Break-induced replication drives large-scale genomic amplifications in cancer cells

Reviewer #1 (Public review):

Summary:

In this report, the authors investigate the mechanisms underlying large-scale genomic amplification (DIGA) induced by genome-wide DNA double-strand breaks (DSBs), such as those generated by ionizing radiation (IR).

Strengths:

The authors demonstrate that DSB-induced DIGA does not require origin re-licensing but is dependent on proteins involved in break-induced replication (BIR). This finding represents a major strength of the study, as it reveals a previously unrecognized mechanism of DSB-induced genomic amplification. Additional strengths include the demonstration that DIGA is promoted by DNA end resection and suppressed by the 53BP1-RIF1-shieldin pathway. The authors also show that SET8 and SUV4-20H1 have opposing effects on CDT1 overexpression-induced and IR-induced re-replication. Furthermore, the finding that the extent of DIGA in cancer cells correlates with sensitivity to IR has potential implications for cancer treatment.

Weaknesses:

However, more comprehensive studies are needed to strengthen the conclusions. Although IR- or DSB-induced DIGA was observed in multiple cancer cell lines, the overall mechanisms underlying why DIGA is more pronounced in certain cell lines but not others remain unclear. Beyond p53 status, additional factors that determine DIGA susceptibility should be investigated. In addition, the evidence that DIGA-associated DNA synthesis occurs within the same cell cycle is not yet sufficient. For instance, a time-course experiment (EdU versus DAPI) should be performed after IR to determine when DIGA initiates relative to normal S-phase DNA replication. A parallel analysis of re-replication at different time points following MLN4924 treatment would provide a useful comparison. Cyclin B and phospho-histone H3 (Ser10) levels should be monitored to define cell cycle stage. Additional evidence supporting a BIR-dependent mechanism, such as demonstrating conservative DNA synthesis and mapping DIGA sites following AsiSI-induced DSBs, would be needed.

Overall, this study provides new insights into the mechanisms driving genome-wide amplification following DSB formation. Additional mechanistic studies will further strengthen the conclusions and broaden the impact of this work.

Reviewer #2 (Public review):

Summary:

In this study, Benamar et al. investigate whether DNA double-strand breaks induce extensive abnormal DNA synthesis in cancer cells and whether this response contributes to the cytotoxic effects of ionizing radiation and other DNA-damaging treatments. The authors refer to this process as double-strand-break-induced genomic amplification (DIGA). Mechanistically, they propose that insufficient protection of broken DNA ends permits excessive resection, followed by RAD51-dependent strand invasion and RAD52-POLD3-POLD4-dependent DNA synthesis resembling break-induced replication.

Overall, the study addresses an interesting and potentially important question. Break-induced replication was originally defined as a pathway that repairs one-ended DNA breaks through extended synthesis from an invaded homologous template. Studies in yeast established that this process involves a migrating DNA-synthesis bubble, conservative inheritance of newly synthesized DNA, frequent template switching, and high mutagenicity. Related forms of DNA synthesis have subsequently been described in mammalian cells at collapsed replication forks, telomeres, under-replicated mitotic regions, and some transcription-associated DNA breaks.

The important advance of this study is the proposal that double-strand-break-associated DNA synthesis, with several features of break-induced replication, can become sufficiently extensive to cause a measurable increase in total cellular DNA content and that this synthesis correlates with radiation-induced cell death. However, the physical structure, genomic distribution, and extent of the additional DNA have not yet been directly established.

Strengths:

Radiation-induced breaks are generally considered in relation to DNA repair, chromosome rearrangements, checkpoint activation, mitotic failure, senescence, and cell death. The possibility that broken DNA ends can also initiate extensive DNA synthesis provides a potentially important additional link between defective break repair, genome amplification, and treatment-induced cytotoxicity.

The authors examine this phenomenon using several complementary approaches and comparisons across multiple cancer cell lines. The finding that several double-strand-break-inducing treatments produce a similar response suggests that the phenotype is not restricted to ionizing radiation or to a single cellular background.

The authors also make efforts to distinguish DIGA from canonical origin-dependent re-replication, such as that caused by CDT1 stabilization and inappropriate origin relicensing. They find that, following irradiation, CDT1 is degraded rather than stabilized and that CDT1 depletion does not suppress DIGA. Perturbation of ORC1 or ORC2 also has little effect on the phenotype. In addition, depletion of SET8 or loss of SUV4-20H1 increases, rather than decreases, DIGA.

The authors carry out a systematic analysis of DNA-end protection and processing. Loss of ATM, RNF8, RNF168, 53BP1, RIF1, or Shieldin components enhances DIGA, whereas depletion or inhibition of CtIP, MRE11, EXO1, RAD51, RAD52, POLD3, or POLD4 suppresses it. These experiments support a model in which insufficient end protection permits excessive DNA-end resection, followed by strand invasion and recombination-associated DNA synthesis involving factors linked to break-induced replication.

Overall, the findings place DIGA within a growing body of evidence that recombination-associated DNA synthesis can extend well beyond short repair patches. The study attempts to connect the processing of double-strand breaks with abnormal DNA synthesis, increased cellular DNA content, and the cytotoxic response to radiation.

Major weaknesses and suggested experiments:

(1) The main evidence for DIGA is the appearance of cells with greater-than-G2/M DNA content by flow cytometry. BrdU incorporation shows that active DNA synthesis occurs within this population, and the density-gradient experiments provide further evidence for newly synthesized DNA. However, these methods do not reveal the physical nature of the proposed genomic amplification-for example, which genomic regions are copied or how long the synthesis tracts are.

This is important for the central conclusion of the study because mammalian break-induced replication can produce genomic duplications, but the extent of synthesis depends strongly on the type of DNA lesion. Previous work has shown that repair of damaged replication forks can produce segmental genomic duplications (Costantino et al., 2014; PMID: 24310611). By contrast, recent measurements at defined two-ended double-strand breaks suggest that mammalian break-induced-replication tracts can be relatively restricted (Li et al., 2021; PMID: 33470420; Shah et al., 2024; PMID: 39368985).

Thus, the large increase in total DNA content detected by flow cytometry in this study would require many simultaneous synthesis events, very long synthesis tracts, or an alternative process such as whole-genome duplication. The authors should therefore characterize the additional DNA directly. Whole-genome sequencing of sorted cells with greater-than-G2/M DNA content could be particularly informative.

(2) The genetic dependencies are consistent with synthesis initiated from resected DNA ends, but they do not directly demonstrate that the new DNA synthesis begins at double-strand breaks. DNA damage could indirectly alter replication-origin activity, cell-cycle progression, or DNA synthesis at genomic regions distant from the original lesions. Increased DNA content alone therefore does not establish amplification initiated directly at DNA breaks.

(3) MLN4924-induced re-replication is used throughout the study, but it is unclear whether the authors have combined MLN4924 with ionizing radiation and measured the resulting DNA synthesis. This experiment could clarify whether conventional re-replication and DIGA are independent, overlapping, or mechanistically connected.

(4) The results involving canonical non-homologous end joining are complex. Loss or inhibition of DNA-PKcs increases DIGA, whereas loss of XRCC4 or XLF strongly suppresses it, and loss or inhibition of LIG4 has a weaker suppressive effect. The authors suggest that XRCC4 and XLF stabilize broken DNA ends and thereby permit the synthesis reaction independently of final ligation. This is an interesting model, but the current evidence does not yet establish it. XRCC4 and XLF can affect end synapsis, break persistence, chromosome fusion, resection, and cell-cycle progression. Their loss could therefore reduce DIGA through several indirect mechanisms. The authors should measure DNA-end resection, RAD51 loading, persistence of double-strand breaks, and recruitment of RAD52 or POLD3 in XRCC4-, XLF-, and LIG4-deficient cells. Complementation with separation-of-function mutants that differentially affect XRCC4-XLF end bridging and LIG4 recruitment could test whether end stabilization, rather than ligation, is important. Without this, the opposing effects of upstream and downstream non-homologous end-joining components remain difficult to understand.

(5) The correlation between the amount of DIGA and radiation sensitivity across cancer cell lines is interesting. However, cell lines differ in many factors that influence radiation responses, including p53 status, apoptosis, checkpoint activity, ploidy, homologous recombination capacity, and proliferation rate. Matched models would provide stronger evidence. Conversely, restoring DNA-end protection in a DIGA-prone cancer cell line could test whether this reduces susceptibility. These experiments would help determine whether DIGA is a general property of cancer cells or a feature of particular repair-defective genetic backgrounds.

Reviewer #3 (Public review):

Summary:

In this study, the authors show that exposure of cancer cells to DSBs induced enzymatically or by ionizing radiation (IR) results in large-scale genomic amplifications (DIGA: DSB-induced genomic amplifications) that are detectable by FACS. This phenomenon was observed in diverse cancer cell lines and was shown to be limited by p53 in paired cell lines. In principle, DIGA could result from re-initiation of DNA synthesis within the same cell cycle, mitotic segregation errors, or repair-associated DNA synthesis. The authors conduct experiments to distinguish between these possibilities and conclude that DIGA is the result of extensive RAD51-dependent DNA synthesis.

Strengths:

The observation of genome amplification in response to DSBs specifically in cancer cells is striking, particularly at high IR doses or by AsiSI endonuclease induction. At 9 Gy, almost 50% of cells have a >4N DNA content. The correlation between high levels of DIGA and reduced clonogenic survival of melanoma cells suggests that DIGA contributes to IR-induced cytotoxicity. The authors convincingly show that DIGA occurs in a single cell cycle and is not due to chromosome mis-segregation at mitosis. Because DIGA is partially dependent on resection nucleases, POLD3, POLD4, RAD52 and RAD51, the authors conclude that DIGA results from repair synthesis by a BIR-like process.

There are significant weaknesses in the study:

(1) It is unclear to this reviewer how BIR-like synthesis could increase genomic DNA copy number by 50% or more, as indicated by the FACS plots. In the case of AsiSI, there are ~150 sites that are efficiently cleaved in U2OS cells. Each of these sites would need to prime extensive synthesis by an inefficient migrating D-loop mechanism. Studies of BIR-like synthesis at DSBs in U2OS cells have shown that repair tracts are fairly short, around 3-10 kb in length. Thus, it is hard to rationalize how BIR at a few hundred DSBs could initiate such large changes in genome size.

(2) The BrdU-FACS plot shown in Figure S4 shows that most of the cells have an 8N content 48 and 72 h after AsiSI induction and do not have much BrdU incorporated. This finding would suggest that genomic DNA doubling is mostly independent of de novo synthesis. Also, there does not seem to be a continuum from 4N up to 8N in these plots as one might expect from variable length DNA synthesis tracts.

(3) The requirement for XRCC4, XLF and LIG4 to promote DIGA is puzzling if synthesis occurs by a BIR-like mechanism. Although there is evidence that DNA synthesis at DSBs can be initiated by homology-directed strand invasion and terminated by NHEJ, this mechanism would limit the extent of synthesis to a few kb at each site. One might have expected an increase in DIGA in the absence of core NHEJ factors because of the increased number of DSBs available for BIR.

(4) Although the authors rule out re-replication as a contributor to IR-induced DIGA, the FACS profiles of MLN4924 and 9 Gy treatment look remarkably similar, and both would be inhibited by aphidicolin treatment. A previous study showed that DSBs induce endoreduplication in Arabidopsis (Adachi et al. PNAS 2011).

Author response:

We thank the reviewers for their careful and constructive evaluation of our study. We are encouraged that the reviewers recognized the significance of DSB-induced genomic amplification (DIGA) and the evidence implicating DNA-end processing and recombination-associated DNA synthesis in this response.

To our knowledge, this study provides the first description of DIGA as a large-scale increase in genomic DNA content following the induction of DSBs in cancer cells and represents an initial effort to define factors that regulate this phenomenon. The present work shows that DIGA can be induced by several sources of DSBs, involves de novo DNA synthesis, is genetically distinguishable from canonical CDT1-dependent origin re-licensing, is regulated by pathways controlling DNA-end protection and resection, and requires RAD51, RAD52, POLD3, and POLD4.

At the same time, we agree that many important questions remain regarding the physical organization and genomic distribution of the additional DNA, the sites from which synthesis originates, the length and number of synthesis tracts, and the full determinants that render some cancer cells more susceptible to DIGA than others. We view these as important questions that arise from the initial characterization of this previously unrecognized phenotype and that will require substantial additional investigation.

Reviewer #1:

We agree that p53 status alone does not explain the considerable variation in DIGA observed among the cancer cell lines examined. Our experiments using isogenic HCT116 cells identify p53 as one factor capable of limiting DIGA, while the genetic studies implicate DNA-end protection and resection pathways as additional determinants. The present data, however, do not establish which of these or other pathways account for the differences among individual cancer cell lines. Defining the molecular basis for this variability will require systematic comparison of DIGA-prone and DIGA-resistant cells.

The reviewer also raises an important question regarding the temporal relationship between DIGA and normal S-phase DNA replication. Our conclusion that the increase in DNA content involves de novo synthesis within the same cell-cycle interval is supported by BrdU incorporation in cells with >4N DNA content, the persistence of DIGA when progression through mitosis is blocked by nocodazole, and the detection of newly synthesized DNA in synchronized irradiated cells. These experiments do not, however, define precisely when DIGA-associated synthesis begins relative to normal S-phase replication. More detailed time-resolved analysis will be required to establish this relationship.

We also agree that the present findings support a BIR-like mechanism rather than providing a complete physical reconstruction of classical BIR. The dependence of DIGA on DNA-end resection, RAD51, RAD52, POLD3, and POLD4 provides genetic evidence for recombination-associated DNA synthesis with features of BIR. Direct determination of synthesis-tract architecture, template usage, and genomic distribution will be required to define the underlying synthesis mechanism more completely.

Reviewer #2:

We agree that direct characterization of the additional DNA represents an important next step in understanding DIGA. The current study demonstrates a substantial increase in cellular DNA content, de novo DNA synthesis within the >4N population, and dependence on factors involved in DNA-end resection, strand invasion, and BIR-associated synthesis. These experiments do not determine which genomic regions are amplified or the length of individual synthesis tracts. Genomic analysis of cells undergoing DIGA should help determine whether the observed increase in DNA content reflects numerous amplification events, extensive synthesis from a subset of sites, or a different organization of the additional DNA.

We appreciate the reviewer highlighting the study by Costantino et al. (2014), which provided important evidence that BIR-associated repair of damaged replication forks can generate segmental genomic duplications in human cells. The genomic alterations characterized in that study, however, arose under a substantially different experimental setting. Costantino et al. induced replication stress through cyclin E overexpression and analyzed copy-number alterations accumulated over a three-week period in clonally derived cells. Among these alterations, amplifications smaller than 200 kb were reduced following depletion of POLD3 or POLD4, leading the authors to propose that this subset of segmental duplications may represent BIR events, whereas larger amplifications and deletions could involve other repair mechanisms.

DIGA, however, differs from these previously described alterations in several readily observable respects. DIGA develops over approximately one to three days following acute induction of DSBs by IR or AsiSI and produces increases in total cellular DNA content sufficiently large to be detected directly by flow cytometry. Thus, the two phenomena differ in their mode of induction, kinetics, and scale. At the same time, the involvement of POLD3 and other recombination-associated factors in both settings raises the possibility that they share aspects of the underlying DNA-synthesis machinery. The present data do not establish whether the additional DNA in DIGA consists of numerous segmental duplications, substantially longer synthesis products, or another genomic configuration.

We also agree that our experiments do not directly demonstrate that DIGA-associated DNA synthesis initiates precisely at individual DSB sites. The ability of AsiSI-generated DSBs to induce DIGA, together with its dependence on DNA-end resection, RAD51, RAD52, POLD3, and POLD4, links the phenomenon closely to DSB processing. Direct mapping of newly synthesized DNA relative to defined DSBs will ultimately be required to determine where DIGA-associated synthesis originates.

The reviewer asks whether MLN4924-induced re-replication and DIGA have been examined simultaneously. We have not examined this combination. Our distinction between these processes instead rests on their different genetic requirements. In particular, depletion of CDT1 strongly suppresses MLN4924-induced re-replication but does not suppress IR-induced DIGA, and DIGA is stimulated while rereplication is inhibited by the depletion of SET8. These observations argue against canonical CDT1-dependent origin re-licensing as the mechanism underlying DIGA, although they do not exclude more complex interactions between replication and DSB-associated DNA synthesis.

We agree that the effects of XRCC4, XLF, and LIG4 are mechanistically intriguing and not yet fully understood. The present experiments establish that loss of XRCC4 or XLF, and to a lesser extent LIG4, suppresses DIGA, whereas loss or inhibition of DNA-PKcs enhances it. Stabilization of broken DNA ends by XRCC4/XLF is one possible interpretation, but effects on end resection, DSB persistence, repair-pathway choice, or other functions of these proteins could also contribute. The opposing effects of different components of the NHEJ machinery therefore identify an important mechanistic question that remains to be resolved.

Finally, we agree that the correlation between DIGA and radiation sensitivity across the melanoma cell-line panel does not by itself demonstrate causality. The data establish an association between the propensity to undergo DIGA and sensitivity to IR. Because these cell lines differ in multiple additional properties that may influence the radiation response, matched models in which DIGA can be selectively altered will be important for determining the extent to which DIGA itself contributes to radiation-induced loss of proliferative capacity.

Reviewer #3:

We agree that the magnitude of the increase in DNA content is one of the most interesting unresolved features of DIGA. Previous analyses of BIR-associated synthesis at defined mammalian lesions have generally described synthesis events considerably smaller than the total increase in DNA content observed here. Our experiments do not establish the length or number of individual synthesis events responsible for DIGA. We therefore use the term BIR-like to describe the genetic requirements of the process rather than to imply that each DSB gives rise to a single exceptionally long BIR tract. Determining how many genomic sites participate and how much DNA is synthesized at individual sites will be important for understanding how the large increase in total DNA content is generated.

Regarding the AsiSI BrdU experiments, it is important to note that BrdU was provided as a one-hour pulse immediately before harvesting. BrdU signal at 48, 72, or 96 hours therefore reports DNA synthesis occurring during that particular one-hour interval and does not measure the cumulative DNA synthesis that preceded the measurement. Consequently, relatively modest BrdU incorporation in cells that have already accumulated high DNA content does not indicate that the preceding increase occurred independently of DNA synthesis. Conversely, these experiments alone do not define the physical mechanism by which the additional DNA accumulated.

We agree that the requirement for XRCC4, XLF, and LIG4 is unexpected under a simple model of BIR. As noted above, the present experiments establish this genetic relationship but do not define its molecular basis. The differential effects of DNA-PKcs and downstream NHEJ factors suggest that individual NHEJ components may influence DIGA through functions that are not adequately represented by viewing the pathway simply as a linear ligation reaction.

Finally, we agree that the similarity between the flow-cytometric profiles produced by IR and MLN4924, as well as their common sensitivity to aphidicolin, does not by itself distinguish the underlying mechanisms. The distinction in the current study instead derives from their different genetic requirements, particularly the dependence of MLN4924-induced re-replication on CDT1 compared with the lack of such a requirement for DIGA, together with the differential effects of factors involved in DNA-end protection, resection, and DSB repair. These observations argue that DIGA is not simply the consequence of canonical origin re-licensing (i.e., rereplication or endoreduplication), while leaving open the possibility that additional replication-associated mechanisms contribute to the phenotype.

In summary, we appreciate the reviewers highlighting several important mechanistic questions raised by our findings. We view these questions as natural extensions of the initial discovery and characterization of DIGA. The present study identifies a large-scale DSB-associated increase in genomic DNA content and establishes important roles for DNA-end protection, resection, strand invasion, and BIR-associated factors in regulating this response. Determining the genomic architecture of the additional DNA, the sites and molecular intermediates from which synthesis originates, and the cellular determinants of DIGA susceptibility will be important goals for future studies.