Ultraviolet-induced RNA:DNA hybrids interfere with chromosomal DNA synthesis

Abstract Ultraviolet (UV) induces pyrimidine dimers (PDs) in DNA and replication-dependent fragmentation in chromosomes. The rnhAB mutants in Escherichia coli, accumulating R-loops and single DNA-rNs, are generally resistant to DNA damage, but are surprisingly UV-sensitive, even though they remove PDs normally, suggesting irreparable chromosome lesions. We show here that the RNase H defect does not cause additional chromosome fragmentation after UV, but inhibits DNA synthesis after replication restart. Genetic analysis implies formation of R-loop-anchored transcription elongation complexes (R-loop-aTECs) in UV-irradiated rnhAB mutants, predicting that their chromosomal DNA will accumulate: (i) RNA:DNA hybrids; (ii) a few slow-to-remove PDs. We confirm both features and also find that both, surprisingly, depend on replication restart. Finally, enriching for the UV-induced RNA:DNA hybrids in the rnhAB uvrA mutants also co-enriches for PDs, showing their co-residence in the same structures. We propose that PD-triggered R-loop-aTECs block head-on replication in RNase H-deficient mutants.


INTRODUCTION
Hard ultraviolet (UVC, henceforth 'UV') is perhaps the best-studied DNA-damaging treatment, with the spectrum of UV-induced DNA lesions predominantly comprising cyclobutane pyrimidine dimers and 6-4 photoproducts (1). In UV-irradiated Escherichia coli cells incubated in the dark to avoid photoreactivation, both types of pyrimidine dimers (PDs) are exclusively removed by the UvrABC-promoted nucleotide excision-repair (NER) (2). Functionally, NER is further subdivided into global genomic repair (GGR, operating anywhere in the genome) and transcriptioncoupled repair (TCR, preferentially removing lesions from actively-transcribed genes) (3). TCR is triggered by RNA polymerase stalling at a PD in the template strand and is initiated by Mfd translocase, which displaces RNA polymerase and recruits NER enzymes to remove the transcription-stalling PD (4,5).
Rapid removal of UV-induced PDs from DNA by NER is even more important for DNA replication, since after sublethal UV exposure of growing E. coli, DNA synthesis cannot restart for about 20 min (6,7), while transcription and translation are barely affected (8). The consequences of replication forks encounters with PDs and with excision intermediates of PD removal are much more serious and lead to formation of either blocked singlestrand gaps (so-called 'daughter-strand gaps') (9,10) or disintegrated replication forks (10,11). In E. coli, blocked ss-gaps are either closed by the RecFOR recombinational repair pathway, or by its backup via translesion DNA synthesis, mostly by Pol V across PDs, but also by Pol IV (10). Disintegrated replication forks are reassembled by the RecBC-pathway of recombinational repair (10); its absence in the recBC mutants reveals that UV induces high levels of chromosome fragmentation (12).
Since everything seems to be known about how UV damages DNA and what chromosomal consequences of this damage are, at least in E. coli,--it comes as a surprise to find a strong UV-sensitivity in mutants that have all the known UV-repair pathways intact, as this suggests existence of yet-to-be-characterized UV-induced chromosome lesions. We have serendipitously found an unexpectedly strong UV-sensitivity in the rnhA rnhB double mutant of E. coli (henceforth rnhAB), deficient in both RNase H enzymes, ribonucleases attacking the RNA moiety of the RNA:DNA hybrids (RDHs). The RNase Hdeficient mutants accumulate single rNs and R-loops in their genomic DNA, with dramatic consequences. RNase H-deficient mutants are inviable in higher eukaryotes (13,14) and show various defects in lower eukaryotes (15,16) and bacteria (17,18). Escherichia coli has two RNase H enzymes, with distinct specificities ( Figure 1A, top). RNase HI encoded by the rnhA gene, removes R-loops and ≥4 nt rN-runs embedded in DNA (R-tracts) (19,20). RNase HII, encoded by the rnhB gene, also called the 'junction ribonuclease', incises the 5 RNA-DNA3 junction within dsDNA, leaving a single rN on the 5 cleaved end and thus initiating removal of single DNA-rNs, and also R-tracts ( Figure 1A) (21,22). Unlike analogous eukaryotic RNase H2 enzymes, the prokaryotic RNase HII shows no activity against RDHs lacking RNA-DNA junctions (unless metallated with Mn 2+ instead of Mg 2+ ) (23,24). Therefore, in contrast to the higher eukaryotes, E. coli rnhB mutants, although accumulating measurable density of single rNs in the genome, show no growth defects or other gross phenotypes, indicating that single rNs at this density do not interfere with DNA replication (18). In contrast, bacteria deficient in RNase HI (rnhA mutants in E. coli) grow slower and suffer from unscheduled initiation of the chromosome replication and synthetic lethality in combination with various defects in the DNA metabolism (25)(26)(27). This implicates RNA:DNA hybrids in general and (protein-free) R-loops in particular as impediments to replication,--an idea especially popular in eukaryotes (28,29). However, protein-free R-loops are rather unstable and should disassemble when a replication fork approaches with the accompanying wave of positive supercoiling; in contrast, Rloops associated with transcribing RNA-polymerases are argued to represent a considerable challenge (30). The lethality of unresolved conflict of replication with headon transcription was recently demonstrated (31,32) and was proposed to be due to formation of R-loop-anchored transcription elongation complexes (R-loop-aTECs), which can be disassembled only by RNase H enzymes, becoming irreparable chromosome lesions in their absence (30).
The relatively modest phenotypes of the rnhA single mutants in E. coli are remarkably exacerbated by the rnhB defect: the resulting completely RNase H-deficient rnhAB double mutants grow extremely slowly, produce filamentous cells, are highly induced for SOS-response and, because of high levels of chromosome fragmentation, depend on recombinational repair (18). To explain severe chromosomal problems of the rnhAB double mutants, we argued that R-loops in them are converted into a common substrate for both RNase HI and RNase HII,--the so-called R-tracts ( Figure 1A, top), which are then transformed, via R-gaps, into double-strand breaks (18). However, we could not detect the expected R-tracts in plasmid DNA from rnhAB mutants,--and so we sought conditions that would kill the rnhAB mutants, reasoning that under these lethal conditions, the elusive R-tracts could be amplified and become detectable.
In some E. coli strains expressing a 'steric gate' mutant of the translesion DNA polymerase Pol V with the increased capacity for DNA-ribonucleotide incorporation, NER was implicated in removal of single DNA-rNs (33). Moreover, the PD-sensor of NER, the UvrA protein, was also shown to recognize rNs in dsDNA substrates in vitro (33), although this finding was later disputed (34). In support of the idea that NER removes misincorporated ribonucleotides in the rnhAB mutants, we did observe an additional growth defect in the rnhAB uvrA triple mutant, although the density of single DNA-rNs in this mutant remains the same as in its rnhAB (UvrA+) parent (18). It is while verifying the extreme UV-sensitivity of the uvrA mutants, that we unexpectedly found the high UV sensitivity of the rnhAB (NER+) mutants ( Figure 1B and Supplementary Figure S1A).
Our initial reasoning about the UV sensitivity of rnhAB NER+ mutant envisaged a combination of pre-existing R-loops or R-tracts with induced PDs to produce lethal 'UV-induced R-lesions' (UVRLs) ( Figure 1A, bottom). However, after testing various models of UVRL formation, we conclude that UVRLs represent R-loop-aTECs, formed as a result of transcribing RNA polymerases stalling at PDs, which are subsequently aggravated by restart of DNA synthesis.

Bacterial strains
Escherichia coli strains (all K-12) are described in Supplementary Table S1, while plasmids are described in Supplementary Table S2. Strain construction was by P1 transduction (35) or by deletion-replacement method with the following removal of the antibiotic resistance by pCP20 (36). Deletions-replacements were confirmed by PCR. The recA, recBCD, recF and uvrA mutants were confirmed by their characteristic UV-sensitivities, while the dnaC mutants were verified as unable to grow at 42 • C.
Primers used for making deletion-replacements, PCR amplification, sequencing and verification of indicated chromosomal loci are listed in the Supplementary Table S3 (Primers).

Media and growth conditions
Cells were grown in LB broth [10 g tryptone, 5 g yeast extract, 5 g NaCl per liter (pH 7.2), with NaOH] or on LB plates (15 g agar per liter of LB broth). The growth temperature was 28 • C unless otherwise indicated in the description of experiments. When screening for mutations linked to antibiotic-resistant genes or when the cells were carrying plasmids, the media were supplemented with the required antibiotic: 100 g/ml ampicillin, 50 g/ml kanamycin, 10 g/ml tetracycline or 10 g/ml chloramphenicol. 1 mM IPTG was used for UvrA overproduction from pSRK10-1.

Viability tests
Overnight cultures of the tested strains were diluted 100fold in the morning and grown in fresh LB to OD 600 ∼0.2, after which the treatment was delivered.
UV treatment and survival. Five 10-fold serial dilutions of six cultures were made in sterile 1% NaCl solution and spotted by 10 l in one row for each dilution on several square Petri dishes with LB agar. Spots were dried, and the plates were exposed to UVC-light (254 nm) in Hoefer UVC 500 UV cross-linker. All manipulations were performed under yellow light (F15T8-GO lamp, General Electric) to avoid photoreactivation. The time to deliver a particular UV dose was calculated from the measurements of UV irradiation with UVC Digital Light Meter (model UV512C, General). Plates were developed overnight in the dark at 28 • C, and the still pin-prick colonies were counted under a stereomicroscope. The survival was determined as the ratio of a culture titer after a particular UV exposure to the titer of the untreated culture.
Rifampicin pretreatment. One half of the culture was treated with 100 g/ml of rifampicin for 5 min, the other half was left untreated. To remove rifampicin, cells were pelleted and resuspended in fresh LB twice, followed by UV exposure of the culture serial dilutions on LB plates.
DNA damaging treatments. A specific amount of the agent was added directly to exponentially growing cultures at 28 • C. The treatments were: 10 mM MMS for 30 min, 10 g/ml mitomycin C for 30 min, 10 mM hydrogen peroxide for 15 min, 30 g/ml nalidixic acid at 37 • C (two time points). Aliquots of cultures were taken, serially diluted in 1% NaCl and spotted on LB agar to enumerate survivors. The plates were incubated overnight at 28 • C. The survival was determined as the ratio of the colony forming units at a specific treatment time or agent concentration to the colony forming units before the treatment (zero time) or without the agent.
Processing cultures to determine the density of DNAribonucleotides, thymine dimers or RNA/DNA hybrids (RDHs) Strains, including those harboring appropriate plasmids, were grown in 20 ml of LB to OD 600 of 0.4, collected by centrifugation, resuspended in 4 ml of 1% NaCl. 1 ml of cells was taken to purify DNA as no UV control, the other 3 ml were mixed with 12 ml of 1% NaCl and transferred to an open sterile glass tray (14 cm × 23 cm), placed under the GE germicidal lamp emitting 254-nm UV-light, and exposed to UV-light with slow shaking of the tray on a platform. After irradiation, the cells were diluted (1:1) with 2× LB without NaCl and transferred to a sterile flask for post-irradiation recovery at 37 • C with vigorous shaking. Seven-milliliter samples were removed at specific times, placed on ice, then collected by centrifugation at 7000 rpm for 5 min and processed according to the 'total genomic DNA isolation' protocol. Samples designated for PD density analysis by plasmid nicking were first mixed with four volumes of icecold 0.01M KCN, 10% pyridine stop solution (37) and processed according to the plasmid isolation protocol.
Plasmid DNA isolation. Plasmid DNA was extracted according to the small-scale alkaline lysis plasmid isolation protocol with all steps carried out on ice (18) followed by LiCl purification step (38).
'Total genomic ' DNA isolation. Cell lysates were prepared according to Brij lysis procedure (39). This method was crucial to obtain highly reproducible results with the S9.6 antibodies. In particular: after UV exposure and post-irradiation recovery (described above), cells were pelleted and resuspended in 0.25 ml of ice-cold solution of 30% sucrose, 0.05M Tris-HCl, pH 8.0. Freshly prepared lysozyme, 0.05 ml of 5 mg/ml in 0.25 M Tris-HCl, pH8.0, was added to the cell suspension, which was kept on ice for 5 min with gentle mixing. 0.1 ml of 0.25 M EDTA pH 8.0 was added for another 5 min on ice with occasional gentle swirling. To lyse the cells, 0.4 ml of detergent mixture (1% Brij58, 0.4% sodium deoxycholate, 0.0625 M EDTA and 0.05M Tris-HCl, pH 8.0) and RNase A to the final concentration 20 g/ml were added and gently mixed. The samples were kept on ice for 30-60 min followed by three organic extractions: 0.8 ml phenol with 40 l chloroform, followed by 0.8 ml phenol/chloroform (1:1), followed by 0.8 ml chloroform. The final aqueous phase was transferred into fresh tube and precipitated with salt and ethanol.

Quantification of ribonucleotides and PDs by the plasmid relaxation method
Density of ribonucleotides in DNA was calculated after plasmid DNA was treated with RNase HII (NEB) or RNase HI (Life Technologies) as previously described (18).
PD density was calculated in plasmid DNA treated with T4 endonuclease V (NEB). In particular, 100-200 ng of purified plasmid DNA was treated in 20 l of 1× T4 PDG reaction buffer (25 mM sodium phosphate (pH 7.2), 100 mM NaCl, 1 mM EDTA, 1 mM dithiothreitol) containing either no enzyme or 3 units of T4 PDG (NEB) at 37 • C for 30 min. After enzymatic reactions the plasmid intermediates were analyzed in 1% agarose gels followed by Southern hybridization with a radioactive probe made from the appropriate plasmid DNA by random priming (18). The signals from supercoiled and relaxed plasmid species were quantified by Phosporimager (FujiFilm FLA-3000, Fuji).
Average density of nicks in a plasmid DNA was derived from the remaining supercoiled DNA representing zero class of the Poisson distribution after the enzymatic treatment. Our calculations steps were (40): (i) the radioactivity signal in the supercoiled monomer band was divided by the sum of the radioactivity in the supercoiled and relaxed monomer bands in the lane to determine the fraction of the supercoiled species, (ii) average number of nicks was calculated according to the formula -ln(F treated /F untreated ), where F treated is a fraction of supercoiled band from the lane with the enzyme treatment, F untreated is a fraction of supercoiled band from the lane of the plasmid treated with buffer-only to take into account the background nicking. The density of nicks in a plasmid was calculated by dividing the double-stranded-plasmid length (in nucleotides) per average number of nicks; then the E. coli genome (9.2 × 10 6 nt) was divided by the density of nicks to get the number of PDs per genome.

Quantification of RDHs and PDs in the chromosomal DNA
Cells were grown and UV-irradiated as described above. DNA was extracted by the total genomic DNA isolation method.
For RDH signal detection: 500 ng of the DNA was split in half and run on two gels (0.8% agarose in TAE buffer) under identical conditions. One gel was stained with ethidium bromide, bands were visualized and then the gel was treated for Southern analysis as before (18). The second gel was soaked in 0.5× TBE buffer, and DNA was transferred onto Amersham Hybond-N + membrane (GE Healthcare) by electric transfer using a Trans-blot cell (Bio-Rad) for 16 h in 0.5× TBE buffer, 20 V with cooling. After electric transfer, the DNA was UV-crosslinked to the membrane, and standard western blotting procedure was applied. Briefly, blocking of the membrane was done for 1 h at room temperature with rotary platform shaker with 5% non-fat milk dissolved in TTBS buffer (0.02 M Tris-HCl, pH 7.5, 0.9% NaCl, 0.05% v/v Tween 20), then the solution was exchanged for the buffer with the primary antibodies (Anti-DNA-RNA Hybrid, clone S9.6 monoclonal antibodies (Millipore)) diluted 1:5000 in TTBS, 5% milk buffer and the membrane was probed at room temperature for 1 h, followed by three washes 10 min each with TTBS buffer. Secondary antibodies (Anti-Mouse IgGperoxidase antibody, Sigma-Aldrich) were diluted to 1:10 000 in TTBS, 5% milk buffer and applied to the membrane at room temperature for 1 h, followed by three washes 10 min each with TTBS buffer. Signal was developed with Supersignal West Femto Maximum Sensitivity Substrate kit (Thermo Scientific). Imaging was done with CCD camera of the Bio-RAD ChemiDoc XRS system and Bio-Rad Quantity One software. Image analysis and quantification of the TIFF files was done with ImageQuant TL program (GE HealthCare Life Sciences).
For thymine dimer signal detection, the same procedure as for the RDH detection was followed, with some modifications. Briefly, 500 ng/lane genomic DNA samples were analyzed by gel electrophoresis. The gels with the DNA samples prepared to be analyzed by Western blotting were treated as for the Southern analysis, since the Anti-Thymine Dimer antibodies (KTM53, Kamiya Biomedical MC-062) bind single stranded DNA. Vacuum transfer of the DNA to the Amersham Hybond-N + membrane (GE Healthcare) was carried out, the membrane was baked for 2 h at 80 • C (instead of UV-crosslinking) and probed with Anti-Thymine Dimer antibodies diluted 1:5000 as primary antibodies.

Calculations
In each experiment, Western signal of a particular strain was divided by the corresponding Southern genomic DNA signal to obtain either RDH density or PD density value. The RDH density for each strain at a particular post-UV time point was normalized to the RDH density value of the rnhAB log culture within the same experiment. Similarly, the PD density value for each post-UV time point was normalized to the PD density value in the same strain measured at 5 min post-UV (which was taken for 100%).

Measurement of the DNA synthesis rate
Cells from 2 ml log cultures grown at 30 • C to OD 600 = 0.35 were centrifuged, washed once with 2 ml 1% NaCl and resuspended in 1.5 ml 1% NaCl, 0.01% Triton X100 to yield OD 600 = 0.4. For zero time point: 300 l of the sample was mixed with 300 l of 2× LB without NaCl and the rate of DNA synthesis was measured as below. The rest of the sample (1.2 ml) was transferred to a sterile Petri plate and spread in a thin layer. UV irradiation was performed with Hoefer UVC 500 UV cross-linker. Following irradiation, 1 ml was transferred to a tube with 1 ml of 2× LB without NaCl. Tubes were shaken for up to 3 h at 37 • C in the dark, and 200 l aliquots were removed for analysis at the indicated times.
To measure the rate, 200 l of the sample was mixed with prewarmed at 37 • C 200 l LB containing 1 Ci of [methyl-3 H] thymidine (MP Biomedicals) and 0.4 g of thymine.
The reaction was carried out for 3 min and then stopped by addition of 5 ml of ice-cold 5% TCA. The tubes were kept on ice throughout the experiment. Samples were processed as described (12). During post-irradiation recovery, cultures were diluted 2-fold at 50, 90 and 150 min, if needed to keep cultures in logarithmic growth. Dilutions were taken into account after the amount of 3 H was determined by scintillation counting.

Chromosome fragmentation by pulsed-field gels
Overnight cultures were diluted 100-fold in LB supplemented with 5 Ci/ml of 32 P-orthophosphoric acid and grown to the OD 600 of 0.35 at 28 • C. At this point, the medium was changed to 1% NaCl, 0.01% Triton X100 for UV irradiation in Petri plates, as described above for the procedure to measure DNA synthesis rate. Following irradiation, 1 ml was transferred to a tube with 1 ml of 2× LB without NaCl supplemented with the 32 P-orthophosphoric acid and shaken at 37 • C for 2 h in the dark. No-UV control samples were treated the same way, but after 1 h recovery time at 37 • C the cultures were diluted 4-fold in fresh LB with the 32 P-orthophosphoric acid, to keep them growing exponentially. The chromosomal DNA preparation in agarose plugs, treatments and conditions for pulse-field gel electrophoresis, as well as quantification of the chromosomal breakage were done as before (41).

RNase sensitivity tests of the RDH
Treatments of 0.5 g of the total genomic DNA with RNase HI (Takara) or RNaseHII (NEB) were performed as described (18). It was critical to run the reaction samples on the gel followed up by electric transfer to separate RNase HI from the substrate. The enzyme tightly binds to the substrate (even at 0 • C in water) and completely blocks immunodetection if a reaction mix is directly used in the dot-blot procedure.
RNaseA treatment. Treatment of 0.5 g of total genomic DNA was in 20 l of the reaction buffer [20 mM Tris-HCl (pH 8.0), 40 mM KCl, 8 mM MgCl 2 , 1 mM dithiothreitol] containing either no enzyme or 5 g/ml of RNase A at 37 • C for 15 min. If inhibition of RNaseA activity toward dsRNA was desired, the reaction mixture was made 0.5 M for NaCl.

DNA/RNA hybrid immunoprecipitation assay (DRIP)
Growth of the cultures and UV-irradiation were performed as described above in the section 'Processing cultures to determine ribonucleotide density, thymine dimer density and RNA/DNA hybrids after UV irradiation'. Total nucleic acids were purified as described in the section: 'Total genomic DNA isolation'.
For each DRIP experiment, 8-10 g of DNA were digested with various enzymes: BamHI + EcoRI, or HaeII, or HaeIII in the 1xCutSmart Buffer (NEB) at 37 • C, followed by extractions with phenol, phenol/chloroform, chloroform and ethanol precipitation.
After digestion with the restriction enzymes, the control DNA samples were treated with RNase HI (Takara). To verify completion of digestion and RNase HI treatment, 0.3-0.5 g of the digested samples were run on 0.7% agarose gel.
The DRIP procedure was as described (31) with small modifications. The DRIP mixture was assembled in 0.4 ml volume by mixing 0.04 ml of 10× IP Binding buffer (200 mM Tris pH 7.5, 9% (1.55 M) NaCl, 0.5% Tween 20) and DNA sample in TE (10 mM Tris-HCl, pH 8.0, 1 mM EDTA) with 1/20 of the DNA sample removed beforehand to use as the INPUT control). S9.6 monoclonal antibodies (Millipore) were added (1 l antibodies per 1 g DNA) to the DRIP mixture and incubated at 4 • C overnight with gentle rotation. Next morning, Protein A-Sepharose beads (BioVision) were prepared by washing them in 1× IP binding buffer (0.5 ml of buffer per 0.1 ml of the beads slurry) three times for 5 min each with gentle rotation at room temperature and precipitation at 2000 rpm for 1 min. 0.04 ml of the prepared ProteinA-Sepharose beads were added to 0.4 ml of DRIP mixture, and the suspension was incubated for 4 h at 4 • C with gentle rotation. The DRIP-beads mixture was pelleted at 2000 rpm for 1 min. The supernatant was removed, 0.1 ml was twice ethanol/salt precipitated, resuspended in 0.12 ml CHIP/Proteinase buffer [10 mM Tris pH 8.0, 1 mM EDTA, 0.67% SDS, 420 g/ml Proteinase K (Roshe)] and processed in parallel to the DRIP fractions (below) to generate the S9.6-DNA-unbound (Eluate) fraction. The pelleted DRIP-beads (the DRIP fraction) were washed in 1 ml of 1× IP Binding Buffer three times with the additional wash in 1 ml TE buffer. After washes, 0.12 ml of the CHIP/Proteinase buffer were added to the DRIPbeads and incubated at 55 • C for 45 min with shaking the tubes in a Eppendorf Thermomixer at 800 rpm. Beads were collected by centrifugation at 7000 rpm for 1 min, and the supernatant (the DRIP fraction) was transferred to a new tube. The DRIP and Eluate DNA was phenol-chloroform extracted, mixed with glycogen (Thermo Scientific) at 0.2 g/l final concentration, made 0.5 M for NaCl, 2.5 volumes of ethanol were added, and after mixing incubated for 2-18 h at −20 • C, then centrifuged at room temperature for 30 min at 16 000 × g. DNA pellets were dissolved in 30 l TE buffer and analyzed by dot blotting.
All DNA samples were split into two fractions: 5 l for RDH signal and 25 l for PD and DNA signal analysis. For RDH western analysis, 5 l of DNA after being mixed with 45 l of TE buffer, were applied directly to the membrane (Hybond N+, GE). For PD and DNA blotting, 25 l sample was denatured for 15 min in 0.1M NaOH at 37 • C, then neutralized with 0.2 M Tris-HCl pH 8.0, followed by addition of TE (25 l DNA, 25 l 0.2 M NaOH, 50 l 0.4 M Tris-HCl, 100 l TE). All DNA samples were applied to the membrane using Hybrid-Dot Manifold (Life Technologies, Inc). For RDH signal analysis, the membrane was UVcrosslinked, while the membrane for PD and DNA signal determination was baked overnight at 80 • C. Detection of the RDH and PD signals was done by western analysis with the corresponding antibodies, as described above in 'Quantification of the RDHs and PDs in the chromosomal DNA'.
Nucleic Acids Research, 2021, Vol. 49, No. 7 3893 After western analysis, the membrane with PD signal was washed off for 30 min at 37 • C with 2% SDS in 50 mM Tris-HCl, and then the standard Southern hybridization protocol was performed with the total genomic DNA as the radioactive probe (generated by random priming). DNA images were generated with phosphor-imaging screens using Typhoon FLA 7000 (GE Healthcare) and calculated with ImageQuant TL program (GE HealthCare Life Sciences).
Enrichment calculations. RDH densities in three fractions (Input, DRIP, Eluate) were calculated by dividing the Western RDH signals to the corresponding Southern genomic DNA signals. RDH-enrichment was calculated as the ratio of the RDH densities in the DRIP or Eluate fractions to the RDH density of the corresponding INPUT fraction. The same procedure was applied to calculations of the PD-enrichment from PD densities.

Co-enrichment of RDHs and PDs
Cells growth, UV-irradiation and DNA extraction was done as described in 'Quantification of RDHs and PDs in the chromosomal DNA'. 10 g of DNA was digested with HaeII in 200 l reaction in the 1xCutSmart Buffer (NEB) at 37 • C, followed by extractions with phenol, phenol/chloroform, chloroform and ethanol precipitation. DNA samples were split in half and run on two agarose gels. Treatment of the gels, processing of the membranes and western analysis were as described in 'Quantification of RDHs and PDs in the chromosomal DNA', the only difference being that DNA from both gels was transferred to the membranes by electric transfer. Western signal imaging was done with iBright CL1000 Imaging System (Invitrogen). Image analysis and quantification of the TIFF files was done with ImageQuant TL program (GE HealthCare Life Sciences).
Calculation of co-enrichment profiles. The profile of the lane is divided into 16 equal fractions from the well to the 0.2-0.5 kb bottom of the lane. The density values for RDH and PD are calculated for each fraction via dividing western signal by Southern (DNA) signal. The RDH density for 30 min point is then normalized to the density of RDH from the corresponding fractions of the log sample (normalized RDH density = RDH 30 min density/RDH log density). Similarly, the PD density of the 30 min point is normalized to the density of PD from the 5 min post-UV sample (normalized PD density = 30 min PD density/5 min PD density). This normalization step permits meaningful comparison of data points from different experiments. The co-enrichment value for a particular fraction within the lane is calculated by multiplying the normalized PD and RDH densities of this fraction.

UV-sensitivity of rnhAB mutants is not due to defects in global NER or recombinational repair
Since it was proposed that UvrA excises single DNA-rNs in the rnhAB mutants (33,42), the UV-sensitivity of the rnhAB mutants ( Figure 1B and Supplementary Figure S1A) could be due to UvrA protein titration by these DNA-rNs, making rnhAB mutants partial uvrA phenocopies. Yet, overexpression of the uvrA+ gene from a plasmid, that complemented UV-sensitivity of the uvrA control strain, failed to improve UV sensitivity of the rnhAB mutant ( Figure 1C). We also assessed the global NER capacity of the rnhAB mutant by isolating plasmid DNA from cells irradiated with 36 J/m 2 UV and treating it in vitro with T4 Endonuclease V (T4 PD glycosylase), that has also AP-lyase activity (43), nicking DNA at PD sites and thus relaxing supercoiled plasmid ( Figure 1D), facilitating determination of the density of PDs in the plasmid DNA. At 5 min post-UV, this density translates into ∼1500 PDs per E. coli genome (∼42 PD per 1 J/m 2 ) ( Figure 1E). In the uvrA mutant, we observed no PD removal within the first hour, confirming its NER-deficiency (the 2 h partial PD 'disappearance' likely reflects replication of PDfree plasmid). In contrast, in all NER-proficient strains independently of their rnhAB status, we observed the same rate of PD removal from the plasmid DNA: about half of PDs was removed by 20 min, and almost all of them were removed by 60 min after UV ( Figure 1E), demonstrating normal global NER capacity in the rnhAB mutants.
Since rapidly growing rnhAB double mutants form DSBs, induce substantial SOS response and rely on recombinational repair to reassemble fragmented chromosome (18), rnhAB mutants could be sensitive to UV because of an overwhelmed recombinational repair. If so, the rnhAB mutants should be also sensitive to all kinds of DNA damage, as recombinational repair mutants are (10). To test this possibility, we treated the rnhAB and recBCD mutants, along with the control strains, with various DNAdamaging agents. The rnhAB double mutant was indeed somewhat sensitive to DNA gyrase inhibitor nalidixic acid at 37 • C (Supplementary Figure S1B). At the same time, the rnhAB mutant showed WT resistance against the DNA methylation agent methyl methanesulfonate (MMS), the DNA alkylating and crosslinking agent mitomycin C, and the oxidative agent hydrogen peroxide (while control mutants were killed by 2-6 orders of magnitude (Supplementary Figure S1C-E). Thus, RNase H-deficient E. coli mutants are not depleted for their global NER or recombinational repair capacities and therefore are generally DNA repair-proficient. We reasoned that some UV-induced PDs combined with RDHs accumulating in the RNase H-deficient mutants could create irreparable chromosome lesions (UVRLs, Figure 1A), --and set out to understand their nature.

Translesion synthesis and the RecFOR pathway do not generate or repair UVRLs
Due to high SOS-induction in the rnhAB mutants (18), the SOS-induced TLS polymerases could contribute to formation of the UVRLs by bypassing pyrimidine dimers with insertion of rNs instead of dNs (44), generating an R-tract or an R-patch (a short R-tract or a mixed run of rNs and dNs) (Figure 2A). In particular, Pol V polymerase is known to readily use rNTPs for translesion synthesis (45). If PD is then removed by NER from the template This model predicts: (i) an improved UV-survival of the rnhAB mutants with additional mutations in Pol IV (dinB) and Pol V (umuCD) polymerases; (ii) a synergistic UV sensitivity of the rnhAB recF triple mutant, in which RecFOR-dependent recombinational repair of daughterstrand gaps is blocked (10) (Figure 2A and Supplementary Figure S2), leaving TLS as the only pathway to fill the ss-gaps to produce more R-patches. Contrary to these predictions, the rnhAB umuCD dinB quintuple mutant shows slightly elevated sensitivity at higher UV doses, which parallels the modest sensitivity of the umuCD dinB triple mutant itself at these doses ( Figure 2B) (46) likely reflecting formation of lesions processed only by TLS. Further, we observed no increase in the density of ribonucleotides in plasmid DNA, isolated from growing or UV-irradiated rnhAB or rnhAB uvrA mutants, tested for presence of single rNMPs or R-patches in the plasmid relaxation assay (18) with either RNase HII ( Figure 2C) or RNase HI (Supplementary Figure S3).
Finally, although the rnhAB recF mutant is more sensitive to UV than either recF or rnhAB mutants, there is no synergy between the recF and rnhAB defects, as UV-survival of the triple rnhAB recF strain is similar to the product of the two mutant survivals ( Figure 2D), suggesting that handling of the UVRLs in the rnhAB mutants is not channeled through the RecF-dependent repair of blocked ss-gaps (therefore, UVRLs are not blocked ss-gaps). Thus, our genetic analysis suggests that TLS is not responsible for generating UVRLs in the rnhAB mutants, whereas the RecFOR repair of blocked ss-gaps neither mends UVRLs, nor participates in their formation.

UVRLs are not resolved into double strand breaks in the rnhAB mutants
Even sublethal doses of UV cause significant chromosome fragmentation in E. coli via replication fork disintegration (12). The broken replication forks in wild type E. coli are robustly reassembled by the RecBCD-promoted recombinational repair (10), so the fragmented chromosomes are only detectable in the UV-sensitive recBC mutants, as linear DNA species in pulsed-field gels (12). If UVRLs in the rnhAB mutants cause replication fork disintegration, for example initiated by attempted repair of PD across R-loop ( Figure 2E), then there will be synergistic decrease in UV-survival of the rnhAB recBC mutant, compared to both the rnhAB and recBC mutants, as well as additional fragmentation in UV-irradiated cultures of the rnhAB recBC mutant compared to the Nucleic Acids Research, 2021, Vol. 49, No. 7 3895 UV-induced chromosome fragmentation in the recBC mutant (12).
We showed before that the rnhAB mutant in AB1157 background is not temperature sensitive, but rnhAB recBC(Ts) mutants are synthetic lethal at 37 • C (nonpermissive temperature for recBC(Ts)) and accumulate DSBs after 4 h of growth at this temperature (18). To avoid this DSB formation and lethality, we incubated plates with UV-irradiated rnhAB recBC(Ts), rnhAB and recBC(Ts) cells at 38 • C for only 2 h and shifted them to 28 • C to allow colony formation. We detected no additional UV-sensitivity of the rnhAB recBC mutant compared with the rnhAB mutant ( Figure 2F), indicating epistasis of the rnhAB defect over the recBC defect.
We also measured DSB accumulation with pulsed-field gels in the rnhAB, recBC(Ts) and rnhAB recBC(Ts) strains exposed to 2, 16 or 36 J/m 2 of UV doses, followed by 2 h of post-irradiation recovery in the growth medium at 37 • C. We confirmed the same level of fragmentation after 16 and 36 J/m 2 of UV in the recBC mutant (12), detected low level chromosome fragmentation in the rnhAB double mutant (as expected, because the strain is DSB-repair proficient (18)) and found no differences in the chromosome fragmentation between the recBC and rnhAB recBC mutants at any UV dose ( Figure 2G). We conclude that the putative UVRLs in the rnhAB mutant are not resolved into double strand breaks, at least within the first 2 h after UV. The possibility that formation of UVRLs itself could depend on RecBC was tested later (Figure 4); initial UVRLs were found to form independently of RecBC.

Post-UV DNA synthesis recovery is inhibited in the rnhAB mutants
Since UVRLs do not kill via chromosome fragmentation, we measured DNA synthesis rates in growing versus UV-irradiated cultures to test weather UVRLs inhibit replication restart after UV. The rate of DNA synthesis in the unirradiated rnhAB mutant cultures is about two times slower than in the WT cultures ( Figure 2H, left), likely reflecting the 40% viability of the rnhAB mutants (18). After sublethal doses of UV irradiation, DNA synthesis in WT cells is known to be blocked for about one generation, until most PDs are removed by NER,--and then resumes at a slightly faster rate (6,7). We compared kinetics of the DNA synthesis restart after 36 J/m 2 UV in three strains: WT, uvrA and rnhAB mutants. The WT culture behaved as expected: its DNA synthesis was blocked for 30 min, then suddenly resumed ( Figure 2H). Again as expected, in the uvrA mutant, the post-UV DNA synthesis never recovered ( Figure 2H), reflecting the inability of this mutant to remove PDs. In the rnhAB mutant, the post-UV DNA synthesis was blocked for 50 min (about twice as long as in WT cells), but then recovered from 50 to 90 min, only to level off after 120 min and to continue at this reduced level ( Figure 2H). We concluded that in the rnhAB mutants, delayed and incomplete recovery of the post-UV DNA synthesis must reflect formation of UVRLs, which inhibit progress of replication forks, but without causing chromosome fragmentation.

UVRLs are formed by transcription due to PDs in DNA
The inability of the rnhAB mutants to continue the restarted DNA synthesis after UV could reflect formation of Rloop-aTECs (18,30), which permanently stall replication forks in the head-on orientation in the absence of RNase H (31,32). We tested the contribution of transcription to the UV sensitivity of the rnhAB mutant by pretreating cells with rifampicin, which blocks transcription initiation, while allowing the ongoing transcription to finish (47). A 5 min pretreatment of growing cultures with 100 g/ml rifampicin and its subsequent removal before UV-irradiation made the wild type cells mildly sensitive to UV, most likely via the residual rifampicin interference with SOS-induction (48) ( Figure 3A). In contrast, rifampicin pretreatment elevated UV resistance of the rnhAB mutant to the (rifampicinreduced) level of the wild type strain ( Figure 3A), strongly implicating transcription after UV in the formation of lethal UVRLs. Such rifampicin pre-treatment does not change the UV-sensitivity of recBC or recF mutants (Supplementary Figure S4A), arguing for its specific effect in the rnhAB mutant, but it does alleviate the ruvABC mutant, again implicating transcription in its UV-sensitivity.
In the RNase H-deficient mutant, transcription likely interferes with replication via formation of R-loops. To see if R-loops could be a part of UVRLs, we overproduced the RecG, a duplex DNA pump with some R-loop disassembling activity (49), whose overexpression in recBC, recF or ruvABC mutants had either no or small effect (Supplementary Figure S4B). Overexpression of RecG from a plasmid indeed improved UV resistance of both the rnhAB double and the rnhA single mutants to a similar extent ( Figure 3B), suggesting that R-loops indeed participate in UVRLs. But then how could PDs in DNA cause formation of R-loops, if the plasmid-nicking assay indicates normal kinetics of DNA-PD removal in the rnhAB mutants ( Figure  1DE)?
Since ribosome-free mRNA stimulates R-loop formation (50), and since ribosomes stall at mRNA lesions (51), UV could induce R-loop-aTECs if ribosomes stall on PDs in nascent mRNA, yielding ribosome-free mRNA behind TECs ( Figure 3C). This logic predicts that inhibition of translation elongation with chloramphenicol treatment before UV irradiation would further exacerbate UVsensitivity of the rnhAB mutants. However, contrary to this expectation, we found no additive effect of chloramphenicol on the UV-sensitivity of the rnhAB mutants (Supplementary Figure S5A). Moreover, we observed an extreme UV-sensitivity of the uvrA rnhAB triple mutant: at UV doses as low as 0.6 J/m 2 , when rnhAB, uvrA, uvrA rnhA and uvrA rnhB strains are still fully UVresistant, the triple mutant is already killed almost 20fold ( Figure 3D, Supplementary Figure S5B). This synergy of the uvrA and rnhAB defects strongly argues against ribosome stalling at PDs in mRNA, since UvrA protein recognizes PDs only in the context of duplex DNA (52). Synergy of two mutations inactivating distinct enzymes usually means that one of the enzymes 'repairs/removes' a particular damage, while the other one 'prevents formation' of the same damage (53). Since overexpression of UvrA failed to relieve UV sensitivity of the rnhAB mutant (Figure 1C), and therefore removal of UVRLs must be the function of RNase H enzymes, it follows that NER must prevent UVRL formation by excising PDs from duplex DNA.

UVRLs must be R-loops formed at stalled Transcription Elongation Complexes
Therefore, it must be transcription-elongation complexes (TECs) blocked by PDs in template DNA that initiate formation of R-loop-aTECs ( Figure 3E). RNA pol II stalling in eukaryotic cells was proposed to induce Rloops behind, after spliceosome eviction from mRNA (54). Our earlier observation that UvrA overexpression does not improve UV resistance of the rnhAB mutant ( Figure  1C) suggests that PDs that cause UVRLs are inaccessible to NER,--for example being masked by stalled RNA polymerases ( Figure 3E). Mfd protein promotes TCR by removing stalled RNA polymerase from the blocking PD and recruiting UvrAB to the now accessible PD (4). Such Mfd action predicts that the mfd rnhAB mutants should be more sensitive to UV. By themselves, mfd mutants are not UV sensitive ( Figure 3F); however, the mfd defect indeed exacerbates the UV sensitivity of the rnhAB mutants ( Figure 3F), supporting the idea that PD-stalled RNA polymerases contribute to UVRL formation.
One more process that removes stalled R-loop-aTECs is the long-range RNA polymerase backtracking by the UvrD helicase (55). GreA and GreB are elongationprocessivity factors that minimize this UvrD-promoted RNA polymerase backtracking by restarting transcription (55). This logic predicts that the greA and greB defects, via eliminating control over backtracking, should make RNA polymerase vulnerable to UvrD removal and therefore should relieve the UV-sensitivity of rnhAB mutants. We found that greA greB rnhAB mutant is indeed more UVresistant than the rnhAB mutant ( Figure 3G), further supporting the idea that UVRLs are R-loop-aTECs.
If UVRLs are R-loop-aTECs, whose strength reflects the high stability of transcription complex, then destabilizing stalled RNA polymerase in E. coli with the rpoB*35 mutation (56) should suppress the UV sensitivity of the RNase H-deficient cells. Indeed, we found the rpoB*35 mutation to be the strongest suppressor of UV-sensitivity of the rnhAB mutant ( Figure 3H). Moreover, even though the post-UV lag of the chromosomal DNA synthesis in the rhnAB rpoB*35 mutant is as long as in the rnhAB mutant, its replication completely recovers after 60 min, in contrast to rnhAB mutant ( Figure 3I). At the same time, the rpoB*35 defect neither accelerates the repair of PD dimers by global NER (Figure 1E), nor reduces the density of DNA-rNs in the rnhAB mutants (Supplementary Figure S6).
Thus, our UV-sensitivity and DNA synthesis rate data strongly argue for the formation of UV-induced R-loop-aTECs at PDs in DNA, with RNase HI and RNase HII being the main enzymes to remove them (Supplementary Figure S7). Moreover, processes destabilizing transcribing RNA polymerases and R-loops alleviate the UV-sensitivity of rnhAB mutants, further supporting this scenario (Supplementary Figure S7). Importantly, the idea of PD-dependent R-loop-aTECs ( Figure 3E) generates two testable predictions about chromosomal DNA in the UV-irradiated rnhAB mutants: (i) it should accumulate RDHs representing R-loop-aTECs; (ii) it should preserve a small fraction of NER-resistant PDs (presumably those that cause R-loop-aTEC accumulation).

UV-induced transcriptional RDHs accumulate in rnhAB mutants
To identify the predicted UV-induced RDHs in the rnhAB mutants, we probed chromosomal DNA with the RDHspecific antibodies S9.6 (57). To minimize the potential RNA contamination in our assays, we removed the bulk of RNA with RNase A treatment during cell lysis and ran the DNA samples in two identical agarose gels, followed by electric transfer of the nondenatured samples from one gel to nylon membrane for Western analysis with S9.6 antibodies ( Figure 4A, top). To quantify RDH density, we normalized the S9.6 signal in the chromosomal DNA band to the corresponding chromosomal DNA signal from Southern analysis of the other gel ( Figure 4A, bottom). Since the S9.6 signal is barely detectable in the RnhA+ strains, yet is readily detected in the growing rnhAB mutants, for a reliable comparison between strains and experiments we normalized each RDH density value within the experiment to the RDH density value of the growing rnhAB culture ( Figure 4B, E-I).
First, we compared the RDH density in growing cells to the one from UV-irradiated cells, recovered post-UV for 1 h ( Figure 4B). RDH signal is detected even in unirradiated growing cultures, with the maximum value in the rnhAB mutants ( Figure 4B, the green bars). After 1 h of post-UV recovery, RDH density increases in all strains: ∼4-fold in WT and rnhB single mutant cells, 9fold in the rnhA single mutant, ∼13-fold in the rnhAB double mutant, ∼5-fold in the rnhAB rpoB* triple mutant ( Figure 4B, green versus purple bars). We conclude that, in various rnh mutants, their UV-sensitivity correlates with both the relative and the absolute accumulation of RDHs.
The UV-induced RDH signal from genomic DNA is removed by in vitro treatment with RNase HI, but not with RNase HII (Figure 4CD), suggesting its Rloop nature. At the same time, UV-induction of RDH signal is completely blocked by rifampicin ( Figure 4E), indicating its generation as a result of transcription. This RNase HI-sensitivity of the RDH signal (also see Supplementary Figure S8), its dependence on transcription ( Figure 4E and Supplementary Figure S9A), as well as its distribution over the chromosome (Supplementary Figure  S9C) and resistance to high temperatures (Supplementary Figure S10) are all consistent with its being extended RDHs, likely long R-loops (see the Supplement for a complete description of this characterization).

Accumulation of RDHs depends on restart of DNA replication
Since UV-irradiated recBC rnhAB mutant cells failed to induce additional chromosome fragmentation over the recBC mutant background ( Figure 2G), there was a possibility that the recBC defect interferes with formation of UVRLs in the rnhAB mutant, yielding lower RDH density. However, we found no difference between RDH signals between the rnhAB and recBC rnhAB mutants, either in growing cells or 30 min after UV, although the RDH signal 60 min past UV was indeed ∼2-fold lower in the recBC background ( Figure 4F). Since the post-UV replication recovery happens around this time in the rnhAB mutant ( Figures 2H or 3I), and since RecBCD enzyme is important for post-UV replication recovery (12), the decrease in RDH signal due to the recBC defect suggested involvement of DNA replication in generation of the maximal UV-induced RDH signal in the rnhAB mutants. Indeed, blocking replication (re)initiation with the dnaC mutation (58) decreases RDH signal at 60 min post-UV ( Figure 4G),--confirming that active replication forks are required to generate maximal UV-induced RDH signal.
To find the time of maximal accumulation of RDH signal, we performed the time course in various mutants ( Figure 4HI). We found that: (i) although there is no additional RDH signal at 10 min after UV, the significant increase in RDH signal at 30 min post-UV happens when no DNA synthesis is detected yet (cf. Figure 2H); (ii) maximum RDH signal in all strains is at 60 min post-UV--by this time DNA replication is either fully recovered (in WT cells, Figure 2H) or starts recovering (in the rnhAB and rnhAB rpoB35* mutants, Figure 3I),--supporting the idea that replication restart is required to amplify RDH; (iii) once at the maximal level, RDH signal remains relatively stable in the rnhAB and rnhAB mdf mutants for 1 or 2 h, while decreasing to the background levels in the WT, rnhA and rhnAB rpoB*35 strain ( Figure 4HI). Importantly, rpoB*35 mutation decreases both formation (at 30 min post-UV) and accumulation (at 60 min post-UV) of RDHs and also facilitates their removal (at 120 min post-UV) ( Figure 4I), which is in agreement with the UV viability and rifampicin treatment data, and shows that stable (WT) RNA polymerase is critical for the formation and stability of UV-induced RDHs.
Finally, the maximal RDH signal in the rnhAB mutants after 36 J/m 2 of UV requires active NER, as it is reduced several times in the rnhAB uvrA mutant (Supplementary Figure S11). However, at 9× lower UV dose of 4 J/m 2 , the RDH signal increase in the rnhAB uvrA triple mutant becomes similarly dramatic ( Figure 4I), suggesting that, by removing PDs, NER promotes RDH signal increase indirectly, likely by relieving UV-inhibition of DNA synthesis (12).

A small fraction of PDs is resistant to NER in the rnhAB mutants
Our genetic and physical data so far strongly support the idea that, after UV exposure, the ongoing transcription spawns accumulation of RDHs, implying they are R-loop-aTECs stalled at pyrimidine dimers ( Figure 3E). In the rnhAB mutants, these R-loop-aTECs cannot be removed and limit the post-UV recovery of DNA replication. This scenario predicts that, while the bulk of PDs in the rnhAB mutants is removed without problem, the few PDs which cause the RDH increase should be masked from NER, by either stalled RNA polymerases or by the associated R-loops ( Figure 5A). To test the prediction that a small fraction of PDs would be removed slowly from the chromosomal DNA of UV-ed rnhAB mutant cells, we determined kinetics of PD repair in WT, uvrA, rnhAB, rnhAB rpoB35* and rnhAB mfd mutants. In contrast to our previous plasmid-based measurement of the global NER capacity ( Figure 1DE), this time we measured PD removal in the chromosomal DNA by anti-pyrimidine dimer-specific antibodies (PD-antibodies) ( Figure 5B, top).
The PD-antibody signals were normalized to the corresponding DNA signals ( Figure 5B, bottom) to calculate PD-densities. The PD-density from the 5 min post-UV sample was taken for 100% to calculate subsequent PD disappearance due to repair at 30, 60, 120 and 180 min ( Figure 5C). Controls show either complete removal of PDs within 60 min in WT cells or PD persistence in the NERnegative (uvrA) mutants (Supplementary Figure S12A). The rnhAB uvrA mutant shows no PD removal either ( Figure 5C). At the same time, in both the WT and rnhAB rpoB35* cells more than 90% of the chromosomal PDs are removed by 30 min (kinetics of PD removal is the same in the rnhB rpoB* RNase HI+ mutant (Supplementary Figure  S12B versus Figure 1E)), while only background signal is detected by 60 min post-UV. In contrast, in the rnhAB and rnhAB mfd mutants, PD repair slows down after 30 min post-UV, with ∼10% of all PDs still persisting by 1 h and ∼5% still detectable by 2 h (Figure 5C).
We conclude that: (i) in the rnhAB mutants, 5-10% of PDs in the chromosomal DNA are removed slowly, if at all; (ii) in this and also in the rnhAB mfd mutants, at 60 min post-UV recovery, the highest RDH signal correlates with retention of ∼10% PDs in the chromosomal DNA ( Figures  4HI versus 5C); (iii) in the 30-60 min interval in these two mutants, the gradual PD removal correlates inversely with the RDH accumulation, suggesting that R-loops expand, while PDs are slowly yet continuously removed; (iv) the rnhAB rpoB35* mutant repairs PDs like WT,--apparently because of the unstable RNA polymerase. Interestingly, up to 30 min post-UV PD repair proceeds at similar rates in all NER+ strains ( Figure 5C),--yet after that time the rnhAB mutants restart post-UV replication and at the same time slow down PD removal,--suggesting it is the arrival of replication forks that slows down removal of the remaining PDs; (v) no Mfd-dependent PD removal is observed in the rnhAB mutant.

RDH fraction is enriched for PDs
If some PDs stall transcribing RNA polymerases, thus avoiding detection by NER and also inducing RDH formation behind stalled TECs, --these NERresistant PDs should be in proximity of the RDHs whose formation they have instigated ( Figure 5A). If, so, enriching chromosomal DNA of the UV-irradiated rnhAB mutants for RDHs should also enrich this DNA for PDs ( Figure  6A).
To test the idea of PD proximity to UV-induced R-loop-aTECs, we performed DNA:RNA immunoprecipitation (DRIP) of the chromosomal DNA isolated from rnhAB mutant 60 min post-UV, at the maximum of the RDH signal ( Figure 4H). The recovered DRIP fraction was analyzed for both RDH and PD densities by dividing the antibody signals by the corresponding DNA signal, and further normalizing it to the RDH and PD densities of the input DNA, to calculate the enrichment factor. As a negative control, the same analysis was applied to the flow-through fraction of DNA that failed to bind to S9.6 antibodies in the DRIP protocol (we call it 'eluate'), in which RDH signal depletion was expected.
We started DRIP enrichment with the rnhAB mutant irradiated with 36 J/m 2 , but detection of the remaining PD signal at 60 min post-UV was unreliable. Therefore, we switched to the rnhAB uvrA strain exposed to either 4 or 8 J/m 2 UV (Figure 4I), to stabilize PDs. However, since in this excision-deficient strain, PDs were expected to be present in both the RDH-containing DRIP fractions and the eluate fractions, this made PD enrichment sensitive to the size of RDHs. Therefore, we performed DRIP analysis with the chromosomal DNA fragments of various sizes, as illustrated by the agarose gel ( Figure 6B, left), with the logic explained in the table ( Figure 6C).
Initially we assumed that R-loops were mostly short and within ∼1 kb of PDs that induced them. Therefore, we expected that cutting DNA to small ≤1 kb fragments would facilitate enrichment for PD signal after DRIP. The RDH size distribution was revealed by Western hybridization of the genomic DNA cut to various average fragment sizes ( Figure 6B, C). Unexpectedly, we found that the RDH size distribution matched the actual fragment size distribution only for the EcoRI + BamHI cutting ( Figure 6B, C). For HaeII and HaeIII cutting the average RDH signal size was much longer than the average DNA fragment size ( Figure 6B, C). These longer-than-expected RDHs in the chromosomal DNA significantly limited the theoreticallypossible enrichment factors for PDs in the DRIPped DNA ( Figure 6C, the rightmost column).
As expected, the DRIP fraction was enriched for RDH, ∼10-fold for the uncut DNA and 30-50-fold for the digested DNA ( Figure 6D). The enrichment of the DRIP fractions was validated by depletion of the RDH signal in the corresponding eluate fractions, so compared to the eluate, the overall enrichment (DRIP/eluate) was increased to 15-500, depending on the cutting ( Figure 6D). Since only ∼1% of RDH signal survives RNase HI treatment ( Figure 6B), we expected no enrichment after DRIP of the RNase HI-treated samples. However, S9.6 antibodies still provided some 15-fold enrichment of the RNase HI-treated DNA relative its eluate fraction (compared to 95-fold enrichment for untreated samples) (Supplementary Figure  S13A), indicating an RDH-independent background of the procedure. Our attempts to reduce it by pre-treating the beads with glycogen or salmon-sperm DNA were unsuccessful, suggesting that it is coming from S9.6 antibody itself and is the intrinsic background of the DRIP procedure.
Next we measured enrichment for PDs in the DRIPped samples ( Figure 6E). Since the eluate fractions of various digestions expectedly showed similar levels of 'no enrichment', we have combined them all in the 'combined eluate' value ( Figure 6E, the far right). The observed PD enrichment by DRIP turned out to be somewhat higher than the theoretical expectations ( Figure  6C), based on the idea of PD proximity to RDHs ( Figure  6E, the purple bars versus gray bars). We found no PD The RDH density was determined in the DRIP fraction, the corresponding eluate fraction, and in the original INPUT fractions (the latter was used for normalizing the first two to calculate enrichment). Note the logarithmic scale of Y-axis. (E) DRIP enriches for PDs. The density of PDs was determined in the DRIP, the eluate and the input fractions from 'D' and then normalized to the input values (purple bars). Since all the eluates showed the same density of PDs (as expected), they were all averaged together. Theoretically-expected enrichment based on the density of PDs and on the average sizes of RDH distributions in panel C is shown by gray bars for comparison. signal enrichment without cutting, a 2-fold enrichment with EcoR + BamH cutting, a 3-5-fold enrichment with HaeII and HaeIII cutting, the latter two values being both significantly different from the (no) PDenrichment value of the combined eluate ( Figure 6E) (also see Supplementary Figure S13B, C for an alternative normalization yielding the same conclusion). RNase HI treatment of the chromosomal DNA before DRIP is expected to decrease the DRIP enrichment for both RDH and PDs. Indeed, after RNase HI treatment of the HaeII-cut DNA we found statistically-significant 2-fold decrease of PD-enrichment in the DRIPed fraction ( Figure  6E). We conclude that enrichment for RDH with DRIP causes co-enrichment for PDs, indicating proximity of PDs to RDHs relative to the non-enriched genomic DNA.

Co-enrichment of PDs with RDHs
Although DRIP did enrich for PDs in the chromosomal DNA of the rnhAB mutants, both the low theoretical limits of this enrichment ( Figure 6C) and the technical issues of DRIP background made us seek a simpler and more robust procedure to further characterize and quantify the RDH-PD co-enrichment. We have noticed that, in the gel-separated HaeII-digested chromosomal DNA of UV-irradiated rnhAB mutants, the longest genomic DNA pieces (6-10 kb) both accumulate RDHs preferentially and repair PDs slower than the bulk of this DNA, which has a mode ∼1.2 kb (Supplementary Figure S14ABC). To test whether UV-induced RDHs indeed co-localize with the remaining PDs in the HaeII-digested genomic DNA, we partitioned these lane profiles, from wells to gel  Figure  S14D, the bottom graph).
Since we could not apply this calculation for the control WT strain at 60 min post-UV, because, even though it did have detectable RDHs ( Figure 4ABH), all PDs were already removed ( Figures 1E and 5C),--we limited this analysis to the time 30 min post-UV. Such quantification for WT cells shows an almost uniform co-enrichment across the lane with an average non-peak value of ∼ 0.5 (which can be taken for the procedure background in excision repairproficient cells) and a small peak of ∼2.0 in fraction #6 ( Figure 7A). In contrast, the rnhAB mutant shows a sharp co-enrichment peak of ∼29 in the same fraction #6 and the average non-peak co-enrichment of ∼1.0 ( Figure 7A; also see the linear version of this plot in Supplementary Figure S14E). Thus, while the non-peak rnhAB profile shows 2× higher RDH-PD co-enrichment over the WT background, its peak fraction #6 features ∼15× higher coenrichment. Interestingly, this peak co-enrichment of the rnhAB mutant was mostly driven by the RDH differences with the WT strain ( Figure 7B), as their PD density profiles were similar ( Figure 7C).
While the wild type cells provided the 'low background' of RDH-PD co-enrichment, the nucleotide-excisiondeficient uvrA mutants, with their extremely high and evenly-distributed PD signal ( Figure 7C) yet low and uniform RDH signal ( Figure 7B and Supplementary Figure S11), should provide the 'high background' of RDH-PD co-enrichment. Indeed, the uvrA mutant developed no co-enrichment peak, but showed a coenrichment plateau around the average of 2.0 ( Figure  7A), which also coincided with the co-enrichment peak of fraction #6 in the WT cells. This confirms not only the 'high background' estimate for RDH-PD co-enrichment, but also the reality of ∼15× higher RDH-PD co-enrichment peak in the rnhAB mutant (Supplementary Figure S14E).
Calculations of co-enrichment, based on an alternative way to normalize fraction densities that reduces runto-run variations, have reached the same conclusions (Supplementary Figure S15). Curiously, profiles in Figure  7C indicate that PD densities in fraction #6 are similar in all three strains, suggesting that PDs in this fraction are protected from excision in the excision-proficient cells. Since 60 min post-UV, PDs completely disappear in WT cells, while modestly decreasing in the rnhAB mutant ( Figure  5C), their subsequent persistence in this mutant must be due to its much higher RDH density ( Figure 7B). Overall, the sharp co-accumulation of RDHs with the slow-torepair PDs in the longest size fraction in the HaeIIcut chromosomal DNA of UV-irradiated rnhAB mutants implies their physical co-localization and suggests that PDs are the cause of RDH accumulation.

DISCUSSION
Ultraviolet irradiation (UV) is, perhaps, the best understood DNA-damaging treatment, and not only because it is the most frequent natural form of DNA damage, but mostly because it generates a narrow class of DNA lesions, primarily pyrimidine dimers (PDs), that are removed by the best-studied and ubiquitous form of repair, the nucleotide excision repair (NER) (52). UV is known to kill bacteria when there are too many UV-lesions in DNA (saturating excision repair) (8), or via irreparable chromosome fragmentation (saturating recombinational repair) (12). Our investigation of the unexpected UV-sensitivity of the RNase H-deficient mutant of E. coli revealed a novel lethal UV-induced DNA lesion of a complex nature that forms in three distinct stages ( Figure 7D). The first stage is UV-induced PD formation in a coding strand of an expressed gene. At the second stage, this PD blocks the progress of a TEC, with two consequences: (i) excision repair of this PD is prevented; (ii) the nascent mRNA forms a small R-loop behind the stalled TEC (this step is postulated, as it is hard to detect). These presumably short original R-loops near PDs require something else to become the chromosomal lesions that would be able to inhibit normal DNA synthesis after UV. The third stage of the UVRL formation reveals itself by amplification of RDH signal, coincident with the post-UV replication restart--therefore we propose that it is the arrival of head-on replication forks, that turns the original short R-loops into long R-loop-anchored transcription-elongation complexes (R-loop-aTECs) ( Figure 7D). Such R-loop elongation is expected from reports that head-on replication-transcription conflict promotes R-loop formation (31,32). It is speculated that the positive supercoiling between the two polymerases could induce compensatory gyrase action on the back of R-loop, generating hyper-negative supercoiling promoting R-loop elongation (30). We propose that such PD-induced long RDHs become lethal DNA structures in the RNase H-deficient mutants, making the affected DNA duplex impassable for the replication machinery.

The long search for the right model
Before arriving at the current model ( Figure 7D) explaining the surprising UV-sensitivity of the RNase H-deficient E. coli mutant that still has functional NER and recombinational repair, we experimentally tested and disqualified other models of the postulated UV-RDH relationship yielding lethal DNA lesions, UVRLs ( Figure 1A). The first one was a simple possibility that UV-independent preexisting RDHs deplete either NER or recombinational repair, making the rnhAB mutant cells phenocopies of the, correspondingly, uvr or rec mutants. Our earlier observation that the uvrA rnhAB strain grows slower than rnhAB strain (18) did imply that NER helps repair RDHs formed during regular growth, and therefore NER could be distracted by them and become less efficient to repair PDs. However, normal kinetics of PD removal from bulk DNA ( Figure 1DE) of the rnhAB mutants and their WT-sensitivity to various DNA damaging treatments (Supplementary Figure S1CDE) was inconsistent with them becoming NER-or recombinational repair-defective phenocopies.
The second model envisioned DNA polymerase translesion synthesis through PDs introducing R-patches across them, inhibiting subsequent DNA synthesis in both strands (Figure 2A). However, inactivation of both translesion DNA polymerases failed to alleviate UVsensitivity of the rnhAB mutants, while the density of DNA-rNs after UV in these mutants failed to increase ( Figure 2BC),--both observations being inconsistent with this model.
The third model, an aggravation of a milder scenario in Figure 2E, envisioned that aborted repair of a PD in the DNA strand of an R-tract results in replication fork explosion (Supplementary Figure S16), which forms irreparable double-strand breaks (40). However, its supposedly more frequent 'sister' scenario, replication fork collapse at the R-loop across the NER-generated ss-gap ( Figure 2E), predicted not only synergy between UVsensitivity of the rnhAB and recBC defects, but also more chromosome fragmentation in the rnhAB recBC mutants after UV. Neither effect was observed experimentally ( Figure 2FG), disqualifying the model of irreparable double-strand breaks of Supplementary Figure S16.
The forth model was stimulated by our observation that recovery of the post-UV DNA synthesis in the rnhAB mutants is only partial (Figure 2H), suggesting a persistent interference, which we linked to UV-concurrent transcription generating R-loops ( Figure 3AB). At first we envisioned stalling of translating ribosomes at PDs in mRNA, leaving the downstream mRNA empty and thus stimulating formation of R-loops, which should be stable in the rnhAB mutants ( Figure 3C). However, PDs in mRNA predicted no additional UV-effect of the rnhAB inactivation in the highly UV-sensitive NER-deficient uvrA mutant, because NER removes PDs only from duplex DNA (52). Therefore, this attractive idea was disqualified by our finding of the synergistic interactions between the uvrA and rnhAB defects at the very low UV doses ( Figure 3D), which instead strongly suggested that the culprits were PDs in DNA.
It is only when we envisioned TECs stalled at PDs in template DNA forming R-loop-aTECs ( Figure 3E), that block subsequent replication through the region, had we finally arrived at a model whose predictions consistently matched our subsequent experimental observations. In particular, the model predicted that making RNA polymerase unstable should eliminate the UV-sensitivity of the rnhAB mutants ( Figure 3H) and alleviate the inhibition of post-UV DNA synthesis ( Figure 3I). It also predicted copious formation of UV-induced RDHs (Figure 4), a small fraction of repair-resistant PDs ( Figure 5) and proximity of PDs to RDHs (Figures 6 and 7). But there was one important aspect of the mechanism of the lethal UV-induced R-lesions, revealed by our experimentation, that our model failed to anticipate,--which is discussed next.

The role of DNA replication in RDH formation
Our initial model ( Figure 3E), according to which Rloops formed due to TECs stalling at PDs, predicted new RDH formation right after UV. Indeed, it should take transcribing RNA polymerase in E. coli (moving at ∼50 nt/s (59,60)) about 20 seconds to traverse an average gene in E. coli (∼1 kb (61)). Even the longest genes or operons in E. coli should be transcribed in less than 5 min. If five more minutes is generously given for completion of Rloop formation,--then a significant RDH signal should be induced within 10 min after UV. Therefore, we were surprised to see absolutely no increase in RDHs 10 min after UV ( Figure 4HI). The significant increase in RDHs was detected in another 20 min, 30 min post-UV, and reached its maximum at 60 min post-UV ( Figure 4HI), as if post-UV RDH amplification depended on yet another factor, which was missing 10 min after UV and was still weak 30 min after UV.
The only factor missing during that time but restored 60 min after UV in the rnhAB mutants was DNA synthesis ( Figure 2H). Indeed, inhibiting DNA synthesis restart with the dnaC(Ts) defect significantly reduces Nucleic Acids Research, 2021, Vol. 49, No. 7 3903 RDH amplification ( Figure 4G), confirming replication forks as the third critical component in the formation of lethal UVRLs. It was reported for both bacteria and human cells that the head-on transcription-replication conflict stimulates formation of pervasive R-loops in the transcribed gene involved in the conflict (31,32). We propose that the same mechanism explains the UV-induced RDH accumulation, especially since excess of replication forks is expected in our cells. Indeed, besides the restarted regular forks, in the UV-irradiated rnhAB mutants there should be additional forks due to two types of stable DNA replication (62): (i) the constitutive one observed in the rnhA mutants; (ii) the inducible one observed in recombinational repair-proficient cells after any type of DNA damage and induction of SOS.

Do R-loops cause double-strand breaks?
Whenever replication is inhibited, double-strand breaks tend to form (63,64), and we were initially surprised to find that inhibition of DNA replication after UV in the rnhAB mutants did not yield (additional) fragmentation ( Figure 2G). In particular, this finding contradicts the widely-accepted view that replication fork collision even with protein-free R-loops generates DSBs (28,29). The key distinction could be accumulation of positive supercoiling.
It was argued before that transcription elongation complexes (TECs) with R-loops behind form stable structures (R-loop-aTECs) that in the rnhAB mutants could block head-on replication forks through accumulation of positive supercoiling, but without necessary generating double-strand breaks (30). Indeed, E. coli gyrB mutants with enhanced positive supercoiling did not require recombinational proteins for viability, suggesting that forks stalling by positive supercoiling does not result in doublestrand breaks (65).
However, lethality and replication fork breakage was observed at the sites of transcription-replication collision in recB mutants in which rrn operons were inverted relative to oriC. Interestingly, under these conditions overexpression of RNase HI enzyme did not rescue lethality of the strain, suggesting that the conflict in this case did not involve Rloops, but was a direct collision between replication fork and RNA polymerases transcribing rDNA (66).

Features of the UV-induced RDHs
To detect the UV-induced RDHs quantitatively, we used S9.6 RDH-specific monoclonal antibodies (57,67), normalizing the S9.6 signal of the chromosomal DNA band from western blot to the chromosomal DNA signal from Southern blot; since the latter varies depending on the strength of the probe, we further normalized the resulting RDH densities within every experiment to the RDH density in growing rnhAB cells (Figure 4). Although quantitative comparison of mostly qualitative Western signals is non-trivial (68), our RDH signal quantification proved accurate and reproducible, being able to reliably detect the differences between WT and rnhAB mutant strains even during normal growth, when the RDH density is low. An important contributor to this success could be our method of genomic DNA preparation, as not all DNA isolation methods are R-loop-friendly (57). Running genomic DNA in agarose gels instead of doing direct dot-hybridization also helped to reduce irreproducibility, likely by removing signal-generating or -inhibiting contaminations.
What is the nature of S9.6 signal that forms so abundantly in UV-irradiated RNase H-deficient cells? First, S9.6 antibodies besides RNA:DNA hybrids also detect dsRNA, although 5-times less efficiently (57,69). Our S9.6 signal is completely removed by in vitro treatment with RNase HI, but is mostly resistant to ssRNA and dsRNAremoving high-salt or low-salt treatments with RNase A (Supplementary Figure S8),--therefore it should represent RNA:DNA hybrids, either R-loops or R-tracts ( Figure  1A, top). Indirectly, since we failed to detect UV-induced accumulation of R-tracts in plasmid DNA (Supplementary Figure S3), the chromosomal S9.6 signal most likely represents R-loops, rather than R-tracts. This tentative conclusion will have to be tested in the future.
The decreased strength and the complete disappearance with time of the UV-induced RDH signal in the rnhA single mutant contrast with high RDH signal accumulation and stability in the rnhAB double mutants, highlighting the role of RNase HII in RDH removal. Interestingly, as we find no RNase HII sensitivity of the RDH signal in the chromosomal band in vitro, in vivo removal of RDHs by RNase HII apparently requires help of other enzymes, for example DNA pol I, that should be able to utilize nick generated by RNase HII. Of note, since RNase HII in vitro does not attack RNA:DNA hybrids lacking RNA-DNA junctions (unless Mn 2+ is substituted for Mg 2+ ), the in vivo requirement for RNase HII inactivation for maximal accumulation of RDH signal suggests a non-R-loop nature of RDHs. However, it could simply be that the RNA strand of R-loop may have patches of incorporated DNA nucleotides long enough to be recognized by RNase HII as proper substrates; RNA polymerases are known to have a limited discrimination against DNA precursors (70,71).
The chromosomal RDH structures are thermally stable, since we observe complete RDH signal disappearance only at 90 • C, when chromosomal dsDNA also completely denatures (Supplementary Figure S10C). Cutting chromosomal DNA with restriction nucleases reveals relatively smooth genomic distribution of the UV-induced RDH signal and estimated the average size of RDHs as 3-4 kb ( Figure 6B), --much longer than we anticipated, but consistent with their thermal stability. The length of an R-loop could be estimated with whole-genomic sequencing after sodium bisulfate treatment to turn cytosines into uracils in the single-stranded DNA displaced by the Rloop. Clusters of genes in the range of 2-10 kb with the increased bisulfate sensitivity were identified even for the WT E. coli (50).

The critical link between unexcised PD and RDH
The corollary of our model ( Figure 5A) is the proximity of UV-induced RDHs to the PDs that caused them. To identify the R-loop-aTECs blocked at PDs we first followed kinetics of PD removal from the chromosomal DNA using PD-specific antibodies. While the rnhAB mutants removes PDs from plasmid with WT kinetics (by 60 min post-UV) ( Figure 1DE), and PD removal from the chromosomal DNA in the rpoB35* rnhAB mutant and in WT cells is likewise complete by 60 min post-UV ( Figure 5C), we found that 10% of PDs still remains by 60 min post-UV in the chromosomal DNA of the rnhAB and mfd rnhAB mutants ( Figure 5C), when the RDH signal actively accumulates in these cells ( Figure 4H, I), potentially linking the RDH signal accumulation with the lingering PDs.
We developed a detection procedure to directly test this logic ( Figure 6A): using DRIP, we recovered the RDH-enriched DNA fragments, to subsequently analyze them for the density of PDs relative to either the input DNA, or the eluate DNA that failed to bind S9.6. The RDH-enriched fraction indeed showed the expected or higher PD densities relative to the eluate fraction, and both RDH and PD enrichments were reduced by RNaseHI treatment (Supplementary Figure S13A and 6E), supporting proximity of RDHs and PDs in the chromosomal DNA. In other words, even though UVinduced RDHs turned out to be quite long (lowering theoretically-possible enrichment), we still found DNA around them to be enriched for PDs, validating the proposed scheme ( Figure 6A).
One weak point of the DRIP enrichment procedure is that it critically depends on not only the high specificity of S9.6 antibody for RDHs, but also on the absence of recognition of DNA with PDs; unfortunately, we found that both are not exactly true for this antibody. First, our DRIP still captures substantial amount of DNA even from WT cells, which have almost no detectable RDH signal ( Figure 4A,B, H). Moreover, we could not reduce this background of non-specific dsDNA by pretreatment of protein-A beads with either glycogen or salmonsperm DNA, suggesting that it is coming directly from S9.6 binding to DNA. Second, and even more troublesome for our objective, we found that such DRIP enrichment of control DNA from RNase H + (WT) cells is further increased by UV-irradiation. Therefore, the observed small but significant RNase HI-resistant 'enrichment' of PD ( Figure 6E) was unfortunate, but not exactly unexpected. There are multiple reports that S9.6 can precipitate non-R-loop DNA structures, reducing the specificity of DRIP (57,72).
We eventually solved the problem of co-enrichment detection differently, by size-fractionating RDH and PD signals of the HaeII-cut chromosomal DNA and by showing that accumulation of the RDH signal happens exactly in the same size fraction (6-10 kb) in which PDs are protected from excision ( Figure 7). The procedure also reveals detectable RDH-PD co-enrichment in WT cells that happens in the same size fraction ( Figure 7A),--but RDHs are actively removed in these RNase H-proficient cells, ultimately making the transcription-stalling PDs vulnerable to excision repair, which resolves the potentially-lethal situation.

Conclusion
In this work, we present multiple lines of evidence to strongly implicate R-loop-aTECs as lethal R-lesions induced by UV in the rnhAB mutants, in two identifiable stages after the formation of PD. Indeed, the connection between R-loop and the PD has to be via transcribing RNA polymerase, while RDH amplification is induced only after arrival of replication forks ( Figure 7D). Thus, instead of a novel yet simple structure (like R-tract), Rlesions after UV turn out to have a complex structure and a fittingly-complex mechanism of formation. From a different perspective, UV light, as a DNA-damaging agent, continues to surprise, first with pyrimidine dimers and their nucleotide-excision repair, then with chromosome fragmentation and its recombinational repair, now with Rloop-aTECs and their disassembly by RNase H enzymes. The next question in this line of research should be the chromosomal position of UV-induced RDHs,--addressed by genomic techniques.