Dicer promotes genome stability via the bromodomain transcriptional co-activator BRD4

RNA interference is required for post-transcriptional silencing, but also has additional roles in transcriptional silencing of centromeres and genome stability. However, these roles have been controversial in mammals. Strikingly, we found that Dicer-deficient embryonic stem cells have strong proliferation and chromosome segregation defects as well as increased transcription of centromeric satellite repeats, which triggers the interferon response. We conducted a CRISPR-Cas9 genetic screen to restore viability and identified transcriptional activators, histone H3K9 methyltransferases, and chromosome segregation factors as suppressors, resembling Dicer suppressors identified in independent screens in fission yeast. The strongest suppressors were mutations in the transcriptional co-activator Brd4, which reversed the strand-specific transcription of major satellite repeats suppressing the interferon response, and in the histone acetyltransferase Elp3. We show that identical mutations in the second bromodomain of Brd4 rescue Dicer-dependent silencing and chromosome segregation defects in both mammalian cells and fission yeast. This remarkable conservation demonstrates that RNA interference has an ancient role in transcriptional silencing and in particular of satellite repeats, which is essential for cell cycle progression and proper chromosome segregation. Our results have pharmacological implications for cancer and autoimmune diseases characterized by unregulated transcription of satellite repeats.


<b>REVIEWER COMMENTS</B>
Reviewer #1 (Remarks to the Author): Gutbrod et al. examine the possible role of Dicer in nuclear RNAi in mouse embryonic stem cells (mESCs). In addition to its well-established roles in post-translational gene silencing, RNAi is required for some transcriptional gene silencing events and genome stability. These nuclear functions are best understood in fission yeast, plants, and C. elegans but whether RNAi also contributes to maintenance of genome stability or transcription regulation in mammalian cells has remained unclear. Some early previous studies observed loss of DNA methylation and centromeric silencing, increased telomere recombination, and chromosome segregation defects in Dicer-/-mESCs (Fukagawa et al., 2004, Nature Cell Biology;Kanellopoulou et al, 2005, Genes Dev) but later studies suggested that these defects were likely indirect and due to loss of specific miRNA(s) that regulate DNA methyltransferase expression (Benetti et al., 2008, NSMB;Sinkkonen et al., 2008, NSMB -not cited). Gutbrod et al. use a conditional knockout of the RNaseIII domain of Dicer in order to avoid ambiguities that may result from accumulation of suppressor mutation(s) that would allow Dicer-/-cells to overcome cell cycle arrest and/or growth defects. They report the following main observations. (1) Dicer -/-mESCs have strong proliferation and chromosome segregation defects that induce the interferon response; (2) a CRISPR-cas9 genetic screen to overcome the viability defect of Dicer-/-cells uncovers mutation in transcription activators, including the Brd4 transcriptional co-activator and the Elp3 histone acetyltransferase; and (3) mutations in the second bromodomain of Brd4 and its fission yeast homolog rescue Dicer-dependent silencing and chromosome segregation defects. Based on these findings they propose that RNAi roles in transcriptional gene silencing of DNA repeats is conserved from fission yeast to mammals. A conserved role for RNAi in these nuclear events would be of great general interest. However, the authors have not ruled that the effects they observe are distinct from previously observed effects that were concluded to be indirect. Some of the findings described by the authors are also inconsistent with this major claim and experiments that would establish direct mechanistic links between different systems are lacking.
Specific comments 1. One of my main concerns is that the authors have not ruled out that the effects they observe following conditional Dicer deletion are indirect. Previous studies suggest that the heterochromatin defects of Dicer-/-cells were due to loss of the miRNA pathway. One of the arguments the authors use against the possibility that the miRNA pathway is responsible for the effects they observe is based on the differences in % lagging/bridging chromosomes and micronuclei between Dicer-/-day 5 KO or clones (propagated for some time) and Dcgr8-/-(Drosha) clones. The authors make an important point about Dicer-/-cells may accumulate suppressors. This may also apply to Dgcr8-/-mESCs. They should perform a conditional KO of Dcgr8 similar to what they have done for Dicer to rule out a role for the miRNA pathway. Their results also do not explain the previous observation that the proliferation and DNA methylation defects of Dicer-/-cells could be rescued by transfection with miR-290 cluster miRNA (Benetti et al., 2008, NSMB;Sinkkonen et al., 2008, NSMB).
2. The small RNA analysis data ( Figure S3C) is inconsistent with a conserved mechanism of RNAi in both fission yeast and mammals proposed by the authors. They do not detect any Dicer-dependent small RNAs that map to major satellite repeats as a model based on conservation would suggest. In contrast, they see accumulation of small RNAs which are Dicer-independent and likely represent degradation products of derepressed major satellite repeat RNAs. These data are more consistent with an indirect effect due to loss of DNA methylation or a non-canonical role for Dicer. The authors should test their conditional Dicer-/-cells have lower levels of DNA methylation. In addition, the authors should test whether the proliferation and chromosome segregation defects of Dicer-/-can be rescued by transfection with WT Dicer versus RNaseIII domain 1 and 2 catalytically dead Dicers.
3. The recovery of Brd4 and Elp3 mutations as suppressors of Dicer-/-is a nice result. However, the data regarding major satellite repeats as a direct target of Dicer and Brd4 is only suggestive. It is equally possible that the effect is indirect or mediated through the other 97 genic targets that show increased expression in Dicer-/-and are suppressed by Brd4-( Figure S8E and Figure 4). The data supporting the idea that the reduction in viability of Dicer-/-mESCs is in part due to the accumulation of dsDNA in the cytoplasm (Fig. 3L) is interesting and consistent with chromosome segregation defects and micronuclei accumulation in Dicer-/-cells. Unfortunately, the later may still be indirect effects due to downregulation of DNMTs in Dicer-/-cells.

Reviewer #2 (Remarks to the Author):
This study by Gutbrod and colleagues revisits Dicer knockout ES cells and rescue clones to describe clear suppressors, presenting strong associations uncovering the molecular basis for the phenotypic defects observed and suggests a conserved mode of action for Dicer in silencing centromeric repeats. This is an important finding, which will have a significant impact on the field. Overall, it is a very well implemented, controlled and presented study. For these reasons, in my view the manuscript warrants publication, providing my comments below are addressed.
However, in my view, the story lacks a little in coherence including a number of implied mechanistic links between the molecular activities of the identified suppressors and thus does not provide significant new insights into mechanism. I would suggest that addressing the points below will strengthen the impact of this study, and provide a firmer basis for many of the suggested causal relationships alluded to in the discussion.

Major:
There is an apparent inconsistency between Figures 2 and 3. If the Crd4 BD1 domain is essential as suggested by Figure 2, why doesn't the Brd4 siRNA or JQ1 have proliferation defects in wt cells ( Figure  S7) and thus no suppressive function in Dicer knockouts ( Figure 3) as for the BD1 mutations ( Figure 2B)? The Minor satellites are, as frequently the case, left in the shadow of the Major satellites. For example, Figure S8C is not referenced at all in the text. This Figure shows that Minor satellites are similarly misregulated as Major Satellites in Dicer knockouts and their transcription is also rescued by Brd4 and Elp3 mutations. Is strand specific transcription of Minor satellites also affected in a similar way as Major Satellites? Is Brd4 enriched on Minor Satellites in Dicer Knockouts?
The authors suggest that the partial rescue of MajSat transcription by Brd4 and Elp3 mutations is important for the suppressive activity of these mutants. However only associations are provided as evidence here, and thus 'Dicer1-/-Chromosomal Defects Depend on Major Satellite Transcription' in title of this section should be toned down. As the authors discuss, the altered MajSat transcription could be merely a consequence of cell cycle differences and irrelevant to the phenotypic outcomes, particularly because Brd4 and Elp3 have global effects on transcription (97 likely direct Brd4 direct targets in this context). To provide convincing evidence that the Dicer phenotype is dependent on MajSat transcription, they should directly downregulate MajSat reverse strand transcription, using a TALE or dCas9 fused to KRAB for example. For coherence, it would also be interesting to determine whether this affects Cohesin loading, which itself rescues proliferation ( Figure 3I) (see next comment). Further, the reversal of transcription orientation of Major Satellites is intriguing, but again only correlated with phenotypic outcomes. Is the reported effect of ectopic expression of satellite transcripts on chromosome mis-segregation, strand specific? Are forward or reverse MajSat transcripts differentially associated with chromatin, and/or formation of DNA:RNA hybrids? Can they artificially switch major satellite transcription orientation to the forward strand in Dicer knockouts (using a similar dCas9/TALE strategy) and demonstrate a rescue of the phenotypes?
While the identification of three categories of Dicer KO suppressors (transcription, H3K9 methylation and chromosome segregation) and their seemingly consistent associations is an important result, the mechanistic and functional link between them is under-investigated. Which, if any one of these activities, is the downstream mediator of phenotypic defects in Dicer knockouts? More evidence should be provided for the molecular associations between them, to determine causal relationships. For example, RAD21 binding at centromeres is not altered in Dicer single KOs compared to WT cells (S9C). How then can Cohesin binding at centromeres be responsible for the phenotypic defects observed in Dicer single KOs, as alluded to by the authors? Another implication is that the Suv39h1 and Ehmt1 suppressive effects occur through Cohesin. Is RAD21 loading at pericentromeic regions decreased in Suv39h1/Dicer double mutants? What about Major and Minor Satellite transcription? Is reverse orientation MajSat transcription also observed in this context? Minor: Page 5: 'both inhibition and overexpression of Aurora B' sentence not finished. Figure S7A is incorrectly cited as S7C Reviewer #3 (Remarks to the Author): In the manuscript "Dicer promotes genome stability via the bromo-domain transcriptional co-activator BRD4" the authors address the role of Dicer in heterochromatic silencing. They find that deletion of Dicer leads to upregulation of major satellite repeats and reduced viability, which can be rescued by deletion of transcription activators such as Brd4. Furthermore, perturbations in transcription by chemical agents or siRNAs increase viability. This genetic interaction is conserved in fission yeast as well. Although the study offers new insights in Dicer role in heterochromatic silencing in mammalian cells, it does not go beyond genetic interactions. The study is missing more mechanistic data showing what actually Dicer does at these repeats and how it prevents Brd4 recruitment.
1) The authors should provide more mechanistic data showing that Dicer is actually involved in silencing of major satellite repeats. The authors should show if Dicer activity is required for silencing and reduced viability. Although, the authors do not detect Dicer dependent small RNAs, it is possible that Dicer activity plays a role in the process. Alternatively, Dicer might directly interact with transcription machinery to regulate transcription in the repeats.
2) The data suggest that Dicer might somehow regulate transcription, but it is unclear if this is the case. The authors should at least determine if Dicer is localized to the repeats and if it interacts with transcription machinery.
3) The authors show that major satellite repeats are upregulated in Dicer deletion cells. This suggest that Dicer is involved in silencing of those repeats. They observe that in Dicer/Brd4 double mutants the transcripts are still heavily upregulated, however, the opposite strand is now being transcribed. To me it is unclear how this switch in transcription happens and how it suppresses interferon response. The authors should provide more data that the transcription from the opposite strand is resulting in increased viability. Interesting observation is also that in wild type cells both + and -strands from repeats are transcribed, while in Dicer deletion only + strand and in Dicer/Brd4 deletion cells only -strand is transcribed. The authors should comment why Dicer deletion upregulates + strand and silences -strand. 4) The authors claim that "pericentromeric transcripts in Dicer mutants had strong strand specificity, which was reversed in the triple mutant ( Fig. 5C)". Unfortunately, I do not see this in the data. In all the data presented forward strand is expressed to a higher level than reversed.

REVIEWER COMMENTS
Reviewer #1 (Remarks to the Author): Gutbrod et al. examine the possible role of Dicer in nuclear RNAi in mouse embryonic stem cells (mESCs). In addition to its well-established roles in post-translational gene silencing, RNAi is required for some transcriptional gene silencing events and genome stability. These nuclear functions are best understood in fission yeast, plants, and C. elegans but whether RNAi also contributes to maintenance of genome stability or transcription regulation in mammalian cells has remained unclear. Some early previous studies observed loss of DNA methylation and centromeric silencing, increased telomere recombination, and chromosome segregation defects in Dicer-/-mESCs (Fukagawa et al., 2004, Nature Cell Biology;Kanellopoulou et al, 2005, Genes Dev) but later studies suggested that these defects were likely indirect and due to loss of specific miRNA(s) that regulate DNA methyltransferase expression (Benetti et al., 2008, NSMB;Sinkkonen et al., 2008, NSMB -not cited). Gutbrod et al. use a conditional knockout of the RNaseIII domain of Dicer in order to avoid ambiguities that may result from accumulation of suppressor mutation(s) that would allow Dicer-/-cells to overcome cell cycle arrest and/or growth defects. They report the following main observations. (1) Dicer -/-mESCs have strong proliferation and chromosome segregation defects that induce the interferon response; (2) a CRISPR-cas9 genetic screen to overcome the viability defect of Dicer-/-cells uncovers mutation in transcription activators, including the Brd4 transcriptional co-activator and the Elp3 histone acetyltransferase; and (3) mutations in the second bromodomain of Brd4 and its fission yeast homolog rescue Dicer-dependent silencing and chromosome segregation defects. Based on these findings they propose that RNAi roles in transcriptional gene silencing of DNA repeats is conserved from fission yeast to mammals. A conserved role for RNAi in these nuclear events would be of great general interest. However, the authors have not ruled that the effects they observe are distinct from previously observed effects that were concluded to be indirect. Some of the findings described by the authors are also inconsistent with this major claim and experiments that would establish direct mechanistic links between different systems are lacking.
We thank the reviewer for recognizing the broad appeal of our study and for identifying potential inconsistencies that we have addressed below.
1. One of my main concerns is that the authors have not ruled out that the effects they observe following conditional Dicer deletion are indirect. Previous studies suggest that the heterochromatin defects of Dicer-/-cells were due to loss of the miRNA pathway. One of the arguments the authors use against the possibility that the miRNA pathway is responsible for the effects they observe is based on the differences in % lagging/bridging chromosomes and micronuclei between Dicer-/-day 5 KO or clones (propagated for some time) and Dcgr8-/-(Drosha) clones. The authors make an important point about Dicer-/-cells may accumulate suppressors. This may also apply to Dgcr8-/-mESCs. They should perform a conditional KO of Dcgr8 similar to what they have done for Dicer to rule out a role for the miRNA pathway. Their results also do not explain the previous observation that the proliferation and DNA methylation defects of Dicer-/-cells could be rescued by transfection with miR-290 cluster miRNA (Benetti et al., 2008, NSMB;Sinkkonen et al., 2008, NSMB).
1. We agree that the loss of microRNAs has effects on Dicer1-/-cells. However, we and others have provided significant genetic evidence that microRNAs are not driving the majority of the phenotypes we observe.
 The generation of Dgcr8 -/-cells by Wang et al. 2008 show that the proliferation defects of freshly derived Dgcr8-/-cells is very mild and therefore these cells are significantly less likely to accumulate suppressors as the selective pressure is much less than Dicer1-/-mESCs. The commercially available cells we used are low passage number and likely to reflect the effects seen in Wang et al. 2008.  Wang et al. 2008 also make this conclusion: "In addition, Dicer1 knockout ES cells seem to have a more profound initial proliferation defect that is overcome over time, presumably due to additional genetic events. By contrast, Dgcr8 knockout ES cells show a stable and more subtle proliferation defect. These differences in the phenotypes of Dicer1 and Dgcr8 knockout ES cells suggest that Dicer has miRNA-independent roles in ES cell function. Dgcr8 knockout ES cells provide a means to identify these roles."  The major phenotypes we describe -dysregulation of major satellite repeat transcription, chromosome segregation defects, severe proliferation defects, and a significant increase in apoptosis -have never been reported to occur in mESCs that have lost other RNAi components such as Drosha, Dgcr8, or Argonautes in our work or others. We therefore consider these phenotypes to be generally microRNA-independent. However, we acknowledge that microRNA-regulated gene expression changes likely play a modifying role for some of these phenotypes such as the subset of the proliferation defect that derives from the loss of cell cycle-regulating microRNAs, which can explain some of the cell cycle defects observed in Dicer1 -/-cells and most of the cell cycle defects observed in Dgcr8-/-cells.  Recently, identical phenotypes of major satellite transcription increase (also strandspecific) and segregation defects were found in Dicer1-/-germ cells (Yadav et al. 2020, Nucleic Acids Research) which have much more severe defects compared to Dgcr8-/germ cells (Zimmerman et al. 2014, PLOS One), extending these observations to additional cell types  Finally, similar Dicer phenotypes also occur in S. pombe, an organism without microRNAs or DNA methylation. While we recognize that microRNA-driven gene expression changes likely provide additional modification to the phenotypes we observe in mESCs, the origin of these transcriptional phenotypes is conserved and microRNAindependent.
2. The mechanism proposed by Benetti et al. 2008, NSMB andSinkkonen et al. 2008, NSMB is that miR-290 targets and represses Rbl2, a factor which represses DNA methyltransferases. A reduction in DNMTs in Dicer1-/-cells is then thought to lead a reduction in DNA methylation at various elements in the genome depending on the study. Additionally, Sinkkonen et al. 2008 show that transfection of miR-290 does show a mild rescue of proliferation defects in Dicer1-/-cells. However, a number of molecular studies argue that loss of methylation does not explain the effects that we describe for the following reasons. We thank the reviewer for suggesting this potential alternative model and have included an addition to the introduction to reflect our consideration for other readers: In Dicer1 -/-mESCs, widely differing phenotypes have been reported 26-28 and one explanation might be the accumulation of mutations that allow stalled Dicer1 -/cells to proliferate 28 . While changes in DNA methylation were initially hypothesized to be partially responsible for proliferation defects 34,35 , follow-up studies have demonstrated little change in DNA methylation levels in Dicer1 -/-mESCs which do not explain the phenotypes obverved 36 . Genetic suppressors arise in Dicer mutants of fission yeast when they exit the cell cycle and as RNAi becomes essential, resulting in the selection and outgrowth of suppressed strains 18 .
2. The small RNA analysis data ( Figure S3C) is inconsistent with a conserved mechanism of RNAi in both fission yeast and mammals proposed by the authors. They do not detect any Dicerdependent small RNAs that map to major satellite repeats as a model based on conservation would suggest. In contrast, they see accumulation of small RNAs which are Dicer-independent and likely represent degradation products of derepressed major satellite repeat RNAs. These data are more consistent with an indirect effect due to loss of DNA methylation or a noncanonical role for Dicer. The authors should test their conditional Dicer-/-cells have lower levels of DNA methylation. In addition, the authors should test whether the proliferation and chromosome segregation defects of Dicer-/-can be rescued by transfection with WT Dicer versus RNaseIII domain 1 and 2 catalytically dead Dicers.
We appreciate the opportunity to more clearly state our proposed mechanism. Our work presented here shows that the majority of the phenotypic effects we describe are independent of small RNAs. As the reviewer points out we make a point of stating that we do not see Dicerdependent major satellite small RNAs. Thus, the conserved mechanism we propose is not the well-described small RNA-directed chromatin formation mechanism that occurs in yeast and plants (but has not been shown to occur in mammals), but rather a conserved non-canonical role for Dicer that occurs at the transcription of pericentromeric repeats.
Our data indicates that Dicer plays a critical, direct role in the processing of pericentromeric transcription in both mouse and yeast.  Upon the loss of Dicer, transcription of this region increases abnormally, in a strand specific manner as we and others have observed (Yadav et al. 2020, NAR), this results in failure of cohesion, chromosome segregation defects, cytosolic DNA, and an increase in interferon signaling all of which reduce proliferation and increase apoptosis. We observe the same effects in S. pombe (an organism without microRNAs) and show they can be rescued by targeting the exact same factors that directly regulate pericentromeric transcription in both yeast and mammals.  As stated above, our model is not consistent with a global loss of DNA methylation and even a local loss of DNA methylation at major satellite repeats has not been observed in  The inducible Dicer1-/-alleles that we generate are a loss of both RNase III domains. We predict that both RNaseIII domains are likely required for processing substrates associated with these phenotypes. A more detailed examination is beyond the scope of this work.
Lastly, we provide additional data strongly supporting our model that direct regulation of transcription rather than post-transcriptional regulation is critical for generating and suppressing the phenotypes.  We see no major satellite-derived small RNAs that are lost upon mutating Dicer1 (Fig.  S3C).  We can suppress the viability and proliferation defects by simply inhibiting RNA Pol II with low doses of alpha-amanitin (Fig. 3C).  Reducing Argonaute expression does not suppress the viability defects (new Figs. S3 D-G, shown below).
We have also added a sentence to the text to address this new data: The lack of DICER1-dependent siRNA is consistent with the absence of an RdRP. Furthermore, the knockdown of all mouse Argonaute proteins simultaneously did not affect the proliferation of Dicer1 -/-mESCs in a viability assay (Figs. S3 D-G) also demonstrating that small RNA are unlikely to be the primary driver of the Dicer1 -/phenotypes. We also performed ChIP-seq, and observed a modest reduction in H3K9me3 at some ERVs, namely IAP and ETn, as well as at LINE1 transposable element loci (Figs. S4A, S4B, S4D, and S5B), as described previously 44 , which might be related to the loss of microRNA.
3. The recovery of Brd4 and Elp3 mutations as suppressors of Dicer-/-is a nice result. However, the data regarding major satellite repeats as a direct target of Dicer and Brd4 is only suggestive. It is equally possible that the effect is indirect or mediated through the other 97 genic targets that show increased expression in Dicer-/-and are suppressed by Brd4- (Figure S8E and Figure  4).
Our data described here as well as data published by others have shown conclusively that DICER1 and BRD4 directly target the major satellite repeat.  A direct interaction between the DICER1 protein and the major satellite genomic loci as well as transcripts has been detected by ChIP and RNA pull-down respectively in mouse cells (Hsieh et al. 2011 Nucleic Acids Research).  Additionally, in mouse spermatogenesis the loss of DICER1 induces major satellite transcription in a strand-specific manner (as we observe) and furthermore, a direct association between the DICER1 protein and the major satellite genomic loci was detected using a combined immunofluorescence/in situ hybridization microscopy approach as well as ChIP-PCR (Yadav et al. 2020 Nucleic Acids Research).  We also observe a significant increase in major satellite transcription (it is in fact the top upregulated transcript) when we ablate DICER1 activity. Therefore, the detection of direct regulation in multiple cell types and increased transcriptional output upon DICER1 loss across three cell types in mouse is convincing evidence that DICER1 directly targets the major satellite repeats.  Our combined ChIP-seq and RNA-seq datasets demonstrate that BRD4 is clearly bound to the major satellite genomic loci (Fig. 4A)  . This paper convincingly shows that direct perturbations of BRD4 activity with JQ1 or siRNAs reduced pericentromeric transcription as we have.  Thus, we and others have shown that BRD4 is localized to pericentromeric repeats and directly regulates the transcription of these repeats across evolution.
While we propose a direct regulation of the major satellite by DICER1 and BRD4 the other 97 genes that fit the same expression pattern could be playing a role as well and we thank the reviewers for bringing up this point. We have now closely interrogated each of these 97 candidates and passed them through a series of filters as follows.  Initially, we intersected our RNA-seq dataset with RNA-seq from both Drosha knockout mESCs (Georgakilas et al. 2014 Nature Comms) and Dgcr8 knockout mESCs (Cirera-Salinas et al. 2017 J. Cell Biology) in order to determine which of these genes were microRNA targets and thus not likely to generate these phenotypes as they do not occur in other microRNA-pathway mutants. This removed 34 of the genes.
 Furthermore, as we observed the same phenotypes and the same genetic relationships in S. pombe we would expect a true modifying locus to be conserved. We found only 7 genes that were not microRNA targets and were conserved in fission yeast. Of these 7, none of them were misregulated in Dicer mutants in S. pombe (Hansen et al. 2005, MCB).  In parallel, we did a deep literature search of all 97 genes to identity any that had been associated with chromosome segregation and we found that none had direct mechanistic links to this phenotype.  We did find that one of these 97 genes -Prdm16, an H3K9me1 methyltransferase -does act at the centromeric heterochromatin, but is functionally redundant with Prdm3 (Pinheiro et al. 2012, Cell), which is not differentially expressed in our datasets and therefore Prdm16 is not a candidate to drive these phenotypes.  Therefore, none of the other candidate genes had a strong link to chromosomal defects and this led us to focus on the pericentromeric satellite repeats themselves, as increased transcription has been shown to lead to chromosomal abnormalities in mouse cells (Zhu et al. 2018, Molecular Cell) and these transcripts showed by far the greatest magnitude of misregulation in Dicer1 -/-cells while this misregulation was also rescued by the Brd4 suppressors.
We have added to the main text to further detail our treatment of the 97 candidates: 97 genes were upregulated upon Dicer1 mutation, downregulated upon Brd4 and Elp3 mutation, and were located near a BRD4 ChIP-seq peak, suggesting they were direct targets of both DICER1 and BRD4. We closely examined all 97 of these candidates through -(i). Crossreferencing Dgcr8 and Drosha knockout mESC RNA-seq datasets to eliminate microRNA target genes, (ii). Searching for homologs in S. pombe as Dicer-dependent centromeric silencing is evolutionarily conserved, and (iii). A deep literature search to identify roles in chromosome segregation that could be generating this critical phenotype. None of the protein-coding candidates we examined passed these filters.
In contrast, the major satellite transcript was the most upregulated transcript in Dicer1 -/cells, and activation of major satellite transcription has been shown to cause chromosome segregation defects in mouse cells 56 .
The data supporting the idea that the reduction in viability of Dicer-/-mESCs is in part due to the accumulation of dsDNA in the cytoplasm (Fig. 3L) is interesting and consistent with chromosome segregation defects and micronuclei accumulation in Dicer-/-cells. Unfortunately, the later may still be indirect effects due to downregulation of DNMTs in Dicer-/-cells.
We appreciate the reviewer recognizing the consistency of phenotypes. As we described in our response above, our RNA-seq data showed we do not observe a downregulation of DNMTs upon the loss of Dicer1. Therefore, we propose the segregation defects leading to dsDNA in the cytoplasm is a direct effect of DICER1 acting on transcription of the pericentromeric repeat.

Reviewer #2 (Remarks to the Author):
This study by Gutbrod and colleagues revisits Dicer knockout ES cells and rescue clones to describe clear suppressors, presenting strong associations uncovering the molecular basis for the phenotypic defects observed and suggests a conserved mode of action for Dicer in silencing centromeric repeats. This is an important finding, which will have a significant impact on the field. Overall, it is a very well implemented, controlled and presented study. For these reasons, in my view the manuscript warrants publication, providing my comments below are addressed.
However, in my view, the story lacks a little in coherence including a number of implied mechanistic links between the molecular activities of the identified suppressors and thus does not provide significant new insights into mechanism. I would suggest that addressing the points below will strengthen the impact of this study, and provide a firmer basis for many of the suggested causal relationships alluded to in the discussion.
We thank the reviewer for recognizing the importance of this study to the field and we hope our response will sufficiently address the concerns regarding a lack of coherence between mechanistic links.
There is an apparent inconsistency between Figures 2 and 3. If the Brd4 BD1 domain is essential as suggested by Figure 2, why doesn't the Brd4 siRNA or JQ1 have proliferation defects in wt cells ( Figure S7) and thus no suppressive function in Dicer knockouts (Figure 3) as for the BD1 mutations ( Figure 2B)?
We appreciate the reviewer pointing out the seemingly conflicting data on bromodomain 1.
 We have found that in wild type mESCs CRISPR targeting of BD1 does lead to more deleterious effects than targeting BD2 -cells containing BD1-targeting guides drop out of the population more readily than BD2-targeting guides and this effect continues progressively over time ( Figure 2B).  Despite these negative effects, we find that from days 8-12 of the Dicer1-/-timecourse targeting BD1 does suppress the Dicer1-/-proliferation defects, but deeper into the timecourse (between 12-16 days) this effect is overcome by the negative consequences of specifically targeting BD1 ( Figure 2B).  While the insertions/deletions generated in BD1 by Cas9 are deleterious, the inhibition of BRD4 by JQ1, which binds both BD1 and BD2, or siRNA do not inhibit proliferation or viability over the 2-3 day timeframe of the luciferase-based MT assays (Figure 3). We hypothesize that it is the different targeting modality (permanent indels vs transitory inhibition) and different time frame that underly this apparent discrepancy. As observed by others, this shows that BD1 clearly has more essential functions in the cell than BD2, which does in fact tolerate loss-of-function mutations. Additionally in S. pombe, the bdf1 and bdf2 double mutant is synthetic lethal and the domain responsible for the synthetic lethality is the BD1 of bdf2 as shown by our tetrad analysis.  Importantly, we see the opposite relationship with Elp3 in Dicer1-/-cells. The Cas9 targeted mutations in the histone methyltransferase domain are excellent suppressors of the Dicer1 proliferation and viability defects, but siRNA treatment (at multiple concentrations) does not show suppression in the MT viability assays.  Therefore, it is critical to use multiple mechanisms of perturbation that have differing effects over time in order to more completely understand a gene's relationship to proliferation.
We have added further explanation to the text to address this point: Most significantly, the Dicer1 -/viability defect was strongly rescued by the small molecule inhibitor JQ1, which specifically inhibits BRD4 and its paralogs (Fig. 3D). While targeting BRD4 with siRNAs or JQ1 does inhibit both bromodomains, we do not observe the deleterious effects of targeting BD1 with these modalities in either Dicer1 -/or wild type mESCs (Figs. 3D, 3E, S7B, and S7C). This is likely due to the transitory and less efficient nature of these treatments in comparison to generating frameshift mutations with CRISPR as in Figure 2.
The Minor satellites are, as frequently the case, left in the shadow of the Major satellites. For example, Figure S8C is not referenced at all in the text. This Figure  We appreciate the reviewer pointing this out. We have now also examined the regulation of the minor satellite repeat transcripts extensively.
 We did see a similar pattern of expression level change, but the relative expression and the fold change was much lower. In our RNA-seq data we again saw a similar pattern, but the effect was orders of magnitude greater for the major satellite transcripts as we saw in our RT-qPCR (see data below and new figure Fig S8D).  The very low read count (<10 in some samples) for the minor satellite prevented us from determining strand specificity in our RNA-seq data.  We also did not see BRD4 enrichment at the minor satellite consensus in our ChIP-seq and ChIP-qPCR data.
We have incorporated this additional figure and have added to the text to highlight these results: In contrast, the major satellite transcript was the most upregulated transcript in Dicer1 -/cells, and activation of major satellite transcription has been shown to cause chromosome segregation defects in mouse cells 56 . The minor satellite transcript on the other hand was very lowly expressed and relatively stable in all conditions (Figs. S8C and S8D). The abundance of the major satellite transcripts increased dramatically over the culture time-course, but was reduced in viable clones and in Dicer1 -/-d8 cells with Brd4 or Elp3 mutations (Fig. 4C).
The authors suggest that the partial rescue of MajSat transcription by Brd4 and Elp3 mutations is important for the suppressive activity of these mutants. However only associations are provided as evidence here, and thus 'Dicer1-/-Chromosomal Defects Depend on Major Satellite Transcription' in title of this section should be toned down. As the authors discuss, the altered MajSat transcription could be merely a consequence of cell cycle differences and irrelevant to the phenotypic outcomes, particularly because Brd4 and Elp3 have global effects on transcription (97 likely direct Brd4 direct targets in this context). To provide convincing evidence that the Dicer phenotype is dependent on MajSat transcription, they should directly downregulate MajSat reverse strand transcription, using a TALE or dCas9 fused to KRAB for example.
We have edited the title to reflect the reviewer's comments: This is likely due to the transitory and less efficient nature of these treatments in comparison to generating frameshift mutations with CRISPR as in Figure 2.

Dicer1 -/-Chromosomal Defects are Associated with Major Satellite Transcription
In order to investigate the mechanism of suppression, we performed BRD4 ChIP-seq in wild type and Dicer1 -/-mESCs. As discussed above (in response to Reviewer 1) we have now closely examined all 97 of these genes and found the major satellite repeat to be by far the most likely locus generating these phenotypes.
We agree that a direct perturbation of major satellite transcription would provide convincing evidence for its role in chromosome segregation. We have now attempted downregulation of the major satellite transcripts with ASOs at many concentrations spanning 5 orders of magnitude and targeting each strand individually as well as combined, but have not observed any amelioration of the viability/proliferation defects in Dicer1 -/-mESCs (below, new Fig S7L-M) potentially due to the massive magnitude of upregulation that we observe upon the loss of Dicer1 (Figs 1D, 4C, S8B). However, another group has shown that CRISPR activation of major satellite transcription in mouse cells does lead to chromosomal defects exactly as we describe (Zhu et al. 2018, Molecular Cell). This effect is conserved in humans and promotes cancer. Combined, these experiments indicate that it is the act of transcription rather than the transcripts themselves that is driving the phenotype. Most convincingly, we identify transcriptional activators that directly bind the major satellite loci (BRD4) as the strongest suppressors of the viability/proliferation defects as well as the chromosomal segregation defects in both mouse and yeast cells. These results strongly support our identification of major satellite transcription as the driver of the chromosomal phenotypes.
We have added these figures and updated the text: This transcript was also strongly downregulated in transcriptomes from Brd4 BD2-/-Dicer1 -/and Elp3 HAT +/-Dicer1 -/clonal double mutants (Fig. S8B).). Targeting of the major satellite transcripts with antisense oligonucleotides (ASOs) did not suppress the viability and proliferation defects of Dicer1 -/-mESCs across a range of concentrations or with strand-specific ASOs tested either individually or in combination (Figs. S7L and S7M). However, the mutation of Brd4 strikingly reversed strand-specific transcription of the satellite transcripts in cultured Dicer1 -/cells, closely resembling transcription in viable clones that had undergone the strong selection (Fig.  4B). Along with reduced BRD4 occupancy at satellite loci (Fig. S9A), and a reduction in elongating Pol II (Fig. S9B), chromosomal defects of Dicer1 -/cells were also rescued by Brd4 BD2-/and Elp3 HAT +/- (Fig. 4D). We further detected a substantial reduction of RAD21 at both the major and minor satellite loci in Brd4 BD2-/-Dicer1 -/double mutants (Fig. S9C)  findings that BRD4 interacts directly with RAD21 in human cells 57 and in D. melanogaster 58 as well as with NIPBL in humans 57,59 .
While we appreciate the suggestion of the reviewer to use CRISPRi dCas9-KRAB we found there to be a number of issues with this strategy.
 KRAB domains induce H3K9me3-dependent silencing, and we have found that H3K9me2/3 is increased in Dicer1-/-cells, rather than decreased (Fig S4C). Therefore, we predict that targeting this region with a KRAB domain may not induce further increases and transcriptional silencing.  This strategy generates heterochromatin indiscriminately of strand and will not allow us to test the strand-specific effect we observe as the reviewer suggests.  The hundreds of thousands of loci (though only a fraction of these are transcriptionally active) represent a major challenge for the CRISPRi system. The design of sgRNAs that target all of the sequence variations of major satellite transcripts and use the standard NGG PAM sequence in the AT-rich major satellite has been a major challenge, especially as the promoter region is unknown (Zhu et al, 2018). The level of reduction we observe in Brd4 suppressor mutants may be difficult to mimic with this strategy.
For coherence, it would also be interesting to determine whether this affects Cohesin loading, which itself rescues proliferation ( Figure 3I) (see next comment). Further, the reversal of transcription orientation of Major Satellites is intriguing, but again only correlated with phenotypic outcomes. Is the reported effect of ectopic expression of satellite transcripts on chromosome mis-segregation, strand specific? Are forward or reverse MajSat transcripts differentially associated with chromatin, and/or formation of DNA:RNA hybrids? Can they artificially switch major satellite transcription orientation to the forward strand in Dicer knockouts (using a similar dCas9/TALE strategy) and demonstrate a rescue of the phenotypes?
As detailed above we have tried extensively to use strand-specific perturbations such as ASOs to suppress the phenotype, but have been unable to demonstrate an effect on the viability and proliferation defects. While it has been clearly demonstrated that major satellite transcripts can associate with the chromatin and heterochromatin factors such as SUV39H1 (Comacho et al. 2017, eLife andJohnson et al. 2017, eLife) we observe very little change in heterochromatin state despite the massive upregulation of major satellite transcripts. This has led us to propose that it is the act of transcription rather than the transcripts themselves that is driving the phenotype. Thus, we can easily suppress the phenotypes by targeting transcription factors such as Brd4.
We have also considered a model in which DNA:RNA hybrids play a role. However, it has recently been demonstrated that BRD4 actually prevents the accumulation of R-loops (Lam et al. 2020Nature Comms, Edwards et al. 2020. Therefore, it is unlikely that the primary mechanism generating these phenotypes is related to DNA:RNA hybrids.
We have added to the text to reflect this consideration: We considered the possibility that the major satellite transcripts generate DNA:RNA hybrids or R-loops that lead to deleterious effects in Dicer1 -/-mESCs. However, our strongest suppressor, Brd4, has been shown to prevent the accumulation of R-loops 80,81 and so mutations would likely only further increase the accumulation of R-loops in Dicer1 -/cells. In the absence of RdRP, and consistent with the lack of small RNAs, there was no decrease in H3K9me2/3 at the pericentromere, while there was a decrease of H3K9me3 at retrotransposons that have corresponding small RNAs 44 .
Finally, as mentioned above the CRISPRi dCas9-KRAB/TALE strategies would not generate strand-specific disruption of major satellite transcription and would not allow us to test this hypothesis. We conclude that mutation of transcription factors such as Brd4 and Elp3 may be the only experimental paradigm powerful enough to combat the high levels of major satellite transcription in a strand-specific way.
While the identification of three categories of Dicer KO suppressors (transcription, H3K9 methylation and chromosome segregation) and their seemingly consistent associations is an important result, the mechanistic and functional link between them is under-investigated. Which, if any one of these activities, is the downstream mediator of phenotypic defects in Dicer knockouts? More evidence should be provided for the molecular associations between them, to determine causal relationships. For example, RAD21 binding at centromeres is not altered in Dicer single KOs compared to WT cells (S9C). How then can Cohesin binding at centromeres be responsible for the phenotypic defects observed in Dicer single KOs, as alluded to by the authors? Another implication is that the Suv39h1 and Ehmt1 suppressive effects occur through Cohesin. Is RAD21 loading at pericentromeic regions decreased in Suv39h1/Dicer double mutants? What about Major and Minor Satellite transcription? Is reverse orientation MajSat transcription also observed in this context?
We appreciate that additional clarification and interpretation of our results would help readers understand our model. While we do identify H3K9 methyltransferases as suppressors the K9me2/3 ChIP-seq and immunofluorescence experiments (Figs. S4 and S5) do not change significantly. This leads us to conclude that transcription is the critical mechanism underlying these phenotypes. To avoid being too vague in our interpretations, we have altered the text to reflect this.
In ES cells, we have shown that the Dicer1 -/viability defect is due to transcription of the centromeric satellite repeats, and can be rescued by hypomorphic mutations in transcription factors Brd4 and Elp3 or by inhibiting Pol II. In support of our findings, nuclear-localized DICER1 has been found associated with satellite repeats and their transcripts, and with the nuclear protein WDHD1 in complex with Pol II 29 and regulates transcription at this locus in multiple mouse cell types 29,73 . Dicer has been reported to interact with Pol II in Drosophila 74 and in human HEK293 cells, where it also prevents the accumulation of dsRNAs from satellite repeats 75 . Additionally, there are well-described interactions between Dicer and the transcriptional machinery in S. pombe 12,76,77 .
The fact that Brd4 and Elp3 were the strongest suppressors of the Dicer1 -/phenotype, which was also suppressed with low doses of -amanitin, strongly implies transcription as the underlying mechanism. We found strand-specific accumulation of major satellite transcripts in Dicer1 -/-mESCs that was reversed by Brd4 BD2-/mutations and by selection for viability that generated our clonal lines.
In fact, we find the clearest associations between classes of suppressors occur with transcription factors and chromosome segregation factors. While RAD21 does not accumulate in Dicer1 mutants as the reviewer points out, the mutation of Brd4 does significantly reduce RAD21 occupancy at the major satellite loci and overall reduction of RAD21 levels with siRNAs suppresses the viability and proliferation defects of the Dicer1 -/-mESCs (Fig. 3I). In fact, a missense mutation in the second bromodomain of BRD4 has been shown to cause a version of the cohesinopathy (Cornelia DeLange Syndrome) through the loss of cohesin loading by NIPBL in the BRD4 mutant (Olley et al. 2018 Nat. Genetics). Again, we propose it is not necessary for RAD21/Cohesin to over-accumulate and be reduced by targeting BRD4 or RAD21 directly, but rather that the defects caused by the loss of DICER1 at this region can be ameliorated by reducing Cohesin at the pericentromere.
While it is possible RAD21 loading is reduced by targeting H3K9 methyltransferases as well, we have not experimentally tested this hypothesis and we felt this investigation lies outside of the scope of this study. We found a much stronger effect in suppressing with transcription factors like BRD4 and decided to focus on this aspect.

Minor:
Page 5: 'both inhibition and overexpression of Aurora B' sentence not finished. Figure S7A is incorrectly cited as S7C We thank the reviewer for bringing our attention to these editorial errors and they have been corrected.
Reviewer #3 (Remarks to the Author): In the manuscript "Dicer promotes genome stability via the bromo-domain transcriptional coactivator BRD4" the authors address the role of Dicer in heterochromatic silencing. They find that deletion of Dicer leads to upregulation of major satellite repeats and reduced viability, which can be rescued by deletion of transcription activators such as Brd4. Furthermore, perturbations in transcription by chemical agents or siRNAs increase viability. This genetic interaction is conserved in fission yeast as well. Although the study offers new insights in Dicer role in heterochromatic silencing in mammalian cells, it does not go beyond genetic interactions. The study is missing more mechanistic data showing what actually Dicer does at these repeats and how it prevents Brd4 recruitment.
We thank the reviewer for their comments and hope we have demonstrated that we have indeed gone beyond only describing genetic interactions in this study.
1) The authors should provide more mechanistic data showing that Dicer is actually involved in silencing of major satellite repeats. The authors should show if Dicer activity is required for silencing and reduced viability. Although, the authors do not detect Dicer dependent small RNAs, it is possible that Dicer activity plays a role in the process. Alternatively, Dicer might directly interact with transcription machinery to regulate transcription in the repeats.
Our model is for a non-canonical role of Dicer in regulating the transcription of the major satellite repeats rather than the well-described small RNA-dependent silencing mechanism. As the reviewer points out we do not detect Dicer-dependent small RNAs and we have now additionally found that targeting Argonautes does not suppress the viability and proliferation defects (see response to Reviewer 1's point #2). We have clearly shown that Dicer is involved in directly regulating the transcription of pericentromeric repeats in both mouse and yeast. We find that the loss of Dicer leads to massive upregulation of pericentromeric repeat transcripts in both organisms (Figs. 1D, 4C, S8B, 5C). We have also demonstrated that the loss of Dicer leads to reduced viability in both organisms (Figs. S1B, 3A, 5A, and 5B). Additionally, the fact that mutations in the same transcription factors that directly bind and control transcription of the pericentromeric repeats suppress so strongly in two evolutionarily distant organisms is suggestive of a conserved mechanism. Finally, we have pointed to other studies supporting this direct regulation of major satellite transcripts across cell types in mouse:  A direct interaction between the DICER1 protein and the major satellite genomic loci as well as transcripts has been detected by ChIP and RNA pull-down respectively in mouse cells (Hsieh et al. 2011 Nucleic Acids Research).  Additionally, in mouse spermatogenesis the loss of DICER1 induces major satellite transcription in a strand-specific manner (as we observe) and furthermore, a direct association between the DICER1 protein and the major satellite genomic loci was detected using a combined immunofluorescence/in situ hybridization microscopy strategy as well as ChIP-PCR (Yadav et al. 2020 Nucleic Acids Research).
Nuclear interaction between DICER1 and RNA Pol II has been reported through the WDHD1 protein (Hsieh et al. 2011 Nucleic Acids Research). Additionally, there are well-known interactions between Dicer and transcription in S. pombe (Djupedal et al. 2007, Genes & Dev, Reyes-Turcu et al. 2011, NSMB, and Zaratiegui et al. 2011 We have updated the text to reflect this.
In ES cells, we have shown that the Dicer1 -/viability defect is due to transcription of the centromeric satellite repeats, and can be rescued by hypomorphic mutations in transcription factors Brd4 and Elp3 or by inhibiting Pol II. In support of our findings, nuclear-localized DICER1 has been found associated with satellite repeats and their transcripts, and with the nuclear protein WDHD1 in complex with Pol II 29 and regulates transcription at this locus in multiple mouse cell types 29,73 . Dicer has been reported to interact with Pol II in Drosophila 74 and in human HEK293 cells, where it also prevents the accumulation of dsRNAs from satellite repeats 75 . Additionally, there are well-described interactions between Dicer and the transcriptional machinery in S. pombe 12,76,77 .
2) The data suggest that Dicer might somehow regulate transcription, but it is unclear if this is the case. The authors should at least determine if Dicer is localized to the repeats and if it interacts with transcription machinery.
As detailed above other groups have shown across mouse cell types a direct interaction between the DICER1 protein and both the major satellite genomic loci by ChIP and the RNA transcripts and that a loss of DICER1 at these loci leads to upregulation of major satellite transcripts (as we observe).
3) The authors show that major satellite repeats are upregulated in Dicer deletion cells. This suggest that Dicer is involved in silencing of those repeats. They observe that in Dicer/Brd4 double mutants the transcripts are still heavily upregulated, however, the opposite strand is now being transcribed. To me it is unclear how this switch in transcription happens and how it suppresses interferon response. The authors should provide more data that the transcription from the opposite strand is resulting in increased viability.
Interesting observation is also that in wild type cells both + and -strands from repeats are transcribed, while in Dicer deletion only + strand and in Dicer/Brd4 deletion cells only -strand is transcribed. The authors should comment why Dicer deletion upregulates + strand and silences -strand.
We observe a massive upregulation of the major satellite transcripts in Dicer1 mutants and a very significant reduction in major satellite transcripts in the Dicer1/Brd4 double mutants, though as the reviewer points out this is still upregulated relative to wild types. We did note the strand-specific trend that appears in the Dicer1 -/-escaping clones and double mutants with Brd4, but we feel a full description of the mechanism of strand specificity is a subject for a follow up study. We have attempted to repress the major satellite transcripts in a strandspecific manner with ASOs, but this had not led to an improvement in viability despite extensive variation across a range of concentrations and target sites (see data in response to Reviewer 2's point 3).
We would like to point out that our data does not suggest that only the + strand is being transcribed in Dicer1 -/-cells or that only the -strand is being transcribed in suppressed cells, but rather that there is an enrichment for these strands. The data being plotted in Fig. 4B is a log2 transformation of the ratio of forward to reverse. We found upregulation of both