Methylated cis-regulatory elements mediate KLF4-dependent gene transactivation and cell migration

Essential revisions:

1) New experimental evidence is required to fully establish the DNA methylation dependent binding of wild type and the mutant form. Control experiments with Klf4 mutant lacking DNA binding potential altogether needs to be conducted; DNA methylation and motif analysis of the cell systems need to be performed.

To fully examine the DNA methylation status of KLF4 WT and R458A ChIP-seq peaks, we performed whole genome bisulfite sequencing to decode the methylome of U87 cells and combined the DNA methylome data separately with the KLF4 WT and R458A ChIP-seq datasets. We found that 66% of the KLF4 WT-specific ChIP-seq peaks showed a high methylation level (e.g., β > 60%) at CpG sites, while only 36% of the ChIP-seq peaks shared by KLF4 WT and R458A reached a similar CpG methylation level (p = 3.7e-223). Different cutoffs for defining high methylation levels did not alter this observation (Figure 3). Therefore, the KLF4 WT-specific ChIP peaks are enriched for highly methylated CpGs. In the revised manuscript, we have incorporated the new data in Figure 3F.

Next, we carried out motif analysis to identify enriched and highly methylated 6-mer DNA motifs in WT-specific ChIP-seq peaks, as well as in shared ChIP-seq peaks. At a cutoff of β > 60% CpG methylation we found 10 methylated 6-mer motifs that were significantly over-represented in the WT-specific peaks (p = 6.6e-37). Many of them share sequence similarity to the motif 5’-CCCGCC (Figure 3G, of which the methylated form was reported to be recognized by KLF4 in our previous study (Hu et al., eLife 2013). In contrast, the peaks shared by WT and R458A were found to be enriched for different motifs (e.g., 5’-AAAAGGAA and 5’- GAGTTGAA). The new data is now in Figure 3G. Taken together, these new data confirmed that the KLF4 WT-specific ChIP-seq peaks were enriched for highly methylated KLF4 binding motifs.

We respectfully disagree that “control experiments with Klf4 mutant lacking DNA binding potential altogether needs to be conducted” because it would not further strengthen the conclusion of this study. In a previous study, the Goldberg group nicely showed that induction of a KLF4ΔC (lacking the entire KLF4 DNA binding domain) in mouse retinal ganglion cells could no longer suppress the axon outgrowth as the WT did (Moore et al., Science 2009 326(5950):298-301). Similarly, induction of the R458A mutant on the same background showed the same phenotype as KLF4ΔC (Goldberg; personal communication), suggesting the removal of the entire DNA binding domain of KLF4 did not cause any additional phenotype. Even if a truncated KLF4 mutant might cause additional phenotypes, it is not the focus of this study. In our opinion, the R458A mutant is the most relevant comparison with KLF4 WT because it loses the ability to bind to methylated DNA motifs while maintaining binding activity to KLF4’s canonical motif (Hu et al., eLife 2013). Using a truncated KLF4 mutant that lacks the entire DNA binding domains would not help us to identify mCpG-dependent KLF4 targets because we would not be able to distinguish KLF4-mCpG binding from the canonical KLF4 binding activity.

2) Causality of DNA methylation dependent binding of Klf4 on the genome and induction of some target genes needs to be more fully established. As pointed by reviewer #3, the presence of the binding within 10kb of the promoter does not prove direct targeting relationship. New experimental data should be provided to support the notion that a target gene is directly regulated by Klf4's binding to the methylated DNA near the gene. This could be CRISPR/cas9 mediated genome editing or other means.

We thank the reviewers for this constructive suggestion. On the basis of the reviewers’ suggestion, we selected two genes, namely RAC1 and RHOC, because 1) they are known to play a crucial role in cell migration; 2) we found that they were associated with WT-specific ChIP-seq peaks and genome-wide bisulfite sequencing confirmed that these peaks contained highly methylated CpGs; and 3) these peaks are in close proximity to the coding regions of RAC1 (-922 bp to TSS) and RHOC (-507bp to TSS). In a newly added case of RAC1, we first showed that induction of KLF4 WT, but not the R458A mutant, induced RAC1 expression at both mRNA and protein levels using RT-PCR and immunoblot analysis, respectively (Figure 5—figure supplement 1A and B). Using a primer pair flanking RAC1 ChIP-seq peak, we confirmed that this fragment could be ChIPed substantially better with the KLF4 WT than the R458A mutant (Figure 5—figure supplement 1 C). To further establish the causality of DNA methylation-dependent binding of KLF4 on induction of its target genes, we performed bisulfite sequencing against this cis-regulatory element with and without the 5-Aza treatment. We observed that the methylation level of four of the six CpGs was reduced from ~100% to an undetectable level after 5-Aza treatment (Figure 5—figure supplement 1D). Most importantly, removal of methylation almost completely abolished KLF4 WT binding to this cis-regulatory element (Figure 5—figure supplement 1E), and expression of RAC1 was also significantly repressed (p < 0.001; Figure 5—figure supplement 1F).

In the case of RHOC, we carried out additional assays to further establish the causal relationship. We performed ChIP-PCR against its cis-regulatory element (-507 bp to TSS) with and without 5-Aza treatment and observed that reduction of methylation in this region also substantially reduced recruitment of KLF4 WT. Taken together, these new results demonstrate that KLF4 WT activated the expression of these two genes via binding to their highly methylated cis-regulatory elements nearby the coding regions. The new data is added as panel Figure 5E and Figure 5—figure supplement 1).

To examine the impact of DNA demethylation on the global expression level, we performed RNA-seq analysis and compared changes in gene expression profiles with and without 5-Aza treatment. We observed that in the absence of KLF4 WT induction the overall expression levels of the 116 KLF4 WT-specific target genes were not significantly affected by 5-Aza treatment (Figure 4), as expected. However, when KLF4 WT was induced, the expression level of 101 of the 116 genes was reduced in 5-Aza treated cells as compared with untreated cells (Figure R4, right panel; p = 1.2e-14 compared to all expressed genes), suggesting that DNA methylation is responsible for the transcription activation of these 101 genes. The new data is in Figure 4F and G.

In our opinion, using CRISPR/Cas9-mediated genome editing may not help strengthen the causality because the mCpG-dependent KLF4 DNA binding activity is also sequence-specific (Hu et al., eLife 2013). In other words, the use of CRISPR/Cas9 to mutate the DNA sequence (i.e., CpG sites) of interest would not assist us to de-couple the effects of DNA sequence and DNA methylation on the KLF4 binding activity. Ideally, a CRISPR-guided site-specific DNA demethylation approach would be appropriate; however, this method might not guarantee maintenance of low DNA methylation level due to the endogenous DNA methyltransferase activity.

3) The treatment of cells by 5-Aza is inducing global DNA demethylation, and the observed phenotypes in the cell system can be attributed to non-specific effects. The authors need to discuss this aspect of the experiments to avoid drawing un-substantiated conclusions.

We agree with the reviewers and have added the following discussion to address this concern in Discussion section.

“We are fully aware that 5-Aza is only a sub-optimal approach to examine the effect of specific methylation sites, because it induces global demethylation and may have some non-specific effects (Christman et al., 2002). However, we have multiple lines of evidence showing that the observed phenotypes are due to specific interactions between KLF4 and methylated sites. (1) In our control experiments when KLF4 R458A cells were treated with 5-Aza, we did not see changes in gene expression and cell migration. (2) In the focused studies of RHOC and RAC1, we showed that DNA methylation is responsible for KLF4 binding and gene activation. RHOC and RAC1 are major GTPase involved in cell migration and motility. (3) Comparison of ChIP-seq between KLF4 WT and R458A showed that WT-specific binding sites are enriched for highly methylated sites. Given the phenotypic differences between KLF4 WT and R458A, the result suggested that the observed phenotypes in WT are likely due to specific interactions between KLF4 and DNA methylation sites. (4) Our new RNA-seq data in the presence of 5-Aza also indicated that 5-Aza globally reversed the up-regulation of many KLF4-mCpG targets.”

4) Statistical methods need to be better described and experimental data need to be more rigorously interpreted. Please check their comments carefully and make sure to remove or modify the particular statements in the revised manuscript.

We thank reviewers for the constructive suggestion on the statistics analyses. We have added more rigorous statistical analyses throughout the revised manuscript according to the reviewers’ suggestions. Detailed responses can be found in the specific points raised by each reviewer. Please refer to our responses to reviewer #1’s major points 1 and 2; reviewer #2’s major points 1, 2, 3 and 7; reviewer #3’s major points 1 and 2. All of these new analyses have been incorporated into the revised manuscript.

Reviewer #1

[…] Major problems:

1) Differential binding of WT and R458A Klf4 in GBM cells: Is this due to failure of the mutant protein to bind methylated DNA? Unfortunately, the evidence presented is weak: the authors performed Bisulfite Sequencing on 15 sites and found that 10 of them contain methylated CpG. Are these DNA sequences recognized by WT Klf4 but not mutant klf4? EMSA should be performed to establish this point. Further, 15 is too small a number to generalize to the rest of the differential binding sites. It is important to carry out base-resolution analysis of DNA methylation in these cells and determined if differential DNA binding by WT and R458A Klf4 in the GBM cells is indeed correlated with presence of DNA methylation at the Klf4 binding motifs in these regions.

We appreciate the reviewer’s insightful suggestion. We have performed the whole genome bisulfite sequencing in GBM cells and combined this new dataset with KLF4 WT and R458A mutant ChIP-seq datasets. Please see our responses to essential revision #1 and #2 above for more details. Here are some highlights that are relevant to answer the reviewer’s questions: 1) WT-specific ChIP-seq peaks contained CpGs of a much higher methylation level than those peaks shared by the WT and R458A mutant (Figure 3) DNA motif analyses against those WT-specific peaks identified 10 significantly enriched motifs, all of which contained at least one CpG with significantly higher methylation level than those identified among the shared peaks (Figure 3G). More importantly, in-depth in vivo studies demonstrated that removal of DNA methylation at the cis-regulatory elements in the promoters of two cell migration genes (RAC1 and RHOC) abolished the recruitment of KLF4 WT to these promoters and repressed their gene transcription (Figure 5—figure supplement 1). Taken together, these new results demonstrated that KLF4 WT activated the expression of these two genes via binding to their highly methylated cis-regulatory elements nearby the coding regions.

2) The authors claimed that 116 genes are activated by mCpG-dependent KLF4 binding activity. Again, the evidence to support this claim is weak. Besides the reasons outlined in #1, there is also a need to establish the causality of KLF4 binding to methylated DNA in the cells and activation of these genes. Since the authors observed partial loss of cell adhesion and migration after 5-AZA treatment, it is important to show that the induction of these 116 genes is affected by 5-AZA in the direction that is consistent with their being targeted by Klf4.

The reviewer suggested a very important way to strength our findings. Please see our responses to essential revision #2 above for more details. To examine the impact of demethylation on global expression level, we performed RNA-seq analysis and compared changes in gene expression profiles with and without 5-Aza treatment. We observed that in the absence of KLF4 WT induction the overall expression levels of the 116 KLF4 WT-specific target genes were not significantly affected by 5-Aza treatment (Figure 4), as expected. However, when KLF4 WT was induced, the expression levels of 101 of the 116 genes were significantly reduced in 5-Aza treated cells as compared with untreated cells (Figure 4B B, p = 1.2e-14 compared to all genes), suggesting that DNA methylation is responsible for the transcription activation of these 101 genes.

3) The 5-AZA treatment leads to partial loss of migration of cells expressing WT Klf4. Is this due to DNA methylation dependent binding of Klf4? It is necessary to demonstrate loss of DNA methylation and loss of Klf4 binding at some sites after treatment.

The answer is yes. Please see our response to essential revision #2.

On the basis of the reviewers’ suggestion, we selected two genes, namely RAC1 and RHOC, because 1) they are known to play a crucial role in cell migration; 2) we found that they were associated with WT-specific ChIP-seq peaks and genome-wide bisulfite sequencing confirmed that these peaks contained highly methylated CpGs; and 3) these peaks are in close proximity to the coding regions of RAC1 (-922 bp to TSS) and RHOC (-507 bp to TSS), respectively. In the case of RAC1, we first showed that induction of WT, but not the R458A mutant, could induce RAC1 at both mRNA and protein levels using RT-PCR and immunoblot analysis, respectively (Figure 5—figure supplement 1A and B). Using a primer pair flanking RAC1 ChIP-seq peak, we confirmed that this fragment could be ChIPed substantially better with the KLF4 WT than the R458A mutant (Figure 5—figure supplement 1C). To further establish causality of DNA methylation-dependent binding of KLF4 on induction of its target genes, we performed bisulfite sequencing against this cis-regulatory element with and without the 5-Aza treatment. We observed that the methylation level of four of the six CpGs was reduced from ~100% to an undetectable level after 5-Aza treatment (Figure 5—figure supplement 1D). Most importantly, removal of methylation almost completely abolished KLF4 WT binding to this cis-regulatory element (Figure 5—figure supplement 1E), and expression of RAC1 was also significantly repressed (p < 0.001; Figure 5—figure supplement 1F). The same set of assays was also carried out to analyze the cis-regulatory elements of RHOC and the same results were obtained. Taken together, these new results demonstrated that KLF4 WT activated the expression of these two genes via binding to their highly methylated cis-regulatory elements nearby the coding regions.

Reviewer #2:

[…] Major points:

1) The authors observe KLF4 ChIP-seq in WT and R458A transfected cells. There are some key aspects that need to be included. How many / percentage of canonical and methylated sites are specific or shared between WT and R458A cells. A representation of the motifs needs to be addressed. Figure 3E, the authors need to provide the input track to the screenshot; it seems to be the background is high. Some other similar screenshots of genes as supplementary figures will be useful. MACS 1.4 has been used to call peaks for KLF4 while again MACS2 has been used to call peaks while comparing with the acetyl mark. It would be better if the authors use the same pipeline to call peaks, as they are different. Why do the authors need to call a broad peak in this case?

We appreciate the reviewer’s positive comments and insightful suggestions to strengthen our findings. As discussed in essential point #1 above, we have performed the whole genome bisulfite sequencing and combined this dataset with our ChIP-seq datasets. We observed that 10 methylated 6-mer motifs, each containing one CpG of high methylation level (β>0.6), were significantly enriched in the KLF4 WT-specific peaks, while distinct motifs without CpG sites were found enriched in the shared ChIP-seq peaks. Please see essential point #1 for more details.

We calculated numbers of WT-specific and shared peaks, respectively, including one of non-methylated motifs (5’-AAAGGA) and methylated (5’-CCCGCC) sites. We found that the non-methylated motif was significantly overrepresented in shared peaks (fold enrichment = 2.9, p = 7.7e-41), whereas the methylated motif was significantly enriched in WT-specific peaks (fold enrichment = 2.4, p = 3.0e-139).

We have also provided the input track and screenshot of some KLF4-mCpG target genes (see Figure 3—figure supplement 1). The new data is now in Figure 3E and Figure 3—figure supplement 1B and C.

We agree that MACS2 is widely used for calling ChIP-seq peaks. In general, the results by MACS 1.4 were quite similar to those obtained using MACS2. However, in our case some peaks observed obvious to eye-balling (for example, the left one of twin peaks associated with RHOC in Figure 3E) could not be identified by MACS2. Therefore, we decided to use MACS1.4 for better sensitivity. For histone modifications in this paper, we used MACS2 to call broad peaks without any problem.

2) The R458A mutant binds to ~1200 regions, the authors need to provide details regarding these regions.

We now provide the chromosomal locations of the WT and R458A shared peaks listed in Figure 3—source data 33.

3) The methods do not indicate whether biological replicates were performed for the sequencing based assays, in that case reproducibility is questionable. The authors can also provide a supplementary summary table for the mapped reads for all the sequencing experiments.

We appreciate the reviewer’s comments. We have added the summary tables in Figure 3—source data 1 and 2 for the mapped reads for all the sequencing experiments in our supplementary table and analyzed correlation of RNA-seq reads between the biological replicates. As showed in Figure 3—figure supplement 1Abelow, the reproducibility is very high (c.c. > 0.8) between our two experiments. The new data is now in Figure 3—figure supplement 1A.

4) Figure 1A: KLF4 expressed at a much higher level in the mutant version and can be observed consistently across the doses, still the authors find loss of physiological functions and KLF4 binding peaks needs explanation. The authors need to include a negative control or some KLF4 silencing assay to show WT KLF4 is directly correlated to the observed physiological changes such as cellular migration / wound healing / adhesion.

We apologize for the poor explanation of the cell system we used in our study. The purpose of using tet-on system, i.e., doxycycline (Dox)-inducible cell lines for KLF4 functional study, was to have an internally-controlled system, such that when the cells are not treated with Dox, they behave like the parental cells, therefore serves as a control cell line to study the biological function of KLF4. Compared to their own control cells, the induction level of KLF4 R458A was a bit higher than that of KLF4 WT, consistent with as our more quantitative RNA-seq data indicating the fold change of KLF4 WT and KLF4 R458A are 20 and 27 fold, respectively. However, the differences in expression level did not affect the ability of KLF4’s binding to methylated CpGs, hence cellular phenotype changes. In the revised version, we specifically explained the tet-on cell system we used (Results section). Moreover, as the reviewer suggested, we did cell migration assays in the non-transfected U87 cells (control) and found that non-transfected U87 cells behave the same as the non-Dox treated U87 KLF4 WT cells. Please refer to our response to reviewer #2, major point 8 later.

5) Figure 2B: Comparatively lesser focal location of F-actin and Vaculin can also be noted in R458A cells it will be helpful to if the authors show fields with similar number of cells as that of the WT cells.

In the revised version, we have chosen more representative photographs to replace Figure 2B. These images support our conclusion that after Dox treatment KLF4 WT induced more focal locations as shown by F-actin and vinculin staining, and the cells were more spread out.

6) Figure 5F subsection “KLF4-mCpG interactions activate RHOC via chromatin remodelling” It is evident treatment of WT cells with 5-aza and doxycycline induces expression of RHOC, so it is apparent expression of RHOC is not absolutely methylation / KLF4 dependent. The authors should explain this in their text. Authors should have R458A version of the figure.

We redid the WB experiments together with KLF4 R458A mutant cells as the reviewer suggested. It showed that 5-Aza itself had minimal effect on the baseline expression of RHOC expression (Figure 5). In KLF4 WT cells, 5-Aza blocked RHOC upregulation induced by KLF4 WT. In KLF4 R458A cells, 5-Aza had no effect on RHOC expression under similar conditions. The data is now in Figure 5G in the revised manuscript.

7) What is the methylation status of the WT specific KLF4 chip peaks sharing acetylation mark.

The reviewer raised a very good point. Of 1,773 WT-specific KLF4 chip peaks sharing acetylation mark, 978 (55.2%) contain CpG sites of >60% methylation level.

Reviewer #3:

[…] Major points:

1) […] In addition, two technical modifications of these analyses are needed to further avoid bias. First, the comparisons among binding sites, induced genes, marks etc need to be restricted to genes that are expressed at all in these cells – it is possible to get high overlaps among unrelated regulators only because everything tends to pile up on open chromatin and active promoters. Similarly, the GO analyses later in the paper should be done taking the induced genes as a background. KLF4 overexpression induces migration and morphological changes, which could explain why genes related to this process are expressed. It is possible that KLF4 only directly induces a small number of genes, and the rest of what we are seeing is a secondary effect.

We appreciate the points raised by the reviewer and performed additional statistical analyses according to the reviewer’s suggestions. We have calculated the statistical significance of the overlapped genes. We focused on 12,824 genes that were expressed (FPKM > 0.5) in U87 cells. Among them, 308 (2.4%) up-regulated genes were identified. In the meantime, we observed that KLF4 mCpG-dependent binding peaks were associated with 1,072 expressed genes, of which 65 were up-regulated genes. The ratio of 6.1% is significantly higher than the ratio of expressed genes not associated with KLF4 binding sites (fold enrichment = 2.5; p = 1.2e-12). Adding KLF4 mCpG binding sites in enhancer region, we finally identified 116 targets out of 2,601 genes, indicating significantly over-representation (fold enrichment = 1.9; p = 5.0e-13). In the revised manuscript, we added this new analysis in the text (subsection “Identification of direct targets of mCpG-dependent KLF4 interactions in GBM cells”). In addition, we performed whole genome bisulfite sequencing, integrated it with the ChIP-seq datasets, and re-analyzed the data (see our response to essential point #1). We also obtained RNA-seq data after 5-Aza treatment and re-analyzed the entire dataset (see our response to essential point #2). To firmly establish the causality between mCpG-dependent KLF4 binding and target gene activation, we performed in-depth assays to examine impact of de-methylation on the recruitment of KLF4 WT to the promoter regions of two target genes in conjunction with changes in their RNA level (see our response to essential point #2).

2) It seems possible that global DNA demethylation could attenuate the ability of cells to do many things, and thus the results shown in Figure 2C and D could be nonspecific. Are there other perturbations that can trigger the same responses that are being assayed, and are these responses intact in aza-treated cells? If not, then the effects cannot really be attributed to a specific effect of KLF4 binding.

Please see our response to essential point #3. As far as we know, 5-Aza treatment is the only effective means to reduce global DNA methylation level. In response to essential point #3, we have added discussion of 5-Aza treatment to the manuscript. To examine the impact of DNA demethylation on the global expression level, we performed RNA-seq analysis and compared changes in gene expression profiles with and without 5-Aza treatment. We observed that in the absence of KLF4 WT induction the overall expression levels of the 116 KLF4 WT-specific target genes were not significantly affected by 5-Aza treatment (see our response to essential point #2).

3) I do not see that Figure 5 adds anything to the paper. It is simply a reiteration of assays already used earlier on a global scale, but using PCR assays. The same issues I describe above in major points 1 and 2 also apply to these results – I believe that the final statement "Therefore, high methylation level in RHOC promoter is essential for KLF4 WT to activate RHOC transcription" is rather an overstatement. The first sentence is better supported: "Several lines of evidence supported that DNA methylation mediated RHOC activation by KLF4 WT". On the whole, however, this is the same argument that has already been made in genome-wide analyses, with no new evidence. I would suggest that this section is dispensible, since it only confirms what would be expected from the rest of the manuscript.

We respectfully disagree with the reviewer. Figure 5 is not redundant because it serves four purposes. First, it addressed the causality issue raised in essential point #2. Second, focused studies are essential for rigorous validation of high throughput analyses, which often contains many false positives. Third, RHOC plays a crucial role in cell migration/adhesion pathways. The discovery that the gene was activated via mCpG-mediated KLF4 binding activity is significant and of great interest in cancer biology. Most importantly, results showed in Figure 5 established a connection between dynamic changes of histone modifications and mCpG-dependent KLF4 binding activity in cells, which offered new molecular mechanism underlying mCpG-dependent gene activation. To address the concern of overstatement raised by the reviewer, we now offer more experimental and computational evidence as described in our responses to essential points #1 and #2. We believe that our in-depth study as illustrated in Figure 5 is well justified.

4) The association of active marks with induced promoters and with many of the Klf4 sites is unsurprising – this would be seen even if the induction of the supposed "target genes" was indirect, and the KLF4 binding was coincidental. Indeed, the fact that there are 3,593 novel H3K27ac peaks, but only 274 overlap KLF4, shows that the two are largely dissociated. Thus, the statement in the Discussion that "Our study reveals that mCpG-dependent binding activity of KLF4 serves as the link between methylated CpG and changes in chromatin status" is not supported by the data shown. It is "a" link, but not "the" link.

We have changed this sentence to "Our study reveals that mCpG-dependent binding activity of KLF4 serves as a link between methylated CpG and changes in chromatin status" (Discussion section, paragraph five).