Competitive endogenous RNA is an intrinsic component of EMT regulatory circuits and modulates EMT

The competitive endogenous RNA (ceRNA) hypothesis suggests an intrinsic mechanism to regulate biological processes. However, whether the dynamic changes of ceRNAs can modulate miRNA activities remains controversial. Here, we examine the dynamics of ceRNAs during TGF-β-induced epithelial-to-mesenchymal transition (EMT). We observe that TGFBI, a transcript highly induced during EMT in A549 cells, acts as the ceRNA for miR-21 to modulate EMT. We further identify FN1 as the ceRNA for miR-200c in the canonical SNAIL-ZEB-miR200 circuit in MCF10A cells. Experimental assays and computational simulations demonstrate that the dynamically induced ceRNAs are directly coupled with the canonical double negative feedback loops and are critical to the induction of EMT. These results help to establish the relevance of ceRNA in cancer EMT and suggest that ceRNA is an intrinsic component of the EMT regulatory circuit and may represent a potential target to disrupt EMT during tumorigenesis.

far as I know that are reports for both decreased and increased overall miRNA abundance in tumors. Is there evidence that the overall miRNA abundance in the cells studied is indeed low compared to hepatocytes or other cell types where the ceRNA hypothesis was studied before? This should be cited or shown. 5) Which isoform of FN1 is induced? According to TargetScan there are two isoforms and only one of them is targeted by mir-200.
Reviewer #2 (Remarks to the Author) In this paper, the authors investigate the importance of ceRNA regulation within the context of EMT. To do so, they characterize two specific examples of ceRNA-miRNA pairs in two different cell lines, and conclude that ceRNA regulation occurs and plays an important role in EMT. Furthermore, evaluation of miRNA and ceRNA stoichiometry provides some support for ceRNA in this specific cancer cell line context. C omments: Major: 1. The authors provide some evidence pointing to a relevant ceRNA mediated mechanism modulating the dynamics of EMT. However, like numerous previous studies they fall short of proving this genetically. The authors should capitalize on their cell system and mutate the miR-21 and miR-200c binding sites (by single point mutations) in the TGFBI and FN1 genes, respectively, using readily available genome editing approaches and test if this affects EMT. These are key experiments that that are essential for the conclusions of the manuscript. 2. The authors, throughout the manuscript, measure EMT markers and study cell lines in surrogate assays, without determining of whether the postulated ceRNA-miRNA-target gene networks are relevant in in vivo conditions (i.e. metastasizing and non-metastasizing mouse or human tumors). 3. A recent study by Denzler et al. (Mol C ell. 64: 565-579, 2016) found that cooperative binding of proximal sites for the same or different miRNAs can increase potency and therefore make ceRNA effects more likely. If such closely-spaced sites exist should be investigated and discussed. 4. In lines 104-116, the authors state that there is a discrepancy between a long miR-200c halflife and rapid ZEB1/C DH1/C DH2 dynamics, concluding that unknown mechanisms must modulate miR-200c expression in order to resolve this issue. This view seems reductionist and ignores factors other than miR-200c that may regulate ZEB1 and its downstream effectors, such as SNAIL1 (Dave et al., 2011, JBC ). 5. There is some contradiction related to the point at which FOXP1 reaches equilibrium -is it 48h (line 157) or 96h (line 168)? This is relevant to evaluating the dynamics of the system. 6. Does miR-21 regulate other EMT genes? The authors state that miR-21 knockdown leads to enhanced , but there is no experimental evidence that this is occurring through FOXP1 repression. In order to make this claim, it should be experimentally shown that mutation of the FOXP1 miR-21 binding site prevents enhanced migration/invasion upon miR-21 knockdown. 7. There is no explanation as to why different cell lines are used to study the miR-21-TGFBI and miR-200c-FN1 regulatory networks. Expression levels of both networks in both cell lines would be informative to help understand stoichiometric requirements for such ceRNA events to occur, or explain why they do not occur under specific circumstances. Furthermore, the authors should discuss the evidence that the observed expression changes/dynamics are relevant for certain proimary (as opposed to cell lines) cancer types.
Minor: 1. Lines 292-301 should be rewritten more clearly, as it is not clear to what "transcriptional repression" refers (presumably to ZEB1 repression of miR-200c?), nor how the time points were selected (how is the point of "maximal ceRNA potency" calculated?).

Reviewer #3 (Remarks to the Author)
The manuscript by Du et al reports that TGFbeta and Fibronectin mRNA can act as ceRNAs in the control of EMT in A549 lung adenocarcinoma cells and MC F10A immortalized mammary cells respectively. The authors show that the level of expression of miR-21 and miR-200C are modulated in the presence of their corresponding ceRNA. The kinetics model initially developed by Tian et al., (2015) has been modified to introduce the contribution of the ceRNA. The findings are of interest to further unravel mechanisms driving epithelial cell plasticity. Remarks A 549 adenocarcinoma cells are known to be morphologically very heterogeneous with clusters of epithelial-like cells scattered among cells with a pronounced mesenchymal morphology. MC F1OA has a basal-like phenotype as most normal-like mammary epithelial cells and again raises issues in terms of epithelial mesenchymal transition. The authors should at least discuss this issue. The result section describing Figure 6 needs to be considerably improved; Each frame should be described accurately. C learly the reversibility to an epithelia-like morphology is not obvious.

Reviewers' comments:
Reviewer #1 (Remarks to the Author): The authors describe ceRNA effects of specific genes during the induction of EMT, which is an important process in tumor biology. The subject is overall of interest. The paper is well written and data presented very clearly. The first parts, pertaining to the roles of FOXP1 in EMT and the regulation of FOXP1 by miR-21, as well as the parts where the authors study transcriptional regulation of TGFBI and NF1, are solid and rigorous. The main claims in the paper and most of the novelty pertain to the ceRNA hypothesis, and they are substantially more questionable, as described below.
Major comments: 1) The authors describe potential ceRNA activity taking place during EMT, driven by high expression of single target genes-TGFBI for mir-21 and FN1 for mi-200c. In both cases ceRNA activity is reported to be driven by a single strong target site. The authors claim that these effects are plausible because the target abundance of these genes approaches the miRNA abundance, though recent computational analysis and data (reviewed in Jens and Rajewsky Nature Reviews Genetics 2015) suggest that even such levels for a single mRNA carrying a single target site are unlikely to be consequential, as the number of other binding sites for the miRNA in the transcriptome and overall target occupancy need to be considered.
Jens and Rajewsky also provide specific criteria which need to be met for the ceRNA effect to be likely. The paper by Boson et al. from the Sharp lab that the author cite (ref. 11) also concludes that competition is unlikely when the miRNA levels are high, as they are for mir-21 and mir-200c (>1,000 copies per cell, according to the authors). In the discussion, the authors claim that "A central debate concerning the physiological relevance of ceRNA is whether the expression level of ceRNAs could approach the abundance of miRNAs regulating a critical biological process.", but actually the debate is more about whether a single gene, no matter how abundant, with a single target site, can compete against the pool of thousands of sites found in the rest of the expressed transcriptome. Do the changes observed by the authors meet the criteria set forth in the published quantitative models of ceRNA function from the Rajewsky, Bartel and Sharp labs? That seems unlikely, and if they do, it should be specifically explained and discussed Response: We thank reviewer for pointing out this critical issue of our study, which we didn't address with enough details in our original manuscript. As the reviewer points out, the issue is "whether a single gene, no matter how abundant, with a single target site, can compete against the pool of thousands of sites found in the rest of the expressed transcriptome.". We specifically addressed this issue by performing mathematical modeling analyses as suggested by the reviewer using the models developed by the Rajewsky and We then first adopted the model of Rajewsky lab and calculated miRNA binding site occupancy during EMT using the "simplified model" as described by the authors. The mathematical modeling clearly showed that the dynamic expression change of FN1 or TGFBI during EMT can substantially change the MRE occupancies. Specifically, without TGFBI, the miR-21 site occupancy is consistently over 90% during the entire course of EMT in A549 cells. In sharp contrast, the miR-21 site occupancy in FOXP1 dropped to around 65-70% (targetScan-based MREs) at 12 to 36 hours when the MRE from TGFBI is included (Updated Fig3B). Simulations using pictar-based MREs generate a similar pattern in MRE dynamics, where the inclusion of TGFBI MREs induced a drop of site occupancy at 12 to 36 hours into EMT, albeit of a smaller magnitude (site occupancy dropped from ~97% to 88%, supplementary Fig. 4D). Although the site occupancy dynamics during EMT in MCF10A cells is different, a clear reduction in site occupancy induced by FN1 is also observed. When the MRE from FN1 is excluded, the miR-200c site occupancy declines gradually from around 90% at the start of EMT, but still above 80% at 120 hours into EMT using pictar-based MREs. In sharp contrast, the addition of FN1 MREs dramatically accelerated the reduction of miR-200c site occupancy, and the miR-200c site occupancy dropped from 85% without FN1 to around 55% with FN1 (pictar-based MRE) at 72 hours into EMT (Fig. 4B). A similar albeit smaller effect of FN1 is also observed with targetScan-based MREs, where the site occupancy dropped from ~50% to 35% at 72 hours into EMT ( Supplementary Fig. 7C).
We also performed simulations using the model from the Sharp lab, which is mathematically equivalent to the "simplified version" of the Rajewsky model. We implemented a Python routine based on the scipy library to solve the equations based on the Sharp model. As expected, identical changes of site occupancies were obtained despite the different solvers utilized. Hence these data are not included in the updated manuscript.
Although the simulated reduction in site occupancy owing to ceRNA is only upto 30%, representing a mild ceRNA effect, site occupancy changes at the estimated magnitude could indeed has a potent downsteam effect owing to the following observations. First, the MREs were estimated by miRNA binding site prediction programs, which are known to have  (2):234-245.). However, the miRNA binding sites for FN1 and TGFBI are experimentally validated. Hence, we expect that the MRE from FN1 or TGFBI could represent an even greater portion of the total MREs than current estimation, and consequently, their expression changes could lead to a larger change in site occupancy, and a bigger ceRNA effect. Secondly, and more importantly, the ceRNA interactions are tightly coupled with double negative feedback loops between ZEB-miR-200c and FOXP1-miR-21 in our study. A critical characteristic of double negative feedback loops is that they can generate switch like behavior, where a small change in input leads to dramatic expression changes of coupled molecules, a phenomenon known as hypersensitivity. Thus it is highly likely that the ceRNA signals generated by the site occupancy changes are amplified by the feedback loops, resulting in substantial changes in the downstream molecules during EMT, which is consistent with our experimental observations and mathematical modeling results.
Taken together, these new analyses demonstrated that a single highly induced mRNA can indeed display noticeable ceRNA effects in these EMT models, given that the MREs from other RNAs are of smaller number, and that the ceRNA signals are tightly coupled with the hypersensitive double feedback loops. We have incorporated these modeling results into  2) As the ceRNA effects of individual mRNAs with a single binding site are not plausible, the authors must very convincingly how that direct competition is indeed happening in their setting. Currently, this is shown using individual reporter assays, but these may suffer from indirect effects. Further, the authors used siRNAs for knockdown of TGFBI and FN1, which might saturate the RISC machinery, thus indirectly affecting microRNA efficacy. To address this, the authors need to: a) show using RNA-seq or microarray analysis that there is specific de-repression of target genes containing seed sites for miR-21 and miR-200c in the two systems when TGFBI and FN1 are perturbed(this is best done using Sylamer analysis); b) show what happens when the ceRNAs are targeted with a different method -ideally, by mutating the specific target sites in TGFBI and FN1 using CRISPR (as suggested by Jen and Rajewsky in their review) or using CRISPRi for target silencing, which would not suffer from potential RISC overloading effect; c) show that the effects of TGFBI on FOXP1 and of FN1 on ZEB1 are dependent on the microRNA pathway (e.g., by deleting Dicer and showing the effect is abrogated).

Response:
We agree that more convincing evidences are needed to support our conclusion and have performed additional experiments/analyses as suggested by the reviewer.
First, we carried out new experiments to show that there is specific de-repression of target genes containing seed sites for miR-21 and miR-200c in the two systems when TGFBI and FN1 are perturbed. Although Sylamer analysis can provide a systematic answer, it is not suitable in our case. For both miR-21 and miR-200c, the estimated ceRNA effects from our response to comment # 1 is mild, suggesting that the direct gene expression changes induced by ceRNA effects will not be dramatic. RNA-seq or microarray are not robust to detect small differential expressions changes, and consequently, Sylamer analysis probably will not produce reliable results. To combat this issue, we performed more sensitive qPCR analyses on a large set of genes with experimentally validated miR-21 or miR-200c binding sites. To control for indirect effect, we selected genes with experimentally validated binding sites for other high expressing miRNAs (the second highest expressing let-7 in A549, and the highest expressing miR-385 in MCF10A) but lacking binding sites for miR-21 or miR-200c. Consistent with our model, we consistently observed expression changes in genes with MREs for miR-21 or miR-200 when perturbing TGFBI or FN1, and no significant changes in control genes, confirming that the observed ceRNA effects are direct and highly specific. The qPCR results have been included in supplementary Figure 6 and 10.
Second, we have utilized CRISPR/Cas9 technology to mutate the seed region of miR-21 binding site in TGFBI 3'UTR and seed region of the miR-200c binding site in FN1 3'UTR, and repeated the same assays originally performed with siRNAs. Consistent with the original siRNA data, knocking out binding site in FN1 or MCF10A completely abolishes ceRNA effect, confirming that the observed ceRNA effect is not due to RISC overloading, but require the functions miRNA binding sites in putative ceRNA molecules (updated Fig. 3-4, supplementary Fig. 3, 5-6).
Finally, we also deleted Dicer in A549 (via CRISPR/Cas9) and MCF10A (siRNA based) and confirmed that the ceRNA effects of TGFBI or FN1 are abrogated without mature miRNAs, confirming that the effects of TGFBI on FOXP1 and of FN1 on ZEB1 are dependent on the microRNA pathway. Unfortunately, despite multiple attempts, we couldn't obtain viable MCF10A clones that harbor DICER mutation using CRISPR/Cas9-based approaches, and have to settle for siRNA based approach to delete DICER in MCF10A cells.
(updated Fig. 3-4, supplementary Fig. 3, 5-6) 3) The ceRNA effect is supposed to happen by competition for miRNA binding, but regular sites are not supposed to reduce the expression level of the miRNA. How do the authors explain the increase in miR-21 levels when using siTGFBI in Fig. 3C, which is abrogated when the TGFBI 3'UTR is used? Especially since mir-21 is not reduced when TGFBI is induced naturally during EMT (Fig. 3B)? Similarly, how is the induction in miR-200c levels in Fig. 4B explained? Response: This is a very sharp observation. Indeed, the ceRNA effects are not supposed to reduce miRNA expression. However, the system analyzed in our study is very special because the ceRNA interactions are directly coupled with double negative feedback loops regulating EMT. Hence, the observed miRNA expression changes are direct outcomes of the double negative feedback loops between ZEB1-miR-200c and FOXP1-miR-21. Consistent with a ceRNA effect of TGFBI, knocking down TGFBI will release more functional miR-21 molecules without changing the total number of miR-21 molecules, which will in turn lead to a reduced level of FOXP1, which is confirmed by results in Fig 3C. Because FOXP1 and miR-21 form a double negative feedback loop (which we also confirmed in Figure 2), less FOXP1 naturally leads to upregulated expression of miR-21.
As for the reason why miR-21 is not reduced when TGFBI is induced naturally during EMT, it is expected when mechanism other than TGFBI is included into the picture. The expression of miR-21 is regulated simultaneously by multiple mechanisms, including repression by FOXP1 as shown in Figure 2, induction via TGF-beta stimulation through SMAD based mechanism (Davis et al.,Nature. 3;, and other undocumented mechanisms. Hence the level of miR-21 is the combined effects of multiple mechanisms, and lack of repression when TGFBI is induced during EMT is expected since studies have reported that TGF-beta treatment can strongly induce miR-21 via SMAD based mechanisms, which is also confirmed by our results. This is also consistent with the downregulation of miR-21 when TGFBI is knock down, because other factors influence miR-21 levels are undisrupted while FOXP1 activity is reduced. The upregulation of miR-21 actually highlights the critical role of TGFBI as an ceRNA during EMT, whose effects are indispensable for abolishing the inhibitory effect of miR-21 to induce effective EMT in A549 cells as shown by our analyses.
Finally, the observed upregulation of miR-200c can be explained similarly to miR-21: knocking down FN1 abolished its ceRNA effects, which lead to more functional miR-200c.
More functional miR-200c leads to lower level of ZEB1, which in turn de-repress miR-200c expression owing to the double negative feedback loop between ZEB1 and miR-200c, resulting in a higher level of miR-200c.
The coupling of the ceRNA effect with double negative feedback loops can also explain why perturbing miRNA activities through ceRNA, which are generally considered to be less potent than these of TFs, can lead to substantial expression changes of gene in EMT.
Previous study demonstrated that the double negative feedback loop in EMT generate bistability, a manifestation of which is the switch like hypersensitive response of induced gene expression. Critically, the timing of upregulation of ceRNA activity is aligned with the timing of the switch, which is confirmed by simulation shown in Figure 5. Hence, a relative small effect induced by ceRNA can be readily amplified by the bi-stability switch, resulting in substantial gene expression changes.
Minor comments: 4) The authors claim that cancer cells "typically express lower levels of miRNAs" (Page 4), but as far as I know that are reports for both decreased and increased overall miRNA abundance in tumors. Is there evidence that the overall miRNA abundance in the cells studied is indeed low compared to hepatocytes or other cell types where the ceRNA hypothesis was studied before? This should be cited or shown.

Response:
We agree that our original description is biased and have updated the references to reflect the fact that both decreased and increased overall miRNA abundance in tumors are reported.
More importantly, experimental data confirmed that the overall miRNA abundance in the cells used in our study is indeed lower compared to hepatocytes or other cell types. In the cancerous A549 cells, miRNA-seq data demonstrated that miR-21 is the highest expressing miRNA. Critically, absolute quantification showed that the absolute count of miR-21 molecules in resting A549 cells is 2313±200 copies/cell, and 4220±384 copies/cell at 96h after TGF-beta stimulation, which is about two order of magnitudes lower than the highest expressing miRNA in hepatocytes (miR-122, 1.2 x 10 5 copies/cell). Moreover, similar to reported miRNA expression landscapes, the overall miRNA expression in A549 cells is dominated by a few highly expressed miRNAs and the vast majority of miRNAs are expressed at very low level. In fact, in A549 cells the total number of reads mapped to miR-21 alone is about 20% of all mapped reads, which put the total number of miRNA molecules in A549 cells at about 10,000 copies/cell (assuming a linear relationship between the number of mapped miRNA-SEQ reads and the absolute count of corresponding molecules). Using a similar strategy, we estimates that there are about 4.4 x 10 5 miRNA molecules in hepatocytes.
Thus experimental data suggested that the total number of miRNAs in A549 cell is about 44 times lower than the total number of miRNAs in hepatocytes. Although the mechanism of broad miRNA downregulation in A549 cells is not clear, MYC, the oncogene that is highly A different scenario is observed in the benign MCF10A cells. Briefly, miRNA-seq data shows that miR-200c is the 15 th highest expressing miRNA in MCF10A (211,178 rpkm) and its absolution quantification is 1396±240 copies/cell (resting) and 1094±39 copies/cell (96h after TGF-beta stimulation). The highest expressing miRNA in MCF10A is miR-378a-3p at 5,886,078 rpkm. Assuming a linear mapping between absolute quantification and miRNA-seq, we estimate there are about 38,910 miR-378a-3p molecules in MCF10A cells, and about 2 x 10 5 total miRNA molecules in MCF10A cells. Consequently, the overall miRNA is about 50% lower than the total number of miRNAs in hepatocytes (4.4 x 10 5 ).
The fact that miR-200c is only the 15 th highest expressing miRNA in MCF10A cells raises a natural question: whether the other 14 high expressing miRNAs influence the ceRNA interactions mediated by miR-200c. Because the high expressing miRNAs (such as miR-378a-3p) outnumber FN1 by an order of magnitude, if any of these miRNAs also bind to ZEB1, then FN1 will not be able to modulate ZEB1 activity via ceRNA-based mechanism.
Critically, the targetScan results demonstrates that none of the 14 high expressing miRNAs  In this paper, the authors investigate the importance of ceRNA regulation within the context of EMT. To do so, they characterize two specific examples of ceRNA-miRNA pairs in two different cell lines, and conclude that ceRNA regulation occurs and plays an important role in EMT. Furthermore, evaluation of miRNA and ceRNA stoichiometry provides some support for ceRNA in this specific cancer cell line context.

Comments:
Major: 1. The authors provide some evidence pointing to a relevant ceRNA mediated mechanism modulating the dynamics of EMT. However, like numerous previous studies they fall short of proving this genetically. The authors should capitalize on their cell system and mutate the miR-21 and miR-200c binding sites (by single point mutations) in the TGFBI and FN1 genes, respectively, using readily available genome editing approaches and test if this affects EMT.
These are key experiments that that are essential for the conclusions of the manuscript.

Response:
We agree with review that genetic experiments are critical to support our results, which is also pointed out by reviewer #1.
As described in our response to comment #2 of reviewer #1, we have mutated the binding sites by CRISPR/Cas9 and performed new experiments, which confirmed that FN1 and TGFBI indeed demonstrate ceRNA effect during EMT.
2. The authors, throughout the manuscript, measure EMT markers and study cell lines in surrogate assays, without determining of whether the postulated ceRNA-miRNA-target gene networks are relevant in in vivo conditions (i.e. metastasizing and non-metastasizing mouse or human tumors).

Response:
We agree that in vivo data will substantially support our models. To this aim, we performed survival analyses of three independent lung cancer data sets. Critically, in all analyzed data sets, patients expressing a higher level of TGFBI consistently demonstrate a worse clinical outcome with a higher risk of relapse-free survival, supporting the notion that overexpression TGFBI promotes EMT and metastasis. We didn't perform similar analyses with FN1 because the miR-200s families of miRNAs are routinely downregulated in breast cancer, rendering the FN1-miR-200c axis irrelevant in breast cancer. The survival analyses results have been added into Figure 3H. Denzler et al. (Mol Cell. 64: 565-579, 2016) found that cooperative binding of proximal sites for the same or different miRNAs can increase potency and therefore make ceRNA effects more likely. If such closely-spaced sites exist should be investigated and discussed.

A recent study by
Response: This is a very relevant point to examine. Unfortunately, no such proximal sites exist in the analyzed mRNAs. Although pictar predicted two miR-200c binding sites in the 3'UTR of FN1, these two predicted sites are > 200 bp apart. Critically, only one of the two predicted site is functional. The functional site has been previously experimentally validated and its function also supported by our data. The other site is not functional because knocking out the site by CRISPR/Cas9 has no impact on its putative ceRNA effects (supplementary Figure 9B and 10A). Moreover, predicted secondary structure demonstrated that seed region of the predicted nonfunctional site locates in stem regions, further suggesting that the site is not functional (supplementary Figure 9F).  Response: These issues are critical for successfully utilizing A549 and MCF10A as cellular models for EMT. We didn't discuss these because we are following the practice of the These technical details have been addressed in the updated manuscript and should help other researchers using A549 or MCF10A as models for EMT.
The result section describing Figure 6 needs to be considerably improved; Each frame should be described accurately. Clearly the reversibility to an epithelia-like morphology is not obvious.

Response:
We have split Figure 6 into two separated figures (A549 data in Figure 6, and MCF10A data in Figure 7). In the new figures, each frame is individually labeled and provides a clear and unbiased reference. We have also revised the manuscript and described the results with greater detail and improved accuracy according to the improved figures, which should make the reversibility to an epithelia-like morphology clear and easy to follow.