A novel class of early replicating fragile sites that contribute to genome instability in B cell lymphomas

Genome-wide mapping of replication-induced DNA damage

Single strand DNA (ssDNA) mapping has been used to localize origins of replication in yeast (Feng et al., 2006). To identify potential sites of fork collapse, we first profiled the location and extent of ssDNA genome-wide using chromatin immunoprecipitation (ChIP) with an anti-RPA antibody (Figure 1). RPA associates with ssDNA at stalled forks near early firing origins when fork movement is inhibited by HU (Tanaka and Nasmyth, 1998).

Freshly isolated mouse B cells are arrested in the G₀ phase of the cell cycle (Figure 1A). Upon stimulation with LPS/IL4, cells synchronously enter into the cell cycle so that by 22 hours, approximately 8% of cells have entered S phase, while at 28 hours over 30% are in S/G2 phases (Figure 1A). To profile early replication origins, we treated cells at 22 hours with 10 mM HU for 6 hours to fully arrest cells at G₁/S (Figures 1A and 1B). We then performed ChIP-Seq of RPA in both untreated and HU-treated cells at 28h (Figures 1A and 1B). Two independent experiments showed reproducibility of genome-wide RPA association in HU-treated cells (Figure S1A). We generated profiles of RPA in untreated and treated samples, centered on individual RPA-bound sites (Figure S1B), and observed a marked increase in the intensity of RPA in HU-treated B cells relative to untreated cells where 5939 out of 11942 genomic regions (49.7%) displayed more than a 4-fold increase in RPA recruitment. In addition to the 53% overlap of RPA-associated regions between HU-untreated vs. treated cells, we also observed that 1441 regions were present only in HU-treated samples (Figure S1B). These HU dependent ssDNA regions may correspond to the firing of new replication origins to compensate for inefficient replication.

To confirm that RPA recruitment maps early replication zones, we used the Repli-Seq approach (Hansen et al., 2010) to identify replication origins in B cells during HU arrest. Approximately 12,000 early activating replication origins across the murine B cell genome were identified (Figure S1C). By comparing the distribution of BrdU incorporation relative to the individual RPA-occupied genomic regions, we observed association of BrdU incorporation with nearly 80% of RPA-bound regions (Figure S1C). Moreover, more than 86% of RPA/BrdU enriched genomic sites coincided with previously mapped early replicating regions in the mouse B cell line CH12 (Stamatoyannopoulos et al., 2012) (p(permutation) < 1×10⁻⁵, Figure S1D). Thus, HU-arrested B cells exhibited an enrichment of RPA at early replicating zones, consistent with an early S phase cell cycle arrest (Figure 1A).

Early replicating regions are associated with accessible chromatin configuration (MacAlpine et al., 2004). In agreement with this, we found that more than 67% of RPA-bound regions in HU-arrested cells reside within intragenic sequences (Figure S1E and S1I), a frequency significantly higher than expected (p(permutation) <1×10⁻⁵). Moreover, RPA preferentially associated with DNaseI hypersensitive sites (DHS) and euchromatic promoters marked by H3K4me3 (Figure S1F). Finally, we measured transcriptional activity in HU-treated B cells directly by genome-wide RNA sequencing. We observed high transcription activity within the RPA-occupied genomic regions as shown by the aggregated pattern of RNA-Seq centered on those regions (Figure S1G). Moreover, 6100 RPA-bound RefSeq genes exhibited significantly higher average mRNA abundance than those that did not show RPA binding (p< 1×10⁻¹⁶, Figure S1H). Thus, HU-induced RPA recruitment in early S phase maps to actively transcribed genes that show the hallmarks of euchromatin..

Replisome stalling in response to HU triggers the activation of the ATR kinase (Ward and Chen, 2001), which protects forks from collapse (Cimprich and Cortez, 2008). and leads to phosphorylation of H2AX (γ-H2AX) (Ward and Chen, 2001), which colocalizes with RPA (Petermann et al., 2010). To examine the relative distribution of γ-H2AX and RPA genome-wide, we carried out ChIP-Seq with an antibody that recognizes γ-H2AX (Figure S1A) and examined their profiles with respect to the center of RPA-bound sites. γ-H2AX-associated genomic regions were much broader than RPA, but these regions overlapped with 93% of RPA-bound sites marking ssDNA in HU-treated cells (Figure 1C), consistent with the finding that γ-H2AX marks stalled forks even prior to DSB formation (Petermann et al., 2010). γ-H2AX/RPA enriched loci may therefore correspond to a combination of stalled and broken replisomes.

Cells deficient in homologous recombination (HR) pathway components, such as XRCC2, often accumulate spontaneous chromosome breaks and exhibit hypersensitivity to HU (Sonoda et al., 1998). Consistent with increased spontaneous DNA damage at replication forks, untreated XRCC2^−/− cells exhibited accumulation of γ-H2AX at similar genomic regions and at almost similar levels observed in HU-treated WT B cells (Figure S2A–C). 90% of γ-H2AX –associated genomic regions in untreated XRCC2^−/− cells correlate with the regions enriched for this protein in HU-treated WT B cells (Figure S2B), and nearly 80% of the regions with enriched γ-H2AX observed in HU-treated WT B cells overlapped with those seen in HU-treated XRCC2^−/− cells (Figure S2C). These data indicate that XRCC2 deficiency leads to increased endogenous levels of replication stress mostly at the same loci where HU induces replication fork stalling and/or breakage in WT cells.

RPA, BRCA1 and SMC5 colocalization marks the sites of replication stress in early replicating zones

Like XRCC2, BRCA1 and members of the structural maintenance of chromosome (SMC) family have been implicated in promoting replication fork restart (Schlacher et al., 2012; Stephan et al., 2011). To determine whether HR proteins bind to a subset of stalled forks marked by RPA and γ-H2AX, we also defined the genome-wide profile of BRCA1 and SMC5. We confirmed BRCA1 and SMC5 ChIP-Seq efficacy by observing their association at both Sμ and Sγ1 in 53BP1^−/− cells, where the breaks in IgH persist unrepaired and undergo extensive resection (Figure S3A) (Bothmer et al., 2010; Bunting et al., 2010; Yamane et al., 2011; Yamane et al., 2013).

We then determined the localization of BRCA1 and SMC5 in HU-arrested B cells. Two independent experiments showed reproducibility of genome-wide BRCA1 and SMC5 association (Figures S3B and S3C). To identify the RPA genomic sites co-occupied by the HR proteins BRCA1 and SMC5, we plotted the distribution of their binding with respect to the center of individual RPA-bound regions. Overall, 2204 regions spanning 10 kbp on average showed RPA/BRCA1/SMC5 triple co-localization (Figures 1D and 1E). We found that RPA was recruited to more than 88% of genomic sites exhibiting BRCA1 and SMC5 association (Figure 1E). Furthermore, genome-wide analysis of RPA/BRCA1/SMC5 profiles in untreated cells revealed more than a 21% increase in the number of genomic regions occupied by these three proteins after HU treatment (Figure S4A). Nevertheless, 48% of RPA/BRCA1/SMC5 triple co-localizations were common between the unperturbed and HU-arrested B cells (Figure S4A). Therefore, we hypothesized that chromatin with concomitant RPA, BRCA1 and SMC5 binding might correspond to regions undergoing replication fork collapse both in response to replication stress and during normal DNA replication. Given that our analysis focused on early replicating sites, which contrasts with late replicating CFSs, we designated these regions as “Early Replication Fragile Sites” (ERFS).

We then characterized ERFSs to determine if they share common underlying primary sequence characteristics. Indeed, these loci were enriched at known repetitive elements, including LINE L2, SINE, DNA transposons and tRNA elements (p(permutation) < 1×10⁻³, Figure 1F) which are known replication fork barriers (Mirkin and Mirkin, 2007). Furthermore, ERFSs showed significantly higher G and C nucleotide content compared to the whole mouse genome, in contrast to CFSs that are enriched in A+T sequences (p(Wilcoxon) < 1×10⁻¹⁶, Figure S4B). 26% of the ERFSs regions overlapped with CpG islands, which are highly enriched at translocation breakpoints in B cell lymphoma (Tsai et al., 2008). Conversely, CpG islands covered approximately 400,000 nucleotides within these regions (p(permutation) < 1×10⁻⁵, Figure 1G). As anticipated, ERFSs clustered at early replication origins (Figure S4C), and over 66% of the loci overlapped with intragenic or promoter sequences of RefSeq annotated protein coding genes (p(permutation) < 1×10⁻³, Figures S4D–E). Moreover, ERFSs are more transcriptionally active relative to flanking genomic regions shown by relative mRNA enrichment by RNA-Seq (Figure 1H). Indeed, more than 86% of the RefSeq annotated genes with ERFSs are among the highest transcribed genes (p(binomial) < 1×10⁻¹⁶, Figure S4F). Finally, ERFSs were significantly enriched in gene pairs that are transcribed in converging or diverging directions (see methods), such as the convergent transcription pair of IKZF1 and FIGNL1 shown in Figure 2A. Compared to expected values, ERFSs were at least 2 times more likely to localize in regions containing gene pairs exhibiting convergent and/or divergent gene pairs ((p(permutation) < 1×10⁻⁵, Figure 1I).

Replication is organized into discrete zones (30–450 kbp in size) containing multiple replication origins that exhibit similar replication timing (Costa and Blow, 2007). Similarly, approximately 80% of the ERFSs are within 300 kbp of one another (Figure 2A; Figure S4G). We therefore integrated these neighboring clustered ERFSs and removed those with footprints less than 10 kbp to define 619 triple co-localized hotspot regions (Figure 2B; Table S1). Interestingly, whereas these hotspots were distributed throughout the genome, the density of hotspots on autosomes was higher than on the sex chromosomes, which have a lower gene density (Figure 2B). An examination of the top 15 hotspots based on a ranked statistics of RPA/BRCA/SMC5 binding strength showed that 9 out of the 15 regions contained gene clusters with at least 3 genes, and 12 out of 15 exhibited divergent/convergent gene pairs (Figure 2A and 2C, Table S1). Of note, 8 out of 15 hotspots are also rearranged in B cell lymphomas (Figure 2C; Table S2), suggesting a possible link between ERFSs, genome rearrangements and cancer (see below).

Early S phase arrest by HU induces DNA damage at ERFSs but not at CFSs

DNA damage at CFSs is visualized by conventional cytogenetic analysis of metaphase chromosomes (Durkin and Glover, 2007). To investigate whether the ERFSs defined by RPA/BRCA1/SMC5 binding are prone to actual breakage, we again treated cells with 10 mM HU, released them into fresh medium overnight, and examined metaphase spreads. Chromatid breaks, chromosome breaks and rearrangements could be discerned in 20–60% of WT cells after HU treatment (Figure S2D). To determine whether ERFSs are more sensitive to breakage under replication stress than regions lacking RPA/BRCA1/SMC5 binding (i.e. coldspots), we hybridized metaphases with BAC probes corresponding to six ERFS hotspots (MHCII, GIMAP, SWAP70, BACH2, IKZF1 and FOXP1) (Figures 3A and 3B), two coldspots (CNTNAP4 and SLITRK6) and two CFSs (FRA8E1 and FRA14A2). For each of the six ERFS hotspots a total of at least forty chromosome aberrations were counted (Table S3). Notably, all 6 ERFS hotspots displayed chromosome aberrations in metaphases from HU-treated samples (Figure 3B). In contrast, neither of the cold regions or CFSs was broken under the same conditions (Figure 3B). Overall, 8–15% of the total damage localized to individual ERFS hotspots, representing a significant fraction of the total damage (Figure 3B). DNA lesions were observed on either the centromeric or telomeric sides of ERFS-specific hybridized BAC (Figure S2E), suggesting that an ERFS represents a potentially large fragile genomic region.

Aberrations at ERFS hotspots were also detected in XRCC2^−/− cells treated with HU (Figure 3C). XRCC2^−/− cells are more sensitive to HU than WT, as evidenced by the higher level of total damage in these cells (Figure S2D). Breaks at MHCII, GIMAP, SWAP70, BACH2, IKZF1 and FOXP1 were found in 5–10% of HU-treated XRCC2^−/− cells compared with 1–6% of WT cells damaged in these regions (Table S3). Nevertheless, the frequency of ERFS-specific instability relative to the total damage was similar in XRCC2^−/− and WT cells (Figures 3B and 3C). Interestingly, spontaneous breaks in the vicinity of the GIMAP hotspot were detectable spontaneously in XRCC2^−/− cells (Figures 3C and 3D; Table S3), which is consistent with increased γ-H2AX observed in unchallenged XRCC2-mutant cells (Figure S2A).

None of the eight CFSs defined in mouse (Helmrich et al., 2006) were among our 619 ERFS hotspots (Table S1). Consistent with this, DNA aberrations at two of the most expressed CFSs in mouse lymphocytes, FRA14A2 and FRA8E1 (Helmrich et al., 2006) were undetectable in HU-treated WT samples (Figure 3B). Absence of CFS expression could be explained by the fact that high concentrations of HU stall replication forks in early S phase (Figure 1A), whereas CFSs replicate late (Durkin and Glover, 2007). Conversely, we found that overnight treatment with low doses of aphidicolin (0.2 μM for 20 h) induced damage at the CFSs FRA14A2 and FRA8E1, while the ERFSs GIMAP and SWAP70 were largely insensitive (Figures 3E and 3F). These data are consistent with the idea that ERFS arise from fork collapse during early replication, while breakage at CFSs arises from a failure to replicate (Debatisse et al., 2012), and the two forms of replication stress induce distinct types of recurrent DNA lesions.

ATR inhibition promotes ERFS and CFS expression

The ATR kinase protects the genome from chromosomal aberrations at late replicating CFSs, (Durkin and Glover, 2007) and is essential for stabilizing stalled forks and facilitates fork restart in early S phase (Cimprich and Cortez, 2008). To confirm that ATR inactivation induces CFSs and determine whether it similarly leads to damage at ERFSs, we treated asynchronous B cells on day 2 with 1 μM of a recently described ATR inhibitor (ATRi) (Toledo et al., 2011). We found that approximately 2.5% and 7.0% of the total chromosomal aberrations localized to the two CFSs, FRA8E1 and FRA14A2, respectively (Figure 4A). ATR deficiency also led to chromosomal aberrations at ERFSs at a similar frequency (Figure 4A; Table S3). Moreover, ERFSs and CFSs were both damaged in XRCC2^−/− cells treated with ATRi (Figure 4A). Thus, the rupture of un-replicated regions at CFSs and fork collapse at ERFSs are similarly sensitive to ATR inhibition.

Transcriptional activity can increase ERFS fragility

As described above, ERFSs are enriched in regions with high transcriptional activity (Figure 1H; Figure 2C; Figure S4F; Table S1). To determine the contribution of transcriptional activity to individual ERFSs, we focused on loci with tissue-specific transcription patterns. SWAP70 is a B cell specific developmental regulator whereas genes within the GIMAP cluster are expressed both in B and in T cells (Figure 4B; Figure S5A). Treatment with ATRi led to a similar frequency of damage at GIMAP in B and T cells, consistent with insignificant changes in gene expression between the two cell types (Figure 4C). In contrast, damage near SWAP70 was 3-fold lower in T than in B cells (Figure 4C; Table S3), which correlated with the decreased transcription of SWAP70 in T cells (Figure 4B). Nevertheless, the replication timing near SWAP70 was similar in both cell types (Figure S5B). To further delineate the role of transcription on ERFS breakage, we used SWAP70^−/− mice in which 2.7 kbp including the first exon and part of the 5′ untranslated region is removed (Borggrefe et al., 2001), allowing us to compare the fragility of ERFSs in the same genomic region in knockout B cells. We determined that SWAP70 mRNA in SWAP70^−/− B cells was reduced by approximately 4-fold relative to WT (Figure 4D). Moreover, DNA damage near SWAP70 was approximately 2.5 fold lower in SWAP70^−/− relative to WT B cells (Figure 4E). In contrast, DNA damage near GIMAP remained at a similar level both in WT and SWAP70^−/− cells (Figure 4E). While our data indicates that high level of transcription contributes to the breakage of some ERFSs, other molecular features including repetitive elements (Figure 1F), covalently bound protein complexes and RNA:DNA hybrids might also be sources of ERFS fragility.

Oncogenic stress can trigger ERFS and CFS fragility

Oncogene deregulation is thought to compromise genome integrity preferentially at CFSs (Bartek et al., 2007; Halazonetis et al., 2008), and CFS deletion has been associated with various cancers (Bignell et al., 2010). To determine whether oncogenic stress similarly induces DNA damage at ERFSs, we overexpressed c-myc in B cells, as it has been implicated in regulating replication initiation and origin firing (Dominguez-Sola et al., 2007). XRCC2^−/− cells were utilized to increase the amount of replicative stress and DNA damage as a result of decreased HR (Figure S2D). c-myc overexpression lead to induction of p53 (Figure 5A), which correlated with an approximately 1.6-fold increase in overall DNA damage in XRCC2^−/− cells overexpressing c-myc compared to empty vector (EV) infected cells (Table S3). Moreover, 7.3% of the total breaks generated in c-myc overexpressing cells were found near SWAP70, compared to 2.4% of total breaks at this ERFS in EV-infected B cells (Figure 5B). Similarly, out of 43 breaks observed in c-myc-infected cells, 3 (7%) were found at the GIMAP cluster and 3(6.7%) were found near BACH2. c-myc overexpression also induced breaks at FRA8E1, showing a 2-fold relative increase in breaks relative to EV-infected cells (Figure 5B). Thus, DNA damage induced by c-myc overexpression can occur at ERFSs and CFSs.

ERFS fragility is AID independent

Mutations and DSBs at various oncogenes including c-myc are due to AID off-target activity (Robbiani et al., 2008). Recently, a number of genome-wide studies in primary B cells mapped AID-induced DNA translocation events, and identified several novel hotspots for AID-dependent translocations at non-Ig genes (Chiarle et al., 2011; Kato et al., 2012; Klein et al., 2011). Among these translocation hotspots, MHCII, GIMAP, IKZF1, PVT1, ETS1, IRF4 and NFkB1 were located within top 15 ERFS hotspots in this study, whereas the IgH locus (the physiologic target of AID) was not ranked high among the list (Figure 2C; Table S1). To determine if AID contributes to ERFS fragility, we stimulated WT and AID knockout B cells with LPS/IL4 for two days, and then treated them with ATRi overnight. These conditions induce robust AID-dependent DNA damage simultaneously with replication stress. We probed metaphases with BACs spanning the IgH locus, the GIMAP cluster, and IKZF1- all AID translocation hotspots- as well as BACH2, SWAP70, FOXP1 and BCL2 (Figure S2E)- ERFSs that are frequently rearranged in B cell lymphoma (Figure 2C; Table S1; Table S2). In WT, the IgH locus was damaged in 3.8% of cells, but the frequency of IgH specific instability did not increase with ATRi (Figure S2F), despite the fact that ATRi greatly increased overall damage (Table S3). Upon ATRi treatment, the frequency of breaks at the ERFSs GIMAP, IKZF1, BACH2, SWAP70, FOXP1 and BCL2 were elevated to the levels similar to those observed at the IgH in activated B cells (Figure S2F). Breaks at some ERFSs were even spontaneously detected (FOXP1 and GIMAP, Figure S2F).

To determine if AID expression contributes to aberrations observed at ERFSs, we next analyzed their breakage frequency in AID^−/− cells. Unlike WT cells, IgH breaks were absent in AID^−/− cells. In contrast, all ERFSs exhibited similar levels of breakage both in WT and AID^−/− cells (Figure 5C; Table S3). Therefore, whereas IgH breaks in B cells are entirely AID-dependent, the breakage of ERFSs is AID-independent. Altogether, these data suggest that some recurrent rearrangements in B cell lymphoma are due to AID-independent replicative stress at ERFSs.

Genome instability at ERFSs is observed in mouse models and human cancer

Among the top 15 ERFS hotspot that break in response to AID-independent replication stress, we have identified three partners that recurrently translocate to IgH in lymphomas: BACH2, FOXP1, and BCL2 (Table S2). We hypothesized that if AID-dependent DSBs in G1 persisted into early S phase, translocations between AID-dependent breaks and ERFS might be detectable. To test this, we examined cells transgenically overexpressing AID and are simultaneously deficient for 53BP1 (IgκAID/53BP1^−/−), thus allowing the persistence of G1 IgH breaks into S phase where they could be joined to ERFSs. Indeed, 26% and 7% of IgκAID/53BP1^−/− B cells carried IgH locus and BACH2 breaks, respectively (Figure 5D). These breaks are fusogenic, since IgH- and BACH2- associated translocations to unidentified partner chromosomes were found in 7.3% and 1.2% of the metaphases respectively (Figure 5D). Importantly, we also detected one IgH/BACH2 translocation among 750 cells (Figure 5E), reminiscent of the translocations observed in human B cell lymphoma (Kobayashi et al., 2011). Thus, AID-dependent breaks generated in G1 (Petersen et al., 2001) can join to ERFS breaks triggered in early S phase.

A hallmark of cancer genomes is widespread copy number changes, insertions and deletions. To determine whether deletions and/or amplifications at ERFSs are a general feature of the B cell lymphoma genome, we compared our ERFSs with high resolution copy number changes detected in biopsies of patients with diffuse large B cell lymphoma (DLBCL), the most common type of non-Hodgkins lymphoma (Lenz et al., 2008). A total of 190 “minimal common regions” (MCRs) were found among 203 biopsies, carrying a gain or a loss of a chromosomal region ranging in size from 5 kbp to 21Mbp (Lenz et al., 2008). Mouse ERFS coordinates were overlaid onto the human genome using two methods, yielding 2205 syntenic regions (Figure S6B–D). Notably, 51.6% of the MCRs observed in primary DLBCL overlapped with syntenic ERFS regions (p(permutation)<1×10⁻⁴, Figure 5F,). Moreover, 20.4% of ERFSs overlapped with MCRs, 32% higher than expectation (p(permutation)< 1×10⁻⁶, Figure 5G). Surprisingly, ERFS were deleted or amplified in DLBCL at least 81% more frequently as compared to CFSs, despite their cancer-specific propensity for breakage (Figure 5G). Moreover, our analysis indicated that the DLBCL copy number alterations exhibited 2-fold higher correlation with B-cell ERFSs compared to deletions and/or amplifications in T-lineage acute lymphoblastic leukemia (Figure S6A) (Zhang et al., 2012a). Finally, by examining homozygous deletions in cancer genomes (Bignell et al., 2010), we found that 25 out of 64 genes known to contribute to oncogenesis coincide with ERFSs (p(hypergeometric) < 6×10⁻²⁰, Figure 5H; Table S4). Based on these findings, we conclude that ERFSs are a significant feature of the mutational landscape of diffuse large B cell lymphomas, and potentially other cancers.

Replicative stress at ERFSs contributes to genome instability in B cells

While AID has been implicated in B cell translocations (Gostissa et al., 2011), very little is known about the mechanisms of chromosomal breakage at several IgH-partner loci including BCL2, BACH2 and FOXP1. Besides programmed DNA damage, replication-based mechanisms are a major contributor to chromosomal instability in cancer (Liu et al., 2012). Activated B cells are among the most rapidly dividing mammalian cells (Zhang et al., 1988), which potentially exposes them to high endogenous levels of replicative stress. Here, we have used a genome-wide approach to identify a subset of early replicating regions in the B cell genome that are particularly vulnerable to fork collapse and contribute to rearrangements in B cell malignancies. In our model, ERFS breaks can occur after the generation of unrepaired AID-induced breaks in G1, and the two breaks could recombine during S or G2. Alternatively, ERFS damage might persist through mitosis resulting in DNA breaks in the subsequent G1 phase when AID is predominantly active. In either case, we suggest that AID-mediated DSBs in G1 (Petersen et al., 2001), together with replication stress-induced damage at recurrent loci, can coordinately drive B cell lymphoma initiation and progression (Figure 6).

ERFS versus CFS

CFSs are considered to be the most replication stress-sensitive sites in the genome (Durkin and Glover, 2007). Although no single mechanism accounts for CFS instability, it is hypothesized that a number of different characteristics may contribute to their fragility including co-occurrence with very large genes, late replication, low density of replication origins, high A–T content and sequences prone to form secondary structures, histone hypoacetylation and a condensed chromatin structure (Helmrich et al., 2011; Jiang et al., 2009; Letessier et al., 2011; Ozeri-Galai et al., 2011). In stark contrast to CFSs, our identified ERFSs replicate early, have an open chromatin configuration, and are origin-, gene-, and G-C-rich.

Despite these diametrically opposite properties, both CFS and ERFS fragility are increased by ATR inhibition (Figure 4A), oncogenic stress (Figure 5B) and deficiencies in HR (Figure 3C) (Bartek et al., 2007; Durkin and Glover, 2007; Halazonetis et al., 2008). These conditions decrease the rate of fork progression but concomitantly increase the density of replication initiating events (Bester et al., 2011; Daboussi et al., 2008; Dominguez-Sola et al., 2007; Shechter et al., 2004) which might contribute to the damage at both CFSs and ERFSs respectively. The decrease in fork speed hinders the completion of replication at CFSs, either because of the scarcity of origins near CFSs (Letessier et al., 2011), the heterochromatic nature of the regions that would limit accessibility of DNA replication and/or DSB repair machineries (Jiang et al., 2009), or because of the interference between transcription and replication at very large genes (Helmrich et al., 2011). While additional origins are not activated near CFSs upon replication stress (Letessier et al., 2011), an increase in origin activity at early replicons might paradoxically contribute to genome instability at ERFSs. For example, increasing the replication initiation events near highly transcribed gene clusters with divergent and/or convergent gene pairs could increase conflicts between DNA replication and transcription machineries. The higher density of activated origins at ERFSs would also be expected to prematurely deplete nucleotide pools (Bester et al., 2011), thereby increasing the probability of subsequent fork stalling and collapse. These two outcomes of replication stress are likely to be linked since increased replication initiation and depletion of nucleotide supplies slows replication (Bester et al., 2011; Jones et al., 2012), while slow fork progression causes activation of dormant origins (Ge et al., 2007), and both incomplete replication and increased origin firing are monitored by ATR activity (Shechter et al., 2004). In conclusion, increased initiating events at ERFSs and a paucity of replication initiation at CFSs could both challenge replication fidelity.

ERFSs and cancer

Oncogenic stress is a major driving force in the early stages of cancer development (Halazonetis et al., 2008); nevertheless, the factors that trigger replicative stress in vivo remain unclear. In the case of B cell lymphomas, oncogenic stress can be initiated by the activity of AID, which by targeting non-Ig genes such as c-myc (Robbiani et al., 2008), leads to c-myc/IgH translocations and consequent aberrant c-myc expression. This form of AID induced oncogenic stress or high levels of proliferative activity in activated B cells could generate DNA damage at ERFSs (Figure 6).

Altogether 103 AID hotspots (Chiarle et al., 2011; Klein et al., 2011) —including the GIMAP cluster, MHCII locus, and IKZF1—were also identified as ERFS hotspots in this study (Table S1). It is possible that the overlap observed between a subset of off-target AID sites and ERFSs is due to common underlying features of these loci. For example, AID is recruited to ssDNA regions (Chaudhuri and Alt, 2004), which are also generated during replicative stress; AID-dependent DSBs and ERFSs are also both enriched in repeat elements (Staszewski et al., 2011). In addition, chromosomal regions with the highest transcriptional activity have the highest AID-dependent translocation density (Chiarle et al., 2011; Klein et al., 2011), and early origins and translocations frequently reside near transcription start sites and RNA polymerase II binding sites (Chiarle et al., 2011; Klein et al., 2011). Thus, these euchromatic regions could serve both as AID targets in G1 and also be susceptible to fork collapse during early S phase.

A number of different hypotheses have been put forward about the mechanisms that promote recurrent translocations in mature B cell lymphomas. These include recurrent genomic damage by AID, random DNA damage followed by selection, and a non-random three-dimensional organization of the genome (Chiarle et al., 2011; Hakim et al., 2012; Klein et al., 2011; Zhang et al., 2012b). To date, replication stress-induced DNA damage has been associated with late-replicating CFS. By using an alternative experimental approach for the discovery of fragile site expression during early replication, we have identified a novel source of recurrent AID-independent DNA breaks that may play a mechanistic role in some of the most common genome rearrangements during B cell lymphomagenesis. Because transcriptional activity and replication timing of a genomic region vary among different cell lineages (Hansen et al., 2010), different sets of ERFSs might also account for recurrent chromosomal rearrangements in cancers of distinct cellular origins.

Mice

XRCC2^−/−(Frappart et al., 2009), 53BP1^−/− (Ward et al., 2004), Igκ AID (Robbiani et al., 2009), AID^−/− (Muramatsu et al., 2000) and SWAP70^−/− (Borggrefe et al., 2001) mice have been described. SWAP70^−/− and WT control mice used in Figure 4D are C57BL/6 background, all other mice are 129/Sv x C57BL/6 background.

ChIP-Seq, Repli-Seq, RNA-Seq, DHS I Mapping, and FISH analysis

ChIP-Seq and RNA-Seq procedures were performed as in (Yamane et al., 2011), Repli-Seq was performed as described in (Hansen et al., 2010). DHSI mapping was performed as described (Sekimata et al., 2009), and FISH analysis is described in (Callen et al., 2007). For detailed methods, see Extended Experimental Procedures.

BACs

Individual BACs to ERFSs were identified using NCBI clone finder, and purchased from BACPAC For complete list of BAC probes used in FISH experiments see Extended Experimental Procedures.

Retroviral infection

Cells were infected with pMX-c-Myc-IRES-GFP or empty vector and GFP-positive cells were sorted as described (Robbiani et al., 2008).

Statistical and computational analyses

Detailed description is available in Extended Experimental Procedures.