, 240 mM EDTA and 3 mg ml−1 proteinase K). Radiolabeled digestion products were separated on a 1% agarose gel at 5 V cm−1 and detected by phosphorimaging.Amplicon sequencing of collapsed productsSamples from Fig. 7 experiments were quenched at 120 min, purified as above and processed for amplicon sequencing. Purified DNA (~2.8 ng or ~10 ng) was amplified in a 20-μl reaction using the NEB Phusion high-fidelity PCR kit (final: 1× Phusion HF buffer, 200 μM dNTPs, 500 μM each of SCP1 (5′-GACGTTGGAGTCCACGTTCTTTAATAGTG-3′) and SCP2 (5′-AATGGCGAATGGAAATTGTAAGCGTT-3′) primers and 0.4 U of Phusion DNA polymerase). Amplification followed a touchdown PCR program using the following settings: denaturation at 98 °C for 30 s, followed by ten cycles of denaturation at 98 °C for 10 s, annealing at 72 °C for 15 s (decreasing by 0.5 °C per cycle) and extension at 72 °C for 10 s. Subsequently, seven cycles were performed with denaturation at 98 °C for 10 s, annealing at 67 °C for 15 s and extension at 72 °C for 10 s, followed by a final extension at 72 °C for 7 min. PCR products were purified using the NEB Monarch PCR and DNA cleanup kit according to the manufacturer’s instructions, quantified on a 1% agarose 1× TBE gel stained with SYBR gold and sequenced by Genewiz Amplicon-EZ next-generation sequencing. Furthermore, unreplicated and unnicked pSSBLEAD plasmid (~10 ng) was amplified and sequenced under identical conditions. This was performed to establish a ‘mutation event’ baseline. Sequencing results from this plasmid were subtracted from the experimental sequencing data (that is, background subtraction of pSSBLEAD was performed).Sequencing analysis of collapsed productsTo analyze sequencing reads, agentic coding (Claude Opus 4.6, Anthropic) was used to generate and implement a custom analysis pipeline. The pipeline performs paired-end read merging, semiglobal alignment to the 276-bp reference template, mutation calling (insertions, deletions and substitutions), classification of mutations relative to the tetO repeat array and detection of palindromic reads containing inversion junctions. Reads were assigned to 20 prespecified mutation bins. Condition means were compared using Welch’s t-test with Benjamini–Hochberg correction applied independently across all 20 bins for each comparison. The complete pipeline source code, configuration parameters, reference sequences and intermediate outputs are provided on Zenodo (https://doi.org/10.5281/zenodo.19687676)95.Pipeline validation used two independent strategies. First, we performed orthogonal reimplementation; all merged reads were realigned using EMBOSS needle96 and independently reclassified using minimal regex-based descriptors derived without access to the original pipeline code, following N-version programming principles97,98,99,100. Second, we performed round-trip verification; classification labels were stripped from pipeline outputs and an independent analysis recovered the classification scheme from the binned reads alone, analogous to round-trip correctness testing for code generated using a large language model101,102. Minimal classifiers (regex patterns and length constraints) were derived for each mutation category from the recovered bin characterizations and applied as a cross-check103. Validation inputs and outputs are provided on Zenodo (https://doi.org/10.5281/zenodo.19687676)95.End labeling of DNADigested DNA products were run alongside a radiolabeled DNA ladder. Radiolabeled ladder was generated by treating a 2-μg DNA ladder (NEB) with 10 U of T4 PNK (NEB) in 1× PNK reaction buffer (NEB) supplemented with 1.33 mM [γ-32P]ATP in a 20-μl volume. The reaction proceeded for 45 min at 37 °C and then purified using the Monarch PCR and DNA cleanup kit (NEB) according to the manufacturer’s instructions.Materials availabilityPlasmids and antibodies generated in this study are available from the corresponding author upon request.Reporting summaryFurther information on research design is available in the Nature Portfolio Reporting Summary linked to this article.Data availabilityRaw sequencing data generated in this study were deposited to the National Center for Biotechnology Information Sequence Read Archive under BioProject PRJNA1454974. Processed data, intermediate analysis files and all other datasets supporting the findings of this study were deposited to Zenodo (https://doi.org/10.5281/zenodo.19687676)95. Any additional information is available from the corresponding author on request. Source data are provided with this paper.Code availabilityAll custom code used for sequencing analysis (paired-end read merging, semiglobal alignment, mutation calling and classification and palindromic read detection) and for pipeline validation (orthogonal reimplementation using EMBOSS needle, round-trip verification and minimal regex-based classifiers) was deposited to Zenodo alongside the associated data (https://doi.org/10.5281/zenodo.19687676)95 under an open-source license approved by the Open Source Initiative.ReferencesCaldecott, K. W. Causes and consequences of DNA single-strand breaks. Trends Biochem. Sci. 49, 68–78 (2024).Article CAS PubMed Google Scholar Lindahl, T. & Barnes, D. E. Repair of endogenous DNA damage. Cold Spring Harb. Symp. Quant. Biol. 65, 127–133 (2000).Article CAS PubMed Google Scholar Nielsen, I. et al. A Flp-nick system to study repair of a single protein-bound nick in vivo. Nat. Methods 6, 753–757 (2009).Article CAS PubMed Google Scholar Strumberg, D. et al. Conversion of topoisomerase I cleavage complexes on the leading strand of ribosomal DNA into 5′-phosphorylated DNA double-strand breaks by replication runoff. Mol. Cell Biol. 20, 3977–3987 (2000).Article CAS PubMed PubMed Central Google Scholar Mayle, R. et al. Mus81 and converging forks limit the mutagenicity of replication fork breakage. Science 349, 742–747 (2015).Article CAS PubMed PubMed Central Google Scholar Vrtis, K. B. et al. Single-strand DNA breaks cause replisome disassembly. Mol. Cell 81, 1309–1318 (2021).Article CAS PubMed PubMed Central Google Scholar Kuzminov, A. Single-strand interruptions in replicating chromosomes cause double-strand breaks. Proc. Natl Acad. Sci. USA 98, 8241–8246 (2001).Article CAS PubMed PubMed Central Google Scholar Cortés-Ledesma, F. & Aguilera, A. Double-strand breaks arising by replication through a nick are repaired by cohesin-dependent sister-chromatid exchange. EMBO Rep. 7, 919–926 (2006).Article PubMed PubMed Central Google Scholar Vilenchik, M. M. & Knudson, A. G. Endogenous DNA double-strand breaks: production, fidelity of repair, and induction of cancer. Proc. Natl Acad. Sci. USA 100, 12871–12876 (2003).Article CAS PubMed PubMed Central Google Scholar Bhowmick, R. et al. RAD51 protects human cells from transcription-replication conflicts. Mol. Cell 82, 3366–3381 (2022).Article CAS PubMed Google Scholar Henckaerts, E. et al. Site-specific integration of adeno-associated virus involves partial duplication of the target locus. Proc. Natl Acad. Sci. USA 106, 7571–7576 (2009).Article CAS PubMed PubMed Central Google Scholar Ortega, P. et al. Mechanism of DNA replication fork breakage and PARP1 hyperactivation during replication catastrophe. Sci. Adv. 11, eadu0437 (2025).Article CAS PubMed PubMed Central Google Scholar Hanzlikova, H. et al. The importance of poly(ADP-ribose) polymerase as a sensor of unligated Okazaki fragments during DNA replication. Mol. Cell 71, 319–331 (2018).Article CAS PubMed PubMed Central Google Scholar Fugger, K. et al. Targeting the nucleotide salvage factor DNPH1 sensitizes BRCA-deficient cells to PARP inhibitors. Science 372, 156–165 (2021).Article CAS PubMed PubMed Central Google Scholar Serrano-Benitez, A. et al. Unrepaired base excision repair intermediates in template DNA strands trigger replication fork collapse and PARP inhibitor sensitivity. EMBO J. 42, e113190 (2023).Article CAS PubMed PubMed Central Google Scholar Huang, M. E. et al. C-to-G editing generates double-strand breaks causing deletion, transversion and translocation. Nat. Cell Biol. 26, 294–304 (2024).Article CAS PubMed Google Scholar Feng, Y. L. et al. DNA nicks induce mutational signatures associated with BRCA1 deficiency. Nat. Commun. 13, 4285 (2022).Article CAS PubMed PubMed Central Google Scholar da Costa, A., Chowdhury, D., Shapiro, G. I., D’Andrea, A. D. & Konstantinopoulos, P. A. Targeting replication stress in cancer therapy. Nat. Rev. Drug Discov. 22, 38–58 (2023).Article PubMed Google Scholar Farmer, H. et al. Targeting the DNA repair defect in BRCA mutant cells as a therapeutic strategy. Nature 434, 917–921 (2005).Article CAS PubMed Google Scholar Simoneau, A., Xiong, R. & Zou, L. The trans cell cycle effects of PARP inhibitors underlie their selectivity toward BRCA1/2-deficient cells. Genes Dev. 35, 1271–1289 (2021).Article PubMed PubMed Central Google Scholar Kimble, M. T. et al. Repair of replication-dependent double-strand breaks differs between the leading and lagging strands. Mol. Cell 85, 61–77 (2025).Article CAS PubMed Google Scholar Pavani, R. et al. Structure and repair of replication-coupled DNA breaks. Science 385, eado3867 (2024).Article CAS PubMed PubMed Central Google Scholar Elango, R. et al. Two-ended recombination at a Flp-nickase-broken replication fork. Mol. Cell 85, 78–90 (2025).Article CAS PubMed Google Scholar Vriend, L. E. M. et al. Distinct genetic control of homologous recombination repair of Cas9-induced double-strand breaks, nicks and paired nicks. Nucleic Acids Res. 44, 5204–5217 (2016).Article CAS PubMed PubMed Central Google Scholar Cortés-Ledesma, F., Tous, C. & Aguilera, A. Different genetic requirements for repair of replication-born double-strand breaks by sister-chromatid recombination and break-induced replication. Nucleic Acids Res. 35, 6560–6570 (2007).Article PubMed PubMed Central Google Scholar Hashimoto, Y., Puddu, F. & Costanzo, V. RAD51- and MRE11-dependent reassembly of uncoupled CMG helicase complex at collapsed replication forks. Nat. Struct. Mol. Biol. 19, 17–24 (2011).Article PubMed PubMed Central Google Scholar Xu, Y. et al. DNA nicks in both leading and lagging strand templates can trigger break-induced replication. Mol. Cell 85, 91–106 (2025).Article CAS PubMed Google Scholar Li, S. et al. PIF1 helicase promotes break-induced replication in mammalian cells. EMBO J. 40, e104509 (2021).Article CAS PubMed PubMed Central Google Scholar Lee, R. S., Twarowski, J. M. & Malkova, A. Stressed? Break-induced replication comes to the rescue! DNA Repair 142, 103759 (2024).Article CAS PubMed PubMed Central Google Scholar Bosco, G. & Haber, J. E. Chromosome break-induced DNA replication leads to nonreciprocal translocations and telomere capture. Genetics 150, 1037–1047 (1998).Article CAS PubMed PubMed Central Google Scholar Davis, A. P. & Symington, L. S. RAD51-dependent break-induced replication in yeast. Mol. Cell Biol. 24, 2344–2351 (2004).Article CAS PubMed PubMed Central Google Scholar Morrow, D. M., Connelly, C. & Hieter, P. ‘Break copy’ duplication: a model for chromosome fragment formation in Saccharomyces cerevisiae. Genetics 147, 371–382 (1997).Article CAS PubMed PubMed Central Google Scholar Malkova, A., Ivanov, E. L. & Haber, J. E. Double-strand break repair in the absence of RAD51 in yeast: a possible role for break-induced DNA replication. Proc. Natl Acad. Sci. USA 93, 7131–7136 (1996).Article CAS PubMed PubMed Central Google Scholar Malkova, A., Naylor, M. L., Yamaguchi, M., Ira, G. & Haber, J. E. RAD51-dependent break-induced replication differs in kinetics and checkpoint responses from RAD51-mediated gene conversion. Mol. Cell Biol. 25, 933–944 (2005).Article CAS PubMed PubMed Central Google Scholar Hu, Q. et al. Break-induced replication plays a prominent role in long-range repeat-mediated deletion. EMBO J. 38, e101751 (2019).Article CAS PubMed PubMed Central Google Scholar Al-Zain, A. M. & Symington, L. S. The dark side of homology-directed repair. DNA Repair 106, 103181 (2021).Article CAS PubMed Google Scholar Minocherhomji, S. et al. Replication stress activates DNA repair synthesis in mitosis. Nature 528, 286–290 (2015).Article CAS PubMed Google Scholar Bhowmick, R., Minocherhomji, S. & Hickson, I. D. RAD52 facilitates mitotic DNA synthesis following replication stress. Mol. Cell 64, 1117–1126 (2016).Article CAS PubMed Google Scholar Sonneville, R. et al. TRAIP drives replisome disassembly and mitotic DNA repair synthesis at sites of incomplete DNA replication. eLife 8, e48686 (2019).Article CAS PubMed PubMed Central Google Scholar Dilley, R. L. et al. Break-induced telomere synthesis underlies alternative telomere maintenance. Nature 539, 54–58 (2016).Article CAS PubMed PubMed Central Google Scholar Zhang, T. et al. Break-induced replication orchestrates resection-dependent template switching. Nature 619, 201–208 (2023).Article CAS PubMed PubMed Central Google Scholar Li, L., Scott, W. S., Khristich, A. N., Armenia, J. F. & Mirkin, S. M. Recurrent DNA nicks drive massive expansions of (GAA)n repeats. Proc. Natl Acad. Sci. USA 121, e2413298121 (2024).Article CAS PubMed PubMed Central Google Scholar Westhorpe, R., Roske, J. J. & Yeeles, J. T. P. Mechanisms controlling replication fork stalling and collapse at topoisomerase 1 cleavage complexes. Mol. Cell 84, 3469–3481 (2024).Article CAS PubMed PubMed Central Google Scholar Fábián, Z. et al. PARP1-dependent DNA–protein crosslink repair. Nat. Commun. 15, 6641 (2024).Article PubMed PubMed Central Google Scholar Johansson, E. & Diffley, J. F. X. Unchecked nick ligation can promote localized genome re-replication. Curr. Biol. 31, R710–R711 (2021).Article CAS PubMed PubMed Central Google Scholar Pierce, A. J., Hu, P., Han, M., Ellis, N. & Jasin, M. Ku DNA end-binding protein modulates homologous repair of double-strand breaks in mammalian cells. Genes Dev. 15, 3237–3242 (2001).Article CAS PubMed PubMed Central Google Scholar Mao, Z., Bozzella, M., Seluanov, A. & Gorbunova, V. DNA repair by nonhomologous end joining and homologous recombination during cell cycle in human cells. Cell Cycle 7, 2902–2906 (2008).Article CAS PubMed PubMed Central Google Scholar Karanam, K., Kafri, R., Loewer, A. & Lahav, G. Quantitative live cell imaging reveals a gradual shift between DNA repair mechanisms and a maximal use of HR in mid S phase. Mol. Cell 47, 320–329 (2012).Article CAS PubMed PubMed Central Google Scholar Wang, Z. et al. DNA polymerase θ (POLQ) is important for repair of DNA double-strand breaks caused by fork collapse. J Biol Chem 294, 3909–3919 (2019).Article CAS PubMed PubMed Central Google Scholar Whelan, D. R. et al. Super-resolution visualization of distinct stalled and broken replication fork structures. PLoS Genet. 16, e1009256 (2020).Article CAS PubMed PubMed Central Google Scholar Balmus, G. et al. ATM orchestrates the DNA-damage response to counter toxic non-homologous end-joining at broken replication forks. Nat. Commun. 10, 87 (2019).Article CAS PubMed PubMed Central Google Scholar Britton, S. et al. ATM antagonizes NHEJ proteins assembly and DNA-ends synapsis at single-ended DNA double strand breaks. Nucleic Acids Res. 48, 9710–9723 (2020).Article CAS PubMed PubMed Central Google Scholar Saini, N. et al. Migrating bubble during break-induced replication drives conservative DNA synthesis. Nature 502, 389–392 (2013).Article CAS PubMed PubMed Central Google Scholar Wilson, M. A. et al. Pif1 helicase and Polδ promote recombination-coupled DNA synthesis via bubble migration. Nature 502, 393–396 (2013).Article CAS PubMed PubMed Central Google Scholar Donnianni, R. A. & Symington, L. S. Break-induced replication occurs by conservative DNA synthesis. Proc. Natl Acad. Sci. USA 110, 13475–13480 (2013).Article CAS PubMed PubMed Central Google Scholar Atari, A., Jiang, H. & Greenberg, R. A. Mechanisms and genomic implications of break-induced replication. Nat. Struct. Mol. Biol. 32, 1871–1882 (2025).Article CAS PubMed PubMed Central Google Scholar Asai, T., Bates, D. B. & Kogoma, T. DNA replication triggered by double-stranded breaks in E. coli: dependence on homologous recombination functions. Cell 78, 1051–1061 (1994).Article CAS PubMed PubMed Central Google Scholar Costantino, L. et al. Break-induced replication repair of damaged forks induces genomic duplications in human cells. Science 343, 88–91 (2014).Article CAS PubMed Google Scholar Sotiriou, S. K. et al. Mammalian RAD52 functions in break-induced replication repair of collapsed DNA replication forks. Mol. Cell 64, 1127–1134 (2016).Article CAS PubMed PubMed Central Google Scholar Vontalge, E. J., Kavlashvili, T., Dahmen, S. N., Cranford, M. T. & Dewar, J. M. Control of DNA replication in vitro using a reversible replication barrier. Nat. Protoc. 19, 1940–1983 (2024).Article CAS PubMed PubMed Central Google Scholar Walter, J., Sun, L. & Newport, J. Regulated chromosomal DNA replication in the absence of a nucleus. Mol. Cell 1, 519–529 (1998).Article CAS PubMed Google Scholar Räschle, M. et al. Proteomics reveals dynamic assembly of repair complexes during bypass of DNA cross-links. Science 348, 1253671 (2015).Article PubMed PubMed Central Google Scholar Hoogenboom, W. S., Klein Douwel, D. & Knipscheer, P. Xenopus egg extract: a powerful tool to study genome maintenance mechanisms. Dev. Biol 428, 300–309 (2017).Article CAS PubMed Google Scholar Conwell, S. C., Cranford, M. T., Kavlashvili, T. & Dewar, J. M. Replication fork collapse in vitro using Xenopus egg extracts. Methods Enzymol. 672, 317–338 (2022).Article CAS PubMed Google Scholar Wu, R. A. et al. TRAIP is a master regulator of DNA interstrand crosslink repair. Nature 567, 267–272 (2019).Article CAS PubMed PubMed Central Google Scholar Deng, L. et al. Mitotic CDK promotes replisome disassembly, fork breakage, and complex DNA rearrangements. Mol. Cell 73, 915–929 (2019).Article CAS PubMed PubMed Central Google Scholar Carreira, A. et al. The BRC repeats of BRCA2 modulate the DNA-binding selectivity of RAD51. Cell 136, 1032–1043 (2009).Article CAS PubMed PubMed Central Google Scholar Long, D. T., Räschle, M., Joukov, V. & Walter, J. C. Mechanism of RAD51-dependent DNA interstrand cross-link repair. Science 333, 84–87 (2011).Article CAS PubMed PubMed Central Google Scholar Benson, F. E. & West, S. C. Substrate specificity of the Escherichia coli RuvC protein. Resolution of three- and four-stranded recombination intermediates. J. Biol. Chem. 269, 5195–5201 (1994).Article CAS PubMed Google Scholar Walter, J. & Newport, J. Initiation of eukaryotic DNA replication: origin unwinding and sequential chromatin association of Cdc45, RPA, and DNA polymerase α. Mol. Cell 5, 617–627 (2000).Article CAS PubMed Google Scholar Reginato, G. et al. HLTF disrupts Cas9–DNA post-cleavage complexes to allow DNA break processing. Nat. Commun. 15, 5789 (2024).Article CAS PubMed PubMed Central Google Scholar Brinkman, E. K. et al. Kinetics and fidelity of the repair of Cas9-induced double-strand DNA breaks. Mol. Cell 70, 801–813 (2018).Article CAS PubMed PubMed Central Google Scholar Schimmel, J., van Schendel, R., den Dunnen, J. T. & Tijsterman, M. Templated insertions: a smoking gun for polymerase θ-mediated end joining. Trends Genet 35, 632–644 (2019).Article CAS PubMed Google Scholar Ceccaldi, R., Rondinelli, B. & D’Andrea, A. D. Repair pathway choices and consequences at the double-strand break. Trends Cell Biol 26, 52–64 (2016).Article CAS PubMed Google Scholar Winterhalter, C., Stratton, K. J., Fenyk, S. & Murray, H. Rescuing the bacterial replisome at a nick requires recombinational repair and helicase reloading. Nat. Commun. 16, 11633 (2025).Article CAS PubMed PubMed Central Google Scholar Olivieri, M. et al. A genetic map of the response to DNA damage in human cells. Cell 182, 481–496 (2020).Article CAS PubMed PubMed Central Google Scholar McVey, M. & Lee, S. E. MMEJ repair of double-strand breaks (director’s cut): deleted sequences and alternative endings. Trends Genet. 24, 529–538 (2008).Article CAS PubMed PubMed Central Google Scholar Bhargava, R., Onyango, D. O. & Stark, J. M. Regulation of single-strand annealing and its role in genome maintenance. Trends Genet. 32, 566–575 (2016).Article CAS PubMed PubMed Central Google Scholar Wyatt, D. W. et al. Essential roles for polymerase θ-mediated end joining in the repair of chromosome breaks. Mol. Cell 63, 662–673 (2016).Article CAS PubMed PubMed Central Google Scholar Sallmyr, A. & Tomkinson, A. E. Repair of DNA double-strand breaks by mammalian alternative end-joining pathways. J. Biol. Chem. 293, 10536–10546 (2018).Article CAS PubMed PubMed Central Google Scholar Decottignies, A. Microhomology-mediated end joining in fission yeast is repressed by Pku70 and relies on genes involved in homologous recombination. Genetics 176, 1403–1415 (2007).Article CAS PubMed PubMed Central Google Scholar Verma, P. & Greenberg, R. A. Noncanonical views of homology-directed DNA repair. Genes Dev. 30, 1138–1154 (2016).Article CAS PubMed PubMed Central Google Scholar Standage-Beier, K., Zhang, Q. & Wang, X. Targeted large-scale deletion of bacterial genomes using CRISPR-Nickases. ACS Synth. Biol. 4, 1217–1225 (2015).Article CAS PubMed PubMed Central Google Scholar Mendez-Dorantes, C., Bhargava, R. & Stark, J. M. Repeat-mediated deletions can be induced by a chromosomal break far from a repeat, but multiple pathways suppress such rearrangements. Genes Dev. 32, 524–536 (2018).Article CAS PubMed PubMed Central Google Scholar Ben-Tov, D. et al. Uncovering the dynamics of precise repair at CRISPR/Cas9-induced double-strand breaks. Nat. Commun. 15, 5096 (2024).Article CAS PubMed PubMed Central Google Scholar So, A. et al. RAD51 protects against nonconservative DNA double-strand break repair through a nonenzymatic function. Nucleic Acids Res. 50, 2651–2666 (2022).Article CAS PubMed Google Scholar Jakobsen, K. P. et al. Minimal resection takes place during break-induced replication repair of collapsed replication forks and is controlled by strand invasion. Cell Rep. 26, 836–844 (2019).Article CAS PubMed Google Scholar Cooke, M. B. et al. Transcription–replication collisions trigger high-fidelity replication reset. Nucleic Acids Res. 53, gkaf1227 (2025).Article CAS PubMed PubMed Central Google Scholar Hyrien, O. & Méchali, M. Plasmid replication in Xenopus eggs and egg extracts: a 2D gel electrophoretic analysis. Nucleic Acids Res. 20, 1463–1469 (1992).Article CAS PubMed PubMed Central Google Scholar Dewar, J. M., Low, E., Mann, M., Räschle, M. & Walter, J. C. CRL2Lrr1 promotes unloading of the vertebrate replisome from chromatin during replication termination. Genes Dev. 31, 275–290 (2017).Article CAS PubMed PubMed Central Google Scholar Sparks, J. & Walter, J. C. Extracts for analysis of DNA replication in a nucleus-free system. Cold Spring Harb. Protoc. https://doi.org/10.1101/pdb.prot097154 (2019).Barrows, J. K. et al. BRD4 promotes resection and homology-directed repair of DNA double-strand breaks. Nat. Commun. 13, 3016 (2022).Article CAS PubMed PubMed Central Google Scholar Phillips, A. F. et al. Single-molecule analysis of mtDNA replication uncovers the basis of the common deletion. Mol. Cell 65, 527–538 (2017).Article CAS PubMed Google Scholar Kavlashvili, T. & Dewar, J. M. Approaches to monitor termination of DNA replication using Xenopus egg extracts. Methods Mol. Biol. 2444, 105–123 (2022).Article CAS PubMed Google Scholar Conwell, S. C., Patel, K. V. N., Weeks-Pollenz, S. J. & Dewar, J. M. Raw data and code from analysis of single and convergent replication fork collapse. Zenodo https://doi.org/10.5281/zenodo.19687676 (2026).Rice, P., Longden, I. & Bleasby, A. EMBOSS: The European Molecular Biology Open Software Suite. Trends Genet. 16, 276–277 (2000).Article CAS PubMed Google Scholar Zook, J. M. et al. An open resource for accurately benchmarking small variant and reference calls. Nat. Biotechnol. 37, 561–566 (2019).Article CAS PubMed PubMed Central Google Scholar Olson, N. D. et al. PrecisionFDA Truth Challenge V2: calling variants from short and long reads in difficult-to-map regions. Cell Genom. 2, 100129 (2022).Article CAS PubMed PubMed Central Google Scholar Avizienis, A. The N-version approach to fault-tolerant software. IEEE Trans. Softw. Eng. SE-11, 1491–1501 (1985).Article Google Scholar Prodan, A. et al. Comparing bioinformatic pipelines for microbial 16S rRNA amplicon sequencing. PLoS ONE 15, e0227434 (2020).Article CAS PubMed PubMed Central Google Scholar Cassidy, A.S. et al. Mitigating hallucinations and omissions in LLMs for invertible problems: an application to hardware logic design automation. Preprint at https://doi.org/10.48550/arXiv.2512.03053 (2025).Allamanis, M., Panthaplackel, S. & Yin, P. Unsupervised evaluation of code LLMs with round-trip correctness. In Proceedings of the 41st International Conference on Machine Learning (eds Salakhutdinov, R. et al.) (PMLR, 2024).Cohen, W.W. Fast effective rule induction. In Proceedings of the 12th International Conference on International Conference on Machine Learning (eds. Prieditis, A. & Russell, S. J.) (Morgan Kaufmann, 1995).Download referencesAcknowledgementsJ.M.D. discloses support for the research of this work from the National Institutes of Health (R35GM128696). D.C. discloses support for the research of this work from the National Institutes of Health (R01ES030575). D.T.L. discloses support for the research of this work from the National Institutes of Health (R35GM119512). M.T.C. discloses support for the research of this work from the National Institutes of Health (F32GM148024). Additional support was provided by the National Institutes of Health through grants T32CA009582 and T32GM139800 (S.C.C.), T32ES007028 (S.J.W.P. and M.T.C.) and P50CA098131 (J.M.D.). The funders had no role in study design, data collection and analysis, decision to publish or preparation of the manuscript.Author informationAuthors and AffiliationsDepartment of Biochemistry, Vanderbilt University School of Medicine, Nashville, TN, USASara C. Conwell, Khushi V. N. Patel, Savannah J. Weeks-Pollenz, Steven N. Dahmen, Matthew T. Cranford, David Cortez & James M. DewarDivision of Biology and Biological Engineering, California Institute of Technology, Pasadena, CA, USAWilliam G. DunphyDepartment of Biochemistry and Molecular Biology, Medical University of South Carolina, Charleston, SC, USADavid T. LongAuthorsSara C. ConwellView author publicationsSearch author on:PubMed Google ScholarKhushi V. N. PatelView author publicationsSearch author on:PubMed Google ScholarSavannah J. Weeks-PollenzView author publicationsSearch author on:PubMed Google ScholarSteven N. DahmenView author publicationsSearch author on:PubMed Google ScholarMatthew T. CranfordView author publicationsSearch author on:PubMed Google ScholarWilliam G. DunphyView author publicationsSearch author on:PubMed Google ScholarDavid T. LongView author publicationsSearch author on:PubMed Google ScholarDavid CortezView author publicationsSearch author on:PubMed Google ScholarJames M. DewarView author publicationsSearch author on:PubMed Google ScholarContributionsJ.M.D. conceptualized and supervised the project. S.C.C. performed all work except where noted otherwise. K.V.N.P. performed the experiments in Extended Data Figs. 6g–k and 7a,c–f. J.M.D. and S.J.W.-P. performed the sequencing analysis. S.N.D. generated the pCRISPR5X plasmid. M.T.C. purified the TetR. D.C. supervised M.C. D.T.L. purified and validated BRC peptides. W.G.D. provided the anti-BLM antibodies used in reviewer experiments. The paper was written by J.M.D. and S.C.C.Corresponding authorCorrespondence to James M. Dewar.Ethics declarationsCompeting interestsThe authors declare no competing interests.Peer reviewPeer review informationNature Structural & Molecular Biology thanks the anonymous reviewers for their contribution to the peer review of this work. Peer reviewer reports are available. Primary Handling Editor: Dimitris Typas, in collaboration with the Nature Structural & Molecular Biology team.Additional informationPublisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.Extended dataExtended Data Fig. 1 Analyses of leading and lagging collapse in Xenopus egg extracts.(a) Encounter with an SSB on the leading strand template, results in a blunt seDSB (leading collapse). (b) Encounter with an SSB on the lagging strand template, results in an seDSB with an overhang (lagging collapse). (c) Encounter with a lagging SSB can also result in continued unwinding and a deDSB. (d) Fork convergence at an SSB can also form a deDSB. (e) Template design with five tandem SSBs. To induce leading collapse, pSSBLEAD contains 5x tandem Nb.BsmI nicking sites (‘N1’ to ‘N5’) each flanked by 1x tetracycline operator (tetO) sequence either side. To induce lagging collapse, pSSBLAG contains the reverse complement of the tetO/SSB array from pSSBLEAD. (f) Pre-replicated control plasmid schematic and normalization procedure. Schematic depicting the DNA structures from Fig. 1a and the loading control, replicated prior to pSSBLEAD. (g) Quantification of total DNA synthesis from Fig. 1b normalized to the maximum signal across all time points and conditions. Total DNA synthesis remained unperturbed prior to collapse. Mean ± S.D., n = 9 independent experiments. (h) Quantification of θ structures from Fig. 1b. θ structures (converging fork intermediates) were abolished upon TetR addition. Mean ± S.D., n = 9 independent experiments. (i) Quantification of collapse efficiency from Fig. 1b. Collapse efficiency approached 100% (calculated as described in methods). Mean ± S.D., n = 9 independent experiments. (j) Quantification of nicked plus supercoiled circular monomers (CMs) from Fig. 1b. Mean ± S.D., n = 9 independent experiments. (k) Overexposure of Fig. 1b (bottom) showing that scCMs form after collapse (lane 10) but not without collapse (lane 5). (l) Quantification of σ structures (σs), high molecular weight products (HMw), and circular monomers (CMs) from Fig. 1b. Mean ± S.D., n = 9 independent experiments. (m) Schematic of expected XmnI-SacI digestion products from Fig. 1e. Control digestion products arise from digestion of the pre-replicated control plasmid. Fragment lengths (bp) and expected radioactive signal relative to the ‘no collapse’ control are indicated. (n) Table of expected DNA synthesis across the collapse region (%) for various collapse and repair efficiencies. See Supplementary Table 2 for a worked example. (o) Quantification of total DNA synthesis from Fig. 1i normalized to the maximum signal across all time points and conditions. Mean ± S.D., n = 3 independent experiments. (p) Quantification of θ structures from Fig. 1i. Mean ± S.D., n = 3 independent experiments. (q) Quantification of collapse efficiency (see methods) from Fig. 1i. Mean ± S.D., n = 3 independent experiments. (r) Quantification of σ structures (σs) and high molecular weight products (HMw) from Fig. 1i. Mean ± S.D., n = 3 independent experiments. (s) Side-by-side quantification of σ structures (σs) from Fig. 1b, i. Mean ± S.D., n = 9 independent experiments for leading collapse and n = 3 for lagging collapse. (t) Side-by-side quantification of high molecular weight products (HMw) from Fig. 1b, i. Mean ± S.D., n = 9 independent experiments for leading collapse and n = 3 for lagging collapse. (u) Side-by-side quantification of collapse region DNA synthesis following lead and lag collapse, as originally depicted in Fig. 1g, n. Mean ± S.D., n = 9 independent experiments for leading collapse and n = 6 for lagging collapse. One-way ANOVA with Dunnett’s test. The 95% confidence interval of the difference (TetR − Expected No Repair) was −11.187 to 5.674. Maximum undetectable restart, from the upper bound, was 8.54%.Source dataExtended Data Fig. 2 Lack of restart is not due to suppression by TetR or the LacR barrier.(a) To address whether TetR binding interferes with seDSB repair, pSSBLEAD was replicated as in Fig. 1b, lanes 6-10. Tetracycline (TC) was added at the onset of replication (TC-0), at 60 minutes (TC-60), or not added (NA control). Samples were separated on an agarose gel and analyzed by autoradiography. TC addition at the onset of replication blocked formation of σs (lanes 1, 3, 5), indicating efficient TetR displacement. (b) Quantification of θ structures from (a). Mean ± S.D., n = 3 independent experiments. (c) Quantification of σ structures (σs) from (a). Mean ± S.D., n = 3 independent experiments. TC-60 did not impact σ structure resolution, indicating residual TetR did not impair seDSB resolution. (d) Purified DNA samples from T = 120 in (a) were analyzed as in Fig. 1f. (e) DNA synthesis in the collapse region was quantified from (d) as in Fig. 1g. Mean ± S.D., n = 3 independent experiments. One-way ANOVA with Dunnett’s test. Post-collapse TC did not increase DNA synthesis, indicating TetR does not block restart. (f) To test whether complex intermediates masked detection of DNA synthesis in the collapse region (Fig. 1e-g), individual DNA strands were analyzed. DNA samples were generated as in Fig. 1f. Graph depicts total DNA normalized to the maximum signal across all time points and conditions. Mean ± S.D., n = 4 independent experiments. (g) Samples from (f) were then separated on an alkaline denaturing agarose gel to separate individual DNA strands then visualized by autoradiography. The broken arm was visible, demonstrating that DNA synthesis did not restart at the seDSB. (h) DNA synthesis in the collapsed region was quantified from (g) as in Fig. 1g. Control fragments could not be resolved in (g), so values were normalized to ‘Lin1’ from the corresponding native gels. Mean ± S.D., n = 4 independent experiments. One-way ANOVA with Dunnett’s test. Denaturing alkaline gel electrophoresis confirmed the absence of restart and revealed stable seDSB intermediates (arm) persisting for at least two hours. Expected no-repair synthesis was lower than in Fig. 1g due to reduced overall DNA synthesis in this experiment set (see f). (i) Plasmid DNAs containing leading SSBs were replicated in extract with LacR and either Tet Buffer or TetR, as in Fig. 1a. pSSBLEAD (pSSBLEAD-386 bp) has the central nick site 386 bp from the lacO array. Plasmid pSSBLEAD-686 bp has the nick site 686 bp away. (j) Samples from (i) were separated on an agarose gel and visualized by autoradiography. (k) Quantification of σ structures (σs) from (j). Mean ± S.D., n = 3 independent experiments. (l) Quantification of high molecular weight products (HMw) from (j). Mean ± S.D., n = 3 independent experiments. (m) Quantification of total DNA synthesis from (j) normalized to the maximum signal across all timepoints and conditions. Mean ± S.D., n = 3 independent experiments. (n) Schematic depicting the assay for restart of DNA synthesis in the collapse region. (o) Purified DNA samples from T = 120 in (j) were digested with XmnI and SacI, then separated on a native agarose gel and visualized by autoradiography. (p) Quantification of collapse region DNA synthesis from (o) as in (n). Signal was normalized to control fragment ‘Lin1’. ‘Expected No Repair’ values account for replication efficiency and collapse efficiency. In Xenopus egg extracts origins are delocalized and variable2, as in human cells9, and plasmids typically use multiple origins3. Thus a small fraction of plasmids is expected to undergo convergent collapse, which elicits efficient repair (see Figs. 4–5), and this fraction will be greater for pSSBLEAD-686 bp where the tetO nick sites are further from the LacR array. To correct for origin firing between the tetO and lacO arrays ‘Expected No Repair + Conv. Adjust.’ accounts for the contribution of converging fork-dependent repair arising from stochastic origin firing between the TetR-bound SSBs and the LacR barrier. Mean ± S.D., n = 3 independent experiments. Two-tailed unpaired t-tests and two-way ANOVA with Šídák’s test. ** P = 0.0040 (q) Purified DNA samples from T = 120 in (j) were separated on an alkaline denaturing agarose gel and visualized by autoradiography. The persistent arm fragment confirmed the absence of detectable restart at the seDSB, and end-to-end fusion products (E-E) were also detected. (r) Quantification of collapse region DNA synthesis from (q) as in (n). Signal was normalized to control fragment ‘Lin1’ from the corresponding native gel. Expected values are as described in (p). Increasing the distance between the collapse site and the LacR array did not promote restart. A small increase in synthesis (p,r) was attributable to stochastic fork convergence, not restart, per convergence-adjusted analysis. Mean ± S.D., n = 3 independent experiments. Two-tailed unpaired t-tests and two-way ANOVA with Šídák’s test. *P = 0.0138 and ** P = 0.0020.Source dataExtended Data Fig. 3 Analyses of RAD51 function during single fork collapse.(a) Quantification of DNA synthesis from Fig. 2b normalized to the maximum signal across all time points and conditions. Mean ± S.D., n = 3 independent experiments. (b) Quantification of nicked plus supercoiled circular monomers (CMs) from Fig. 2b. Mean ± S.D., n = 3 independent experiments. (c) Purified DNA from T = 120 in Fig. 2b was digested with XmnI and SacI to analyze collapse region synthesis. Digested samples were separated on an agarose gel and visualized by autoradiography. (d) Quantification of DNA synthesis in the collapse region from (c) normalized to control fragment ‘Lin1’. Mean ± S.D., n = 3 independent experiments. One-way ANOVA with Dunnett’s test. *** P = 0.0004 (e) Quantification of E-E fragments from (c), normalized to vehicle condition. Mean ± S.D., n = 3 independent experiments. One-way ANOVA with Dunnett’s test. * P = 0.0300 (f) Quantification of DNA synthesis from Fig. 2f normalized to the maximum signal across all time points and conditions. Mean ± S.D., n = 3 independent experiments. (g) Purified samples from T = 120 in Fig. 2f were digested with XmnI and SacI, then separated on an agarose gel and visualized by autoradiography. (h) Quantification of DNA synthesis in the collapse region from (g) normalized to control fragment ‘Lin1’. Mean ± S.D., n = 3 independent experiments. One-way ANOVA with Dunnett’s test. (i) Quantification of E-E fragments from (g), normalized to vehicle condition. Mean ± S.D., n = 3 independent experiments. One-way ANOVA with Dunnett’s test. (j) pSSBLead was replicated in Xenopus egg extract to induce leading collapse as in Fig. 1a. (k) Samples from (j) were purified by either phenol:chloroform extraction, to recover total DNA, or spin column purification to exclude high molecular weight DNA. Purified DNA samples were separated on an agarose gel and visualized by autoradiography. High molecular weight species were undetectable after column extraction but all other species were detected, confirming that column purification effectively excluded high molecular weight species. n = 3 independent experiments. (l) Samples from (j) were digested with AlwNI, then separated on an agarose gel, and visualized by autoradiography. Red arrows indicate fragments enriched in phenol:chloroform purified samples. n = 3 independent experiments. (m) Phenol:chloroform samples from (j) were treated with either AlwNI or Topo II then separated on an agarose gel and visualized by autoradiography. HMw species were resolved by AlwNI but not topo II, indicating they were not catenanes. n = 3 independent experiments. (n) pSSBLag was replicated to induce lagging collapse as in Fig. 1h. (o) Purified samples from (n), as in (k). (p) Samples from (n) digested with AlwNI, as in (l). n = 3 independent experiments. (q) Samples from (n) were restriction digested, separated on an agarose gel, and visualized by autoradiography. Colored arrows indicate end-to-end linear fragments from Fig. 2i. Asterisk indicates collapsed arm fragments. n = 3 independent experiments.Source dataExtended Data Fig. 4 Analysis of microhomology at end-to-end fusion junctions following leading collapse.(a) Example of an end-to-end fusion junction detected by sequencing at reference position 69/68. The junction (indicated by the arrow) falls within a tetO sequence (indicated in green). Forward (fwd) and reverse (rev) microhomology (MH) scores are expressed as base pair matches per 10 bp window on each side of the junction. Sequencing reads are indicated alongside predicted junction sequences. (b) As in (a), a representative end-to-end fusion junction at reference position 76/61. (c) Distribution of forward microhomology scores across all palindromic junctions identified within the tetO array (n = 22, mean=1.7 bp matches/10). For visualization of junction positions refer to Fig. 2k. (d) Distribution of reverse microhomology scores across all palindromic junctions identified within the tetO array (n = 22, mean=2.3 bp matches/10). For visualization of junction positions refer to Fig. 2k. (e) Distribution of microhomology asymmetry across all palindromic junctions (n = 22, mean = -0.5). A central score demonstrates no significant directional bias in microhomology usage at end-to-end fusion junctions. For visualization of junction positions refer to Fig. 2k.Source dataExtended Data Fig. 5 Resolution of leading, but not lagging, collapse by extensive resection.(a) Quantification of DNA synthesis from Fig. 3f normalized to the maximum signal across all time points and conditions. Mean ± S.D., n = 3 independent experiments. (b) Samples from T = 120 in Fig. 3f were purified and restriction digested with AlwNI. Digested samples were then separated on an agarose gel and visualized by autoradiography. The collapsed arm fragment was readily detectable following MRE11 depletion, indicating that degradation of the collapsed arm was blocked. (c) Quantification of collapsed arm products from (b). Mean ± S.D., n = 3 independent experiments. Data were analyzed by unpaired two-sided t-test. *** P = 0.0002 (d) Quantification of high molecular weight products (HMw) from Fig. 3f. Mean ± S.D., n = 3 independent experiments. (e) Quantification of full length products from (b). Mean ± S.D., n = 3 independent experiments. Data were analyzed by unpaired two-sided t-test. *** P = 0.0008 (f) Leading strand collapse was induced as in Fig. 1a, except extracts were supplemented with BRC4WT at T = 0 and with either vehicle or aphidicolin (Aph) at T = 22.5 min. Addition of BRC4WT blocks D-loop formation such that it does not compete with the secondary collapse pathway. Later addition of aphidicolin was used to specifically identify the effects of inhibited resynthesis following seDSB formation. Schematic depicts the predicted effect of aphidicolin on secondary collapse where the inhibition of resynthesis is predicted to shift the balance toward resection, promoting the enhanced formation of circular monomers (CMs). (g) Samples from (f) were separated on an agarose gel and visualized by autoradiography. (h) Quantification of total DNA synthesis from (g) normalized to the maximum signal across all timepoints and conditions. Reduced DNA synthesis in aphidicolin-treated conditions confirms effective inhibition of DNA polymerase activity, while the overall low level of nCM formation in the vehicle control compared to Fig. 1b indicates nonspecific inhibition of nCM formation by the vehicle (DMSO). Mean ± S.D., n = 3 independent experiments. (i) Quantification of nicked circular monomers (nCMs) from (g). Mean ± S.D., n = 3 independent experiments.Source dataExtended Data Fig. 6 Analyses of convergent collapse in Xenopus egg extracts.(a) Quantification of DNA synthesis from Fig. 4b normalized to the maximum signal across all time points and conditions. Mean ± S.D., n = 4 independent experiments. (b) Quantification of collapse efficiency from Fig. 4b. Mean ± S.D., n = 4 independent experiments. Collapse efficiency was calculated as described in methods. (c) Convergent fork collapse was induced by replication of pSSB as in Fig. 4a. (d) DNA samples from T = 120 minutes in (c) were purified by either phenol:chloroform extraction, to recover total DNA, or by spin column purification to preferentially exclude high molecular weight DNA. To investigate whether erroneous end-to-end fusions were formed, purified DNA was restriction digested with XmnI, AlwNI, or SapI, which should cut pSSB only once. Products of digestion were then separated on an agarose gel. Erroneous end-to-end fusions should have resulted in products greater than and smaller than the length of full length products (4661 bp), similar to Fig. 2i-j. However, exclusively linear products were formed in all conditions and few other species were generated, indicating that end-to-end fusions were absent. n = 6 independent experiments. (e) pSSB was replicated as in Fig. 4b lanes 6-10. DNA intermediates were purified, digested with AlwNI, then separated on an agarose gel. Replication and collapse intermediates were readily converted to full length linear products, indicating that the ends were efficiently repaired. n = 3 independent experiments (f) Quantification of HMw products at T = 120 min across multiple independent experiments in Vehicle, BRC4WT-, and BRC4MUT-treated conditions. BRC4WT treatment resulted in a modest but significant reduction in HMw compared to vehicle, while BRC4MUT treatment did not significantly differ from vehicle. Mean ± S.D., n = 16 independent experiments. Data were analyzed by one-way ANOVA and Dunnett’s multiple-comparison method. * P = 0.0414 (g) Convergent fork collapse was induced as in Fig. 4i. A tight time course was performed to analyze the effects of BRC peptide treatment on collapsed σs and linear (Lin) products. (h) Convergent collapse was induced as in (g) in the presence of Vehicle, BRC4WT, or BRC4MUT. Samples were collected at the indicated timepoints, separated on an agarose gel, and visualized by autoradiography. (i) Quantification of σ structures from (h). Mean ± S.D., n = 3 independent experiments. (j) Quantification of linear products (Lin) from (h). Mean ± S.D., n = 3 independent experiments. (k) Quantification of high molecular weight products (HMw) from (h). Mean ± S.D., n = 3 independent experiments.Source dataExtended Data Fig. 7 Analyses of RAD51-requirement following convergent fork collapse.(a) Convergent fork collapse was induced as depicted in Fig. 4a except in either mock- or RAD51-immunodepleted extracts. (b) Western blot confirming immunodepletion of RAD51 from Xenopus egg extract. Rd1, Rd2, and Rd3 indicate sequential rounds of depletion beads. (c) Convergent collapse was induced as in (a) in mock- or RAD51-immunodepleted extracts. Samples were collected at the indicated timepoints, separated on an agarose gel, and visualized by autoradiography. (d) Quantification of σ structures from (c). Mean ± S.D., n = 3 independent experiments. (e) Quantification of linear products (Lin) from (c). Mean ± S.D., n = 3 independent experiments. (f) Quantification of high molecular weight products (HMw) from (c). Loss of RAD51 does not impair HMw formation following convergent collapse, demonstrating that resolution of collapsed deDSBs occurs independently of RAD51. Mean ± S.D., n = 3 independent experiments. (g) pSSB was replicated as in Fig. 4b, lane 10, then treated with RuvC or left untreated. RuvC treatment did not resolve high molecular weight species following convergent collapse (lanes 1-2), in contrast to leading and lagging collapse in Fig. 2d, h. n = 3 independent experiments. (h) DNA samples were prepared and purified as in (g) then treated with either AlwNI or topo II. High molecular weight products were resolved by AlwNI but not topo II, indicating that they were not composed of catenanes. n = 3 independent experiments. (i) Quantification of DNA synthesis from Fig. 4j normalized to the maximum signal across all time points and conditions. Mean ± S.D., n = 3 independent experiments. (j) Purified samples from T = 120 of Fig. 4j were digested with XmnI and SacI to analyze DNA synthesis in the collapse region (as in Fig. 4f-h). Digested samples were separated on an agarose gel then visualized by autoradiography. (k) Quantification of DNA synthesis in the collapsed region from (j) normalized to control fragment ‘Lin1’. Mean ± S.D., n = 3 independent experiments. Data were analyzed by one-way analysis of variance (ANOVA) and Dunnett’s multiple-comparison method. No detectable effect on collapse region DNA synthesis was observed, indicating that completion of DNA synthesis occurred independently of RAD51.Source dataExtended Data Fig. 8 Analyses of nCas9 induced collapse in Xenopus egg extracts.(a) Quantification of DNA synthesis from Fig. 5b. Mean ± S.D., n = 3 independent experiments. (b) Quantification of nicked plus supercoiled circular monomers (CMs) from Fig. 5b. Mean ± S.D., n = 3 independent experiments. (c) Quantification of σ structures plus linear products (collapsed molecules) from Fig. 5b. Mean ± S.D., n = 3 independent experiments. (d) Quantification of θ structures from Fig. 5b. Mean ± S.D., n = 3 independent experiments. (e) Quantification of collapse efficiency (see methods) from Fig. 5b. Mean ± S.D., n = 3 independent experiments. (f) Leading collapse was induced using nCas9 as in Fig. 5b lanes 6-10 with vehicle, BRC4WT, or BRC4Mut peptide. DNA samples were separated on an agarose gel and visualized by autoradiography. (g) Quantification of high molecular weight products (HMw) from (f). Mean ± S.D., n = 3 independent experiments. (h) Quantification of DNA synthesis from (f) normalized to the maximum signal across all time points and conditions. Mean ± S.D., n = 3 independent experiments. (i) Purified DNA samples from T = 120 in (f) were analyzed as in Fig. 5f to monitor synthesis in the collapse region. (j) Synthesis in the collapse region was quantified from (i) as in Fig. 5g. Mean ± S.D., n = 3 independent experiments. One-way ANOVA with Dunnett’s test. (k) Convergent collapse was induced using nCas9 as in Fig. 5b lanes 16-20 with vehicle, BRC4WT, or BRC4Mut peptide. DNA samples were separated on an agarose gel and visualized by autoradiography. (l) Quantification of high molecular weight products (HMw) from (k). Mean ± S.D., n = 3 independent experiments. (m) Quantification of DNA synthesis from (k) normalized to the maximum signal across all time points and conditions. Mean ± S.D., n = 3 independent experiments. (n) Purified DNA samples from T = 120 in (k) were analyzed as in Fig. 5f to monitor synthesis in the collapse region. (o) Synthesis in the collapse region was quantified from (n) as in Fig. 5g. Mean ± S.D., n = 3 independent experiments. One-way ANOVA with Dunnett’s test. (p) Leading collapse and convergent collapse were induced by replicating pCRISPR. Purified DNA products (Fig. 5b, lanes 10, 20) were treated with RuvC then separated on an agarose gel and visualized by autoradiography. HMw products are mostly sensitive to RuvC for leading collapse (lanes 1-2) but mostly insensitive for convergent collapse (lanes 3-4) as observed for collapse at stabilized SSBs (Fig. 2d,h and Extended Data Fig. 7g). n = 3 independent experiments. (q) Purified DNA was obtained as in (p), treated with AlwNI or TopoII, separated on an agarose gel and visualized by autoradiography. HMw species were resolved by AlwNI but not topo II, indicating they were not composed of catenanes, as observed for collapse at stabilized SSBs (Extended Data Fig. 3m). n = 3 independent experiments. (r) Quantification of nicked circular monomers (nCM) from Fig. 5b, lanes 1-10. Mean ± S.D., n = 3 independent experiments. (s) Lagging collapse was induced by replication of pCRISPR as depicted in Fig. 5a but with guides targeting the lagging strand template. Replication samples were separated on an agarose gel and visualized by autoradiography. (t) Leading collapse was induced using nCas9H840A as in Fig. 5a but using pCRISPR1X containing only a single target sequence. Replication samples were separated on an agarose gel and visualized by autoradiography. (u) Quantification of θ structures from (t). Mean ± S.D., n = 6 independent experiments. (v) Quantification of nicked plus supercoiled circular monomers from (t). Mean ± S.D., n = 6 independent experiments. (w) Quantification of collapse efficiency (see methods) from (t). Mean ± S.D., n = 6 independent experiments.Source dataExtended Data Fig. 9 Analyses of MRE11 and CtIP immunodepletion.(a) Western blots confirming independent immunodepletion of MRE11 and CtIP from Xenopus egg extract. Rd1, Rd2, and Rd3 indicate sequential rounds of depletion beads. Reciprocal Western blotting for MRE11 and CtIP revealed that the proteins were not co-depleted. (b) Quantification of DNA synthesis from Fig. 6b. Mean ± S.D., n = 4 independent experiments. (c) Quantification of the ratio of linear products ‘Lin’ to nicked plus supercoiled circular monomers (CMs). (d) Quantification of DNA synthesis from Fig. 6j. Mean ± S.D., n = 4 independent experiments.Source dataExtended Data Fig. 10 tetO and tetO fragment insertions arising from fork collapse.(a) Size distribution of tetO insertions (repeat + fragment) at repair junctions following convergent fork collapse in vehicle- or Cdc7i-treated conditions, expressed as a percentage of total sequencing reads. Mean ± S.D., n = 3 independent experiments. (b) Size distribution of tetO fragment insertions at repair junctions following convergent fork collapse in vehicle- or Cdc7i-treated conditions. Insertions of 7-8 bp are the predominant fragment size across conditions. Mean ± S.D., n = 3 independent experiments. (c) Mapping of 7-8 bp tetO fragment insertions to the terminal tetO repeat (highlighted in green) at repair junctions following convergent fork collapse in vehicle- or Cdc7i-treated conditions. Insertions cluster predominantly at reference positions 193, 195, and 198, consistent with microhomology-mediated templating from the terminal tetO repeat. Mean ± S.D., n = 3 independent experiments. (d) Frequency of mutation classes for the parental plasmid used in Fig. 7. Abbreviations: Ins, insertions; Del, deletions; Sub, substitutions. (e) Examples of tetO fragment insertions detected by amplicon sequencing at repair junctions following convergent fork collapse. Three insertion events (at reference positions 193, 195, and 198) are shown mapped to two distinct parental sequence contexts: the non-nick region (top) and the nick-proximal region (bottom). Microhomology (MH) and homology (H) associated with each insertion event is indicated. Bold nucleotides indicate the homologous bases. (f) Size distribution of total tetO insertions (repeat + fragment) at repair junctions following convergent fork collapse in mock- or MRE11-immunodepleted extracts. Mean ± S.D., n = 2 independent experiments. (g) Size distribution of tetO fragment insertions at repair junctions following convergent fork collapse in mock- or MRE11-immunodepleted extracts. Mean ± S.D., n = 2 independent experiments. (h) As in (c), for repair junctions following convergent fork collapse in mock- or MRE11-immunodepleted extracts. Mean ± S.D., n = 2 independent experiments. (i) Size distribution of total tetO insertions (repeat + fragment) at repair junctions following single fork collapse in vehicle- or Cdc7i-treated conditions. Mean ± S.D., n = 3 independent experiments. (j) Size distribution of tetO fragment insertions at repair junctions following single fork collapse in vehicle- or Cdc7i-treated conditions. Mean ± S.D., n = 3 independent experiments. (k) As in (c), for repair junctions following single fork collapse in vehicle- or Cdc7i-treated conditions. Mean ± S.D., n = 3 independent experiments.Source dataSupplementary informationSupplementary Information (download PDF )Supplementary Tables 1 and 2.Reporting Summary (download PDF )Peer Review File (download PDF )Source dataSource Data Fig. 1 (download XLSX )Statistical source data.Source Data Fig. 1 (download PDF )Unprocessed gels and western blots.Source Data Fig. 2 (download XLSX )Statistical source data.Source Data Fig. 2 (download PDF )Unprocessed gels and western blots.Source Data Fig. 3 (download XLSX )Statistical source data.Source Data Fig. 3 (download PDF )Unprocessed gels and western blots.Source Data Fig. 4 (download XLSX )Statistical source data.Source Data Fig. 4 (download PDF )Unprocessed gels and western blots.Source Data Fig. 5 (download XLSX )Statistical source data.Source Data Fig. 5 (download PDF )Unprocessed gels and western blots.Source Data Fig. 6 (download XLSX )Statistical source data.Source Data Fig. 6 (download PDF )Unprocessed gels and western blots.Source Data Fig. 7 (download XLSX )Statistical source data.Source Data Extended Data Fig. 1 (download XLSX )Statistical source data.Source Data Extended Data Fig. 1 (download PDF )Unprocessed gels and western blots.Source Data Extended Data Fig. 2 (download XLSX )Statistical source data.Source Data Extended Data Fig. 2 (download PDF )Unprocessed gels and western blots.Source Data Extended Data Fig. 3 (download XLSX )Statistical source data.Source Data Extended Data Fig. 3 (download PDF )Unprocessed gels and western blots.Source Data Extended Data Fig. 4 (download CSV )Statistical source data.Source Data Extended Data Fig. 5 (download XLSX )Statistical source data.Source Data Extended Data Fig. 5 (download PDF )Unprocessed gels and western blots.Source Data Extended Data Fig. 6 (download XLSX )Statistical source data.Source Data Extended Data Fig. 6 (download PDF )Unprocessed gels and western blots.Source Data Extended Data Fig. 7 (download XLSX )Statistical source data.Source Data Extended Data Fig. 7 (download PDF )Unprocessed gels and western blots.Source Data Extended Data Fig. 8 (download XLSX )Statistical source data.Source Data Extended Data Fig. 8 (download PDF )Unprocessed gels and western blots.Source Data Extended Data Fig. 9 (download XLSX )Statistical source data.Source Data Extended Data Fig. 9 (download PDF )Unprocessed gels and western blots.Source Data Extended Data Fig. 10 (download XLSX )Statistical source data.Rights and permissionsOpen Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.Reprints and permissionsAbout this article