TUG1-mediated R-loop resolution at microsatellite loci as a prerequisite for cancer cell proliferation

Oncogene-induced DNA replication stress (RS) and consequent pathogenic R-loop formation are known to impede S phase progression. Nonetheless, cancer cells continuously proliferate under such high-stressed conditions through incompletely understood mechanisms. Here, we report taurine upregulated gene 1 (TUG1) long noncoding RNA (lncRNA), which is highly expressed in many types of cancers, as an important regulator of intrinsic R-loop in cancer cells. Under RS conditions, TUG1 is rapidly upregulated via activation of the ATR-CHK1 signaling pathway, interacts with RPA and DHX9, and engages in resolving R-loops at certain loci, particularly at the CA repeat microsatellite loci. Depletion of TUG1 leads to overabundant R-loops and enhanced RS, leading to substantial inhibition of tumor growth. Our data reveal a role of TUG1 as molecule important for resolving R-loop accumulation in cancer cells and suggest targeting TUG1 as a potent therapeutic approach for cancer treatment.

localization of the TUG1 lnc RNA is driven by intron retention (Dumbovic et al, Nat Commun, 2021, https://doi.org/10.1038. In the present paper, TUG1 cDNA with no intron was used to map the site of interactions with DHX9 and RPA32 in the nucleus. The authors should discuss about the possibility of the intronic sequence that is bound to TUG1. 4. It is reported that DHX9 is localized in the nucleolus functionally (Thacker et al, Sci Rep, 2020, https://doi.org/10.1038/s41598-020-75160-z, Leone et al, EMBO Rep, https://www.ncbi.nlm.nih.gov/pmc/articles/PMC5494521/) . Please comment on your TUG1-FISH and DHX9 IF data on this aspect. Did you see the nucleolar localization, in a population of cells?
Reviewer #2 (Remarks to the Author): In the manuscript entitled "TUG1-mediated R-loop resolution at microsatellite loci as a prerequisite for cancer cell proliferation", Suzuki and Kondo claim that upregulation of TUG1 resolved the problem of R-lop accumulation during DNA replication stress and led to increased tumor growth. In general, the individual data points should be displayed. There should be a quantification of WBs. The author should also highlight the biological replicate of experiments. The experimental models are inconsistent (for example, in Fig 2, U2OS was investigated, and in Fig. 3, Hela was used). The author should explain why the various cell lines were used for different experiments. Too many abbreviations were used, and the author did not provide the full name of some of the abbreviations. The gating of cell cycle experiments should be presented in all figures. There are plenty of visualizations. However, from my point of view, some simple tables with useful statistics are more informative. The claim that targeting TUG1 coupled with DNA damaging agents provides a novel strategy for the treatment of cancers is not well experimentally supported, since only one cancer cell line was investigated in in vivo treatment. The author studies the function of lncRNA. Which transcriptional variant of TUG1 did they focus on? Especially in a publication from Dumbovic et al. shows that TUG1 has a bimodal distribution of fully spliced cytoplasmic and intron-retained nuclear transcripts.
1. Fig. 1: They only used mitotic shake-off to obtain synchronized cell cultures. They should also use retrospective synchronization achieved by flow sorting to validate their fining. Especially to better detect and characterize the cells in S-Phase, the author should use BrdU or Edu staining for accurately measuring DNA replication. The investigated FxCycle Violet staining is only for bivariate analysis of total DNA content. 2. Fig. 1A shows the fold change of lncRNA expression after two hours of treatment. How many DMSO samples were used as a control? 3. Fig. S1A: could the author explain what mVenus-positive means? 4. Fig. S1A: could the author provide the statistical analysis of FACS data? 5. Fig. 1D: are all these data statistics non-significant? The author could use the subcellular fraction isolation and rt-qPCR to validate smFISH results. 6. Fig. S1F: how many times was siRNA KD processed? Since there is only one western blot figure without statistics, it looks like only one time KD was processed. 7. Fig. 1G: since ATR inhibitor VE-821 is investigated, WB of ATR expression should be added. 8. Fig. S1H showed the comparison between tumor and normal. Could the author clarify what is expected under the definition of normal cells (normal brain? Blood cells?...) 9. Fig. 2B: how did the author analyze co-localization? By counting. How many times did the author process the experiment? It looks like the investigation was processed only once, and multiple nuclei were counted. 10. Fig. 2C and E, where is the corresponding quantification? 11. Fig. S2A, more information about plasmids and developing stable overexpression should be provided in the method section. 12. Fig 3B, please provide the quantification of replicas. 13. Fig. 3E: in the main text, Fig. 3F should be changed to Fig. 3E. for the conclusion, the author needs to calculate the ratio since the input protein amount is inconsistent. 14. Fig. 4B: Why the DRB treatment was investigated? The author should explain the purpose of this experiment. 15. Fig. 4C: the experiment was repeated three times. Therefore, it should be only three data points in the figure. It is incorrect to show all 900 fibers and process the static analysis using 900 data points. 16. Fig. S4G: it should be a separate table demonstrating the percentage of different R-loop regions. 17. Fig. 6A and B: did the experiment process only one time? Please provide the static analysis. 18. Fig.6D: please provide the results of TUG1#2 as well. 19. Fig. 6E. which statistical analysis was used? It is significant only for 7 days. 20. Fig. 6G: increased H2AX percentage of cells doesn't directly indicate the increase of apoptosis. 21. Fig. 7: patients with GBM containing a methylated MGMT promoter benefited from TMZ treatment. What is the methylation status of MGMT in LN229?
Reviewer #3 (Remarks to the Author): Suzuki and colleagues report that the taurine upregulated gene lnRNA plays a role in the resolution of replication stress induced R-loops and propose that its downregulation might be employed as a potent therapeutic approach for treating cancer.
While a role for TUG1 in R-loop resolution would be novel, the case made here is not compelling. Nor was it proposed how TUG1 facilitates R-loop resolution other than proposed interaction with DHX9 that one assumes must be distinct from the mechanism through which DHX9 resolves other R-loops.
Specific criticisms: 1. Fig 1D plots  references Chk2 a downstream target of ATM? 3. A major criticism of this manuscript is the reliance on co-localization data. It is not clear what colocalization means here. Only few representative images are shown, often with foci false coloured with red and green. Colocalization should at least in some way be detected as a yellow colour resulting from the superimposed green and red foci. These are not that abundant with many green and red foci not colocalised and more often than not colocalization is only partial. More importantly, colocalization,, particularly when indicative of a physical interaction as suggested in the manuscript for TUG1 and DHX9, would result not only in a coincidence of the foci (i.e. in the same place) but also a codistribution (i.e. the shape and density of the two signals would vary co-ordinately). This does not seem to be the case. How was colocalization measured, what were the criteria used in scoring (by eye or by computer analysis?) and how was it analysed? This is critical if these data are to establish a compelling case for proper colocalization. Numerical data is not presented for all the colocalization studies (e.g. Fig 2E). 4. The interaction between TUG1 and RPA is performed after overexpression of RPA subunits (Fig 2D). However the authors already made the point that their observations are cell cycle dependent and the over-expression of RPA will disrupt this. The experiment should rely on the endogenous properly regulated RPA proteins 5. In Fig 2E the authors use a catalytically dead form of RNAseH-GFP to bind DNA:RNA hybrid. This tool is likely to detect hybrid generated as replication primers as well as R-loops. How much of the signal they measure is due to this? Does the signal go away if they knock down primase in the cells? 6. Depletion of TUG1 reduces the speed of replication elongation by a small but seemingly significant amount. What happens to replication fork progression if cells are induced for RS between the periods of labelling? i.e. in response to DNA damage treatment? The images in Suppl Fig 3C are not high quality as DNA fibres appear overlaid and difficult to measure. There seems to be an issue with combing. 7. Fig 4D suggests that the cell cycle is prolonged in cells depleted of TUG1 and that this is due to an extended S-phase. It would be preferable to see FACS plots of PI and BrdU incorporation as this gives a better measure of ongoing nucleotide incorporation and the completion of replication. Notwithstanding the lower replication progression in cells lacking TUG1, do the authors know that replication delay is caused by slower replication fork progression or whether the replicative issues cause a proportion of cells to arrest/ die during S-phase? The latter would fit with the data reported later showing proliferation assays for normal v TUG1 depleted cells. This might also be supplemented by survival curves in response to the RS inducing agents used in the study. 8. The authors label their cell cycle plots as cells that are 2n and 4n, which indicates diploid and tetraploid. Do they mean 2c and 4c which is a measure of diploid DNA content (non-replicated v replicated genomes) 9. Fig 4G -the authors propose that 'data indicated that TUG1 resolves R-loops via DHX9, which is induced by transcription'. This is an overstatement as it is based solely on a correlation. 10. The role of DHX9 in R loop metabolism is currently unclear with competing assessments of its role in R-loop resolution (DOI: 10.1016/j.celrep.2018.04.025 and DOI: 10.1038/s41467-018-06677-1). In both papers however, knockdown of DHX9 resulted in a decrease in global R-loops. Therefore it cannot necessarily be inferred that an interaction with DHX9 might promote R-loop resolution. More analysis is required including genomic approaches with depleted DHX9 and or expression of DHX9 helicase defective mutant. 11. The DRIP-seq experiments seem to be the strongest data and speak directly to the different localization and quantity of R-loops 12. The authors report that loss of TUG1 induces DNA damage and apoptosis. They measure damage in the form of DNA breaks with the phosphorylated form of histone H2AX or directly as a tail moment. What proportion of these DSB are specific RS induced DNA breaks and not DSB caused by fragmentation during apoptosis? 13. The authors suggest that treatment of glioblastoma cells with TMZ and depletion of TUG1 has a synergistic effect. While there is an increased effect this does not at all look like synergy, perhaps additive though.
Overall there are a number of issues to be addressed before the current manuscript gets close to establishing its hypothesis. At the moment the manuscript is structured as if the authors are trying to show a role for TUG1 in R-loop resolution, rather than critically evaluating whether or not it does.
We thank all three reviewers for their constructive comments. In order to address these important comments, we have performed many more experiments and found that the new data consistently support the original findings.

Reviewer #1 :
In this manuscript, the authors investigated the long non-coding RNA, TUG1, and found that it resolves R-loops, and interacts with RPA and DHX9 helicase at certain loci with specific repeats, in cancer cells. Depletion of TUG1 results in replication stress (RS), accumulation of R-loop and DNA damage, and apoptosis. Mouse xenograft experiments showed that a combination therapy with ati-TUG-1DDA and TMZ may be a promising therapy for glioblastoma. The findings are potentially interesting. The authors performed numerous experiments, and most were well-conducted. The paper is well written mostly. However, some critical experiments and explanations are missing, and interpretations are sometimes confusing. Several issues below should be addressed.
Thank you for the constructive comments. In order to address these important comments, we have performed additional experiments.
Major comments: 1. There is a difference of FISH data in the most part of this paper and their previous paper.
Their single molecule FISH data in this paper gives an impression that most TUG1 is in the nucleus. However, a regular FISH in Figure 7G and data in their previous paper (Katsushima et al, Nature Commun, 2016, https://www.nature.com/articles/ncomms13616) show that substantial TUG1 molecules are in the cytoplasm. What makes the difference? The authors should clarify the cellular distribution of TUG1. They should make a direct comparison of the nuclear and cytoplasmic TUG1 in their subcellular fractionation, in Fig.1D. Also, they should be careful about interpretation of direct and indirect cellular effects under the TUG1 knockdown.
TUG1 is generally located both in the nucleus and cytoplasm, as was reported in previous studies including ours (Cabili et al., 2015Genome Biol. DOI: 10.1186/s13059-015-0586-4, Katsushima et al., 2016. We found that newly transcribed TUG1 appeared particularly in the nucleus two hours after either CPT or HU treatment, followed by increased TUG1 molecules in the cytoplasm at four hours (Fig. 1C, 1D). The single molecule FISH data in Fig. 1C are representative images showing nuclear localization of TUG1 two hours after CPT or HU treatments (Fig. 1C, 1D). As suggested, we further performed a direct comparison of the nuclear and cytoplasmic TUG1 in subcellular fractions using qRT-PCR ( Supplementary   Fig. S1D).
As for direct and indirect cellular effects, we studied cells after four hours of TUG1 knockdown (direct effect) by Slot blot (Fig. 4B), DNA fiber assay (Fig. 4C), and DRIP-seq (Fig. 5). These experiments clearly showed that TUG1 KD induced R-loop accumulation and slowed fork progression (RS) within this time frame. Subsequently, after 24 hours of TUG1 KD, we observed both direct and indirect effects as a consequence of accumulated unresolved R-loops and RS, such as an increase of γ-H2AX signals (Fig. 6A, 6B), DNA fragmentation ( Fig. 6C), suppressed cell growth (Fig. 6D), and apoptosis (Fig. 6E). We have now explained that a short period of ASO treatment is useful to understand the direct effects of TUG1 KD in the Results section. Thank you for this helpful suggestion. In order to show TUG1 binding to the TUG1 KDsensitive site, we conducted CARPID-based ChIP-qPCR, because this assay represents the detection of a physiological interaction of TUG1 with the target sites. This assay showed that TUG1 and proteins in its proximity were significantly enriched at the TUG1-sensitive regions (BCL2 and C22orf34) ( Supplementary Fig. S6J). We have added these data to  We thank the reviewer for this important comment. In our DRIP-seq data, we recognized that the level of R-loop induced by co-treatment with TUG1-ASO and CPT is lower than with CPT treatment alone. This is probably due to a combination of TUG1 KD and 2 hours of CPT treatment decreasing cellular metabolism including transcription and DNA replication,  formation, DHX9/RPA binding, R-loop resolution, and apoptosis. It is confusing, because RS is caused by CPT/HU treatment (page 6, "…induced RS by treatment with hydroxyurea (HU) or camptothecin (CPT)", where TUG1 is overexpressed. In addition, RS is also caused by TUG1 depletion (Fig. 7A, Summary." Depletion of TUG1 lead to overabundant R-loops and enhanced RS.").
We are sorry for any confusion in our manuscript. As suggested, we have now added a model Minor comments: 1. The Summary starts with "Oncogene-induced DNA replication stress (RS)…". There is no experiment clearly and directly addressing the oncogene-induced DNA RS in this manuscript, therefore, it should be rewritten.
We corrected the first sentence accordingly in the Summary.
2. On page 16, the second from the last line: " Supplementary Fig. S1J" is not presented in the current manuscript.
We sincerely apologize for this and have corrected it.
3. A paper from Dr. Rinn's research group was published in Nat Commun, describing that nuclear localization of the TUG1 lncRNA is driven by intron retention (Dumbovic et al., Nat Commun, 2021, https://doi.org/10.1038/s41467-021-23221-w). In the present paper, TUG1 cDNA with no intron was used to map the site of interactions with DHX9 and RPA32 in the nucleus. The authors should discuss about the possibility of the intronic sequence that is bound to TUG1.
Thank you for this comment. This point was also raised by reviewer #2. Dumbovic et al.
showed that both fully spliced and intron-retained transcripts are present in the nucleus, while only spliced transcripts exist in the cytoplasm of HeLa cells. In the revised manuscript, we examined spliced and intron-retained transcripts by smFISH using a probe set targeting exon 2, which detects both spliced and intron-retained transcripts (ViewRNA Probe Set; Assay ID: VA1-11879, Thermofisher), and a probe set targeting both intron 1 and intron 2, which can detect only intron-retained transcripts (ViewRNA Probe Set; Assay ID: VF6-6000434, Thermofisher). After CPT treatment, R-loops (RNH1 D210N -GFP) colocalized with both fully spliced and intron-retained transcripts. These data suggest that exon regions are sufficient for the resolution of R-loops. However, further work is needed to define the functional difference between fully spliced and intron-retained transcripts. We have added these data and sentences to Supplementary

Reviewer #2
In the manuscript entitled "TUG1-mediated R-loop resolution at microsatellite loci as a prerequisite for cancer cell proliferation", Suzuki and Kondo claim that upregulation of TUG1 resolved the problem of R-lop accumulation during DNA replication stress and led to increased tumor growth. In general, the individual data points should be displayed. There We agree with this comment. To consistently perform experiments using HeLa cells, we have added data from newly established HeLa-RPA32-GFP and HeLa-RPA70-GFP cell lines for the RIP experiments in Fig. 2D. Now HeLa cells were used throughout the in vitro study. We also used U2OS cells in addition to HeLa cells particularly in DNA damage-related experiments because U2OS has been well characterized and widely used in the analysis of DNA repair due to its intact DNA damage response. Finally, we strove to apply our new findings more to the clinical context, particularly for glioma, a representative orphan disease and refractory cancer. For this, glioma cell lines LN229, T98G and U251MG, were investigated to assess the effectiveness of targeting TUG1 as a model for the treatment of glioma, in addition to the some of the key experiments related to TUG1 function in vitro. We have added these data and sentences to Fig. 2D, the Materials and Methods, and the Results.
O-5. Too many abbreviations were used, and the author did not provide the full name of some of the abbreviations.
We apologize for this inconvenience. We rechecked the text and made sure that abbreviations are fully spelled out when they first appear in the text. In addition, a list of abbreviations was also added as Supplemental Table S8. Thank you for this comment. According to the suggestion, in addition to the super-resolution microscopy visualizations, we added a summary of the results to Supplementary Table S2. O-8. The claim that targeting TUG1 coupled with DNA damaging agents provides a novel strategy for the treatment of cancers is not well experimentally supported, since only one cancer cell line was investigated in in vivo treatment.
We agree with this comment. We additionally conducted therapy with antiTUG1-DDS and TMZ in the U251MG glioma xenograft mouse model. This combination therapy most effectively suppressed U251MG tumor growth, as was also seen in LN229 glioma cells. We added these data and sentences to Supplementary showed that both fully spliced and intron-retained transcripts are present in the nucleus, while only spliced transcripts exist in the cytoplasm of HeLa cells. In the revised manuscript, we examined spliced and intron-retained transcripts by smFISH using a probe set targeting exon 2, which detects both spliced and intron-retained transcripts (ViewRNA Probe Set; Assay ID: VA1-11879, Thermofisher), and a probe set targeting both intron 1 and intron 2, which can detect only intron-retained transcripts (ViewRNA Probe Set; Assay ID: VF6-6000434, Thermofisher). After CPT treatment, R-loops (RNH1 D210N -GFP) colocalized with both fully spliced and intron-retained transcripts. These data suggest that exon regions are sufficient for the resolution of R-loops. However, further work is needed to define the functional difference between fully spliced and intron-retained transcripts. We have added these data and sentences to Supplementary Fig. S2D, Material and Methods, Results, and Discussion.
1. Fig. 1: They only used mitotic shake-off to obtain synchronized cell cultures. They should also use retrospective synchronization achieved by flow sorting to validate their fining.
Especially to better detect and characterize the cells in S-Phase, the author should use BrdU or Edu staining for accurately measuring DNA replication. The investigated FxCycle Violet staining is only for bivariate analysis of total DNA content.
Thank you for this important comment. As pointed out, we synchronized the HeLa/Fucci2 cells by mitotic shake off and used the cells for analysis after 13 hours. In order to better detect and characterize the cells in S-Phase, we used EdU staining for accurately measuring DNA replication along with FxCycle Violet-based flow cytometric analysis ( Supplementary   Fig. S1A). We validated the finding that after 13 hours of mitotic shake off, EdU-positive cells (i.e., in S phase) were significantly enriched by this method (>90%). We have added these data and sentences to Supplementary Fig. S1A, Materials and Methods, and the Results.
2. Fig. 1A shows the fold change of lncRNA expression after two hours of treatment. How many DMSO samples were used as a control?
We are sorry for this inadequate explanation. Cells were treated with either DMSO, HU, or CPT and examined in duplicate. We have corrected these sentences in Materials and Methods and Supplementary Table S1. This comment is related to your comment #1. In the revised manuscript, we showed that after 13 hours of mitotic shake off, a majority of cells (>90%) was EdU-positive (S phase). We have added the statistical analysis to Supplemental Fig. S1A and explanatory sentences to Materials and Methods. 5. Fig. 1D: are all these data statistics non-significant? The author could use the subcellular fraction isolation and rt-qPCR to validate smFISH results.
Thank you for this suggestion. We have statistically analyzed the data and added the results to Fig. 1D. We have also validated the smFISH results with RT-qPCR after isolating subcellular fractions in Supplementary Fig. S1D. 6. Fig. S1F: how many times was siRNA KD processed? Since there is only one western blot figure without statistics, it looks like only one time KD was processed.
We apologize for this insufficient presentation. We have performed this series of experiments in triplicate. We have now added quantifications of the gel images with statistical analysis to Supplementary Fig. S1G (originally Fig. S1F) in the revised version. 7. Fig. 1G: since ATR inhibitor VE-821 is investigated, WB of ATR expression should be added.
We agree with this comment. The Western blotting of p-ATR and ATR has been added to 9. Fig. 2B: how did the author analyze co-localization? By counting. How many times did the author process the experiment? It looks like the investigation was processed only once, and multiple nuclei were counted. This is an important point and was also raised by reviewer #3. We have analyzed superresolution images using ImageJ software. This software contains a function "Analyze Particles" that extracts the contours of the particle at each wavelength. Then we automatically counted the number of particles and particles that overlapped with other particles in different colors. We have added this information to the Materials and Methods. We have conducted the super-resolution microscopy analysis in triplicate. Four to 10 cells were examined in individual experiments. Fig. 2B, 2C, and 2E show data from a representative single experiment. The independent experimental data are summarized in Supplementary Table S2. We added information on the biological replicates and the number of examined cells to the figure legends.
10. Fig. 2C and E, where is the corresponding quantification?
We have added the quantitative analysis of the co-localization to Fig. 2C and 2E. 11. Fig. S2A, more information about plasmids and developing stable overexpression should be provided in the method section.
We have added this information to the Materials and Methods.

Fig 3B, please provide the quantification of replicas.
We have added quantification of the Western blotting signals to Fig. 3B. 13. Fig. 3E: in the main text, Fig. 3F should be changed to Fig. 3E. for the conclusion, the author needs to calculate the ratio since the input protein amount is inconsistent.
We sincerely apologize for this typo. We also quantitatively analyzed the Western blots throughout the manuscript including Fig. 3E. 14. Fig. 4B: Why the DRB treatment was investigated? The author should explain the purpose of this experiment. 15. Fig. 4C: the experiment was repeated three times. Therefore, it should be only three data points in the figure. It is incorrect to show all 900 fibers and process the static analysis using Thank you for this comment. According to the suggestion, we now show the data from a single representative experiment from four cell lines in Fig. 4C. In addition, each triplicate from the DNA fiber assay is summarized in Supplementary Table S4. The statistical analysis is added for the independent experiments. 16. Fig. S4G: it should be a separate table demonstrating the percentage of different R-loop regions.
Thank you for this comment. We have added the data to Supplementary Table S5 in addition to Supplementary Fig. S6F (originally Fig. S4G). Fig. 6A and B: did the experiment process only one time? Please provide the static analysis.

17.
We quantified band densities of the Western blot gel images, which were obtained from three independent experiments (i.e., in triplicate), and are shown as a bar graph with statistics ( Fig.   6A). We also performed flow cytometry in triplicate and statistically analyzed the data (Fig.   6B).
We have added neutral comet assay data using TUG1#2 ASO to Supplementary Fig. S8D. 19. Fig. 6E. which statistical analysis was used? It is significant only for 7 days.
In the original manuscript, we conducted statistical analyses only on day seven using the Student's t-test. In the revised version, we performed statistical analyses on all the proliferation curves. We performed global curve-fitting by nonlinear regression analysis (Exponential Malthusian growth) using GraphPad Prism 9.4.1 software. The best-fit k (the rate constant) value was calculated for each data set (the growth of cells treated with Ctrl ASO, TUG1#1, and TUG1#2), carried out in triplicate. One-way ANOVA and Tukey's multiple comparison tests were newly used for statistical analysis. We have added a sentence to the Materials and Methods and new statistics to Fig. 6D (Fig. 6E in the original version). 20. Fig. 6G: increased H2AX percentage of cells doesn't directly indicate the increase of apoptosis.
We agree with this comment. Accordingly, we have added the data showing that 48 hours of TUG1 depletion did not induce apoptosis in TIG3 to Fig. 6F.  Cells were seeded at 1 × 10 3 cells/well into 96-well plates and treated with different concentrations of TMZ for 7 days. Cell viability was assessed using Cell Count Reagent SF (Nacalai Tesque, Japan).

Reviewer #3
Suzuki and colleagues report that the taurine upregulated gene lnRNA plays a role in the resolution of replication stress induced R-loops and propose that its downregulation might be employed as a potent therapeutic approach for treating cancer.
While a role for TUG1 in R-loop resolution would be novel, the case made here is not compelling. Nor was it proposed how TUG1 facilitates R-loop resolution other than proposed interaction with DHX9 that one assumes must be distinct from the mechanism through which DHX9 resolves other R-loops.
Thank you for these important comments. To the extent possible, these have been addressed by the addition of new data.
Specific criticisms: 1. Fig 1D plots  The results of single molecule FISH in Fig. 1D showed a value relative to the median number of TUG1 molecules in the S/G2 nucleus at 0 h. We apologize for this and have now corrected the figure legends and added the significance assessments.
We thank the reviewer for this insightful question about TUG1 foci and RS sites. After CPT treatment, around half of TUG1 foci colocalized with R-loops (RNH1 D210N -GFP). Proliferating cell nuclear antigen (PCNA), which is a co-factor of DNA polymerase-delta (Bravo et al., Nature, 1987, DOI: 10.1038 showed striking colocalization with about half of the TUG1/RNH1 D210N -GFP foci ( Supplementary Fig. S2B) (Please also see comment #5). We also labeled newly synthesized DNA with EdU to visualize the locations where replication occurs. TUG1 foci also colocalize with EdU signals (Supplementary Fig. S2C) (Please also see comment #5). In view of the fact that TUG1 foci colocalize with pRPA32 ( Fig. 2A), these data indicate that TUG1 is located at the RS site, which consists of a stalled fork and an Rloop. We have added the new data and a sentence to Supplementary Fig. S2C, Fig. S2D  We agree with this comment. The Western blotting of p-ATR/ATR and p-ATM/ATM has been added to Fig. 1G. and Supplementary Fig. S8C, respectively. 3. A major criticism of this manuscript is the reliance on co-localization data. It is not clear what co-localization means here. Only few representative images are shown, often with foci false coloured with red and green. Co-localization should at least in some way be detected as a yellow colocv ur resulting from the superimposed green and red foci. These are not that abundant with many green and red foci not colocalised and more often than not colocalization is only partial. More importantly, co-localization,, particularly when indicative of a physical interaction as suggested in the manuscript for TUG1 and DHX9, would result not only in a coincidence of the foci (i.e. in the same place) but also a co-distribution (i.e. the shape and density of the two signals would vary co-ordinately). This does not seem to be the case. How was co-localization measured, what were the criteria used in scoring (by eye or by computer analysis?) and how was it analysed? This is critical if these data are to establish a compelling case for proper co-localization. Numerical data is not presented for all the colocalization studies (e.g. Fig 2E).
Thank you for this criticism. This important point was also raised by reviewer #2. STORM microscopy has a high resolution of less than 40 nm, based on detecting single photoswitchable fluorophores in a labeled sample in time and precisely localizing them at much higher optical resolution than can be attained by confocal imaging systems (more than Based on the aforementioned idea, we have analyzed super-resolution images using ImageJ software. This software contains a function "Analyze Particles" that extracts the contours of the particle at each wavelength. Then the number of particles and particles that overlapped with other particles in different colors are automatically counted. We have added this information to the Materials and Methods. Quantification of the co-localization analysis in Fig. 2C and 2E has also been added. Finally, in addition to the super-resolution microscopy visualizations, we added a summary of the results to Supplementary Table S2. 4. The interaction between TUG1 and RPA is performed after overexpression of RPA subunits (Fig 2D). However the authors already made the point that their observations are cell cycle dependent and the over-expression of RPA will disrupt this. The experiment should rely on the endogenous properly regulated RPA proteins.
Thank you for this comment. According to this suggestion, we have tested the four antibodies against RPA70 and four against RPA32 listed below, but none of these antibodies worked in the RIP assay. However, although our RIP experiments used overexpressed RPA subunits, we detected a consistent interaction between TUG1 and endogenous RPA32 by the alternative CARPID technology in vivo ( Supplementary Fig. S3D). We also validated the interaction between endogenous pRPA32 and TUG1 by CLIP assay (Fig. 3D). Therefore we believe that TUG1 does indeed interact with RPA. hybrid. This tool is likely to detect hybrid generated as replication primers as well as R-loops.
How much of the signal they measure is due to this? Does the signal go away if they knock down primase in the cells?
Thank you for this comment. According to the suggestion, we knocked down the primases PRIM1 and PRIM2, together with RNH1 D210N -GFP expression. However, unfortunately, these treatments proved lethal to the cells, despite testing several different conditions. In response to comment #1, we showed that TUG1/RNH1 D210N -GFP foci colocalize with PCNA and EdU, indicating that TUG1 is located at stalled forks ( Supplementary Fig. S2B, S2C). Consistent with this, RNH1 D210N -GFP spots were clearly increased by CPT treatment, which induces Rloops (Fig. 2E). As the reviewer commented, RNaseH is involved in removal of RNA primers from Okazaki fragments during DNA replication. However, the length of RNA primers is generally short, about 10-20 nt. Therefore, the signal intensity of RNaseH-GFP at the RNA primers might be weaker relative to that at R-loops, the length of which is generally more than a few kb.
6. Depletion of TUG1 reduces the speed of replication elongation by a small but seemingly There seems to be an issue with combing.
According to this comment, we strove to improve the combing process and count only those fibers that did not overlay ( Fig. 4C and Supplementary Fig. S4C) 7. Fig 4D suggests that the cell cycle is prolonged in cells depleted of TUG1 and that this is due to an extended S-phase. It would be preferable to see FACS plots of PI and BrdU incorporation as this gives a better measure of ongoing nucleotide incorporation and the completion of replication. Notwithstanding the lower replication progression in cells lacking TUG1, do the authors know that replication delay is caused by slower replication fork progression or whether the replicative issues cause a proportion of cells to arrest/ die during S-phase? The latter would fit with the data reported later showing proliferation assays for normal v TUG1 depleted cells. This might also be supplemented by survival curves in response to the RS inducing agents used in the study.
Thank you for this comment. Four hours after TUG1 knock down (ASO transfection), we had shown R-loop accumulation (Slot blot in Fig. 4B and DRIP-seq in Fig. 5) and slowed fork progression (DNA fiber assay in Fig. 4C) in the original manuscript. According to the above suggestion, we further performed FCM analysis of DNA content and EdU incorporation after four hours of TUG1 KD (Supplementary Fig. S4D-G). The FCM analysis showed slight but significant reduction of EdU incorporation at this time ( Supplementary Fig. S4D, Fig. S4E).
Consequently, the cells showed an increase of γ-H2AX signals from 6 hours after TUG1 KD ( Fig. 6A), DNA fragmentation at 24 h (Fig. 6C), and apoptosis at 48 h (Fig. 6E). Consistent with this, after 24 hours of TUG1 KD, EdU incorporation was dramatically decreased ( Supplementary Fig. S4F, Fig. S4G). These data indicate that replication delay is mainly caused by slowed replication fork progression in the early response to TUG1 depletion (i.e. after four hours), but at a later time point, in addition to the replication delay, a proportion of cells undergo arrest and death during S phase, caused by DNA damage, that contributes to suppressing cell growth (Fig. 6D).
8. The authors label their cell cycle plots as cells that are 2n and 4n, which indicates diploid and tetraploid. Do they mean 2c and 4c which is a measure of diploid DNA content (nonreplicated v replicated genomes) Thank you for this helpful comment. We have corrected the labels in Fig. 4D, 6B, 6F, Supplementary Fig. S1A, S4D, S4H, S8B accordingly.
9. Fig 4G -the authors propose that 'data indicated that TUG1 resolves R-loops via DHX9, which is induced by transcription'. This is an overstatement as it is based solely on a correlation.  (Fig. 4F), while tethering the deletion mutant of TUG1-PP7, which DHX9 cannot bind, did not suppress R-loop formation (Fig. 4G). Furthermore, depletion of DHX9 increased the amount of R-loops in this experimental setting (Fig. 4H). Consistent with this, we found that depletion of DHX9 increased R-loops at the TUG1-sensitive CA repeat regions BCL2 and C22orf34 ( Supplementary Fig. S5F). These data strongly suggest that DHX9 11. The DRIP-seq experiments seem to be the strongest data and speak directly to the different localization and quantity of R-loops Thank you for this encouraging comment. Our DRIP-seq data indicated that TUG1 specifically functions to resolve R-loops in CA repeat regions. In the revised version, we further report that DHX9 KD increased R-loop accumulation in TUG1-sensitive loci ( Supplementary Fig. S6I). We propose that the TUG1-RPA-DHX9 interaction is a novel lncRNA-mediated mechanism for regulating R-loops at certain loci. We have added sentences to that effect to the Discussion.
12. The authors report that loss of TUG1 induces DNA damage and apoptosis. They measure damage in the form of DNA breaks with the phosphorylated form of histone H2AX or directly as a tail moment. What proportion of these DSB are specific RS induced DNA breaks and not DSB caused by fragmentation during apoptosis?
This is an important point. In the original manuscript, we carefully excluded cells at a late apoptosis stage from the measurements in the neutral comet assay by ignoring small comet heads and enlarged tails. In the revised version, we also depleted TUG1 with/without inhibition of apoptosis by the caspase inhibitor, Z-VAD-FMK, for 24 hours, and then performed neutral comet assays. No significant difference in the tail moment between cells treated with or without Z-VAD-FMK was observed ( Supplementary Fig. S8E). Therefore, we conclude that the DSBs we detected were mostly induced by RS, and not by apoptosis. We have added these data and a sentence to Supplementary Fig. S8E Thank you for the constructive comments. We now believe that our new data convincingly support the original findings.

REVIEWER COMMENTS
Reviewer #1 (Remarks to the Author): The authors satisfactory addressed all of the points I raised.
There is only one subtle point left, which is that they missed including the label "Supplementary Figure  11" in a list on the cover page of their supplementary materials file.
Basically, I suggest that the revised version is eligible for acceptation.
Reviewer #2 (Remarks to the Author): The reversed paper thoroughly answered all my questions. Significantly, the author strongly updated the quality of all figures with suitable statistical analyses and clear explanations. The reversed version of the manuscript presents a high-quality science that meets Nature communication's quality.
Reviewer #3 (Remarks to the Author): The authors have responded to specific points raised. My responses are below.
1. Fig 1D plots Fig. 1D showed a value relative to the median number of TUG1 molecules in the S/G2 nucleus at 0 h. We apologize for this and have now corrected the figure legends and added the significance assessments. We thank the reviewer for this insightful question about TUG1 foci and RS sites. After CPT treatment, around half of TUG1 foci colocalized with R-loops (RNH1D210N-GFP). Proliferating cell nuclear antigen (PCNA), which is a co-factor of DNA polymerase-delta (Bravo et al., Nature, 1987, DOI: 10.1038/326515a0) showed striking colocalization with about half of the TUG1/RNH1D210N -GFP foci ( Supplementary Fig. S2B) (Please also see comment #5). We also labeled newly synthesized DNA with EdU to visualize the locations where replication occurs. TUG1 foci also colocalize with EdU signals ( Supplementary Fig. S2C) (Please also see comment #5). In view of the fact that TUG1 foci colocalize with pRPA32 ( Fig. 2A), these data indicate that TUG1 is located at the RS site, which consists of a stalled fork and an Rloop. We have added the new data and a sentence to Supplementary Fig. S2C, Fig. S2D 3. A major criticism of this manuscript is the reliance on co-localization data. It is not clear what co-localization means here. Only few representative images are shown, often with foci false coloured with red and green. Co-localization should at least in some way be detected as a yellow colocv ur resulting from the superimposed green and red foci. These are not that abundant with many green and red foci not colocalised and more often than not colocalization is only partial. More importantly, co-localization,, particularly when indicative of a physical interaction as suggested in the manuscript for TUG1 and DHX9, would result not only in a coincidence of the foci (i.e. in the same place) but also a co-distribution (i.e. the shape and density of the two signals would vary co-ordinately). This does not seem to be the case. How was co-localization measured, what were the criteria used in scoring (by eye or by computer analysis?) and how was it analysed? This is critical if these data are to establish a compelling case for proper co-localization. Numerical data is not presented for all the colocalization studies (e.g. Fig 2E).
Thank you for this criticism. This important point was also raised by reviewer #2. STORM microscopy has a high resolution of less than 40 nm, based on detecting single photoswitchable fluorophores in a labeled sample in time and precisely localizing them at much higher optical resolution than can be attained by confocal imaging systems (more than 200 nm). We employed super-resolution microscopy to investigate molecular interactions between TUG1 and proteins. Based on the aforementioned idea, we have analyzed super-resolution images using ImageJ software. This software contains a function "Analyze Particles" that extracts the contours of the particle at each wavelength. Then the number of particles and particles that overlapped with other particles in different colors are automatically counted. We have added this information to the Materials and Methods. Quantification of the co-localization analysis in Fig. 2C and 2E has also been added. Finally, in addition to the super-resolution microscopy visualizations, we added a summary of the results to Supplementary Table S2.
Obviously, a certain amount of image analysis needs to be taken on trust. However, the amount of colocalization in the graphs does not seem to align with visual inspection of the images. For example in Fig 2 none of the images look like there is 60% colocalization of pRPA with Tug1.
4. The interaction between TUG1 and RPA is performed after overexpression of RPA subunits (Fig 2D). However the authors already made the point that their observations are cell cycle dependent and the over-expression of RPA will disrupt this. The experiment should rely on the endogenous properly regulated RPA proteins.
Thank you for this comment. According to this suggestion, we have tested the four antibodies against RPA70 and four against RPA32 listed below, but none of these antibodies worked in the RIP assay. However, although our RIP experiments used overexpressed RPA subunits, we detected a consistent interaction between TUG1 and endogenous RPA32 by the alternative CARPID technology in vivo ( Supplementary Fig. S3D). We also validated the interaction between endogenous pRPA32 and TUG1 by CLIP assay (Fig. 3D). Therefore we believe that TUG1 does indeed interact with RPA.
That is a pity. CARPID is a proximity labelling system and less convincing. Why would no RPA32 antibody work with RIP. RPA is not a particularly limiting protein. Fig 2E the authors use a catalytically dead form of RNAseH-GFP to bind DNA:RNA hybrid. This tool is likely to detect hybrid generated as replication primers as well as R-loops. How much of the signal they measure is due to this? Does the signal go away if they knock down primase in the cells?

In
Thank you for this comment. According to the suggestion, we knocked down the primases PRIM1 and PRIM2, together with RNH1D210N-GFP expression. However, unfortunately, these treatments proved lethal to the cells, despite testing several different conditions. In response to comment #1, we showed that TUG1/RNH1D210N -GFP foci colocalize with PCNA and EdU, indicating that TUG1 is located at stalled forks ( Supplementary Fig. S2B, S2C). Consistent with this, RNH1D210N -GFP spots were clearly increased by CPT treatment, which induces Rloops (Fig. 2E). As the reviewer commented, RNaseH is involved in removal of RNA primers from Okazaki fragments during DNA replication. However, the length of RNA primers is generally short, about 10-20 nt. Therefore, the signal intensity of RNaseH-GFP at the RNA primers might be weaker relative to that at R-loops, the length of which is generally more than a few kb. I am still not convinced of RNH1D210N -GFP as a tool, used in isolation of other robust mechanisms. I have seen knockdown of primases with RNH1D210N-GFP expression reported at a meeting where quite a lot of GFP signal was lost. I'm not sure if that study was published yet.
6. Depletion of TUG1 reduces the speed of replication elongation by a small but seemingly significant amount. What happens to replication fork progression if cells are induced for RS between the periods of labelling? i.e. in response to DNA damage treatment? The images in Suppl Fig 3C are not high quality as DNA fibres appear overlaid and difficult to measure. There seems to be an issue with combing.
According to this comment, we strove to improve the combing process and count only those fibers that did not overlay ( Fig. 4C and Supplementary Fig. S4C depletion induces DNA damage caused by accumulated R-loops, reducing the speed of replication elongation (please also see the following comment #7). In the revised version, we now explain the cause of the slowdown of replication fork progression by TUG1 depletion in the Results section.
It is not clear for the fibre assay why the authors have used both CldU and IdU. The whole point of using two nucleotides is to incorporate a treatment between the two to see the effect of a treatment on fork speed etc. so that altered speed will be given by ratio of green over red. i.e. It is an internal control.
General issue