Stochastic variation in the FOXM1 transcription program mediates replication stress tolerance

  1. Hendrika A. Segeren
  2. Kathryn A. Wierenga
  3. Frank M. Riemers
  4. Elsbeth A. van Liere
  5. Bart Westendorp

Reviewed by

Read the full article See related articles

Listed in

Log in to save this article

Abstract

Oncogene‐induced replication stress (RS) is a vulnerability of cancer cells that forces reliance on the intra‐S‐phase checkpoint to ensure faithful genome duplication. Inhibitors of the intra‐S‐phase checkpoint kinases ATR and CHK1 have been developed, but resistance to these drugs remains problematic. Understanding drug tolerance mechanisms is impeded by analysis of bulk samples, which neglect tumor heterogeneity and often fail to accurately interpret cell cycle‐mediated resistance. Here, by combining intracellular immunostaining and single‐cell RNA‐sequencing, we characterized the transcriptomes of oncogenic RAS‐expressing cells with variable levels of RS when challenged with a CHK1 inhibitor combined with gemcitabine. We identified 37 genes differentially expressed between tolerant and sensitive cells, including several FOXM1 targets. While complete knockdown of FOXM1 impeded cell proliferation, partial knockdown protected cells against DNA damage, and improved recovery from drug‐induced RS. Remarkably, knockdown of individual FOXM1 target genes UBE2C and MKI67 also mitigated DNA damage, uncovering unanticipated roles for these in the replication stress response. Our results suggest that low levels of FOXM1‐dependent gene expression during S and G2 phase protects cells against excessive DNA damage during drug‐induced replication stress.

Article activity feed

  1. Version published to 10.1002/1878-0261.13819
  2. Review Commons

    Note: This response was posted by the corresponding author to Review Commons. The content has not been altered except for formatting.

    Learn more at Review Commons

    Reply to the reviewers

    We want to thank both reviewers for their thorough and constructive review of our manuscript. Below, we have re-iterated their comments followed by an explanation of how we have revised the manuscript to address this.

    Reviewer #1 (Evidence, reproducibility and clarity (Required)):

    This manuscript presented by Segeren et al. applied an interesting HRASG12V inducible cell model to study the mechanism of cellular resistance to replication stress inducing agents. They also employed a novel reversible fixation technique which allows them to FAC sort cells according to their replication stress levels before applying single cell sequencing analysis to the same cell populations. By comparing cells with low levels of replication stress to cells with high levels of replication stress, they found that reduction in gene expression of FOXM1 target genes potentially protects cells against replication stress induced by CHK1i plus gemcitabine combination. Overall, this is a very interesting study. However, the following points should be addressed prior to publication:

    Major:

    1. Figure 3E and 3F showed two lists of differentially expressed genes in γH2Ax low cells. However, instead of arbitrarily extracting the FOXM1 target genes and TP53 targeted genes, it would be appreciated if the author could perform an unbiased and unsupervised gene set enrichment analysis such as Enrichr.

    As recommended, we performed an enrichment analysis using Enrichr to identify transcriptional programs associated with the we used the genes that were downregulated in the γH2AX-low cells. FOXM1 appeared as a prominent hit in different databases (both experimental and computational). We have included the lists of differentially expressed genes as an additional supplemental table (Table S1) and have included the Enrichr results as Table S3 (i.e. CHEA and ENCODE). We have described our results in lines 198-200 of the revised manuscript.

    1. At the experiment design stage, the authors also included HRASG12V status as a test condition because they previously found that HRASG12V mutation induces basal level replication stress and they would like to include this condition to study the adaptation to replication stress (line 110). However, the difference in HRASG12V negative and HRASG12V positive cells was not followed up in the later part of the paper. Can they show lists of differentially expressed genes identified under HRASG12V negative conditions as well (in the same format of Figure 3E and 3F) and comment on the differences as well?

    In the original manuscript, we included heatmaps of differentially expressed genes in the control cells in Figure S2. For improved clarity, we have modified this figure so that the heatmaps are labeled "Control cells". In the revised manuscript, we have also included Table S2, which lists the differentially expressed genes between yH2AX low and yH2AX high control cells, and Table S3, which lists the Enrichr results obtained based on these gene lists.

    We observed FOXM1 target genes in both the control and HRASG12V cells. Thus, the mechanism we identify does not appear to be specific to oncogenic Ras expression. We discuss this in lines 221-225. Because there were no other notable differences between the gene sets, we do not focus on this in the manuscript.

    1. In line 194 and in Figure S2B, the authors claimed that ANLN, HMGB2, CENPE, MKI67, and UBE2C demonstrated co-expression, but other genes displaying similar correlation scores were not commented (such as F3, CYR61, CTGF, etc). To avoid being biased at the analysis stage, the authors should define clearly what the cut-off of correlation score is and why only co-expression of ANLN, HMGB2, CENPE, MKI67, and UBE2C were mentioned.

    As suggested, we explain now in the revised manuscript that we focused on gene clusters consisting of at least 3 genes, that had a correlation coefficient greater than or equal to 0.4 with at least one other gene within the clusters. This cutoff is typically defined as representing a "moderate to good" correlation in biological data (Overholser, Sowinski, 2008). To make clear which clusters correlating gene sets passed these criteria, we have also highlighted these genes in Figure S3B. This returned the cluster we had already identified as FOXM1 targets, and as well spotted by the reviewer, a larger cluster which included* F3, CYR61, CTGF, SERPINE1, ANKRD1, KRTAP2-3, UGCG, and AMOTL*. Our Enrichr analysis did not identify any putative transcription factors linking the genes in this larger cluster. We are still interested to identify the putative transcription regulation mechanism linking these genes in future studies, but this is beyond the scope of the current manuscript. We have described these observations in lines 211-218.

    1. In line 215, instead of validating CENPE, UBE2C, HMGB2, ANLN, and MKI67 individually, the authors decided to validate FOXM1 instead, because they believe all the aforementioned genes are targets of FOXM1, therefore, validating FOXM1 alone would suffice. Again, this makes the validation process also biased. CENPE, UBE2C, HMGB2, ANLN, and MKI67 should be validated individually because they might sensitize cells to replication stress via different mechanisms. Besides, if all these genes were identified together because they are FOXM1 target genes, why did the authors not identify FOXM1 itself as a differentially expressed gene from the single cell sequencing? The sequencing only analyzed the S/G2/M cells, expression of FOXM1 should be detected easily.

    We agree with the reviewer that the omission of individual FOXM1 target genes in the validation process makes a biased impression. Therefore we ordered siRNAs against CENPE, UBE2C, HMGB2, ANLN, and MKI67. Similar to the other DE genes in the original mini-screen we first knocked down these genes using the siRNA Smartpools (pools of 4 individual siRNAs against each genes). Here, we observed a decrease in γH2AX signal compared to drug-treated cells transfected with all 5 Smartpools compared to drug-treated cells transfected with control siRNA. We next moved on to the deconvolution step of the screen, where we transfected cells with 4 individual siRNA against each gene. Here, we observed inconsistent effects of ANLN, CENPE, and HMGB2 when comparing the individual siRNAs, which all produce efficient knockdown of their target genes. But interestingly, for both MKI67 and UBE2C, each of the 4 individual siRNAs similar decreased yH2AX signal, though it was not as strong as the decrease observed when FOXM1 is knocked out. Understanding the exact mechanism of how MKI67 and UBE2C reduce replication stress is beyond the scope of this paper, but we hypothesize that, as with FOXM1, it is likely linked to their role in promoting progression through the cell cycle. These results are shown in Figures S5, and we mention these remarkable findings in the revised abstract and discuss these in the light of the recent literature in the Discussion section (lines 275-286).

    Then, we also addressed the comment about FOXM1 not being changed in the single cell RNA-seq analysis. We could indeed readily detect FOXM1 expression our single-cell RNA sequencing data. The difference in expression did not change significantly in cells sorted according to γH2AX level (Figure 4C). Because FOXM1 is highly regulated post-translationally, we hypothesized that an increase in the (active) protein is correlated to increased replication stress rather than transcript levels. This was indeed the case and we further explain our experiment to test this hypothesis in response to Point #6 (results are displayed in Figure 4D and described in lines 201-209).

    1. As pointed out by the author in the Discussion, single cell sequencing is not good at differentiating the causes from the consequences. The author tried to validate many of the differentially expressed genes in γH2Ax low cells. However, the fact that only FOXM1 knockdown passed the validation and deconvolution pointed out that the great majority of the identified genes are not the cause of the sensitivity change to replication stress inducing agents but likely the consequences. Therefore, in Figure S2C and S2D, it would be better that the authors could just name the genes as 'downregulated genes' in Figure S2C and 'upregulated genes' in Figure S2D. Taking into consideration that the expression change in the great majority of these genes are just consequences of sensitivity change to replication stress, defining them as 'potentially sensitizing' genes and 'potentially conferring resistance' genes is rather misleading.

    We agree that the way we originally labeled these plots may have been misleading. We have renamed then to "Downregulated in yH2AXlow" and "Upregulated in yH2AXlow", as recommended by the reviewer.

    1. To better prove that FOXM1 is the leading cause of the sensitivity to CHK1i+Gemcitabine induced replication stress, can the authors show the FOXM1 expression status in the tolerant cell population identified in Figure 1B (lowest panel)? Alternatively, can they plot FOXM1 expression level in the same tSNE plots shown in Figure 3B to 3D to see whether some of the γH2Ax low populations also show reduced FOXM1 expression?

    *FOXM1 *expression levels were not increased with gH2AXhigh versus gH2AXlow HRASG12V cells in the single cell RNA-sequencing data (Figure 4C in revised manuscript). However, as mentioned in our answer to point #4 we performed an additional experiment, which showed a strong positive correlation between phospho-FOXM1 and γH2AX (as measured by flow cytometry) in S-phase cells (Figure 4D). This indicates that the active form of the FOXM1 indeed increases as yH2AX levels increase, consistent with the observed increase in FOXM1 target genes. These results are described in lines 201-209.

    1. Clonogenic survival assay in Figure 4D was not quantified properly in Figure 4E. To rule out the siFOXM1 mediated growth/survival defects and to only focus on the siFOXM1 mediated resistance to CHK1i+Gemcitabine, the survival rate (intensity percent in this case) of CHK1i+Gemcitabine treated condition should be normalized against the survival rate of the Vehicle condition. E.g., the intensity percent of the siSCRAMBLE after treatment should be divided by the intensity percent of the untreated siSCRAMBLE; the intensity percent of the si#1 after treatment should be divided by the intensity percent of the untreated si#1, and so on. If the authors would like to show siFOXM1 induced growth/survival defects, they can still present the left part of the Figure 4E (the Vehicle group).

    Originally, we chose to show the absolute IntensityPercent for all groups, without normalizing to the untreated group, because we wanted to also highlight the FOXM1-mediated changes in growth. We agree that normalizing the IntensityPercent of the drug-treated group to the vehicle group better highlights the siFOXM1-mediated resistance. We have therefore re-analyzed the data and presented it this way in Figure 5E (described in lines 293-295). We moved our original Figure 4E to a new supplemental figure (Figure S4B) to still point out the effects of siFOXM1 on cell growth in untreated cells.

    Minor:

    1. In line 176, the author claimed that 'Interestingly, rare cells treated with CHK1i + gemcitabine are located within the untreated cell cluster (Fig. 3C)'. However, it is not as obvious where these cells are in the plot, especially to people who are new to tSNE plots. It would be appreciated if the authors could label these cells by circling them with red lines and make the point stronger.

    Rather than circling these points (we thought this would make the plot too "busy"), we have created an inset that zooms in on the region where we see the untreated cells within the untreated cell cluster. Within the inset, we use arrows to point out the cells we are referring to. This can be seen in our updated Figure 3C.

    1. In Figure S2B, it will be ideal to label clearly which genes are upregulated genes and which are downregulate.

    On the x-axis of the heatmap, we have drawn lines to separate the downregulated and upregulated genes.

    1. In line 50, the word 'multifaced' needs to be corrected to 'multifaceted'.

    Thank you for catching this, we have fixed it.

    1. It is unclear what 'underly drug resistance' means in line 150.

    We have reworded this sentence so that is more clear. It is now written as follows: "we aimed to identify gene-expression programs that mediate the low level of RS in a subset of cells, which could potentially mediate drug resistance". This change is in lines 155.

    1. It is advised that the phrase 'cell cycle position' could be changed to 'cell cycle phase' or 'cell cycle stage'.

    We purposefully used the phrase "cell cycle position" because we wanted to emphasis gradient-like progress through the cell cycle rather than a discrete distinction from one-phase to the next. We have reworded the text slightly to now say "position within S-phase" (lines 163, 187, 191, 208), since all the cells we are interested in are in S phase, but some are further through S phase than others.

    1. In line 185, the word 'in' after 'within' can be removed.

    Thank you for catching this, we have fixed it.

    1. In line 194, 'Among genes downregulated in γH2AXlow cells, the expression of ANLN, HMGB2, CENPE, MKI67 and UBE2C correlated' is missing an 'are' in front of the word 'correlated'.

    Thank you for catching this, we have fixed it.

    1. In line 239, Fig.SC3 should be Fig. S3C.

    Thank you for catching this, we have fixed it.

    1. FOXM1 is known as a crucial gene for G2/M transition. Therefore, FOXM1 knockdown cells are expected to be mostly arrested at the G2/M interface. Therefore, in line 244, it is incorrect to say stronger FOXM1 knockdown induced a 'lower proportion of cells in G2 phase'. In fact, as shown in Figure 4C, cells are accumulating in G2 phase (peaking around 11M on the DAPI axis) and depleted from G1 phase (peaking around 7M).

    We have reworded this to say that there is "a higher proportion of cells in S-phase and a less distinct G2 peak" (lines 270-271). The DAPI profiles of the scrambled, siFOXM1 #1, and siFOXM1 #2 conditions all show an S-phase "valley" between a G1 and G2 peak (the valley sits at about 8M-9M). In the siFOXM1 #3 and siFOXM1 #4 conditions, we no longer see this valley, therefore we interpret this as cells still in S-phase. If they had progressed from S-phase into G2 phase, we expect that we would again see this "valley" to the left of a clear G2 peak. In the figure below, we overlayed DNA content histograms of the different FOXM1 targeting siRNAs with the scrambled siRNA to demonstrate this point more clearly.

    Reviewer #1 (Significance (Required)):

    Advance: The study reported a novel reversible fixation technique which can lead to potentially good citations. However, the findings from the single cell sequencing alone fell short in novelty to reach high impact because FOXM1 has been reported to impact on cellular sensitivity to CHK1 inhibition mediated replication stress (PMC7970065). Moreover, the study did not provide mechanistic explanation to the observed phenotype but only validated the finding from the sequencing, and the gene of focus (FOXM1) was not originally identified from the sequencing, slightly undermining the paper's foundation. To make it a better paper. the authors need to be less biased when it comes to data analysis and interpretation.

    Audience: People who are interested in basic research in cell cycle, DNA damage, cancer, chemotherapy would be interested.

    My expertise: Cancer, DNA damage, cell cycle

    Reviewer #2 (Evidence, reproducibility and clarity (Required)):

    Summary:

    Replication stress activates ATR and CHEK1 kinases as part of the inter S phase DNA damage response. CHEK1 kinase inhibitors (CHK1i) have been shown to induce an accumulation of unresolved replication stress and widespread DNA damage and cell death caused by replication catastrophe, and are therefore under clinical evaluation. At the same time, CHEK1 inhibition results in the activation of CDK1 and FOXM1 and premature expression of G2/M genes (Saldivar et al., 2018 Science). FOXM1-drivent premature mitosis has been shown to be required for the replication catastrophe and CHK1i sensitivity (Branigan et al., 2021 Cell Rep.). In this study, Segeren and colleagues set out to investigate the mechanisms of replication stress tolerance. They used CHK1i inhibitors in combination with the DNA-damaging chemotherapeutic agent Gemcitabine and oncogenic HRASG12V expression to increase replication stress. The authors utilized an intriguing setup of combined immunofluorescence staining followed by single cell RNA-seq analysis to overcome limitations of bulk cell analyses. In particular, the authors sought to identify genes that are differentially regulated in replication stress-tolerant cells compared to sensitive cells. However, even single cell analyses can be confounded by differences in cell cycle distribution. To mitigate this, the authors selected mid S-phase cells for their analysis. While this may not have completely eliminated minor differences in cell cycle progression, the authors identified FOXM1-regulated G2/M cell cycle genes, among others, that were down-regulated in the tolerant cells. When the authors followed up on the effect of these genes on replication stress tolerance, they identified FOXM1 knockdown as the only robust mediator of replication stress tolerance.

    Major comments:

    The authors observed that cell cycle distribution could be a major confounding factor in their single cell analysis and attempted to reduce this variation by selecting mid S-phase cells based on the DAPI signal. The authors then chose to compare gH2AXlow and gH2AXhigh subpopulations of RPE-HRASG12V cells because their "DAPI signal was comparable" (line 181-184). However, their data show that these subpopulations also show differences in their DAPI signal distribution, with gH2AXlow cells tending to have lower DAPI signals than gH2AXhigh cells (Supplementary Figure 2A). Thus, the major confounding factor that the authors sought to remove seems to have prevailed and it remains possible that the difference in cell cycle gene expression is merely due to differences in cell cycle progression of the individual cells. Given that DAPI information seem to be readily available for the individual cells, the authors should normalize their analysis to the DAPI signal to remove this potential confounding effect or clearly state this potential limitation.

    We agree that indeed it is very challenging to fully disentangle the influence of cell cycle distribution on our analysis. And indeed, the γH2AXlow HRASG12V cells have slightly reduced median DNA content compared to γH2AXmid and γH2AXhigh. However, this was not the case in the RPE control cells, and we still found that FOXM1 target genes were strongly enriched in the γH2AXhigh cells (Fig S2C and Table S4). Therefore, it is highly unlikely that bias in S-phase position distributions does not explain our results. Nevertheless, to be transparent about this write in the Results on lines 192-193 the following: "The other groups all showed similar DAPI intensities, although gH2AXlow RPE-HRASG12V cells showed a slight but statistically significant reduction compared to their gH2AXhigh counterparts (Fig. S2A)".

    In our subsequent experiments to assess the relationship between phospho-FOXM1 (representing the transcriptionally active protein) and γH2AX, we observed that though there was a strong correlation between pFOXM1 and γH2AX, there was no correlation between phospho-FOXM1 and DAPI (Figure 4D-E). We therefore would like to point out that although our readout for replication stress inevitably increases as cells progress through DNA replication, heterogeneity in phospho-FOXM1 levels cannot be explained by position in S-phase. These results are described in lines 203-209.

    Finally, we do not think it would be statistically appropriate to use the DAPI signal (generated by fluorescence intensity as measured by the flow cytometer) as a normalization factor for our gene expression data.

    Minor comments:

    The findings of Saldivar et al., 2018 Science and Branigan et al., 2021 Cell Rep. should be mentioned in the introduction.

    As recommended, we mentioned both these papers in the introduction. In line 62, we cite the Branigan paper as showing that modulation of cell cycle regulators is a strategy used by cancer cells to resist replication stress. In lines 63-65, we reference them as follows: "The RS response is tightly linked with cell cycle progression, as multiple intra S-phase checkpoint kinases play a role in curtailing proteins involved in the S-G2 transition (Branigan et al., 2021, Saldivar et al., 2018)."

    The authors conclude that "cell cycle position can be a major confounding factor when evaluating the transcriptomic response to RS." It should be noted that stochastic differences in the cell cycle distribution of bulk cells are perhaps the best-known confounder in single cell analyses (see, for example, Buettner et al., 2015 Nat. Biotechnol.).

    We chose to reference the Buettner paper to justify our decision to select only cycling cells in our scRNA seq approach. Our reference to the paper, and to the fact that cell cycle distribution is a major confounder in single cell analysis, is in lines 138-140.

    Supplementary Figure 2A: The median should be added to the violin plots.

    As suggested, we have added medians to the violin plots. In addition, we added details on statistical analysis.

    The statement "Differential expression analysis revealed 19 genes that were significantly downregulated in gH2AXlow RPE-HRASG12V cells, suggesting that elevated levels of these genes are correlated with sensitivity to RS-inducing drugs" refers to Figure 3E and Table S1. However, Table S1 lists the "key resources" and does not seem to be related to this statement. A table showing log2fold-changes and FDR values should be added and referenced here.

    We have generated tables with the fold change values of differentially expressed genes between the yH2AX low and yH2AX high cells. These are found in Table S1 (for HRAS G12V cells) and Table S2 (for Control cells) in the supplementary file of the revised manuscript. The "key resources" has been moved to Table S5.

    The statement "Remarkably, Braningan and co-workers observed no effect of full FOXM1 deletion on cell cycle progression" seems somewhat inconsistent with what has been stated and assessed in that study. The authors may want to replace "progression" with "distribution". A reduction in proliferation is commonly observed when FOXM1 levels are reduced.

    In addition, the authors may want to consider that their addition of HRASG12V and Gemcitabine may contribute to a more substantial S phase checkpoint response.

    We agree with the reviewer that a reduction in proliferation is commonly observed when FOXM1 levels are reduced (Barger et al., 2021, Cheng et al., 2022, Yang et al., 2015, Wu et al., 2010), but in Branigan et al., they see no decrease in proliferation with knockout of FOXM1. They state "There were no apparent differences in the growth rate of the LIN54 and FOXM1 KO versus EV cells over 10 days (Figure 1G)". Though they do not elaborate on why they see this unexpected response, we suspect a permanent full knockout of FOXM1 could cause compensatory adaptation in their cell lines. In our experiments, we perform transient knockdowns, so cells may not have the time to adapt to the loss of FOXM1 and obtain compensatory mechanisms that would allow them to continue cycling as rapidly as control cells treated with non-targeting siRNA.

    However, we decided to remove this from the Discussion section, as it seemed to interrupt the discussion about the potential mechanisms underlying protection against DNA damage by FOXM1 depletion.

    The statement that "the mechanism by which high FOXM1 activity is a prerequisite to accumulate DNA damage in S-phase during CHK1 inhibition remains to be uncovered" seems to neglect that premature mitosis has been suggested as a mechanistic cause (Branigan et al., 2021 Cell Rep.). It would be helpful if the authors could elaborate on this.

    In our discussion, we do already emphasize the described role of FOXM1 in promoting premature mitosis (lines 330-337), but we argue that in our experimental conditions we are observing another - previously undescribed- role for FOXM1 in promoting replication stress during S phase. We previously observed with live cell imaging that CHK1i + gemcitabine does not cause premature mitosis in RPE-HRASG12V cells (published in Segeren et al. Oncogene 2022, Figure 5). Instead, these cells typically showed a cell cycle exit from G2. This makes it highly unlikely that premature mitosis is the reason why these cells would accumulate excessive DNA damage. We realize now that it was an important omission not to elaborate on this and have added this clarification to the Discussion (lines 341-345 in revised manuscript). In addition, we have removed a few lines of less important text (about the lack of direct effect of FOXM1 KO in the Branigan paper; see answer to previous point) to improve clarity and readability.

    Reviewer #2 (Significance (Required)):

    General assessment: The strength of the study is the intriguing methodology of combined immunofluorescence followed by single cell RNA-seq. The limitations are that this methodology does not seem to fully solve the stated problems. In addition, the study is essentially limited to confirming previous findings.

    Advance: The study strengthens current knowledge but provides essentially no advance. The authors confirm existing knowledge with an additional approach. While this is not an advance in itself, it is important to the community.

    Audience: I felt that the study would appeal to a basic science audience. In particular, the CHK1i and intra S-phase checkpoint areas, with limited interest beyond that.

    My relevant expertise lies in transcriptomics, gene regulation and the cell cycle.

    Reference list

    Barger, C.J., Chee, L., Albahrani, M., Munoz-Trujillo, C., Boghean, L., Branick, C., Odunsi, K., Drapkin, R., Zou, L. & Karpf, A.R. 2021, "Co-regulation and function of FOXM1/RHNO1 bidirectional genes in cancer", *eLife, *vol. 10, pp. 10.7554/eLife.55070.

    Branigan, T.B., Kozono, D., Schade, A.E., Deraska, P., Rivas, H.G., Sambel, L., Reavis, H.D., Shapiro, G.I., D'Andrea, A.D. & DeCaprio, J.A. 2021, "MMB-FOXM1-driven premature mitosis is required for CHK1 inhibitor sensitivity", *Cell reports, *vol. 34, no. 9, pp. 108808.

    Cheng, Y., Sun, F., Thornton, K., Jing, X., Dong, J., Yun, G., Pisano, M., Zhan, F., Kim, S.H., Katzenellenbogen, J.A., Katzenellenbogen, B.S., Hari, P. & Janz, S. 2022, "FOXM1 regulates glycolysis and energy production in multiple myeloma", *Oncogene, *vol. 41, no. 32, pp. 3899-3911.

    Overholser, B.R. & Sowinski, K.M. 2008, "Biostatistics primer: part 2", *Nutrition in clinical practice : official publication of the American Society for Parenteral and Enteral Nutrition, *vol. 23, no. 1, pp. 76-84.

    Saldivar, J.C., Hamperl, S., Bocek, M.J., Chung, M., Bass, T.E., Cisneros-Soberanis, F., Samejima, K., Xie, L., Paulson, J.R., Earnshaw, W.C., Cortez, D., Meyer, T. & Cimprich, K.A. 2018, "An intrinsic S/G(2) checkpoint enforced by ATR", *Science (New York, N.Y.), *vol. 361, no. 6404, pp. 806-810.

    Segeren, H.A., van Liere, E.A., Riemers, F.M., de Bruin, A. & Westendorp, B. 2022, "Oncogenic RAS sensitizes cells to drug-induced replication stress via transcriptional silencing of P53", *Oncogene, *vol. 41, no. 19, pp. 2719-2733.

    Wu, Q., Liu, C., Tai, M., Liu, D., Lei, L., Wang, R., Tian, M. & Lu, Y. 2010, "Knockdown of FoxM1 by siRNA interference decreases cell proliferation, induces cell cycle arrest and inhibits cell invasion in MHCC-97H cells in vitro", *Acta Pharmacologica Sinica, *vol. 31, no. 3, pp. 361-366.

    Yang, K., Jiang, L., Hu, Y., Yu, J., Chen, H., Yao, Y. & Zhu, X. 2015, "Short hairpin RNA- mediated gene knockdown of FOXM1 inhibits the proliferation and metastasis of human colon cancer cells through reversal of epithelial-to-mesenchymal transformation", *Journal of experimental & clinical cancer research : CR, *vol. 34, no. 1, pp. 40-1.

    We want to thank both reviewers for their thorough and constructive review of our manuscript. Below, we have re-iterated their comments followed by an explanation of how we have revised the manuscript to address this.

  3. Review Commons

    Note: This preprint has been reviewed by subject experts for Review Commons. Content has not been altered except for formatting.

    Learn more at Review Commons

    Referee #2

    Evidence, reproducibility and clarity

    Summary:

    Replication stress activates ATR and CHEK1 kinases as part of the inter S phase DNA damage response. CHEK1 kinase inhibitors (CHK1i) have been shown to induce an accumulation of unresolved replication stress and widespread DNA damage and cell death caused by replication catastrophe, and are therefore under clinical evaluation. At the same time, CHEK1 inhibition results in the activation of CDK1 and FOXM1 and premature expression of G2/M genes (Saldivar et al., 2018 Science). FOXM1-drivent premature mitosis has been shown to be required for the replication catastrophe and CHK1i sensitivity (Branigan et al., 2021 Cell Rep.). In this study, Segeren and colleagues set out to investigate the mechanisms of replication stress tolerance. They used CHK1i inhibitors in combination with the DNA-damaging chemotherapeutic agent Gemcitabine and oncogenic HRASG12V expression to increase replication stress. The authors utilized an intriguing setup of combined immunofluorescence staining followed by single cell RNA-seq analysis to overcome limitations of bulk cell analyses. In particular, the authors sought to identify genes that are differentially regulated in replication replication stress-tolerant cells compared to sensitive cells. However, even single cell analyses can be confounded by differences in cell cycle distribution. To mitigate this, the authors selected mid S-phase cells for their analysis. While this may not have completely eliminated minor differences in cell cycle progression, the authors identified FOXM1-regulated G2/M cell cycle genes, among others, that were down-regulated in the tolerant cells. When the authors followed up on the effect of these genes on replication stress tolerance, they identified FOXM1 knockdown as the only robust mediator of replication stress tolerance.

    Major comments:

    The authors observed that cell cycle distribution could be a major confounding factor in their single cell analysis and attempted to reduce this variation by selecting mid S-phase cells based on the DAPI signal. The authors then chose to compare gH2AXlow and gH2AXhigh subpopulations of RPE-HRASG12V cells because their "DAPI signal was comparable" (line 181-184). However, their data show that these subpopulations also show differences in their DAPI signal distribution, with gH2AXlow cells tending to have lower DAPI signals than gH2AXhigh cells (Supplementary Figure 2A). Thus, the major confounding factor that the authors sought to remove seems to have prevailed and it remains possible that the difference in cell cycle gene expression is merely due to differences in cell cycle progression of the individual cells. Given that DAPI information seem to be readily available for the individual cells, the authors should normalize their analysis to the DAPI signal to remove this potential confounding effect or clearly state this potential limitation.

    Minor comments:

    The findings of Saldivar et al., 2018 Science and Branigan et al., 2021 Cell Rep. should be mentioned in the introduction.

    The authors conclude that "cell cycle position can be a major confounding factor when evaluating the transcriptomic response to RS." It should be noted that stochastic differences in the cell cycle distribution of bulk cells are perhaps the best-known confounder in single cell analyses (see, for example, Buettner et al., 2015 Nat. Biotechnol.).

    Supplementary Figure 2A: The median should be added to the violin plots.

    The statement "Differential expression analysis revealed 19 genes that were significantly downregulated in gH2AXlow RPE-HRASG12V cells, suggesting that elevated levels of these genes are correlated with sensitivity to RS-inducing drugs" refers to Figure 3E and Table S1. However, Table S1 lists the "key resources" and does not seem to be related to this statement. A table showing log2fold-changes and FDR values should be added and referenced here.

    The statement "Remarkably, Braningan and co-workers observed no effect of full FOXM1 deletion on cell cycle progression" seems somewhat inconsistent with what has been stated and assessed in that study. The authors may want to replace "progression" with "distribution". A reduction in proliferation is commonly observed when FOXM1 levels are reduced. In addition, the authors may want to consider that their addition of HRASG12V and Gemcitabine may contribute to a more substantial S phase checkpoint response.

    The statement that "the mechanism by which high FOXM1 activity is a prerequisite to accumulate DNA damage in S-phase during CHK1 inhibition remains to be uncovered" seems to neglect that premature mitosis has been suggested as a mechanistic cause (Branigan et al., 2021 Cell Rep.). It would be helpful if the authors could elaborate on this.

    Significance

    General assessment: The strength of the study is the intriguing methodology of combined immunofluorescence followed by single cell RNA-seq. The limitations are that this methodology does not seem to fully solve the stated problems. In addition, the study is essentially limited to confirming previous findings.

    Advance: The study strengthens current knowledge but provides essentially no advance. The authors confirm existing knowledge with an additional approach. While this is not an advance in itself, it is important to the community.

    Audience: I felt that the study would appeal to a basic science audience. In particular, the CHK1i and intra S-phase checkpoint areas, with limited interest beyond that.

    My relevant expertise lies in transcriptomics, gene regulation and the cell cycle.

  4. Review Commons

    Note: This preprint has been reviewed by subject experts for Review Commons. Content has not been altered except for formatting.

    Learn more at Review Commons

    Referee #1

    Evidence, reproducibility and clarity

    This manuscript presented by Segeren et al. applied an interesting HRASG12V inducible cell model to study the mechanism of cellular resistance to replication stress inducing agents. They also employed a novel reversible fixation technique which allows them to FAC sort cells according to their replication stress levels before applying single cell sequencing analysis to the same cell populations. By comparing cells with low levels of replication stress to cells with high levels of replication stress, they found that reduction in gene expression of FOXM1 target genes potentially protects cells against replication stress induced by CHK1i plus gemcitabine combination.

    Overall, this is a very interesting study. However, the following points should be addressed prior to publication:

    Major:

    1. Figure 3E and 3F showed two lists of differentially expressed genes in γH2Ax low cells. However, instead of arbitrarily extracting the FOXM1 target genes and TP53 targeted genes, it would be appreciated if the author could perform an unbiased and unsupervised gene set enrichment analysis such as Enrichr.
    2. At the experiment design stage, the authors also included HRASG12V status as a test condition because they previously found that HRASG12V mutation induces basal level replication stress and they would like to include this condition to study the adaptation to replication stress (line 110). However, the difference in HRASG12V negative and HRASG12V positive cells was not followed up in the later part of the paper. Can they show lists of differentially expressed genes identified under HRASG12V negative conditions as well (in the same format of Figure 3E and 3F) and comment on the differences as well?
    3. In line 194 and in Figure S2B, the authors claimed that ANLN, HMGB2, CENPE, MKI67, and UBE2C demonstrated co-expression, but other genes displaying similar correlation scores were not commented (such as F3, CYR61, CTGF, etc). To avoid being biased at the analysis stage, the authors should define clearly what the cut-off of correlation score is and why only co-expression of ANLN, HMGB2, CENPE, MKI67, and UBE2C were mentioned.
    4. In line 215, instead of validating CENPE, UBE2C, HMGB2, ANLN, and MKI67 individually, the authors decided to validate FOXM1 instead, because they believe all the aforementioned genes are targets of FOXM1, therefore, validating FOXM1 alone would suffice. Again, this makes the validation process also biased. CENPE, UBE2C, HMGB2, ANLN, and MKI67 should be validated individually because they might sensitize cells to replication stress via different mechanisms. Besides, if all these genes were identified together because they are FOXM1 target genes, why did the authors not identify FOXM1 itself as a differentially expressed gene from the single cell sequencing? The sequencing only analyzed the S/G2/M cells, expression of FOXM1 should be detected easily.
    5. As pointed out by the author in the Discussion, single cell sequencing is not good at differentiating the causes from the consequences. The author tried to validate many of the differentially expressed genes in γH2Ax low cells. However, the fact that only FOXM1 knockdown passed the validation and deconvolution pointed out that the great majority of the identified genes are not the cause of the sensitivity change to replication stress inducing agents but likely the consequences. Therefore, in Figure S2C and S2D, it would be better that the authors could just name the genes as 'downregulated genes' in Figure S2C and 'upregulated genes' in Figure S2D. Taking into consideration that the expression change in the great majority of these genes are just consequences of sensitivity change to replication stress, defining them as 'potentially sensitizing' genes and 'potentially conferring resistance' genes is rather misleading.
    6. To better prove that FOXM1 is the leading cause of the sensitivity to CHK1i+Gemcitabine induced replication stress, can the authors show the FOXM1 expression status in the tolerant cell population identified in Figure 1B (lowest panel)? Alternatively, can they plot FOXM1 expression level in the same tSNE plots shown in Figure 3B to 3D to see whether some of the γH2Ax low populations also show reduced FOXM1 expression?
    7. clonogenic survival assay in Figure 4D was not quantified properly in Figure 4E. To rule out the siFOXM1 mediated growth/survival defects and to only focus on the siFOXM1 mediated resistance to CHK1i+Gemcitabine, the survival rate (intensity percent in this case) of CHK1i+Gemcitabine treated condition should be normalized against the survival rate of the Vehicle condition. E.g., the intensity percent of the siSCRAMBLE after treatment should be divided by the intensity percent of the untreated siSCRAMBLE; the intensity percent of the si#1 after treatment should be divided by the intensity percent of the untreated si#1, and so on. If the authors would like to show siFOXM1 induced growth/survival defects, they can still present the left part of the Figure 4E (the Vehicle group).

    Minor:

    1. In line 176, the author claimed that 'Interestingly, rare cells treated with CHK1i + gemcitabine are located within the untreated cell cluster (Fig. 3C)'. However, it is not as obvious where these cells are in the plot, especially to people who are new to tSNE plots. It would be appreciated if the authors could label these cells by circling them with red lines and make the point stronger.
    2. In Figure S2B, it will be ideal to label clearly which genes are upregulated genes and which are downregulate.
    3. In line 50, the word 'multifaced' needs to be corrected to 'multifaceted'.
    4. It is unclear what 'underly drug resistance' means in line 150.
    5. It is advised that the phrase 'cell cycle position' could be changed to 'cell cycle phase' or 'cell cycle stage'.
    6. In line 185, the word 'in' after 'within' can be removed.
    7. In line 194, 'Among genes downregulated in γH2AXlow cells, the expression of ANLN, HMGB2, CENPE, MKI67 and UBE2C correlated' is missing an 'are' in front of the word 'correlated'.
    8. In line 239, Fig.SC3 should be Fig. S3C.
    9. FOXM1 is known as a crucial gene for G2/M transition. Therefore, FOXM1 knockdown cells are expected to be mostly arrested at the G2/M interface. Therefore, in line 244, it is incorrect to say stronger FOXM1 knockdown induced a 'lower proportion of cells in G2 phase'. In fact, as shown in Figure 4C, cells are accumulating in G2 phase (peaking around 11M on the DAPI axis) and depleted from G1 phase (peaking around 7M).

    Significance

    Advance:

    The study reported a novel reversible fixation technique which can lead to potentially good citations. However, the findings from the single cell sequencing alone fell short in novelty to reach high impact because FOXM1 has been reported to impact on cellular sensitivity to CHK1 inhibition mediated replication stress (PMC7970065). Moreover, the study did not provide mechanistic explanation to the observed phenotype but only validated the finding from the sequencing, and the gene of focus (FOXM1) was not originally identified from the sequencing, slightly undermining the paper's foundation. To make it a better paper. the authors need to be less biased when it comes to data analysis and interpretation.

    Audience:

    People who are interested in basic research in cell cycle, DNA damage, cancer, chemotherapy would be interested.

    My expertise:

    Cancer, DNA damage, cell cycle

  5. Version published to 10.1101/2024.03.26.585806v1 on bioRxiv