Endogenous DNA damage at sites of terminated transcripts

Abstract

DNA damage promotes mutations that fuel cancer, ageing and neurodegenerative diseases^1-3, but surprisingly, the causes and types of damage remain largely unknown. There are three identified mechanisms that damage DNA during transcription: collision of RNA polymerase (RNAP) with the DNA-replication machinery head-on and co-directionally^4-6, and R-loop-induced DNA breakage^7-10. Here we identify novel DNA damage reaction intermediates^11,12 and uncover a fourth transcription-related source of DNA damage: endogenous DNA damage at sites of terminated transcripts. We engineered proteins to capture single-stranded DNA (ssDNA) ends with 3' polarity in bacterial and human cells. In Escherichia coli, spontaneous 3'-ssDNA-end foci were unexpectedly frequent, at one or more per cell division, and arose via two identifiable pathways, both of which were dependent on DNA replication. A pathway associated with double-strand breaks was suppressed by overexpression of replicative DNA polymerase (pol) III, suggesting competition between pol III and DNA damage-promoting proteins. Mapping of recurrent 3'-ssDNA-ends identified distinct 3'-ssDNA-end-hotspots, mostly unrelated to double-strand breaks, next to the 5'-CCTTTTTT transcription-terminator-like sequence. These 3'-ssDNA-termini coincide with RNA 3'-termini identified by DirectRNA sequencing¹³ or simultaneous 5' and 3' end RNA sequencing (SEnd-seq)¹⁴ and were prevented by a mutant RNAP that reads through terminators. Our findings reveal that transcription termination or pausing can promote DNA damage and subsequent genomic instability.

Extended Data Figure 1 |. Construction and validation of E. coli DExI fusion proteins.

a, Diagram of the chromosomal inducible non-tagged DExI, DExI-GFP, DExI-mCherry and DExI-FLAG fusion constructs used in this study, strains SMR22601, SMR22593/SMR24774, SMR24800 and SMR24776, respectively. The doxycycline- and temperature-inducible expression cassettes were engineered at a non-genic site in the E. coli chromosome (Methods), in cells that carry the wild-type gene in its native locus. Because DExI is dominant to the wild-type gene, DExI phenotypes are observed here, as previously. The doxcycline-inducible expression system consists of the phage PN25 constitutive promoter-driven tetR repressor gene and the inducible P N25tetO promoter-driven DExI fusions (left and middle). The temperature-inducible expression system consists of the phage lambda cIts857 temperature-sensitive repressor gene and lambda PR promoter-driven DExI fusion (right). b, DExI-mCherry fusion forms fewer spontaneous foci than DExI-GFP. DExI focus appearance frequencies with indicated fusions in log-phase cells. Mean ± range, n = 2 experiments. c, Coomassie-stained image of recombinant DExI-GFP-6xHis (herein referred to as DExI-GFP). Representative image of two, with similar results. For gel source data, see Supplementary Figure 1. d-f, DExI-GFP binds only 3’-ssDNA-end-bearing substrates detectably. d, DExI-GPF binds ssDNA, but not dsDNA or ssRNA. Left, the polynucleotide binding properties of DExI-GFP were examined by electrophoretic mobility shift assay (EMSA). 1 nM ³²P-radiolabeled ss/dsDNA and ssRNA probes were incubated with indicated concentrations of DExI-GFP. Shift of polynucleotide relative to no-protein control indicates protein binding. Right, quantification from three experiments: Whereas the binding to ssDNA with a 3’ end has a dissociation constant (Kd) of 0.6 ± 0.4 nM (KDs averaged from n = 2 experiments, mean ± range) the binding to blunt dsDNA; ssRNA; and 5’-ssDNA overhanging tail are undetectable (ND) even with 50 μM DExI-GFP protein, indicating that their binding by DExI-GFP is weaker than 50 μM, at least 680 times weaker than to 3’-ssDNA ends. This was also the case for binding to e, circular ss or dsDNA, mean ± S.E.M for n = 3 experiments for ssDNA. f, binding to closed circular dsDNA, nicked circular dsDNA and dsDNA nicked at the consensus sequence. EMSA analysis of 350 ng circular ss or dsDNA substrates was performed with indicated concentrations of DExI-GFP or human replication protein A (RPA). The latter was used as a positive control for DNA binding. Representative image of n = 6, n = 6, n = 3 experiments for ssDNA, intact and nicked dsDNA, respectively. For gel source data, see Supplementary Figure 1. g, DExI production mimics the phenotype of cells lacking Exo I and Exo VII, supporting its role blocking Exo VII degradation of DNA. Data from. Assays of “RecF-pathway” homologous recombination (HR) of linear phage lambda DNAs show that the sbcB15 (or DExI) mutation mimics the phenotype of cells that carry both the deletion of the Exo I (∆xonA) and of Exo VII (∆xseA) simultaneously, and so lack all known single-strand-dependent 3’ exonuclease activity in E. coli. Removal of Exo VII 3’- and 5’-single-strand-specific exonuclease (∆xseA) simultaneously with Exo I (∆xonA) restores HR proficiency to recBC sbcC cells as well as the SbcB15-mutant (DExI) protein does. [xthA encoding Exo III, an apurinic-apyrimidinic (AP) endonuclease, made no detectable contribution.] Diagram (above): two phage lambda genotypes co-infected (“crossed”) in the strains indicated below (single lines represent double-stranded DNA molecules). The lambda lack their own red HR system (bio1 substitution or ∆b1453) and HR-related genes encoded in the nin region (∆nin5) and so recombine via E. coli proteins. The lambda replicate via rolling circle and so do not require any HR for formation of lambda genome multimers, required for lambda genome packaging (and so for progeny phage, per). These data imply that the inability of ∆xonA to restore HR to recBC sbcC cells results from Exo VII action, and that blocking Exo VII with SbcB15-mutant (DExI) protein prevents Exo VII-inhibition of HR. HR is quantified as the titer of lambda E+ S+ recombinant particles among Ets3 parental phage from crosses with limiting Ets parent, 50-times fewer of than of the other lambda genotype employed, using methods per^,. These are means ± s.e.m. of ratios, normalized to the value from the recBC sbcC sbcB15 strain done in parallel for each genotype with the numbers of independent experiments for each bar as follows, from left to right: n = 6, n = 5, n = 6, n = 3, n = 3. ***p ≤ 0.001, one-way ANOVA with multiple comparison test.; n.s., not significant. h, How DExI promotes and Exo I, Exo VII, and SbcCD hairpin endo/exonuclease inhibit “RecF-pathway” recombination at DSB ends. Cells that lack the RecBCD double-strand endo/exonuclease require both an sbcCD null mutation and DExI (SbcB15 protein) to restore UV-resistance and HR proficiency (and Fig 1b). Above: in ∆recB DExI-producing cells: DExI (SbcB15 protein, blue) prevents Exo I and Exo VII ssDNA-dependent exonucleases (red notched circle) from degrading 3’-ssDNA ends^, (and g, above). The SbcCD hairpin endo/exonuclease (orange notched circle) degrades DNA with hairpin secondary structures independently of Exo I and Exo VII action. Exo I degrades 3’-ssDNA ends and Exo VII degrades ssDNA ends of 3’ or 5’ polarity. Below: in ∆recB ∆sbcC DExI cells, blocking degradation of 3’-ssDNA ends with DExI protein and preventing hairpin endo/exo activity by sbcC deletion is proposed to allow persistence of the single-stranded DNA with 3'-ends, required for RecA loading and homology-directed repair (HDR). i, Efficiency of DExI-GFP focus formation after gamma irradiation. The efficiency of DExI focus formation depends on whether each DSB contains one or two resected DSB ends. Thus, it would be 0.5 foci per DSB if these breaks were mostly one-ended, or 0.25 foci per DSB if they were mostly two-ended. In E. coli cells, the directional distribution of nuclease-impeding Chi sites leads to more rapid directional nucleolytic destruction of one of the two chromosome arms or DSB ends than the other, and the ratio of one-to-two-ended species is not known. j, Representative example (above) and quantification of western blots for Exo I-GFP under the native xonA promoter or DExI-GFP driven by the chromosomal inducible P_N25tetO promoter (means ± s.e.m., n = 6 experiments for all except the rightmost value, for which n = 3). From these data we can approximate DExI concentrations. On induction, about 94-times more DExI is produced than native Exo I, also present; and uninduced, about 1.7 times more DExI than native Exo I. For gel source data, see Supplementary Figure 1. E. coli strains used: b: SMR24774, SMR24800; g: SMR121, SMR1700, SMR1701, SMR1775, SMR3108; i: SMR24774; j, SMR27431, SMR27378, SMR24774

Extended Data Figure 2 |. E. coli DExI rapidly accumulates at laser-induced damage in human cancer cells.

a, Diagram of EmGFP-DExI induction and live-cell imaging after laser micro-irradiation in U2OS cells. b, EmGFP-DExI accumulates at sites of laser-induced DNA damage. Scale bar, 10 μm. c, Quantification of b. EmGFP fluorescence intensity at damage sites was calculated by comparing the EmGFP-DExI intensity at damaged versus undamaged sites within the same cell as a function of time. Mean ± s.e.m. n = 12 cells.

Extended Data Figure 3 |. Spontaneous 3’-ssDNA-end foci are frequent.

a, Representative images of cells grown in microfluidic chamber, images that record complete rounds of cell division. DExI- GFP-producing cells revealed short- or long-lived spontaneous foci, whereas no foci were observed in GFP-producing cells. b, DExI inhibits cell division. Scatter plot of division times in indicated strains. “no foci” indicates induced cells that are producing DExI-GFP or GFP, but had no visible focus at the time photographed. Cumulative data from n = 3 experiments, one-way ANOVA with multiple comparison test. c, Dose response of increased DExI-GFP induction with increasing inhibition of growth rate. n = 3 experiments, two-sided Student’s t test, asterisks from left to right,**p = 0.0010, **p = 0.0013, ***p = 0.0002, ***p = 0.0001. d, Reduction of chromosome number in Exo I-producing cells/flow cytometry histogram of DNA content (DAPI fluorescence intensity) of Exo I-producing cells after replication run-off. These controls for the replication run-off assays in Fig. 3g produce wild-type Exo I and showed fewer chromosome copies per cell than the wild-type (Fig. 3g), even though they did not display the failure to complete replication seen in DExI-producing cells (Fig. 3g). These data support the interpretation that the spontaneous 3’-ssDNA-ends detected by DExI (Figs. 1 and 3) are generated normally by replication, not as a result of the mutant DExI protein, and so are rapidly degraded in Exo I-producing cells. That is, the spontaneous 3’-ssDNA-ends detected by DExI (Figs. 1 and 3) are not an artifact of producing DExI, and its persistent binding to DNA. e, Although the frequency of mutations is increased by DExI, DExI induced little change in the sequences of cycA gene mutations compared with spontaneous mutations. Summary of mutation sequences from 49 independent clones each, isolated from each of spontaneous and DExI-induced mutants. The two blue dots indicate mutations that occurred together in one mutant isolate, and the two green dots in a different mutant isolate. Each of the rest occurred singly in each sequenced isolate. f, Additional controls for Fig. 3i. No significant reduction of spontaneous mutation rate by rpoB2 or ∆greA ∆greB in cells without DExI. Mean ± s.e.m. n = 4 experiments; one-way ANOVA, ns not significant. g, Increased (not reduced) mutagenesis from increased production of pol IIIα suggests that suppressing the pol IIIα-suppressible DSB pathway may have increased the contribution to mutagenesis of the RNAP-associated pathway, which promotes mutations at cycA. Given that the RNAP-associated pathway requires replication (Fig. 3a), the mutagenesis increase might result from the proposed STABD hypothesis (Fig. 5g), which requires two events to form the RNAP-promoted 3’-ssDNA ends trapped by DExI: (i) RNAP cleavage of the DNA, and (ii) replisome displacement of the cleaved strand (Fig. 5g center picture). Perhaps increased processivity of the replisome with pol IIIα overproduction increases the probability of the second event (displacement), which would increase the frequency of DExI trapping of the 3’ end created by RNAP, and therefore mutagenesis. Mean ± s.e.m. n = 4 experiments; one-way ANOVA with multiple comparison test. Asterisks from left to right,**p = 0.007, *p = 0.038, *p = 0.021. E. coli strains used: a: SMR22427, SMR24774; b: SMR22427, SMR24774; c: SMR26532, SMR26551; d: SMR23317; f: SMR3070, SMR26532, SMR26553, SMR27374, g: SMR26532, SMR26551, SMR27366.

Extended Data Figure 4 |. Production of the DNA pol IIIα subunit reduces spontaneous 3’-ssDNA-ends by reducing DSBs.

a, Western-blot analyses confirm that doxycycline (dox) induces pol IIIα production, and show its leaky expression with no dox inducer. Representative image of two with similar results. b, Pol IIIα levels have little effect on cell growth rate. The growth curve (optical density, OD = 600 nm) of DExI-GFP- and pol IIIα-producing cells with indicated Dox concentrations. Saturated overnight cultures diluted in M9 glucose medium with different concentrations of doxycycline were grown at 30°C for 1.5 hours, then shifted to 37°C for DExI-GFP induction, mean ± range, n = 2 experiments. c, The corresponding DExI-GFP foci, examined after 4 hours. n = 3 experiments (0 dox), n = 2 experiments (1 ng/ml), n = 2 experiments (10 ng/ml), n = 3 experiments (100 ng/ml), mean ± SEM, *p = 0.017, two-sided Student’s t test. d, Excessive, but not normal levels of DNA pol III replication errors promote 3’-ssDNA-end foci. We manipulated levels of the ε subunit of pol III, encoded by the dnaQ gene, which performs replication “proofreading” (removal) of mis-incorporated bases in replication. Deletion of dnaQ increased DExI foci, indicating that excessive persistence of misincorporated bases can induce DNA damage containing 3’-ssDNA-ends. Mean ± s.e.m. n = 3 experiments, **p = 0.0012, two-sided Student’s t test. e, However, DnaQ overproduction had little effect, implying that, normally, proofreading by DnaQ/ε does not cause most spontaneous 3’-ssDNA-ends; proofreading activity is sufficient to prevent those ends. n = 2 experiments. f, Trans-lesion synthesis (TLS) DNA polymerases are not required for spontaneous 3’-ssDNA-end focus formation. Frequencies of DExI-GFP foci in strains that lack one or more TLS DNA polymerase. n = 2 experiments, mean ± range. g, Western blot analysis of the pol IIIα subunit and Gam amounts in indicated strains. Note that the apparent decrease in the pol IIIα subunit in rpoB2 cells would be expected to increase foci whereas spontaneous foci were decreased in rpoB2 cells (Fig. 4a, f). Representative blots. h, Models for generation of pol IIIα-suppressible, DSB-dependent spontaneous 3’-ssDNA-ends. We hypothesize that (i) high levels of the highly processive DNA pol III reduce replication stalling or stabilize forks until blocks are removed; (ii) conversely, low pol III levels, e.g., by stochastic variation between cells, might allow fork reversal which liberates nascent 3’-ends of the leading strand at a DSB end; (iii) reversed forks can be cleaved by Holliday junction-specific endonucleases, creating a DSB end, (iv) then DSB-end resection produces 3’-ssDNA-ends. (v) DSB ends might also arise from single-strand interruptions. Fork reversal has been associated with transcription termination. However, the pausing/termination-promoted, rpoB2-suppressible component of 3’-ssDNA-end foci was mostly non-overlapping with the DSB (Gam-suppressed) component (Fig. 4a), implying that most pausing/termination-related 3’-ssDNA-ends are not likely to be at reversed forks, which possess DSB ends. By any of many models, including those above, pol IIIα might block or out-compete clamp interactions with other proteins (repair or synthesis) that contribute to spontaneous DSBs that acquire 3’-ssDNA-ends by resection. i, Single deletions of ΔgreA or ΔgreB show wild-type levels of DExI-GFP foci. n = 4 experiments, mean ± s.e.m. For gel source data, see Supplementary Figure 1. E. coli strains used: a: SMR26382; b: SMR26382; c: SMR23252, SMR26382; d: SMR24774, SMR26631; e: SMR26649, SMR26651, SMR26653, SMR26655; f: SMR24774, SMR26371, SMR26373, SMR26374; g: SMR23252, SMR26382, SMR26483, SMR26485, SMR26487, SMR26488, SMR26385, SMR26486; i: SMR24774, SMR25015, SMR25016.

Extended Data Figure 5 |. ThreeSSeq 3’-ssDNA-end genomic profiling with DExI detects dispersed, reproducible, hotspots.

a, Integrated genome viewer (IGV) views of signals from ThreeSSeq with DExI in the indicated strains reveal hotspots throughout the genome. Signals were normalized by Reads Per Kilobase per Million mapped reads (RPKM). For calling of peaks and all quantification, the DExI signal (negative control, background) is subtracted from the signal with DExI-FLAG to obtain the DExI-FLAG-specific reads. b, Two DExI-FLAG experiments are more correlated with each other than with the DExI control. Heatmap of Pearson correlations between the experiments. c, Venn diagrams of peak overlap between two replicates in DExI-FLAG-producing, otherwise wild-type cells. d, Heatmaps of ThreeSSeq signals within ± 300 nt regions around peak summits of 113 TOP-strand and 111 BTM-strand 3’-ssDNA-end hotspot sites. Sites were centered along peak summits and sorted by signal intensity. Signals were normalized by RPKM. e, The distribution of 175 sequence-specific 3’-ssDNA-end sites (shown as black dots) throughout the genome. The leading- and lagging-strand in each replichore are indicated. f, Sequence-specific 3’-ssDNA-end hotspots are distributed randomly throughout the genome. To test the randomness of the 3’-ssDNA-end-hotspot distribution, we sampled 175, 92 and 83 genomic sites, corresponding to the number of all DExI-FLAG-immunoprecipitated sites throughout the genome, the number of TOP-strand sties and the number of BTM-strand sites, randomly 1000 times, and calculated the average distance of neighboring sites. The barplots display the distribution of the sample means, and the arrows indicate the average distances of tested data, including all 175 sites (left, p = 0.10), 92 TOP-strand sites (center, p = 0.48), and 83 BTM-strand sites (right, p = 0.16). The distribution of actual 3’-ssDNA-end hotspots was not different from the 1000 random samples. Strains used: a: SMR24776, SMR24777; b: SMR24776, SMR24777; d: SMR24776.

Extended Data Figure 6 |. The rpoB2 RNA-polymerase read-through mutation reduces spontaneous ThreeSSeq 3’-ssDNA-end-hotspot signals at the terminator-like consensus sequence.

a, Heatmap of ThreeSSeq signal in rpoB2, phage Mu Gam-producing, and dnaE pol IIIα subunit-producing cells across 175 spontaneous sequence-specific 3’-ssDNA-end hotspot sites. Signals are normalized by RPKM and ordered according to signal intensity in the wild-type cells. Signals are displayed along ± 300-nt region around, and centered along, peak summits. Whereas each rpoB2 data set differs significantly from the WT done in parallel with it (b, below), the data from Gam production and pol IIIα production do not differ from the WT. b, Violin plot showing ThreeSSeq signals of 175 3’-ssDNA-end sites in indicated strains. Read counts were normalized by RPKM using the function dba.count in R package DiffBind. One-way ANOVA with multiple comparison test. n = the total cumulative dots from 2 experiments. c, Integrative genomic viewer (IGV) views of ThreeSSeq signals in indicated strains at two 3’-ssDNA-end hotspot sites. d, CRISPR-ablation of consensus sites (C-sites) that flank cycA reduce DExI-induced mutagenesis at cycA. Mean ± s.e.m. n = 4 experiments. One-way ANOVA with multiple comparison test, asterisks from left to right, p < 0.0001 (GraphPad software does not provide exact p value), p = 0.0005, p = 0.001. E. coli strains used: a: SMR24776, SMR26369, SMR26525, SMR26366; SMR24776, SMR26369; d: SMR27380, SMR27381, SMR26532, SMR26551, SMR27384, SMR27385.

Extended Data Figure 7 |. Endogenous 3’-ssDNA-ends at consensus sites are associated with termination of unannotated transcripts.

a, Classification, and distribution of recurrent 3’-ssDNA-end sites relative to annotated genes. 3’-ssDNA-ends are defined as: non-template (blue; located on the sense strand within a gene), intergenic (yellow), and template (mageneta; located on the antisense strand within a gene). Arrows represent open reading frames. b, Subclassification and distribution of intergenic 3’-ssDNA-end sites. They are defined as: 3’ UTR (located on the extended sense strand of an upstream gene); 5’ UTR (located on the extended sense strand of a downstream gene); both and neither. c, Bicyclomycin treatment has little effect on DirectRNA-seq signal at the 3’-ssDNA-end sites, indicating Rho-independent transcript terminations. Average profile of DirectRNA-seq signal in bicyclomycin- (BCM)-treated cells versus untreated cells.

Extended Data Figure 8 |. DNA endonuclease activity of RNAP at the ThreeSSeq consensus sequence, the rpoB2-sensitive component.

a, Transcription complexes were assembled with E. coli core RNAP on nucleic acid scaffolds composed of template DNA (T DNA, gray), non-template DNA (NT DNA, blue), and RNA (red). The RNA, which anneals to the T DNA strand, defines the geometry of the RNAP-DNA complex. In the absence of RNA, RNAP binds in either of two orientations: binding either the blue or gray DNA strand as template. In panel b, only one scaffold orientation is shown: blue strand on top. But in all DNA substrates with neither protein nor RNA bound, RNAP binds either top or bottom strand as template. b, Indicated DNA strands were labeled at the 5’ end with γ³²P-ATP and T4 PNK (label position shown by a yellow star). Assembled complexes were incubated at 37°C for 15, 30, and 60 min; E. coli RfaH, which specifically binds to the GCGGTAGC element in the NT DNA, was present where shown. After quenching, DNA products were analyzed on a 10% urea-acrylamide gel and visualized by phosphorimaging. A dashed line indicates samples from different gels. DNA ladder was made by mixing DNA oligonucleotides of indicated lengths and labeling the mixture with γ³²P-ATP and T4 PNK; the 30-mer was present in 3x molar excess. The red X indicates the position of two apparent cleavage products in the non-template DNA; the resulting products migrate as 26- and 27-mers. Each experiment was performed three times with similar results. Primary data are provided in the Source Data File. Two negative controls indicate consensus-sequence specificity of the cleavage: (i) the 5’- labeled T-DNA, which would show cleavage in the bottom gray strand if cleavage were non-specific and could occur next to the 5’ACCCGCGT sequence there; and (ii) the 5’-labeled NT AA DNA, which carries the 5’AATTTTTT sequence in place of the 5’CCTTTTTT consensus, at the same position, with same strand labeled. Neither of these substrates elicits cleavage 5’ of those sequences as is seen with the active consensus sequence in labeled strand (leftmost, 5’-labeled NT-CC DNA). For gel source data, see Supplementary Figure 1.

Extended Data Figure 9 |. Consensus-sequence sites with little DNA damage display transcript termination at nearby upstream sites.

a, DirectRNA-seq reveals two termination peaks (RNA 3’-ends) near sites of the consensus sequence (CS) at which ThreeSSeq did not detect DNA damage. Average profile of DirectRNA-seq and ThreeSSeq signals around the CSs of 2375 sites on the TOP (left) and 2277 sites on the BTM (right) strand. DirectRNA-seq signals that mapped to TOP (green) and BTM (yellow) strands, along -300-nt to +100-nt regions, around the CSs, are superimposed on ThreeSSeq signals (TOP: orange; BTM: purple). b, Examples of integrative genomic viewer (IGV) views of DirectRNA-seq signals (green) and ThreeSSeq signals (orange) at two sites that have the CS but are not detected by ThreeSSeq and, c, at two sites at which 3’-ssDNA-ends are detected. d, Model for DNA-damage prevention at the “cold” sites of the CS, following the idea from Zhou and Martin on the effects of multiple RNAPs, or the lack thereof, on RNAP dynamics. The model can explain our data that a closely upstream RNAP termination site may prevent the ThreeSSeq-detectable DNA damage. An upstream RNAP might occlude the CS downstream, or displace an RNAP at the downstream CS, so that the DNA cannot be cleaved when the transcript ends. Alternatively (not shown), more termination at the upstream terminator leaves fewer RNAPs and transcripts to end at the downstream site.

Figure 1 |. DExI-fluorescent-protein fusions detect 3’-ssDNA-ends in E. coli.

a, Illustration of DExI-GFP binding to a 3’-ssDNA-end. b, Functional DExI fusions restore ultra-violet light (UV) resistance in the ΔrecB ΔsbcC double mutant, per^,, ΔΔ, ΔrecB ΔsbcC; means ± range, n = 2 experiments. c, DExI-GFP binds 3’-ssDNA-ends preferentially. Electrophoretic mobility shift assays (EMSA) of ³²P-radiolabeled DNA substrates (asterisk-marked), concentrations of DExI-GFP indicated. Representative image of three, with similar results. d, Strategy for varying DSB numbers in living cells with I-SceI endonuclease and engineered I-SceI cut sites (I-sites). e, Representative images of data quantified in f: spontaneous DExI-GFP foci (left) or foci after I-SceI cut. Scale bar, 2 μm. f, Quantification of spontaneous and post-I-SceI DExI-GFP foci; means ± s.e.m. n = 3 experiments, p = 0.014. g, Frequencies of I-SceI-induced DExI-GFP foci (spontaneous foci subtracted); means ± s.e.m. n = 3 experiments, p = 0.045. h, Frequencies of phage Mu GamGFP (DSB) foci; means ± range n = 2 experiments. i, Spontaneous DExI-mCherry foci co-localized with SSB-Ypet, implying DExI binding in ssDNA. Means ± s.e.m. n = 3 experiments, p = 0.006 and 0.026, respectively.. For f, g, i, two-sided paired t test. j, Representative images of co-localization of DExI-mCherry foci with SSB-YPet foci (≤ 50kb apart). Scale bar, 1 μm. k, Percentage of DExI-mCherry foci co-localized with SSB-YPet foci; means ± range, n = 2 experiments. For gel source data, see Supplementary Figure 1.

Figure 2 |. DExI labels DNA damage in human cancer cells.

a, Representative images of EmGFP-DExI and γH2AX accumulation in laser-induced DNA-damage tracts in human U2OS cancer cells. b, Percentage of the cells that display co-localized EmGFP-DExI- and γH2AX-positive laser lines. Means ± s.e.m., n = 3 experiments. c, EmGFP-DExI foci are induced and localized to γH2AX-positive DNA-damage sites (DSBs) following ionizing-radiation (IR) exposure. Cells were analyzed by immunofluorescence with the indicated labels at 6 hr post-IR. Representative images of EmGFP-DExI and γH2AX foci in IR or not-treated (NT) cells, and their d, quantification of co-localization in cells with >10 foci. e, EmGFP-DExI foci label hydroxyurea-(HU)-induced lesions. Representative images of EmGFP-DExI and γH2AX foci in NT and HU-treated cells. f, Quantification of e as in d. For e and f, mean ± range, n = 2 experiments. Scale bars, 10 μm.

Figure 3 |. Most spontaneous DNA damage with 3’-single-stranded ends occurs during chromosome replication.

a, Blocking E. coli replication initiation reduced DExI-mCherry foci at 42°C, at which dnaATS cells fail to initiate replication, representative images; b, focus frequencies normalized by DNA content (DAPI fluorescence intensity). Mean ± s.e.m. n = 3 experiments, one sided Student’s t test: p = 0.046. c, Distribution of DExI-GFP focus lifespans measured by time-lapse microfluidic imaging; n = 30. d, Representative images of short- and long-lived DExI-GFP foci. Extended Data Fig. 3a shows images spanning entire cell division cycles. e, Generation-dependence of spontaneous DExI-GFP foci; n = 3 microfluidic microcolonies, mean ± s.e.m. f, Diagram of “replication-runoff” assay. Replication initiation (and cell division) are inhibited by treatment with rifampicin and cephalexin, while the ongoing cycles of replication are completed. DExI inhibits DNA replication elongation leading to—g, reduced DNA from incomplete chromosome replication. Flow cytometry histograms of DNA content/DAPI fluorescence intensity of wild-type and DExI-producing cells after runofft. h, Whole-genome-sequencing (WGS) shows DExI reduction of DNA reads (failure to complete replication runoff) throughout the genome; origin, 0 Mb. Median-normalized WGS reads in indicated strains. i, DExI-stabilization of spontaneous 3’-ssDNA ends increased mutation rates. Dependence on GreA GreB transcription-proofreading factors; reduction by pausing- and termination-impaired rpoB2 mutant RNAP. Additional controls in Extended Data Figure 3f. Mean ± s.e.m. n = 9 experiments; one-way ANOVA with multiple comparison test. Asterisks from left to right, p = 0.0001, p = 0.0002, p = 0.0029.

Figure 4 |. DNA pol III-suppressible DSBs and a separate transcription-related process underlie spontaneous 3’-ssDNA-end foci.

a, Production of phage Mu Gam or the replicative DNA polymerase (pol) IIIα subunit, or use of the RNAP rpoB2 (terminator read-through) mutant reduces spontaneous 3’-ssDNA-end foci. Triple, Gam rpoB2 PN25tetOdnaE; mean ± s.e.m., n = 3 experiments; one-way ANOVA with multiple comparison test. Asterisks from left to right, p = 0.035, p = 0.027, p = 0.007, p = 0.0007, p = 0.0001. b, GamGFP foci in wild-type cells, representative image. Scale bar, 5 μm. c, Pol IIIα subunit production reduced spontaneous GamGFP foci (DSBs), indicating that Pol IIIα inhibits the spontaneous DSB-generating pathway. Mean ± s.e.m., n = 3 experiments; two-sided Student’s t test, p = 0.048. d, R-loop removal by RNase H1 protects against 3’-ssDNA-ends, but e, overproduction of RNase H1 has little effect on spontaneous DExI-GFP focus numbers. For d and e, mean ± range, n = 2 experiments. f, Percentages of different causes of spontaneous 3’-ssDNA-ends, essentially all of which are also DNA-replication-dependent (Fig. 3a,b). g, Transcription-proofreading factors GreA and/or GreB are required for formation of some spontaneous 3’-ssDNA-end DExI-GFP foci. Mean ± s.e.m. n = 3 experiments; one-way ANOVA with multiple comparison test. Asterisks from left to right, p = 0.016, p = 0.016.

Figure 5 |. A transcription-associated consensus sequence.

a, ThreeSSeq genomic profiling. Strand-specific sequencing libraries from immunoprecipitated fragments. 3’-ssDNA-end hotspots are predicted to form asymmetrical peaks with steep sides to the right for top-strand peaks (TOP, orange), and left for bottom-strand peaks (BTM, purple). b, Heatmaps of signals in ± 300 nt regions around peak summits show predicted asymmetry in 175 sites, centered along peak summits. Normalized by Reads Per Kilobase per Million mapped reads (RPKM). c, The motif 5’-CCTTTTTT is enriched among and abuts recurrent 3’-ssDNA-ends. Average ThreeSSeq signal within ± 300 nt regions around the first C of the consensus sequence for 92 TOP-strand (left) and 83 BTM-strand (right) sites. d, Position of 3’-ssDNA-ends relative to RNA 3’-ends detected by by DirectRNA-seq. The 3’-DNA-ends are of any sequence but occur immediately 5’ to the consensus sequence. RNA transcripts of the same polarity detected by DirectRNA-seq terminate 3’ to the consensus sequence. e, Average DirectRNA-seq signal (gray, left; yellow, right) overlaid with ThreeSSeq signal (400 nt around the first C of the consensus sequence). f, rpoB2 mutation reduced transcript termination at a consensus site. The qRT-PCR fragments are 233 bp 3’ of the upstream primers, and 290 bp 5’ of the downstream primers. n = 6 experiments, mean ± s.e.m.; two-sided Student’s t test. Asterisks from left to right, p = 0.039, p = 0.034. g, STop-And-Break DNA (STABD) hypothesis for 3’-ssDNA-end formation. We propose that the consensus sequence RNA pauses transcription, sometimes causing termination and RNAP clamp opening, which allows the DNA non-template strand to anneal to the template in the RNAP active site. RNAP proofreading nuclease might then cleave the non-template DNA strand. DNA replication is suggested to displace the 3’-DNA-end making it single-stranded, in agreement with our finding that the spontaneous ends are replication dependent (Fig. 3a, b).